8th International Meeting

8
th
International Meeting on
Statistical Climatology
University of Lneburg
12 to 16 March 2001

Schedule, Abstracts, Participants
1

STOCHASTIC CLIMATE MODELS 18
To the nature of red noise in the climatic series, G.V. Alekseev (Arctic and Antarctic Research
Institute, St.-Petersburg, Russia) 18
Frequency Modulation in the Instrumental Record, P. Carl (Institute for Freshwater Ecology and Inland
Fisheries, Research Association Berlin of the Leibniz Society, Berlin, Germany) 18
Statistical algorithms for solving inverse problems on predictability of regional climate, Anatoly I. Chavro, Yegor
V. Dmitriev (Institute for Numerical Mathematics RAS, Russia) 18
Validating Interannual Variability in Atmospheric General Circulation Models, Carsten S. Frederiksen
1
and
Xiaogu Zheng
2
(
1
Bureau of Meteorology Research Centre, Melbourne, Australia;
2
National Institute of
Water and Atmospheric Research, Wellington New Zealand) 19
The character of prediction utility in dynamical systems relevant to climate and weather, Richard Kleeman
(Courant Institute for Mathematical Sciences) 19
Stochastic dynamics in a simple model of seasonal open ocean deep convection, Till Kuhlbrodt (Potsdam Institute
for Climate Impact Research, Climate System Dept., Potsdam, Germany) 19
Prediction of ENSO using an optimized mixture of linear models, Frank Kwasniok (Leibnitz-Institut fr
Atmosphrenphysik, Khlungsborn, Germany) 20
Simple Stochastic Models of the Thermohaline Circulation, Adam Monahan (Humboldt-Universitaet zu Berlin) 20
STATISTICS, MODELS AND DATA ASSIMILATION 20
Improvement of prediction using SSA methods in addition to MEM: Case of NAO study, Sylvia Antunes
1
, H.
Oliveira Pires
1,2
e Alfredo Rocha
2
(
1
Instituto de Meteorologia, R. C ao Aeroporto, Lisboa,
2
Universidade
de Aveiro) 20
On the interrelationships between multisite climatic temperature forecasts obtained from downscaling regional
sea-level-pressure fields, Ulrich Callies (Institute for Hydrophysics, GKSS Research Center; Geesthacht,
Germany) 20
Model Error and Uncertainty, Peter Challenor (Southampton Oceanography Centre, Southampton) 21
The Winter Temperature Downscaling Experiments Using Canonical Correlation Analysis in China, Youmin
Chen (Agrometeorology Institute, Chinese Academy of Agric Sciences, Beijing, China) 21
On-line Gaussian Processes and their Potential in Data Assimilation, Dan Cornford, Lehel Csato and Manfred
Opper (Computer Science, Aston University, Birmingham) 21
Statistical Downscaling of Largescale GCM Data to Local Ice Accumulation in Greenland, Traute Crueger, Hans
v. Storch (GKSS Research Center, Geesthacht, Germany) 21
Spatialisation of 24 hours precipitation with 100 years return period, Mojca Dolinar
1
, Damijana Kastelec
2

(
1
Hydrometeorological institute of Slovenia, Ljubljana, Slovenia;
2
University of Ljubljana, Biotehnical
Faculty, Ljubljana, Slovenia) 22
Bayesian estimate of halphen type a law's parameters, S. El Adlouni
1
, A. Bouziane, D. Ouazar, A. Zoglat
1
(
1

Faculte des Sciences-Rabat; Ecole Mohammadia d'Ingenieurs, Rabat, Morroco) 22
STOCHASTIC CLIMATE MODELS 23
EMAD: An Empirical Model of the atmospheric Dynamics, Juan Pedro Montavez and Jin Song von Storch
(Meteorological Institute of Hamburg University) 23
2
On the Natural Regularity and Predictability of Terrestrial and Extra-Terrestrial Parameters, Nityanand Singh
(Indian Institute of Tropical Meteorology, Pashan, Pune, India) 23
Stochastic Tropical Atmospheric Boundary-Layer Dynamics: Convective Downdrafts and Recovery, Discharge-
Recharge Processes, and 1/f-Noise, Jun-Ichi Yano and Richard Blender (Meteorolo-gisches Institut,
Universitaet Hamburg, Hamburg, Germany) 23
Variational Assimilation of botanical proxy Data for the Reconstruction of Palaeoclimate, Christoph Gebhardt
(Meteorologisches Institut der Universitt Bonn, Germany); Norbert Khl (Institut fr Palontologie,
Universitt Bonn, Germany); Andreas Hense (Meteorologisches Institut der Universitt Bonn, Germany);
Thomas Litt (Institut fr Palontologie, Universitt Bonn, Germany) 24
How realistic is GCM-simulated precipitation? Nikolaus Groll, Martin Widmann (Institute of Hydrophysics,
GKSS Research Centre, Geesthacht, Germany) 24
Homogenisation and Interpolation of incomplete climatological Data with simple physically based Models,
Andreas Hense, Christoph Gebhardt (Meteorologisches Institut, Universitt Bonn,Bonn, Germany) 25
Trends of Maximum and Minimum Temperatures in Northern South America, Ramon A. Quintana-Gomez
(Department of Earth Sciences, Universidad de los Llanos Ezequiel Zamora, Barinas, Estado Barinas,
Venezuela) 25
Investigation of Spatial and Temporal Ozone Immission Structures under Stochastic-Deterministic Model
Assumptions, Heike Hoffmann (Graduate School of Spatial Statistics, Freiberg, University of Mining and
Technology, Germany) 25
Reduced-order schemes for data assimilation in oceanographiy based on the Kalman filter, I. Hoteit and D.T.
Pham (Laboratoire de Modelisation et Calcul, Grenoble, France) 25
Cloudiness and sea surface temperature: tackling the problem of missing values, Richenda Houseago-Stokes and
Peter Challenor (Southampton Oceanography, Centre Southampton, UK) 26
Removing errors in simulated distributions of climate elements using a non-parametric adjustment procedure,
Radan Huth
1
, Jan Kysel
1,2
(
1
Institute of Atmospheric Physics, Prague, Czech Republic;
2
Dept. of
Meteorology and Environment Protection, Charles University, Prague, Czech Republic) 26
Simplified EOFs - three alternatives to rotation, Ian Iolliffe (University of Aberdeen, UK); Mudassir Uddin
(University of Karachi, Pakistan); Karen Vines (Open University, UK) 26
Reconstruction of large-scale climate, J.M. Jones (GKSS Research Centre, Geesthacht, Germany) 27
On a probabilistic interpretation of variational data assimilation and model error estimation, Gennady A. Kivman
(Alfred-Wegener-Institut for Polar- and Marine Research, Bremerhaven, Germany) 27
Statistical presentation of wind wave climate, Leonid Lopatoukhin
1,2
, Valentine Rozhkov
1
, Alexander
Boukhanovsky
1,3
, (
1
State Oceanographic Institute;
2
St. Petersburg State University (Dep. Oceanology);
3

Institute of High Performance Computing and Data Base, Russia) 27
Model-based spatial modelling of climatological data, J. Mateu
1
, P. Juan
1
, C. Antolin
2,3
, C. Ano
3
28
(
1
Department of Mathematics, Universitat Jaume I, E-12071 Castellon, Spain;
2
Department of Vegetal Biology,
Universitat de Valencia, Burjassot, Spain;
3
CIDE-Centre for Desertification, Cami de la Marjal s/n.,
Albal (Valencia), Spain) 28
EXTREME VALUE ANALYSIS 28
Regional GEV-growth curves of yearly precipitatation depths in Belgium for durations ranging from 10 minutes
to 30 days, Daniel Gellens, Gaston Demare (Royal Meteorologuical Institute of Belgium, Brussels,
Belgium) 28
3
Extreme precipitation events in southeastern South America and the relationships with summertime large-scale
South American Stationary Systems (SASSY), Leila M. V. Carvalho
1,2
, Charles Jones
1
, Brant Liebmann
3
(
1
Institute for Computational Earth System Sciences (ICESS), University of California, Santa Barbara,
California, USA;
2
Dept. of Atmospheric Sciences, University of Sao Paulo, Sao Paulo, Brazil;
3
NOAA-
CIRES Climate Diagnostics Center, Boulder, Colorado,USA) 28
Diagnosis of variations in the climate-related flood risk - Some results from the Interior Western United States,
Shaleen Jain, Martin Hoerling, Gary Bates, Randall Dole (NOAA/CIRES Climate Diagnostics Center,
Boulder, CO) 28
The Climatological Probability of Drought in Northern Province of South Africa, Tiba Kabanda (School of
Environmental Sciences, University of Venda, Thohoyadou, South Africa) 29
Statistical Analysis of Point Discharge Current at Different Time Scale Over the Pune Region, S.S.
Kandalgaonkar, M.I.R. Tinmaker, M.K. Kulkarni (Indian Institute of Tropical Meteorology, Pune, India)
29
A Dedacal Study of Premonsoon Seasons Thunderstorms Over Pune Region by Using Markov Chain Model, S.S.
Kandalgaonkar, M.K. Kulkarni, M.I.R. Tinmaker (Indian Institute of Tropical Meteorology, Pune, India)
29
Trends of Atlantic wave extremes as simulated in a 40-year wave hindcast using kinematically reanalyzed wind
fields, Xiaolan Wang* and Val R. Swail (Network Strategies Division, AMWSD Meteorological Service of
Canada, Downsview Ontario, Canada) 30
Hurricane Landfall Winds Along the United States Gulf and Atlantic Coasts: Monte Carlo Simulations, Gary D.
Skwira*, Douglas A. Smith+ and Richard E. Peterson* (Wind Science and Engineering Program, Texas
Tech University, Lubbock, TX, USA; *Department of Geosciences, +Department of Civil Engineering) 30
Simulation of ocean wave climate changes through wind variance changes on different timescales, E. Bauer ,R.
Weisse ( PIK, Potsdam; GKSS, Geesthacht, Germany)
1
2 1 2
30
Frequency and Intensity of Mediterranean Cyclones in a doubled CO2 scenario, P.Lionello (University of Lecce),
F.Dalan (University of Padua) 30
Extreme waves and surge events in a doubled CO2 scenario in the Adriatic Sea. P.Lionello (University of Lecce)
E.Elvini, A.Nizzero (University of Padua) 31
Extreme value analysis of absolute minimum temperature and minimum dew-point temperature at Sodankyl,
Finland, Jaakko Helminen, Niklas Sjblom (Finnish Meteorological Institute (FMI), Helsinki, Finland) 31
Observed Changes in Temperature Extremes of Korea, Won-Tae Kwon, Hee-Jeong Baek*, Byung-Choel Choi
(Meteorological Research Institute, Korea Meteorological Administration; * Seoul National University,
Korea) 32
The role of atmospheric large-scale circulation over Europe and North Atlantic in sea level variability on the
Polish coastal zone, Miroslaw Mietus (Institute of Meteorology and Water Management, Maritime
Branch, Gdynia, Poland) 32
Space-time variograms and a functional form for total air pollution measures, De Iaco, S.
1
, Myers, D. E.
2
, Posa,
D.
3,4
(
1
Universita di Chieti, Pescara Italy
2
University of Arizona, Tucson AZ - USA;
3
Universit di
Lecce, Lecce - Italy;
4
IRMA - CNR, Bari - Italy) 32
Performance of four-dimensional variational assimilation and dynamical instability, experiments on a quasi-
geostrophic model, Carlos Pires (CGUL, - Centro de Geofisica da Universidade de Lisboa, Lisboa,
Portugal) 33
A multisite self-consistent Weather Generator for Impact Studies in Portugal, Budong Qian, Joo Corte-Real,
Hong Xu (ICAT, Faculty of Sciences, University of Lisbon, Portugal) 33
Monthly mean forecasts of rainfall and temperature in Portugal, Joao Rio, Carols Pires (CGUL, - Centro de
Geofisica da Universidade de Lisboa, Lisboa, Portugal) 33
4
Sensitivity Studies on Wind Measurement Errors in Mesoscale Model Simulations, Radhika Ramachandran,
Prakash M Dolas (Space Physics Laborator, Vikram Sarabhai Space Centre, 34
Trivandrum, India) 34
Building the bridge between sequential data assimilation and geostatistics: the stationarity issue, Julien Snegas,
Hans Wackernagel (Centre de Geostatistique, Ecole des Mines de Paris, Fontainebleau, France) 34
Public Debate and Climate Model Products: Whereof must one be silent? Leonard A. Smith (Centre for the
Analysis of Time Series, London School of Economics, London, UK) 34
Multiple Analysis of Series for Homogenization (MASH). Seasonal Application of MASH. Automatic using of
Meta Data, Tamas Szentimrey (Hungarian Meteorological Service, Budapest, Hungary) 35
A data assimilation method with a coupled ocean-land-atmosphere model and its impact on the ocean state and on
medium-range pre-dictions, Clemente A. S. Tanajura, Konstantin P. Belyaev (Laboratorio Nacional de
Computao Cientifica (MCT/LNCC), Petrpolis, RJ, Brazil) 35
Ben P. Kirtman (Center for Ocean-Land-Atmosphere Studies (IGES/COLA), Calverton, MD, USA) 35
Assessing Model Skill in Climate Simulations, Karl E. Taylor (Program for Climate Model Diagnosis and
Intercomparison (PCMDI), Lawrence Livermore National Laboratory, Livermore, CA, USA) 36
XXth Century Behaviour of Temperature and Precipitation Extremes in Hungary, Jnos Mika, Tams
Szentimrey (Hungarian Meteorological Service, Budapest, Hungary); Judit Bartholy, Rita Pongrcz
(Etvs Lrnd University, Dept. Meteorology, Budapest, Hungary); Lszl Szeidl (Department of
Mathematics, University of Pcs, Hungary) 36
Homogenized Daily Temperatures for Trend Analyses in Extremes over Canada, Lucie Vincent, Xuebin Zhang,
Barrie Bonsal and Bill Hogg (Climate Research Branch, Environment Canada) 36
The analysis of extreme temperature distributions on the example of Poland, Joanna Wibig (Department of
Meteorology and Climatology University of Lodz) 37
Comparison of 20- and 50-year return values in GCM-simulated, downscaled and observed temperature series,
Jan Kysel
1,2
, Radan Huth
1
(
1
Institute of Atmospheric Physics AS CR, Prague, Czech Republic;
2
Dept. of
Meteorology and Environment Protection, Charles University, Prague, Czech Republic) 37
Test on the probability distribution of maximum air temperatures, Jana Jurekov
1
, Jan Picek
2
, Jaroslava
Kalvov
1*
(
1
Charles University in Prague,
2
Technical University in Liberec, Czech Republic; *Dept. of
Meteorology and Environment Protection, Faculty of Mathematics and Physics, Charles University,
Praha, Czech Republic) 37
Three Approaches to determine extreme years of Global Atmospheric Temperature, Luis Gimeno
1,*
, Juan Ael
1
,
Higinio Jorge
1
, Pedro Ribera
1
, David Gallego
2
, Ricardo Garcia
2
, Emiliano Hernandez
2
(
1
Universidad de
Vigo. Facultad de Ciencias de Orense, Orense, Spain;
2
Universidad Complutense de Madrid. Facultad de
Ciencias Fisicas, Madrid. Spain; *Corresponding author) 38
Bayesian Statistical Modelling of Nonlinear Climate Processes, E. P. Campbell, Y. Li (CSIRO Mathematical and
Information Sciences, , Wembley, Australia) 38
Modelling and characterizing extreme drought events, Jess Abaurrea, Ana Carmen Cebrian (Dpto. Metodos
Estadsticos, Universidad de Zaragoza, Ed. Matemticas. Pedro Cerbuna, 12., Zaragoza, Spain,) 38
Extremes in Instationary Time Series, Jrgen Grieser (Institute for Meteorology and Geophysics, University
Frankfurt, Germany) 39
Comparison of Several Methods for the Estimation of Extreme Value Distribution in Time Series with Trends,
Henrique Oliveira Pires, Joao Pestana Ferreira (Instituto de Meteorologia, Lisboa, Portugal) 39
The Possible Influence of Large Volcanic Eruptions on El Nino, Philippe Naveau (National Center for
Atmospheric Research, Boulder, USA) 39
5
SIMULATION MODELS 39
Simplified atmosphere models with realistic variability, Ulrich Achatz (Leibniz-Institut fuer Atmosphaerenphysik
an der Universitaet Rostock e.V., Kuehlungsborn, Germany); J.D. Opsteegh (KNMI, De Bilt, The
Netherlands) 39
Modelling of Temporal Rainfall by the Use of Stochastic Point Process Models in View of Scaling, Konrad Bogner
(Institute of Hydrology, University of Agricultural Research,Vienna, Austria) 40
Statistical Modelling of Tornado Occurrence, Harold E. Brooks (NOAA/National Severe Storms Laboratory
Norman, Oklahoma) 40
A method to extend the temporal validity of nested regional climate simulations, Udo Busch and Dietrich Heimann
(DLR Oberpfaffenhofen, Institut fr Physik der Atmosphre, Germany) 40
Stochastic generation of daily precipitation, Aristita Busuioc
1
, Hans von Storch
2
and Rodica Tomozeiu
1
(1
National Institute of Meteorology and Hydrology, Bucharest, Romania; 2 Institute of Hydrophysics,
GKSS Research Center, Geestacht, Germany) 41
Statistical Downscaling of Daily, Multi-Site Precipitation in Tropical and Sub-Tropical Climates, Stephen P.
Charles
1
,
Bryson C. Bates
1
, Stephen J. Crimp
2
and James P. Hughes
3
(
1
CSIRO Land and Water,
Wembley, Australia;
2
Queensland Department of Natural Resources, Indooroopilly, Australia;
3
Department of Biostatistics, University of Washington, Seattle, USA) 41
Stochastic Weather Generator Model for Simulation of Multiple Meteorological Time Series, Ildik Dobi-
Wantuch, Jnos Mika (Hungarian Meteorological Service, Budapest, Hungary); Lszl Szeidl
(Department of Mathematics, University of Pcs, Hungary) 42
Interdiurnal and interannual variability in stochastic daily weather generator: Modelling and the role in
agricultural and hydrologic studies, Martin Dubrovsky
1
, Josef Hejzlar
2
, Jan Kysely
3
, Josef Buchtele
4
,
Zdenek Zalud
5
, Daniel Marvan
6
(
1
Institute of Atmospheric Physics AS CR, Hradec Kralove, Czech
Republic (CR),
2
Hydrobiological Institute AS CR and Faculty of Biological Sciences USB, Ceske
Budejovice, CR;
3
Institute of Atmospheric Physics AS CR, Prague, CR;
4
Institute of Hydrodynamics AS
CR, Prague, CR;
5
Mendel University of Agriculture and Forestry, Brno, CR;
6
Faculty of Mathematics
and Physics, Charles University, Prague, CR) 42
Temperature predictability in the greater Mediterranean area, R. Garca, E. Hernndez, T. Muoz, P. Ribera*
and L. Gimeno* (Dto Fsica de la Tierra II, Universidad Complutense, Madrid Spain; Dto Fsica
Aplicada, Universidad de Vigo, Campus de Orense, Orense, Spain) 42
A new validation scheme for the evaluation of climate model outputs, F.-W. Gerstengarbe, M. Kcken, P.C.
Werner (Potsdam-Institute for Climate Impact Research, Potsdam, Germany) 43
The construction of multi-site and multi-variate scenarios for Mediterranean regions based on Monte Carlo
simulations with a conditional weather generator, C.M. Goodess and J.P. Palutikof (Climatic Research
Unit, University of East Anglia, Norwich, UK) 43
Reconstructing large-scale variability from paleoclimatic evidence by means of pattern nudging, Martin Widmann
1
, Hans von Storch
1
, Ingo Kirchner
2
(
1
GKSS Research Center, Geesthacht, Germany,
2
Max-Planck-
Institute for Meteorology, Hamburg, Germany) 44
Spatio-temporal patterns of Meteosat Satellite Thermal Infrared Derived Convective Clouds over Ethiopia as
Described by PCA & MVC, Abebe Yeshanew (Senior Meteorologist, National Meteorological Services,
Addis Ababa, Ethiopia) 44
Skill of seasonal hindcasts as a function of the ensemble size, F.W. Zwiers and V.V. Kharin (Canadian Centre for
Climate Modelling and Analysis, Victoria, BC, Canada) 44
6
Determining extreme sets in multivariate distributions, onika. Lakatos (Hungarian Meteorological Service); I.
Matyasovszky (Department of Meteorology, Etvs Lornd University, Budapest) 44
The Probability Of An Extreme In A Correlated Time-Space Domain, Albert R. Boehm 45
Climate extremes in Germany and its relation to the objective weather type classification of the German Weather
Service, Peter Bissolli, Udo Baum (Deutscher Wetterdienst, Offenbach, Germany) 45
Statistical methods and case studies investigating extreme climatology of the Carpathian Basin, Judit Bartholy ,
Laszlo Szeidl ( Dept. of Meteorology Eotvos Lorand University, Budapest Hungary; Dept. of
Mathematics University of Pecs, Hungary)
1
2 1 2
45
SIMULATION MODELS 46
Simulation of multivariate time series of meteorological data, Asa Forsman, Claes Bjrklund and Anders Grimvall
(Department of Mathematics, Linkping University, Linkping, Sweden) 46
Modelling the probability of precipitation using the Marcov chain MonteCarlo technique, Martin Kappler
(Fachbereich Statistik, Universitt Dortmund); Andreas Hense (Meteorologisches Institut Universitt
Bonn); Siegfried Schach (Fachbereich Statistik Universitt Dortmund) 46
Reproduction of variance in statistical downscaling: inflation vs. noise addition, Radan Huth
1
, Jan Kysel
1,2

(
1
Institute of Atmospheric Physics, Prague, Czech Republic;
2
Dept. of Meteorology and Environment
Protection, Charles University, Prague, Czech Republic) 46
Simulating meteorological data by block resampling, Anders Nordgaard, Asa Forsman, Anders Grimvall
(Department of Mathematics, Linkping University, Linkping, Sweden) 47
The Amazon climate: A multiscale and nonstationary processes, Guillermo O. Obregn, Carlos Nobre (Centro de
previso de Tempo e Estudos Climticos - CPTEC / Instituto Nacional de Pequisas Espaciais INPE,
Cachoeira Paulista, So Paulo Brazil) 47
First indication of anthropogenic climate change in local wave conditions in the North Sea, Arnt Pfizenmayer
(Institute for Coastal Research, GKSS Research Centre; Geesthacht, Germany) 47
Ensemble Predictions with a Limited-Area Mesoscale Atmospheric Model - Implications for the Interpretation of
its Direct Model Output, Susanne Theis, Andreas Hense (Meteorologisches Institut der Universitt Bonn,
Bonn, Germany); Ulrich Damrath, Volker Renner (Deutscher Wetterdienst, Offenbach, Germany) 47
Modeling stochastic structure of local climate as downscaling of GCM fields, 1,2Marina Timofeyeva and 1Rachael
Craig. (Department of Geology, Kent State University, Kent, Ohio; UCAR/NOAA/National Weather
Service, Silver Spring, Maryland) 48
Numerical experimentation with regional atmospheric models, Hans von Storch, Ralf Weisse (Institute of Coastal
Research, GKSS Research Centre, Germany) 48
DECISION MAKING AND DETECTION OF ANTHROPOGENIC CLIMATE CHANGE 49
Exploiting the emerging signal of anthropogenic climate change in probabilistic climate forecasting, Myles Allen
(Space Science & Technology Department, Rutherford Appleton Laboratory, Chilton, Didcot, UK) 49
Significance of Anthropogenic Radiative Forcings in Simulation of Temporal Pattern of Surface Air Temperature
in Monsoon Climate, Ajay Singh (School of Management, Indian Institute of Technology- Bombay, Powai,
India) 49
Decadal variations of Greenland precipitation over the past 1400 years, K. K. Andersen and P. D. Ditlevsen,
(Department of Geophysics, Copenhagen, Denmark) 49
Lessons taught from the Homogenization of 200 US First Order Stations, Auguste C. Boissonnade, Ph.D.
1
and
Larry Heitkemper
2
(
1
Risk Management Solutions, Inc. Menlo Park, California, USA;
2
EarthSat
Corporation, Washington D.C., USA) 49
7
North Atlantic osccillation projection on regional climate predictability, Roxana Bojariu and Daniela Paliu
(National Institute of Meteorology and Hydrology, Bucharest, Romania) 50
What influences decision-makers in drought prone areas when allocating water resources? Bernard Cornlis
(Dpartement de Gomatique Universit de Lige) 50
Assessing uncertainty in anthropogenic signal estimates using space-time and trend-based optimal detection
approaches, Nathan Gillett, Gabi Hegerl, Myles Allen 50
Applications of Statistical Climatology in Finance, Stephen Jewson, Anders Brix (Risk Management Solutions) 51
Trends in northern hemisphere winter cyclone activity based on a comparison of cyclone indices, Christopher J.
Paciorek (Department of Statistics, Carnegie Mellon University); James S. Risbey (Department of
Engineering and Public Policy, Carnegie Mellon University); Richard D. Rosen (Atmospheric and
Environmental Research, Inc. Cambridge, Massachusetts); Valerie Ventura (Department of Statistics,
Carnegie Mellon University) 51
Signal analysis of the atmospheric mean temperature 500/1000 hPa north of 55 N between 1949 and 1994, Heiko
Paeth, Andreas Hense (Meteorologisches Institut, Universitt Bonn, Bonn, Germany) 51
Expert Assessment of Uncertainties in Detection and Attribution of Climate Change, James Risbey (University of
Utrecht, Department of Science, Technology and Society, Utrecht, The Netherlands) 52
A Maximum Entropy Approach for Extracting Forecast Probabilities from Finite Ensembles, M.S. Roulston ,
L.A. Smith and C. Ziehmann ( Mathematical Institute, Oxford University, UK; Institute of Physics,
Potsdam University, Germany)
1
1 2 1 2
52
Coupled sea ice and atmosphere variability in the Weddell Sea, Antarctica, Silvia A. Venegas (Danish Center for
Earth System Science (DCESS), Niels Bohr Institute for Astronomy, Physics and Geophysics, University
of Copenhagen, Denmark); Mark R. Drinkwater (Oceans/Sea-Ice Unit, ESTEC Earth Sciences Division
(APP-FSO), Noordwijk, The Netherlands) 52
A Multi-Index Bayesian Approach to Climate Change Detection and Attribution, Klaus Hasselmann, Reiner
Schnur (Max Planck Institute of Meteorology, Hamburg, Germany) 53
Changes in statistical properties of daily precipitation in transient climate change experiments with a coupled
atmosphere-ocean GCM, V.A. Semenov and L. Bengtsson (Max Planck Institute for Meteorology,
Hamburg, Germany) 53
Near Two Centuries Monsoon Rainfall Variations across India: Reconstructed Past and Predicted 10-year Future,
N.A. Sontakke (Indian Institute of Tropical Meteorology, Pashan, India) 53
Statistical detection of the anthropogenic greenhouse signal in observed global and regional climate data fields, T.
Staeger, J. Grieser, C.-D. Schnwiese (Institut fr Meteorologie und Geophysik, J.W. Goethe Universitt
Frankfurt a.M.) 54
Trends in the Diurnal Temperature Range in the CCCma Coupled Model, Daithi A. Stone, Andrew J. Weaver
(School of Earth and Ocean Sciences, University of Victoria, Canada) 54
Detection of the causes of recent trends in radiosonde temperatures, P.W.Thorne
1
, P.D.Jones
1
, S.F.B.Tett
2
,
M.R.Allen
3
, D.E.Parker
2
, T.D.Davies
1
, and T.J.Osborn
1
(
1
Climatic Research Unit, UEA, Norwich, UK.;
2
Hadley Centre for Climate Prediction & Research, Meteorological Office, Bracknell, Berkshire, UK;
3
Rutherford Appleton Laboratory, Chilton, Didcot, UK.) 54
Signal Analysis and Detection Studies Using Neural Networks, Andreas Walter (Institute of Meteorology and
Geophysics, Johann-Wolfgang Goethe University, Frankfurt, Germany) 55
Potentially Predictable Patterns, Xiaogu Zheng
1
and Carsten S. Frederiksen
2
(1 National Institute of Water and
Atmospheric Research, Wellington, New Zealand; 2Bureau of Meteorology Research Centre, Melbourne,
Australia) 55
Extratropical cyclone variability in the Northern Hemisphere winter from the NCEP/NCAR Reanalysis data, Olga
Zolina, Sergey Gulev, Sergey Grigoriev (P.P.Shirshov Institut of Oceanology, RAS, Moscow) 55
8
POSTER CONTRIBUTIONS TO THE 8 INTERNATIONAL MEETING ON STATSITICAL
CLIMATOLOGY 56
An Explanation for the Multidecadal, Winter Rainfall Decline Over South-Western Australia, Bryson C. Bates ,
Stephen P. Charles , James P. Hughes , and Edward P. Campbell ( CSIRO Land and Water, Wembley,
WA 6913, Australia; Department of Biostatistics, University of Washington, Seattle WA 98195, USA;
CSIRO Mathematical and Information Sciences, Wembley, WA 6913, Australia)
1
1 2 3 1
2
3
56
Statistical Downscaling to Monthly Precipitation in the Northwestern of South America (Colombian Region),
Nestor R. Bernal S. (Nacional University of Colombia) 56
Better Quality of StatisTical Downscaling Models Without Preceding PC-Analyses For Predictands? Example:
Rainfall Assessments for Namibia U. Beyer (Institute of Geography, University of Wuerzburg,
Wuerzburg) 56
Multiple Non-linear Statistical Model for Estimating Mean Maximum Urban Heat Island in a Medium-Sized
Town Szeged, Hungary, Zsolt Bottyn, Jnos Unger (Department of Climatology and Landscape Ecology,
University of Szeged, Szeged, Hungary) 57
Application of factor analysis for quantification of climate forcing processes in the Baltic Sea region, Arunas
Bukantis Dept. Of Hydrology and Climatology of Vilnius University, Vilnius Lithuania) 57
A New Approach of Interpolation and Prediction of Climate Time Series, Hong-Xing Cao, Xiao-Jing Jia (Chinese
Academy of meteorological Sciences, Beijing ,China) 57
Wind Field Retrieval From Scatterometer Data, Dan Cornford, David J. Evans and Ian T. Nabney 58
Homogenization and Quality Control in State Space Models, Thomas K. Friedli, Evi Schuepbach, Tania Steiner,
Jrg R. Hsler (Department of Mathematical Statistics and Actuarial Sciences, University of Berne, Bern,
Switzerland) 58
Statistical Analysis of Modern Climate Changes observed in the West and East Antarctic, Jagovkina, S. V. ,
Lagun, V. E. ( Main Geophysical Observatory, St.-Petersburg, Russia; Arctic and Antarctic Research
Institute, St.-Petersburg, Russia)
1
2 1 2
58
Statistical Forecasts of Summertime South American Stationary Systems (SASSY), Charles Jones
1
and Leila M. V.
de Carvalhol
2
(1 Institute for Computational Earth System Sciences (ICESS), University of California,
Santa Barbara, CA; 2 Dept. of Atmospheric Sciences, University of So Paulo, Brazil) 59
Using of cluster analysis in processing of correlation coefficients, Stanislava Kliegrov, Ladislav Metelka (Czech
Hydrometeorological Institut, Hradec Krlov Svobodn Dvory, Czech Republic) 59
A new look at the temporal structure of a time series, Ken-Chung Ko (Department of Geography, National
Kaohsiung Normal University, Kaohsiung, Taiwan) 59
Extreme value analysis applied on precipitation time series in the Czech Republic, Kveton, V. (Czech
Hydrometeorological Institute, Prague, Czech Republic) 59
Statistical climatology and interannual variability of the extratropical cyclone parameters in the global
atmosphere, Victor E. Lagun (Air-Sea Interaction Department, Arctic and Antarctic Research Institute,
St. Petersburg, Russia) 59
Spatial Organization of Decadal and Bidecadal Rainfall Fluctuation on Southern North America and Southern
South America, Omar Abel Lucero, Norma C. Rodriguez (Instituto Nacional del Agua and National
University of Cordoba, Cordoba, Argentina) 60
Interrelationship between Interdecadal Fluctuations in the Southern Oscillation and the Sea Surface Temperature
of the Tropical Pacific, Omar Abel Lucero, Norma C. Rodriguez (Centro de la Region Semirida (INA),
and National University of Cordoba.Cordoba, Argentina) 60
Variability of annual pressure and precipitation on Iberian Peninsula, M Dolores Manso Orgaz* and Liliana
Caramelo** (*Depart. of Physics. University of Aveiro, Portugal; ** Depart. of Physics University of
Trs-os-Montes and Alto Douro . Vila Real) 60
9
Reproduction of dry and wet Periods by Stochastic Daily Precipitation Generators, D. Marvan
1
, M. Dubrovsky
2
(
1
Faculty of Math. and Physics, Charles University, Prague, Czech Republic;
2
Institute of Atmospheric
Physics, Hradec Kralove, Czech Republic) 61
Semi-parametric statistical approaches for the analysis of climatological data, J. Mateu
1
, C. Alvarez2, T. Sanfeliu
2
,
M.M. Jordan
3
(
1
Department of Mathematics;
2
Department of Experimental Sciences, Universitat Jaume
I, E-12071 Castellon, Spain;
3
Department of Experimental Sciences, Universitat Miguel Hernandez,
Elche, Spain) 61
Statistical Methods in climatology: an application to the longest precipitation series in Portugal, Solange
Mendona Leite (Geophysical Centre of the University of Lisbon, Portugal) 61
Spatial patterns of the five-day rainfall all over North-Western Iberian peninsula, M. Andrs (Atmospheric
Physics Department, University of Salamanca, Spain); Clemente Toms (Atmospheric Physics
Department, University of Salamanca, Spain); Fernando de Pablo (Atmospheric Physics Department,
University of Salamanca, Spain); Solange Mendona Leite (Geophysical Centre of the University of
Lisbon, Portugal) 62
The effects of large scale phenomena on local climate variability over Romanian territory, Daniela Mihaela Paliu,
Roxana Bojariu, (National Institute of Meteorology, Hydrology and Water Management, Bucharest,
Romania) 62
Neural networks to predict atmospheric profiles, Nazario D. Ramirez-Beltran (Department of Industrial
Engineering, University of Puerto Rico, Mayaguez); Jorge Gonzalez (Department of Mechanical
Engineering, University of Puerto Rico, Mayaguez); Felipe Chong Rengifo (Department of Industrial
Engineering, University of Puerto Rico, Mayaguez) 62
Mid-term prediction of the daily maximum/minimum temperatures in South Korea using Dynamic Linear
Models, Keon Tae Sohn (Pusan National University, Korea); Jeong Hyung Lee (Dong-A University,
Korea); Byungsoo Kim (Inje University, Korea) 62
Statistics in Agricultural Climatology adapting a workshop for African needs, S. M. Gathara, C. Oludhe, S. B.
Otengi, P. K. Runanu, R. D. Stern
63
Changes of large-scale atmosphere circulation during XX century and their influence on the climate in Ukraine,
V.F. Martazinova, T. Sverdlik (Ukrainian Hydrometeorological Institute, Kiev, Ukraine) 63
Validating a Downscaling Model of the Stochastic Structure of Daily Temperature, Marina Timofeyeva
1
, Rachael
Craig
2
(
1
UCAR, NOAA/NWS, Silver Spring, MD;
2
Dept. of Geology, Kent State University, Kent, OH) 63
Statistical Decomposition of European Temperature Time Series, Silke Trmel, Jrgen Grieser, C.-D. Schnwiese
(Institut fr Meteorologie und Geophysik, Universitt Frankfurt, Frankfurt /M., Germany) 64
Coupled sea ice and atmosphere variability in the Weddell Sea, Antarctica, Silvia A. Venegas (Danish Center for
Earth System Science (DCESS), Niels Bohr Institute for Astronomy, Physics and Geophysics, University
of Copenhagen, Denmark); Mark R. Drinkwater (Oceans/Sea-Ice Unit, ESTEC Earth Sciences Division
(APP-FSO), Noordwijk, The Netherlands) 64

10

MONDAY 12.3.2001

Plenary session: Stochastic Climate Models
9:00 Greg Holloway
9:45 Comment: J in-Song von Storch
10:00 Comment: Andreas Hense

10:15 Coffee break

10:45 Claude Frankignoul
11:15 Norm McFarlane

11:45 Lunch
13:00 Poster introduction

Session 1: Stochastic Climate Models
14:30 G.V. Alekseev:
To the nature of red noise in the climatic series
14:50 P. Carl:
Frequency Modulation in the Instrumental Record
15:10 Anatoly I. Chavro:
Statistical algorithms for solving inverse problems on predictability of regional climate
15:30 Carsten S. Frederiksen:
Validating Interannual Variability in Atmospheric General Circulation Models

15:50 Coffee break

16:20 Richard Kleeman:
The character of prediction utility in dynamical systems relevant to climate and weather
16:40 Till Kuhlbrodt:
Stochastic dynamics in a simple model of seasonal open ocean deep convection
17:00 Frank Kwasniok:
Prediction of ENSO using an optimized mixture of linear models
17:20 Adam Monahan:
Simple Stochastic Models of the Thermohaline Circulation

Session 2: Statistics, Models and Data Assimilation
14:30 Sylvia Antunes:
Improvement of prediction using SSA methods in addition to MEM: Case of NAO study
14:50 Ulrich Callies:
On the interrelationships between multisite climatic temperature forecasts obtained from
downscaling regional sea-level-pressure fields
15:10 Peter Challenor:
Model Error and Uncertainty
15:30 Youmin Chen:
The Winter Temperature Downscaling Experiments Using Canonical Correlation Analysis in
China

15:50 Coffee break

16:20 Dan Cornford:
On-line Gaussian Processes and their Potential in Data Assimilation
16:40 Traute Crueger:
Statistical Downscaling of Largescale GCM Data to Local Ice Accumulation in Greenland
17:00 Mojca Dolinar:
Spatialisation of 24 hours precipitation with 100 years return period
17:20 S. El Adlouni:
Bayesian estimate of halphen type a law's parameters

18:00 Icebreaker-Party at the university

TUESDAY 13.3.2001

11
Plenary session: Statistics, Models and Data Assimilation
9:00 Geir Evensen
9:45 Comment: Carlos Pires
10:00 Comment: Christian Lajaunie

10:15 Coffee break

10:45 Valerie Isham
11:15 J ean-Paul Chils

11:45 Lunch

Session 1: Stochastic Climate Models
13:00 J uan Pedro Montavez:
EMAD: An Empirical Model of the atmospheric Dynamics
13:20 Nityanand Singh:
On the Natural Regularity and Predictability of Terrestrial and Extra-Terrestrial Parameters
13:40 J un-Ichi Yano:
Stochastic Tropical Atmospheric Boundary-Layer Dynamics: Convective Downdrafts and
Recovery, Discharge-Recharge Processes, and 1/f-Noise

13:00 Christoph Gebhardt:
Variational Assimilation of botanical proxy Data for the Reconstruction of Palaeoclimate
13:20 Nikolaus Groll:
How realistic is GCM-simulated precipitation?
13:40 Andreas Hense:
Homogenisation and Interpolation of incomplete climatological Data with simple physically
based Models
14:00 Ramon A.Quintana-Gomez
Trends of Maximum and Minimum Temperatures in Northern South America
14:20 Heike Hoffmann
Investigation of Spatial and Temporal Ozone Immission Structures under Stochastic-
Deterministic Model Assumptions
14:40 I. Hoteit
Reduced-order schemes for data assimilation in oceanographiy based on the Kalman filter

15:00 Coffee break

15:30 Richenda Houseago-Stokes
Cloudiness and sea surface temperature: tackling the problem of missing values
15:50 Radan Huth
Removing errors in simulated distributions of climate elements using a non-parametric
adjustment procedure
16:10 Ian Iolliffe
Simplified EOFs - three alternatives to rotation
16:30 J .M. J ones
Reconstruction of large-scale climate
16:50 Gennady A. Kivman
On a probabilistic interpretation of variational data assimilation and model error estimation
17:10 Leonid Lopatoukhin
Statistical presentation of wind wave climate
17:30 J . Mateu
Model-based spatial modelling of climatological data

Session 3: Extreme Value Analysis
13.00 Daniel Gellens
Regional GEV-growth curves of yearly precipitatation depths in Belgium for durations
ranging from 10 minutes to 30 days
12
13.20 Leila M. V. Carvalho
Extreme precipitation events in southeastern South America and the relationships with
summertime large-scale South American Stationary Systems (SASSY)
13.40 Shaleen J ain
Diagnosis of variations in the climate-related flood risk - Some results from the Interior
Western United States
14.00 Tiba Kabanda
The Climatological Probability of Drought in Northern Province of South Africa
14.20 S.S. Kandalgaonkar
Statistical Analysis of Point Discharge Current at Different Time Scale Over the Pune Region
14:40 S.S. Kandalgaonkar
A Dedacal Study of Premonsoon Seasons Thunderstorms Over Pune Region by Using Markov
Chain Model
15.00 Xiolan Wang
Trends of Atlantic wave extremes as simulated in a 40-year wave hindcast using kinematically
reanalyzed wind fields

15.30 Coffee break

15.50 Gary D. Skwira
Hurricane Landfall Winds Along the United States Gulf and Atlantic Coasts: Monte Carlo
Simulations
16.10 Eva Bauer
Simulation of ocean wave climate changes through wind variance changes on different
timescales
16.30 P.Lionello
Frequency and Intensity of Mediterranean Cyclones in a doubled CO2 scenario
16:50 P.Lionello
Extreme waves and surge events in a doubled CO2 scenario in the Adriatic Sea
17.10 Niklas Sjblom
Extreme value analysis of absolute minimum temperature and minimum dew-point
temperature at Sodankyl, Finland
17.30 Won-Tae Kwon
Observed Changes in Temperature Extremes of Korea

WEDNESDAY 14.3.2001

Plenary session: Extreme Value Analysis
9:00 Adri Buishand
9:45 Comment: Hans Alexandersson
10:00 Comment: David Stephenson

10:15 Coffee break

10:45 Lynda Chambers
11:15 Richard Chandler
11:45 Rick Katz

12:15 Lunch

13:30 Miroslaw Mietus
The role of atmospheric large-scale circulation over Europe and North Atlantic in sea level
variability on the Polish coastal zone
13:50 S. De Iaco
Space-time variograms and a functional form for total air pollution measures
14:10 Carlos Pires
13
Performance of four-dimensional variational assimilation and dynamical instability,
experiments on a quasi-geostrophic model
14:30 Budong Qian
A multisite self-consistent Weather Generator for Impact Studies in Portugal
14:50 J oao Rio
Monthly mean forecasts of rainfall and temperature in Portugal
15:10 Radhika Ramachandran
Sensitivity Studies on Wind Measurement Errors in Mesoscale Model Simulations

15:30 Coffee break

16:00 J ulien Snegas
Building the bridge between sequential data assimilation and geostatistics: the stationarity
issue
16:20 Leonard A. Smith
Public Debate and Climate Model Products: Whereof must one be silent?
16:40 Tamas Szentimrey
Multiple Analysis of Series for Homogenization (MASH). Seasonal Application of MASH.
Automatic using of Meta Data
17:00 Clemente A. S. Tanajura
A data assimilation method with a coupled ocean-land-atmosphere model and its impact on the
ocean state and on medium-range predictions
17:20 Karl E. Taylor
Assessing Model Skill in Climate Simulations

13:30 J anos Mika
XXth Century Behaviour of Temperature and Precipitation Extremes in Hungary
13.50 Lucie Vincent
Homogenized Daily Temperatures for Trend Analyses in Extremes over Canada
14.10 J oanna Wibig
The analysis of extreme temperature distributions on the example of Poland
14.30 J an Kysel
Comparison of 20- and 50-year return values in GCM-simulated, downscaled and observed
temperature series
14.50 J aroslava Kalvov
Test on the probability distribution of maximum air temperatures
15.10 Luis Gimeno
Three Approaches to determine extreme years of Global Atmospheric Temperature

15:30 Coffee break

16.00 E. P. Campbell
Bayesian Statistical Modelling of Nonlinear Climate Processes
16.20 J ess Abaurrea
Modelling and characterizing extreme drought events
16.40 J rgen Grieser
Extremes in Instationary Time Series
17.00 Henrique Oliveira Pires
Comparison of Several Methods for the Estimation of Extreme Value Distribution in Time
Series with Trends
17.20 Philippe Naveau
The Possible Influence of Large Volcanic Eruptions on El Nino

14
Session 4: Simulation Models
13:30 Ulrich Achatz
Simplified atmosphere models with realistic variability
13:50 Konrad Bogner
Modelling of Temporal Rainfall by the Use of Stochastic Point Process Models in View of
Scaling
14:10 Harold E. Brooks
Statistical Modelling of Tornado Occurrence
14:30 Udo Busch
A method to extend the temporal validity of nested regional climate simulations
14:50 Aristita Busuioc
Stochastic generation of daily precipitation
15:10 Stephen Charles
Statistical Downscaling of Daily, Multi-Site Precipitation in Tropical and Sub-Tropical
Climates

15:30 Coffee break

16:00 Ildiko Dobi-Wantuch
Stochastic Weather Generator Model for Simulation of Multiple Meteorological Time Series
16:20 Martin Dubrovsky
Interdiurnal and interannual variability in stochastic daily weather generator:
Modelling and the role in agricultural and hydrologic studies
16:40 R. Garcia
Temperature predictability in the greater Mediterranean area
17:00 Friedrich Wilhelm Gerstengarbe
A new validation scheme for the evaluation of climate model outputs
17:20 Clare Goodess
The construction of multi-site and multi-variate scenarios for Mediterranean regions based on
Monte Carlo simulations with a conditional weather generator

THURSDAY 15.3.2001

Plenary session: Simulation Models
9:00 J im Hughes
9:45 Comment: Francis Zwiers
10:00 Comment: Hans von Storch

10:15 Coffee break

10:45 Ruben Pasmanter
11:15 Ed Waymire

11:45 Lunch

13:00 Martin Widmann
13:20 Abebe Yeshanew
Spatio-temporal patterns of Meteosat Satellite Thermal Infrared Derived Convective Clouds
over Ethiopia as Described by PCA & MVC
13:40 Francis Zwiers
Skill of seasonal hindcasts as a function of the ensemble size

13.00 Monika Lakatos
Determining extreme sets in multivariate distributions
15
13:20 Albert R. Boehm
The Probability Of An Extreme In A Correlated Time-Space Domain
13.40 Peter Bissolli
Climate extremes in Germany and its relation to the objective weather type classification of
the German Weather Service
14.00 J udit Bartholy
Statistical methods and case studies investigating extreme climatology of the Carpathian Basin

Session 4: Simulation Models
13:00 Anders Grimvall / Asa Forsman
Simulation of multivariate time series of meteorological data
13:20 Andreas Hense / Martin Kappler
Modelling the probability of precipitation using the Marcov chain MonteCarlo technique
13:40 Radan Huth
Reproduction of variance in statistical downscaling: inflation vs. noise addition
14:00 Anders Nordgaard
Simulating meteorological data by block resampling
14:20 Guillermo Obregon
The Amazon climate: A multiscale and nonstationary processes
14:40 Arnt Pfizenmayer
First indication of anthropogenic climate change in local wave conditions in the North Sea

15:00 Coffee break

15:30 Susanne Theis
Ensemble Predictions with a Limited-Area Mesoscale Atmospheric Model - Implications for
the Interpretation of its Direct Model Output
15:50 Marina Timofeyeva
Modeling stochastic structure of local climate as downscaling of GCM fields
16:10 Hans von Storch
Numerical experimentation with regional atmospheric models

Session 5: Decision Making and Detection of Anthropogenic Climate Change
13:00 Myles Allen
Exploiting the emerging signal of anthropogenic climate change in probabilistic climate
forecasting
13:20 Ajay Singh
Significance of Anthropogenic Radiative Forcings in Simulation of Temporal Pattern of
Surface Air Temperature in Monsoon Climate
13:40 Katrine Krogh Andersen
Decadal variations of Greenland precipitation over the past 1400 years
14:00 Auguste Boissonnade
Lessons taught from the Homogenization of 200 US First Order Stations
14:20 Roxana Bojariu
North Atlantic osccillation projection on regional climate predictability
14:40 Bernard Cornlis
What influences decision-makers in drought prone areas when allocating water resources?

15:00 Coffee break

15:30 Nathan Gillett
Assessing uncertainty in anthropogenic signal estimates using space-time and trend-based
optimal detection approaches
15:50 Stephen J ewson / Anders Brix / Rodrigo Caballero
Applications of Statistical Climatology in Finance
16:10 Christopher J . Paciorek
Trends in northern hemisphere winter cyclone activity based on a comparison of cyclone
indices
16
16:30 Heiko Paeth
Signal analysis of the atmospheric mean temperature 500/1000 hPa north of 55 N between
1949 and 1994
16:50 J ames Risbey
Expert Assessment of Uncertainties in Detection and Attribution of Climate Change
17:10 Christine Ziehmann
A Maximum Entropy Approach for Extracting Forecast Probabilities from Finite Ensembles

19:00 Conference Dinner at Hotel Seminaris

FRIDAY 16.3.2001

Plenary session: Decision Making
9:00 Granger Morgan
9:45 Comment: Myles Allen
10:00 Comment

10:15 Coffee break

10:45 Klaus Hasselmann
11:15 Mark Berliner
11:45 Chris Forest

12:15 Lunch

Session 5 Decision Making and Detection of Anthropogenic Climate Change
13.20 Silvia A. Venegas
Coupled Sea Ice and Atmosphere Variability in the Weddell Sea, Antarctica
13:40 Reiner Schnur
A Multi-Index Bayesian Approach to Climate Change Detection and Attribution
14:00 V.A. Semenov
Changes in statistical properties of daily precipitation in transient climate change experiments
with a coupled atmosphere-ocean GCM
14:20 N.A. Sontakke
Near Two Centuries Monsoon Rainfall Variations across India: Reconstructed Past and
Predicted 10-year Future
14:40 Tim Staeger
Statistical detection of the anthropogenic greenhouse signal in observed global and regional
climate data fields

15:00 Coffee break

15:30 Daithi Stone
Trends in the Diurnal Temperature Range in the CCCma Coupled Model
15:50 P.W. Thorne
Detection of the causes of recent trends in radiosonde temperatures
16:10 Andreas Walter
Signal Analysis and Detection Studies Using Neural Networks
16:40 Xiaogu Zheng
Potentially Predictable Patterns
17:00 Olga Zolina
Extratropical cyclone variability in the Northern Hemisphere winter from the NCEP/NCAR
Reanalysis data

17
Stochastic Climate Models

To the nature of red noise in the climatic series,
G.V. Alekseev (Arctic and Antarctic Research In-
stitute, St.-Petersburg, Russia)

It is considered the forming of low-frequency com-
ponents in the climatic spectra of atmospheric cha-
racteristics that was named low-frequency weather
noise. The base of approach is statistical model of
weather variability that include individual weather
disturbances (IWD). This IWD are re-presented
with help of self-model function including constant
shape function and set of IWD parameters
(amplitude, duration, time of IWD start). It is
known that spectrum of a process described of such
model depends on spectrum shape of individual
disturbance. On the base of experiments with
proposed model was shown that as a rule spectrum
of many IWD with duration T has a maximum at
periods much more than T. Averaging of time
series including IWD amplifies this effect. In this
connection it should be noted that low-frequency
rise of climatic spectra could be more consequence
of transfer of local structures energy to the low-
frequency part of spectra in response to linear pro-
cedures of spectral analysis and averaging than real
low-frequency variability.

Frequency Modulation in the Instrumental Re-
cord, P. Carl (Institute for Freshwater Ecology
and Inland Fisheries, Research Association
Berlin of the Leibniz Society, Berlin, Germany)

Multiple cyclic forcing and the climate system`s
internal oscillatory capabilities (or nature), from
intraseasonal to decadal and longer, may bear
complex, evolving modal structures as found in
energy density spectrograms (resp. 'scalograms`)
throughout the instrumental record. Customary
methods used to visualize features in the time-fre-
quency plane, including "Short-Time" (windowed)
Fourier and Wavelet Transforms (STFT, WT), are
suited to provide a qualitative view of these evo-
lutions. The question arises to what extent such
apparent structural nonstationarities reflect anhar-
monic (maybe anthropogenic) climate drifts and/or
just higher-order cyclostationarity comprising the
frequency domain itself via systematic frequency
modulation (FM). As a basic conception of commu-
nication, FM of a carrier mode might be a principle
at work in climatic signal transmission as well---
and might even hint at topological constraints (i.e.
geometrical structures in the back) of the system`s
phase space trajectories. Recent paleoclimatic
studies hint at harmonic FM of Milankovich cycles,
and the challenging departure since about the 40th
from apparent astronomical control of the phase of
the seasonal cycle, as found in temperature records
of the instrumental period, deserves further clari-
fication.
Reconstruction from time series of the score behind
the 'sounds' of climate and climatic change (and
inference on respective 'castings') calls for high-re-
solution, quantitative, joint analyses of multiple sig-
nal attributes among which the dimensions of mo-
dulation must not be neglected. Adaptive waveform
("Matching Pursuit"; MP) analysis using a highly
redundant dictionary of frequency-modulated
Gabor atoms has been applied to this end to a
representative set of climatic time series of the
19/20th centuries. FM-MP "structure books" cove-
ring all parameters of these (non-orthogonal) de-
compositions - a sort of empirical time series mo-
dels in a seven-dimensional signal space - are pro-
vided and discussed. Comparison is made to the
WT which yields a time-scale energy distribution
'as is', and to the STFT which provides a spectro-
gram that relies (as the FT heavily does) on mutual
interference of Fourier modes. Like the STFT, the
MP method interprets the energy density distri-
bution in terms of its dictionary, but overcom-
pleteness of the latter and intrinsic provision of
nonstationarity relaxes massive interferences to
modest here. Unlike WT, the MP approach bears
"scale" as a truly independent parameter, resulting
in localizations that only depend on the scale of a
(partial) signal, not on its specific position in the
fre-quency domain; spectrograms provide high re-
solution throughout.
Deep FM is generally found to fit best into WT
energy density distributions of the instrumental re-
cord as far as considered. This is a surprise which
might underscore the potential topological aspects
mentioned. Though the signal space has to be fur-
ther extended to include harmonic amplitude modu-
lation (AM) and modulated drifting modes at least,
and the high demand on resolution poses resource
problems which call for massively parallel
computation, another surprise at the present stage
of analysis is a clear phase coincidence between
leading modes of both insolation and temperatures
since the 70th. The phase of leading 'slow' modes
points to decadal scale cooling ahead (but a con-
servative projection is intrinsic to the method). The
problem of Global Warming is 'preserved', how-
ever, in second-rank modes of growing signal
energy which should be taken seriously with a view
on their next warming phase.

Statistical algorithms for solving inverse problems
on predictability of regional climate, Anatoly I.
Chavro, Yegor V. Dmitriev (Institute for
Numerical Mathematics RAS, Russia)

New algorithms for solving inverse problems on the
restoration of small-scale structure of geophysical
fields in separate regions by large, space- and time-
averaged fields are proposed. The algorithms are
multi-dimensional statistical models for forecasting
small-scale structure of most probable anomalies of
geophysical fields in separate regions using space-
averaged values of global fields (or low-frequency
component part of the spectrum), which can be
predicted by the general circulation models with
rough resolution.
18
The problems are solved in the frameworks of line-
ar models. It was supposed that anomalies of regio-
nal climate of field f are connected with global ave-
raged field x as x =Af+n (1) where A is a linear
operator and n is the error of such representation.
An operator R is constructed on the base of a priory
information such that acting on equation (1) one
can get a best (in a sense of mean-square error) esti-
mation of the regional field f=Rx.
To construct R we used the methods of reduction
[1] or statistical regularization [2].The proposed
methods also allow to control the reliability of sta-
tistical models for solving the inverse problem. The
results of numerical experiments of the restoration
of the regional values of such geophysical fields as
mean surface temperature, high-frequency compo-
nent of H500 field and precipitation. The proposed
methods allow to restore over 70% of anomalies of
geophysical fields using the large-scale values pre-
dicted by the general circulation models.

Validating Interannual Variability in Atmospheric
General Circulation Models, Carsten S.
Frederiksen
1
and Xiaogu Zheng
2
(
1
Bureau of
Meteorology Research Centre, Melbourne,
Australia;
2
National Institute of Water and
Atmospheric Research, Wellington New Zealand)

This paper is an effort to explore general statistical
procedures for validating the inter-annual variabili-
ty in ensembles of atmospheric general circulation
model simulations forced by observed sea surface
temperatures (SSTs). The method relies on being
able to decompose the interannual variance of a
seasonal mean of daily meteorological values into
three parts: a forced component, a low frequency
internal source component and a weather-noise
component. How well models reproduce the inter-
annual variability of the components of seasonal
means is a topic of great interest. Ideally, in asses-
sing, or validating, a model's simulation of climate
variability, the emphasis should be on the extent to
which the forced component of variation is correct-
ly simulated since this is potentially predictable.
Here, we propose an estimation method for the
correlation coefficient between the SST-forced
components of simulated and observed seasonal-
means. The technique has been applied to DJ F and
J J A seasonal means of 200hPa geopotential heights
using daily data from an ensemble of 4 runs of the
Bureau of Meteorology Research Centre (BMRC)
AGCM for the period 1958 - 1991. We have taken
as our "proxy" observed data the daily 200hPa
geopotential height field from the National Centre
for Environmental Prediction (NCEP) and National
Centre for Atmospheric Research (NCAR) reana-
lyses over the same period.

The character of prediction utility in dynamical
systems relevant to climate and weather, Richard
Kleeman (Courant Institute for Mathematical
Sciences)

A new parameter of dynamical system predicta-
bility is introduced which measures the potential
utility of predictions. It is shown that this parameter
satisfies a generalized second law of thermody-
namics in that for Markov processes utility declines
monotonically to zero at very long forecast times.
Expressions for the new parameter in the case of
Gaussian prediction ensembles are derived and a
useful decomposition of utility into dispersion
(roughly equivalent to ensemble spread) and signal
components (roughly equivalent to ensemble mean
squared) is introduced. Earlier measures of predic-
tability are shown to have only considered the dis-
persion component of utility. A variety of simple
dynamical systems with relevance to climate and
weather prediction are introduced and the behavior
of their potential utility is analyzed in detail. For
the (stochastic) climate systems examined here, the
signal component is at least as important as the
dispersion in determining the utility of a particular
set of initial conditions. The simple ``weather'' sys-
tem examined (the Lorenz system) exhibited diffe-
rent behavior with the dispersion being more im-
portant than the signal at short prediction lags. For
longer lags there appeared no relation between
utility and either signal or dispersion. On the other
hand there was a very strong relation at all lags
between utility and the location of the initial con-
ditions on the attractor.

Stochastic dynamics in a simple model of seasonal
open ocean deep convection, Till Kuhlbrodt
(Potsdam Institute for Climate Impact Research,
Climate System Dept., Potsdam, Germany)

Aspects of open ocean deep convection variability
are explored with a two-box model of convective
mixing in a water column. In order to locate the
model in a region of parameter space relevant to the
real ocean, it is fitted to observational data of the
Labrador Sea. Seasonal cycles in the boundary
conditions allow for a realistic model fit without
changing the basic dynamics of the model. The
model suggests that Labrador Sea in the years
1964--1974 is close to a bistable regime where con-
vection is either òn' or òff'. The role of high-fre-
quency atmospheric variability in triggering con-
vection events is simulated by adding a noise term
to the surface buoyancy forcing. Depending on the
underlying deterministic model state, the effect of
the added noise is very different. First, when the de-
terministic model state is convective, the noise
leads to shorter convection events that are, how-
ever, more efficient in cooling the deep ocean.
Second, when the deterministic state is non-
convective, the noise induces convection events,
and for an optimal noise level a maximum vertical
density gradient is achieved. Third, for the bistable
deterministic state the noise leads to jumps between
the two model states of annual convection and
stable stratification. Here, the dependence of the
mean residence times in either state on model
parameters is studied. This reveals a mechanism for
19
scale interaction between the synoptic and the
decadal timescale.
The results highlight the crucial role of the high-
frequency surface forcing for the ocean's long-term
variability and its climatological mean state.

Prediction of ENSO using an optimized mixture of
linear models, Frank Kwasniok (Leibnitz-Institut
fr Atmosphrenphysik, Khlungsborn,
Germany)

A novel statistical prediction scheme for ENSO is
presented. The predictor-predictand relationship is
modelled by a mixture of linear regression models.
Each model is associated with a probability distri-
bution describing its domain of influence in phase
space. The joint probability distribution of predic-
tors and predictands is approximated by interpo-
lation among the models. The locations and the
prediction coefficients of the models are
determined by maximizing the likelihood of
observing a given learning data set using an
expectation-maximization procedure.
Unlike many other techniques, the method offers
forecasts of prediction errors. Two applications of
the concept described above to prediction of sea
surface temperature indices are presented. Firstly, a
linear ENSO prediction model is constructed from
observational data which takes into account the
annual cycle by conditioning the regression
coefficients on the phase of the annual cycle. Se-
condly, a state dependent modelling approach is
tried, i. e., the model coefficients depend on some
ENSO-related variables. A cross-validated predic-
tion skill is estimated and compared with that of
other prediction schemes.

Simple Stochastic Models of the Thermohaline
Circulation, Adam Monahan (Humboldt-
Universitaet zu Berlin)

Since the pioneering work of Stommel, simple
deterministic models have played a central role in
shaping our understanding of the thermohaline cir-
culation (THC). In particular, THC dynamics is
often analysed within the context of dynamical sys-
tems theory. The presence of even weak stochastic
perturbations can qualitatively alter the character of
variability in simple models of the THC. This talk
will address some of these effects of stochasticity,
including relative stabilisation of regimes and
stochastic induction of sensitivity to initial condi-
tions.

Statistics, Models and Data
Assimilation

Improvement of prediction using SSA methods in
addition to MEM: Case of NAO study, Sylvia
Antunes
1
, H. Oliveira Pires
1,2
e Alfredo Rocha
2

(
1
Instituto de Meteorologia, R. C ao Aeroporto,
Lisboa,
2
Universidade de Aveiro)

Singular Spectral Analysis (SSA) is a method that
allows the separation of series in isolated compo-
nents some identified as trend, other as components
with characteristics of periodicity. Using selected
significative components, grouped or isolated, this
method includes the reconstruction of series.
Reconstructed components are noisily reduced and
present a regular behaviour and these are the cha-
racteristics that permit the improvement in pre-
diction.
The method consists in fitting AR predictors esti-
mated by Maximum Entropy Methods (MEM) to
selected reconstructed components rather to the ori-
ginal series. In order to guarantee the independence
of data the predictors are estimated using half of the
series and tested in the remaining part. The pro-
cedure is applied separately for each of the com-
ponents and the final forecast is obtained by
summing the forecasts corresponding to each com-
ponent.
The North Atlantic Oscillation index, obtained u-
sing the Island pressure series and the recently
corrected pressure series of Lisbon is used to per-
form this study.

On the interrelationships between multisite clima-
tic temperature forecasts obtained from down-
scaling regional sea-level-pressure fields, Ulrich
Callies (Institute for Hydrophysics, GKSS Re-
search Center; Geesthacht, Germany)

A case study of evaluating the information content
of climatic temperature forecasts at four European
stations, obtained from downscaling regional sea-
level-pressure information, is presenteded. Fore-
casts are informative in the sense of this study if a
great proportion of observed temperature variability
can be attributed to these forecasts subsequent their
post-calibration based on a linear regression model.
As observations at different locations are generally
correlated, each forecast delivered for one
particular site will be informative about other sites
as well. Even when the respective local forecasts
are found to be best adapted to the conditions at a
particular station, it will generally be benecial to
establish a multiple-calibration scheme that
involves additional information from forecasts
delivered for other locations. The technique of
fitting conditional mo-dels is used to measure
incremental information obtained by consulting
several forecasts. Its application also allows the
determination whether additional information
gained from forecasts delivered for other sites is
genuine or if it is merely mediated by correlated
observations.
Results of conditional independence modelling are
con-trasted with the results of other techniques that
describe the inter-relationship between data by in-
troducing artificial variables obtained from opti-
mum coordinate transformations (empirical ortho-
gonal function and canonical correlation analyses).
20

Model Error and Uncertainty, Peter Challenor
(Southampton Oceanography Centre,
Southampton)

Climate research is increasingly dependent upon
com-plex numerical models. While such models
give reasonable estimates of the state of the climate
system as yet we do not have any measure of un-
certainty, such as confidence limits, on such esti-
mates. One approach is to run the models many
times and use the resulting ensemble to evaluate the
uncertainty. However realistic models are expen-
sive to run and it is impossible to create ensembles
that are large enough to allow the uncertainty to be
evaluated properly. In this paper an alternative is
proposed. A Bayesian approach is taken and the
output of the climate model is represented by a
Gaussian process. A limited number of model runs
are needed to characterise the Gaussian process
which is then used to calculate the probability
density function (pdf) of a model output given the
pdf of some model inputs (which could be the for-
cing function or uncertain internal model para-
meters). The technique is illustrated with a simple
climate model (where we can run a large enough
ensemble to validate the method). The extension of
such methods to large state-of-the-art models is
discussed.

The Winter Temperature Downscaling Experi-
ments Using Canonical Correlation Analysis in
China, Youmin Chen (Agrometeorology Institute,
Chinese Academy of Agric Sciences, Beijing,
China)

High quality monthly temperatures at 147 stations
in China and the NCAR/NCEP reanalysis circula-
tion data during winter (the average of Dec. J an.
Feb) of 1951/52-1997/98 are applied in this paper.
With the canonical correlation analysis (CCA) tech-
nique being employed, the downscaling model with
East Asia circulation as predictor and the monthly
temperature as predictand were established. Some
experiments with different PC truncations and
different CCA modes retained were conducted in
order to find the optimum downscaling model. It is
found that as many as possible PC truncation and 4-
6 CCA modes can make the better model skill. The
optimum model was found with 13 PC truncation in
predictors and 8 PC truncation in predictand and
also 5 CCA modes retained. The validation results
with cross-validation method were compared be-
tween the models with and without detrending to
the raw data. There are markedly different model
skills for different sub-area of this country, chan-
ging from 10% to 70% in Brier-based score. The
distribution of model skill score is rather consistent
with topography characteristics, in that the northern
China and eastern China have a relatively higher
skill for its easy impact subject to circulation, while
in the area where the sharp orographic relief can
impede airflow from Siberian high, the model skill
is low. Therefore the model results indirectly reflect
the impact of topography on downscaling skill. The
downscaling model partially capture the tempera-
ture trend change even when the raw data are de-
trended, but a better-simulated trend can be reached
if the raw data trends are retained for developing
models, which can lead to a better downscaling
skill.

On-line Gaussian Processes and their Potential in
Data Assimilation, Dan Cornford, Lehel Csato
and Manfred Opper (Computer Science, Aston
University, Birmingham)

Recently, methods have been proposed that use on-
line algorithms for training Gaussian processes,
based on a Bayesian probabilistic framework. The
online method is an iterative algorithms for upda-
ting the mean function and covariance kernel,
which removes the requirement for large matrix in-
version. When combined with a sparsity heuristic,
the algorithm has been shown to produce small ge-
neralisation errors on standard regression and
classification problems. In this paper we illustrate
the application of this method for assimilating scat-
terometer observations to retrieve wind fields using
an online algorithm. We show how the algorithm
may be adapted to cope with the multi-modal
nature of the distribution of the local wind vector
given the local scatterometer observations. The
local wind vector probability distribution is
modelled using a mixture density network. Since
the method used will track a local Gaussian mode
in the posterior distribution of the wind field given
the scatterometer observations, we show how the
algorithm could be used to sample from this
posterior distribution. We go on to show how
sparsity can be used to select a restricted set of
spatial locations, which capture the variability in
the full wind field model but reduces redundant
computational and storage requirements. We
suggest this may be a useful pre-processing tool for
general data assimilation systems where the
observations have spatially variable redundancy,
which is often the case for remotely sensed
observations. By reducing the number of data
points required to represent the field in question in
a pre-processing step more time is available for a
more complete data assimilation system.

Statistical Downscaling of Largescale GCM Data
to Local Ice Accumulation in Greenland, Traute
Crueger, Hans v. Storch (GKSS Research Center,
Geesthacht, Germany)

Object of the work is the creation and investigation
of local 'Artificial Ice Core' accumulation estimated
from GCM data. Using GCM runs performed with
different CO2 concentrations and solar forcing, the
natural and anthropogenic influence on ice accu-
mulation shall be investigated. For estimation of the
artificial ice accumulation a statistical downscaling
model is developed. In this presentation the statisti-
21
cal technique and an application for one north
Greenland ice core are described.
The statistical technique is based on a multiple line-
ar regression using yearly accumulations as pre-
dictant and principle components (PCs) of seasonal
averaged largescale fields as predictors. In order to
separate temperature from dynamic effects on pre-
cipitation the streamfunction (500hPa) and tem-
perature (500hPa, 700hPa) fields are tested as pre-
dictors.
We vary the boundaries and the months of the
fields, whose mean are used for the EOF expansion.
This way the patterns with the highest correlations
between their PCs and the ice accumulation are
found. The PCs of these patterns are used as pre-
dictors for the multiple linear regression. The para-
meters for which the explained variance is highest
(after cross-validation) are part of the final form of
the model.
The data used to develop the model are the monthly
NCEP-Reanalysis and yearly accumulations of
Greenland ice cores from 1948-1992. The cores
were obtained and analyzed by the Alfred Wegener
Institute for Polar and Marine Research Bremer-
haven, Germany (AWI). For one core from the
northern slope of the Greenland ice sheet the
greatest part (65% explained variance after
validation) of the yearly variability of the ice core
accumulation is described by the third and second
EOF pattern of the 500hPa streamfunction field
averaged from May to August. The area of the
streamfunction field reaches from northeast Ameri-
ca to west Asia. The patterns correspond well to the
synoptic conditions leading to precipitation in
northern Greenland investigated otherwise. The
influence of the temperature on ice accumulation in
northern Greenland seems to be small.
An 'Artificial Ice Core' is estimated by applying the
regression model to a 1000 year coupled atmos-
phere/ocean GCM control run.

Spatialisation of 24 hours precipitation with 100
years return period, Mojca Dolinar
1
, Damijana
Kastelec
2
(
1
Hydrometeorological institute of
Slovenia, Ljubljana, Slovenia;
2
University of
Ljubljana, Biotehnical Faculty, Ljubljana,
Slovenia)

The information about heavy precipitation amount
and the probability of its occurrence is valuable
especially in hydrology and civil engineering. Since
the network of precipitation registration stations is
not dense, it is very difficult to provide re-
presentative data to the customer. In order to be
able to respond on customer requests we wanted to
make a digital map of 24-hour precipitation with
100 years return period with resolution 1 km.
For analysis we used the data from

* 360 stations with daily precipitation measure-
ment (precipitation is measured once per day at 7
a.m.)
* 50 stations with precipitation registration in-
struments (precipitation is measured every five
minutes)

Gumbel method was used for calculating 24-hours
precipitation with 100 years return period for all
stations with registration instruments. As the first
step we analysed the correlation between 24-hours
precipitation with 100 years return period and
extreme daily precipitation. The preliminary spatial
analysis was made to find out the dependence of
extreme precipitation on some geographical varia-
bles. In general, a positive trend was evident on
altitude. The subregions were defined regarding the
residuals (difference to station measurements) of
regression with altitude.
Different methods of spatial interpolation were
tested:

* cokriging - taking the maximum daily pre-
cipitation as covariable
* universal kriging - taking the maximum daily
precipitation as external drift

Cross-validation technique was used to determine
the appropriate variogram model. The kriging pre-
dictions were calculated in the grid with 1 km reso-
lution. We used the altitude data from the digital re-
lief model with the same grid. For the kriging pre-
dictions we considered different influential sur-
roundings and the most efficient was chosen with
cross-validation technique.

Bayesian estimate of halphen type a law's para-
meters, S. El Adlouni
1
, A. Bouziane, D. Ouazar,
A. Zoglat
1
(
1
Faculte des Sciences-Rabat; Ecole
Mohammadia d'Ingenieurs, Rabat, Morroco)

The Halphen distributions family has some sta-
tistical properties that make them appropriate for
modeling some hydrological series. In particular,
the Halphen type A law is a useful tool in extreme
value analysis.
Despite of the large set of distributions used in
hydrology, none is judged globally adequate for
adjusting hydrological series. The 3-parameter dis-
tributions (Halphen type A, B, and B-1) developed
by Halphen (1941) meet some specific require-
ments for fitting these series. However, the litera-
ture about these distributions is very restrictive and
it seems that, except the papers by Halphen (1941),
Halphen (1955) and Morlat (1956), there is no work
that has rigorously explored in detail their mathe-
matical and statistical properties.
In their paper Perreault et al. (1997) present a com-
plete study and pointed out some interesting mathe-
matical and statistical properties of the Halphen
type A distribution. This is a member of the expo-
nential distributions family, so that the maximum
likelihood estimation of its parameters is appro-
priate and easy to proceed. As a measure of esti-
mators accuracy we consider the bias and stan-
dardized error. These two statistics involve the
parameters and their estimators, we reviewuate
22
them using simulations. In this work we give a
method for generating Halphen type A distributed
samples. Using these samples we study some
parameters estimators properties. In the second part
we use a Halphen type A law with a Bayesian
framework, to improve the estimations of the maxi-
mum likelihood, especially for samples with small
size. To compute bayesian estimators we uses Mon-
te Carlo Markov Chains algorithms (MCMC). To
conclude we point out some pertinent remarks.

Stochastic Climate Models

EMAD: An Empirical Model of the atmospheric
Dynamics, Juan Pedro Montavez and Jin Song
von Storch (Meteorological Institute of Hamburg
University)

In this work we present a hybrid model consisting
of an ocean GCM (HOPE-G) and an empirical mo-
del of the atmospheric dynamics (EMAD). The
goal is to extract, in a most complete way, the air-
sea interaction processes represented by a fully
coupled AO-GCM and express the statistical
properties of these processes in terms of a simple
model (EMAD).
EMAD is formulated for the global ocean, i.e. in-
cluding the regions covered by sea ice. The model
is able to generate the heat, fresh water and
momentum fluxes to drive the ocean model, and to
response to the surface conditions given by the
ocean. The formulation of the model consist of two
parts, a deterministic part formulated in a truncated
EOF space and a stochastic part formulated in the
physical space. The model is fitted to 200 years of
daily data from an integration with the coupled
ECHO-G GCM. The skill of the model is better
than a persistence model.
The model is able to produce realistic responses of
the atmosphere to oceanic anomalies related to
ENSO and polynyas. Finally a coupled HOPE/-
EMAD run has been performed. Using cross-corre-
lation functions between SST and surface fluxes, it
is shown that the hybrid model is able to reproduce
the most features of the air-sea interaction pro-
cesses obtained in the fully coupled ECHO-G
GCM.

On the Natural Regularity and Predictability of
Terrestrial and Extra-Terrestrial Parameters,
Nityanand Singh (Indian Institute of Tropical Me-
teorology, Pashan, Pune, India)

Extrapolation of dominant modes of fluctuations
after fitting suitable mathematical function to the
observed long period time series is one of the
approaches to long-term weather or short-term cli-
mate prediction. Experiences suggest that reliable
predictions can be made from such approaches
provided the time series being modelled possesses
considerable regularity. Choice of the suitable
function is also an important task of the time series
modelling-extrapolation-prediction (or TS-MEP)
process. Perhaps equally important component of
this method is the development of effective filtering
module. The filtering mechanism should be such
that it effectively suppresses the high frequency, or
unpredictable, variations and carves out the low
frequency mode, or predictable, variation of the
given series. By incorporating a possible solution to
those propositions a new TS-MEP method has been
developed in this paper. A Variable Harmonic
Analysis (VHA) has been developed to decompose
the time series into sine and cosine waveforms for
any desired wave resolution within the data length
(or fundamental period). In the Classical Harmonic
Analysis (CHA) the wavelength is strictly an
integer multiple of the fundamental period. For
smoothing the singular spectrum analysis (SSA)
has been applied. The SSA provides the mechanism
to decompose the series into certain number of
principal components (PCs) then recombine the
first few PCs, representing the dominant modes of
variation, to get the smoothed version of the actual
series.
Twenty-four time series of terrestrial and extra-
terrestrial parameters, which visibly show strong
regularity, are considered in the study. They can be
broadly grouped into five categories: (i) inter-
annual series of number of storms/depressions over
the Indian region, seasonal and annual mean
northern hemisphere land-area surface air tempera-
ture and the annual mean sunspot number (chosen
cases of long term/short trends or oscillation); (ii)
monthly sequence of zonal wind at 50-mb, 30-mb
and 10-mb levels over Balboa (representative of
quasi-biennial oscillation); (iii) monthly sequence
of surface air temperature (SAT) over the Indian
region (strongly dominated by seasonality); (iv)
monthly sequence of sea surface temperature (SST)
of tropical Indian and Pacific Oceans (aperiodic
oscillation related to El Ni0/La Nia); (v) se-
quence of monthly sea level pressure (SLP) of se-
lected places over ENSO region (seasonality and
aperiodic oscillation). Best predictions are obtained
for the SLP followed by SAT and SST due to
strong seasonality and/or aperiodic oscillations. The
predictions are found satisfactory for the lower stra-
tospheric zonal wind over Balboa, which displays
quasi-periodic oscillations. Because of a steep
declining trend a reliable prediction of number of
storms/depressions over India is possible by the use
of the method. Prediction of northern hemisphere
surface air temperature anomaly is not found satis-
factory.

Stochastic Tropical Atmospheric Boundary-Layer
Dynamics: Convective Downdrafts and Recovery,
Discharge-Recharge Processes, and 1/f-Noise,
Jun-Ichi Yano and Richard Blender (Meteoro-
lo-gisches Institut, Universitaet Hamburg, Ham-
burg, Germany)

It was recently shown by the authors (manuscript
submitted to J . C.) by analyzing TOGA-COARE
sonde data that the tropical convective variability
23
measured as CAPE (convective available potential
energy) varies as 1/f-noise for the time scale of 1-
100 days. Variability of the surface moisture and, to
a lesser extent, temperature are responsible for this
1/f-noise behavior of CAPE. It is speculated that
stochasticity of convective downdrafts, which sud-
denly dry and cool the surface boundary-layer, and
the following sudden recovery is the main physical
mechanism. This mechanism may be further idea-
lized as a discharge-recharge process, which can be,
furthermore, regarded as a type of self-criticality. In
order to generate a 1/f-noise from a random series
of such pulse-like events, these time intervals must
be close to Gaussian distribution (rather than Pois-
son) with the mean interval corresponding to the
minimum period for the 1/f-noise. The present
paper reports on the statistical analysis of the
TOGA-COARE research vessel surface obser-
vations, which are frequent enough (1-15 min) to
resolve the individual downdraft events.

Assimilation

Variational Assimilation of botanical proxy Data
for the Reconstruction of Palaeoclimate,
Christoph Gebhardt (Meteorologisches Institut
der Universitt Bonn, Germany); Norbert Khl
(Institut fr Palontologie, Universitt Bonn,
Germany); Andreas Hense (Meteorologisches
Institut der Universitt Bonn, Germany); Thomas
Litt (Institut fr Palontologie, Universitt Bonn,
Germany)

The terrestrial palaeorecords of botanical fossils
(e.g. pollen, macro fossils) serve as proxy data
which provide important information that can be
used for the reconstruction of near-surface palae-
oclimate. However, a spatially coherent recon-
struction is often hampered by a relatively sparse
number of fossil sites and by factors influencing
each fossil record on a regional scale. In this case, a
reconstruction of climatological fields requires ad-
ditional information like a background field or dy-
namical constraints. We present a simple two-di-
mensional variational assimilation of the proxy data
resulting in a reconstructed climate which is con-
sistent with the observed palaeovegetation as well
as with large-scale features of the atmospheric dy-
namics.
In a first step, transfer functions quantifying the re-
lation between climate and the botanical proxy data
are found by combining a gridded climatology of
30-year (1961--1990) monthly means with distri-
bution maps of the recent vegetation. This results in
Bayesian transfer functions describing conditional
probabilities for the existence of the important spe-
cies given certain climatic states. Thus a most pro-
bable geographical distribution of each species can
be related to a given climate. The deviation of this
conditional vegetation from the actually observed
palaeovegetation is described in a cost function
which is minimized in dependence of the climatic
state by a variational approach. The implementation
of the conditional probabilities leads to a non-qua-
dratic cost function interpretable as a maximum-li-
kelihood function. A simplified dynamical model
and a first-guess field of the palaeoclimate are in-
plemented as weak constraints. The reconstructed
climate is found at the minimum of the cost func-
tion.
The method is tested on European palaeobotanical
data to reconstruct temperatures of two time slices
of the Eemian interglacial ( 125-115 kyears BP)
which are characterized by relatively stable climatic
and vegetational conditions similar to the present
time.

Nikolaus Groll, Martin Widmann (Institute of Hy-
drophysics, GKSS Research Centre, Geesthacht,
Germany)

Our goal is to understand how useful GCMs are for
estimating regional precipitation variability from
changes in the synoptic-scale flow.
The first step is to assess how well the NCEP re-
analysis data captures precipitation variability. The
rationale behind this is that the NCEP reanalysis
can be thought of as a proxy for an ideal general
circulation model (GCM) which nearly perfectly
represents the synoptic-scale flow, but in which
areas of complex topography are not sufficiently
resolved. Reanalysis precipitation is entirely cal-
culated according to the model physics and para-
meterisations, and is not adjusted to observations.
Precipitation fields from the NCEP reanalysis have
therefore been evaluated using high-resolution grid-
box datasets for precipitation in Oregon and Wa-
shington, as well as over the Alps. The reanalysis
captures precipitation varability on time scales of
months to years accurately. Nevertheless there are
over- and underestimations of the seasonal means,
which affect not only spatial scales up to a few
model gridcells over Oregon and Washington, but
also the total area mean over the Alpine region.
These systematic biases can be mainly attributed to
the poor representation of the topography in the
NCEP model.
To determine how well precipitation variability is
represented in GCMs that are used for climate
change studies, the ECHAM4 model has been
investigated. In contrast to the NCEP Reanalysis
the sequence of atmospheric states in the ECHAM4
control run is not directly related to the observed
states. In order to allow for a comparison of
simulated and observed precipitation, we developed
an analog-based resampling method (ABR). This
method selects for each observed day the most
similar day from the control run (with respect to the
SLP field over a specified region). The observed
precipitation can then be compared with the simu-
lated precipitation of the analog day. ABR is cur-
rently applied to SLP fields over the North Atlantic
Ocean and Europe in order to compare ECHAM4
24
simulated precipitation with observed precipitation
over the Alps.
Our results show that simulated precipitation might
be a good predictor for regional precipitation down-
scaling.

Homogenisation and Interpolation of incomplete
climatological Data with simple physically based
Models, Andreas Hense, Christoph Gebhardt
(Meteorologisches Institut, Universitt
Bonn,Bonn, Germany)

A characteristic of climatological data sets like the
Comprehensive Ocean Atmosphere Data set
COADS is the incompleteness in space and time.
Furthermore, even data sets like the Climate
Research Unit (CRU) temperature data set may
suffer from insufficient sampling on the sub-month-
ly time scale leading to unrepresentative data with
respect to neighbouring grid points.
We will present a simple but physically based inter-
polation method ("data model") to interpolated and
homogenize existing climatological data sets up to
an a-priori selected spatial scale. For the COADS
1950-1994 in the Atlantic we analyse wind (vector
and velocity) and sea level pressure using an opti-
mal averaging procedure and an Ekman wind mo-
del as a dynamic constraint. The data model is
formulated through a variational approach. The first
result is that the artificial trends in the wind data set
can be reduced taking into account the pressure
information but unfortunately as a second result it
turns out that the simple Ekman model is insuf-
ficient even on the monthly time scale due to the
omission of curvature effects. The interpolation and
homogenization of the CRU temperature data is
again done in a variational framework. A two di-
mensional advection diffusion model defines modes
which are used as a projection basis. The ampli-
tudes are estimated from the available data under
the constraint that the resulting tendencies of the
advection-diffusion model are minimal in the least
square sense. This data model is such efficient that
Monte Carlo methods like cross validation can
provide a quality control by detecting unrepre-
sentative data and estimating an a-posteriori error
measure of the derived complete fields.
Results will be presented for the period 1881-1998
and the area 2.5S to 77.5N.

Trends of Maximum and Minimum Temperatures
in Northern South America, Ramon A. Quintana-
Gomez (Department of Earth Sciences,
Universidad de los Llanos Ezequiel Zamora,
Barinas, Estado Barinas, Venezuela)

Investigation of Spatial and Temporal Ozone
Immission Structures under Stochastic-Deter-
ministic Model Assumptions, Heike Hoffmann
(Graduate School of Spatial Statistics, Freiberg,
University of Mining and Technology, Germany)

Spatial and temporal properties of ozone immission
structures are studied under stochastic-deterministic
model assumptions. The given data are the half-
hour averages of pollutants and meteorological pa-
rameters at 30 monitoring stations from 1995 to
1999. The sites are irregularly distributed over Sa-
xony, concentrated in the strongly loaded areas
(with respect to primary pollutants), in the cities
and along the moutain ranges along the Czech-
German boundary (Ore Mountains). The target of
the work is the production of maps of the ozone
load for half-hour intervals.
In the given case spatial interpolation is rather diffi-
cult because of the heterogenity of the monitoring
network. This concerns the spatial distribution of
the stations as well as their representativeness,
which differs for city stations, outskirts stations and
land stations. Originally the monitoring network
was constructed to observe the maximum concen-
trations of the primary pollutants like nitrogen
oxides or sulphur dioxide; therefor stations are mis-
sing at positions where extreme point concentra-
tions of ozone are expected (in the lee of the big
cities). Big problems also result from the fact that
some stations close together (e.g. three stations in
the city of Leipzig) are located in regions of dif-
ferent air-chemical regimes, what leads to a special
kind of trend in the data.
Despite these difficult circumstances, the Saxonian
ozone data could be nevertheless spatio-temporally
trend-corrected by methods such as meanpolishing
following Cressie so that finally geostatistical pro-
cedures for stationary processes could be used. In
particular, a reasonable variogram could be esti-
mated showing sufficient correlations between the
ozone values at neighbouring stations. The long
range of the variogram of more than one hundred
kilometers is in accordance with what is known
about the spatial process of ozone distribution.
These spatial investigations were combined with
temporal analyses. In particular, the temporal inves-
tigations were carried out for the whole data set as
well as for single stations respectively seasons,
leading to interesting relationships which are useful
for the prediction of the ozone distribution.

Reduced-order schemes for data assimilation in
oceanographiy based on the Kalman filter, I.
Hoteit and D.T. Pham (Laboratoire de Modeli-
sation et Calcul, Grenoble, France)

Brute-force implementation of the extended Kal-
man (EK) filter in realistic ocean models is not
possible because of its prohibitive cost. To remedy
to this problem, different degraded forms of the EK
filter, which basically reduce the dimension of the
system through some kind of projection onto a low
dimensional subspace, have been proposed. The
main principle of our contribution is to study the
usefulness of the evolution in time of the reduced-
order state space. We will therefore compare the
theoretical and practical aspects of the singular
evolutive extended Kalman (SEEK) filter intro-
duced by Pham et al. and the reduced-order
25
extended Kalman (ROEK) filter introduced by
Cane et al.. Indeed, the reduced state space of the
SEEK filter is let to evolve in time according to the
model dynamic while the evolution of the reduced
space of the ROEK filter is replaced, and thus re-
mains fixed in time, by the evolution of the error in
the reduced dynamic defined by the reduced space.
On the other hand, the SEEK and the ROEK filters
remain expensive in real operational oceanography.
Hoteit et al. have proposed different degraded
forms of the SEEK filter to reduce its cost. Their
approach consists in simplifying the evolution of
the reduced space, which is the most expensive part
of the SEEK filter. To reduce the cost of the ROEK
filter, we introduce the use of an autoregressive
statistical model to let the error evolve in the
reduced space, which is the most expensive part of
this filter. Finally, using a twin experiments ap-
proach, these filters have been implemented and
tested by assimilating altimetric data in a realistic
setting of the OPA model in the tropical Pacific
ocean.

Cloudiness and sea surface temperature: tackling
the problem of missing values, Richenda
Houseago-Stokes and Peter Challenor
(Southampton Oceanography, Centre
Southampton, UK)

Sea surface temperature from infra red radiometers
on satellites is an important climate variable. Un-
fortunately, because of cloud, the data set contains
many missing values. This makes it difficult to ap-
ply Principal Component Analysis to look at the
dominant modes of variability. We investigate two
solutions to this problem. The first is to interpolate
the missing data using spatial optimal interpolation
on the non-missing data and apply standard me-
thods to evaluate the empirical orthogonal functions
(EOF's). The other method is to use the recently de-
veloped method of Probabilistic Principal Compo-
nent Analysis and apply the EM algorithm. This
algorithm enables us to estimate the EOF's and the
missing values simultaneously. To test the two me-
thods we compare them on data where we have si-
mulated missing values by deleting points at
random and thus know the true results. We then ap-
ply them to fifteen years of sea surface temperature
data in the North Atlantic from the Advanced Very
High Resolution Radiometer (AVHRR) on US sa-
tellites.

Removing errors in simulated distributions of cli-
mate elements using a non-parametric adjustment
procedure, Radan Huth
1
, Jan Kysel
1,2
(
1
Institute
of Atmospheric Physics, Prague, Czech Republic;
2
Dept. of Meteorology and Environment
Protection, Charles University, Prague, Czech
Republic)

We propose a method that allows one to adjust a
simulated (GCM-simulated or downscaled) distri-
bution of any climate element to its observed cou-
nterpart. The method is non-parametric and consists
in fitting equally-spaced percentiles in the observed
and simulating distributions and linearly inter-
polating between them. The only assumption it
makes is that percentiles of a distribution (not
values themselves) are simulated correctly.
The method is applied to daily temperature and
precipitation at several sites in central Europe, si-
mulated by two GCMs and a downscaling proce-
dure. The statistical moments (up to the fourth) of
the simulated temperature distributions after the
adjustment are very close to those observed, even if
the simulated distributions themselves were seve-
rely distorted relative to observations. On the other
hand, the temporal and spatial structure of the
simulated temperatures (e.g., lag-1 autocorrelations,
day-to-day variability, spatial autocorrelations, fre-
quency and duration of extreme periods) are not
affected by the adjustment. For precipitation, the
numbers of dry days are corrected after the ad-
justment and the probabilities of a wet-to-wet and
dry-to-wet transitions are considerably improved.
A modification of the method is applied to 2xCO2
GCM simulations. It assumes that GCMs correctly
simulate the temperature difference between the
future and present climate for the equally spaced
percentiles. We show that the method successfully
removes GCMs errors from the temperature series.
In validation studies of GCM-simulated or down-
scaled climate variables for the purposes of impact
assessments, the main emphasis is put on the first
two statistical moments of their distributions. We
argue that a detailed validation of them is of se-
condary importance as one can easily adjust their
distributions to fit the observations. Instead, the va-
lidation studies should put much more emphasis on
temporal and spatial structure of the variables, for
which such an adjusting method does not exist.

Simplified EOFs - three alternatives to rotation,
Ian Iolliffe (University of Aberdeen, UK);
Mudassir Uddin (University of Karachi,
Pakistan); Karen Vines (Open University, UK)

Principal component analysis (PCA) is widely used
in atmospheric science, and the resulting empirical
orthogonal functions (EOFs) are often rotated to aid
interpretation. We describe three methods that
provide alternatives to the standard two-stage pro-
cedure of PCA followed by rotation. The techni-
ques are illustrated on a simple example involving
Mediterranean sea surface temperature.
The first method compresses the usual two stages
into one, by successively maximizing a criterion
that combines variance and simplicity. The second
technique is based on an idea from regression, the
LASSO (Least Absolute Shrinkage and Selection
Operator). Our technique adds a constraint on the
sum of absolute values of elements in an EOF to
the usual PCA optimisation problem, which has the
effect of driving some elements to zero. Our third
approach seeks high variance EOFs such that the
elements of each EOF are proportional to (hope-
fully low-valued) integers. The approach achieves
26
this by applying a series of variance maximising
but "simplicity-preserving" transformations to an
initial set of particularly simple, but not necessarily
high variance, initial EOFs.
Each technique is shown to provide different sim-
plified interpretations for the major sources of
variation in a data set. Comparisons will be made
between the three methods and with standard PCA
and its rotated version. We conclude that all three
techniques have advantages compared to the usual
two-stage rotation.

Reconstruction of large-scale climate, J.M. Jones
(GKSS Research Centre, Geesthacht, Germany)

It is aimed to reconstruct those large-scale patterns
of extratropical circulation which are strongly lin-
ked to variability in proxy data. Current work is
centered on the Antarctic circumpolar circulation.
Using canonical correlation analysis we found pro-
mising relationships between tree ring chronologies
from Argentina, Chile and New Zealand, and large-
scale climate patterns representing the Antarctic
circumpolar circulation for sea level pressure and
850hPa geopotential height, taken from the NCEP
reanalysis. This work shall also be extended to
Northern Hemisphere circulation modes. The suita-
bility of other proxy records, such as ice core and
coral data, for incorporation into these reconstruc-
tions is also being investigated. Preliminary ana-
lysis of Greenland ice core accumulation shows
links with large-scale Northern Hemisphere atmos-
pheric circulation. These reconstructions are the
first stage of DATUN (Data Assimilation through
Upscaling and Nudging). The aim of the DATUN
method is to obtain a physically based, best guess
for the large-scale states of the atmosphere during
the Late Holocene with annual temporal resolution.
The second stage of this methodology is nudging
the large-scale variability of a GCM towards the
results from the upscaling.

On a probabilistic interpretation of variational
data assimilation and model error estimation,
Gennady A. Kivman (Alfred-Wegener-Institut for
Polar- and Marine Research, Bremerhaven, Ger-
many)

Estimating the model error statistics is a funda-
mental problem of variational data assimilation
(VAR). It cannot be solved without understanding
of the VAR in probabilistic terms. A serious de-
terrent to the use of the probability theory is that
probabilities for the data and for the model va-
riables are conceptually diverse quantities. In envi-
ronmental studies, the former may be treated as
relative frequencies while the latter generally ex-
press degrees of believe. Thus, it is difficult to give
an interpretation to probabilities on the joint data-
model state space and to view the VAR as an
application of Bayes' theorem.
Restoration of the probability interpretation of the
VAR may be achieved by transferring the infor-
mation provided by the data onto the model state
space as another degree of believe. Then, having
two probability distributions over the same space,
we may apply the probability theory to the pro-
babilities themselves. This leads to a natural cri-
terion of compatibility of the data with the model
solution for given data and model uncertainties.
Contrary to the chi2 criterion, it is applicable to any
probability distributions.
As the probabilities on the model state space are
viewed as degrees of belief, the principle of maxi-
mum entropy should be employed to estimate
statistical parameters of the model errors since it
yields the least unbiased solution for assigning pro-
babilities. It turns out that if the compatibility cri-
terion is fulfilled, the prediction made with the
VAR may be expressed as the mean with respect to
an operator valued probability measure of what is
told by the data and by the model. Maximization of
a generalization of Shannon's entropy for such
probabilities is proposed to use for estimating the
model error statistics. The performance of this
approach is compared with other techniques widely
used in the VAR such as the maximum likelihood
and the generalized cross-validation.

Statistical presentation of wind wave climate,
Leonid Lopatoukhin
1,2
, Valentine Rozhkov
1
,
Alexander Boukhanovsky
1,3
, (
1
State
Oceanographic Institute;
2
St. Petersburg State
University (Dep. Oceanology);
3
Institute of High
Performance Computing and Data Base, Russia)

1. Existence of 40 years (1957-1996) of reanalysis
of meteorological data and verified spectral wave
models allows to calculate the ensemble of spatial-
temporal wind and wave fields in a grid points.
2. Vector wind field V(r,t) is specified by mean
mV(r,t), variance DV(r,t), covariance KV(r,?,t,?),
where r is vector of spatial coordinates, ? vector
of displacement between the grids, t- time, ? -lag.
Invariant of tensors DV(?), KV(?) are used for
cartographic presentation. The annual variations of
wind speed are represented as the expansions of
mV (t), DV(t), KV(t,?) by the trigonometric basis in
accordance with the theory of periodically corre-
lated random processes.
3. A set of 9 values (h,?,?)i , is used for parame-
terization of complicated fields of wind wave and
swell spectra S(?,?,r,t). Here i=1, 2, 3 indexes for
wave height h, periods ? and directions ? for wind
waves (i=1) and swells (i=2, 3), calculated by mean
of the moments mjk(j,k =0, 1, 2) of truncated
spectra S(?,?) and peak frequency. Such a change
of S(?,?) to a set of 9 random values ? simplify the
treatment the results of calculations. Besides,
applying of both the theory of non-random function
S(?,?) =S(?,?,?) of random arguments ? and met-
hod of statistical linearization propose the approach
for classification of climatic wave spectra.
4. Annual maxima of wave heights and wind speed
are recorded in different months and from different
directions. The ensemble of values (h, ?) ? (|V|,?) is
27
a mixture of distribution and Monte-Carlo simula-
tion of random values and function is used.
5. Synoptic variability of waves is connected with a
sequence of storms and weather windows and it pa-
rametric representation (in accordance with a
theory of random outlier) may be given in a system
? of four random values {h+,h- ,?,?}. Where h+,h-
the value of outlier (up and down), ?,? - duration
of outlier. Method of estimations of wind and wave
extreme with different return periods is proposed
and based on the results of analysis of distributions
?.

Model-based spatial modelling of climatological
data, J. Mateu
1
, P. Juan
1
, C. Antolin
2,3
, C. Ano
3
(
1
Department of Mathematics, Universitat Jaume
I, E-12071 Castellon, Spain;
2
Department of
Vegetal Biology, Universitat de Valencia,
Burjassot, Spain;
3
CIDE-Centre for
Desertification, Cami de la Marjal s/n., Albal
(Valencia), Spain)

The problem of estimation and prediction of a spa-
tial stochastic process, observed at irregular loca-
tions in space, is considered. The statistical model
proposed is a mixture of two components, one in-
volving a non-parametric term, accounting for a de-
terministic trend given by exogenous variables, and
a parametric component defining the purely spatial
random variation. Several well known statistical
models such as the full interactive model, ridge-re-
gression or neural-network regression are particular
cases of the proposed modelization. Field variables
usually show spatial dependence among observa-
tions which is an important drawback to traditional
statistical methods. However, geostatistical tech-
niques analyze and describe the spatial dependence
and quantify the scale and intensity of the spatial
variation providing the essential spatial information
for local estimation. Geostatistics is principally the
application of regionalized variable theory.
The estimation process is developed in a two-step
procedure, first addressing the estimation of the
non-parametric component, and then the estimation
of the parametric term from the residual stochastic
process. The proposed methodology is applied to
model the spatial process of rain intensity (rainfall
erosivity) in a mediterranean region. Some other re-
lated processes like elevation or distance to the sea
are also considered. These variables are considered
to be highly important in the mechanisms and pro-
cesses of erosion by water in Mediterranean en-
vironments.

Extreme Value Analysis

Regional GEV-growth curves of yearly preci-
pitatation depths in Belgium for durations ran-
ging from 10 minutes to 30 days, Daniel Gellens,
Gaston Demare (Royal Meteorologuical Institute
of Belgium, Brussels, Belgium)

A regional L-moments approach has been used to
study the yearly extreme k-days amounts in Bel-
gium. Two different datasets originating from two
networks are available for this purpose: the dense
climatological network yielding yearly extremes of
k-days precipitation amounts (with k= 1, ..., 30
days); and the sparser hydrometeorological network
equipped with recording raingauges with a time
step of 10 minutes. The latter dataset provided the
yearly extremes for durations ranging from 10 mi-
nutes through 7 days.
The regional homogeneity of the distribution being
verified, the growth curves of the regional GEV-
distributions have been established. This allows for
the comparison of the extreme value distributions
based upon yearly extremes from the two datasets
(including an "experimental" assessment of the
Herschfield factor H) and the sharp determination
of the curve parameter of the GEV-distribution for
a large range of durations.

Extreme precipitation events in southeastern
South America and the relationships with sum-
mertime large-scale South American Stationary
Systems (SASSY), Leila M. V. Carvalho
1,2
,
Charles Jones
1
, Brant Liebmann
3
(
1
Institute for
Computational Earth System Sciences (ICESS),
University of California, Santa Barbara,
California, USA;
2
Dept. of Atmospheric Sciences,
University of Sao Paulo, Sao Paulo, Brazil;
3
NOAA-CIRES Climate Diagnostics Center,
Boulder, Colorado,USA)

The South Atlantic Convergence Zone (SACZ) is
characterized by a wide range of temporal varia-
bility, from synoptic to intraseasonal and inter-
annual scales. Our recent study has objectively
identified the occurrence of South American Sta-
tionary Systems (SASSY) during austral summer.
SASSY events are characterized by persistence of
two or more days and large geographic extension
over southeast Brazil. This study investigates the
occurrence of extreme precipitation events in south-
east South America and the relationships with the
SASSY. Daily Outgoing Longwave Radiation
(OLR) and precipitation data derived from surface
stations from the states of the South and Southeast
Brazil (1979 to 1999) are used to determine the
frequency of occurrence of extreme precipitation
events. Whenever the daily precipitation for a cer-
tain period exceeds a given percent of the
climatological mean for that period, it is considered
as an event. This presentation will focus on the
spatial and temporal variability of the SASSY in
association with extreme precipitation events.

Diagnosis of variations in the climate-related
flood risk - Some results from the Interior Wes-
tern United States, Shaleen Jain, Martin
Hoerling, Gary Bates, Randall Dole
(NOAA/CIRES Climate Diagnostics Center,
Boulder, CO)

28
In the interior western United States, spring floods
(i.e., peak streamflow) reflect an interesting integra-
tion of the seasonal-to-interannual climate varia-
tions (through winter snowpack development) and
Springtime weather patterns (responsible for the
snowmelt). Consequently, trends and year-to-year
variations in the flood magnitude and timing may
be cast in a climate-related risk assessment frame-
work. We present a detailed empirical-statistical
assessment of this climate-flood problem, with
special emphasis on the sensitivity of flood peaks
to: (a) large-scale climate signal during winter and
Spring, and their potential predictability, and (b)
regional hydrologic features (e.g., elevation and
drainage area of river basins).
The observed flood probability distribution is exa-
mined for preferred tail behavior, consequent to the
known modes of seasonal-to-interannual climate
variability. Some potential application areas for
such studies (including water resources manage-
ment, stream ecology, hydraulic design, and emer-
gency response) are discussed.

The Climatological Probability of Drought in
Northern Province of South Africa, Tiba Kabanda
(School of Environmental Sciences, University of
Venda, Thohoyadou, South Africa)

This paper presents a study of the climatology of
drought in the Northern Region of South Africa.
Drought is described as a rare and severe event,
therefore the frequency of occurrence and intensity
need to be explained by using a criterion that will
distinguish normal dry spells from real drought.
The paper is intending to describe such methods.
The study examines the geographical variability of
dry spells in the area of study and analyses drought
(seasonal) in detail to reveal its frequency, cover-
age, and persistence. Rainfall stations in the region
are the point-source data samples that represent the
study region. All stations were intercorrelated to
produce blocks of highly correlating sites. The site
characteristics depict different physical proportions
such as geographical location and elevation. Rain-
fall data for 40 years is used in this study.
The study employs standardised precipitation Index
(SPI) in the analysis because it allows the deter-
mination of the rarity of an episode in terms of pro-
bability. The results obtained from the study reveal
spatial-temporal distribution of drought within the
study area.

Statistical Analysis of Point Discharge Current at
Different Time Scale Over the Pune Region, S.S.
Kandalgaonkar, M.I.R. Tinmaker, M.K. Kulkarni
(Indian Institute of Tropical Meteorology, Pune,
India)

Thunderstorm is an important weather phenomenon
for the understanding of the many issues related to
the atmospheric electricity. As the storm forms and
dissipates several complex cloud physical and elec-
trical processes such as charge separation, preci-
pitation formation, occurrence of lightning and
point discharge current are taking place. The non-
linear conduction current flowing from the atmos-
phere to the surface of the earth when the electric
field is sufficiently strong is called Point Discharge
Current (PDC). In the present paper minute interval
data of Point Discharge Current (PDC) collected for
a total of 65 thunderstorm days occurred during 6
years period (1972-1977) at Pune (18 32'N, 73
51'E, 559 m asl) has been subjected to statistical
analysis namely frequency distribution at different
time scale (distribution during the total diurnal
period). The total diurnal period (0000 - 2359 IST)
is divided into 3 distinct time intervals i.e. after-
noon (1400-2100 IST); night time (2100-0759 IST)
and day time (0800-1400 IST). These three time
intervals are practically consistent with the require-
ment of separating the diurnal period into convec-
tive and non-convective atmospheric regimes.
Gaussian Model (computer base), is applied and the
results suggested that the distribution is normal and
the chi-square value obtained (5.2008) is less than
the standard table value (9.542) for 22 degrees of
freedom. From this result it is seen that the hypo-
thesis applied to the present data set holds good.
Since it is a normal distribution the data has been
subjected to Fischer's test for the skewness and
kurtosis. The results thus obtained revealed that the
distribution is found to be normal with negative
skewness and exhibits the property of leptokurtic
distribution. The sustained occurrence of negative
and positive PDC at the end of the storm day in
second and third category of time interval may be
due to the presence of space charges of either po-
larity.

A Dedacal Study of Premonsoon Seasons Thun-
derstorms Over Pune Region by Using Markov
Chain Model, S.S. Kandalgaonkar, M.K.
Kulkarni, M.I.R. Tinmaker (Indian Institute of
Tropical Meteorology, Pune, India)

Probabilistic description of the thunderstorm phe-
nomena in the premomonsoon season over Pune
(18 32'N, 73 51'E, 559 m asl) region has been
studied with the help of Markov Chain Models. The
daily thunderstorm data of pre-monsoon season for
a period of 11 years i.e. from (1970-1980) have
been subjected to the Markov Chain models of first,
second and third order. Use of Akaike's Information
Criterion (AIC) clearly indicated that the first order
Markov Chain Model is the best which described
the atmospheric process of thunderstorm pheno-
mena over the Pune region. This is based on the
decision procedure depending upon the principle of
maximum likelihood. The steady state probabilities
and mean recurrence time of dry and wet days
have also been calculated for first and second order
models for each separately and also for the total of
11-years. These computations revealed that the ob-
served and theoretical values of steady state
probabilities realistically consistent.

29
Trends of Atlantic wave extremes as simulated in
a 40-year wave hindcast using kinematically
reanalyzed wind fields, Xiaolan Wang* and Val R.
Swail (Network Strategies Division, AMWSD
Meteorological Service of Canada, Downsview
Ontario, Canada)

In this study, both seasonal and monthly extremes
of wave height in the North Atlantic are analyzed.
The analysis is based on a 40-year (1948-1997) nu-
merical hindcast using an intensive kinematic
reanalysis of wind fields. Changes in the ocean
wave extremes are identified by performing the
Mann-Kendall test, and are further related to
changes in the atmospheric circulation (sea level
pressure) by means of Redundancy Analysis. The
month-to-month variability of trends is also re-
vealed.
The results show that the trend patterns of wave ex-
tremes vary greatly from month to month. Gene-
rally, changes are most extensive in J anuary and
March for the northern North Atlantic, and in J uly
and October for the central and/or subtropical North
Atlantic. In J anuary and March, significantincreases
of wave extremes in the northeast North Atlantic
are closely related to changes in the North Atlantic
Oscillation (i.e., the deepening of the Icelandic low
and the strengthening of the Azores high) during
the last four decades. In J une, J uly and November,
the trend pattern features significant increases of
wave extremes in the ocean off the North America
coast; and it features a large area of significant in-
creases in the subtropical North Atlantic in
October. Changes in the other months are of little
field significance. Generally, increases of ocean
wave height are associated with an anomaly low
pressure over the region.
The results also indicate that the ocean off the
North America coast has roughened in August, No-
vember and December during the recent decades,
experiencing higher wave extremes.
Since the kinematic reanalysis of winds correctly
re-assimilates in situ surface wind data, intensifys
wind fields in extratropical storms as necessary,
and includes tropical cyclone boundary layer winds,
the resulting wave hindcast shows greater rates of
changes than those identified from a wave hindcast
using the NCEP reanalyzed winds for the same pe-
riod.
Especially, more significant increases are detected
for the ocean off the Canadian coast in summer,
and for the central North Atlantic in fall.

Hurricane Landfall Winds Along the United
States Gulf and Atlantic Coasts: Monte Carlo
Simulations, Gary D. Skwira*, Douglas A. Smith+
and Richard E. Peterson* (Wind Science and
Engineering Program, Texas Tech University,
Lubbock, TX, USA; *Department of Geosciences,
+Department of Civil Engineering)

While hurricane landfalls along the United States
Gulf and Atlantic Coasts have had a greater impact
in recent years, they remain a relatively infrequent
phenomenon. At a particular location the record of
past occurrence is too sparse to provide meaningful
statistics to use in planning for future landfalls.
In order to exploit the historical database more
extensively, Monte Carlo simulations have been
carried out. Using tropical cyclone characteristics
compiled by Ho et al. (1987), probability density
functions were assembled for storm speed and di-
rection, central pressure and radius of maximum
wind. After random selection of parameters from
the probability density functions, a tropical cyclone
wind field model (Holland et al. 1991) was ini-
tialized to obtain the horizontal winds.
Simulated propagation of the model cyclone then
allowed the time history of wind speed and direc-
tion at landfall to be tabulated at specific locations.
Repeated random sampling allowed a 1000-storm
climatology to be generated.
The simulation results are portrayed for 20 sites
along the United States Atlantic and Gulf coasts.
The contrasting simulated climatologies are dis-
cussed in terms of the meteorological and physical
settings as well as the impacts for planning and
design.

Simulation of ocean wave climate changes
through wind variance changes on different
timescales, E. Bauer
1
,R. Weisse
2
(
1
PIK, Potsdam;
2
GKSS, Geesthacht, Germany)

The observed mean ocean wave height in the North
Atlantic has grown significantly over the past 30-40
years. Since no corresponding growth of the mean
wind forcing was observed, the effect of wind
variance changes on two different timescales was
tested. Test cases were conducted with the wave
model WAM using ECMWF-Reanalysis wind
fields. The varying balance of the wave generation
and wave dissipation processes suggested to esti-
mate the effect of storm frequency changes on daily
timescales and of high-frequency wind variance
changes on hourly timescales. First, the storm fre-
quency changes were simulated by modifying the
time intervals between succeeding storm events.
Second, the high-frequency wind variance was
simulated by a quasi-realistic wind generator deve-
loped from measurements with 20-minutes time
resolution. The two simulation studies revealed a
mean wave height growth with decreasing storm
frequency on the one hand, and with increasing
high-frequency wind forcing on the other hand. To
explain the observed wave height growth in the
North Atlantic support is provided by independent
observations of growing duration of westerly
weather patterns in connection with the intensifying
North Atlantic Oscillation (NAO) during that
period. So, the first simulation helps to identify
mechanisms inducing wave climate changes where-
as the second simulation yields an estimate of
modelling uncertainty.

Frequency and Intensity of Mediterranean Cyclo-
nes in a doubled CO2 scenario, P.Lionello
30
(University of Lecce), F.Dalan (University of
Padua)

The analysis is based on two 30-year long global
simulations carried out with the ECHAM 4 model
at the DMI (Danish Meteorological Institute): a
CTR (control) experiment, which reproduces the
present climate, and a CO2 experiment which
simulates the effect of a doubled atmospheric CO2
content. Both experiment were carried out at T106
resolution.
An assessment of the variations in the cyclonic ac-
tivity is, obviously, important. The atmospheric cy-
clones are a potential source of major damages in
the Mediterranean area, because of their association
with storm surges, high wind waves, intense preci-
pitation and floods.
The method is based on the search of the pressure
minima in the T106 slp (sea level pressure) fields,
available at time intervals of 6 hours. The search is
based on the partitioning of the slp fields in depres-
sions by the identification of sets of steepest
descent paths leading to the same slp minimum.
To each depression has been associated a trajectory,
obtained by following the location of its slp mini-
mum in time. The procedure results in a list of
cyclones, their trajectories, initial and final points,
and the sequences of their pressure minima, of the
areas covered, and of the maximum geostrophic
wind at the sea surface.
The analysis shows that: 1) the total number of cy-
clones in the CTR experiment is in acceptable
agreement with the ERA-15 data and it can be
considered an acceptable reproduction of the pre-
sent climate. 2) There is a statistically significant
decrease in the total number of cyclones in the CO2
experiment. This suggests that the CO2 doubling
would result in a milder climate in the Mediter-
ranean Area.
The interannual variability of the cyclonic activity
remains comparable to the difference between the
two scenarios, so that a mild year of the present
scenario and a stormy year of the doubled CO2
scenario have an equivalent number of cyclones.
No significant change in the geographic distribution
of the cyclones has been found. The simulation
with CO2 doubling presents a few intense events
that suggest a possible intensification of extreme
cyclones. The number of events is, anyway, too
small to have a reliable statistical significance.

Extreme waves and surge events in a doubled CO2
scenario in the Adriatic Sea. P.Lionello
(University of Lecce) E.Elvini, A.Nizzero
(University of Padua)

The Venetian littoral, the flat northern coast of the
Adriatic Sea, is particularly vulnerable to extreme
marine events. An intensification or a change of the
regimes of storm surges and waves, due to the in-
creased CO2 content in the atmosphere, would
require the reorganization of the coastal defenses
for the protection of an economically important
area and a unique historical heritage.
This analysis is based on 30 year long simulations
of the wave field and of the storm surge in the
Adriatic Sea using the information provided by the
global (T106) time slice simulations of the CTR
and CO2 scenario carried out by DMI.The wave
field simulations have been carried out using the
WAM model at 1/6 degs. resolution. The surge
simulations have been carried out with a barotropic
shallow water model at 1/12 degs resolution.
A crucial element for the accurate reproduction of
surge events and of the wave field is the quality of
the surface wind. Unfortunately, the T106 resolu-
tion is absolutely inadequate for the reproduction of
the surface wind field in the Adriatic Sea. There-
fore, it is not possible to use directly the T106 wind
fields for the evaluation of surges and waves
scenarios.
The wind fields used for the regional simulations
have been obtained by statistical downscaling of the
T106 slp fields. The downscaling was based on the
CCA analysis of the ERA-15 T106 slp fields and of
the regional surface wind fields produced by the
BOLAM (BOlogna Limited Area Model) at a re-
solution of 25km.
The 30 year long simulation carried out with WAM
and with the surge model have been used for an
analysis of the distribution of the extreme values.
The results show a significant diminished intensity
of the extreme events. It appears that the dimi-
nished cyclonic activity results also in a diminished
danger of coastal floods (ignoring the effect of the
sea level rise) and a diminished intensity of the
wave regime. The evaluation of the surge scenarios
should be considered cautiously, because of the
systematic underevaluation (not yet quantitatively
assessed) of surge events at the Venetian littoral
present in the CTR scenario.

Extreme value analysis of absolute minimum tem-
perature and minimum dew-point temperature at
Sodankyl, Finland, Jaakko Helminen, Niklas
Sjblom (Finnish Meteorological Institute (FMI),
Helsinki, Finland)

On J anuary 29, 1999 new record low absolute mini-
mum temperatures were reached at several sites in
Finnish Lapland with severe impacts on local ener-
gy supply. This led to the need to re-estimate the
corresponding recurrence values. The results based
on the peak-over-threshold (POT) approach of the
absolute minimum temperature at Sodankyl for
periods 1908-1998 and 1908-2000 will be
presented and compared. The other interesting
question in this context to be asked is how low the
absolute minimum temperature can reach all in all.
As such limit the minimum dew-point temperature
will be considered as it represents the temperature
value at which the formation of shallow ice crystal
fog will stop further cooling due to long-wave
radiation at the ground. To clarify the question of
the lowest possible absolute minimum temperature
values the POT-analysis results of the minimum
dew-point temperature will be presented and
discussed as well.
31

Observed Changes in Temperature Extremes of
Korea, Won-Tae Kwon, Hee-Jeong Baek*,
Byung-Choel Choi (Meteorological Research
Institute, Korea Meteorological Administration; *
Seoul National University, Korea)

Changes in mean, maximum and minimum tem-
perature with daily resolution are calculated with 14
stations data (with relatively long records) in
Korea. In the twentieth century, temperature in
Korea is increasing and especially in Seoul, the
minimum temperature has shown a large increase.
Since 1980's the trends in minimum, maximum, and
as well as mean temperatures become steeper than
the earlier decades. Ullung-do, an Island station,
shows no trend until 1980's but indicates an
increasing trend since then. The temperature
increases during the twentieth century are between
1C for rural and island stations, and 2C for
large metropolitan areas. Another prominent
change occurs in daily temperature ranges, showing
decreasing trend due to a large increase in
minimum temperature. We also analyzed the
changes in extremely warm and cold days, and
freezing days (T < 0C) and hot summer days
(Tmin >25C). The numbers of extremely warm
day and hot summer day are increased and this
trend is more prominent during 1990's. The
numbers of extremely cold day and freezing day
have decreasing trends and the trends show sharp
drops in 1950's and 1980's. And total numbers of
extremely warm and cold days are increasing with
time, which suggests that the climate of Korea has
become more extreme in recent decades. We will
present the detailed results from station data and
also mid-tropospheric data.

Assimilation

The role of atmospheric large-scale circulation
over Europe and North Atlantic in sea level varia-
bility on the Polish coastal zone, Miroslaw Mietus
(Institute of Meteorology and Water Management,
Maritime Branch, Gdynia, Poland)

An extensive study concerning the relationship be-
tween large-scale pressure field over Europe and
North Atlantic and climatic and sea level variability
on the Polish coast has been carried out for the time
span of the last century and in case of expected
increase of the CO2 concentration.
EOF analysis of mean sea level on the Polish coast
shows that there is generally one dominating mode
of spatial variability explaining more than 90 per
cent of variance in each season (only in summer 87
per cent). The leading mode presents increasing
deviation from the long-term mean when moving
from the western part of the coast towards east. The
second mode explains 7-8 per cent of variance.
Variability of the first principal components of
annu-al time series contains statistically significant
trends.
The first pair of canonical patterns for annual mean
explains 23 per cent of variance of regional sea le-
vel pressure and 95 per cent of the sea level va-
riance at Polish coast. The correlation between pat-
terns equals 0.74. The first canonical map of sea
level pressure presents the Northern and Central
Europe under the influence of pressure system with
center over Finland. The rest of continent and At-
lantic Ocean are under the influence of a weaker
and more extend pressure system with center over
Atlantic Ocean, west of Brittany. Such pattern is
responsible for the changes of water amount in the
Baltic Sea. The corresponding canonical map of sea
level is similar to its well-known first empirical
orthogonal function.
Monthly and daily variability of sea level were re-
constructed by means of canonical correlations
patterns or redundancy patterns. They remain in
good and very good agreement with in situ obser-
vations (from the years 1947-95) in all seasons of
the year expect of summer. Correlations coeffi-
cients between reconstructed time series and in situ
measurements reach to 0.88. Reconstructed time se-
ries reveal positive trends in the nearly 100 years
1899-1995. Reconstructed positive trends are
caused by an intensification of westerly flow over
the Baltic Sea Basin and due to increasing stability
of westerly flow. An increase in the range of sea
level variability is also confirmed by tendency in
increase of higher level percentiles.
In case of increasing CO2 concentration a contin-
uation of nowadays-existing mean sea level rise is
expected. ECHAM1/LSG pre-industrial experiment
shows that there is expected slight increase of sea
level rise in the respect to control run. In the end of
6th decade of 21st century a jump of sea level in
range of 4-6 cm may have place. ECHAM3 time-
slice experiment shows that the mentioned rise
might be additionally accelerated by 0.7-0.9 cm
decade-1 in the respect to 1xCO2 concentration.
The same tendencies are expected for minimum and
maximum sea level values. The expected increase
of percentiles should be lower the increase of mean
sea level, however range of their variability should
be greater.

Space-time variograms and a functional form for
total air pollution measures, De Iaco, S.
1
, Myers,
D. E.
2
, Posa, D.
3,4
(
1
Universita di Chieti, Pescara
Italy
2
University of Arizona, Tucson AZ - USA;
3
Universit di Lecce, Lecce - Italy;
4
IRMA - CNR,
Bari - Italy)

There are different ways to quantify Total Air
Pollution (TAP) for a region. Spatial averages, over
a region of interest, or temporal averages, for a
specified time period, could be considered for each
contaminant or for a weighted average of several
contaminants. Such averages might be interpreted
as the total mass of contamination for a time period
and a geographic region. Knowing the possible
32
sources and their individual emissions would allow
a comparison to determine if pollution is flowing
into or out of a region. The comparison might also
be used to determine if there are un-identified sour-
ces.
As an application of quantifying total air pollution,
data for NO, NO2 and CO collected in the Milan
District, Italy are used to generate a combined in-
dicator of pollution, TAP1. The weightings are ob-
tained by multiple PCA analyses of the daily con-
centration levels. The first components are treated
as data in space-time and a space-time product-sum
variogram is fitted to the data. Using this model and
the dual form of kriging, i.e., Radial Basis Func-
tions, a functional form for the TAP1 concen-
trations inspace-time could be obtained. Using this
functional form, spatial and temporal integrals pro-
duce measures of total air pollution over a region in
space for a specified time period. Using the speci-
fied legal limits on the three contaminants, a limit is
generated for TAP1. In turn this can be used to
identify the region in space-time where the maxi-
mum for TAP1 is exceeded.

Performance of four-dimensional variational assi-
milation and dynamical instability, experiments
on a quasi-geostrophic model, Carlos Pires
(CGUL, - Centro de Geofisica da Universidade de
Lisboa, Lisboa, Portugal)

We perform four-dimensional variational assimi-
lation (4DVAR) of artificially generated noisy ob-
servations on a multi-level quasi-geostrophic model
in the perfect model setting. The quality of analyses
and corresponding forecasts are specially improved
with the extension of the assimilation period Ta up
to 8-10 days (Pires et. al. 1996; Swanson et. al.,
1998). One obtains an explicit formula of the
covariance matrix C of the assimilation error,
showing its dependence on the assimilation period
length Ta, the flow instability and the temporal
observation density. An approximation of the
leading eigen structure of C is obtained using the
linear and adjoint models. This approximation is
compared to that obtained by statistical estimation
of C trough the generation of a large set of different
realizations of observation errors. The test is
performed for a blocking case in the Atlantic sector
where the present is put a few days before the
blocking onset. The 4DVAR assimilated solutions
are able to track the blocking transition if Ta is
sufficiently extended to the past. The eigen-
structure of C tends to the space spanned by the
asymptotic Lyapunov vectors at the present time.
However the assymptotic limit of C is obtained in
practice for a finite useful assimilation period Tau
depending much on the recent dynamical instability
of the flow during the assimilation period. Weak
instabilities are favorable to small 4DVAR analysis
errors and large useful assimilation periods,
especially if that occurs recently in the assimilation
period. We access the analysis errors, data
assimilability and predictability for different
patterns of dynamical instability, measured by the
leading finite singular values of the linear
propagator. The effect of observation density is also
accessed, especially near the tracking of large flow
instabilities. In a real setting, this study can help to
put limits of data assimilability and predictability of
4DVAR assimilated solutions.

A multisite self-consistent Weather Generator for
Impact Studies in Portugal, Budong Qian, Joo
Corte-Real, Hong Xu (ICAT, Faculty of Sciences,
University of Lisbon, Portugal)

Through previous studies done in past years,
downscaling of precipitation and temperature for
Portugal, from large-scale output of GCMs inclu-
ding both control and transient (scenario) runs, has
essentially been achieved (Corte-Real et al., 1999).
However, previous downscaling studies were de-
veloped site by site, not taking into account in a
consistent way, the interrelationships between pre-
cipitation and temperature, while some models
designed for impact studies, such as hydrological
models, require spatially distributed precipitation
scenarios consistent with temperature scenarios as
well. Stochastic weather generators have been de-
veloped either to model climate elements (such as
precipitation, temperature and radiation in many
studies) at a given site or to simulate multisite
distributions of a specific element such as preci-
pitation (Wilks, 1999). In this contribution, a sto-
chastic downscaling model, developed for conti-
nental Portugal, for obtaining spatially distributed
regional precipitation scenarios consistent with
temperature scenarios and conditioned on daily
circulation patterns is presented and validated. Im-
plications of the model for impact studies are also
discussed.

Monthly mean forecasts of rainfall and tempe-
rature in Portugal, Joao Rio, Carols Pires (CGUL,
- Centro de Geofisica da Universidade de Lisboa,
Lisboa, Portugal)

We assess the quality of monthly mean forecasts of
rainfall and temperature in Portugal using a simple
and low-cost statistical model. The statistical pre-
dictors are extracted from the NCEP Reanalysis of
the daily 1000hPa geopotential height field (z1000).
EOF analysis is performed on the North Atlantic-
Europe area selecting the leading 10 PCs at the
forecast time as predictors. The monthly cumulated
rainfall and the monthly mean temperature are
computed in three complementary regions: north,
center and south, from which we compute the
distribution terciles. That quantities are then classi-
fied in one of the three classes: above, normal, be-
low. The forecast is based on the analogue method
using the angle between the predictors vector as
similarity function. The search for analogues is
performed in cross validation on a daily basis from
data within a period centered at the same month of
the forecast time. If the angle is below a given
threshold then that day is considered an analogue.
33
The predicted class is the most populated from the
pool of chosen analogues. The forecast method is
the one-step PC method presented by Vautard et. al.
(1996). Contingency tables with forecast/observed
classes are built for each region and variable from
which we compute the LEPS (Liner Error In Proba-
bility Space) to access the forecasts quality. This
score vanishes for constant or random forecasts and
it is one for correct forecasts. The results show
higher scores for the cumulated precipitation, es-
pecially in winter months, which is associated to
the occurrence of large-scale patterns favourable to
frontal rainfall. In contrast, precipitation in summer
which is mostly associated to local conditions
which are not captured by the chosen oredictors.
Further work is ongoing as we intend to increase
the number of data stations and the predictor
information (e.g. using sea surface temperature in
the Atlantic).

Sensitivity Studies on Wind Measurement Errors
in Mesoscale Model Simulations, Radhika
Ramachandran, Prakash M Dolas (Space Physics
Laborator, Vikram Sarabhai Space Centre,
Trivandrum, India)

Large errors in radiosonde sounding data find a
mention in literature from time to time. Pressure
levels have to be determined accurately for obtai-
ning heights through the hydrostatic relation, for
example. Temperature and humidity along with the
winds are the other independent fields measured by
a radiosonde.
We have initiated wind forecast studies for the
south west monsoon regime, comprising the months
of June, July and August in India .The domain
centre of the study is the coastal station Chennai
(13.1 N, 80.3 E), located off the Bay of Bengal.
The wind regime at Chennai during this period is
characterised by low level westerly jet centred
around the height of 2.5 km and the tropospheric
easterly jet (TEJ ) with core centred around 16 km.
The average maximum in the westerly regime is
about 10 m/s while in the TEJ , the average easterly
maximum is about 35 m/s. The upper tropospheric
westerlies extend at least up to 35 km in the
stratosphere. TEJ has an azimuth very close to 90
with a steadiness factor of 98 % while the low -
level westerly jet has an azimuth close to 285 with
a steadiness factor of 94 %.
Presently we are conducting 2D and 3D simulations
using the mesoscale Advanced Regional Predic-
tion Systems (ARPS) model. This is a non-hydro-
static, fully compressible primitive equation model
in a generalised terrain following co-ordinate sys-
tem developed by the Centre for Analysis and
Prediction of Storms (CAPS), Oklahoma, USA. In
this study we will discuss the effect of the errors in
the initialising wind field on the wind forecast.

Building the bridge between sequential data assi-
milation and geostatistics: the stationarity issue,
Julien Snegas, Hans Wackernagel (Centre de
Geostatistique, Ecole des Mines de Paris, Fon-
tainebleau, France)

A central issue in spatio-temporal prediction is the
modeling of the spatial covariance. This problem is
solved differently in sequential data assimilation
and in geostatistics. In the first approach, regardless
of the algorithm used, the spatial structure is propa-
gated forward in time by means of the physical
model (Bennett, 1992, Evensen, 1997), so that the
spatial covariance at a given time step is the result
of an initial structure and the dynamics of the
system. In the second approach, the inference of the
spatial structure relies on an assumption of spatial
stationarity (or homogeneity) which makes it pos-
sible to compute an experimental covariance on the
basis of a single realization. However, the assump-
tion of stationarity is often too strong, and several
extensions have been proposed, like the Intrinsic
Random Functions (Matheron, 1973) or more
recently a spatial deformation model (Guttorp and
Sampson, 1994).
The present paper points out the respecting ad-
vantages of these two types of methodologies and
indicates how they can benefit from each other.
From a geostatistical point of view, the update algo-
rithm used in sequential data assimilation can in-
deed be seen as a very powerful tool to generate a
nonstationary spatial covariance, liable to take into
account local variations like those induced by
topography, meteorology, time dependent variables
or boundary conditions. Conversely, stationary
spatial covariances are of ten used in data assi-
milation to model the system noise, for which at
least the stationarity assumption should be tested.
We show how a reasonable use of the physical
properties of the variables can serve, in both fields,
the modeling of the spatial structure, especially
when the available data are scarce in space.
Several examples from hydrodynamics will be
given to illustrate these features.

Public Debate and Climate Model Products:
Whereof must one be silent? Leonard A. Smith
(Centre for the Analysis of Time Series, London
School of Economics, London, UK)

Climate modelling holds distinction in the physical
sciences in that it applies quantitative methods to
produce detailed predictions which can only be re-
fined and evaluated in-sample. This raises the justi-
fied scepticism of statisticians, although climate
scientists can hardly be held responsible for the
time constants set by the problem at hand. They
can, however, be held accountable for which model
products (in particular the level of detail in any
forecast) are supplied to policy makers and to the
public. This paper argues for a robust attempt to
establish which model products might prove re-
alistic (i.e. what are the smallest space and time
averages can be discussed?).
34
A statistician may well ask why it has taken so
many years to fit the Earth's global mean surface
temperature (~128 points) given models with 2^20
degrees-of-freedom and an uncounted number of
parameters? The response, of course, is to appeal to
the Laws of Physics: the number of free degrees-of-
freedom are many fewer than the number of varia-
bles; both variables, parameterisations, and para-
meter values are constrained by "the physics" and
the resulting simulation yields a realistic reproduc-
tion of the entire planet's climate system to within
reasonable bounds. But what bounds? exactly?
This paper sketches a framework with which to
quantify those bounds. We define, for example, a
temporal credibility ratio: the smallest time-step in
the model divided by the the shortest duration over
which some model variable must be averaged if it is
to compare favourably with observations. For a sin-
gle model run under transient forcing scenario,
there are good reasons that this ratio must be small
on statistical grounds; most of these reasons vanish
if an ensemble of runs are made. The statistics of
bounding boxes in moderate dimensional spaces is
discussed, and the likely number of ensemble mem-
bers required to capture reality is noted. This pro-
vides a direct measure, for example, of the
(in)ability of a model to provide insight on regional
changes: if a model cannot capture regional va-
riations in the data on which the model was con-
structed (that is, in-sample) claims that out-of-
sample predictions of such regional averages
should be used in policy making are vacuous.
Related points will also be touched upon, including
questions such as resource allocation: is it better to
have the best guess available (with no meaningful
quantitative estimate of its reliability whatsoever),
or to estimate the likely range of outcomes, further
qualified by evidence of how well this particular
event (climate variable) can be simulated in this
particular model?

Multiple Analysis of Series for Homogenization
(MASH). Seasonal Application of MASH. Auto-
matic using of Meta Data, Tamas Szentimrey
(Hungarian Meteorological Service, Budapest,
Hungary)

The MASH (Multiple Analysis of Series for Homo-
genization) method is a relative homogeneity test
procedure that does not assume the reference series
are homogeneous. Possible break points and shifts
can be detected and adjusted through mutual com-
parisons of series within the same climatic area.
The candidate series is chosen from the available
time series and the remaining series are considered
as reference series. The role of series changes step
by step in the course of the procedure. Several
difference series are constructed from the candidate
and weighted reference series. The optimal weigh-
ting is determined by minimizing the variance of
the difference series, in order to increase the
efficiency of the statistical tests. Providing that the
candidate series is the only common series of all the
difference series, break points detected in all the
difference series can be attributed to the candidate
series.
A new multiple break points detection procedure
has been developed which takes the problem of sig-
nificance and efficiency into account. The sig-
nificance and the efficiency are formulated accor-
ding to the conventional statistics related to type
one and type two errors, respectively. This test
obtains not only estimated break points and shift
values, but the corresponding confidence intervals
as well. The series can be adjusted by using the
point and interval estimates. Since a MASH pro-
gram system has been developed for the PC, the
application of this method is relatively easy. The
paper focuses on new developments of MASH
which are connected with two special problems of
the homogenization of climatic time series. One of
themis the relation of monthly, seas onal and
annual series.
The problem arises from the fact, that the signal to
noise ratio is probably less in case of monthly series
than in case of derived seasonal or annual ones.
Consequently the inhomogeneity can be detected
easier at the derived series although we intend to
adjust the monthly series.
The second problem is connected with the usage of
meta data in the course of homogenization proce-
dure. The developed version of MASH system
makes possible to use the meta data information - in
particular the probable dates of break points -
automatically. The decision rule is based on the
confidence intervals given for the break points.

A data assimilation method with a coupled ocean-
land-atmosphere model and its impact on the
ocean state and on medium-range pre-dictions,
Clemente A. S. Tanajura, Konstantin P. Belyaev
(Laboratorio Nacional de Computao Cientifica
(MCT/LNCC), Petrpolis, RJ, Brazil)
Ben P. Kirtman (Center for Ocean-Land-Atmos-
phere Studies (IGES/COLA), Calverton, MD,
USA)

A data assimilation method based on the well-
known Kalman filter scheme is considered with the
COLA coupled ocean-land-atmosphere model. This
method is applied to assimilate in situ observations
of surface and subsurface temperatures from the
PIRATA project. Results of the assimilation pro-
duce a new corrected temperaturesalinity ocean
state, which can be used to estimate different
oceanographic quantities. In particular, the present
paper contains new reconstructed meridional heat
fluxes, and focus on the tropical Atlantic Ocean.
One-month forecast experiments during 1999 have
been performed. The impact of the assimilation on
the forecasted temperature profiles of the upper 500
meters is investigated.

35
Assessing Model Skill in Climate Simulations,
Karl E. Taylor (Program for Climate Model Diag-
nosis and Intercomparison (PCMDI), Lawrence
Livermore National Laboratory, Livermore, CA,
USA)

Although skill scores in the weather forecast
community are routinely used to assess model per-
formance, this practice is rather uncommon in the
climate modeling community. Skill scores have
been proposed, but there is little agreement or even
understanding of the relative merits of various
scores. Most skill scores are related in some way to
the common second-order statistics that charac-
terize the degree of pattern correspondence between
simulated and observed fields (e.g., correlation,
RMS difference, and ratio of variance). A family of
new measures of model skill is proposed, which
allows one to specify the relative importance placed
on succcessfully simulating the correct amplitude
of variations compared to obtaining a high
correlation. The scores account for the limits to
potential agreement between models and data that
result from unforced (internal) variabilty and obser-
vational errors. The use of these skill scores is
illustrated by applying them to output from the
Atmospheric Model Intercomparison Project
(AMIP), in which models were forced by observed
sea surface temperatures and sea ice cover. An
assessment of the relative merits of about 20
models participating in the project will be given,
both in terms of their ability to simulate the
climatological mean state and the year to year
deviations from that state. An ensemble of inde-
pendent realizations of the simulated climate ob-
tained from a single model will then be used to
assess the statistical signficance of apparent diffe-
rences in model skill and to estimate fundamental
limits to agreement that must result from unforced
variability. The limitations of using a single mea-
sure to quantify a model's skill in simulating an ob-
served field will also be discussed, and the use of a
new diagram to simultaneously show several mea-
sures of skill will be described.


XXth Century Behaviour of Temperature and
Precipitation Extremes in Hungary, Jnos Mika,
Tams Szentimrey (Hungarian Meteorological
Service, Budapest, Hungary); Judit Bartholy, Rita
Pongrcz (Etvs Lrnd University, Dept.
Meteorology, Budapest, Hungary); Lszl Szeidl
(Department of Mathematics, University of Pcs,
Hungary)

Monthly extremes are analysed at ten stations
representing 93,000 sq. km of the country. Ten and
30 years statistics of four characteristics are
parallelly computed and evaluated. These are: the
mean; the year of new record setting; the range
between the two extremes of opposite sign in the 10
or 30 years period; and frequency of extremes
defined as 10 % of the empirical distribution.
Four considerations are quantified, namely the
long-term changes, characterised by variations of
the 10 years statistics; the representativity of pre-
vious 30 years periods; the effect of homo-
genisation; and the spatial synchronity of extremes
at the 10 stations in Hungary.

Some typical questions from the 4 x 4 ones:
* Which decades exhibited much more (less)
record settings than the theoretical 1/N ?
* Did the frequency of extremes change
monotonously or in correlation to the global trends
?
* Did the homogenisation change the distribution
of new records or the 10 % extremes ?
* Can we estimate the monthly ranges from
previous (i.e. 10 years old) 30 years periods ?
* Do the new monthly record setting occur para-
llelly in such a relatively small area ?

For preliminary conclusion, one can establish that,
although even the 30 years periods exhibited con-
siderable variability in Hungary, neither the fre-
quency, nor the departure from the mean increased
with time or with the global temperature in the past
century.

Homogenized Daily Temperatures for Trend Ana-
lyses in Extremes over Canada, Lucie Vincent,
Xuebin Zhang, Barrie Bonsal and Bill Hogg
(Climate Research Branch, Environment Canada)

A database of long-term and homogenized monthly
temperatures has been created for climate change
analyses over Canada. A technique based on re-
gression models was developed to identify "in-
homogeneities" in individual annual maximum and
minimum temperature time series using sur-
rounding stations. The inhomogeneities were often
step changes caused by factors such as changes in
the station exposure, location, instrumentation, and
observing program. Using this technique, it was
possible to divide the tested series into homo-
geneous segments at the identified steps. Monthly
adjustment factors, derived from the regression
models, were applied to bring each segment into
agreement with the most recent homogeneous part
of the series.
Recent work has involved the adjustments to daily
maximum and minimum temperatures. Reliable
daily datasets are essential for the accurate des-
cription of spatial and temporal characteristics of
extreme temperature related variables. Adjustment
of daily temperatures has raised several concerns
mainly due to the large variability in day to day
temperature, particularly at high latitude stations.
An approach was investigated using linear inter-
polation between "target values" derived from the
monthly adjustment factors. This approach has two
main advantages. First, it doesnt require the use of
a daily reference series (which is often difficult to
produce), and second, the average of the daily
36
adjustments over a particular month is equal to the
monthly adjustment factor. Therefore,
homogenized daily temperatures compatible with
the homogenized monthly datasets can be obtained.
To assess the impact of these daily adjustments on
extreme temperatures, trends in the percentage of
days above (below) the 95th (5th) percentile have
been computed for maximum and minimum daily
temperatures both before and after adjustment.
Seasonal trend maps were then examined for
various periods of time. Preliminary results show
that the inhomogeneity having the greatest impact
on the trends during the past five decades is
associated with the 1961 change to observing
window at principal stations. An overall assessment
of Canadian temperature extremes indicates no
consistent trends in the number of extreme hot
summer days, but significant trends toward fewer
days with extreme low minimum temperatures du-
ring the last century.

The analysis of extreme temperature distributions
on the example of Poland, Joanna Wibig
(Department of Meteorology and Climatology
University of Lodz)

The surface air temperature of the world has
warmed by about 0.3-0.6oC since the 19th century.
This warming trend should be associated with the
variation of daily extreme temperatures. There are
indications of the rise of the daily minimum and
maximum temperature in some areas. Brzdil at al.
(1996) have shown that in Poland the linear in-
crease in mean seasonal daily maximum tempe-
rature is slightly higher than that of daily minimum
temperature in all seasons but autumn during the
period 1951-1990. Wibig (2000) has shown the
increase of minimum temperature at some stations
in Poland in a longer period. Therefore the changes
are not uniform even on the area of Poland. The
question arise: has the frequency of extreme events
changed during the present century? If so, the
extremes in temperature can have significant social
and economic impacts. Nicholls et al. (1996) states
that although there is no evidence for global
changes they probably exist on the regional scale.
These changes can manifest not only as a shift of
the mean value but also as variation in other
parameters of the extreme values distribution. The
aim of this paper is to fit the theoretical distribution
to the records of daily extreme temperature and the
of extreme temperature for the longer period
(season, year) and analyze the long-term course of
parameters of such distributions. Several distri-
butions and fitting methods were tested and the
results were compared.

Comparison of 20- and 50-year return values in
GCM-simulated, downscaled and observed tem-
perature series, Jan Kysel
1,2
, Radan Huth
1

(
1
Institute of Atmospheric Physics AS CR, Prague,
Czech Republic;
2
Dept. of Meteorology and
Environment Protection, Charles University,
Prague, Czech Republic)

Extreme weather and climate events severely in-
fluence ecosystems and human society. High and
low temperature extremes are among the most fre-
quently investigated extreme events. Nevertheless,
relatively little work has been done as regards ex-
tremes of surface temperature in GCM and down-
scaling studies even though it appears to be clear
that impacts of climate change would result mainly
from changes in climate variability and extreme
events. This study concentrates on the comparison
of 20- and 50-year return values of daily maximum
and minimum temperatures in (i) observation, (ii)
GCM simulated control climates (ECHAM3 and
CCCM2 GCMs are employed), (iii) statistical
downscaling from observation, and (iv) statistical
downscaling from GCMs. As for the area involved,
several sites in central Europe and the nearest
gridpoints corresponding to the stations are
analyzed. Observations cover the period 1961-
1990.
The downscaling method used is a stepwise mul-
tiple regression of 500 hPa height and 1000/500
hPa thickness gridpoint values over most of Europe
and adjacent Atlantic Ocean. Two methods of
enhancing the downscaled variance to become
equal to that observed are compared, namely the
inflation of variance and white noise addition.
Since downscaled temperature series reproduce the
observed means and variances, for a fair
comparison between GCMs and downscaling the
distributions of GCM-produced temperatures were
re-sized to have the observed mean and standard
deviation.
Extreme value analysis is performed by fitting the
Gumbel distribution and/or the generalized extreme
value (GEV) distribution to the sample of annual
extremes of maximum and minimum temperature.
Various methods of the estimation of parameters of
extreme value distribution are intercompared,
including the method of L-moments and maximum
likelihood method.
The comparison among GCM-simulated, down-
scaled and observed 20- and 50-yr return values of
temperature does not appear to be sensitive to what
statistical method of the estimation of parameters of
extreme value distribution is used, although in-
dividual return values are influenced by the choice
of the method. The skill of both GCMs in repro-
ducing extreme high and low temperatures is
limited. The statistical downscaling from GCMs
tends to improve the results although it generally
yields extremes that are too moderate compared to
observed values (this holds for downscaling both
from observation and GCMs). White noise addition
in downscaling from observation leads to more
realistic return values of maximum and minimum
temperature than variance inflation.

Test on the probability distribution of maximum
air temperatures, Jana Jurekov
1
, Jan Picek
2
,
Jaroslava Kalvov
1*
(
1
Charles University in
37
Prague,
2
Technical University in Liberec, Czech
Republic; *Dept. of Meteorology and
Environment Protection, Faculty of Mathematics
and Physics, Charles University, Praha, Czech
Republic)

If we are interested in the extreme values of tem-
peratures or precipitation, the tails of the underlying
probability distribution are more interesting than its
central part. Hence, a specific test on the shape of
tails is desirable. After a collection of observations,
we would first like to decide whether the
distribution is light- or heavy-tailed, and what can
be the value of its Pareto index. More specifically,
our problem is to decide whether the summer daily
maximum temperatures at the Czech Republic
stations belong to the domain of attraction of the
Frchet rather than of the Gumbel extreme value
distributions. In accordance with the previous
investigation of the authors, we suppose that the
vector of the summer daily maximum temperatures
X(n) follows the autoregression model of order 1,
[Image] (1)
t=0, 1, 2, , where { et } is a sequence of
independent random variables, identically dis-
tributed with distribution function F and density
(innovation density) f, generally unknown. The
hypothesis states that - log(1 - F(x)) ~m log x as
x , where 0<m m0, m0 prescribed. A finite
value of m means that the distribution function F
has a heavy right tail and belongs to the domain of
attraction of the Frchet distribution. The proposed
test is based on the tail behavior of the extreme
autoregressive quantiles of the model (1). The
asymptotic null distribution of the test criterion is
normal and the test is consistent against the
alternative of lighter tails as well as of the
exponential tails. The test is applied to the series of
the daily maximum temperatures at the Czech
Republic station in the period 1961 - 1998.

Three Approaches to determine extreme years of
Global Atmospheric Temperature, Luis Gimeno
1,*
,
Juan Ael
1
, Higinio Jorge
1
, Pedro Ribera
1
, David
Gallego
2
, Ricardo Garcia
2
, Emiliano Hernandez
2
(
1
Universidad de Vigo. Facultad de Ciencias de
Orense, Orense, Spain;
2
Universidad Complutense
de Madrid. Facultad de Ciencias Fisicas, Madrid.
Spain; *Corresponding author)

Global atmospheric temperature is usually com-
puted from the average of a combination of land air
temperature anomalies and sea surface temperature
anomalies . Both components of the dataset are ex-
pressed as anomalies from 1961-90, as this makes
computation much easier. In this work a daily 2.5
latitude x 2.5 longitude grid box temperatures
from NCAR-NCEP reanalysis at three pressure
levels (850, 500 and 200 hPa) and for the period
1958-1998 were used to determine extreme years of
global atmospheric temperatures by means of three
different approaches: 1) The difference between the
world surface area with positive temperature ano-
malies from 1961-90 and the world surface area
with negative temperature anomalies from 1961-90,
2) The global number of hot days (number of days
with temperatures exceeding the 90 percentile from
1961-1990) and the number of cold days (number
of days with temperatures being lower than the 10
percentile from 1961-1990 and 3) The percentage
of world area with annual temperature exceeding
the 90 percentile from 1961-1990 and the per-
centage of world surface area with annual tem-
perature being lower than the 10 percentile from
1961-1990.

Bayesian Statistical Modelling of Nonlinear
Climate Processes, E. P. Campbell, Y. Li (CSIRO
Mathematical and Information Sciences, ,
Wembley, Australia)

Nonlinear statistical methods have enjoyed rapid
development during the 1990s, and offer great
potential as a tool for climatologists seeking rela-
tionships of predictive value. A key issue still to be
resolved however is predictor selection. In the time
series context this requires selection both of optimal
predictors and their lags. A powerful new metho-
dology for model selection within a Bayesian
framework, known as reversible jump Markov
chain Monte Carlo, has emerged in the statistical
literature in the late 1990s. We show how this
methodology can be used to identify and fit non-
linear time series models, and illustrate the
approach using threshold models for monthly rain-
fall. We show how our approach can be generalised
to select nonparametric nonlinear time series mo-
dels.

Modelling and characterizing extreme drought
events, Jess Abaurrea, Ana Carmen Cebrian
(Dpto. Metodos Estadsticos, Universidad de
Zaragoza, Ed. Matemticas. Pedro Cerbuna, 12.,
Zaragoza, Spain,)

The aim of this work is to develop a stochastic mo-
del to analyze the occurrence and severity of
meteorological drought events. The description of
the characteristics of droughts by probability distri-
butions provides measures such as the inter-drought
recurrence time, the expected duration or deficit
and the return values, which are helpful in water
resource management.
In order to define a drought event we use an opera-
tional definition based on the comparison of a sto-
chastic process, s(t), describing the hydrological
state of the system and a threshold, u(t), which re-
presents a critical level for the process. For the
analysis of long-term meteorological droughts
presented in this work, we use as signal the
monthly moving annual rainfall. The proposed
model, based on results from Extreme Value
Theory and Excess over Threshold methods, is a
compound Poisson cluster process, CPClP, which
combines a Poisson Cluster process, to represent
the drought occurrence, with a vector series
composed by three random variables -duration,
38
deficit and maximum intensity- to describe the
drought magnitude. Six Spanish monthly rainfall
series of about one hundred years long are studied
to check this model.
The fitted CPClP model is also used as the basis to
characterize the largest drought event to occur in a
given period of time. This characterization requires
the knowledge of the distribution of the maximum
in samples of random size. Results about this issue
in samples with Poisson size are presented for some
of the most frequent distributions in environmental
applications.

Extremes in Instationary Time Series, Jrgen
Grieser (Institute for Meteorology and
Geophysics, University Frankfurt, Germany)

Extreme values of meteorological variables threaten
human life and property. Thus, knowledge of the
statistical features of extremes is important. The
concerned variables are the waiting time distribu-
tion, return period and risk. These variables are
easily obtained if the underlying meteorological
variable is independent and identically distributed.
Otherwise the probability to exceed a threshold is
not constant. It is demonstrated that the widely used
return periods are not a useful variable in this case.
However, if the meteorological variable itself has a
certain temporal structure also the probability to
exceed a threshold changes in time in a determined
way.
It is shown that the statistical features of extremes
can be drawn from the statistical features of the
meteorological variable itself. The latter one may
be taken from observations or from model runs.
Two real world examples demonstrate that the
statistical features of extreme values may change
considerably even if the mean features of the under-
lying meteorological variable have only slightly
changed.

Comparison of Several Methods for the Esti-
mation of Extreme Value Distribution in Time
Series with Trends, Henrique Oliveira Pires, Joao
Pestana Ferreira (Instituto de Meteorologia,
Lisboa, Portugal)

The method of Peaks Over Threshold (POT) for
extreme value statistics, where the number of peaks
is modelled by a Poisson distribution and the value
over the threshold by an exponential distribution is
compared with methods based on extreme value
asymptotic distribution theory.
Estimation is made using Gumbel distribution (ex-
treme value Type I) with just a location and a dis-
persion parameters and, also, using the generalised
von Mises distribution where there is also a shape
parameter; von Mises-J enkinson distribution re-
duces to Gu mbel when this shape parameter is
null.
The influence of trends, which is essential in a
changing climate, is evaluated by fitting, using
maximum likelihood estimation, the Gumbel model
and the generalised von Mises-J enkinson distri-
bu-tion, but allowing for a linear variation of the lo-
cation parameter with time. The maximum like-
lihood equations are solved by numerical methods.
The maximums of the likelihood achieved can then
be compared. As an example we use a 100 years
time series of the air temperature in Lisbon where
there is a very significant increasing trend.

The Possible Influence of Large Volcanic Erup-
tions on El Nino, Philippe Naveau (National
Center for Atmospheric Research, Boulder, USA)

The El-Nino Southern Oscillation (ENSO) is the
largest source of inter-annual climatic variability on
a global scale. To determine if large tropical erup-
tions influence the ENSO by introducing an ex-
ternal forcing associated with a cooling period, or if
deviations from the El-Nino cycle are only due to
random noise is of primary interest to the paleo-
climate and climate change communities. Hence,
the aim of this talk is to investigate the statistical
association between the ENSO and volcanic erup-
tions.
To study the historical behavior of El-Nino events
over a long period of time 1525-1987, we use the
data set constructed by Quinn and Neal (1995). To
assess volcanic activities from 1525 to 1991, we
work with the VEI (Volcanic Explosivety Index)
that describes the occurrences and the intensity of
volcanic eruptions (Newhall and Self 1982, updated
by Simkin and Siebert 1995).
To investigate the existence of a potential link be-
tween these two signals, we develop a variety of
statistical models that can take into account the
special characteristics of our data (e.g., volcanic
eruptions seem to occur in clusters). The most
complex model is a Hidden Markov Model (HMM)
that is based on two main components: the ob-
servations and some hidden atmospheric states. The
advantages of such a process is its flexibility and its
capabilities to model a variety of dependence struc-
tures between the ENSO and VEI. Finally, to
remove the risk of over-parameterization, we
compare different HMM's to simpler models
(independent Bernouilli processes and Markov
Chains) by using model selection techniques.

Simulation Models

Simplified atmosphere models with realistic vari-
ability, Ulrich Achatz (Leibniz-Institut fuer
Atmosphaerenphysik an der Universitaet Rostock
e.V., Kuehlungsborn, Germany); J.D. Opsteegh
(KNMI, De Bilt, The Netherlands)

The realism of climate simulations by general cir-
culation models is partly due to their use of an
enormous number of degrees of freedom. Unfor-
tunately this leads to an (ever increasing) loss of
transparency of specific model results and low
integration speeds, hampering both progress on
39
basic understanding of the climate system and the
applicability of GCMs for climate impact assess-
ment studies. There is therefore a need for com-
plementary climate models of low or intermediate
complexity which can reproduce the atmosphere's
behavior based on simplified descriptions. In the
light of its important role it seems desirable that
internal climate variability is also captured. Up to
now there is no low- or intermediate-complexity
climate model which satisfies this demand. Here
progress might be brought by so-called reduced
models utilizing realistic dynamics in combination
with an optimal set of basis patterns and a good
parameterization of the impact of scales and
processes not explicitly resolved. The patterns are
chosen such that they can give a rather complete
description of the climate attractor but simul-
taneously neglect highly improbable realizations of
unrealistic states. A good candidate for this are
empirical orthogonal functions (EOF). Achatz and
Branstator (1999) have recently shown that the
empirical etermination from a GCM data set of
linear parameterizations of the effect of ageos-
trophy, unresolved vertical and horizontal scales,
and unresolved physics can improve a reduced
quasigeostrophic two-layer model projected onto
EOFs such that it is able to reproduce the internal
variabiity of the GCM not only in midlatitudes but
also in the tropics. In oder to also include the
nonlinear tropical dynamics we have built on this
result and developed a corresponding model class
based on the primitive equations. Comparison with
the climatology of a conventional GCM
(ECHAM3) shows good agreement. Models based
on 500 and as few as 30 EOFs will be discussed
with respect to the simulation of the seasonal
dependence of mean state and fluxes, and to the
reproduction of recurrent anomalies.

Modelling of Temporal Rainfall by the Use of Sto-
chastic Point Process Models in View of Scaling,
Konrad Bogner (Institute of Hydrology, University
of Agricultural Research,Vienna, Austria)

Clustered rectangular point process models and va-
rious modifications are investigated in detail regar-
ding the assessment of fit. Besides testing the
agreement between observed properties and those
used in fitting the model, but examined at different
levels of aggregation, the generalized extreme value
analysis is applied to compare simulated and
observed extreme rainfall. In former publications a
lack of fit for extreme values of hourly aggregation
levels is reported sometimes. However the in-
vestigated modifications, like the incorporation of
third order moments in the fitting procedure,
indicate an improvement and better agreement for
small time intervals also. Furthermore assessment
of fit is made for simulated series of sub-hourly
time resolution considering the extreme values.
There are considerable differences between obser-
ved and simulated properties at these levels of
aggregation. Therefore the method of wavelet trans-
form is applied to reveal the self similarity and
fractal pattern of rainfall time series. The
combination of these analysis of stochastic point
process models and the scale of fluctuation will
facilitate to improve the disaggregation possibilities
and to identify the limits of applicability of point
process models.

Statistical Modelling of Tornado Occurrence,
Harold E. Brooks (NOAA/National Severe Storms
Laboratory Norman, Oklahoma)

Tornadoes are rare events at any location. Possible
changes in the distribution over time could have
significant effects on public safety. Unfortunately,
the quality of data collection has changed drama-
tically over the years in many countries (e.g., an-
nual tornado reports in Germany have increased by
a factor of 5-10 in the last 50 years). As a result,
changes in the "true" distribution of tornadoes can
be hard to distinguish from changes in the repor-
ting. It appears, however, that the conditional distri-
bution of tornadoes by intensity is relatively
consistent in many countries for tornadoes rated F2
or higher on the Fujita damage scale, decreasing
logarithmically with increase F-scale value. This
consistency allows us to make some estimate of the
degree of underreporting in different locations. For
instance, it appears likely that only about 15% of
French tornadoes are currently in their database.
A 75-year record of significant tornadoes (those
rated F2 or higher) in the United States has
provided input to a variety of statistical models of
tornado occurrence that allow us to estimate the
range of variability. The first model is a Markov
chain model of occurrence of at least one
significant tornado anywhere in the United States
on a daily basis. It provides estimates of the
probability of long runs of consecutive days with or
without tornadoes. The second model uses an
estimate of the probability of a significant tornado
on a daily basis on a grid in the US, based on a
simple kernel density estimation technique. Those
daily probabilities are then modelled using a beta
distribution, and the beta distributions used to
generate a long model run that provides estimates
of tornado occurrence and, of particular interest, the
interannual variability.

A method to extend the temporal validity of nested
regional climate simulations, Udo Busch and
Dietrich Heimann (DLR Oberpfaffenhofen,
Institut fr Physik der Atmosphre, Germany)

This study introduces a statistical-dynamical extra-
polation scheme (SDE) for nested regional climate
model simulations. The method is based on a
cluster analysis weather-type classification scheme.
Predictor is the 500~hPa geopotential height and
predictands are the regional surface temperature
and precipitation.
For the validation of the scheme a 30-year GCM
simulation of the Hadley Centre with a contin-
uously nested regional climate model (RCM) is
40
used. The basic features of the regional develop-
ment in Central Europe are quite well reproduced
by the SDE. A comparision of the SDE and the
direct RCM output shows only a deviation of
0.13~K for the decadel mean winter temperature
and 8~\% for the decadel mean winter precipitation
of the greater Alpine Region. The SDE estimation
of the interannual variability of the mean winter
precipitation shows an excellent reproduction for
the North Sea area and a slight underestimation for
the greater Alpine area.
The results of the SDE are two dimensional
available with the resolution of the nested regional
climate model. With the application of the SDE the
regional climate model has to be nested only for a
few years to get 30-year climatological statistics.
This render the possibility to obtain regional scale
information for the length of GCM experiments.
30-year GCM time-slice experiments with different
CO
2
concentrations could be compared within the
regional scale. The comparison and validation of
regional models with different model physics and
parametrizations is possible for a higher number of
model runs including the possibility to cover time
spreads of climatological periods (i.e. 30-years).

Stochastic generation of daily precipitation,
Aristita Busuioc
1
, Hans von Storch
2
and Rodica
Tomozeiu
1
(1 National Institute of Meteorology
and Hydrology, Bucharest, Romania; 2 Institute
of Hydrophysics, GKSS Research Center,
Geestacht, Germany)

The paper presents a stochastic model conditioned
by the large-scale climatic distributions to generate
daily precipitation amount. With such a model,
numerous statistically equivalent but in its details
different realization of long sequences of daily
rainfall can be generated. The model uses a first
order Markov chain combined with a downscaling
model based on the canonical correlation analysis
(CCA) to link the local precipitation with large-
circulation. Precipitation occurrence is described by
a two-state, first-order Markov chain. The pre-
cipitation either occurs or it does not (the two
states) and the conditional probability of pre-
cipitation occurrence depends only whether preci-
pitation occurred in the previous day. There are two
parameters describing the precipitation occurrence
process: the transition probabilities of a wet day
following a dry day and the probability of a wet day
following a wet day. Variation of precipitation
amount on wet days is characterized using gamma
distribution, which has two parameters: the shape
parameter and the scale parameter. Therefore the
stochastic precipitation generation model depends
on four parameters. To link the four parameters by
the large-scale circulation, represented by the sea
level pressure on the European scale, a regression
model based on the canonical correlation analysis
have been used. This model has been tested for the
winter (J anuary) daily precipitation amount at the
Bucharest station over 1901-1994. It was found
that, for the independent data set (1901-1950), two
model parameters (mean precipitation amount for
wet days and the probability of a wet day following
a wet day) are not very well estimated over 1901-
1930 interval. This result could be explained by the
fact that before 1931 these parameters describe
another precipitation regime (as urbanization effect)
or these parameters are no so strongly dependent on
the large-scale circulation. Testing of model for
other months (seasonally stratified) and other
station placed in same geographical conditions are
in progress.

Statistical Downscaling of Daily, Multi-Site Pre-
cipitation in Tropical and Sub-Tropical Climates,
Stephen P. Charles
1
,
Bryson C. Bates
1
, Stephen J.
Crimp
2
and James P. Hughes
3
(
1
CSIRO Land and
Water, Wembley, Australia;
2
Queensland Depart-
ment of Natural Resources, Indooroopilly, Austra-
lia;
3
Department of Biostatistics, University of
Washington, Seattle, USA)

Regional climate simulation is becoming increa-
singly important as the scientific and policy
community require realistic projections of the
potential regional impacts of climate change. Alt-
hough general circulation models (GCM) and limi-
ted area models (LAM) simulate large-scale atmos-
pheric processes reasonably well, they tend to over-
estimate the frequency and under-estimate the
intensity of daily precipitation, and thus fail to
reproduce the observed precipitation statistics at the
spatial and temporal scale required for regional
impacts assessment. Statistical downscaling can
overcome these limitations by relating regional-
or point-scale precipitation to large-scale atmos-
pheric fields. The nonhomogeneous hidden Markov
model (NHMM) is a statistical downscaling model
previously shown to reproduce daily precipitation
probabilities, spatial patterns in precipitation
occurrence, wet and dry spell length statistics, and
the probability distributions of precipitation
amounts at multiple sites in temperate midlatitude
regions (Pacific north-west USA and south-western
Australia).
Here, we apply the NHMM to tropical and sub-tro-
pical regions in Australia where precipitation is
generated by a much wider variety of mechanisms
including monsoon, coastal and upper troughs,
tropical cyclones, fronts, and localised convective
systems. The literature contains very few examples
of statistical downscaling in tropical regions. A
suite of observed atmospheric predictors (derived
from NCEP/NCAR Reanalysis data), encompassing
measures of atmospheric flow, moisture content,
and stability, provide NHMMs that reproduce the
observed summer and winter precipitation statistics
for tropical, sub-tropical continental, and sub-
tropical maritime regions in Queensland, Australia.
Future work will compare downscaled simulations
generated from the observed atmospheric predictors
with those generated from the corresponding LAM
atmospheric predictors.

41
Stochastic Weather Generator Model for Simu-
lation of Multiple Meteorological Time Series,
Ildik Dobi-Wantuch, Jnos Mika (Hungarian
Meteorological Service, Budapest, Hungary);
Lszl Szeidl (Department of Mathematics,
University of Pcs, Hungary)

This work presents the development of a daily,
stochastic weather generator point-model, which is
based on precipitation occurrences, as a driving
parameter. The model, simulating diurnal series of
maximum, mean and minimum temperature, preci-
pitation, sunshine duration, cloudiness, relative
humidity, wind speed and air pressure, is re-
commended for use in climate change impact
studies in Hungary.
The first problem, to solve, is the suitable modeling
of wet and dry sequences, since, as we are going to
discuss, some often used simulation approaches do
not appropriately work in our climate. Considering
the frequently occurring long dry periods in Hun-
gary, duration of sequences without measurable
precipitation is modeled by mixed Poisson and
geometric distributions.
Another important aspect is the preservation of the
first four moments and the correlation structure of
the empirical time series. In order to retain these
statistical features for all variables, we applied
various uni- and multi-variate methods.
To prepare the data vectors for further modeling,
first we standardized the series considering the
annual cycle of the means and standard deviations
by applying Fourier transformation. Next, a centra-
lization for conditional averages, depending on the
serial number of the given day within the dry or wet
series, is performed, since this dependence is also
significant in several respects.
Factor analysis and test of normality suggest that
three of the variables, namely maximum tempera-
ture, relative humidity and air pressure, can be
parallelly modeled as a weakly stationary, 3-dimen-
sional process which can be fairly well described
by an AR(1) model. However, the other six climate
elements have non-normal distributions. This
recognition drives us to accept a multivariate exten-
sion of the Johnson distribution. This technique
already matches the necessary statistics for the
given elements and climatic region.

Interdiurnal and interannual variability in sto-
chastic daily weather generator: Modelling and
the role in agricultural and hydrologic studies,
Martin Dubrovsky
1
, Josef Hejzlar
2
, Jan Kysely
3
,
Josef Buchtele
4
, Zdenek Zalud
5
, Daniel Marvan
6
(
1
Institute of Atmospheric Physics AS CR, Hradec
Kralove, Czech Republic (CR),
2
Hydrobiological
Institute AS CR and Faculty of Biological
Sciences USB, Ceske Budejovice, CR;
3
Institute of
Atmospheric Physics AS CR, Prague, CR;
4
Institute of Hydrodynamics AS CR, Prague, CR;
5
Mendel University of Agriculture and Forestry,
Brno, CR;
6
Faculty of Mathematics and Physics,
Charles University, Prague, CR)

Met&Roll is a WGEN-like four-variate (precipi-
tation, solar radiation, temperature maximum, tem-
perature minimum) stochastic daily weather
generator used to provide synthetic weather series
for models simulating growth of crops and forests
and hydrological regime in present and changed
climate conditions. The previous version of the
generator was found not to satisfactorily reproduce
interdiurnal and interannual variability of the
weather series. Two questions are raised in this
contribution: (1) What is an effect of these im-
perfections on results obtained by the models
(growth models, hydrological models) feeded by
the synthetic weather series? (2) How to improve
the generators performance?
In the first part, improvements of Met&Roll aiming
to better reproduction of statistical structure of the
weather series are introduced: (i) To better repro-
duce interdiurnal variability of the weather series,
correlations among solar radiation and daily ex-
treme temperatures are allowed to vary during a
year and the Markov chain of the third order (in-
stead of the first order) is used to model preci-
pitation occurrence. (ii) As the generator was found
to underestimate interannual variability of monthly
means, the underlying mathematical model was
split in two steps: In the first step, time series of
monthly means of the four variables is generated
using first-order four-variate autoregressive model.
In the second step, daily weather series is generated
using the monthly means produced in the first step.
Impact of the new features of the generator on the
statistical structure of synthetic daily weather series
will be demonstrated.
In the second part, effect of the interdiurnal and
interannual variability of daily weather series on
crop yields simulated by growth models (CERES
and WOFOST) and on stream hydrology and
stream water quality simulated by SAC-SMA
hydrological model will be shown.

Temperature predictability in the greater Mediter-
ranean area, R. Garca, E. Hernndez, T. Muoz,
P. Ribera* and L. Gimeno* (Dto Fsica de la
Tierra II, Universidad Complutense, Madrid
Spain; Dto Fsica Aplicada, Universidad de Vigo,
Campus de Orense, Orense, Spain)

Long-range statistical forecasting of annual and
seasonal average temperatures in an area centered
over the Mediterranean Sea has been considered
from two different approaches intended to forecast
annual or seasonal average temperature values se-
veral months in advance. The methods are
Optimate Climate Normals (OCN) and an extension
of it called Climate Normal Functions (CNF). Both
of them attempt to forecast temperatures from the
past behavior of its own time series. They have
been applied to a dataset of monthly averaged tem-
peratures from 151 observatories from Europe and
Africa distributed around the Mediterranean area,
covering the period 1955-1990.
This study makes use of annually and seasonally
averaged values. For every observatory, OCN and
42
FCN have been computed for each averaging
period. Next, cluster analysis has been applied to
classify similitudes in the predictability charac-
teristics. Three predictability patterns have been
identified: in the Northern area and 8 yr period has
been detected; the North Atlantic Oscillation seems
to be the most important influence. In the Eastern
area the predictability coefficients reach their
maximum at the longest average periods. Finally,
the aera closest to the Mediterranean shows a one
year persistence as the most characteristic feature.

A new validation scheme for the evaluation of cli-
mate model outputs, F.-W. Gerstengarbe, M.
Kcken, P.C. Werner (Potsdam-Institute for
Climate Impact Research, Potsdam, Germany)

Model validation plays an important role in the
development of climate models. The validation en-
compasses a simple optic comparison, the use of
generalising measurement figures (e.g. root mean
square error - RMSE) and the application of fully
developed statistical methods. The latter often re-
quire relatively high calculation time so that they
are rarely used for evaluation purposes. Never-
theless, the evaluation of complex correlations can
only be done with such complex methods. In order
to make progress here, a method is presented in the
following which is based on a multivariate pattern
recognition and provides measurement figures for
quality evaluation that can be easily interpreted and
whose spatial distribution and temporal develop-
ment serves to derive information on possible
sources of error.
The question that should be answered by the new
method is: How well can a model represent defined
complex structures of a climate regime?
In a first step the validation conditions are defined
and the selected parameters are grouped in patterns
for a reference field at each grid point of a model
area via n time steps by means of a cluster analysis
algorithm. It has then to be investigated how far the
same parameter combination of the simulated field,
i.e. the climate model output, reflects these patterns.
This is done by a distance measurement which
provides information about the assignment of the
grid points to the clusters of the reference field.
Thus, one receives a yes/no answer: the simulated
value either corresponds with the one of the
reference field (the grid point belongs to the same
cluster as in the analysis ) or the same grid points of
the reference field and the simulated field belong to
different cluster. In order to specify this yes/no
answer, each degree of deviation, standardised to
0;1 (0 no deviation, 1 largest deviation with a
statistical security of 95%) will be determined in
addition.
The effectiveness of the method is presented by a
comparison of the quality of different regional cli-
mate models.

The construction of multi-site and multi-variate
scenarios for Mediterranean regions based on
Monte Carlo simulations with a conditional wea-
ther generator, C.M. Goodess and J.P. Palutikof
(Climatic Research Unit, University of East
Anglia, Norwich, UK)

Empirical methods for downscaling GCM output
offer relatively simple and computationally effi-
cient means of constructing scenarios with a high
spatial and temporal resolution. However, rela-
tively few downscaling studies have addressed the
problem of constructing scenarios which are consis-
tent between sites and between variables on a day-
by-day basis. Here, a three-stage method for the
construction of self-consistent daily temperature
and rainfall scenarios which has been developed
and tested for two study areas in southeast Spain
and southern Italy is described.
First, daily rainfall scenarios were constructed for a
key site in each study area using a conditional
weather generator in which rainfall occurrence de-
pends on both the circulation type and on whether
the previous day was wet or dry. Each simulation
set consists of 1000 runs, i.e. a Monte Carlo
approach is taken. This has a number of advan-
tages:

a range of scenarios can be provided and the
level of uncertainty can be indicated, for
example by calculating quantile values;
ranked pairs can be used as climate scenarios,
thus taking some account of systematic errors
in the downscaling models and underlying
GCMs; and,
non-ranked pairs can be used in climate sen-
sitivity studies, for example to maximise the
changes in a particular season.

Second, daily rainfall scenarios for a network of
sites in each study area were constructed using an
analogue approach, based on a reference scenario
for each key site, i.e. by sampling from the ob-
served data conditional on the circulation type and
whether the key site was wet or dry.
Third, multi-site daily temperature scenarios were
constructed using a transfer function method, with
free atmosphere variables as predictors in multiple
regression models. The principle of consistency is
maintained by constructing transfer functions sepa-
rately for wet days and dry days. The latter infor-
mation was available from the multi-site rainfall
scenarios.
The final scenarios are considered more plausible
than the raw GCM changes. This approach is, how-
ever, affected by a number of problems which are
common to all empirical downscaling approaches
(i.e. stationarity, overdispersion, data availability
and the reliability of GCMs). A number of ways in
which the method could be further refined and
developed to reduce these problems are proposed.
43
Assimilation

Reconstructing large-scale variability from paleo-
climatic evidence by means of pattern nudging,
Martin Widmann
1
, Hans von Storch
1
, Ingo
Kirchner
2
(
1
GKSS Research Center, Geesthacht,
Germany,
2
Max-Planck-Institute for
Meteorology, Hamburg, Germany)

Pattern nudging is the second step of the newly
developed Data Assimilation Through Upscaling
and Nudging (DATUN) technique. The aim of
DATUN is to obtain a physically-based, best guess
for the large-scale states of the atmosphere during
the Late Holocene with annual temporal resolution.
DATUN uses statistical upscaling models to esti-
mate continental-scale and hemispheric climate va-
riability from climate proxy records. In a second
step, which is the subject of this presentation, the
large-scale variability of a GCM is nudged towards
these estimates.
Commonly used nudging procedures relax model
states towards a target field that is prescribed at
every gridpoint. However, this approach is proble-
matic in the context of paleoclimatic reconstruc-
tions, since the historic atmospheric states are only
incompletely known. To overcome this difficulty,
we developed a nudging method that is purely
defined in a pattern space. Only large-scale patterns
are nudged and the synoptic-scale variability is
therefore expected to be undamped but consistent
with the large-scale forcing. Pattern nudging has
been implemented in the ECHAM4 model and
tested using prescribed states of the Arctic Os-
cillation. An evaluation of these test runs will be
presented. This work is undertaken as part of the
German project KIHZ (Klima in Historischen
Zeiten, Climate in Historical Times).

Spatio-temporal patterns of Meteosat Satellite
Thermal Infrared Derived Convective Clouds over
Ethiopia as Described by PCA & MVC, Abebe
Yeshanew (Senior Meteorologist, National
Meteorological Services, Addis Ababa, Ethiopia)

Convective clouds with top temperature at -40 C
thresholds are derived from Meteosat satellite ther-
mal infrared channels over Ethiopia from J uly to
September fro the year 1988, 1989 and 1990. The
duration of convective activity (cold cloud duration,
CCD) maintaining its top cloud temperature at -40
C threshold are summed up to every dekad (8, 9, 10
or 11 days depending on the month considered).
The standardised values of the duration of convec-
tive activity is subjected to principal component
analysis to cull the most dominant and important
signals from the dataset. The first two principal
components are selected based on the break point
of the eigenvalues verse the eigenvector
components plot. The first two principal
components explained 83% of the total variance.
The most dominant spatial mode reveals a dipole
structure whereas the second depicts a tripole
spatial configuration. The two spatial modes of
principal component analysis are used in a
minimum variance cluster analysis (MVCA) to
delineate areas where CCD is homogeneous in
space. Here, the cubic clustering criterion (CCC) is
used to determine the number homogeneous region.
Based on CCC and MVC, four regions of convec-
tive activity regions are defined. The defined con-
vective regions show different temporal charac-
teristics of the duration of convective activities over
the country. The convective activity regime with
maximum CCD is located over western and central
parts of the country. Whereas, the region charac-
terised by minimum occurrence of convective
clouds entails northern, northeastern, southern,
southeastern parts of Ethiopia.

Skill of seasonal hindcasts as a function of the
ensemble size, F.W. Zwiers and V.V. Kharin
(Canadian Centre for Climate Modelling and
Analysis, Victoria, BC, Canada)

Forecast skill as a function of the ensemble size is
examined in a 24-member ensemble of northern
winter (DJ F) hindcasts produced with the second
generation general circulation model of the Cana-
dian Centre for Climate Modelling and Analysis.
These integrations are initialized from the NCEP
reanalyses at 6 hour intervals prior to the forecast
season. The sea surface temperatures that are ap-
plied as lower boundary conditions are predicted by
persisting the monthly mean anomaly observed
prior to the forecast period. The potential pre-
dictability that is attributed to lower boundary
forced variability is estimated.
The forecast skill in the first 2 weeks, which ori-
ginates predominately from the initial conditions, is
greatest for relatively small ensemble sizes. The
forecast skill increases monotonically with the
ensemble size in the rest of the season. The skill of
DJ F 500 hPa geopotential height hindcasts in the
Northern Hemisphere and in the Pacific/North
America sector improves substantially when the
ensemble size increases from 6 to 24. A statistical
skill improvement technique based on the Singular
Value Decomposition method is also more suc-
cessful for larger ensembles.


Determining extreme sets in multivariate distri-
butions, onika. Lakatos (Hungarian
Meteorological Service); I. Matyasovszky
(Department of Meteorology, Etvs Lornd
University, Budapest)

The term extremes is simply defined as events
having small probabilities at the tail(s) of an un-
derlying density function. Consider a multivariate
44
random variable with probability density function
f(x). For any 0<<1 an extreme set T is defined as

T f c :{ ; ( ) } x x < such that f d
T
( ) = x x ,
where c=c(). When the analytical form of f is
kownn the task to obtain T is straightforward.
Without knowing the form, which is a typical case,
an estimate of the density is needed. Therefore, a
procedure is developed to determine T using a
kernel type estimator for f. The methodology is
applied to air temperature and wetness data in order
to define critical dry and warm conditions.

The Probability Of An Extreme In A Correlated
Time-Space Domain, Albert R. Boehm

Consider an area with weather observations at a
limited number of points and over a limited period
of record. Questions: What is the probability during
a specified target period in the future that there will
be:
A: An extreme anywhere in the area?
B: A specified fraction or more will experience an
extreme?
C: An extreme will persist during some fraction of
the target period?
D: Extremes will exist beyond some penalty
function which could be complicated and is fitted to
specific requirements such as crop growth or
heating load.

Case 1: The area is homogeneous and correlation is
isotropic in space and stationary in time. This case
can be handled by fitting the probability of an
extreme at point and modeling correlation structure.
Questions A, B, C and some forms of D can be
answered with the CUB algorithm defined at the
5IMSC 1992. CUB depends on the eigenvalues of
the domain correlation matrix. More complicated
forms of D can be simulated.
Case 2: The area is heterogeneous and the
probability may have a seasonal variation, but cor-
relation is isotropic in space and stationary in time.
In effect, the spatial and time domain consists of
sub regions that differ in their extreme probability
but are correlated. Since the heterogeneous pro-
bability will be less than if all the domain were at
the highest probability but greater than if all the
domain were at the lowest probability, a weighted
mean can be used as an approximation. Since the
correlation is assumed known, a conditional pro-
bability of a lesser sob-domain being extreme given
that a greater sub-domain is not extreme can be
calculated.
Case 3. The domain is undergoing climate change
or fluctuation. This complicated case has an elegant
solution: Take into account all possible changes
(including no change) each weighed by the pro-
bability it will occur.

Climate extremes in Germany and its relation to
the objective weather type classification of the
German Weather Service, Peter Bissolli, Udo
Baum (Deutscher Wetterdienst, Offenbach,
Germany)

Climate extremes can be attributed to special fea-
tures of certain weather types. For the area of
Germany, such weather types can be described
using the objective weather type classification of
the German Weather Service. This classification
provides 40 standard weather types, but, by using
special indices, it is possible to define other special
extreme weather types too.

This presentation discusses the following aspects:
1. how to define climate extremes in Germany,
2. how to define extreme weather types,
3. how to find relations between extreme
weather types and climate extremes.

The definition of these extreme events is mainly
based on frequency distributions of different cli-
mate elements (daily temperature maximum and
minimum, daily wind speed maximum, daily
precipitation amount and daily snow depth) and
weather type indices (indices of advection, cyclo-
nality and humidity). In addition, the spatial varia-
bility of climate elements within Germany is taken
into account. Advantages and disadvantages of
these defined extremes are discussed, from a statis-
tical point of view as well as from a climatological-
practical one.
The relation between weather types and climate
extremes can be defined by probability distri-
butions. Considering that the weather type classi-
fication also provides forecasts of weather types,
the forecast skill of these probability distributions
can be estimated.

Statistical methods and case studies investigating
extreme climatology of the Carpathian Basin,
Judit Bartholy
1
, Laszlo Szeidl
2
(
1
Dept. of Meteo-
rology Eotvos Lorand University, Budapest Hun-
gary;
2
Dept. of Mathematics University of Pecs,
Hungary)

One of the most challenging topics of the global
change research is predicting future changes in
extreme climatological values. An overview of the
statistical methodology will be presented that can
provide scientific and reliable answer to the follo-
wing questions: how the extreme values and their
occurrences can be modelled, what kind of pro-
blems may occur investigating climatic fields of a
given meteorological parameter, etc. The presen-
tation will focus on the extremes of climate of the
Carpathian Basin. Monthly temperature data obser-
ved at 16 stations will be examined, furthermore,
precipitation measurements from 162 and 8 stations
located in Hungary will be used in monthly and
daily scale evaluation, respectively. In this region
the precipitation trends evaluated during last
decades seems to be inconsistent. Beyond extreme
45
value analysis of precipitation, variability and fluc-
tuation characteristics will be determined. Spatial
distributions will be mapped of various statistical
characteristics for all seasons. Time series will be
also analyzed separately as robust and extreme.
Distributions of extreme, robust and full time series
will be compared. Asymptotic approach will be
applied to precipitation and temperature time series
using Gumbel distribution model. Model para-
meters will be estimated by method of moments.
Return values will be determined for pre-set
periods, e.g., 10, 20, 50 years, which thresholds will
be exceeded once in average during that period.
More detailed investigations will be carried out for
the two sensitive regions of the Carpathian Basin,
namely, the drainage basin of the shallow Lake
Balaton and Sio, and the Hungarian Great Plain. In
the present climatic conditions the water balance of
these regions tends to be negative, and therefore it
is likely that the global change may affect them the
most.

Simulation Models

Simulation of multivariate time series of meteoro-
logical data, Asa Forsman, Claes Bjrklund and
Anders Grimvall (Department of Mathematics,
Linkping University, Linkping, Sweden)

A great variety of models used in the environmental
sciences are driven by meteorological data. This
implies that simulations involving synthetic mete-
orological inputs can play an important role in
several different contexts. First, such simulations
can be used to determine the joint distribution of an
arbitrary set of response variables. Secondly,
simulations followed by statistical analyses of the
response to different inputs can be used to elucidate
the dynamic behaviour of complex models. In the
present paper, we discuss how the statistical charac-
teristics of a given multivariate time series of me-
teorological data can be transferred to arbitrary
amounts of synthetic data. In particular, we de-
scribe a procedure involving the following steps: (i)
transformation of the marginal distribution of each
of the meteorological variables to a standardised
normal distribution; (ii) fitting of a vector auto-
regressive model to the transformed data; (iii)
simulation of new observations according to the
fitted auto-regressive model; (iv) transformation of
the data thus generated to synthetic meteorological
data. A nonparametric resampling method that
preserves the rank correlations of the different
meteorological variables is also described. In both
cases, seasonality is handled by dividing the ori-
ginal data sets into subsets representing the
different months of the year. An empirical data set
comprising thirty years of daily data regarding
temperature, precipitation, vapour pressure, wind
speed, and cloudiness was used to illustrate the
methods proposed. Examination of observed and
simulated data revealed only marginal differences
in the statistical characteristics of the two types of
data.

Modelling the probability of precipitation using
the Marcov chain MonteCarlo technique, Martin
Kappler (Fachbereich Statistik, Universitt Dort-
mund); Andreas Hense (Meteorologisches Institut
Universitt Bonn); Siegfried Schach (Fachbereich
Statistik Universitt Dortmund)

Although precipitation simulated in atmospheric
models is calculated deterministically, the inter-
actions between the rain generating processes in
space and time require a probabilistic interpre-
tations of precipitation forecasts. This is especially
true for high resolution weather forecast models
like the Lokal-Modell LM of the German Weather
Service DWD which has presently a horizontal
resolution of 7 km. We will show results for
forecasts of hourly accumulated rain fall from the
LM as input to a postprocessing model which
describes rainfall events as outcomes from Ber-
noulli trials with an unknown probability (PoP
probability of precipitation) to be estimated from
the data.
The postprocessing model is set up as a hierarchical
Bayesian network with three levels. The logit-trans-
formed Bernoulli probability (level 1) is para-
metrized as a normally distributed variable where
the expectation is parametrized (level 2) linearily in
terms of a-priori chosen variables like orographic
height, orographic gradient in wind direction and
the precipitation field in the neighborhood of the 1
hour backward trajectory. Lastly the regression
coefficients (level 3) are characterized by the
uninformative normal density function with expec-
tation zero and a large variance. The Bayesian
network and the specification of the above given
conditional and priori densities allows the calcul-
ation of the a-posterior density for the unknown
parameters conditioned on the data of simulated
precipitation. The parameter estimation is done by a
Gibbs sampling of the a-posteriori density with the
help of a Monte Carlo simulation (Markov chain
Monte Carlo). Data are takenonly from a quarter of
the model domain such that the remaining parts can
be used for validation. The estimated values for the
regression coefficients show that the gradient in
wind direction and the backward trajectories are the
most important external variables. Taking the Brier
skill score for the validation shows that the post-
processing model is clearly better than persistency
or a PoP estimated from a simple spatial average.

Reproduction of variance in statistical
downscaling: inflation vs. noise addition, Radan
Huth
1
, Jan Kysel
1,2
(
1
Institute of Atmospheric
Physics, Prague, Czech Republic;
2
Dept. of
Meteorology and Environment Protection,
Charles University, Prague, Czech Republic)

Statistical downscaling models can explain only a
part of the variance of the original (observed) time
46
series. Therefore, the variance of the downscaled
series is underestimated. Two methods of enhan-
cing the downscaled variance to become equal to
that observed are in common use: (i) inflation of
variance, consisting in multiplying downscaled
anomalies by the ratio of the observed and
downscaled standard deviations, and (ii) noise
addition, consisting in adding a random process
(noise) to the downscaled series. The variance
inflation appears to be a physically doubtful
concept as it assumes that all the local variability
originates from large-scale variability, which
apparently is not correct.
The comparison of the two approaches is conducted
for daily maximum temperatures in May to Sep-
tember and daily minimum temperatures in Novem-
ber to March at six central European stations for a
30-year period. The downscaling method used is a
stepwise multiple regression of 500 hPa height and
1000/500 hPa thickness gridpoint values over
Europe and adjacent parts of the Atlantic Ocean.
The random process added to the time series is a
white noise process. The evaluation is based on (i)
lag-1 autocorrelations, (ii) standard deviation of
day-to-day temperature changes, (iii) characteristics
of persisting extreme events (heat and cold waves),
and (iv) spatial correlations.
It is shown that the downscaling with variance
inflation reproduces persistence characteristics of
minimum temperature correctly, whereas overesti-
mates persistence of maximum temperature. The
addition of a white noise process results in a severe
underestimation of persistence (overestimation of
day-to-day variations) in both seasons. Similar
results apply to spatial correlations.
We argue that although the variance inflation is not
correct from a physical point of view, it results in
much less biased temporal and spatial autocorre-
lations than if a white noise process is added, and
therefore, inflation should be given preference if
one is concerned with temporal and spatial structure
of downscaled quantities.

Simulating meteorological data by block re-
sampling, Anders Nordgaard, Asa Forsman,
Anders Grimvall (Department of Mathematics,
Linkping University, Linkping, Sweden)

Validation of statistical models used to simplify
complex mechanistic models needs large-scale data
generation. In this paper we use a block resampling
method to generate ''new'' series of data from a long
series of multivariate meteorological observations.
The method is based on the matched-block
bootstrap method suggested by Carlstein et al
(1998) where blocks of data are resampled accor-
ding to a Markov chain. The transition probabilities
are computed using a Mahalanobis distance mea-
sure.

The Amazon climate: A multiscale and nonsta-
tionary processes, Guillermo O. Obregn, Carlos
Nobre (Centro de previso de Tempo e Estudos
Climticos - CPTEC / Instituto Nacional de
Pequisas Espaciais INPE, Cachoeira Paulista,
So Paulo Brazil)

The goal of this work is to examine the climatic
complexity of the western part of Amazon basin
related to both the ENOS and NAO phenomenon
based on monthly level of the Negro river at
Manacapur-Manaus (Q) from 1903 to 1999. The
Morlet wavelet transform (WT) and the probability
density function (PDF) are used as a tools. The WT
shows that on Amazon basin exist four regimes of
quasi-periodic oscillation. The 2-3 yr cycles and the
5 yr cycles are in agreement with the ENOS signal.
The quasi-decadal cycles appears linked to the
NAO, principally before 1940, and the quasi-bi-
decadal cycles, remaining any abrupt climatic
change in time, seems to be associated with both
the ENOS and NAO. The PDFs between Q vs SOI,
and Q vs NAO shown distinctive multimodatity
indicating the nonlinear response of Q to these
phenomenona or at least a weak direct relationship.

First indication of anthropogenic climate change
in local wave conditions in the North Sea, Arnt
Pfizenmayer (Institute for Coastal Research,
GKSS Research Centre; Geesthacht, Germany)

In the central North Sea we observe an increase in
the frequency of eastward propagating waves in the
recent four decades. To assess the significance of
this change the wave statistic for this century was
reconstructed with a statistical model. With a linear
multivariate technique (redundancy analysis)
monthly mean air pressure fields over the North
Atlantic and Western Europe were downscaled on
the intramonthly frequency of directional wave
propagation. When compared against this reference,
the recent change appears statistically significant at
the 5% level.
In order to investigate the reason for this local cli-
matic change, reconstruction was compared with
the downscaled results of a control and transient
GCM scenario (ECHAM4-OPYC3) and with re-
sults obtained in a high resolution time slice experi-
ment with increased concentrations of greenhouse
gases and aerosols. Both estimates are qualitatively
consistent with the changes observed in the last
decades. We suggest that the recent increase of
eastward propagation is a local manifestation of
anthropogenic global climate change.

Ensemble Predictions with a Limited-Area Meso-
scale Atmospheric Model - Implications for the
Interpretation of its Direct Model Output,
Susanne Theis, Andreas Hense (Meteorologisches
Institut der Universitt Bonn, Bonn, Germany);
Ulrich Damrath, Volker Renner (Deutscher
Wetterdienst, Offenbach, Germany)

47
The atmosphere is a chaotic system. In principle,
this aspect has always to be taken into account
whenever an atmospheric forecast is produced and
interpreted. One possibility of dealing with the
intrinsic uncertainty of the forecast is the generation
of ensemble predictions. The principle of ensemble
prediction is to forecast the impact of the uncer-
tainty of initial conditions on the uncertainty of the
forecast. This is usually done by running several
numerical forecasts from slightly different initial
states. Certainly, one could also imagine to vary
other parameters than the initial conditions, for
example the boundary conditions at the ground,
such as roughnesslength or albedo. This method is
especially interesting in the context of limited-area,
high-resolution atmospheric models which are nes-
ted within a global model.
The increase of forecast resolution is substantially
governed by the quantitative specification of the
boundary conditions at the ground. Many of these
boundary conditions cannot be quantified with
downright certainty. We have produced ensemble
predictions by slightly perturbing the rough-
nesslength and the orography of a limited-area
mesoscale weather forecast model (horizontal
resolution: 7km). The results reveal that small-scale
features of only a few gridboxes' size are not
deterministically predictable. This implies that
small-scale predictions by deterministic regional
climate models might be subject to a considerable
amount of uncertainty, as well. Whenever the direct
output of a deterministic forecast model contains a
non-negligible amount of stochastic information, it
must be interpreted in a statistical way. Ensemble
predictions, however, are not always realizable,
because they demand an enormous amount of com-
puter capacities. As a temporary solution, we have
developed a method which postprocesses the direct
model output statistically. The method requires the
data of one numerical simulation only. It will be
validated against experimentally conducted en-
semble predictions.

Modeling stochastic structure of local climate as
downscaling of GCM fields, 1,2Marina
Timofeyeva and 1Rachael Craig. (Department of
Geology, Kent State University, Kent, Ohio;
UCAR/NOAA/National Weather Service, Silver
Spring, Maryland)

The Kent State University Local Climate Model
(KSU LCM) is a statistical downscaling procedure
that relates a suite of dependent variables (stochas-
tic structure of local climate) using canonical
correlation analysis to a suite of independent varia-
bles (derived from insolation, a DEM and GCM
velocity, geopotential heights and SST). Our me-
thod differs from the other techniques of down-
scaling (e.g., Zorita and von Storch, 1999) as it uses
a semi-Lagrangian approach to represent the
regional climate controls at the point where climate
is modeled.
The stochastic structure of daily temperature is
represented by (1) parameters of the best fit of in-
solation to maximum daily temperature, (2) para-
meters of the third order autoregressive (AR3)
model fit to the residuals from the insolation model,
(3) variance of the residuals of the AR3 model; and,
(4) mean and variance of temperature range.
Stochastic structure of daily precipitation is repre-
sented by: (1) probability of wet, trace, and dry
days, and (2) parameters (mean and variance) of the
(log-normal) distribution assumed for daily preci-
pitation and conditioned on the estimated proba-
bility of wet days.
The KSU LCM suite of independent variables com-
prises surface, upwind, and downwind variables
computed using CCM3 boundary conditions. Sur-
face variables represent local topography and ra-
diation balance. Upwind variables are: radiation
balance air temperature and its meridional gradient,
average gradient in saturation vapor pressure,
speed, div and curl of velocity as well as orographic
controls. Downwind (30 km) variables represent
blocking effects.
The transfer function was calibrated (using 434
stations in an elevation- and climate-division-
stratified sample) and tested (using 1290 stations)
for climate of the western U.S. and Mexico using a
control run of CCM3 for boundary conditions.
This new model allows high temporal (one day)
and spatial (1km
2
) resolution solutions of climate
for much of western North America for any time
period for which CCM3 solutions are available, e.g.
6ka paleoclimate and future climate scenarios.
Similar solutions can be used for other GCMs when
calibrated with their control runs.

Numerical experimentation with regional atmos-
pheric models, Hans von Storch, Ralf Weisse
(Institute of Coastal Research, GKSS Research
Centre, Germany)

While the need for determining the climatic noise
with the help of multiple runs in numerical ex-
perimentation with global models is acknowledged
since the 1970s, regional atmospheric modelers in
the past have designed experiments usually as a
simple pair of "control" and "experimental" simu-
lation. The effect of the experimental modification
is then considered as being described by the
difference between the two simulations. This
approach assumes implicitly that the state of the
regional atmosphere in such models would be de-
termined by the boundary conditions, that no
"noise" would be generated on regional scales.
However, this assumption is invalid as is shown by
examples. For long-term differences the noise level
is indeed rather small, but for episodes large dif-
ferences in the synoptic configuration (e.g. with
respect to the speed of cyclones) intermittently
emerge.
The problem is demonstrated with the case of the
effect of an interactive ocean surface wave field on
the formation of extratropical storms. In previous
studies, this interaction was claimed to cause
storms to intensify (through increased latent heat
fluxes) or damped (through increased momentum
48
flux) cyclogenesis, but a ensemble simulation
indicates that the found effects are mostly random.

Decision Making and Detection of
Anthropogenic Climate Change

Exploiting the emerging signal of anthropogenic
climate change in probabilistic climate fore-
casting, Myles Allen (Space Science &
Technology Department, Rutherford Appleton
Laboratory, Chilton, Didcot, UK)

Significance of Anthropogenic Radiative Forcings
in Simulation of Temporal Pattern of Surface Air
Temperature in Monsoon Climate, Ajay Singh
(School of Management, Indian Institute of
Technology- Bombay, Powai, India)

Augmentations in greenhouse gas concentrations
since pre-industrial era have led to a positive
radiative forcing of climate, tending to warm the
surface and to produce other changes of climate.
Tropospheric aerosols have led to a negative direct
forcing and possibly also to a negative indirect
forcing of a similar magnitude. Primarily the
negative forcing is focused in particular regions and
subcontinental areas, it can have continental to
hemispheric scale effects on climate patterns. All
India Summer Monsoon plays very critical role in
many of socio-economic activities in the Indian
Subcontinent. Surface air temperature in the
monsoon climate is important for number of rea-
sons including its significance in agriculture,
industry and planning. An accurate and reliable
portrayal of change in the temperature due to cli-
mate change is therefore very important in the
perspective of impact of climate change on socio-
economic system and environmental system.
The ability of current generation of global climate
models, in their long-term simulations, to replicate
the observed atmospheric behavior on a wide range
of spatial and time scales provides support in
applying these models to the regional climate
change projections induced by anthropogenic
radiative forcings. The ability of a coupled atmos-
phere ocean climate model (ECHAM3 +LSG) at
T21 resolution to simulate the temporal pattern in
area-averaged annual mean surface air temperature
over Indian subcontinent in three different numeri-
cal experiments viz.; control, GHG and GHG plus
aerosol is examined. The significance of anthro-
pogenic radiative forcings in the simulation of
temporal pattern of the temperature has been exa-
mined. The model has superior skill in reproducing
the observed trend and interannual variability in
area-averaged annual mean surface air temperature
during the past century in GHG plus aerosol experi-
ment. The models ability to simulate the observed
pattern in the temperature over the area of interest
could be improved by better representation of
meso-scale forcings such as monsoon trough,
complex orography and parameterization scheme
for convection and land surface processes in the
model.

Decadal variations of Greenland precipitation
over the past 1400 years, K. K. Andersen and P. D.
Ditlevsen, (Department of Geophysics,
Copenhagen, Denmark)

In the accumulation zone of the Greenland ice sheet
the annual accumulation rate is registered in the firn
and subsequently in the ice. The annual layer
thickness may be identified through identification
of the annual cycle in the isotopic climate signal
and other annually varying parameters. The snow is
however subject to wind drifting which introduces
rather large 'glaciological noise' to the signal.
We developed a method to estimate the noise in
accumulation records derived from ice cores. This
method was then applied to the annually resolved
accumulation records from four independent deep
ice cores. In this way we have extracted a pre-
cipitation record for Greenland covering the period
from 555 AD to 1973 AD with decadal resolution.

Lessons taught from the Homogenization of 200
US First Order Stations, Auguste C. Boissonnade,
Ph.D.
1
and Larry Heitkemper
2
(
1
Risk Management
Solutions, Inc. Menlo Park, California, USA;
2
EarthSat Corporation, Washington D.C., USA)

Weather conditions have an enormous impact on
agricultural, energy, and indirectly retail
businesses. Study of historical weather
observations is one of the means for assessing the
likelihood of potential impact weather could have
on these businesses. The most common weather
index, Cooling degree-day or Heating degree-day,
used in assessing weather impact is derived from
the daily minimum and maximum temperature at a
given well-manned station over a 5 to 7-month
period. New financial instruments called weather
derivatives, based on the likely future behavior of
the weather index, have been developed and are
currently traded in financial markets for hedges
against the weather risk. All these instruments are
very dependent on historical and future
observations of weather data at given weather
stations.
Large volatility is present in such assessment of
weather risk due to the large uncertainty in pre-
dicting weather beyond a few days but also in the
uncertainty on the quality of recorded observations
for the purpose of weather risk assessment. His-
torical temperature data may contain missing or
erroneous values. Furthermore, changes in location
and/or instrumentation for most weather stations
have caused increase or decrease in temperature
observations. Such discontinuities can make his-
torical data misleading relative to how weather is
being recorded under present day-conditions.
A 1 1/2 year joint project between Risk Mana-
gement Solutions and Earth Satellite Corporation
49
was undertaken for homogenization of time series
for more than half of the available First and Second
Order stations in the US over the time period
(1950-1999). Results of the homogenization study
were reviewed with the local meteorologists in
charge.
This paper presents findings of this project, a
comparative analysis of several methods for de-
tecting and assessing the magnitude of discon-
tinuities and present conclusions regarding the
quality of available meta databases. In addition, a
summary of the impact of instrumentation changes
over the past three decades is presented. Read-
justment of climatological data for particular loca-
tions is also presented. Emphasis on the importance
and problems of homogenization of such time
series in weather risk is made. Suggestions for
further research with potential benefits in a better
assessment of weather risk are also outlined.

North Atlantic osccillation projection on regional
climate predictability, Roxana Bojariu and
Daniela Paliu (National Institute of Meteorology
and Hydrology, Bucharest, Romania)

NAO is the dominant mode of variability over the
Atlantic-European region. A number of studies
have shown that most of cooling in the northwest
Atlantic and the warming across Europe and over
Eurasia since the mid 1970s results from changes in
the NAO. However, a more detailed regional pro-
jection of NAO related climate fluctuations and
predictability is still needed.
In the present study, a NAO related behavior is
identified when the predictive potential of winter is
assessed using a multifield analog technique based
on large and local climate state vectors. The
hindcast experiment have been done for the interval
1962-1989 using seasonal means of temperature
and precipitation anomalies from 62 Romanian
stations, geopotential heights at 500 mb over the
Atlantic-European sector and sea surface tempera-
tures (SSTs) from the North Atlantic. The autumn
thermal anomalies in a large part of Romanian
territory are significantly well predicted from the
previous winter. A larger scale perspective of this
behavior is investigated by performing canonical
correlation analysis (CCA) for air surface tem-
perature over Europe, in November and sea level
pressure (SLP) over Northern Hemisphere in J a-
nuary. Temperature and SLP data for the interval
1961-2000 are extracted from NCEP re-analyses.
This source of autumn predictability over Ro-
manian regions seems to be related to the inter-
annual persistence of one NAO, more pronounced
in the latest decades of 20
th
century.
Many studies have shown that both phases of NAO
are associated with basin-wide changes in the
intensity and location of the North Atlantic jet
stream and storm track. Changes in the intensity
and location of North Atlantic jet stream may be
caused by processes such as changes in sea surface
temperature and/or in the extent of ice sheet at high
latitudes. The NAO physical mechanism involved
in the November/J anuary signals seems to be more
related to Arctic atmospheric processes such as 70
days oscillation identified in the geopotential height
at 500 mb level. The 70 days oscillation might be
triggered or locked up by the seasonal cycle in the
special context of orographic conditions and air-
land interaction over high latitudes areas.

What influences decision-makers in drought
prone areas when allocating water resources?
Bernard Cornlis (Dpartement de Gomatique
Universit de Lige)

At an operational level, computerised decision
support systems implement Bayesian statistics,
fuzzy logic, neural networks, genetic algorithms,
data based or model based management systems,
Nevertheless, these concepts barely leave the tech-
nical and scientific circles. When it comes to hu-
man decision makers, the results of these processes,
if they reach them, is just one more information
which can be taken into account or not. In the field
of climatology, just like in major environmental
issues, lack of information are important, and
induce a great uncertainty. The four techniques
used to appreciate uncertainties of consequences
are probabilities, imprecise probabilities,
possibilities and unknown possibilities. In this case
as well, not all the decision makers have access to
such systems or trust them.
This contribution can be seen as an attempt to for-
malise the way current decision makers deal with a
potential climatic event. Different types of private
and public decision makers, ranging from farmers
to Members of the European Parliament, were
asked during an interview to allocate water re-
sources to their territory of action. Through this
pragmatic exercise, several ideas were tested:
would the spatial characteristics of their territory
influence the repartition; would their decision
behaviour change under uncertainty; would the
integration in time modify their answer. After com-
pleting the allocation scenarios, they were asked to
comment their decisions. Besides, several neutral
questions relating to the use of water, to drought
and to their territory were asked. Analysis of this
survey shows that, instead of basing their decisions
on math models and measured info, lay decision
makers base their decisions on their moral values
and their representation of the issue.

Assessing uncertainty in anthropogenic signal
estimates using space-time and trend-based
optimal detection approaches, Nathan Gillett,
Gabi Hegerl, Myles Allen

The optimal fingerprinting' approach to the de-
tection of anthropogenic climate change involves
the regression of some observed diagnostic onto the
same diagnostic derived from perturbed inte-
grations of a GCM. Both time/space and trend-
based diagnostics have been used in published
studies. While results from both approaches ge-
50
nerally indicate that the influence of greenhouse
gases and sulphate aerosol is detectable in global
surface temperature over the past fifty years, they
have lead to different estimates of the amplitude of
these `signals'. The goal of the present work is to
understand these differences in results. A first step
is to transform those results that are based on
stepwise regression to multiregression. By elimi-
nating other differences in the analysis, relating to
the data used and filtering, we establish the relative
importance for detection of the linear trend and
residual components of the response to the forcings.
An additional reason for differences in results based
on different model simulations relates to the differ-
ing indirect sulphate aerosol response (that due to
changes in cloud properties) in the ECHAM4 and
HadCM3 models. By examining the spatial and
temporal characteristics of this response in each
model, we better characterise the causes of dif-
ferences between published results on the detection
of anthropogenic sulphate aerosols, and assess
which differences are sensitive to the detection ap-
proach used.

Applications of Statistical Climatology in Finance,
Stephen Jewson, Anders Brix (Risk Management
Solutions)

Statistical models of meteorological phenomena are
used widely in the fields of banking, insurance and
re-insurance to help estimate both climatological
and state-dependent financial risk. An overview is
given of methodologies used in this area, including
very large ensemble simulations, optimal combi-
nations of climatological and forecast information,
and final metric optimisation of forecasts.
A number of examples are given including:
1)Real-time hurricane loss forecasting: the need for
ensembles of size in excess of 10,000 and how
these ensembles can be created; how deterministic
forecasts can be used to condition these ensembles.
2)Extratropical wind-storm risk estimation: the
distribution of numbers of wind-storms and the
importance of seriality.
3)Weather derivative pricing methods: daily and in-
dex models for weather derivative pricing and the
pros and cons of different methods for incorpora-
ting forecasts.

Trends in northern hemisphere winter cyclone
activity based on a comparison of cyclone indices,
Christopher J. Paciorek (Department of Statistics,
Carnegie Mellon University); James S. Risbey
(Department of Engineering and Public Policy,
Carnegie Mellon University); Richard D. Rosen
(Atmospheric and Environmental Research, Inc.
Cambridge, Massachusetts); Valerie Ventura (De-
partment of Statistics, Carnegie Mellon
University)

We investigate northern hemisphere extratropical
winter cyclone activity using a variety of indices
calculated from the NCEP/NCAR reanalysis data.
We ask whether these indices provide similar mea-
sures of cyclone activity and what each index
suggests about trends in cyclone activity during
1949-1999. The indices are Eady growth rate at
500 hPa, temperature variance at 500 hPa, the 95
th

percentile of surface wind speeds, and counts of
both storms and intense storms. Eady growth,
temperature variance, and extreme winds are rea-
sonably well-correlated over the storm-track re-
gions and northern parts of North America and
Eurasia, but poorly correlated elsewhere. These
three indices also show moderately-strong corre-
lations with the two storm count indices over much
of the storm-track regions, but only when the storm
index is offset 10 degrees to the north of the other
index. We analyze simple linear regressions of the
indices against time using the False Discovery Rate
(FDR) procedure. Eady growth, temperature va-
riance, and extreme winds show significant in-
creases over the mid-latitude storm-track regions as
well as over parts of Eurasia, while temperature
variance shows decreases over the polar latitudes of
the North Pacific. Extreme winds at the sigma 995
level of the reanalysis model and geostrophic winds
calculated from sea level pressure show unex-
plained jumps in the 1950s that account for most of
the increases seen in the wind index. Poisson
regressions of the two storm count indices against
time are suggestive of increases over the oceanic
storm-tracks, but it is difficult to distinguish these
increases from statistical chance. Given the diffi-
culties with the extreme winds index, it is difficult
to draw conclusions about human impacts from our
results, apart from the evidence for increased
cyclone activity over the storm-track regions.
Results from Global Climate Models (GCMs) that
compare control and enhanced CO2 scenarios
suggest poleward and downstream shifts in the
storm-tracks with increased CO2. The models also
suggest the possibility of fewer but stronger
cyclones. The Eady growth and temperature va-
riance indices provide some evidence for increased
downstream cyclone activity over the storm-track
regions. The storm count indices do not suggest a
decrease in the number of events, but provide weak
evidence for an increase in more intense events
over the storm-tracks.

Signal analysis of the atmospheric mean tem-
perature 500/1000 hPa north of 55 N between
1949 and 1994, Heiko Paeth, Andreas Hense
(Meteorologisches Institut, Universitt Bonn,
Bonn, Germany)

The lower tropospheric mean temperature 500/1000
hPa is examined in the northern hemisphere high
latitude region north of 55 ffi N with regard to a
climate change signal due to anthropogenic climate
forcing as a supplement to previous studies which
concentrated on near surface temperatures. An
observational data set of the German Weather
Service (DWD) is compared with several model
simulations including different scenarios of green-
house gas and sulphate aerosol forcing derived
51
from the two recent versions of the coupled climate
model in Hamburg, ECHAM3/LSG and ECHAM4-
/OPYC. The signal analysis is based on the optimal
fingerprint method by Hasselmann (1979,1993),
which supplies a detection variable with optimal
signal-to-noise ratio. The natural variability mea-
sures are derived from the corresponding long-term
control experiments. From 1970 onward, we find
high trend pattern analogies between the observa-
tional data and the greenhouse-gas induced model
simulations. The fingerprint of this common tem-
perature signal consists of a predominate warming
with maximum over Siberia and a weak cooling
over the North Atlantic reaching an estimated sig-
nificance level of about 99 %. The additional for-
cing by sulphate aerosols decreases the correlation
of this climate change simulation with the ob-
servations. The natural variability constitutes about
30 % of the conforming trend patterns. The signal-
to-noise ratio is best over the oceans while the
tropospheric temperatures over the land masses are
contaminated by strong noise. There is evidence
that the temperature signal is independent of the
initial conditions and the model version.

Expert Assessment of Uncertainties in Detection
and Attribution of Climate Change, James Risbey
(University of Utrecht, Department of Science,
Technology and Society, Utrecht, The
Netherlands)

The issue of whether, and to what extent, human-
induced increases in greenhouse gases have caused
climate changes is contentious. Studies of this issue
have focused first on `detecting' climate change
against the background of natural variability, and
further on àttributing' any detected signal to in-
creases in greenhouse gases or other possible cau-
ses. Communication of the science and role of
uncertainties on this issue has been hindered by the
lack of explicit formal approaches for making over-
all conclusions on detection and attribution. We
have developed a protocol to quantify uncertainties
in each component of the detection and attribution
process and to provide a structured way to make
overall conclusions. Here we present results on
detection and attribution of climate change from
analysis of expert judgements obtained using the
protocol, highlighting areas of convergence and
divergence among experts. We show that, among a
broad spectrum of experts, probabilities of detec-
tion are uniformly very high (>95%) for some lines
of evidence and high (>85%) for others. Further,
there is near consensus that greenhouse forcing is
responsible for more than half the warming in
global mean temperature this century. For the other
lines of evidence examined, greenhouse forcing
makes smaller contributions with more spread a-
mong expert assessments.
A Maximum Entropy Approach for Extracting
Forecast Probabilities from Finite Ensembles,
M.S. Roulston
1
, L.A. Smith
1
and C. Ziehmann
2
(
1
Mathematical Institute, Oxford University, UK;
2
Institute of Physics, Potsdam University, Ger-
many)

A maximum entropy method is presented for in-
corporating prior information, in the form of a
climatological probability distribution, into a pro-
babilistic forecast generated by an ensemble predic-
tion system. The approach is to find the probability
distribution that is as close as possible to the clima-
tology, in the relative entropy sense, while being
consistent with the ensemble according to the chi-
squared test. The result is that low-probability
events, too rare to be expected in an N member en-
semble, are assigned finite probabilities based on
climatology.
An information theoretic skill score is also intro-
duced. This ìgnorance' score is equal to the number
bits that are required to describe truth in a data
compression scheme designed around the probabi-
listic forecast. It is also related to the rate of return
achieved by a gambler betting optimally (propor-
tionally) on possible outcomes.
The maximum entropy method is illustrated in se-
veral chaotic systems. Both perfect and imperfect
ensembles are considered, the latter both under per-
fect and under imperfect models. It is shown that
this method can reduce the variability in forecast
skill obtained when ensembles are small. The
average forecast skill, as reflected by Brier Score or
by the ignorance is largely unaffected. The re-
duction in variance is achieved by systematically
improving the worst forecasts. This reduction
occurs both with perfect and with imperfect en-
sembles.

Coupled sea ice and atmosphere variability in the
Weddell Sea, Antarctica, Silvia A. Venegas
(Danish Center for Earth System Science
(DCESS), Niels Bohr Institute for Astronomy,
Physics and Geophysics, University of
Copenhagen, Denmark); Mark R. Drinkwater
(Oceans/Sea-Ice Unit, ESTEC Earth Sciences Di-
vision (APP-FSO), Noordwijk, The Netherlands)

A frequency-domain singular value decomposition
is performed on 20 years (1979-1998) of satellite
observations of sea ice concentration and drift
together with sea level pressure reanalysis data in
the Weddell Sea, Antarctica, in order to determine
the dominant timescales of variability in the region.
The interannual timescale (periods of 3-4 years) is
found to dominate the ice and atmospheric fluc-
tuations. This variability is a regional manifestation
of the Antarctic Circumpolar Wave in the Weddell
sector of the Southern Ocean.
Interannual variations of sea ice concentration and
drift are driven largely by anomalous atmospheric
patterns, which alternatively enhance or reduce the
meridional wind circulation to the east of the
Antarctic peninsula. Thermodynamic processes (ra-
52
pid ice growth in open water areas), dynamic pro-
cesses (ice ridging/rafting due to ice convergence)
and geographic constraints all play significant roles
in the formation of summer ice concentration
anomalies in the southern Weddell Sea every 3-4
years. These ice anomalies are exported northwards
out of the Weddell basin approximately one year
later, and then advected eastwards before diffusing
in the ice margin.
A low-frequency signal (presumably a quasi-deca-
dal oscillation) is also detected, although its signi-
ficance is fairly speculative in a 20-yr record. Ice
variability on this timescale results from a change
in the shape of the Weddell gyre circulation which
occurred around 1990, likely associated with large-
scale variations of the ocean-atmosphere system.

A Multi-Index Bayesian Approach to Climate
Change Detection and Attribution, Klaus
Hasselmann, Reiner Schnur (Max Planck In-
stitute of Meteorology, Hamburg, Germany)

In the conventional optimal fingerprinting approach
to detection and attribution (D&A) of anthropo-
genic or other externally forced climate change, the
observations are projected onto a low-dimensional
phase space spanned by one or a few expected cli-
mate change signal patterns which are rotated away
from high natural climate noise. A detection test is
performed by assessing whether the null hypothesis
that an observed value in this reduced space is with-
in the interval of natural variability can be rejected
at a certain confidence level or not. This test is most
powerful in a one-dimensional sub-space. For the
attribution of a detected change one tests whether
the detected climate change is consistent with the
predicted signals of alternative candidate mecha-
nisms. The number of signal patterns must be kept
small to achieve statistical significance. The un-
certainties of the predicted signals enter only in the
attribution test.
Here, an alternative, Bayesian approach is presen-
ted which leads to a basically different optimal fil-
tering strategy. One considers the impact of evi-
dence (i.e. observed data) on the subjectively
defined prior probability that the hypothesis of an
externally forced climate change signal is true. The
resulting posterior probability depends on the ratio
of the likelihoods of observing the data for the com-
plementary cases that the hypothesis is true or false.
There exists no longer a formal separation between
detection and attribution, and the statistical un-
cer-tainties of the predicted signal play an equally
important role as those associated with natural va-
riability. There is also no need to project onto a
strongly reduced sub-space spanned only by the
signal patterns of the hypotheses being tested.
Nonetheless, a technique is presented to weight the
observed data such that the impact of the evidence
on the posterior probability is maximized. This
leads also to a dimensional reduction, but one
which is normally less severe.
Results are presented for the simultaneous appli-
cation of the Bayesian D&A framework to several
climate change indices. The approach is particularly
appropriate for indices whose statistical uncertainty
can be estimated only subjectively, such as sea le-
vel, sea ice, glaciers, extremes, etc. As example we
have chosen indices based on annual means and the
annual cycle of surface temperature and precipi-
tation and the diurnal cycle of temperature, as well
as ocean heat content. The results are compared
with the conventional D&A approach.

Changes in statistical properties of daily
precipitation in transient climate change
experiments with a coupled atmosphere-ocean
GCM, V.A. Semenov and L. Bengtsson (Max
Planck Institute for Meteorology, Hamburg,
Germany)

Statistical properties of daily precipitation distri-
bution are analyzed using data of the coupled at-
mosphere-ocean ECHAM4/OPYC3 general circu-
lation model transient climate change experiments.
Two different climate forcing mechanisms are
applied. In the GHG experiment the forcing is due
to the changing concentrations of the greenhouse
gases only. In the GSDIO experiment in addition to
the GHG the direct and indirect effects of the
anthropogenic sulfur aerosol and changes in the
tropospheric ozone are taken into account. Statis-
tical properties of daily precipitation for different
seasons are described by parameters of a simple
statistical model based on the gamma distribution.
The efficiency of the gamma distribution model is
estimated. Global trends of the mean precipitation,
precipitation intensity, wet day probability and the
gamma distribution parameters are analyzed for the
present (1950-2000) and future (2000-2050) cli-
mates.
Particular attention is paid to the extreme preci-
pitation events. Simulated characteristics for the
present climate are compared to the available ob-
servations. For the future climate the differences of
the hydrological cycle and daily precipitation distri-
bution changes due to different climate forcing
mechanisms are investigated.

Near Two Centuries Monsoon Rainfall Variations
across India: Reconstructed Past and Predicted
10-year Future, N.A. Sontakke (Indian Institute of
Tropical Meteorology, Pashan, India)

Long period rainfall reconstructions are vital for
numerous purposes such as climate variability and
prediction, cross comparision with climatological
proxies and so on. Owing to high spatial variability,
the representativeness of the summer monsoon
rainfall for the entire country is limited. In order to
develop an effective system for monsoon rainfall
studies the country has been divided into six zones
i.e. North West India, (NWI), North Central India
(NCI), North East india (NEI), West Peninsular
india (WPI), East Peninsular India (EPI) and South
Peninsular India (SPI). Empirical Orthogonal Func-
tion (EOF) and Cluster Analysis of ubiquitous 306
53
raingauges has been carried out for this classifi-
cation. By applying an objective technique 'selec-
ting a subset of few gauges whose mean showed the
highest CC with the all gauges mean series' total
116 optimum number of gauges have been fixed for
real time updating of the different zonal series and
reconstruction from 1871. The different zonal series
are extended as far as possible backward prior to
1871 by applying the objective technique on the
limited meager available gauges. The NEI and EPI
series are extended back to 1848, the NWI to 1844,
the NCI to 1842, the WPI to 1817 and SPI to 1813.
The all-India series from 1871 onwards is prepared
from the are-weighted mean of the six zones and
well agrees with the series prepared with all the
gauges. The series is extended back to 1813 by
objective technique on the avalilable gauges. The
examination of these series up to 1999 in detail
reveals interesting results in relation to the chan
pattern of monsoon circulation in the present
century with respect to global warming.
The low frequency smoothed series up to 1999
obtained through binomial low-pass filters plus the
series from 2000-2009, filled with the long term
mean is subjected to Variable Harmonic Analysis
(VHA). VHA is a modification of classical har-
monic analysis. An objective technique is applied
to select few harmonics whose linear combination
showed a correlation coefficient of ~0.85 with the
smoothed series. The rainfall amount generated
from the selected harmonics predicts the rainfall
fluctuations from 2000-2009. The rainfall series
actual and predicted from the earliest to 2009 will
be presented.

Statistical detection of the anthropogenic green-
house signal in observed global and regional cli-
mate data fields, T. Staeger, J. Grieser, C.-D.
Schnwiese (Institut fr Meteorologie und
Geophysik, J.W. Goethe Universitt Frankfurt
a.M.)

This study deals with the analysis of climate ob-
servational data of the last 100 years and their
attribution to anthropogenic and natural forcings.
The data include annual, seasonal and monthly
means (totals) of surface air temperature, sea level
pressure and precipitation. The spatial resolution
reaches from global means over area means on a
global scale (only temperature and pressure data) to
grid points on the European scale.
First an EOF analysis of the data fields is per-
formed. Thereafter the PC timeseries are used as
target variables in a sucsessive/multiple regression
analysis, where the following forcings are used as
potential regressors: Greenhouse gases (GHG),
tropospheric sulfate aerosol, explosive volcanism,
solar activity, ENSO and NAO.
For each forcing and target variable tempo-spatial
fields of signal amplitudes and signal-to-noise
ratios are obtained. The unexplained variance was
split up into structured and unstructured compo-
nents, the latter being an estimation for climate
noise.
It can be seen, that the total explained and the
GHG-attributed variance is maximum in case of
global averaged temperature data and is reduced in
less spatial averaged and other than temperature
data (and is minimal in case of European pre-
cipitation).
The detection of the anthropogenic greenhouse sig-
nal suceeds in the global mean temperature data in
1973 with a probability of 99.9%. In European data
NAO is dominant. It is an open question whether
there exists a GHG-NAO coupling on the European
scale, which aggravates the detection of the GHG
signal here. On the other hand the detection of the
GHG signal suceeds much better in the area mean
data on the global scale. In 42 out of 72 areas the
probability for a climate change exceeds 95%.
Solar forcing reveals to be weak in most cases.

Trends in the Diurnal Temperature Range in the
CCCma Coupled Model, Daithi A. Stone, Andrew
J. Weaver (School of Earth and Ocean Sciences,
University of Victoria, Canada)

Recent analyses of the global surface temperature
record have revealed that late 20th century in-
creases have been characterised by the daily mini-
mum temperature increasing at a faster rate than the
daily maximum temperature. These differential
trends have resulted in a decrease in the diurnal
temperature range (DTR), the magnitude of which
is comparable to the increase in mean temperature.
Despite the importance of these trends in the con-
text of climate change, they have not yet been
extensively examined in numerical simulations of
coupled general circulation models due to technical
limitations. The Canadian Centre for Climate Mo-
del-ling and Analysis first generation coupled gene-
ral circulation model has recently been integrated
under a variety of anthropogenic forcing scenarios.
An examination of the temporal trends in the DTR
in these integrations, their spatial characteristics,
their similarity to observations, and their relation to
trends in clouds, specific humidity, aerosols, and
radiative forcing, will be presented.

Detection of the causes of recent trends in
radiosonde temperatures, P.W.Thorne
1
,
P.D.Jones
1
, S.F.B.Tett
2
, M.R.Allen
3
, D.E.Parker
2
,
T.D.Davies
1
, and T.J.Osborn
1
(
1
Climatic Research
Unit, UEA, Norwich, UK.;
2
Hadley Centre for
Climate Prediction & Research, Meteorological
Office, Bracknell, Berkshire, UK;
3
Rutherford
Appleton Laboratory, Chilton, Didcot, UK.)

Previous climate change detection and attribution
studies have considered either changes in surface
temperatures, or zonally averaged radiosonde tem-
peratures. We contend that in so treating the radio-
sonde temperatures these latter studies have ignored
valuable information, as the observed temperature
changes are zonally heterogeneous. In considering
the full fields we hypothesise that the power to
discriminate between competing forcing mecha-
54
nisms will be enhanced. Here initial results are pre-
sented from an optimal detection study that in-
cludes information on longitudinal as well latitu-
dinal variations in upper air temperatures. This
study uses the Hadley Centre's globally gridded
HadRT radiosonde dataset and compares this to two
versions of the Hadley Centre's GCM, HadCM2
and HadCM3. A number of input diagnostics are
utilised in our analysis. We comment upon which
signals are detected and implications for intra- and
inter-model consistency considerations arising from
our analysis.

Signal Analysis and Detection Studies Using
Neural Networks, Andreas Walter (Institute of
Meteorology and Geophysics, Johann-Wolfgang
Goethe University, Frankfurt, Germany)

The climate system is considered to operate in a
complex, non-linear way. The higher the complexi-
ty of a system is, the more complicated statistical
investigations of such a system are. Because of
their capability to model multiple non-linear input-
output relations Neural Network Models (NNM)
are an appropriate tool for investigations of the cli-
mate system. We apply the most widely used NNM
architecture, the Backpropagation Network (BPN),
with and without an additional Momentum-Term to
analyze the dynamics of the climate system on a
global, hemispheric and regional scale and compare
these results to those carried out by a traditional
multiple linear regression (MLR). The NNM is able
to explain up to 88% of variance and the residuals
can be treated as noise. The extracted natural
(Volcanism, El Ni~no, solar activity) and anthro-
pogenic (CO2, Sulfate and combined influence)
signals show reasonable time structures and ampli-
tudes, e.g. BPN global: CO2: +1.11 K, SO4--: -0.36
K, combined CO2 & SO4--: +0.79 K. In a second
step these modeling techniques are applied to the
EOF-transformed representation of global spatio-
temporal temperature anomalies and signal analysis
are perfomed on this sacale, too. Using the signal-
to-noise ratio a detection variable is defined and it
is shown that NNM find a significant higher
anthropogenic influence in larger areas than MLR
does.

Potentially Predictable Patterns, Xiaogu Zheng
1

and Carsten S. Frederiksen
2
(1 National Institute
of Water and Atmospheric Research, Wellington,
New Zealand; 2Bureau of Meteorology Research
Centre, Melbourne, Australia)

A modified EOF analysis is proposed for studying
long-range potentially predictable patterns of
meteorological seasonal mean fields, that is, the
patterns arising from slowly varying external for-
cings and slowly varying internal dynamics. Using
500-hPa heights over the North Pacific/North
America region, it is demonstrated that the spatial
patterns derived by the modified EOF analysis are
more closely related to slowly varying external for-
cings and slowly varying internal dynamics, and
therefore are more potentially predictable than
those derived by the conventional EOF analysis, ro-
tations of EOFs or low-pass filtering. The proposed
modified EOF analysis is particularly suitable for
analysing regional atmospheric fields in the
extratropics.

Extratropical cyclone variability in the Northern
Hemisphere winter from the NCEP/NCAR Reana-
lysis data, Olga Zolina, Sergey Gulev, Sergey
Grigoriev (P.P.Shirshov Institut of Oceanology,
RAS, Moscow)

The winter climatology of Northern Hemisphere
cyclone activity was derived from 6-hourly NCEP/-
NCAR Reanalysis data for the period from 1958 to
1999, using software which provides a better
accuracy of cyclone identification in comparison to
numerical tracking schemes. Cyclone characte-
ristics over the Kuroshio and Gulfstream are very
different than those over the continental North
America and the Arctic. Analysis of Northern
Hemisphere cyclones shows secular and decadal-
scale changes in cyclone frequency, intensity, life
time and deepening rates. The western Pacific and
Atlantic are characterized by the increase in
cyclone intensity and deepening during 42-year
period, although the eastern Pacific and continental
North America demonstrate opposite tendencie s in
most of cyclone characteristics. There is an increase
of the number of cyclones in the Arctic and in the
western Pacific and a downward tendency over the
Gulfstream and subpolar Pacific. Decadal scale
variability in the cyclone activity over the Atlantic
and Pacific exhibits the south-north dipole-like
patterns. Atlantic and Pacific cyclone activity
associated with NAO and PNA is analysed.
Atlantic cyclone frequency demonstrates a high
correlation with NAO and reflects the NAO shift in
the mid 1970s, associated with considerable
changes in European storm tracks. The PNA is
largely linked to the eastern Pacific cyclone
frequencies, and controls cyclone activity over the
Gulf region and North American coast during the
last two decades. Assessment of th e accuracy of
the results and comparison with those derived using
numerical algorithms, shows that biases inherent in
numerical procedures, are not negligible.

55
Poster contributions to the 8 Inter-
national Meeting on Statsitical
Climatology

An Explanation for the Multidecadal, Winter
Rainfall Decline Over South-Western Australia,
Bryson C. Bates
1
, Stephen P. Charles
1
, James P.
Hughes
2
, and Edward P. Campbell
3
(
1
CSIRO
Land and Water, Wembley, WA 6913, Australia;
2
Department of Biostatistics, University of
Washington, Seattle WA 98195, USA;
3
CSIRO
Mathematical and Information Sciences,
Wembley, WA 6913, Australia)

There has been a decline in 'winter' (MayOctober)
rainfall over south-western Australia (SWA) since
about the middle of the 20
th
century. Since the mid
1970s, this decline has had important consequences
for water resources management and agriculture
and forestry. The links between the rainfall decline
and changes in atmospheric circulation are inves-
tigated using the NCEP-NCAR reanalysis dataset
that spans 195898, observed daily rainfall series
for 30 sites across the region, and a nonhomo-
geneous hidden Markov model (NHMM) fitted to
the period 197892 and validated for 195877. The
NHMM defines stochastic conditional relationships
that link daily multi-site precipitation to a discrete
set of weather states (patterns). The transition from
one state to another is conditioned on a relatively
small number of atmospheric variables (predictors)
derived from the reanalysis dataset.
Our analysis consisted of three parts: (1) the use of
a nonparametric jump detection algorithm for i-
dentifying any spurious jumps in the predictor
series; (2) comparisons of the predictor series and
NHMM weather state sequence for 1958 to mid
70s with those for mid70s to 1998; and (3) a sen-
sitivity analysis to investigate whether the rainfall
decline since the mid70s can be attributed to the
behaviour of a single predictor.
Our results indicate that: (1) there are no spurious
jumps in the reanalysis data used; (2) the atmos-
pheric predictor series for the mid70s to 1998 are
quite different to those for 1958 to the mid70s; (3)
there is very strong evidence of changes in the
mean probability of occurrence of two out of six
weather states; and (4) the changes in weather state
frequency, and hence rainfall, are due to a com-
bination of changes in several atmospheric varia-
bles rather than any one predictor. Possible mecha-
nisms underlying the change in atmospheric circu-
lation will be discussed.

Statistical Downscaling to Monthly Precipitation
in the Northwestern of South America (Colom-
bian Region), Nestor R. Bernal S. (Nacional
University of Colombia)

In climate change studies the global circulation
models of Atmosphere (GCMAs) enable one to
simulate the global climate, with the field variables
being represented on a grid ponits 300 km apart.
One particular interest concerns the simulation of
possible changes in rainfall and surface air tem-
perature due to an asummed increase of greenhouse
gases. However , the models yield the climatic
proyections on grid points that in most cases do not
correspond to the sities of major interest. To
achieve local estimates of the climatological va-
riables, methods like the one known as statistical
downscaling are applied. In this article we show a
case in point related to monthly precipitation by
applying canonical correlation analysis (CCA) to
some regions in the Northwestern of South America
(Colombian Region).

Better Quality of StatisTical Downscaling Models
Without Preceding PC-Analyses For Predictands?
Example: Rainfall Assessments for Namibia U.
Beyer (Institute of Geography, University of
Wuerzburg, Wuerzburg)

This contribution presents results of regional rain-
fall assessments for Namibia under conditions of
man-made enhanced greenhouse warming, obtained
in the context of the DFG-post-graduate program
J oint Geoscientific Research in Africa. Relations
between large-scale atmospheric circulation and
Namibian summer-rainfall are established by multi-
ple regression analyses. For this purpose PCs of
geopotential heights (300,500,1000hPa) and month-
ly rainfall data of Namibia (84 stations) are linked
by stepwise multiple regression analyses for every
station or alternatively every PC of the station data,
on a monthly base (November March) during a
30-year calibration period. After verifying these
statistical relations in an independent period, sta-
tions with regression models of sufficient quality
(significant correlations r>0.4 between observed
and modelled data) are selected and used to assess
local rainfall for greenhouse gas-scenarios from
simulated ECHAM4tr-T42 geopotential heights
data. In case of preceding PCA for rainfall data no
selection is possible all PCs have to be included
to calculate future rainfall. Information about model
quality exists, but can not be used to get higher
reliability in this downscaling approach. Indepen-
dent of the procedure, results for climate conditions
of increasing CO
2
-concentrations compared to a
reference period (1961-90) show a light increase of
rainfall in parts of Namibia for December, only
small changes in J anuary and significant increases
in February especially in the eastern regions, but a
distinct decrease in March all over the country.
These findings point to an intensified, more accen-
tuated rainy season, however the amount of rainfall
remain more or less the same under conditions of
enhanced greenhouse warming.

56
Multiple Non-linear Statistical Model for Esti-
mating Mean Maximum Urban Heat Island in a
Medium-Sized Town Szeged, Hungary, Zsolt
Bottyn, Jnos Unger (Department of Climatology
and Landscape Ecology, University of Szeged,
Szeged, Hungary)

This study examines the spatial and quantitative
influence of urban factors on the surface air tem-
perature field of the medium-sized city of Szeged,
Hungary using of mobile measurements under
different weather conditions between March 1999
and February 2000. This city of about 178 000 is
situated in a low, flat flood plain along the Tisza
River.
Efforts have been concentrated on the determi-
nation of spatial distribution of mean maximum ur-
ban heat island (UHI) intensity and modelling it in
space.
During our work we have been determined correct,
existing relationship between rural parameters and
mean max. UHI intensity by means of multiple
non-linear regression analisys.
The results indicate strong relationship exists be-
tween urban thermal excess and land-use features
such as built-up ratio, gradient of built-up ratio,
sky-view factor and distance from citys centre as
well. On the other hand the role of water surface is
negligible in this case.
The model is based on three non-linear regression
equations which are calculated in the studied peri-
ods distinguishing heating and non-heating seasons
within the one-year period in the studied area.
Thus our model is very useful because the predicted
seasonally spatial distribution of mean max. UHI
intensity field differences less than 0.2 C on
average from the measured UHI field .
In the interest of practical application on the basis
of our previous results - we have been developed
this model to predict seasonally mean max. UHI in-
tensity field considering only the geographical po-
sition within the studied area.
The reliability of this model is studying connection
with another city which has similar invariable urban
parameters, like Szeged.

Application of factor analysis for quantification of
climate forcing processes in the Baltic Sea region,
Arunas Bukantis Dept. Of Hydrology and Clima-
tology of Vilnius University, Vilnius Lithuania)

The article represent a discussion of methods based
on factor analysis (FA) and used for identification
of climate-forming factors and quantitative eva-
luation of their importance. The work was per-
formed using the data from seven meteorological
stations in the eastern part of the Baltic Sea region:
average monthly values of different meteorological
parameters (average maximal and average minimal
values of air temperature T1 and T2, atmos-
pheric pressure P, wind speed in the height 10 m
v, monthly precipitation Q, duration of sun-
shine S and partial pressure of water vapour e).
FA has revealed that the initial variables joined into
two main groups meteorological complexes,
whereas the variability of meteorological para-
meters and their mutual relations throughout the
year are determined by two main factors: advection
factor joining T1, T2 and e variables and baric-
radiational - P, S, Q. The portion of dispersion
formed by the both factors in the variability of
meteorological elements make up on the average
70-85% of their total dispersion. In cold seasons
(November-February) the advection factor is the
dominating one, forming 44-54% of the total
dispersion, whereas the importance of the second
factor is 1,5-2 times smaller. In the rest of the year
the influence of baric-radiational factor increases
(forms 30-40% of dispersion in variables) and
almost equals the importance of advection factor.
The remaining 15-25% of dispersion in meteo-
rological elements are, presumably, formed by local
meso- and microfactors. The both meteorological
complexes, as it has turned out during ortho-
gonalization of factors in FA, exist almost inde-
pendent of each other (in cold seasons in parti-
cular), i.e., it is possible to distinguish two most
important types of climate-forming processes in the
eastern part of the Baltic Sea region. One of them is
advection process related with the input of various
air masses whose features are best reflected by air
temperature and humidity. The other process
mostly takes place within one air mass: air mass
transformation, vertical mixing, formation of clouds
and related precipitation and input of solar
radiation. Thus, FA is not confined only to
systematisation of input data. It is helpful in
objectivization of climate-forming peculiarities and,
in this particular case - in confirmation of the
theory about climate-forming processes in the
eastern in the eastern part of the Baltic Sea region
and quantitative evaluation of the importance of the
climate-forming processes themselves.

A New Approach of Interpolation and Prediction
of Climate Time Series, Hong-Xing Cao, Xiao-
Jing Jia (Chinese Academy of meteorological
Sciences, Beijing ,China)

By use of an observed data series a new time series
modeling has been proposed. Firstly, a nonlinear
ordinary differential equation can be retrieved from
the data series based on a criterion of bilateral
difference forecasts, secondly taking the retrieved
differential equation as a dynamic kernal, a model,
which is called data-based mechanistic self-memo-
ry model (abbreviated as DAMSM), can be estab-
lished on basis of the self-memorization principle.
The DAMSM can be applied to an temporal
interpolation of climatic observed data in default of
an observation and to a climate prediction.
Modeling of the inter-annual forecast of the preci-
pitation in summer over Yangtze Delta was carried
out, By utilizing the seasonal precipitation for 1951
to 1998, the retrieved differential equation was
obtained, which is

57
Dx/dt=-0.638x(t-1)+0.569x(t-3)-0.512x(t-
4)+0.281x(t-5)+0.334x(t-7) +0.277x(t-5)2

The hindcast of the DAMSM of the summer
precipitation over the Yangtze Delta was computed
,its RMS ( root mean square) error is 75.3 mm
which is better than that by the autoregressive
(AR) model.
Besides, the models of the aviation transport and
sunspot were established too ,the prediction accu-
racies are all quite satisfactory.

Wind Field Retrieval From Scatterometer Data,
Dan Cornford, David J. Evans and Ian T. Nabney

It is well known that radar backscatter can be used
to estimate the near surface wind vector over the
ocean. In recent years several satellites have been
deployed carrying scatterometers to provide data
for retrieval of wind fields and subsequent
assimilation into numerical weather prediction
models. This paper addresses the question of
whether there is sufficient information in the
scatterometer observations alone to uniquely deter-
mine the "true" surface wind vector field. This is
difficult because the retrieved local wind vectors
have several ambiguous solutions for a given
scatterometer observation. We use a Bayesian
approach to combine the local scatterometer
observations with a global wind field prior. The
prior model is a Gaussian process with parameters
estimated from several years of numerical weather
prediction model data. We show how direct local
inverse models, based on mixture density networks,
can be used efficiently in the retrieval procedure.
The posterior distribution of the wind field has high
dimension, with several distinct modes. We show
how to sample from these distributions and illus-
trate the performance of several sampling algo-
rithms on toy problems. Many of the existing algo-
rithms are shown to perform badly on the toy
problems, because the modes of the posterior are
well separated in the high dimensional space. We
show that in most cases the posterior distribution of
the wind field has two dominant modes one of
which is typically close to the numerical weather
prediction model wind field (which is not used at
any point in the retrieval procedure). We also
discuss the effect of increasing the number of local
observations considered in the retrieval, which may
improve the chance of retrieving the "true" wind
field.

Homogenization and Quality Control in State
Space Models, Thomas K. Friedli, Evi
Schuepbach, Tania Steiner, Jrg R. Hsler
(Department of Mathematical Statistics and
Actuarial Sciences, University of Berne, Bern,
Switzerland)

Long-term surface ozone records may provide
essential information on changes in lower tropos-
pheric ozone over time. The homogeneity of such
records is commonly assessed by analysing the
difference series of neighbouring stations with
statistical tests such as the Buishand, Alexan-
dersson or Easterling-Peterson test. These approa-
ches generally suffer from the lack of readily
available nearby reference stations and hence, these
test procedures were often applied to original
surface ozone records without differencing with a
nearby reference station.
Using surface ozone data at J ungfraujoch (J FJ,
3580 m asl) and Zugspitze (ZUG, 2960 m asl) in
the Alps, we show that high-elevation surface o-
zone records display significant serial correlations
which may seriously affect the absolute and relative
homogeneity testing. Hence, we review some
existing parametric and nonparametric
homogeneity tests and present their performance on
simulated surface ozone records, with inclusion of
multiple outliers and level shifts.
It is evidenced that none of these tests perform
satisfactory, neither for absolute nor for relative
homogeneity testing.
Structural models in state space form provide a
suitable and sufficiently flexible statistical frame-
work for the unified analysis and homogeneity tes-
ting of time series records. Using surface ozone
data at J FJ and ZUG, we show how to estimate, test
and validate structural models in state space form
including multiple structural breaks. The per-
formance of this modelling approach is assessed on
simulated surface ozone series with inclusion of
multiple outliers and level shifts.

Statistical Analysis of Modern Climate Changes
observed in the West and East Antarctic,
Jagovkina, S. V.
1
, Lagun, V. E.
2
(
1
Main Geophy-
sical Observatory, St.-Petersburg, Russia;
2
Arctic
and Antarctic Research Institute, St.-Petersburg,
Russia)

The statistical analysis of temperature, pressure
fields, precipitation and wind speed is executed for
Russian Antarctic stations located in the East
(Mirny) and in the West (Bellingshausen) Hemis-
pheres. This analysis was made on the base of
multiyear current registered meteorological data for
all period of instrumental observations (1956-2000
for Mirny and 1968-2000 for Bellingshausen).
Mirny Observatory is situated on the coast of Cape
Davis at a small protrusion of Mirny Peninsula.
Mirny Observatory is the main base of Russian
studies in the Antarctic. The Bellingshausen station
is situated on the Fildes Peninsula in the south -
western part of King-George Island, being a part of
the South Shetland Isles.
The tendencies of interannual climatic changes are
identified with account of annual, seasonal, month-
ly and synoptic variations of considered parameters.
The traditional methods of regressive and corre-
lative analyses as well as methods of stochastic
analysis of the ensemble of several periodically
correlated random processes are used for inter-
annual oscillation description and low-frequency
trend determination. Obtained results are compared
58
with NCAR/NCEP Reanalysis (1958-1998) data set
and hemispheric temperature anomaly time series.
Statistical features of climatic parameters variation
in Western and Eastern parts of Southern
Hemisphere are demonstrated. Possible causes of
determined climatic tendencies are discussed in
connection with available data about cloudness
regime, eddies synoptic activity and sea ice dyna-
mics.

Statistical Forecasts of Summertime South
American Stationary Systems (SASSY), Charles
Jones
1
and Leila M. V. de Carvalhol
2
(1 Institute
for Computational Earth System Sciences
(ICESS), University of California, Santa Barbara,
CA; 2 Dept. of Atmospheric Sciences, University
of So Paulo, Brazil)

The South Atlantic Convergence Zone (SACZ) is
subjectively described as a convective region exten-
ding from the Amazon basin, passing over
southeast Brazil and running towards the South
Atlantic. The SACZ is characterized by a wide
range of temporal variability from synoptic to intra-
seasonal and interannual scales. Our recent study
has objectively identified the occurrence of South
American Stationary Systems (SASSY) during
austral summer. SASSY events are characterized by
persistence greater than two days and large geo-
graphic extension over southeast Brazil. This pre-
sentation will discuss a statistical method to fore-
cast the large-scale feature of convective activity in
the SASSY region. Daily Outgoing Longwave
Radiation (OLR) from the period J anuary 1979
through December 1999 is used to develop and
validate the statistical model. Examples of real-time
forecasts will also be discussed.

Using of cluster analysis in processing of corre-
lation coefficients, Stanislava Kliegrov, Ladislav
Metelka (Czech Hydrometeorological Institut,
Hradec Krlov Svobodn Dvory, Czech
Republic)

The satisfactory predictors are looked for the long-
term (seasonal) statistical forecasts for the Czech
Republic territory. The relationships among sets of
temperature (monthly value) in the Czech Republic
and sets of SST (Sea Surface Temperature, in grid
points 2x2) in part of the Atlantic were described
by the help of correlation coefficients. For simpli-
fication the cluster analysis was used in next pro-
cessing, when the regions of the ocean with similar
behaviour of significant correlation coefficients
were found. J ust possibility of using of K-means
Clustering Method, that moves objects among
k clusters with the goal (1) minimize variability
within clusters, and (2) maximize variability among
clusters, was explored in the paper. The objects, in
our case, were sets of 12 correlation cofficients
among set of monthly temperature in the Czech
Republic (for one concrete month) and sets of
monthly values of SST (stepwise for the same
month, for preceding month, for more preceding
month, and so on, to month about 11 months pre-
ceding the concrete month) for every grid point.
Cluster analysis was computed for every our
monthly temperature set separately and results
were illustrated in maps.
For some one of these month was number of
clusters clear, for some one was determination of
the number more problematic. J ust analysis of the
polemic cases is main aim of the report.

A new look at the temporal structure of a time
series, Ken-Chung Ko (Department of Geography,
National Kaohsiung Normal University,
Kaohsiung, Taiwan)

This study introduces a method in analyzing the
temporal structure of quasi-periodic oscillations.
The idea of the moving spectra will be adopted to
study the temporal variation of a time series. The
objective of this study is to look at different aspects
of the spectral methods in order to detect the
periodicity that could be hidden in regular spectra.
Results show dramatic changes after applying
normalization procedures and statistical tests to the
moving spectra. Four idealized cases tested by the
moving spectral method reveal that the moving
spectral method can transform the time series into
the temporal spectral structure.

Extreme value analysis applied on precipitation
time series in the Czech Republic, Kveton, V.
(Czech Hydrometeorological Institute, Prague,
Czech Republic)

Yearly extremes of 1- to 7-days precipitation totals
both from 8 long time series and 50 shorter time
series (1961-1998) in the Czech Republic were ana-
lysed. Different observation period and different
estimation method were used. Results are com-
pared. Obtained experiences are discussed.

Statistical climatology and interannual variability
of the extratropical cyclone parameters in the
global atmosphere, Victor E. Lagun (Air-Sea
Interaction Department, Arctic and Antarctic
Research Institute, St. Petersburg, Russia)

The calculation of the global distribution and of the
time variability of cyclone synoptic eddies para-
meters, including the space and time frequencies
estimations, the depth, the linear size, the area, the
velocity, the trajectory variability is made on the
base of twice - daily mean sea level pressure and
geopotential fields extracted from NCEP data sets
for 1980-1998 period. The original methods for
statistical analysis of meteorological fields were
used to define the position of cyclones and anti-
cyclones on the sphere. These numerical methods
are based on traditional definition of synoptic
eddies boundaries as external closed isobars. The
estimations of mean monthly, seasonal, annual and
59
decadal synoptic climatology parameters for both
hemispheres were carried out.
The analysis of different algorithms used for para-
meters estimations in synoptic statistical climato-
logy is executed. The obtained results are compared
with the known estimations for hemispheric para-
meters distributions of synoptic climatology and the
differences in results due to including of small and
mesoscale disturbances parameters in the statistics
of extratropical cyclones are discussed. The tenden-
cy of the extratropical cyclones time variability
under global warming conditions is estimated. The
results demonstrate that the number of mature
extratropical cyclones is increasing during the
current period of global warming, and it is the
important indicator of modern climate change. The
possible reasons of synoptic eddy activity increa-
sing are discussed.

Spatial Organization of Decadal and Bidecadal
Rainfall Fluctuation on Southern North America
and Southern South America, Omar Abel Lucero,
Norma C. Rodriguez (Instituto Nacional del Agua
and National University of Cordoba, Cordoba,
Argentina)

The spatial organization of decadal and bidecadal
components (fluctuations) of annual rainfall is iden-
tified in this research for two regions: 1) southern
South America, and 2) southern North America
(conterminous USA, southeastern Canada and
northern and central Mexico). Findings indicate that
these decadal and bidecadal components have high-
ly-coherent wave-like spatial organization. Two
types of organization of decadal and bidecadal
components of annual rainfall were identified: a
train of propagating fluctuations, and quasi-stan-
ding fluctuations. For decadal components, such
patterns alternate in time. A widespread change in
the spatial organization of decadal component of
annual rainfall took place simultaneously in both
continents in 1932. The bidecadal component is
organized as standing fluctuations in southern
North America; and as travelling fluctuations in
southern South America The spatial pattern of de-
cadal fluctuations of annual rainfall has 12- and 13-
year cycles, and the spatial pattern of bidecadal
fluctuations has predominantly 21- and 22-year cy-
cles.

Interrelationship between Interdecadal Fluctua-
tions in the Southern Oscillation and the Sea Sur-
face Temperature of the Tropical Pacific, Omar
Abel Lucero, Norma C. Rodriguez (Centro de la
Region Semirida (INA), and National University
of Cordoba.Cordoba, Argentina)

Atmospheric fluctuations at decadal and interde-
cadal timescales on the Southern Oscillation and in
the sea surface temperature of the tropical Pacific
are the focus of this research. Analyses are based
on the SOI (Southern Oscillation Index), and on
Wright's SSTI (The SSTI is defined by the average
anomaly of SST on a large part of eastern and cen-
tral Tropical Pacific.) Analyses were carried out
using the integral wavelet transform. This paper
documents 1) The existence of a large amplitude
modulation of the Southern Oscillation (AMSO). 2)
A change in the relative phase between SOI and
SSTI fluctuations at the decadal and bidecadal
timescale, which occurred in mid-1930s.
Signals in the band of wavelet-period smaller than
10 years do not show indication of climate change
throughout the period of study (almost a century-
long data record). The controlling process on the
variability of decadal component of SOI and SSTI
during that period is AMSO (amplitude modulation
of the Southern Oscillation). AMSO produces
maximum amplitude of decadal components of
SOI and SSTI in 1913-1917. Thereafter, these com-
ponents have decreasing amplitude until about
1957. In early-1960s, decadal component of SOI
and SSTI signals starts to amplify. AMSO has a
major influence on the observed values of SOI and
SSTI. The state of AMSO, and contributions from
decadal, bidecadal, and tridecadal component of
SOI and SSTI, account for major changes observed
in this century in the SOI, and in the SSTI.

Variability of annual pressure and precipitation
on Iberian Peninsula, M Dolores Manso Orgaz*
and Liliana Caramelo** (*Depart. of Physics.
University of Aveiro, Portugal; ** Depart. of
Physics University of Trs-os-Montes and Alto
Douro . Vila Real)

In this study we investigate the space and temporal
variation of the annual precipitation and the annual
pressure (at the sea-level), on the Iberian peninsula
(IP) in 40 stations of 39 years. Using the technique
of empirical orthogonal functions (EOFs) to obtain
a regional precipitation and regional pressure. In
order to obtain the structure of temporal variations
we are using to spectral analysis of the principal
components (PCs). We analysed the relationship
between the observed time series and the following
teleconnection indices: Oscillation of the North
Atlantic (Gibraltar extension NAO-G and Ponta
Delgada extension NAO-A) .We found two patterns
for the annual pressures accounts for 86% of total
variance; and four for the annual total precipitation
account for 74% of the total variance. These
leading PCs are associated with the following
patterns: NAO-G, NAO-A, respectively. The iso-
baric gradient and the spectra show coherence of
the first EOF of the annual pressure and of the
annual total precipitation with the trimester of
winter NAO-G. For the reconstruction of the series
is provided a statistical model in function of the
more significant oscillations.

60
Reproduction of dry and wet Periods by Stochastic
Daily Precipitation Generators, D. Marvan
1
, M.
Dubrovsky
2
(
1
Faculty of Math. and Physics,
Charles University, Prague, Czech Republic;
2
Institute of Atmospheric Physics, Hradec
Kralove, Czech Republic)

Stochastic models used to generate the series of wet
and dry days in daily weather generators are mostly
based on Markov chains (MCs). Since the first-
order MCs cannot satisfactorily reproduce the
dis-tribution of the lengths of wet and dry spells (in
addition, other weather characteristics, such as
occurrence of hot and cold periods, derived from
variables generated conditionally on the preci-
pitation occurrence may be distorted too), the alter-
native approaches are employed. These include
higher-order MCs, hybrid-order Markov chains and
serial approach models. The present contribution
aims to assess capability of several models (higher
order Markov chains, hybrid-order Markov chains
and serial approach model) to reproduce selected
statistical characteristics precipitation occurrence
series. The stress will be put on reproducing the
long wet and dry spells. The fit of the synthetic vs.
observed weather series is evaluated in terms of (i)
summary measures (Wilcoxon, chi-square and
Kolmogorov-Smirnov tests), (ii) maximum length
of wet/dry spells, and (iii) frequency of occurrence
of extremely long dry/wet spells.
In the second part of the contribution, the accuracy
of the generators parameters estimated from the
learning (observed) data will be assessed. This
exercise is motivated by the problem, whether it is
better to use simpler model (e.g., first order Markov
chain) with less parameters (and greater accuracy)
being derived from the learning data, or to use more
sophisticated model with more parameters, which
are, however, derived with lower accuracy from the
learning sample. The combined effect of the size of
the learning sample, type of the model and detail-
ness of the representation of the annual cycle of the
generator's parameters (one value per year/sea-
sons/month) on the generators' performance will be
therefore assessed.
The study is supported by Grant Agency of the
Czech Rep., contract 205/97/P159 and by Grant A-
gency of Academy of Sciences of CR, contract
A3042903.

Semi-parametric statistical approaches for the
analysis of climatological data, J. Mateu
1
, C.
Alvarez2, T. Sanfeliu
2
, M.M. Jordan
3
(
1
Department of Mathematics;
2
Department of
Experimental Sciences, Universitat Jaume I, E-
12071 Castellon, Spain;
3
Department of Experi-
mental Sciences, Universitat Miguel Hernandez,
Elche, Spain)

An understanding of the ozone behaviour near
surface layers is essential for a study of pollution
surfaces. Controlling possible sources of air pollu-
tion has become a relevant goal in our increasingly
degraded environment. Moreover, concentrations of
atmospheric trace gases involved in forming ozone
and nitrogen oxides change rapidly with several
meteorological variables such as wind speed and
direction, temperature, humidity and solar
radiation. Thus, good statistical models which can
handle with the intrinsic variability of
climatological and/or meteorological data are
clearly needed.
In this paper we set up the foundamentals to
propose, build and further develop statistical mo-
dels based on semi-parametric approaches. Particu-
larly, a mixed formulation involving a non-parame-
tric component based on locally weighted regres-
sion procedures which accounts for a possible
deterministic trend, and a parametric component re-
pre-senting the purely temporal random variation is
proposed.
The data set analyzed comes from an area with ty-
pical mediterranean environmental conditions, so
that our conclusions can be easily generalized to
general mediterranean polluted environments.

Statistical Methods in climatology: an application
to the longest precipitation series in Portugal,
Solange Mendona Leite (Geophysical Centre of
the University of Lisbon, Portugal)

Instrumental climatic data are limited to the period
from mid-twentieth century for the most regions in
Iberian Peninsula, with a few data sets going back
to late 19th century or early 20th century. Portugal,
however, has one high-quality instrumental weather
observation series going back to the middle 19th
century.
This weather station was selected by an exhaustive
search of documentation regarding instrumentation
and observational practices. Statistical tests of the
reliability of observing practices also have been
applied in the selection process, as well as tests to
detect inhomogeneities. The data set used is the
most reliable available for monitoring precipitation
trends during the last century and a half. With the
availability of this long time series it is now possi-
ble to conduct much more detailed analyses. In this
report we summarise these efforts using the weather
station of Instituto Geofsico Infante D. Lus, in
Lisbon. A trend analysis over the period 1856-1999
indicates a temporal pattern of precipitation variabi-
lity. The statistical characteristics of precipitation
monthly totals are analysed. Maximum entropy
spectral analysis is applied to quantify the relative
importance of decadal variability. An autorregres-
sive model of order 45 is fitted to the 144-year
precipitation series and the agreement between
observed and estimated values is judged. The value
of this model simulation as a source of regional
climate scenarios should be considered in the con-
text of other uncertainties in the global climate
change scenarios which are used to drive them.
These include the inherent variability of the climate
system and the differences in sensitivity and in
patterns of response between climatic elements.

61
Spatial patterns of the five-day rainfall all over
North-Western Iberian peninsula, M. Andrs
(Atmospheric Physics Department, University of
Salamanca, Spain); Clemente Toms
(Atmospheric Physics Department, University of
Salamanca, Spain); Fernando de Pablo (Atmos-
pheric Physics Department, University of Sala-
manca, Spain); Solange Mendona Leite
(Geophysical Centre of the University of Lisbon,
Portugal)

Precipitation is the climatic element which presents
a higher spatial and temporal variability in the
Iberian Peninsula. While the analyses of the tem-
poral variability of the precipitation data sets is
quite easy, the spatial variability is much more
difficult to identify and to interpret, and even more
when orography makes rainfall distributions even
more complex. In this report we present a regiona-
lization of Castilla y Len, a region located in
north-western Spain, using 5-day rainfall data sets.
Daily values were available for 119 weather
stations, during a 6-year period, from 1990 to 1995,
considering the rainfall period between September
and May. The region presents a characteristic
orography which affects the final results. Principal
Component Analysis (VARIMAX rotation) has
been applied as an adequate method of regio-
nalization. Six components were identified as statis-
tically significant, explaining 82.6% of the total
variance of the field. In order to identify which
synoptic systems are related to each component,
those cases in which the score was positive only in
one of the components (while the remaining were
negative) were selected. Thereafter, the synoptic
situations associated with that 5-day group were
analysed, looking for the main one and obtaining
common patterns for each one of the components.
We have also tried to identify the associated
systems which affect the region when there were
several components with positive values for a 5-day
group.

The effects of large scale phenomena on local cli-
mate variability over Romanian territory, Daniela
Mihaela Paliu, Roxana Bojariu, (National In-
stitute of Meteorology, Hydrology and Water Ma-
nagement, Bucharest, Romania)

This paper aims to identify the effects of large scale
variability modes such as North Atlantic Oscillation
(NAO), East Atlantic Pattern (EA), East Atlantic/-
West Russia Pattern (EA/WR), Scandinavian
Pattern (SCA) and Polar/Eurasia Pattern (POL) on
the occurrence of extreme temperature and precipi-
tation anomalies over Romanian territory. The large
scale phenomena are generally responsible for ex-
tended range predictability. However, the complex
orographical characteristics of Romanian area due
to the presence of Carpathian mountains which are
standing against certain types of large scale
circulation and modifies the relation between large
scale phenomena and local variability and predic-
tability.
Data sets consist of observed monthly temperature
and precipitation amounts from 62 Romanian
stations and indices of NAO, EA, EA/WR, SCA,
POL taken from NCEP data. The analyzed time
interval is 1961-1996. Statistical significant rela-
tionships between large and local scales are
identified mapping the correlation coefficients be-
tween the large scale indices and local thermal and
precipitation anomalies. Also, composite maps of
precipitation and temperature anomalies over Ro-
mania territory have been derived depending on the
opposite phases of the large-scale phenomena. The
large/local scale relationships with predictive po-
tential can be further used to develop a regionally-
orientated methodology for extreme event risks
assessment .

Neural networks to predict atmospheric profiles,
Nazario D. Ramirez-Beltran (Department of
Industrial Engineering, University of Puerto Rico,
Mayaguez); Jorge Gonzalez (Department of Me-
chanical Engineering, University of Puerto Rico,
Mayaguez); Felipe Chong Rengifo (Department of
Industrial Engineering, University of Puerto Rico,
Mayaguez)

A methodology to predict climatic variables based
on radiosonde and satellite observations is de-
veloped. The proposed methodology will contribute
to mitigate the existing of very limited radiosonde
data for the Caribbean. Satellite and radiosonde
data from 1986 to 1999 were used to create the cli-
mate information required for a competitive neural
network to identify the climatic observations
associated to a typical day. A feedforward neural
network was also designed to model the rela-
tionships among the inputs and outputs of an
atmospheric dynamic system. Meteorological varia-
bles from two consecutive mandatory pressure
levels were used to train a feedforward neural net-
work for predicting the following variables
geopotential height, air temperature, relative
humidity and wind components in the Caribbean
basin. The designed scheme has the capability of
predicting five climatic variables from the sea
surface up to 17 km of elevation. The developed
neural network methodology was successfully
tested by using satellite and radiosonde
observations for the year 2000. Climatic
observations and neural network predictions were
very close indicating that the proposed
methodology is a reliable tool for predicting atmos-
pheric profiles.

Mid-term prediction of the daily maximum/mini-
mum temperatures in South Korea using Dynamic
Linear Models, Keon Tae Sohn (Pusan National
University, Korea); Jeong Hyung Lee (Dong-A
University, Korea); Byungsoo Kim (Inje
University, Korea)

The mid-term(from +1 day to +10 day) prediction
of the daily maximum/minimum temperatures in
62
South Korea was performed using the Dynmic
Linear Models(DLM). Observations, which are 6-
hour-interval data (including the daily maxi-
mum/minimum temperatures) from September 2 in
1999 to May 31 in 2000, and the simulated outputs
of GDAPS(Global Data Assimilation and Predic-
tion System) used at Korea Meteorological Ad-
ministration, which are upto 240 hour-prediction,
are used as training data of DLM. The DLM with a
prior distribution consists of two equations and
intial distributions; (1) state equation:
Y(t)=F(t)m(t)+v(t), (2) m(t)=G(t)M(t-1)+w(t), (3)
m(0) is a normal-variate with mean a(0) and
variance C(0) where two independent random
variables v(t) and w(t) have normal distributions
with mean 0 and variances V(t) and W(t), respec-
tively. In addition, V(t) has an inverse Gamma dis-
tribution, with two parameters n(0) and s(0), as a
prior. Y(t) is an observation at time t, F(t) consists
of input vaiables and m(t) means a dynamic coef-
ficient vector. For estimating W(t) and m(t) dyna-
mically, the discount factor and the well-known
Kalman filtering are used.
For the selection of optimal DLM, some procedures
are considered as follows; (1) to select initial va-
lues, a(0), C(0), n(0) and s(0), randomly. (2) to de-
termine the transition matrix G(t), (3) to choose the
components of F(t), (4) to find an optimal DLM
under the selected F(t), with changing the value of
discount factor(from 0.01 to 1.00), (5) to determine
the optimal DLM among the all F(t)'s, (6) to
perform the sensitivity analysis for the effect of
initial values, and (7) interval estimation for fore-
casts. RMSE is used as a measure for the optimi-
zation.
As results, we can find that (1) the proposed DLM's
are robust to the initial values from the sensitivity
analysis, (2) the transition matrix G(t) seems to be
an Identity matrix, (3) for the k-lead forecast, the
optimal F(t+k)'s consist of Y(t), the estimate of
Y(t+k) from DLM and the GDAPS(t+k) forecast,
(4) RMSE's are 2.0 - 2.7 (for +1, +2, +3, +4), 2.6 -
3.0(for +5, +6, +7, +8), 3.0 - 3.8(for +9, +10), and
(5) the forecasts using the proposed DLM are re-
markably better than GDAPS forecasts.

Statistics in Agricultural Climatology adapting a
workshop for African needs, S. M. Gathara, C.
Oludhe, S. B. Otengi, P. K. Runanu, R. D. Stern

A 10-week workshop on statistical methods in
agroclimatology (SIAC) was given in Reading UK,
from 1983 to 2000. In May 2000 SIAC was given
for the first time at the Institute for Meteorological
Research and Training (IMTR), in Nairobi, Kenya.
This paper describes the lessons learned from this
venture and the changes that are being made to
tailor the 2001 course more to regional needs.
Many of these changes have followed from a train-
the-trainer workshop run in October 2000. In addi-
tion to strengthening the resource team, this has led
to a range of practical projects, described in the
paper, on both data management and applications
that are being conducted in the 6 months leading to
the 2001 course. The main course software has also
been upgraded.

Changes of large-scale atmosphere circulation
during XX century and their influence on the
climate in Ukraine, V.F. Martazinova, T. Sverdlik
(Ukrainian Hydrometeorological Institute, Kiev,
Ukraine)

The changes of the large-scale atmosphere circula-
tion over Atlantic-Europe region and formed by it
regional circulation and climate in Ukraine during
XX century are considered.
The maps of mean monthly sea level pressure
(SLP) for the 1881-1940, 1951-80, 1980-95 are
analyzed. The patterns of mean SLP differences be-
tween 1881-1940 and 1951-80; 1980-95 and 1951-
80 allowed to find identical tendency of the
atmosphere circulation transformation during peri-
ods of global warming (at the beginning and at the
end of the century) relatively stable period of global
temperature course (at the middle of the century).
The absolute values of the tendency at the end of
the century in twice as much.
The changes of the large-scale and regional atmos-
phere circulation during recent decades are defined
by shifting to the east of climate atmosphere centers
of action (Azores, Siberian highs and European
depression), that forms new weather conditions in
Ukraine. As a result the present weather conditions
on the territory of Ukraine are formed by eastern
periphery of Azores high in all seasons and charac-
terized by warm and wet winters and cool summers.
These conclusions conform to the results of air
temperature and precipitation time series analysis.
Hence this study has showed that of global and
regional climate changes are connected with chan-
ges of the large-scale atmosphere circulation.

Validating a Downscaling Model of the Stochastic
Structure of Daily Temperature, Marina
Timofeyeva
1
, Rachael Craig
2
(
1
UCAR, NOAA/-
NWS, Silver Spring, MD;
2
Dept. of Geology, Kent
State University, Kent, OH)

We model daily maximum temperature with seven
parameters: annual mean, phase shift and amplitude
from solar insolation, weighted parameters (phis) of
an AR(3) model of the residuals from the insolation
model and residual variance (sigma). The model
was calibrated with a stratified sample of instru-
mental records from the western U.S. and Mexico
for 1979-1988. Three stations (high, low and
middle elevations) from each climate division in the
U.S. and from each one-degree grid cell in Mexico
were selected for calibration; for a total of 434.
Boundary conditions are defined by CO2 concen-
tration, sea surface temperature, 500 and 1000 mb
winds and geopotential height from the CCM3 T31
paleoclimate control run. Thirty-two predictor va-
riables represent regional radiative and orographic
controls on temperature for up to 1000km upwind
of the calibration locations as well as local and
63
downwind (blocking) effects. The transfer function
defining the downscaling relation was estimated by
canonical regression.
The transfer function was validated with an inde-
pendent sample of 1290 instrumental records from
the same area and time period. Six separate analy-
ses comprise the validation. We first tested skew-
ness and kurtosis of the distribution of residuals
(observed minus modeled). All variables but phase
and phi(3) passed both tests. Those two variables
showed very slight positive skewness. Secondly we
computed paired t-statistics on randomly selected
subsets of size 208 (giving precision in means of
one degree C). We did not reject the null hypothesis
that model and observed means are equal, except
for one (of three) tests of phase. The third test
considered the significance of correlation of
residuals with each of the independent variables.
One of the seven variables (sigma) passed fewer
(89%) than 95% of the 384 tests performed. To
detect possible spatial bias we examined (test four)
the correlation of residuals with each of latitude,
longitude and elevation. All passed. The fifth test
considered a Q-mode correlation analysis of the
ensemble of independent and dependent variables
at all the stations. Correlation at 92% of the stations
exceeds 0.7. We see no systematic pattern in the
spatial distribution of stations with lower cor-
re-lation. Finally (test six) we examined the spatial
distribution of residuals and found no suggestion of
systematic patterns related to geographic controls.

Statistical Decomposition of European Tempera-
ture Time Series, Silke Trmel, Jrgen Grieser,
C.-D. Schnwiese (Institut fr Meteorologie und
Geophysik, Universitt Frankfurt, Frankfurt /M.,
Germany)

One strategy to analyze climate variability in
meteorological data, used in classical time series
decomposition, is the search for prede ned
components. We decompose monthly temperature
time series into linear as well as nonlinear trends,
the season cycle and its variations, harmonic com-
ponents, episodic components, extreme events and
noise. In contrast to extreme values our extreme
events are not caused by trends or other well de ned
changes in the mean, but they are unlikely to occur
by chance and are independent from changes in the
parameters of the distribution. It is very unlikely
that they are in accordance with the feature of the
time series. For the decomposition a stepwise
regression is used which converges fast to robust
estimaters. Our results show that the trend over
Europe can be regarded as positive and linear, that
positive phase shifts in the season cycle occur
especially in the western and negative ones in the
eastern Europe. In the north and northwest of
Europe positive changes in the amplitude of the
season cycle dominate, whereas negative ones are
found in the eastern part and in central Europe. The
overwhelming majoritiy of extreme events are
detected in winter and represent extreme cold
events. Besides, we found a signi cant increase of
extreme cold events in the second part of the
sampling time.

Coupled sea ice and atmosphere variability in the
Weddell Sea, Antarctica, Silvia A. Venegas
(Danish Center for Earth System Science
(DCESS), Niels Bohr Institute for Astronomy,
Physics and Geophysics, University of
Copenhagen, Denmark); Mark R. Drinkwater
(Oceans/Sea-Ice Unit, ESTEC Earth Sciences
Division (APP-FSO), Noordwijk, The
Netherlands)

A frequency-domain singular value decomposition
is performed on 20 years (1979-1998) of satellite
observations of sea ice concentration and drift
together with sea level pressure reanalysis data in
the Weddell Sea, Antarctica, in order to determine
the dominant timescales of variability in the region.
The interannual timescale (periods of 3-4 years) is
found to dominate the ice and atmospheric fluc-
tuations. This variability is a regional manifestation
of the Antarctic Circumpolar Wave in the Weddell
sector of the Southern Ocean.
Interannual variations of sea ice concentration and
drift are driven largely by anomalous atmospheric
patterns, which alternatively enhance or reduce the
meridional wind circulation to the east of the
Antarctic peninsula. Thermodynamic processes
(rapid ice growth in open water areas), dynamic
processes (ice ridging/rafting due to ice
convergence) and geographic constraints all play
significant roles in the formation of summer ice
concentration anomalies in the southern Weddell
Sea every 3-4 years. These ice anomalies are
exported northwards out of the Weddell basin
approximately one year later, and then advected
eastwards before diffusing in the ice margin.
A low-frequency signal (presumably a quasi-de-
cadal oscillation) is also detected, although its sig-
nificance is fairly speculative in a 20-yr record. Ice
variability on this timescale results from a change
in the shape of the Weddell gyre circulation which
occurred around 1990, likely associated with large-
scale variations of the ocean-atmosphere system.

64
Participants List 8. IMSC
27 February 2001

Ulrich Achatz
Leibniz-Institut fuer Atmosphrenphysik
an der Universitaet Rostock e.V.
Schlossstr. 6
18225 Kuehlungsborn, Germany
e-mail: [email protected]

Christine Achberger
Dept. of Earth Sciences
Physical Geography
Gteborg University
Box 460
40 30 Gteborg
Sweden

Ahmed Adjez
National Climate Center
P. O. Box 45, Hacen Badi
16200 El-Harrach
Algier, Algeria

Nasser ALBALOUSHI
United nations Development programme
UAE/99/001 Project

Abdullah Aldhanhani
UAE/99/001 Project

Genrikh V. ALEKSEEV
AARI Arctic and Antarctic Research Institute
Ocean-Atmosphere Interaction Department
199397,Bering st.,38
St.-Petersburg, Russia
e-mail:[email protected]

Hans ALEXANDERSSON
Swedish Meteorological and Hydrological Institute,
Norrkping, Sweden

Katrine Krogh ANDERSEN
Department of Geophysics
J uliane Maries Vej 30
2100 Copenhagen O, Denmark

Sylvia ANTUNES
Instituto de Meteorologia
R. C ao Aeroporto
Lisboa, Portugal

J udit BARTHOLY
Dept. of Meteorology Eotvos Lorand University
P.O.Box 32
Budapest, H-1518, Hungary

Bryson C. BATES
CSIRO Land and Water
Private Bag
PO Wembley WA 6014 Australia

Eva BAUER
Potsdam Institute for Climate Impact Research (PIK)
PO Box 601203
14412 Potsdam, Germany

Thomas BENGTSSON
Geophysical Statistics Project
National Center for Atmospheric Research
Climate and Global Dynamics Division
P.O. Box 3000, Boulder, CO 80307-3000
USA

Mark BERLINER
Ohio State University, Dept. of Statistics
404 Cockins Hall, 1958 Neil Avenue
Columbus, OH 43210, USA

Nestor R. BERNAL S.
National University of Colombia

Ulrike BEYER
Institute for Geography
University of Wrzburg
Am Hubland
97074 Wuerzburg, Germany

Peter BISSOLLI
FE24c1
Postfach 10 04 65
63004 Offenbach, Germany

Albert BOEHM

Dr.Reinhard BOEHM
Zentralanstalt fuer Meteorologie und Geodynamik
Hohe Warte 38
A-1190 Wien, Austria

Dr. Uwe BOEHM
Universitaet Potsdam
Unstitut fuer Physik
Postfach 60 15 53

Konrad BOGNER
Institute of Hydrology
University of Agricultural Research
Muthgasse 18
1190 Vienna, Austria

Auguste BOISSONNADE
Risk Management Solutions Inc.
149 Commonwealth Drive
Menlo Park, CA 94025
California, USA

Roxana BOJ ARIU
Institutul National de Meteorologie si Hidrologie
Sos. Bucuresti-Ploiesti 97
71 552 Bucharest, Romania
[email protected]

Katarina BORNE
Dept. of Earth Sciences
Physical Geography
Gteborg University
65
Box 460
40 30 Gteborg
Sweden

Zsolt BOTTYN
Department of Climatology and Landscape Ecology
University of Szeged
University of Szeged
P.O. Box 653
6701 Szeged, Hungary

Anders BRIX
Risk Management Solutions Inc.
Menlo Park
California, USA

Harold E. BROOKS
NOAA/National Severe Storms Laboratory
Halley, Circle
Norman, OK, USA

Adri BUISHAND
Royal Netherlands Meteorological Institute (KNMI)
De Bilt, The Netherlands

Arunas BUKANTIS
Dept. of Hydrology and Climatology of Vilnius
University
M.K.Ciurlionio St. 21/27
Vilnius 2009, Lithuania

Aristita BUSUIOC
National Institute of Meteorology and Hydrology
71552 Bucharest, Romania

Ulrich CALLIES
Institute for Hydrophysics
GKSS Research Center
21494 Geesthacht, Germany

Ins CAMILLONI
Dept. of Atmospheric Sciences
University of Buenos Aires
Argentinia

Eddy CAMPBELL
CSIRO Mathematical and Information Sciences
Private Bag No. 5
Wembley, WA 6913, AUSTRALIA

Hong Xing CAO
Chinese Academy of meteorological Sciences,
46 Baishiqiaolu
100081 , Beijing ,China
e-mail : [email protected]

Carmen CAPILLA
Universidad Politecnica,
Depto. Estadistica e Investigacion Operativa Apartado
Apartado 22012-46071 Valencia, Spain

Dr. Peter CARL
Institute for Freshwater Ecology
und Inland Fisheries (IGB)
Forschungsverbund Berlin
Hausvogteiplatz 5-7
10117 Berlin, Germany

Leila M. Vespoli de CARVALHO
Institute for Computational Earth System Science
(ICESS)
University of California
Santa Barbara, CA 93106

Ana C. CEBRIAN
Institut de Statistique
Universit Catholique de Louvain
Voie du Roman Pays, 20
B-1348, Louvain-la-Neuve
Belgium

Peter CHALLENOR
Southampton Oceanography Centre
Southampton, SO14 3ZH, UK

Lynda CHAMBERS
Bureau of Meteorology Research Centre
Melbourne, Australia

Richard CHANDLER
Dept. of Statistical Science
University College
Gower Street
London WC 1E 6BT, UK

Stephen CHARLES
CSIRO Land and Water
Private Bag #5 Wembley WA 6913 Australia

Rosane R. CHAVES
INPE CPTEC - Rod. Pres. Dutra, Km 40
C. Paulista - SP - Brasil - CEP:12630-000, Brazil

Anatoly CHAVRO
Institute for Numerical Mathematics RAS
8, Gubkin St.
Moscow, 117333, Russia

Youmin CHEN
Agrometeorology Institute
Chinese Academy of Agricultural Sciences
30 Baishiqiao Road
Beijing, 100081, China
or: [email protected]

J ean-Paul CHILES
Bureau de Recherches Gologiques et Minires
BRGM 3, avenue Claude Guillemin
BP 6009, 45060 Orlans Cedec 2, France

Bernard CORNLIS
Dpartement de Gomatique Universit de Lige
Place du 20-Aot, 7
4000 Lige, Belgium

Dan CORNFORD
Computer Science
66
Aston University
Aston Triangle
Birmingham B4 7ET, UK

J oao CORTE-REAL
Univ. Lisbon, Portugal

Traute CRGER
Meteorological Institute
University of Hamburg
Germany

Moses DAMOAH
- c/o Rev J oshua Damoah
Private-Mail Bag
Abesim Sunyani-B/A
Ghana W/A
- Kwaku Damoah
48 Yorkshire Road
Mitcham, Surrey CR4 1QD, UK

Utpal .K. DE
J adavpur University
Calcutta, India

Simona De ZOLT Sappadina
Department of Physics G. Galilei
via Marzolo 8
35131 Padova, Italy

Ildiko DOBI-WANTUCH
Hungarian Meteorological Service
1525 Budapest P. O. Box. 38, Hungary

Mojca DOLINAR, Hydrometeorological Institute of
Slovenia, LJ UBLJ ANA, Slovenia

Martin DUBROVSKY
Institute of Atmospheric Sciences
Husova 456
50008 Hradec Kralove
Czech Republic

Salaheddine EL-ADLOUNI
Univ. Mohammed V, Facult des Sciences
Dp. de Mathmatiques et Informatique.
BP. 1014; Rabat Maroc

Geir EVENSEN
Nansen Environmental and Remote Sensing Center,
Edvard Griegsvei 3a,
5037 Solheimsviken, Norway

Frauke FESER
Institute of Hydrophysics
GKSS Research Centre
Geesthacht, Germany

Gtz FLSER
Geesthacht, Germany
e-mail [email protected]

Dr. Ulrich FOELSCHE
Institute for Geophysics, Astrophysics, and Meteorology
University of Graz (IGAM/UG)
Universitaetsplatz 5
A-8010 Graz, Austria

Chris FOREST
Massachusetts Instituts of Technology
Dept. Earth, Atmos. And Plan.
77 Massachusetts Avenue, Cambrigde,
MA 02139-4307 USA

Carsten S. FREDERIKSEN
Principal Research Scientist
BMRC Model Evaluation Group
Bureau of Meteorology Research Centre
Melbourne, Australia

Thomas K. FRIEDLI
Institut fr mathematische Statistik und
Versicherungslehre
Universitt Bern
Sidlerstrasse 5
3012 Bern, Switzerland

R. GARCA
Dto Fsica de la Tierra II
Universidad Complutense
Madrid, Spain

Christoph GEBHARDT, Meteorologisches Institut der
Universitt Bonn
Auf dem Hgel 20
53121 Bonn, Germany

D. GELLENS
Computer Centre
Royal Meteorological Institute of Belgium
Av Circulaire, 3
180 Brussels, Belgium

Friedrich Wilhelm GERSTENGARBE
Potsdam-Institut fr Klimafolgenforschung
POB 601203

Luis GIMENO
Universidad de Vigo, Facultad de Ciencias de Orense,
32004 Orense, Spain
Tfno: 988387208
Fax: 988387159

Clare GOODESS
Climatic Research Unit
University of East Anglia
Norwich NR4 7TJ , UK

J rgen GRIESER
J ohann-Wolfgang Goethe Universitt Frankfurt am Main
Institut fr Meteorologie & Geophysik
Postfach: 11 19 32
D - 60054 Frankfurt a. M.

67
J rgen Grieser
Food and Agriculture Organization of the United Nations
SDRN - C 854 Via delle Terme di Caracalla
00100 Rome, Italy

Anders GRIMVALL
Department of Mathematics
Division of Statistics
Linkping University
SE-58183 Linkping, Sweden

Nikolaus GROLL
Postfach 1160

Peter GUTTORP
University of Washinton
Seattle, WA, USA

Sven HARIG
Alfred-Wegener-Institut
Bussestrae 24
D-27570 Bremerhaven
Germany

Knut HARSTVEIT
Pb 43 Blindern
0313 OSLO, Norway

Urip HARYOKO
Meteorological and Geophysical Agency
J l. Angkasa I No. 2
J akarta, Indonesia

Klaus HASSELMANN
Max-Planck-Institute for Meteorology
Bundesstr. 55, 20146 Hamburg, Germany

Gabi HEGERL
Gabi Hegerl and Tom Crowley
Dept. of Oceanography
Texas A&M
College Station, TX 77843, USA

Gabi HEGERL
University of Washinton
Seattle, WA, USA

Dr. Hermann HELD
Telegrafenberg C4, 14473 Potsdam

J aakko HELMINEN
Finnish Meteorological Institute
Helsinki, Finland
J [email protected]

Andreas HENSE,
Meteorological Institute
University of Bonn
Auf dem Hgel 20
53121 Bonn, Germany

Carme HERVADA i Sala
Fisica i Enginyeria Nuclear
L'escola industrial (UPC)
Colom 1
08222 Terrassa, Spain

Heike HOFFMANN
Graduiertenkolleg Rumliche Statistik
Bernhard-v.-Cotta-Str. 2
09596 Freiberg, Germany

Greg HOLLOWAY
Institute for Ocean Sciences
9860 West Sannich Road
Sidney, B.C. V8L 4 B2, Can.

Ibrahim HOTEIT
Laboratoire de Modelisation et Calcul
Grenoble, France

Richenda HOUSEAGO-Stokes
J ames Rennell Division for Ocean Circulation and
Climate
Southampton Oceanography Centre
Empress Dock
Southampton, SO14 3ZH, UK

J im HUGHES
Dept. of Biostatistics
University of Washington
Mailstop 359931
Seattle, WA 98195

Radan HUTH
Institute of Atmospheric Physics
Prague, Czech Republic

Knut A. IDEN
Norwegian Meteorological Institute (DNMI)
P.O.Box43 Blindern
N-0303 Oslo
Norway

Valerie ISHAM
Dept. Of Statistical Science
University College
Gower Street
London WC1E 6BT, UK

Shaleen J AIN
Cooperative Institute for Research in Environmental
Sciences (CIRES)
University of Colorado
Campus Box 216
Boulder, CO 80303-0216 USA
E-mail: [email protected]
Office: NOAA-CDC/David Skaggs Research Center,
Room 1D-111

Stephen J EWSON
Risk Management Solutions, Inc.
68
Menlo Park
California, USA

Ian J OLLIFFE
University of Aberdeen, UK

Charles J ONES
Assistant Researcher
Institute for Computational Earth System Science
(ICESS)
University of California, Santa Barbara
CA 93106-3060, USA

J ulie M. J ONES
Postfach 1160

Eigil KAAS
Danish Meteorological Institute
Lyngbyvej 100
2100 Copenhagen, DENMARK

Tiba KABANDA
School of Environmental Sciences, University of Venda
Private Bag, X5050, Thohoyadou, 0950, South Africa

J aroslava KALVOVA
Dept. of Meteorology and Environment Protection
Faculty of Mathematics and Physics
Charles University
V Holeovikch 2
180 00 Praha 8, Czech Republic

S.S. KANDALGAONKAR
Physical Meteorology and Aerology Division
Indian Institute of Tropical Meteorology
N.C.L. Post Office, Dr. Homi Bhabha Road
PUNE 411 008, India

Mohammed-Said KARROUK
Universitt Hassan II
Centre de Recherche de Climatologie
BP 8220 Oasis, MA-20103, Casablanca, Morocco

Rick KATZ
Environmental and Societal Impacts Group
Boulder, CO 80307
Colorado, USA

Frank KAUKER
Alfred-Wegener-Institute for Polar and Marine Research
Bussestr. 24
27570 Bremerhaven, Germany

Dr A.S.A. KHALIL
Chief Technical Advisor
UAE/99/001 Project

Gennady A. KIVMAN
Alfred-Wegener-Institut for Polar- and Marine Research
Bremerhaven,, Germany

Richard KLEEMAN
Courant Institute for Mathematical Science
New York University, USA

Stanislava KLIEGROVA
Czech Hydrometeorological Institut
Branch Hradec Krlov
Dvorsk 410
503 11 Hradec Krlov - Svobodn Dvory, Czech
Republic

Ken-Chung KO
Department of Geography
National Kaohsiung Normal University
116 Ho-Ping First Road
Kaohsiung, Taiwan

Till KUHLBRODT
Climate System Dept.
Telegrafenberg, PO Box 60 12 03

Ashwini KULKARNI
Pashan, Pune 411008, India

Venerabilis KULULETERA
Tanzania Meteorological Agency
P.O.Box 3056
Dar es Salaam
Tanzania

Vit KVETON
Czech Hydrometeorological Institute
Na abatce 17
143 06 Prague 4, Czech Republic

Frank KWASNIOK
Leibniz-Institut fr Atmosphrenphysik
Schlossstrasse 6
18225 Khlungsborn, Germany

Won-Tae KWON
Meteorological Research Institute
Korea Meteorological Administration, Korea

J an KYSELY
Institute of Atmospheric Physics

Victor E. LAGUN
Air-Sea Interaction Department
Arctic and Antarctic Research Institute
Beringstr. 38
St. Petersburg 199397, Russia
69

Monika LAKATOS
Hungarian Metrological Service

David R. LEGATES
Center for Climatic Research
University of Delaware
227 A Pearson Hall
Newark, DE, 19716, USA

Piero .LIONELLO
Dept. of Material Sciences
University of Lecce
Via per Arnesano
73100 Lecce, Italy

Ulrike LOEPTIEN
Universitt Kiel
Institut fr Meteorologie

Leonid LOPATOUKHIN
State University (Dep.Oceanography),
199178, St.Petersburg, 10 Line, 33/35, Russia

Omar Abel LUCERO
Omar Abel Lucero
National University of Cordoba
Centro de la Regin Semirida (INA), Argentina
[email protected]
[email protected]

M Dolores MANSO ORGAZ
Depart. of Physics
University of Aveiro,
Portugal

Rol MADDEN
UCAR, Boulder, Colorado, USA

Ileana MARES
Constantin MARES
Bucharest, 71552, ROMANIA
[email protected]

J orge MATEU
University J aume I
Campus Riu Sec
12071 Castellon, Spain

Istvan MATYASOVSZKY
Department of Meteorology
Eotvos Lorand University
1117 Budapest Pzmny Pter stny 1/A
1364 Budapest, Hungary

Norm MCFARLANE
Canadian Centre for Climate Modeling and Analysis
Victoria, Canada

Solange MENDONCA LEITE
Physics Department
University of Tras-os-Montes e Alto Douro
PoBox 202
5000-911 Vila Real, Portugal

Miroslaw MIETUS
Institute of Meteorology and Water Management
Maritime Branch
Gdynia, Poland

J anos MIKA
H-1525 Budapest P. O. Box. 38.
(H-1024 Budapest Kitaibel P. u. 1), Hungary

Adam MONAHAN
Institut fuer Mathematik
Humboldt-Universitaet zu Berlin
Unter den Linden 6
10099 Berlin, Germany

M. Granger MORGAN
Carnegie-Mellon University
5000 Forbes Ave., Pittsburgh
PA 15213, USA

Manfred MUDELSEE
Institute of Meteorology
University of Leipzig
Stephanstr. 3
D-04103 Leipzig, Germany

Peter MLLER
Department of Oceanography
University of Hawaii, USA

Brbel MLLER-KARULIS
Marine Monitoring Centre
Institute of Aquatic Ecology
University of Latvia
8 Daugavgrivas
1007 Riga, Latvia

Donald E. MYERS
University of Arizona
Tucson, AZ 85721, USA

Philippe NAVEAU
National Center for Atmospheric Research,
Boulder, CO, USA,

Lars NERGER
Alfred-Wegener-Institut
fuer Polar- und Meeresforschung
Rechenzentrum
Buergermeister-Smidt-Strasse 20
27568 Bremerhaven, Germany

70
Svetlana NIKULINA
Central Asian Research Hydrometeorological Institute
(SANIGMI)
Tashkent, Uzbekistan

Carlos NOBRE
Centro de Previso de Tempo e Estudos Climaticos
Cachoeira Paulista, Brasil

Anders NORDGAARD
Statistics Division
Linkping University
SE-581 83 Linkping, Sweden

Doug NYCHKA
Boulder, CO, USA

Guillermo O. OBREGON
Centro de Previsao de Tempo e Estudos Climaticos
(CPTEC)
Instituto Nacional de Pesquisas Espaciais (INPE)
Rod. Presidente Dutra, km. 40
12630-000 Cachoeira Paulista, Sao Paulo, Brazil

Timothy J . OSBORN
School of Environmental Sciences
University of East Anglia,
Norwich NR4 7TJ , UK

Taha B. M. J . OUARDA

Christopher J . PACIOREK
Department of Statistics
Carnegie Mellon University
Pittsburgh, Pennsylvania, USA

Heiko PAETH
Meteorologisches Institut der Universitt Bonn
Auf dem Hgel 20
53121 Bonn, Germany

Daniela Mihaela PALIU
71552 Bucharest, Romania

Ruben PASMANTER
Faculty of Science
University of Amsterfdam
The Netherlands

Sarath PEIRIS

Richard E. PETERSON, Chair
Professor of Atmospheric Science
Associate Director
Department of Geosciences
Wind Science and Engineering Program
Texas Tech University
Lubbock, TX 79409 USA

Arnt PFIZENMAYER
Geesthacht, Germany

Carlos Alberto Leitao PIRES
CGUL, - Centro de Geofsica da Universidade de Lisboa
Rua da Escola Politcnica 58
1250 Lisboa, Portugal

Faculdade de Cincias
Departamento de Fsica, Edifcio C8
Rua Ernesto de Vasconcelos - Campo Grande
1749-016 Lisboa Codex - PORTUGAL

Henrique Oliveira PIRES
Instituto de Meteorologia
R. C ao Aeroporto
Lisboa, Portugal

Budong QIAN
ICAT - Faculty of Sciences
University of Lisbon, Portugal
e-mail: [email protected];
[email protected]; [email protected]

Radhika RAMACHANDRAN
Space Physics Laboratory
Vikram Sarabhai Space Centre
Trivandrum - 695 022
INDIA

Nazario D. RAMIREZ-Beltran
Department of Industrial Engineering
University of Puerto Rico
Mayaguez Puerto Rico 00680

Clarence RICHARDSON
Laboratory Director
Grassland Soil & Water Research Laboratory
808 E. Blackland Road
Temple, TX 76502, USA
[email protected]

Pedro RIBERA

J oao RIO
Centro de Geofsica da Universidade de Lisboa;
PORTUGAL

J ames RISBEY
University of Utrecht
Department of Science, Technology and Society
Padualaan 14
3584 CH Utrecht, The Netherlands

Torben SCHMITH
Danish Meteorological Institute
71
Climate Research Division
Lyngbyvej 100
2100 Copenhagen , DENMARK

Vladimir A. SEMENOV
Max-Planck-Institut fr Meteorologie
Bundesstr., 55
20146 Hamburg, Germany

J ulien SNEGAS
Centre de Geostatistique
Ecole des Mines de Paris
77305 Fontainebleau

Amjad Majeed SHAH
Assistant Professor
Statistics Department
Islamia University Bahawalpur, Pakistan

Ajay SINGH
School of Management
Indian Institute of Technology
Bombay,Powai
Mumbai-400076, India

NITYANAND SINGH
Dr. Homi Bhabha Road
Pashan, Pune- 411 008, India

Leonard A. SMITH
Centre for the Analysis of Time Series
London School of Economics
London, WC2A 2AE, UK

Keon Tae SOHN
Pusan National University, Korea

Neelima SONTAKKE
Pashan, Pune-411008, India

Tim SPARKS
Centre for Ecology and Hydrology
Monks Wood, Abbots Ripton
Huntingdon
Cambridgeshire PE28 2LS, UK
e-mail: [email protected] or [email protected]

Tim STAEGER
J .-W. Goethe-Universitdt Frankfurt
Institut fr Meteorologie und Geophysik
Arbeitsgruppe Meteorologische
Umweltforschung/Klimatologie
Postfach 11 19 32
D-60054 Frankfurt a.M., Germany
mail: [email protected]

Christos N. STEFANAKOS
National Technical University of Athens
Dept. of Naval Architecture & Marine Engineering
P.O.BOX 64033
Zografos, 15710
Athens, Greece

Dietrich STOYAN
Institut fr Stochastik
Bergakademie Freiberg, Germany

Daithi STONE
School of Earth and Ocean Sciences
University of Victoria
PO Box 3055, Victoria
BC, V8W 3P6, Canada

Hans von STORCH
Geesthacht, Germany

J in-Song von STORCH
Meteorologisches Institut der Universitt Hamburg
Germany

Tatyana SVERDLIK
Ukrainian Hydrometeorological Institute,
37, Nauki av., Kiev, Ukraine
Phone: (044) 265 87 90
Fax: (044) 265 53 63

Tamas SZENTIMREY
Climatological and Agrometeorological Division
1525--Budapest P.O. Box 38, Hungary

Clemente A. S. TANAJ URA, Laboratorio Nacional de
Computa of Cientfica, Petrpolis, Brazil

Karl E. TAYLOR
P.O. Box 808, L-264
Livermore, CA 94550
email: [email protected]

J onathan TAYLOR
Lancaster University, UK

Susanne THEIS
Meteorologisches Institut der Uni Bonn
Auf dem Hgel 20
53121 Bonn, Germany

Peter THORNE
School of Environmental Sciences
University of East Anglia
Norwich, NR4 7TJ , UK

Marina TIMOFEYEVA
National Weather Service
1325 East West Hwy
Silver Spring, MD 20910 3283

Silke TRMEL
J ohann Wolfgang Goethe Universitt
Institut fr Meteorologie und Geophysik
Robert-Mayer-Str. 1
60054 Frankfurt a.M.,

72
Heikki TUOMENVIRTA
Finnish Meteorological Institute
P.O. Box 503
00101 Helsinki, Finland

Silvia A. VENEGAS
Danish Center for Earth System Science (DCESS)
Niels Bohr Institute for Astronomy, Physics and
Geophysics
University of Copenhagen, Denmark

Lucie VINCENT
Climate Research Branch
Meteorological Service of Canada
4905 Dufferin Street
Downsview, Ontario, Canada
M3H 5T4

Aldo J . VIOLLAZ
Universidad Nacional de Tucuman
Argentina
e-mail: [email protected];
[email protected].

Hans Wackernagel
Ecole des Mines
Paris, France

Andreas WALTER
Institute of Meteorology and Geophysics
J ohann-Wolfgang Goethe University
Frankfurt/Main, Germany

Xiaolan (Leona) WANG
Network Strategies Division, AMWSD
4905 Dufferin Street, Downsview
Ontario, Canada M3H 5T4

Ed WAYMIRE
Oregon State University, Dept. of Mathematics,
Kidder Hall 368, Corvallis, OR 97331-4605, USA

J oanna WIBIG
Dept. of Meteorology and Climatology
University of Lodz
Lipowa 81
90-568 Lodz, Poland
[email protected]

Martin WIDMANN
Postfach 1160

Lawrence WILSON
Environment Canada

Katja WOTH

J un-Ichi YANO
Meteorologisches Institut
Universitt Hamburg
Bundesstrasse 55
D-20146 Hamburg, Germany

Abebe YESHANEW
National Meteorological Services Agency
Addis Abeba, Ethiopia

Xiaogu ZHENG
The National Institute of Water and Atmospheric
Research
301 Evans Bay Parade, Greta Point
P. O. Box 14901
Wellington New Zealand

Christine ZIEHMANN
Universitt Potsdam - Institut fr Physik
Am Neuen Palais 10 - Haus 19
Postfach 601553

Olga ZOLINA
P.P.Shirshov Institute of Oceanology
Russian Academy of Science (IORAS)
36 Nakhimovsky ave.
117851 Moscow, Russia

Francis ZWIERS
Canadian Center for Climate Modelling and Analysis
c/o University of Victoria
PO Box 1700, STN CSC
Victoria, BC V8W 2Y2, Canada
e-mail f zwi er s@ec. gc. ca

73

von Storch, 2, 7, 8, 15, 37, 41

74

8th International Meeting

Uploaded by

Copyright:

Available Formats

8th International Meeting

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

8th International Meeting

Uploaded by

Copyright:

Available Formats

8

You might also like