Reg Environ Change (2014) 14:727–742

DOI 10.1007/s10113-013-0534-3

ORIGINAL ARTICLE

Climate change impact on countrywide water balance in Bolivia

Jorge José Escurra • Vı́ctor Vazquez •
Rita Cestti • Erwin De Nys • Raghavan Srinivasan

Received: 17 February 2013 / Accepted: 31 August 2013 / Published online: 22 September 2013
 Springer-Verlag Berlin Heidelberg 2013

Abstract There is increasing concern about the ongoing dioxide projections derived from the Special Report
reduction in water supplies in the tropical Andes due to Emission Scenario. Overall, model results were close to
climate change effects such as glacier/snow melting satisfactory compared to observations, with some excep-
resulting from rising air temperatures. In addition, extreme tions due to lack of information for expanding the timeline
events and population growth are already directly affecting and improving calibration. Based on the calculation of
life and water renewability in the country. A countrywide three hydrologic indicators, the study identifies basins that
integrated national plan for improving basin-scale water would be the most susceptible to water deficits for a
management in Bolivia is needed to assure water avail- baseline from 1997 to 2008, and in the event of the pro-
ability for agriculture, industry, mining, and human con- jected climate change, to 2050.
sumption. This study aims to provide a modeling tool to
assess Bolivia’s past, current, and future water availability Keywords Hydrologic indicators  Water deficit 
and identify basins at risk of water deficits. The Soil Water Bolivia  Water balance  Climate change impact
Assessment Tool was used to simulate the monthly water
balance from 1997 to 2008, as well as the water balance
projected to 2050 for the entire country. It considers pos- Introduction
sible changes in air temperatures and precipitation pro-
posed by 17 Global Circulation Models as well as carbon Bolivia is confronting problems of water scarcity due to
lack of precipitation, increasing average temperatures
mainly in the Andes mountain range, and social conflicts.
Electronic supplementary material The online version of this
article (doi:10.1007/s10113-013-0534-3) contains supplementary On the other hand, some areas (such as Trinidad, Cobija,
material, which is available to authorized users. and Santa Cruz, among others) are impacted by floods.
Social conflicts linked to water have been also observed
J. J. Escurra (&)  V. Vazquez  R. Cestti  E. De Nys
throughout Bolivian history, with one of the most recent
The World Bank, 1818 H Pennsylvania Av,
Washington, DC, USA and well known being a violent strike in the city of Coc-
e-mail: [email protected] habamba in 2000 against water supply privatization and the
V. Vazquez company Aguas del Tunari Consortium. The reason for the
e-mail: [email protected] strike was the excessive increase in water rates for con-
R. Cestti sumers (Lobina 2000). The conflict represents the central
e-mail: [email protected] tension that exists around water in the contemporary world:
E. De Nys privatization versus the common good under conditions of
e-mail: [email protected] water scarcity (Peredo 2004). Climate change is also cur-
rently affecting glaciers in the Tropical Andes. An analysis
R. Srinivasan
of the records of 268 mountain weather stations (above
Spatial Sciences Laboratory, Texas A&M University,
College Station, TX, USA 4,000 m m.a.s.l) between 1N and 23S along the Andes
e-mail: [email protected] indicates a temperature increase of 0.11 C per decade.

123
728 J. J. Escurra et al.

This value is 0.05 C higher than the global average tem- In 2000, as part of the First Bolivian communication to
perature increase (Bradley et al. 2006). All these changes the United Nations Framework Convention on Climate
as well as others (such as pollution) are expected to neg- Change (UNFCCC), WATBAL was used to analyze pos-
atively impact current and future water availability in sible climate change impacts on two major sub-basins,
Bolivia, as well as affecting the environmental Pirai and La Paz, where two of Bolivia’s main cities are
sustainability. located (Santa Cruz and La Paz). The periods of hydrologic
Several publications have attempted to describe the baseline for Pirai and La Paz were 1976–1992 and
hydrology in parts of the country with complex topography 1985–1996, respectively. Comparison between basins is
and diverse microclimates (Caballero et al. 2002; Ribstein again difficult because of differences in the period of the
et al. 1995; Ronchail et al. 2005; and Roche and Jauregui baseline. After calibration and validation of WATBAL for
1988). However, the first and only attempt to understand the hydrologic baseline, a climate change analysis using
the countrywide water cycle was published in 1992 (Roche the Hadley Center United Model 2 (HadCM2) GCM con-
et al. 1992). This publication produced a hydrologic ana- cluded that:
lysis from the years 1869–1983. The hydrologic parameters
• Only the eastern region of the Pirai sub-basin would be
included precipitation, evapotranspiration (ET), potential
vulnerable to flooding (surplus of water) due to climate
evapotranspiration (ETP), and surface runoff (SR). The
change impacts. The climate change scenario indicated
study concluded that a sub-annual water balance over a
that stream flows would increase in the rainy season;
longer timeline was needed to improve the understanding
however, the western region of the Pirai sub-basin
of Bolivia’s hydrology. In addition, the study indicated that
would be more vulnerable to drought in the rainy and
the endorheic basins of the Altiplano region, located
dry seasons.
around Lake Titicaca, have a higher ETP than precipita-
• The La Paz sub-basin would face a similar situation due
tion, indicating vulnerability to water scarcity in these
to climate change. The higher altitude regions of the
regions.
sub-basin would be vulnerable to drought, especially
In recent years, there have been several studies related
the watershed of Hampaturi reservoir. The study
to climate change impacts on water availability in
predicted that in 2030 water availability in the reservoir
Bolivia, all oriented to specific basins, but there are no
would not satisfy domestic demand, which would be
recent scientific publications to update a countrywide
close to 0.4 Gm3. Lower regions of the sub-basin would
water balance. The Ministerio de Desarrollo Sostenible y
be prone to flooding.
Medio Ambiente (MDSMA) and the United States
Environment Protection Agency (USEPA) analyzed the A climate change assessment using a countrywide water
vulnerability of the forestry, agricultural, and water balance (MDP and VPTyMA 2007) divided Bolivia into
sectors to climate change. This analysis evaluated water the three major basins (La Plata, Amazonas, and Endo-
yields (WY) for five sub-basins. The expected effects of rreica). This study used results from the countrywide
possible climate change were calculated using the Water hydrologic study from 1869 to 1983 of Roche et al. (1992).
Balance Model (WATBAL) to estimate the hydrologic The climate change analysis projected an increase of
impacts of climate change projected by two Global 1.5 C in the entire country, a 15 percent (%) reduction in
Circulation Models (GCMs). The periods used to calcu- the annual precipitation in the Amazonas basin, and a 15 %
late the hydrologic baseline varied between sub-basins, increase in La Plata and Endorreica basins. The study
making inter-basin comparisons difficult (Caine from concluded that the country would face increased water
1960 to 1975, Miguillas from 1970 to 1976, Guadal- deficits under these climate change scenarios, as well as
quivir Alto from 1883 to 1992, Choqueyapu from 1981 intensification of the glacier melting in the La Plata and
to 1990, and Mamore from 1971 to 1981). After cali- Endorreica basins.
brations and validations of WATBAL results versus The present study attempts to sort through historic
observed flows, the impacts of projected climate change information, fill information gaps, and develop a validated
[by the Goddard Institute Space Studies (GISS) and the and calibrated countrywide water balance. The SWAT
United Kingdom Met Office (UKMO-89) GCMs] and model was used to simulate the countrywide water balance
doubled atmospheric carbon dioxide (CO2) concentra- in order to: (1) provide a updated digital knowledge data-
tions on water availability were analyzed. The results base that includes hydrometereologic data, elevation, water
indicated that WY would decrease during the rainy infrastructure, soil characteristic data, land use data, crop
season and increase during the dry season. Authors calendar information, agricultural practices (including
emphasized that the results should only be taken as irrigation), irrigated land areas, glaciers, and socioeco-
relative estimates because of the lack of stream flow data nomic indicators; (2) calculate current and historic values
for validating the results (MDSMA 1997a, b). of actual ET, WY, groundwater recharge (GW), soil

123
Climate change impact 729

moisture (SM), SR, and precipitation; and (3) identify the Bolivia is divided into eleven basins [Food and Agriculture
basins affected by deficits or surpluses of water. After Organization (FAO) 1997]: Pilcomayo-Bermejo, Paraguay,
simulating the historic and present water balance, potential Lago Titicaca, Desaguadero-Poópo-Coipasa, Salar de Uy-
future events (e.g., climate change effects) are incorporated uni, Beni, Madre de Dios-Orthon, Madera, Ichilo-Mamore,
in order to project their possible consequences. Based on Itenez, and Grande (Fig. 1).
the results, constructive interaction among civil society, The description of the hydrology, land cover, and soil
non-governmental organizations (NGOs), government, and texture is shown in S1 (Electronic Supplementary Mate-
experts could be used to create consensus about how to rial). Tables 1 and 2 show the characteristics of the eleven
move forward. main hydrologic regions and reservoirs considered in the
water balance model. The scope of the work does not
consider the Lake Titicaca as part of the study area because
Study area the evaporation of the lake would influence in the average
ET of the basin. This could affect the order of prioritization
Bolivia is located in South America between 14–23S and for basins that are vulnerable to water scarcity.
60–65West (W). Bolivia has an area of 1,098,581 km2. As part of the snow melting analysis, the variation in
The population growth rate is 1.77 % (CIA World Fact- precipitation and mean temperature at 1-km differences of
book 2009). Bolivia has three topographical regions well altitude were calculated from weather stations in B1. These
defined by the Andes mountain range: Andean, Sub- variations, called temperature and precipitation lapse rates,
Andean, and Plains regions. were 4 C and 15 mm, respectively, and are part of the
Bolivia has a wide spectrum of climate conditions due to model parameters to simulate the snow melt. The temper-
the Andes. The average annual precipitation in the Andean ature and lapsed rate were calculated based on the regres-
region is around 500 mm/year, the Sub-Andean region is sion analysis of field data from the closest weather stations
950 mm/year, and the Plains region is close to 1,870 mm/ to the glaciers. Temperature lapse rates in mountain
year (Roche et al. 1992). The per capita freshwater avail- regions vary from 3 to more than 6 C/km (Meyer 1992).
ability for the country was estimated at around 160 m3/ For instance, stations in the front range of Colorado display
capita/year in the year 2000 (Black and King 2009). temperature lapse rate of between 2.8 and 4 C/km (Barry

Fig. 1 Study area which includes eleven basins

123
730 J. J. Escurra et al.

Table 1 Characteristics of the eleven main hydrologic regions (basins) in Bolivia

Basin Area Altitude Mean precipitation Population Number of Number of stations
(km2) (m.a.s.l) from 1994 to (Census sub-basins meteorology,
2008 (mm) 2001) stream gauges

B1 8,378 3,500–5,950 500 973,171 864 14, 2

B2 77,691 3,550–6,550 400 547,339 60 10, 4
B3 60,982 3,350–4,600 250 38,936 656 3, 0
B4 119,702 250–5,900 500 1,222,558 130 20, 1
B5 99,258 150–5,100 850 2,485,437 165 27, 0
B6 238,205 130–3,250 1,200 355,354 150 19, 0
B7 90,803 30–1,155 900 64,950 486 4, 0
B8 118,380 100–6,400 1,500 1,383,505 122 17, 3
B9 119,702 100–2,780 1,700 42,948 213 2, 1
B10 21,591 90–290 1,500 5,025 124 1, 0
B11 157,388 100–4,670 2,000 1,155,102 144 8, 2

Table 2 Characteristics of the six reservoirs lakes considered in the model

Name Rivers Operation Surface area Gross capacity Altitude Longitude Latitude
starting year (km2) (106 9 m3) (m.a.s.l) () ()

Tacagua Tacagua 1961 5.9 31.3 3,735 66.73 18.82

Challviri Chalviri 1936 0.2 2.6 4,434 65.69 19.65
Reque Huancarani 1950 0.1 1.3 3,985 65.95 19.57
Tuni PayaHuaychinta 1977 5.2 21.5 4,437 68.25 16.24
Khara Khota Pauchintani 1989 4.9 14 4,359 68.38 16.17
El Tranque Culpina Mayu 1984 0.25 5 3,026 64.94 20.80

1990). Fifty stations in the Andes of Peru and Ecuador characteristics. The agricultural practices information was
generate a lapse rate of 6.5 C/km between 2,500 and obtained by interviewing an agronomist from the Founda-
4,500 m.a.s.l. (Johnson 1976). tion for the Promotion and Investigation of Andean Pro-
ducts (PROINPA in Spanish).
Agriculture water consumption

Agriculture made up about 12 % of the Bolivian gross Data and methods

domestic product (GDP) in 2010 (INE 2010), and the main
crops are potatoes, rice, corn, soy, barley, quinoa, wheat, The methodology started with the monthly and annual
coffee, cotton, coca, vegetables, and fruits. Soy is one of simulation of the hydrologic indicators from 1997 to 2008.
the major export crops and its export accounts for 30 % of The hydrologic indicators are represented by the precipi-
Bolivia’s GDP attributable to agriculture (INE 2010). The tation, total annual renewable water (TARW), and ET. To
total area equipped with irrigation systems in Bolivia is enable a deep understanding of the water availability of
around 128,000 hectares or ha (Siebert and Döll 2001), and each basin, the country was divided into eleven basins and
this area is mainly located in the following basins: B1, B6, the hydrologic indicators were simulated for each of them.
B5, B2, and B4. 0.6 Gm3 of water is stored in 287 dams in Then, a climate change analysis involving identification of
Bolivia, and irrigation accounts for 85–90 % of water use the wettest and driest scenarios through the calculation of
in the country. Of the total irrigated area, 65 % is irrigated the Climate Moisture Index (CMI) was applied for each
from rivers, 15 % from wells, and 20 % from reservoirs basin. As part of the climate change analysis, outputs from
(Viceministerio de Recursos Hı́dricos y Riego 2010). As 17 GCMs for the Special Report Emission Scenario
part of the study, the four main crops, namely soy, potato, (SRES) A2 in 2050 were used to generate the CMI. Then,
corn, and rice were incorporated into the water balance the average changes in precipitation and mean temperature,
model considering the local agriculture practices and crop from the GCMs selected for the driest and wettest scenario

123
Climate change impact 731

at each basin, were incorporated into the hydrologic model with a scale of 1:250,000 (http://essm.tamu.edu/bolivia/
using the delta change approach. The results of both sce- cambio_climatico_es.htm). A soil map from FAO was
narios help to understand what future problems (deficit/ used, with a scale of 1:5,000,000 (http://essm.tamu.edu/
surplus) each basin could face due to climate change bolivia/cambio_climatico_es.htm). A global irrigation map
impacts. (Siebert et al. 2006) was derived from the FAO’s global
information system on water and agriculture. (http://www.
The hydrologic model tool fao.org/nr/water/aquastat/irrigationmap/index10.stm). Gla-
cier maps from the National Snow and Ice Data Center
The countrywide water balance was developed using the which is a public domain world glacier inventory (http://
2009 version of the Soil and Water Assessment Tool nsidc.org/data/docs/noaa/g01130_glacier_inventory/) were
(SWAT). SWAT is a semi-distributed model, which effi- also used as input. Reservoir locations and characteristics
ciently simulates hydrology and water quality at different were derived from a 2010 national reservoir inventory of
scales. SWAT operates on daily time steps and has been the Ministerio de Medio Ambiente y Agua de Bolivia. The
extensively used to forecast the impact of management operation data of the six reservoirs were supplied by the
practices on water availability, sediments, and agricultural Universidad Mayor de San Andres (UMSA) Geology
chemical yield considering variations in soil characteris- Department. Weather input data (daily precipitation,
tics, land use, crop growth, and irrigation practices (Arnold maximum and minimum temperature, and daily solar
et al. 1998). radiation) were provided by the Servicio Nacional de
SWAT divides the watershed into sub-basins based on Meteorologı́a e Hidrologı́a de Bolivia (SENAMHI) for 115
the topographical characteristics. Later, these sub-basins stations. Because of the scarcity of stations in the southwest
are subdivided in hydrologic response units (HRUs) based and northern parts of the country, additional daily weather
on unique soil, land use, and average longitudinal slope data for 12 stations were obtained from the National
range properties given by the user. The model allows the Aeronautics and Space Administration (NASA) Langley
user to include up to five slope ranges. The calculations at Research Center POWER Project funded through the
each HRU related to the hydrologic parameters and nutri- NASA Earth Science Directorate Applied Science Program
ent transport were developed individually and aggregated (http://earth-www.larc.nasa.gov/cgibin/cgiwrap/solar/agro.
at the sub-basin and basin levels. Water losses were also [email protected]). River discharge data
calculated at each HRU and routed with the surface and required for calibration and validation were obtained from
subsurface runoff into each reach, sub-basin outlet, and SENAMHI for 13 stream gauges with flows that cover the
watershed outlet through the channel network (Neitsch timeline (1994–2008). Two stream gauges were used as
et al. 2002). input flows into the model for rivers that flow from
Precipitation, WY, ET, ETP, SR, SM, and GW were neighboring countries (Desaguadero and Madre de Dios);
obtained considering different formulas. SR, which is therefore, a total of 11 stream gauges were used for cali-
needed to obtain WY, was calculated using the modified bration and validation for 1997–2008. Records of areas
Soil Conservation Service (SCS) curve number equation. cultivated with soy, potato, corn, and rice were obtained
ETP, which is needed to obtain the ET, was calculated from the FAO Statistical Databases (FAOSTAT). With
using the Hargreaves equation (Neitsch et al. 2002). In respect to the projection of the baseline to 2050, monthly
developing countries, this equation is the most used average changes in precipitation and mean temperature
because of the reduced number of meteorological param- from 17 GCMs of 50-km resolution were obtained from
eters that are involved in the formulation (maximum and Santa Clara University (Maurer et al. 2007; www.
minimum temperature and solar radiation), especially for climatewizard.org). Further information related to CO2
daily simulation over a period of time. projections and definitions of emission scenarios were
obtained from the Intergovernmental Panel of Climate
Input data and model setup Change (IPCC) Fourth Assessment Report.
As part of processed data, the Spline Cubic interpolation
The required raw input data were compiled from different method (Shanchun et al. 2006) and Random-Forest algo-
local and international sources. A Digital Elevation Model rithm (Lawler et al. 2006) were used to fill gaps in the
(DEM) was extracted from an Advanced Spaceborne measured daily temperature and precipitation records,
Thermal Emission and Reflection Radiometer Global respectively.
Digital Elevation Model (ASTER-GDEM 2009) public The weather data for each sub-basin are assigned auto-
domain geographic database with a resolution of 90 m matically in SWAT using the closest weather station to the
(http://www.gdem.aster.ersdac.or.jp/). The land use map sub-basin’s center. Due to Bolivian topography and
was from the Superintendencia Agraria of Bolivia in 2000 microclimate characteristics, the closest weather station to

123
732 J. J. Escurra et al.

the center of a sub-basin often does not represent the pre- values higher than 1,800 mm. The 3D map considers the
cipitation of that sub-basin properly. As it is known, rep- annual precipitation average from 1997 to 2008.
resenting the proper precipitation gradient of the entire
country is crucial for reducing the error in the development Calibration and validation setup
of a hydrology model. Therefore, two steps were taken: (1)
the first step consisted of developing an additional Calibration and validation of baseline hydrology were
screening work to determine which station to use for each based on river discharge data from eleven stream gauges.
sub-basin based on local hydrologic studies; (2) the second The model was run monthly for 12 years (1997–2008). The
step consisted of using a coefficient for better representing period from 1997 to 2004 was used for calibration, and the
the precipitation in a specific area which is located between period from 1998 to 2008 was used for validation. Five
15.3 and 17.5S and between 69.7 and 64.2W. model performance evaluation methods were applied
According to studies (Roche et al. 1992; Andrade and (Moriasi et al. 2007): (1) Nash–Sutcliffe efficiency (EF),
Blacutt 2010), in that area, the precipitation abruptly (2) percent bias (PBIAS), (3) the coefficient of variation of
increases from 1,500 to more than 3,000 mm/year. Daily the root mean square error (CVRMSE), (4) mean absolute
precipitation data for the baseline from weather stations or error (ABSERR), and (5) the coefficient of determination
National Aeronautics and Space Administration (NASA) (R2). Equations to calculate the five performance methods
Langley Research Center POWER, which could properly are shown in S2 (Electronic Supplementary Material).
represent this increment, were not possible to obtain. To The ratings for the validation of the hydrologic model are
represent the precipitation gradient, two additional stations classified as follows: (1) it is considered to be ‘‘good’’ when
with daily precipitation from 1994 to 2008 were incorpo- EF is greater than 0.75 (close to the unity), PBIAS is within
rated as input. These two rainfall stations were generated ±20 %, CVRMSE and ABSERR are close to zero, and R2
by multiplying the daily precipitation values of the weather is close to unity (Stehr et al. 2009); (2) the rating is con-
stations: Camata (15.17S 68.77W) and Coripata (16.30S sidered to be ‘‘satisfactory’’ when EF is between 0.36 and
67.60W) by a coefficient of 2.2. The coefficient was 0.75, and PBIAS is between ±20 and ±40 % (Van Liew
obtained based on the comparison of the average monthly et al. 2005). For the study, it is considered to be ‘‘satisfac-
precipitation from 1997 to 2008 of the two weather stations tory’’ when the CVRMSE is between 0.40 and 0.75, the
and the average monthly precipitation from 1995 to 2005 ABSERR is between 10 and 30, and R2 is between 0.45 and
of six weather stations located in the area (Molina 2005). 0.75; (3) the rating is considered to be ‘‘moderate satisfac-
Altitudes of each of the six weather stations were also tory’’ when the EF is between 0.36 and 0.25, the PBIAS is
found and considered in the calculation of the coefficient. between ±40 and ±60 %, the CVRMSE is between 0.75
Figure 2 shows the location of the two rainfall stations, and 0.90, the ABSERR is between 30 and 60, and the R2 is
which were generated by multiplying for a coefficient, at a between 0.35 and 0.45; (4) all the rest of values outside the
3D topography and high annual precipitation map with described ranges are considered ‘‘poor.’’

Climate change analysis

As part of the climate change analysis, 17 GCMs were used

for the SRES A2 from the IPCC Fourth Assessment Report
(2007). The SRES A2 underlies the theme of self-reliance,
increasing population, regionally oriented economic
development, and preservation of local identities, which
represent the Bolivian situation. The nomenclatures of the
17 GCMs and their respective research centers where these
were developed are the following: Bjerknes Centre for
Climate Research (BCM2.0), Canadian Centre for Climate
Modelling and Analysis (CGCM3T47—T47 resolution),
Centre National de Recherches Meteorologiques
(CNRMCM3), Australia’s Commonwealth Scientific and
Industrial Research Organisation (CSIROMk3 and CSI-
ROMk3.5), Meteorological Institute, University of Bonn,
Meteorological Research Institute of KMA, and Model and
Fig. 2 Location of rainfall stations at a 3D map of topography and Data Groupe at MPI-M (ECHO-G), Geophysical Fluid
high annual average precipitation Dynamics Laboratory (GFDLCM2.0 and GFDLCM2.1),

123
Climate change impact 733

United Kingdom Met. Office (HADCM3 and HADGEM1), ETP using the annual mean temperature (ta) from the 17
Goddard Institute for Space Studies (GISSE-R), Institute GCMs was calculated using the Blanney and Criddle
for Numerical Mathematics (INMCM3.0), Institut Pierre equation (Allen and Pruitt 1986). The ta and pa values from
Simon Laplace (IPSLCM4), National Institute for Envi- the A2 SRES 2050 outputs of the 17 GCMs were calcu-
ronmental Studies (MIROC3.2 medres), Meteorological lated based on the ta and pa values from the baseline of the
Research Institute Japan (MRI-CGCM2.3.2), and National Climate Research Unit of the University of East Anglia
Centre for Atmospheric Research of United States (Hulme et al. 2008) and the p (%) and t (C) values
(NCARPCM and NCARCCSM3). obtained from 17 GCMs. Figure 3 shows the ranges of A2
Changes in monthly mean temperatures (C) and pre- SRES 2050 values of CMI from the 17 GCMs for the
cipitation (%) compared with the baseline were obtained eleven basins. The ranges are presented based on four CMI
from each GCM for the A2 SRES 2050 (2045–2065) and values coming from: (1) the 1st quartile (25 %) value; (2)
the GCM’s baseline (1961–1990). These changes in mean the immediate superior to the 1st quartile ([25 %) value;
temperature (t) and precipitation (pp) were spatially dis- (3) 3rd quartile (75 %) value; and (4) immediate inferior to
tributed in a Geographical Information System (GIS) raster the 3rd quartile (\75 %) value.
(1 9 1). Then, these values were aggregated for each of The values of CMI on the graph vary from -0.9 to
the eleven basins with the average weighing area. The CMI -0.1; this indicates that ETP is greater than pa in all
was obtained to select the values of changes in t and pp, basins. Basins with CMI values close to -0.9 are envi-
coming from the A2 SRES 2050 outputs of GCMs, for two ronments where water is scarce. After the CMIs for the 17
climate scenarios: ‘‘driest’’ and ‘‘wettest’’ (McCabe and GCMs were calculated, they were organized in ascending
Wolock 1992). The CMI was calculated using the formula order. The first quartile value (25 %) and its immediate
below (Eq. 1): superior were selected as representative CMIs for the driest
ETP pa scenario. The third quartile value (75 %) and its immediate
CMI ¼ 1  when pa  ETP or CMI ¼ 1 inferior were selected as representative CMIs for the wet-
pa ETP
when pa\ETP test scenario. This selection procedure was used to take the
most representative values, which reflect the boundaries of
ð1Þ
the range of CMIs without considering the maximum and
where ETP and pa are annual potential evapotranspiration minimum CMIs. Because there is still uncertainty in the
(mm) and annual precipitation (mm), respectively. The possible effects of climate change, there are outliers that

Fig. 3 Values of CMI from 17

GCMs (A2 SRES 2050) for the
eleven basins

123
734 J. J. Escurra et al.

could affect the analysis; therefore, it is not advisable to B2, B4, B8, and B11. These comparisons were used to
use the maximum and minimum CMIs. In Fig. 3, B3 pre- calculate the values of the performance indicators as part of
sents the lowest CMI and B9 presents the highest CMI. the validation process.
Average monthly changes in precipitation and mean
temperature for SRES A2 in 2050 from these two driest Hydrologic indicators
and wettest GCMs at each basin were calculated and
incorporated as input over the model baseline (1997–2008). The hydrologic indicators are indexes used to capture the
This method is called the delta change approach and it has water supply and demand at the basins; these indicators are
been widely used in several hydrologic impact studies represented by the precipitation, TARW, and ET. TARW
around the world. It is also known that this approach faces (total natural renewable surface, groundwater, and inflows
difficulties at recreating projected temporal dynamics from other countries) in the year 2000 was calculated for
(fluctuations) of precipitation and heavy precipitation 3114 sub-basins included in the model. This value was
events in the future (Seaby et al. 2013). However, ana- 604 km3/year aggregated for the entire country. The value
lyzing heavy precipitation events is not the scope of the is close to the annual renewable water value (622.5 km3/
work due to the known uncertainty of the GCMs outputs. year in 2000) proposed by The World’s Water (2008). In
Projected CO2 for 2050 was also considered as model addition, the values of annual renewable surface water
input; 510 ppm for SRES A2 was used following the range (TARWS) and groundwater (TARWG) in the year 2007
suggested by the IPCC Fourth Assessment Report (2007). were also compared with the estimates from the same years
SWAT predicts the impacts of changes in CO2 on plant proposed by The Atlas of Water (2009). Values of TARWS
growth though the calculation of radiation use efficiency and TARWG obtained from the model were 0.54 and 0.23
(Neitsch et al. 2002). As part of a study to examine the Gm3, respectively. The proposed Atlas of Water’s esti-
water balance of the Amazon Basin, a sensitivity analysis mates of TARWS and TARWG were 0.59 and 0.13 Gm3,
was achieved to analyze the impact for doubling atmo- respectively. These estimates proposed by The Atlas of
spheric CO2 concentrations (from 325 to 650 ppm). As a Water are also close to the simulated values from the
result, the canopy conductance would decrease by model, increasing the confidence in the results. Other
20–35 % depending on the vegetation characteristics and results of internal renewable groundwater resources and
annual river discharge would increase by between 3 and internally renewable water resources were calculated by
16.5 % (Costa and Foley 1997). This study highlights the the most recent version of the WaterGap Global Hydrology
need to consider CO2 concentrations during the climate Model or WGHM (spatial resolution of 0.5 by 0.5 daily
change assessment for river basin hydrology. time step). These estimates of internal renewable ground-
water and water resources for Bolivia, computed by
WGHM for the climate normal 1961–1990, were 146 and
Results and discussion 321 mm/year, respectively (Döll and Fielder 2008). These
values are also close to the model results (1997–2008) of
Calibration and validation internal renewable groundwater and water resources, which
are 141 and 426 mm/year, respectively.
The calibration procedure was used for each of the eleven Figure 5 shows the annual average of the precipitation,
basins individually, and it was done manually by changing TARW, and ET from 1997 to 2008 at each of the eleven
SWAT parameters, which respond to different hydrologic basins. Reasons for uncertainty in these results could
conditions of the basins. These parameters are shown in S3 include large size of basins, small number of operative
(Electronic Supplementary Material). The validation per- stream gauges for calibration, and relatively short timeline
formance of the validation was made based on the calcu- of the baseline (12 years). In addition, unusual weather
lation of the five performance evaluation methods (Van patterns in the country due to the complex topography
Liew et al. 2005; Moriasi et al. 2007; Stehr et al. 2009). make it difficult to estimate real precipitation. Improving
Figure 1 and Table 3 show the locations of the eleven the countrywide soil characteristic information and
stream gauges and the results of the performance of the groundwater depth data could help to better represent water
validation at each station. The overall results of the com- infiltration through the soil profile and increase accuracy
parison between the observed and simulated flows were for the calculation of the amount of the shallow aquifer
satisfactory based on the average of all the ratings [good recharge. The underestimation of peak flows by SWAT
(G) = 10; satisfactory (S) = 23; moderate satisfactory (Fig. 4) is attributed to the above causes and the fact that
(MS) = 12; poor (P) = 10]. macropore flow was not considered. Because of the lack of
Figure 4 shows the comparison of observed and simu- input data with adequate resolution and these causes of
lated flows at six gauge stations at the following basins: B1, uncertainty, the results are presented as a range of values.

123
Climate change impact 735

Table 3 Validation results for

Basin Station Performance Rating Basin Station Performance Rating
eleven stream gauge stations
indicator/value indicators

B1 Suchez EF/0.09 P B4 Villamontes EF/0.74 S

PBIAS/53.3 MS PBIAS/36.3 S
CVRMSE/0.98 MS CVRMSE/0.75 S
ABSERR/12.6 S ABSERR/99.6 P
R2/0.40 MS R2/0.75 S
Keka EF/0.49 S B8 Santa Rita de EF/0.54 S
B.A.
PBIAS/46.2 MS PBIAS/1.52 G
CVRMSE/0.87 MS CVRMSE/0.48 S
ABSERR/2.06 S ABSERR/79.3 MS
R2/0.68 S R2/0.59 S
B2 Calacoto EF/0.69 S Rurrenabaque EF/0.64 S
PBIAS/-2.31 G PBIAS/28.6 S
CVRMSE/0.49 S CVRMSE/0.44 MS
ABSERR/9.73 G ABSERR/670 P
R2/0.77 G R2/0.82 G
Mauri EF/0.15 P Riberalta EF/-0.85 P
PBIAS/14.3 G PBIAS/75.6 P
CVRMSE/0.76 MS CVRMSE/0.90 MS
ABSERR/7.38 G ABSERR/6407 P
R2/0.43 MS R2/0.83 G
Oruro EF/0.57 S B11 Ichilo EF/0.30 MS
PBIAS/6.36 G PBIAS/25.6 S
CVRMSE/0.64 S CVRMSE/0.59 S
ABSERR/30.3 MS ABSERR/219 P
R2/0.67 S R2/0.45 S
Mamore EF/0.04 P
PBIAS/1.52 G
CVRMSE/0.72 S
ABSERR/3644 P
R2/0.47 S

The range involves maximum, minimum, mean minus the flooding problems. B9, B8, and B3 present some values of
standard deviation (r), and mean plus r values from the the TARW range higher than those of the ET range. On the
yearly ET and TARW obtained from the model. In the case other hand, B5 and B4 have values of ET over 300 mm
of precipitation, the mean was used as the maximum value. greater than TARW.
It is interesting to observe that B11, with the highest Figure 6 shows the annual average values of the pre-
annual average precipitation ([1,800 mm), is in the only cipitation, ET, and TARW from 1997 to 2008 for each sub-
basin where the range of TARW is greater than the range of basin. The locations of the rain gauge stations, used as
ET. The percentage of forest land (forest, forest mixed, and input, are included in the map of precipitation.
wetlands forested) is close to 58 %. B11 is also the most The spatial distribution of the ET and TARW is influ-
affected by illegal and legal deforestation, which has been enced by the precipitation, complex topographical charac-
rapidly increasing since 1991 by more than 2.7 km2/year teristics of the country due to the Andes range of
(Aliaga et al. 2010). If the Amazonian tropical forest is mountains, and temperature. It is also observed that a high
replaced by degraded grass in a model, annual evapo- precipitation gradient has a direct impact on the calculation
transpiration is reduced by 30 % in the region (Nobre et al. of TARW (B11). As might be expected, the spatial distri-
1991). It is possible that the high precipitation and the bution of precipitation and ET appears to be correlated in
reduction in the forested area could justify the difference Fig. 7. The modeling process included evaporation from
between the ranges of TARW and ET, and it could lead to water bodies and from irrigated areas. As a result, we found

123
736 J. J. Escurra et al.

Fig. 4 Comparison between observed and simulated flows by SWAT (1998–2008) at the basins: B1, B2, B4, B8 and B11

a few areas where ET is greater than precipitation, but Hydrologic indicators impacted by climate change
these are small (500,000 ha) and are located where irri-
gated land (close to 10 % of the total area) and lakes are Based on the average for monthly change in temperature
present (excluding Lake Titicaca). Basins located in the (C) and precipitation (%) from these two GCMs for each
Plains region have higher values of ET due to higher scenario, the hydrologic indicators were recalculated and
temperatures and the area of vegetation cover is larger. compared with those obtained from the baseline

123
Climate change impact 737

Fig. 5 Range of hydrologic indicators (1997–2008) for the eleven basins

(1997–2008). Figure 7 shows a map of the changes in % of water deficits is as follows: B4, B2, B7, B3, B10, B1, B5,
the average annual hydrologic indicators coming from the B6, B9, B8, and B11. Projections indicate that these dif-
baseline (1997–2008) and driest/wettest scenarios in 2050 ferences tend to be greater than zero for all the basins
at each sub-basin. (Table 4).
Values of hydrologic indicators in 2050 for each basin
for the wettest and driest scenarios suggest which basins
are more vulnerable to water deficits, considering the
Summary and conclusions
possible impacts of climate change projected by 17 GCMs
for the SRES A2 in 2050.
A hydrologic model for the entire country of Bolivia was
Figure 8 shows the hydrologic indicators (precipitation,
developed, calibrated, and validated from 1997 to 2008.
ET, and TARW) obtained from the model for the baseline
The largest barrier to the development of the study was the
(1997–2008), wettest scenario (2050 SRES A2), and driest
lack of information; however, available information was
scenario (2050 SRES A2). There are no basins in which the
compiled and adjusted based on: (1) algorithms that con-
future climate scenarios project only water surplus. Possi-
sider spatial and temporal current and historical data; (2)
ble water surplus or deficit is projected for B1, B9, and
interviews with local experts; and (3) reviews of govern-
B10. The reason is the tendency of most of the GCMs to
ment and scientific documents. Using the outputs for mean
project a decrease in precipitation at the basin level for the
temperature and precipitation from 17 GCMs for A2 SRES
driest and wettest scenarios. As it is expected, ET for the
2050, the wettest and the driest projections for changes in
wettest and driest scenarios tends to increase due to the
temperature and precipitation at basin level were calcu-
temperature increase projected by the GCMs (1.5–4 C)
lated, based on the calculation of the CMI from each of the
with a few exceptions due to reduction in evaporation in
17 GCMs outputs. The wettest and driest climate scenarios
places where precipitation has been considerably
(as well as the projected change in atmospheric carbon
decreased.
dioxide) were used to simulate hydrologic indicators
The differences between the TARW and precipitation
(precipitation, ET, and TARW) for each scenario at basin
ratio in % (TPr) for the baseline and the mean of TPr for
and sub-basin levels. The conclusions of the study are as
both climate scenarios were calculated at each basin. These
follows:
differences were organized in descendent order to identify
the basins most susceptible to water deficits due to climate • The validation procedure in all the basins (using the
change effects. The order of priority for basins most vul- monthly flows from 1998 to 2008) from eleven
nerable is as follows: B11, B5, B3, B6, B1, B8, B7, B10, hydrometric stations produced satisfactory model per-
B9, B2, and B4. Taking into account only the baseline formance. Moderately satisfactory simulation of basin
period, the order of priority for basins most susceptible to hydrology in basins without stream gauge was achieved

123
738 J. J. Escurra et al.

Fig. 6 Annual Average of the hydrologic indicators (1997–2008) at sub-basin level

using stream data from governmental reports and • As result of the climate change analysis, both future
scientific publications. climate scenarios project a decrease in TARW based on
• As result, from the water balance analysis for the the CMIs from the 17 GCMs, suggesting a drier future.
12 years of baseline, basin B11 (precipita- However, potential isolated problems due to surplus of
tion [ 1,800 mm) is the only basin where a surplus water (floods) cannot be discounted as climate change
of water is projected. On the other hand, B5 and B4 impacts, even in the hypothetical case of the dry
face substantial water deficits (ET— scenario. The order of priority for the basins most
TARW [ 300 mm). This analysis does not consider vulnerable to water deficit due to climate change effects
water consumption for domestic use. is as follows: B11, B5, B3, B6, B1, B8, B7, B10, B9,

123
Climate change impact 739

Fig. 7 Changes in % of annual

average hydrologic indicators
between the baseline
(1997–2008) and the wettest/
driest scenarios (2050 SRES
A2)

123
740 J. J. Escurra et al.

Fig. 8 Comparison of the

hydrologic indicators for the
baseline (1997–2008), wettest
scenario (2050 SRESA2), and
driest scenario (2050 SRESA2)
at each basin

123
Climate change impact 741

Table 4 Difference between the TPr for the baseline (1997–2008) well as several universities in La Paz, Cochabamba and Potosı́ helped
and the mean TPr for the wettest and driest scenarios (2050 SRES A2) to develop the initial knowledge base used in the study. A large group
of other individuals also provided useful comments including Allan
Basins 1: TPr 2: TPr (mean for Difference Jones, Bénédicte Augeard, and Jacob Burke.
(baseline) both scenarios) (2–1)

B11 73 52 21
B5 38 23 15
B3 29 20 8 References
B6 47 40 7
Aliaga J, Villagas H, Leguı́a D (2010) Documento de Trabajo No.
B1 32 27 6 17/10 Deforestación en Bolivia: Una aproximación espacial.
B8 53 49 4 Instituto de Investigaciones Socio Económicas, Universidad
B7 19 14 4 Católica Boliviana, Bolivia
Allen RG, Pruitt WO (1986) Rational use of the FAO Blaney-Criddle
B10 31 27 4
formula. J Irrigation Drain Eng 112(2):139–155
B9 51 47 3 Andrade MF, Blacutt BL (2010) Evaluación del modelo climático
B2 19 16 3 regional PRECIS para le área de Bolivia: comparación con datos
B4 17 16 1 de superficie. Revista Boliviana de Fı́sica 16(16):1–12
Arnold JG, Srinivasan R, Muttiah RS, Williams JR (1998) Large area
hydrologic modeling and assessment part I: model development.
B2, and B4. Taking into account only the baseline J Am Water Resour Assoc 34(1):73–89
Barry R (1990) Changes in mountain climate and glacio-hydrologic
period, the order of priority for the basins most responses. Mt Res Dev 10:161–170
susceptible to water deficit is as follows: B4, B2, B7, Black M, King J (2009) The atlas of water: mapping the world’s most
B3, B10, B1, B5, B6, B9, B8, and B11. The three first critical resource. University of California Press, Berkeley
basins have a TPr value lower than 20 %. None of the Bradley RS, Vuille M, Diaz HF, Vergara W (2006) Threats to water
supplies in the tropical Andes. SCIENCE-NEW YORK THEN
basins projects water surplus as potential climate WASHINGTON, 1755
change impact at basin level; however, this is observed Caballero Y, Jomelli V, Chevallier P, Ribstein P (2002) Hydrological
at sub-basin level. characteristics of slope deposits in high tropical mountains
• It is observed that both scenarios at sub-basin level (Cordillera Real, Bolivia). Catena 47(2):101–116
Central Intelligence Agency (CIA) (2009) The World Factbook 2009.
imply that precipitation would considerably decrease in GPO, Washington. Web. 4 Mar 2009
the southern part of B1. In addition, there is a Centro Digital de Recursos Naturales de Bolivia (2009) Texas A&M
substantial reduction in ET in the southern part of B1 University. 12 Dec 2009 http://essm.tamu.edu/bolivia/
where precipitation has declined. B11 is projected to Costa MH, Foley JA (1997) Water balance of the Amazon Basin:
dependence on vegetation cover and canopy conductance.
have the greatest increase in ET in both climate J Geophys Res Atmos (1984–2012), 102(D20), 23973–23989
scenarios. The TARW in the wettest scenario is also Döll P, Fiedler K (2008) Global-scale modeling of groundwater
reduced due to a decrease or a slight increase in recharge. Hydrol Earth Syst Sci 12:863–885. doi:10.5194/hess-
precipitation and great expected increase in ET (influ- 12-863-2008
Food and Agriculture Organization, FAO (1997) FAO/TCP-BOL/
enced by the temperature). In addition, some sub-basins 6611: Delimitación de Cuencas Hidrográficas de Bolivia. La Paz
(9,700 km2) would face an increase in TARW of more Hulme M, Conway D, Jones PD, Jiang T, Barrow EM, Turney C
than 35 % in the wettest scenario. However, very few (2008) Construction of a 1961–1990 European climatology for
sub-basins would face an increase of more than 35 % in climate change modelling and impact applications. Int J Climatol
15(12):1333–1363
the driest scenario. A decrease in TARW in both Instituto Nacional de Estadı́stica (INE) (2010) Información Demog-
scenarios confirms the conclusion of the climate change ráfica, La Paz
assessment for the countrywide water balance (MDP Johnson AM (1976) The climate of Perú, Bolivia and Ecuador. In:
and VPTyMA 2007). However, problems due to Schwerdtfeger W (ed) Climates of Central and South America.
Elsevier, Amsterdam, pp 147–218
surplus of water (floods) should not be discounted as Lawler JJ, White D, Neilson RP, Blaustein AR (2006) Predicting
potential climate change impacts, even in the case of climate-induced range shifts: model differences and model
the driest scenario. reliability. Glob Chang Biol 12(8):1568–1584
Lobina E (2000) Cochabamba: water war. Focus Public Serv 7(2):5–10
Maurer EP, Brekke L, Pruitt T, Duffy PB (2007) Fine-resolution
Acknowledgments This work was undertaken as part of the climate projections enhance regional climate change impact
Technical Assistance Water-Related Adaptation to Climate Change studies. Eos Trans AGU 88(47):504
and Variability in Bolivia supported by the Environment Unit of the McCabe GJ Jr, Wolock DM (1992) Effects of climatic change and
Latina American Region (LCSEN) of the World Bank, managed by climatic variability on the Thornthwaite moisture index in the
Karin Kemper. Thanks are in order for the efforts and inputs of the Delaware River basin. Clim Chang 20(2):143–153
many members of the assessment team, including Humberto Perotto, Meyer HW (1992) Lapse rates and other variables applied to
Marco Andrade, Bruno Condori, and William Farmer. The Servicio estimating paleoaltitudes from fossil floras. Palaeogeogr Palae-
Nacional de Hidrografia y Metereologia de Bolivia (SENAMHI) as oclimatol Palaeoecol 99:71–99

123
742 J. J. Escurra et al.

Ministerio de Desarrollo Sostenible y Medio Ambiente, MDSMA Roche M, Fernández C, Aliaga A, Peña J, Salas E, Montaño JL (1992)
(1997a) Análisis de Vulnerabilidad y Adaptación de los Balance Hı́drico Superficial de Bolivia. La Paz
Ecosistemas al Cambio Climático y Opciones de Adaptación. Ronchail J, Labat D, Callede J, Cochonneau G, Guyot JL, Filizola N,
La Paz De Oliveir A (2005) Discharge variability within the Amazon
Ministerio de Desarrollo Sostenible y Medio Ambiente (MDSMA) basin. Regional hydrological impacts of climate change-Impact
(1997b) Vulnerabilidades de los Ecosistemas al Posible Cambio assessment and decision making. IAHS Press, IAHS Publ, 296,
Climático y Análisis de Mitigación de gases de Efecto Inver- 21–30
nadero. La Paz Seaby LP, Refsgaard JC, Sonnenborg TO, Stisen S, Christensen JH,
Ministerio de Planificación del Desarrollo (MDP), Viceministerio de Jensen KH (2013) Assessment of robustness and significance of
Planificación Territorial y medio Ambiente (VPTyMA) (2007) climate change signals for an ensemble of distribution-based
El Cambio Climático en Bolivia (Análisis, sı́ntesis de impactos y scaled climate projections. J Hydrol 486:479–493
adaptación). La Paz Shanchun F, Qiuli Z, Jie Q, Pengcheng Z (2006) The temperature
Molina J (2005) Régimen de precipitación en la cuenca de Huarinilla- compensation of high precision pressure sensor based on the
Cotapata. Ecologı́a en Bolivia 40(1):43–55 cubic spline interpolation. J Metrol Soc India 21(3):167–170
Moriasi DN, Arnold JG, Van Liew MW, Bingner RL, Harmel RD, Siebert S, Döll P (2001) A digital global map of irrigated areas—an
Veith TL (2007) Model evaluation guidelines for systematic update for Latin America and Europe. Report A0102, Center for
quantification of accuracy in watershed simulations. Trans Environmental Systems Research, University of Kassel, Kurt
ASABE 50(3):885–900 Wolters Strasse 3, 34109 Kassel
Neitsch SL, Arnold JG, Kiniry JR, Wiliams JR, King KW (2002) Soil Siebert S, Hoogeveen J, Frenken K (2006) Irrigation in Africa,
and water assessment tool theoretical documentation version Europe and Latin America. Update of the digital global map of
2000. GSWRL Report 02-01, BRC Report 02-05, TR-191. Texas irrigation areas to version 4. Frankfurt Hydrology Paper 05,
Water Resources Institute, College Station University of Frankfurt (Main), Germany and FAO, Rome
Nobre CA, Sellers PJ, Shukla J (1991) Amazonian deforestation and Stehr A, Debels P, Arumi JL, Romero F, Alcayaga H (2009)
regional climate change. J Clim 4(10):957–988 Combining discharge data and MODIS imagery for evaluating
Peredo E (2004) Water, privatization and conflict: women from the the performance of SWAT and its snowmelt routine in a small
Cochabamba Valley. Fundación Solon (Bolivia) and Heinrich Andean basin, Chile. J Hydrol Sci 54:1053–1067
Boll Foundation North America, La Paz Van Liew MW, Arnold JG, Bosch DD (2005) Problems and potential
Ribstein P, Tiriau E, Francou B, Saravia R (1995) Tropical climate of autocalibrating a hydrologic model. Trans Am Soc Agric
and glacier hydrology: a case study in Bolivia. J Hydrol 48(3):1025–1040
165(1):221–234 Viceministerio de Recursos Hidricos y Riego (VRHyR) (2010)
Roche M, Jauregui CF (1988) Water resource, salinity and salt yields Inventario Nacional de Presas, Bolivia 2010. La Paz
of the rivers of the Bolivian Amazon. J Hydrol 101(1):305–331

123