FORECAST APPLICATIONS IN AGRICULTURE:

APPROACHES, ISSUES, AND CHALLENGES

JENNIFER G. PHILLIPS and JAMES W. HANSEN

International Research Institute for Climate Prediction
(IRI), PO Box 1000, 61 Rte. 9W, Monell Building,
Palisades, New York 10964-8000, USA

Abstract
We discuss the path along which agricultural
applications research has evolved, and current approaches that
attempt to meet the challenge of utilizing seasonal forecasts in
farm management. Approaches used have evolved over the
period since they first become available in the mid 1980s.
From an early phase of searching for locations around the world
where predictable inter-annual variability influenced crop
outcomes and improving our understanding of the physical
interactions between climate and crop growth, our current
interest rests more in the human-oriented issues of perception of
probabilistic information and implications for decision making.
For small-scale and commercial farmers alike, we argue that
improving the comprehension of uncertainties associated with
climate outcomes and their implications holds the largest hope
for successful applications of forecasts to farm management.

167

1. Introduction
Applications of seasonal climate forecasts have been
contemplated for a number of years now in many places around
the world, with the agricultural sector considered one of the
most likely to receive benefits. Approaches to the promotion
and improved use of forecasts in farm management have
evolved over that time, yet a number of obstacles remain. From
an early stage of optimism about revolutionizing agriculture,
when anticipated benefits to dryland agricultural production
were heralded as the next Green Revolution (Cusack, 1983;
Sah, 1987), we now have a more realistic understanding of the
opportunities and limitations to applying seasonal forecasts, and
have more clearly identified the challenges.
Given the ubiquitous nature of the farming enterprise,
experience can be drawn from a wide range of farming systems
applications, across many continents, cultures, and levels of
development. Although details of the uses of seasonal forecasts
may differ from system to system, we will argue here that many
of the challenges are common to all. All farmers are faced with
managing a complex production system under a range of
constraints. Decisions are made with imperfect information
regarding future states of a number of variables, often
dominated by the uncertainty regarding weather outcomes.
Seasonal climate forecasts constrain the distribution of possible
outcomes of weather more narrowly than the climatological
distribution in the absence of forecasts. Yet the decisionmaking process remains the same (or should remain the same,
as we argue), if the new information is correctly understood.
One of the biggest challenges, then, is ensuring that decision
168

makers have a solid understanding of the expected distribution

of climate outcomes and its implications for their production
options.
In this paper, we present some background on the path
that has led us to the current focus of research, then expand on
the issue of communication and approaches being developed to
address the technology transfer challenge.

2. Early Approaches in Agricultural Applications

Early insights into the potential ramifications of El
Nio-related inter-annual climate variability were discovered
through simple correlation analysis. Given the large-scale
nature of the influence of Pacific sea-surface temperatures
(SSTs) on global climate, teleconnections with seasonal
variation in precipitation at the regional or national scale could
easily be detected through straightforward statistical tools. The
earliest, and now famous, example of this was Sir Gilbert
Walkers work attempting to understand the causes of the
massive drought and ensuing famine in India at the end of the
19th century. Walker was the first to make the connection
between inter-annual oscillations in sea-level atmospheric
pressure in the southern Pacific and the episodic failure of the
Indian monsoon.
Since then, many others have confirmed the oceanatmosphere links, but the next step toward understanding
implications for crop production was slow in coming. Early
correlation work between SSTs and crop production in the
United States (Handler and Handler, 1983) and Australia
169

(Nicholls, 1985), went unnoticed until late 1980s and early

1990s, when evidence of predictability of ENSO was
confirmed (Cane, Zebiak and Dolan, 1986). In 1994, Mark
Cane recognized the ENSO signal in a time series of national
maize yields in Zimbabwe, gaining international attention with
a publication in the journal Nature (Cane et al., 1994). If future
states of ocean surface temperature were predictable, then crop
production (and by extension food shortfalls) would also be
predictable, with large implications for food security
management.
In that early phase of applications work, such analyses
were used mainly as an exploratory tool, looking for ENSO
impacts where they were not easily found by looking only at the
meteorological records (Hansen et al., 2000), or to identify
areas of potential ENSO forecast applications (e.g. Carlson et
al., 1996; Posest et al., 1999; Zubair and Somasundea, 2000).
It has been argued (e.g., Rosenzweig, 1994), somewhat
successfully, that examination of ENSO signals in monthly or
seasonal mean climate statistics can miss agriculturallyimportant climatic responses to ENSO. This is because plant
growth can integrate precipitation and temperature sequences
thereby revealing impacts of global circulation anomalies
through regional yields that might otherwise go undetected.
This turned out to be useful both for revealing regions where
ENSO has an influence, and also for pointing out some of the
complexities of plant-environment-human interactions (e.g. the
difficulties of finding a signal in the yields of paddy rice). In
order to move beyond descriptive studies to understanding
implications for management, other tools were needed.
170

3. Experimenting with Decision Scenarios Using

Crop Simulation Models
Approaches originally developed in the context of
climate change impacts on agriculture using crop simulation
modeling were soon adopted by the research community
interested in inter-annual climate prediction. At the simplest
level, crop models could be driven with historical time series of
observed climate data and results sorted by ENSO phase to gain
insight into the mechanisms of ENSO impacts. For example,
Phillips et al. (1999) combined statistical and modeling
approaches to investigate the impacts of El Nio and La Nia
phases on maize yields in the United States Cornbelt. The
statistical analysis revealed an association between regional
maize yields and Pacific SSTs (R2 = 0.39), but mapping of the
district-level yields by ENSO phase showed high spatial
variability in the response. The use of crop growth simulations
revealed that lower maize yields in La Nia years were
associated with both accelerated development resulting from
higher temperatures in July and August, and also with water
stress during grainfill. As a number of other studies have
pointed out (e.g. Meinke and Hochman, 2000), soil type, depth,
and initial soil moisture influence the severity of the impact.
Studies of this sort carried implications for management but
were not explicitly designed to test decision alternatives.
Other studies have taken this approach one step further,
using simulation models explicitly to investigate potential
management strategies.
This approach has been used
extensively in Australia to assess alternative cropping strategies
(e.g., Meinke et al., 1996 (peanut); Hammer et al., 1996
171

(wheat); Meinke and Hochman, 2000 (wheat and sorghum)).

One important conclusion from this body of work is that
variation in a number of factors, from the temporal distribution
of rainfall to spatial distribution of soil type or initial moisture
conditions, lead to a wide range of yield outcomes within any
given ENSO phase. This recognition has helped to foster
awareness of the importance of the probabilistic nature of
forecasts to making appropriate farm management decisions
(e.g. Meinke and Hochman, 2000; Ferreyra et al., 2001).
In additional to the ability to investigate production
outcomes as a function of ENSO phase and alternatives such as
planting date (Phillips et al., 1998), cultivar (Phillips and
Rajagopalan, 1999), or land allocation (Messina et al., 1999),
simulation allows for cost/benefit analysis and risk assessment.
Hansen (2000) shows the importance of identifying ones risk
tolerance in evaluating outcomes of alternatives under a range
of climate scenarios.
Using the example of farm land
allocation between maize, soybean, and sunflower based on the
work of Messina et al. (1999), he compares the risk-adjusted
value of forecast information between risk neutral decision
makers and those that are risk averse, varying their perception
of the uncertainty in the forecast. He shows that for risk averse
farmers who interpret the forecast deterministically that is,
assuming the mean tendency of an ENSO phase is the expected
outcome - the value of forecast information is negative over the
long run due to exposure to excessive income risk. In
comparison, a risk averse farmer correctly considering the
probabilistic nature of the forecast information responds with
decisions that result in a net positive value over the long run.
172

4. Communicating Forecast Uncertainty and

Implications for Farm Decision Making
Analyses of historic agricultural data, simulation
experiments with decision scenarios, and experience working
with farmers has clearly highlighted the importance of (1) the
variability of impacts of climate fluctuations both in space and
time, and (2) understanding the resulting forecast uncertainties
associated with any skillful but imperfect forecasts, and their
implications for how forecast information is used in farm
management (Phillips et al., 2001; Hansen, 2000; Jones et al.,
2000; Hammer et al., In press).
With respect to forecast uncertainty, the issue is not one
of poor forecasts we are assuming that forecast quality is
high. Rather, it is the intrinsically probabilistic nature of the
forecast, and of climate in general, that is crucial to the decision
making process. We have found that the concept of a
prediction is often taken to mean a deterministic indication of
a future event (Letson et al., 2000; Phillips et al, 2001); this
interpretation is what often leads to poor decisions. Farmers are
generally concerned not only with averages, but with the
stability of production and reliability of income. This is
particularly true of smallholder, resource-poor farmers in
climate-sensitive environments. Farmers employ a range of
risk management strategies to reduce their vulnerability to
consequences of a poor year, generally at the cost of some
reduction of average production, resource use efficiency, and
sometimes sustainability of natural resources. In the absence of
seasonal climate forecasts, farmers clearly factor into their
decisions some mental representation of the distribution of
173

possible weather outcomes derived from their understanding of

local climate. Although there is evidence that this mental
representation can be distorted from the real distribution
(Weber, 1997), it none the less considers a range of outcomes
rather than a single expected event. Thus, in order to support
the appropriate use of seasonal climate forecasts, our task is to
present them clearly as shifts from the climatological
distribution.
Regarding variation in impacts across space and time, it
can be argued that prescriptive advice regarding the correct
usage of a forecast may be difficult to provide (Jones et al.,
2000). More importantly, conveying the range of possible
climate outcomes, and implications for the potential impacts on
crop response may be the most effective aspect to
communicate.
Helping farmers become aware of the
importance of climate variability, and equipping them with the
tools to analyze and assess their own farm condition,
preferences and risk tolerance will help them make use of
probabilistic climate forecasts and manage climate variability.
Although most of the crop simulation studies mentioned
above are based on scenarios depicting large-scale commercial
farmers, the points of greatest concern are arguably relevant to
all scales of farming. While some argue that large-scale, betterendowed farmers are likely to reap most of the benefit from
seasonal forecasts because they have greater access to resources
and information, and hence, greater decision capacity (e.g.,
Broad and Agrawala, 2000), others argue that resource-poor
smallholder farmers will potentially benefit more due to their
greater vulnerability to impacts of climate variability, leading
perhaps to greater motivation to use forecast information (e.g.,
174

Barrett, 1998). The issue of risk may be of greater importance

for small-scale, resource-limited farmers, who appropriately
tend to be more risk averse than relatively wealthy farmers.
Where conditions allow, resource-poor farmers seek to spread
risk by cultivating a wider variety of crops and utilizing offfarm employment more frequently than large-sale farmers,
making their decision environment even more complex. A
focus on communicating the probabilistic nature of the
information seems therefore to be particularly critical in the
case of smallholder farmers.
Communicating a basic understanding of the
uncertainties in forecast information is a current focus of
research on applications of seasonal climate forecasts to
agricultural management at the IRI. To achieve this we are
developing hands-on training materials that address the topic of
understanding climate variability in general, decision making in
the face of climate uncertainty, and the implications of seasonal
forecasts within that probabilistic framework. Crop simulation
modeling is seen as one of the tools that can help convey this
understanding by making potential crop management outcomes
explicit. Moving beyond simple ENSO-based climate scenarios,
we are working on developing methods to express model-based
forecasts of seasonal climate in the format necessary to drive
crop simulations (see IRI, 2000). Understanding how people
perceive uncertainty and make decisions under risk is a
necessary ingredient in developing training programs to help
people use seasonal forecasts wisely.

175

5. Summary
Research on the application of seasonal climate
forecasts has progressed from a focus on the physical impacts
of inter-annual climate variability on crop production to a more
human-centered consideration of how farmers can incorporate
this probabilistic information into decisions to minimize
production risks. It has become clear over the last few years of
working in this field that misinterpretation of what the forecast
represents can actually lead to poorer decisions and increased
exposure to risk. However, it can be shown that correct
interpretation of forecasts, and their wise use in farm
management will lead to improvements in overall agricultural
productivity, with likely benefits to resource use efficiency and
environmental sustainability. The greatest task before us is to
learn, along with the farm community, how to express forecast
information in such a way that the full range of possible
outcomes of various decision alternatives are made explicit, and
that farmers have the tools to apply that information to their
own specific decision context. Increasing awareness of climate
interactions with the farming enterprise will help farmers better
manage climate variability with or without a strong forecast. A
partnership between climate scientists, the agricultural research
community, and farmers on the ground will be necessary to
achieve this goal.

176

References
BARRETT, C.B. The value of imperfect ENSO forecast information:
discussion. Amer. Journal Agr. Econ. v.80, p.1109-1112, 1998.
BROAD, K.; AGRAWALA, S. The Ethiopian food crisis uses and
limits of climate forecasts. Science, v.289, p.1693-1694, 2000.
CANE, M.A.; ESCHEL, G.; BUCKLAND, R.W. Forecasting
Zimbabwean maize yield using eastern equatorial Pacific sea
surface temperature. Nature, v.370, p.204-205, 1994.
CANE, M.A.; ZEBIAK, S.E.; DOLAN, S.C. Experimental forecasts
of El Nio. Nature, v.321, p.827-832, 1986.
CARLSON, R.E.; TODEY, D.P.; TAYLOR, S.E. Midwestern corn
yield and weather in relation to extremes of the southern
oscillation. Journal of Production Agriculture, v.9, p.347352, 1996.
CUSACK, D.F. Introduction. In: CUSACK, D.F., ed. Agroclimate
information for development: reviving the green revolution.
Boulder: Westview Press, 1983. p.xii-xvi.
FERREYRA, R.A.; PODEST, G.P.; MESSINA, C.D.; LETSON,
D.; DARDANELLI, J.; GUEVARA, E.; MEIRA, S. A linkedmodeling framework to estimate maize production risk
associated with ENSO-related climate variability in Argentina.
Agricultural and Forest Meteorology, v.107, p.177-192, 2001.
HAMMER, G.L.; HOLZWORTH, D.P.; STONE, R. The value of
skill in seasonal climate forecasting to wheat crop management
in a region with high climate variability. Australian Journal of
Agricultural Research, v.47, p.717-773, 1996.
HAMMER, G.; HANSEN, J.W.; PHILLIPS, J.; MJELDE, J.W.;
HILL, H.S.J.; LOVE, A.; POTGIETER, A. Advances in
177

application of climate prediction in agriculture. Agricultural

Systems. In press.
HANDLER, P.; HANDLER, E. Climatic anomalies in the tropical
Pacific Ocean and corn yields in the United States. Science ,
v.220, p.1155-1156, 1983.
HANSEN, J.W. Applying climate prediction in a risk management
framework: farm land allocation in the Argentine Pampas. In:
ANNUAL CLIMATE DIAGNOSTICS AND PREDICTION
WORKSHOP, 25., 2000, Palisades, New York. USA.
Proceedings...
HANSEN, J.W.; ROSENZWEIG, C.; GOLDBERG, R. 2000. Global
cereal production and the El Nio-Southern Oscillation. In: 2000
Agronomy Abstracts. Am. Soc. Agronomy, Madison, WI,
USA.
IRI. 2000. Summary Report of the Workshop: Linking Climate
Prediction Model Output with Crop Model Requirements.
IRI-CW/00/2. International Research Institute for Climate
Prediction, Palisades, New York.
JONES, J.W.; HANSEN, J.W.; ROYCE, F.S.; MESSINA, C.D.
Potential benefits of climate forecasting to agriculture.
Agriculture, Ecosystems and Environment v.82, p.169-184,
2000.
LETSON, D.; LLOVET, I.; PODEST, G.; ROYCE, F.; BRESCIA,
V.; LEMA, D.; PARELLADA, G. User perspectives of climate
forecasts: crop producers in Pergamino, Argentina. Climate
Research. Submitted.
MEINKE, H.; STONE, R.C.; HAMMER, G.L. SOI phases and
climate risk to peanut production: a case study for Northern
Australia. International Journal of Climatology, v.16, p.783789, 1996.
178

MEINKE, H.; HOCHMAN, Z. Using seasonal climate forecasts to

manage dryland crops in northern Australia Experiences from
the 1997/98 seasons. In: HAMMER, G.L.; NICHOLLS, N.;
MITCHELL, C., ed. Applications of seasonal climate
forecasting agricultural and natural ecosystems. Dordrecht:
Kluwer, 2000. p.149-165.
MESSINA, C.D.; HANSEN, J.W.; HALL, A.J. Land allocation
conditioned on El Nio-Southern Oscillation phases in the
Pampas of Argentina. Agricultural Systems, v.60, p.197-212,
1999.
NICHOLLS, N. Impact of the Southern Oscillation on Australian
crops. Journal of Climatology, v.5, p.553-560, 1985.
PHILLIPS, J.G.; CANE, M.A.; ROSENZWEIG, C. ENSO, seasonal
rainfall patterns, and simulated maize yield variability in
Zimbabwe. Agricultural and Forest Meteorology, v.90, p.3950, 1998.
PHILLIPS, J.G.; RAJAGOPALAN, B. Use of seasonal climate
forecasts for selecting maize maturity class. In: 1999
Agronomy Abstracts. Am. Soc. Agronomy, Madison, WI,
USA.
PHILLIPS, J.G.; RAJAGOPALAN, B.; CANE, M.A.;
ROSENZWEIG, C. The role of ENSO in determining climate
and maize yield variability in the U.S. Cornbelt. International
Journal of Climatology, v.19, p.877-888, 1999.
PHILLIPS, J.G.; UNGANAI, L.; MAKAUDZE, E. Current and
potential use of seasonal climate forecasts for resource-poor
farmers in Zimbabwe. In: Impacts of El Nio and climate
variability on agriculture. Madison: American Society of
Agronomy, 2001. (ASA. Special Publication, 63).
179

PODEST, G.P.; MESSINA, C.D.; GRONDONA, M.O.; MAGRIN,

G.O. Associations between grain crop yields in Central-Eastern
Argentina and El Nio-Southern Oscillation. Journal of
Applied Meteorology, v.38, p.1488-1498, 1999.
ROSENZWEIG, C. Maize suffers a sea-change. Nature, v.370,
p.175-176, 1994.
SAH, R.K. Tropical economies and weather information. In: FEIN,
J.S.; STEPHENS, P.L., ed. Monsoons. New York: John Wiley
& Sons, 1987. p.105-119.
WEBER, E.U. Perception and expectation of climate change:
precondition for economic and technical adaptation. In:
BAZERMAN, M.H.; MESSICK, D.M., ed. Environment, ethics
and behavior: the psychology of environmental valuation and
degradation. San Francisco: Lexington, 1997. p.314-341.
ZUBAIR, L.; SOMASUNDEA, S. ENSO influences on rice
production in Sri Lanka. In: INTERNATIONAL FORUM ON
CLIMATE PREDICTION, AGRICULTURE AND
DEVELOPMENT. Proceedings... Palisades, New York:
International Research Institute for Climate Prediction, 2000.
p.179-184. IRI-CW/00/1.

180