Agricultural Systems 60 (1999) 197±212
www.elsevier.com/locate/agsy
Land allocation conditioned on El NinÄo-Southern Oscillation
phases in the Pampas of Argentina$
C.D. Messina a, J.W. Hansen b,*, A.J. Hall a
a
IFEVA, Facultad de AgronomõÂa, Universidad de Buenos Aires, 4453 Avda. San Martin, Buenos Aires 1417, Argentina
b
Agricultural and Biological Engineering, University of Florida, PO Box 110570, Gainesville, FL 32611-0570, USA
Received 31 January 1999; received in revised form 16 April 1999; accepted 19 April 1999
Abstract
The El NinÄo-Southern Oscillation (ENSO) contributes to the vulnerability of crop production to climate variability
in the Pampas region of Argentina. Predictability of regional climate anomalies associated with ENSO may provide
opportunities to tailor decisions to expected climate, either to mitigate expected adverse conditions or to take advantage of favorable conditions. Model analysis was used to explore the potential for tailoring land allocation among
crops to ENSO phases at the farm scale in two sub-regions of the Pampas. The model identi®es as a function of risk
preferences and initial wealth the crop mix that maximizes expected utility of wealth at the end of a 1-year decision
period based on current costs and prices, and crop yields simulated for each year of historical weather. The model
reproduced recent land allocation patterns at the district scale under moderate risk aversion, and predicted increasing
diversi®cation with increasing risk aversion. Di€erences in land allocation among ENSO phases were consistent with
known climate response to ENSO, and crop response to water availability. Tailoring land allocation to ENSO phase
increased mean net farm income between US$5 and $15 haÿ1 yearÿ1 relative to optimizing the crop mixture for all
years, depending on location, risk aversion and initial wealth. The relationship between potential value of ENSO
information and risk aversion was not monotonic, and di€ered between locations. Crop mix and information value
also varied with crop prices and initial soil moisture. There are potential ®nancial bene®ts of applying this approach to
tailoring decisions to ENSO phases. # 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Climate prediction; El NinÄo-Southern Oscillation; Risk; Expected utility
1. Introduction
Climate variability is the major source of agricultural production risk in Argentina. The Pampas,
* Corresponding author. Fax: +1-352-392-4092.
E-mail address: [email protected]¯.edu (J.W. Hansen)
$
Florida Agricultural Experiment Station, Journal Series
No. R-06795.
the main agricultural region in Argentina, is characterized by high interannual variability of rainfall
(CV ˆ 28%) and rainfed crop production (Hall et
al., 1992). Climate risk increases toward the western part of the region due to an east±west gradient (1000±600 mm yearÿ1) in precipitation. The
CV of crop yields in the western districts ranges
from 11% for sun¯ower to 20% for soybean.
Median crop losses (i.e. planted area that is not
0308-521X/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S0308-521X(99)00032-3
198
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
harvested) range from 15% for wheat to 35% for
sun¯ower. Farmers' aversion to income risk leads
to conservative management strategies that reduce
mean productivity (Keating et al., 1991; Kingwell,
1994; Thornton and Wilkens, 1998).
The El NinÄo-Southern Oscillation (ENSO) accounts for much of the interannual variability of
the climate of the Pampas (Kiladiz and Diaz,
1989; Halpert and Ropelewski, 1992; Ropelewski
and Halpert, 1996). The ENSO phenomenon
involves two extreme phases characterized by
anomalously warm and cold surface waters in the
eastern tropical Paci®c. During cold events, maximum temperatures and solar irradiance tend to be
higher, and minimum temperature and rainfall
tend to be lower than normal. Warm events show
an opposite but weaker in¯uence. ENSO in¯uences crop yields in the region through its in¯uence on precipitation, temperatures and solar
irradiance (Magrin et al., 1998; Podesta et al.,
1999). While maize, soybean and sorghum yields
tend to be lower than normal during cold events,
sun¯ower yield shows a weaker and opposite
response. Maize is the most responsive of the
major ®eld crops to increases in rainfall during
warm events (Podesta et al., 1999).
The predictability of climate and yield variability associated with ENSO suggests a potential
to tailor agricultural production decisions to either
mitigate the negative impacts of adverse conditions or to take advantage of favorable conditions.
Crop management and land allocation among
alternative crops are the most important options
available to farmers for managing climate risk
(Muchow and Bellamy, 1991; Kingwell, 1994; van
Noordwijk et al., 1994). Most applications of
ENSO-based climate forecasts for managing risk
have focused on management of individual crops
(Hansen et al., 1996; Marshall et al., 1996; Meinke
et al., 1996; Phillips et al., 1998). Attempts to use
ENSO information to optimize land allocation
decisions have generally been at regional, rather
than farm, spatial scales (Adams et al., 1995;
Solow et al., 1998).
The objective of this study is to explore the
potential for managing climatic risk at a farm scale
by tailoring land allocation among crops to ENSO
phases in the Argentine Pampas. We accomplish
this through model analysis of representative
farms at two locations di€ering in rainfall, soils
and present cropping patterns, and by analysis of
the sensitivity of optimal crop mix and information value to factors that vary among farms (i.e.
risk preferences, initial wealth, soil, climate) and
between years (i.e. crop prices, initial conditions).
The research tests three hypotheses under conditions appropriate to the Pampas region. First,
optimal land allocation among crops di€ers
among ENSO phases for a given physical and
economic environment. Second, optimizing crop
mix based on ENSO phase information increases
mean farm income. Third, the physical and economic environment and farmers' aversion to risk
in¯uence the optimal crop mix and potential value
of ENSO information.
2. Land allocation model
Expected utility theory provides a quantitative
framework for characterizing decision makers'
preferences among risky outcomes (Pratt, 1964;
Robison et al., 1984; Hardaker et al., 1997). Nonlinearity of a utility function U of, for example,
wealth W accounts for risk attitudes. Increasing U
implies that the decision maker prefers more
wealth to less. A concave utility function accounts
for aversion to risk by giving low wealth greater
weight than higher wealth. In other words, the
expected utility of wealth is less than the utility
of expected wealth. Aversion to risk is quanti®ed
by the curvature rather than the scale of U. For
a given expected wealth, risk aversion can be
expressed by a coecient of absolute Ra,
Ra …W† ˆ ÿU00 …W†=U0 …W†;
or relative risk aversion (Rr),
Rr …W† ˆ WRa …W† ˆ ÿWU00 …W†=U0 …W†;
where U 0 and U00 are the ®rst and the second
derivatives of U with respect to wealth. Theory
predicts that decision makers seek to maximize
the expected value of U for the distribution of
expected outcomes.
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
We assume that farmers allocate land to cropping enterprises in a way that maximizes the
expected utility of wealth at the end of a 1-year
planning period:
WF ˆ W0 ‡ ;
liability (US$). Variable crop production costs are
factored into P.
For given expected weather conditions, optimal
land allocation is determined by maximizing
expected utility of wealth at the end of the decision
period:
where W0 is initial wealth and is risky farm net
income during the year. The power function,
max
x EfU…WF †g
U…WF † ˆ WF 1ÿRr =…1 ÿ Rr †;
subject to:
…1†
used in this study implies constant relative risk
aversion. Under this assumption, Ra decreases as
W0 increases. Decisions are sensitive to additive
changes in W0 but insensitive to scale (i.e. proportional changes in W0 and ) (Pope and Just, 1991).
Empirical evidence supports the implication of Eq.
(1) that wealthier farmers are less averse to risk
than poorer ones (Young, 1979; Lins et al., 1981;
Pope and Just, 1991; Chavas and Holt, 1990).
We assume that prices are known but weather
is unknown at decision time. Farmers in the
region have access to international future prices
prior to planting, and can ®x their prices by
contract. This initial approach allowed us to
isolate climate from price risks in our analysis.
E€ects of price variability were examined as a
part of a sensitivity analysis (see below). Distributions of net income from each crop enterprise were calculated from constant crop prices
(means for 1986±97) and input costs, and from
yields simulated for each year of historical
weather data within a particular category. Net
farm income (US$) that would result from
weather year i is calculated as:
i ˆ
m
X
xj ij ÿ C ÿ Ti ;
…2†
jˆ1
…3†
where xj is land area (ha) allocated to crop j, Yij
and pij are yield (Mg haÿ1) and gross margin (US$
haÿ1) from crop j, Pj and cj are price (US$ Mgÿ1)
and ®xed production costs (US$ haÿ1) for crop j,
C is ®xed farm costs (US$) and Ti is income tax
ˆ
n
X
U…W0 ‡
iˆ1
m
X
xj ij ÿ C ÿ Ti †=n
…4†
jˆ1
Ax4b;
x50;
where Ui is utility (Eq. (1)) for weather year i, A is
a matrix of technical coecients, x is the vector of
areas allocated to each cropping system, and b is a
vector of farm resource constraints (Lambert and
McCarl, 1985; Chavas and Holt, 1990). In this
case, b is the total farm size (a scalar) and A
becomes a vector of ones to relate land allocated
to each crop to available land. Although this fairly
standard formulation is ¯exible enough to handle
a variety of linear constraints such as availability
of labor or equipment, we consider only nonnegativity and farm size constraints to land
allocation.
The potential value V of ENSO information can
be expressed as the di€erence in expected economic returns to optimal decisions conditioned on
ENSO phases and returns to optimal decisions
based on the historical climatology (Thornton and
MacRobert, 1994; Solow et al., 1998; Wilks and
Wolfe, 1998), or:
Vˆ…
ni
3 X
n
X
X
ij ÿ
k †=n;
iˆ1 jˆ1
ij ˆ Yij Pj ÿ cj ;
199
…5†
kˆ1
where *ij is farm income in year j of ENSO phase
i given optimal crop mix for phase j, and *k is
farm income in weather year k given crop mix
optimized for all n weather years in the historical
record. For ease of comparison, we express V on a
haÿ1 basis.
200
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
3. Approach
3.1. ENSO phases
ENSO events were categorized based on 5month running means of spatially averaged sea
surface temperature (SST) anomalies in the region
of the tropical Paci®c Ocean between 4 N±4 S and
90 ±150 W. A crop year (preceding July to current
June) was classi®ed as a warm (cold) event if
the SST anomalies were 50.5 C (4ÿ0.5 C) for
at least six consecutive months including the
October±December quarter (Sittel, 1994; Trenberth, 1997). The SST index is based on observed
data for the period 1949 to the present. For years
before 1949, the index was derived from reconstructed monthly mean SST ®elds (Meyers et al.,
1999). The period from 1931 includes 13 warm
events (1940, 1951, 1957, 1963, 1965, 1969, 1972,
1976, 1982, 1986, 1987, 1991 and 1997) and 15
cold events (1938, 1942, 1944, 1949, 1954, 1955,
1956, 1964, 1967, 1970, 1971, 1973±1975 and
1988).
3.2. Case study farms
We applied the land allocation model to representative farms at two locations in the Pampas.
Hall et al. (1992) provide a thorough description
of regional climate, soils and cropping systems.
Pergamino is in the climatically favorable eastern
humid region, where the dominant crops are
maize, wheat and soybean. Santa Rosa is in the
drier, climatically marginal western region, where
yields are more variable due to years with water
de®cits. More drought-tolerant cropsÐwheat,
sun¯ower and sorghumÐdominate in this region.
In both regions, wheat is sown in June and July,
maize and sun¯ower in October and November,
and soybean from mid-November through December. Dominant soils include Typic Hapludolls
and Ustipsaments in Santa Rosa, and Vertic
Argiudolls and Typic Hapludolls in Pergamino.
Fig. 1 shows mean monthly precipitation by
ENSO phase at each location.
Assumptions about the case study farms (Table
1) were based on information from AACREA
(Asociacion Argentina de Consorcios Regionales
de Experimentacion Agricola). Livestock is an
important component of farming systems near
Santa Rosa. However, because this study focuses
on crop production, calculations of income, taxes
and initial wealth were based only on that portion
of the farm devoted to crops.
Initial wealth is de®ned as liquid assets, estimated at 60% of the recent value of crop land.
This de®nition is based on the assumption that a
farmer will not sacri®ce future income potential by
selling crop land, but can borrow up to 60% of
land value. We assumed that the farmer owns his
land, and does not carry debt on facilities or
equipment beyond their salvage value. Production
Fig. 1. Mean monthly precipitation for each ENSO phase at (A) Santa Rosa and (B) Pergamino.
175
175
202
202
142
145
c
b
a
600
630
30
26
644
932
Entic Haplustolls
Typic Argiudolls
Santa Rosa
Pergamino
251
305
CV
(%)
Total
(mm yearÿ1)
Plant extractable soil water.
Estimates based on AACREA technical assumptions (Anonymous, 1998) and historical prices (SAGPyA, archives).
Double crop with wheat.
103
139
71
133
827
2200
Land value
(US$ haÿ1)
Crop area
(ha)
Precipitation
PESW a
(mm)
Soil
Location
Table 1
Characteristics of representative farms in Santa Rosa and Pergamino
216
234
Double c
Sole
Sun¯ower
Wheat
Maize
Fixed costs
(US$ haÿ1)
Fixed costs of production b
(US$ haÿ1)
Soybean
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
201
costs for each crop were estimated using technical assumptions made by AACREA (1998) and
historical input prices provided by SAGPyA
(Secretaria de Agricultura, Ganaderia, Pesca y
Alimentacion). Variable production costs include:
(1) harvest costs equal to 8% of crop value; (2)
trading costs equal to 10% of value for maize, 8%
for soybean, 7% for wheat and 6% for sun¯ower;
and (3) transportation costs. Trading costs include
a 2% tax on gross income, drying costs, discounts
for weed seed presence and 3% trader commission. Fixed farm costs include administration
costs and property taxes. Income tax Ti was calculated for each weather year i as:
i > a
b…i ÿ a† ‡ c
Ti ˆ
0
i 4a
where a, b and c are tabulated values designed to
provide an incrementally progressive marginal tax
rate. Mean farmgate prices (Table 2) were estimated from historical prices (1986±97) at the
Rosario port for all crops at Pergamino and for
maize and soybean in Santa Rosa. Wheat and
sun¯ower prices at Santa Rosa are based on prices
for the same period at the Bahia Blanca port. The
sun¯ower price includes an 8% premium for oil
content.
3.3. Crop simulation
Dynamic, process-level crop simulation models
have proven useful for quantifying the interactions between weather variability, management
and the physical environment. They were used in
this study to estimate distributions of crop yields
due to historical climate variability for given
soils parameters and initial conditions, cultivars
and crop management scenarios. Yields were
simulated by the crop models included in version
3.5 of the Decision Support System for Agrotechnology Transfer (Jones et al., 1998): GenericCERES (Ritchie et al, 1998) for maize and wheat,
CROPGRO (Boote et al., 1998) for soybean and
OILCROP-SUN (Villalobos et al., 1996) for sun¯ower. Careful regional calibration and validation
of these models has been performed at INTA
(Instituto Nacional de TecnologõÂa Agropecuaria)
202
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
Table 2
Mean and standard deviation of farmgate prices a (US$ Mgÿ1) for ®eld crops (1986±97)
Crop
Harvest period
Maize
Soybean
Wheat
Sun¯ower
a
Pergamino
Apr.±Jun.
May±Jul.
Dec.±Feb.
Feb.±May
Santa Rosa
Mean
S
Mean
S
97
221
110
180
26
58
34
35
83
195
96
159
27
58
35
30
Source: SAGPyA, archives.
(Baethgen and Magrin, 1995; Travasso and DeleÂcolle, 1995; Meira and Guevara, 1997).
INTA researchers provided characteristic parameter values calculated for the dominant agricultural soils and commonly used crop cultivars in
the Pampas (Meira, Magrin and Travasso, personal communication). Daily weather data for the
1931±97 period (rainfall, maximum and minimum
temperature, bright sunshine duration and limited
periods of solar irradiance) for Pergamino and
Santa Rosa are from the Servicio Meteorologico
Nacional, and have undergone extensive quality
checking (PodestaÂ, University of Miami, personal
communication). Solar irradiance H (MJ mÿ2
dayÿ1) was estimated from bright sunshine duration (S) using the AÊngstroÈm (1924) equation, with
monthly a and b coecients calculated by robust
regression (Lanzante, 1996). When both H and S
were missing (26% of days at Santa Rosa, 25% at
Pergamino) H was generated stochastically using
an adaptation of WGEN (Richardson and Wright,
1984) that incorporates the stochastic solar irradiance model described by Hansen (1999). Typical
management practices for each crop and location
(Table 3) are based on data provided by the farmer
organizations, AACREA and FAA (Federacion
Agraria Argentina). Simulated sowing was conditioned on soil conditions, and occurred as soon
after the dates in Table 3 as soil water content in the
top 10 cm exceeded 50% of plant-extractable soil
water holding capacity, and soil temperature
exceeded 5 C for wheat, 12 C for maize and 14 C
for sun¯ower and soybean.
Soil conditions at sowing vary among years, and
have important e€ects on simulated yields. This is
particularly important for deep soils with high
water holding capacity, such as those in the Pampas region. Average initial soil conditions were
calculated from continuous long-term (1931±97)
sequences simulated for each cropping system.
Table 3
Management assumptions used for crop simulations
Crop
Pergamino
Cultivar
Maize
Soybean
Sole
Double
Wheat
Sun¯ower
a
Santa Rosa
Planting
N fertilizer
Date a
Density
(mÿ2)
`DK 752'
1 Oct.
`A 5435'
`A 5435'
`Oasis'
`Conti 3'
1 Nov.
20 Dec.
5 Jun.
1 Oct.
Cultivar
Planting
Date b
Amount
(kg haÿ1)
Date a
8.0
40
60
`DK 752'
1 Oct.
25.0
30.0
200.0
5.0
ÿ
ÿ
0
ÿ
0
0
75
0
`DM 4700'
`DM 4700'
`Pigue'
`Conti 3'
1 Nov.
15 Dec.
1 Jul.
10 Oct.
N fertilizer
Date b
Amount
(kg haÿ1)
5.5
40
40
25.0
30.0
250.0
4.0
ÿ
ÿ
ÿ
ÿ
0
0
0
0
Density
(mÿ2)
Target date. Simulations permitted delayed sowing due to unfavorable soil temperature and water content for all crops, and
preceding wheat maturity date for double-cropped soybean.
b
Days after planting.
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
+
Simulated soil water, NOÿ
3 and NH4 contents on
31 March, averaged across years and cropping
systems, were used as initial conditions for the
crop simulations used to optimize land allocation.
The 31 March initialization date for soil water and
N-balance simulations is based on the assumption
that soil water content is predictably low shortly
after the summer crop harvest. Starting soil simulations early captures the e€ect of recharge from
rain during the fallow from 31 March to sowing.
3.4. Optimization
The ®rst step in calculating optimal land allocation was to simulate yields for each cropping system and every year of weather data. Gross margins
from each crop enterprise (Eq. (3)) were then calculated from constant production costs (Table 1)
and prices (Table 2), and yields simulated for each
weather year. Eq. (4) was then solved by constrained nonlinear optimization to identify the set
of areas (x) allocated to each crop enterprise that
maximized expected utility. This study used the
generalized reduced gradient algorithm (Lasdon et
al., 1978) included in the Microsoft (1997) Excel
spreadsheet software. Because the utility function
(Eq. (1)) is always concave for Rr>0, the procedure is expected to always identify the global
maximum (Lambert and McCarl, 1985). However,
we used several random starting values of x to
ensure that the solution identi®ed was the global
optimum. The optimization loop iteratively adjusted x subject to constraints, calculated i (Eq. (2))
and U(i) (Eq. (1)) for each weather year i, then
calculated the current value of the objective function as the average utility among weather years.
We repeated the optimization procedure for all
years of weather data and for the years in each
ENSO phase. This provided the two sets of farm
incomes optimized with and without using ENSO
phase information required to estimate the potential value V of ENSO information (Eq. (5)).
3.5. Sensitivity analyses
3.5.1. Risk preferences and initial wealth
The optimization problem (Eqs. (1)±(4)) was
solved for typical farms at Santa Rosa (600 ha)
203
and Pergamino (630 ha) for four levels of relative
risk aversion (Rr=0.0, 1.0, 2.0 and 3.0) representing risk-neutral and slightly, moderately and very
risk-averse decision makers (Hardaker et al.,
1997). Three levels of initial wealth were considered for Pergamino (W0=US$1386, US$1108
and US$831 haÿ1) and Santa Rosa (US$521,
US$417 and US$312 haÿ1) based on the ability to
borrow up to 60% of land value, and 20 and 40%
reductions from that maximum. The reduced
levels of W0 account for indebtedness.
3.5.2. Sensitivity to factors that vary among years
Crop prices and soil conditions, water and nitrogen content at sowing, vary among years and
in¯uence gross margins of individual crop enterprises. To test the in¯uence of these factors on
optimal land allocation and information value, the
land allocation model was solved for a moderately
risk-averse farmer (Rr ˆ 2 and W0 ˆ 1108) at
Pergamino using di€erent crop prices and di€erent
initial soil conditions. Sensitivity of optimal crop
mix and information value to price variability was
explored by solving the land allocation model
using crop prices for individual years from 1986±
97. Sensitivity of optimal crop mix and information value to soil conditions variability was tested
using information, derived from simulation, that
attempted to capture crop and climate e€ects during the preceding growing season and fallow on
soil water and nitrogen content at sowing. We
analyzed sensitivity to these factors by comparing
yields and optimization results simulated using
average soil initial conditions derived for each
preceding cropping system. Soil water, NOÿ
3 and
NH+
4 contents by soil layer were averaged for all
years as described previously from long-term
sequential simulations for wheat, the wheat±
soybean double crop, and the mean among summer crops (i.e. maize, sun¯ower, soybean). We
repeated the procedure using soil initial conditions derived for each preceding ENSO phase.
4. Results and discussion
Crops included in the optimal crop mix di€ered
between the two locations. For moderate risk
204
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
aversion (Rr ˆ 2; W0=US$812 haÿ1 at Pergamino
and US$417 haÿ1 at Santa Rosa), the model predicted optimal farm land allocation for all years
that was quite similar to the actual allocation of
land among cropping systems in the districts corresponding to each location (Alippe, unpublished
report; Cascardo et al., 1988) (Fig. 2). Under the
assumption of risk neutrality, all land was allocated to the cropping system with the highest
mean net return (Table 4). The predicted increase
in diversi®cation among crop enterprises with
increasing aversion to risk is consistent with other
studies (Patten et al., 1988; Kingwell, 1994). As
risk aversion increased from neutrality to Rr ˆ 3,
simulated mean income decreased from US$128
to US$117 haÿ1 in Pergamino and from US$13 to
ÿUS$4 haÿ1 in Santa Rosa, while the standard
deviation of income decreased from US$154 to
US$109 haÿ1 in Pergamino and from US$135
to US$44 haÿ1 in Santa Rosa. This pattern of
increased diversi®cation and decreasing variability
of farm income with increasing risk aversion can
be explained by the imperfect correlation of yields
and, therefore, of gross margins, among crop
enterprises (Table 4). A mixture of risky strategies
will have a lower overall risk (expressed by standard deviation S) than the weighted average of
risk of the individual strategies if their crosscorrelation is less than one (van Noordwijk et al.,
1994). Overall risk of the mixture decreases as
income correlation among crop enterprises
decreases.
The very low mean income simulated for Santa
Rosa suggests that crop production should not be
feasible in that region given historical climate
conditions. The ratio of cultivated to grazed land
has increased in recent decades. Viglizzo et al.
(1995) attribute this trend primarily to an increasing trend in precipitation (Fig. 3A ). Simulations
showed a trend toward increasing yields and farm
income due to changes in climate at Santa Rosa.
Mean farm income simulated for a moderately
risk-averse farmer (Rr ˆ 2, W0=US$417 haÿ1)
increased from ÿUS$19 haÿ1 for the period 1940±
70 to US$43 haÿ1 for 1971±97 (Fig. 3B). The trend
in simulated income helps to explain the observed
Fig. 2. Predicted optimal crop mix at (A) Pergamino and (B) Santa Rosa for risk neutral (RN), moderately risk-averse (MRA) and
very risk averse (VRA) decision makers, and mean land allocation in the surrounding region (REG). Initial wealth was US$417 haÿ1
for Santa Rosa and US$812 haÿ1 for Pergamino.
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
205
Table 4
Mean, standard deviation and linear correlations of simulated gross margins of crop enterprises by ENSO phase for Pergamino and
Santa Rosa (1931±97)
Crop
enterprise
Abbreviation Pergamino
Santa Rosa
Mean
S
(US$ haÿ1)
Linear cross-correlation a
MZ
All years
Maize
Wheat
Sun¯ower
Soybean
Soybean±wheat
MZ
WH
SU
SB
SW
301
240
200
318
299
225
97
98
205
181
Cold phase
Maize
Wheat
Sun¯ower
Soybean
Soybean±wheat
MZ
WH
SU
SB
SW
160
205
158
205
170
216 1.00
122 ÿ0.32
68 0.55
225 0.41
120 ±0.24
Warm phase
Maize
Wheat
Sun¯ower
Soybean
Soybean±wheat
MZ
WH
SU
SB
SW
392
270
192
343
377
228
67
109
236
250
a
WH
SU
SB
SW
Mean
S
(US$ haÿ1)
1.00
104
0.08
1.00
62
0.63
0.02 1.00
69
0.57 ÿ0.01 0.55 1.00
70
0.50
0.38 0.29 0.65 1.00
±
1.00
0.46
0.83
0.95
0.88
2
1.00
29
0.04 1.00
37
±0.09 0.49 1.00
±14
0.51 0.07 0.57 1.00
±
135
1.00
76
0.33 1.00
54
0.50 0.71 1.00
107
0.69 0.58 0.86 1.00
±
Linear cross-correlation a
WH
SU
SB
1.00
0.34
0.40
0.53
±
1.00
ÿ0.07
0.31
±
1.00
0.01
±
1.00
±
135 1.00
49 0.38
45 ±0.15
137 0.00
± ±
1.00
±0.46
0.53
±
1.00
0.19
±
1.00
±
1.00
0.18
0.32
±
1.00
0.20
±
1.00
±
157
55
65
164
±
169
63
71
165
±
MZ
1.00
0.44
0.62
0.62
±
MZ, maize; WH, wheat; SU, sun¯ower; SB, soybean; SW, soybean±wheat.
changes in land use, and reconciles the current
feasibility of crop production with the extremely
low long-term mean simulated income. We chose
to base our analyses of farm-scale land allocation
on the entire weather series (1931±97) to ensure
adequate representation of warm and cold events.
The implications of this decision on current land
use decisions needs further research.
4.1. Optimal land allocation conditioned to ENSO
Fig. 4 shows optimal land allocation among
cropping systems predicted for Pergamino and
Santa Rosa for varying levels of risk aversion. The
di€erences in optimal crop mix among ENSO
phases can be explained by di€erences in crop tolerance to water availability, and are consistent
with known e€ects of ENSO on rainfall in this
region (Fig. 1). Precipitation tends to be low during cold events and high during warm events in the
critical period from November to December when
grain number is determined in maize (Hall et al.,
1992; Cirilo and Andrade, 1994) and sun¯ower
(Cantagallo et al., 1997). Rainfall is less in¯uenced
by ENSO during pod formation in soybean (February), when the crop is most susceptible to water
shortage (Sinclair et al., 1994). The optimal crop
mix for the cold phase was dominated by crops
that either tolerate water stress (i.e. sun¯ower in
Santa Rosa) or avoid periods of water shortage
during critical development stages (i.e. wheat and
soybean in Pergamino). In Pergamino, the majority of land was allocated to maize in the warm
phase. Under conditions of ample rainfall and
average prices (Table 2), maize is generally the
most pro®table crop in the region.
4.2. In¯uence of risk aversion
Under the assumption of risk neutrality, the
optimization procedure usually selected the monoculture with the highest mean gross margin for
206
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
Fig. 3. Annual time series of (A) precipitation and (B) simulated net income at Santa Rosa, 1937±97. Solid lines indicate
means of 1940±70 and 1971±90.
each ENSO phase (Table 4; Fig. 4A, D). The
exception was the cold phase in Pergamino, where
the optimal crop mix included both wheat and
soybean. Monocultures are expected when risk
neutrality is assumed (Mjelde et al., 1996, 1997).
However, the progressive marginal tax rate in
Argentina reduces the marginal value of increasing
income and, therefore, has the same e€ect as risk
aversion (Hardaker et al., 1997). Simulated gross
margins of the two crops were quite similar on the
average, but negatively correlated (Table 4).
Optimal crop mix conditioned to ENSO phase
showed the same tendency toward diversi®cation
with increasing risk aversion (Fig. 4) as the crop
mix optimized for all years (Fig. 2). Under moderate or high risk aversion, the optimal crop mix
always included the most pro®table crop enterprise, and additional enterprises with incomes that
were weakly or negatively correlated with the most
pro®table crop. The optimal crop mix in the cold
phase under moderate risk aversion (Fig. 4B, E) is
easily explained by the negative correlation of
gross margins between wheat and sun¯ower in
Santa Rosa, and between wheat and soybean
in Pergamino (Table 4). Although incomes of
alternative crop enterprises were all positively
correlated in the warm phase, the similarity of
mean income among the crops selected and the
mitigating in¯uence of imperfect correlation on
overall risk still favored diversi®cation under at
least moderate risk aversion.
The in¯uence of risk aversion on the value V
of ENSO information di€ered between the two
locations (Fig. 5). At Pergamino, V increased
monotonically with increasing Rr for the range
considered. However, at Santa Rosa, V decreased
as Rr increased above 1.0. Favorable climate
conditions and relatively greater initial wealth
and mean income in Pergamino generally favored
strategies that take advantage of the greater climatic di€erences among ENSO phases that occur
in Pergamino. In contrast, the more adverse climatic conditions in Santa Rosa favored strategies
that protect farmers from extremely low incomes.
The decrease of V with increasing Rr re¯ects the
high probability of negative net incomes in all
ENSO phases, and the similarity of strategies to
avoid economic loss among ENSO phases. Furthermore, lower initial resource endowment and
mean income results in greater absolute risk
aversion for a given relative risk aversion in
Santa Rosa than in Pergamino. These results
support the lack of monotonic relationship
between risk aversion and the value of information predicted by Hilton (1981). Mjelde and
Cochrane (1988) also found a negative relationship between risk aversion and climate information value when crop management strategies were
optimized for adverse climate conditions, and a
positive relationship when climatic conditions
were more favorable.
4.3. In¯uence of prices
The possible e€ect of prices on optimal crop mix
and the value of ENSO information were explored
by optimizing land allocation for di€erent
observed commodity prices for a moderately riskaverse farmer (Rr ˆ 2, W0 ˆ 1108) at Pergamino.
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
207
Fig. 4. Predicted optimal crop mix by ENSO phase for (A±C) Santa Rosa and (D±F) Pergamino for (A, D) neutral, (B, E) moderate
and (C, F) high levels of risk aversion. Initial wealth was US$417 haÿ1 for Santa Rosa and US$812 haÿ1 for Pergamino.
Changes in relative prices (Table 5) can favor or
exclude crops from the set of feasible options (see
examples in Fig. 6). An increase in the price of one
crop relative to the others (e.g. maize in 1989, Fig.
6A) can reduce the feasibility of the alternatives
and decrease di€erences in optimal crop mix
among ENSO phases. The resulting loss of ¯exibility reduces the potential value of ENSO information. This was the case for 1987±90, 1996 and
1997 price scenarios. The relatively high soybean
prices in 1987, 1988 and 1997 favored monocultures for all ENSO phases, resulting in an
ENSO information value of zero (Table 5). In
contrast, the balance of commodity prices in 1992
(Fig. 6B) enhanced the value of ENSO information by favoring di€erences in optimal crop mix
among ENSO phases. The greatest absolute bene®t from tailoring land allocation to ENSO phases
came from 1994 prices (Fig. 6C), when high sun¯ower prices compensated for its relatively low
productivity.
These results illustrate the important e€ect of
relative prices of alternative crops on optimal land
allocation. Relative pricing can limit the potential
bene®t of ENSO information in particular years.
Decision support applications of this type of optimization model should, therefore, always be based
on the most current price expectations.
208
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
Fig. 5. Predicted value of ENSO information as a function of
relative risk aversion at Pergamino and Santa Rosa. Initial
wealth was US$312 haÿ1 for Santa Rosa and US$831 haÿ1 for
Pergamino.
4.4. Sensitivity to soil initial conditions
Simulated mean initial (31 March) extractable
soil water and mineral N contents were sensitive to
the preceding crop (Table 6). The preceding
ENSO phase in¯uenced initial soil water, but not
N. Simulated crop yields responded signi®cantly
to the di€erent initial conditions (Table 6).
Among the crops considered, maize was the
most sensitive and sun¯ower the least sensitive
to di€erent initial soil conditions due to previous
ENSO phase. Initial soil conditions also in¯uenced optimal land allocation for any single
year. Maize and soybean tended to dominate the
optimal crop mix when initial soil water contents
were high (i.e. following wheat monoculture or
warm events) at Pergamino, while drier initial
conditions (i.e. following wheat±soybean) tended
to favor the wheat±soybean double crop (data
not shown).
Both mean income for the crop mix optimized
without considering ENSO and the potential value
of ENSO information were quite sensitive to differences in initial soil conditions (Table 6). With
the exception of initial conditions associated
with the wheat±soybean double crop, initial conditions in¯uenced mean income and information
value in the same direction. The relatively high
information value following the double crop suggests that ENSO information might help farmers
avoid important losses associated with growing
soybean after wheat during a cold event (data not
shown). These results highlight the sensitivity of
initial soil conditions to conditions in the preceding
Table 5
Historical price scenarios (1986-97) and corresponding predicted ENSO information value for a moderately risk-averse farmer at
Pergamino
Year
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
a
b
Relative crop price a
Information value
Maize
Soybean
Wheat
Sun¯ower
(US$ haÿ1)
%b
0.87
0.89
1.08
1.50
0.90
0.84
0.84
0.83
0.93
0.88
1.46
0.97
0.81
1.15
1.48
1.34
0.72
0.77
0.82
0.92
0.94
0.79
1.04
1.22
0.68
0.79
1.08
1.20
1.30
0.47
0.93
1.01
0.96
0.98
1.64
0.97
0.78
1.07
1.19
1.27
0.88
0.76
0.73
1.04
1.19
0.95
1.01
1.11
4
0
0
1
2
3
11
10
12
9
4
0
6
0
0
0
1
7
19
10
12
14
1
0
Prices relative to observed mean prices for 1986±97.
ENSO information value relative to mean income for the optimal crop mix without using ENSO information.
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
209
Fig. 6. Predicted optimal crop mix by ENSO phases at Pergamino for (A) 1989, (B) 1992 and (C) 1994 prices and moderate risk
aversion.
Table 6
Simulated crop yield as a function of soil initial conditions simulated for each previous crop and previous ENSO phase, and corresponding predicted ENSO information value for a moderately risk-averse farmer and initial wealth of US$1108 haÿ1 at Pergamino
Initial soil a
Mean simulated
Water
(mm)
N
(kg haÿ1)
Crop yield (Mg haÿ1) b
Maize
Soybean
Sun¯ower
Wheat
Previous crop
Wheat
Summer crops
Wheat±soybean
210
134
78
168
57
41
7.9a
6.0b
4.7c
3.0ab
2.6b
2.2c
2.7a
1.7b
1.4b
4.8a
4.1b
3.6b
Previous ENSO phase
Warm
Neutral
Cold
174
124
105
77
73
76
7.0a
6.3ab
5.7b
2.8a
2.6ab
2.4b
2.0a
2.0a
1.9a
4.7a
4.3ab
3.9b
Factor
a
b
c
d
Information value
Income c
(US$ haÿ1)
%d
214
116
60
15
7
12
7
6
20
144
114
88
13
8
7
9
7
8
Mean simulated extractable soil water and mineral N on 31 March.
Means for a given crop followed by di€erent letters di€er signi®cantly ( p ˆ 0:05) by Tukey's multiple comparison test.
Obtained for land allocation optimized without using ENSO information.
ENSO information value relative to mean income for the optimal crop mix without using ENSO information.
growing season, and con®rm the importance of
assumptions about initial soil water content for
obtaining reliable yield simulations. Marshall et
al. (1996) indicated that improved initial soil conditions may increase the potential bene®t of climate forecasts to Australian wheat farmers.
5. Conclusions
The simple model presented in this paper identi®es optimal land allocation by maximizing the
expected value of nonlinear utility of wealth associated with a given distribution of simulated
210
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
yields. The model proved useful in accomplishing
our objective of exploring the potential bene®ts of
tailoring farm land allocation to ENSO phases,
and is a step toward the practical application of
ENSO-based climate prediction to farm-scale land
allocation decisions. The ability of the optimization model to predict actual land use practices at a
district scale strengthens con®dence in the validity
of the approach, input data and assumptions.
Results of this exploratory study complement
previous studies that focused either on optimizing
management for a single crop or cropping patterns
at regional scale, and suggest that the optimization
model has potential as a farm decision support
tool. The structure of the model is readily adapted
to consider other constraints or objectives that
may be relevant to particular farms. Further
research is necessary to explore implications of
other farm constraints (e.g. labor and machinery
capacity) and crop management tailored to ENSO
phase on the potential value of ENSO information.
Predicted optimum land allocation varied
among ENSO phases in a manner that was consistent with known in¯uences of ENSO on precipitation and di€erences in sensitivity to water
availability among crops. Results suggest that tailoring land allocation among crops to ENSO
phases can increase mean farm income by 9% in
average and up to 20%. However, optimal crop
mix and resulting ®nancial bene®ts are quite sensitive to factors that vary among location, farm
characteristics and years. In particular, e€ects of
risk aversion on potential ENSO information
value was nonmonotonic and in¯uenced by location. Changes in prices and risk aversion within
plausible ranges can constrain the value of ENSO
information by restricting the feasibility of alternative crop enterprises. Except for the previous
crop scenario wheat±soybean, initial soil water
content a€ect crop yields, land allocation and
forecast value proportionally to net mean income.
For these reasons, decision support applications of
this type of model should be updated each year
based on current price expectations and soil water
content, and consider the characteristics and
environment of each farm. We emphasize the
importance of using locally validated crop models
and high quality historical weather records in
order to obtain reliable results from the optimization model.
Acknowledgements
We gratefully acknowledge information and
helpful comments provided by J.W. Jones, G.O.
Magrin, M.I. Travasso, E. Guevara, S. Meira, D.
Letson, F. Royce, G. Duarte and two anonymous
reviewers. This work was supported by the grant
from NOAA (Oce of Global Programs) entitled
``Regional Application of ENSO-Based Climate
Forecasts to Agriculture in the Americas'', by a
fellowship from the Inter-American Institute for
Global Change Research, and by the University of
Buenos Aires.
References
AACREA, 1998. Revista CREA 211, 103±113.
Adams, R.M., Bryant, K.J., McCarl, B.A., Legler, D.M.,
O'Brien, J.J., Solow, A.R., Weiher, R., 1995. Value of
improved long-range weather information. Contemporary
Economic Policy 13, 10±19.
AÊngstroÈm, A., 1924. Solar and terrestrial radiation. Quart. J.
Roy. Meteorol. Soc. 50, 121±126.
Baethgen, W.E., Magrin, G.O. 1995. Assessing the impacts of
climate change on winter crop production in Uruguay and
Argentina using crop simulation models. In: Rosenzwig, C.
(Ed.), Climate Change and Agriculture: Analyses of Potential International Impacts (ASA Special Publication No. 59).
ASA, Madison, WI, pp. 207±228.
Boote, K.J., Jones, J.W., Hoogenboom, G., 1998. Simulation
of crop growth: CROPGRO model. In: Peart, R.M., Curry,
R.B. (Eds.), Agricultural Systems Modeling and Simulation.
Marcel Decker, New York, pp. 651±693.
Cantagallo, J., Chimenti, C., Hall, A.J., 1997. Number of seed
per unit area in sun¯ower correlates well with a photothermal quotient. Crop Sci. 37, 1780±1786.
Cascardo, A.R., Pizzarro, J.B., Peretti, M.A., Gomez, P.O.,
1998. Sistemas de produccion predominantes. In: Barsky, O.
(Ed.), La Agricultura Pampeana: Transformaciones productivas y sociales. Fondo de Cultura Economica, Buenos
Aires, Argentina, pp. 95±145.
Cirilo, A.G., Andrade, F.H., 1994. Sowing date and maize
productivity: I. Crop growth and dry matter partitioning.
Crop Sci. 34, 1039±1043.
Chavas, J.P., Holt, M.T., 1990. Acreage decisions under risk: the
case of corn and soybeans. Am. J. Agric. Econ. 72, 529±538.
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
Hall, A.J., Rebella, C.M, Ghersa, C.M., Culot, J.P., 1992. Field
crop systems of the Pampas. In: Pearson, C.J. (Ed.),
Field Crop Ecosystems of the World. Elsevier, Exeter, UK,
pp. 413±450.
Halpert, M.S., Ropelewski, C.F., 1992. Surface temperature
pattern associated with the Southern Oscillation. J. Climate
5, 577±593.
Hansen, J.W., 1999. Stochastic daily solar irradiance for biological modeling applications. Agric. For. Meteorol. 94, 53±63.
Hansen, J.W., Jones, J.W., Magrin, G., Meira, S.G., Guevara,
E.R., Travasso, M.I., DõÂaz, R.A., Marino, M., Hordij, H.,
Harwell, C., PodestaÂ, G., 1996. ENSO e€ects on yields and
economic returns of wheat, corn, and soybean in Argentina.
In: Actas del VII Congreso Argentino y VII Congreso
Latinoamericano e IbeÂrico de MeteorologõÂa. Buenos Aires,
Argentina, 2±6 September, 1996, pp. 89±90.
Hardaker, J.B., Huirne, R.B.M., Anderson, J.R., 1997. Coping
with Risk in Agriculture. CAB International, Wallingford, UK.
Hilton, R., 1981. The determinants of information value:
synthesizing some general results. Management Science 27,
57±64.
Jones, J.W., Tsuji, G.Y., Hoogenboom, G., Hunt, L.A.,
Thornton, P.K., Wilkens, P.W., Imamura, D.T., Bowen,
W.T., Singh, U., 1998. Decision support system for agrotechnology transfer: DSSAT v3. In: Tsuji, G.Y., Hoogenboom, G., Thornton, P.K. (Eds.), Understanding Options
for Agricultural Production. Kluwer Academic Publishers,
Dordrecht, The Netherlands, pp. 157±177.
Keating, B.A., Godwin, D.C., Watiki, J.M., 1991. Optimizing
nitrogen inputs in response to climatic risk. In: Muchow,
R.C., Bellamy, J.A. (Eds.), Climate Risk in Crop Production:
Models and Management for the Semiarid Tropics and Subtropics. CAB International, Wallingford, UK, pp. 329±358.
Kiladiz, G., Diaz, H., 1989. Global climate anomalies associated with extremes in the Southern-Oscillation. J. Climate
2, 1069±1090.
Kingwell, R.S., 1994. Risk attitude and dryland farm management. Agricultural Systems 45, 191±202.
Lambert, D.K., McCarl, A., 1985. Risk modeling using direct
solution of nonlinear approximations of the utility function.
Am. J. Agric. Econ. 67, 846±852.
Lanzante, J.R., 1996. Resistant, robust and non-parametric
techniques for the analysis of climate data: theory and
examples, including applications to historical radiosonde
station data. Intl. J. Climatol. 16, 1197±1226.
Lasdon, L.S., Waren, A.D., Jain, A., Ratner, M., 1978. Design
and testing of a generalized reduced gradient code for
nonlinear programming. ACM Transactions on Mathematical Software 4, 34±50.
Lins, D., Gabriel, S., Sonka, S., 1981. An analysis of the riska
version of farm operators: an asset portfolio approach. West.
J. Agric. Econ. 6, 15±30.
Magrin, G.O., Grondona, M.O., Travasso, M.I., Boullon, D.R.,
Rodriguez, G.R., Messina, C.D., 1998. Impacto del fenomeno
``El NinÄo'' sobre la produccioÂn de cultivos en la regioÂn Pampeana. Instituto Nacional de TencologõÂa Agropecuaria, Instituto de Clima y Agua, Castelar, Argentina.
211
Marshall, G.R., Parton, K.A., Hammer, G.L., 1996. Risk attitude, planting conditions and the value of seasonal forecasts
to a dryland wheat grower. Austr. J. Agric. Econ. 40, 211± 233.
Meinke, H., Stone, C., Hammer, G.L., 1996. SOI phases and
climatic risk to peanut production: a case study for northern
Australia. Intl. J. Climatol. 16, 783±789.
Meira, S., Guevara, E., 1997. Application of SOYGRO in
Argentina. In: Krop€, M.J., Teng, P.S., Aggarwal, P.K.,
Bouman, J., Jones, J.W., Van Laar, H.H. (Eds.), Applications
of System Approaches at Field Level (Vol. 2), pp. 235±242.
Kluwer, Dordrecht.
Meyers, S.D., O'Brien, J.J., Thelin, E., 1999. Reconstruction of
monthly SST in the tropical Paci®c Ocean during 1868±1993
using adaptive climate basis functions. Mon. Wea. Rev.
(in press).
Mjelde, J.W., Cochrane, M.J., 1988. Obtaining lower and upper
bounds on the value of seasonal climate forecasts as a function of risk preferences. Western J. Agric. Econ. 13, 285±293.
Mjelde, J.W., Thompson, N., Nixon, C.J., 1996. Government
institutional e€ects on the value of seasonal climate forecasts.
Am. J. Agric. Econ. 78, 175±188.
Mjelde, J.W., Thompson, T.N., Nixon, C.J., Lamb, P.J., 1997.
Utilizing a farm-level decision model to prioritise future climate prediction research needs. Meteorol. Appl. 4, 161±170.
Muchow, R.C., Bellamy, J.A., 1991. Climate Risk in Crop
Production: Models and Management for the Semiarid Tropics and Subtropics. CAB International, Wallingford, UK.
Patten, L.H., Hardaker, B., Pannel, D.J., 1988. Utility-e€cient
programming for whole-farm planning. Austr. J. Agric.
Econ. 32 (1±2), 88±97.
Phillips, J.G., Cane, M.A., Rosenzweig, C., 1999. ENSO, seasonal rainfall patterns and simulated maize yield variability
in Zimbabwe. Agric. For. Meterol. 90, 39±52.
PodestaÂ, G., Messina, C.D., Grondona, M.O., Magrin, G.O.,
1999. Associations between grain crop yields in centraleastern Argentina and El NinÄo-Southern Oscillation. J.
Appl. Meteorol. (in press).
Pope, R.D., Just, R.E., 1991. On testing the structure of risk
preferences in agricultural supply analysis. Am. J. Agric.
Econ. 73, 743±748.
Pratt, J.W., 1964. Risk aversion in the small and in the large.
Econometrica 32, 122±136.
Richardson, C.W., Wright, D.A., 1984. A Model for Generating Daily Weather Variables. ARS-8, US Department of
Agric., Agric. Res. Serv., Washington, DC.
Ritchie, J.T., Singh, U., Godwin, D.C., Bowen, W.T., 1998.
Cereal growth, development and yield. In: Tsuji, G.Y.,
Hoogenboom, G., Thornton, P.K. (Eds.), Understanding
Options for Agricultural Production. Kluwer Academic
Publishers, Dordrecht, The Netherlands, pp. 79±98.
Robison, L.J., Barry, P.J., Kliebenstein, J.B., Patrick, G.F.,
1984. Risk attitudes: concepts and measurement approaches.
In: Barry, P.J. (Ed.), Risk Management in Agriculture. Iowa
State University Press, Aimes, IA, pp. 11±30.
Ropelewski, C.F, Halpert, M.S., 1996. Quantifying Southern
Oscillation±Precipitation Relationships. J. Climate 9, 1043±
1059.
212
C.D. Messina et al. / Agricultural Systems 60 (1999) 197±212
Sinclair, T.R., 1994. Limits to crop yield? In: Boote, K.J.,
Bennett, J.M., Sinclair, T.R., Paulsen, G.M. (Eds.), Physiology and Determination of Crop Yield. ASA, CSSA and
SSSA, Madison, WI, pp. 509±532.
Sittel, M.C., 1994. Di€erences in the Means of ENSO Extremes
for Maximum Temperature and Precipitation in the United
States (Technical Report No. 94-2). Center for Ocean±
Atmospheric Prediction Studies, Florida State University,
Tallahassee, FL.
Solow, A.R., Adams, R.F., Bryant, K.J., Legler, D.M.,
O'Brian, J., McCarl, B.A., Nayda, W., Weiher, R., 1998. The
value of improved ENSO prediction to U.S. agriculture.
Climatic Change 39, 47±60.
Thornton, P.K., MacRobert, J.F., 1994. The value of information concerning near-optimal nitrogen fertilizer scheduling.
Agricultural Systems 45, 315±330.
Thornton, P.K., Wilkens, P.W., 1998. Risk assessment and
food security. In: Tsuji, G.Y., Hoogenboom, G., Thornton,
P.K. (Eds.), Understanding Options for Agricultural Production. Kluwer Academic Publishers, Dordrecht, The
Netherlands, pp. 329±348.
Travasso, M.I., DeleÂcolle, R., 1995. Adaptation of the CERES-
Wheat model for large area yield estimation in Argentina.
Eur. J. Agron. 4, 347±353.
Trenberth, K., 1997. The de®nition of El NinÄo. Bull. Amer.
Meteor. Soc. 78, 2771±2777.
van Noordwijk, M., Dijksterhuis, G.H., van Keulen, H., 1994.
Risk management in crop production and fertilizer use with
uncertain rainfall; how many eggs in which baskets. Netherlands J. Agric. Science 42, 249±269.
Viglizzo, E.F., Roberto, Z.E., Filippin, M.C., Pordomingo,
A.J., 1995. Climate variability and agroecological change in
the central pampas of Argentina. Agriculture, Ecosystems
and Environment 55, 7±16.
Villalobos, F.J., Hall, A.J., Ritchie, J.T., Orgaz, F., 1996.
OILCROP-SUN: a development, growth and yield model of
the sun¯ower crop. Agron. J. 88, 403±415.
Wilks, D.S., Wolfe, D.W., 1998. Optimal use and economic
value of weather forecasts for lettuce irrigation in a humid
climate. Agric. For. Meteorol. 89, 115±129.
Young, D.L., 1979. Risk preferences of agricultural producers:
their use in extension and research. Am. J. Agric. Econ. 61,
1063±1070.