262 H.S.J. Hill et al. Agricultural and Forest Meteorology 100 2000 261–272
are teleconnected linked to seasonal climate vari- ations in other parts of the world Bjerknes, 1969;
Ropelewski and Halpert, 1986, 1987, 1989; Kiladis and Diaz, 1989. Economic-based studies Marshall
et al., 1996; Mjelde et al., 1997; Hill et al., 1998 have shown these events to have value to decision-makers
because of their ability to forecast climate conditions. Classifying ENSO events and the terminology asso-
ciated with the ENSO phenomenon, however, are not standardized. The objective of this study is to com-
pare different ENSO classification methods to deter- mine which provides greater value to Canadian and
US wheat Triticum aestivum L. producers.
Current methods to classify ENSO events rely on sea surface temperature anomalies, sea surface
air pressure differences across the Pacific, or some combination of these and other weather parameters.
Methods using the Southern Oscillation Index SOI rely on sea surface air pressure differences. There
are at least two SOI-based climate forecasting meth- ods. The most commonly used method classifies SOI
events into three phases 3P: low SOI El Niño, other, and high SOI La Niña Climate Prediction
Center, 1998. Within the three-phase method, two different classification schemes are used in this study.
A second method classifies SOI events into five phases 5P: consistently negative, consistently pos-
itive, rapidly falling, rapidly rising, and consistently zero Stone and Auliciems, 1992. Only these two
SOI-based forecast methods, along with perfect fore- casts, are considered because sea surface temperature
data are not available for a sufficient period to provide meaningful comparisons. This study is the first formal
attempt to compare the economic value of different SOI-based forecast methods.
2. Methodology
Currently, few wheat producers are integrating SOI-based climate forecasts into their management
practices because many are unaware of this informa- tion or lack the knowledge to use this information
Nicholls, 1999. Consequently, data containing pro- ducers’ responses to improved climate forecasts are
not available. One method to overcome data limita- tions is to use biophysical simulation models. Such a
simulation model in combination with decision theory is used to identify the value of the 3P, 5P, and perfect
forecasting methods.
2.1. Crop growth simulation model The crop growth model used in this study is the
CERES-wheat model Godwin et al., 1990. The CERES-wheat model is selected because of its de-
tail and inclusiveness regarding wheat growth and development. This model is a process-oriented, man-
agement level model of wheat crop growth and devel- opment that simulates soil water balance and nitrogen
balance associated with plant growth. The model requires daily weather maximum and minimum tem-
perature and precipitation, soil, and variety-specific genetic characteristics. CERES-wheat has been veri-
fied and used to assess different management strate- gies in various locations Savin et al., 1995; Pecetti
and Hollington, 1997.
2.2. Site descriptions and data Six sites within the major winter wheat producing
regions in the US are modeled, whereas seven sites are modeled to represent major Canadian and US spring
wheat-growing regions. Approximate locations of the sites are given in Fig. 1. Descriptions of the sites
are presented in Table 1. The dominant wheat class grown in each region is modeled. Canadian western
red spring wheat is the only wheat class modeled in Canada. Hard red spring wheat is modeled in North
and South Dakota and Montana. Hard red winter wheat is modeled in Kansas, Oklahoma, and Texas. Soft red
winter wheat is modeled in Illinois and Ohio. Soft white winter wheat is modeled in Washington. Rep-
resentative soil characteristics appropriate for wheat growing are identified in Canada using the Soil Land-
scapes of Canada maps Agriculture Canada, 1986a, b, 1991 and in the US using the Map Unit Use File
MUUF Baumer et al., 1984.
Canadian wheat variable production costs are ob- tained from the provincial wheat crop enterprise bud-
gets for the years 1989–1995 and adjusted to 1997 prices using the Western Canadian producer price in-
dex Saskatchewan Agriculture and Food, 1996, 1997; Alberta Agriculture, 1989–1995; Manitoba Agricul-
ture, 1989–1995. Variable production costs in the US
H.S.J. Hill et al. Agricultural and Forest Meteorology 100 2000 261–272 263
Fig. 1. Locations of spring and winter wheat sites.
Table 1 Site and parameter description for modeled wheat sites
Site name Longitudelatitude northwest
Soil type Wheat class
Julian planting dates
a
Seed rate seedsha
b
US winter wheat Illinois
39–89 Silt loam
Soft red 250, 265, 280
3,500,000 Kansas
38–98 Sandy loam
Hard red 240, 255, 270
3,500,000 Ohio
41–84 Silt loam
Soft red 250, 265, 280
4,750,000 Oklahoma
36–98 Silt clay
Hard red 240, 255, 270
3,500,000 Texas
33–99 Silt clay
Hard red 240, 255, 270
1,875,000 Washington
48–118 Silt loam
Soft white 240, 255, 270
2,500,000 US spring wheat
c
Montana 48–110
Silt loam Hard red
95, 110, 115 3,000,000
North Dakota 48–97
Silt loam Hard red
95, 110, 115 4,750,000
South Dakota 45–98
Silt loam Hard Red
95, 110, 115 4,750,000
Canada spring wheat
d
Carmen, Man. 49–98
Clay loam Western red
120, 135, 150 2,750,000
Aneroid, Sask. 49–108
Clay loam Western red
120, 135, 150 1,875,000
Watson, Sask. 52–106
Clay loam Western red
125, 140, 155 2,750,000
Vermilion, Alta. 52–111
Clay loam Western red
120, 135, 150 2,750,000
a
Day of year in Julian days.
b
Seeds per hectare.
c
Hard red spring.
d
Canadian western red spring.
264 H.S.J. Hill et al. Agricultural and Forest Meteorology 100 2000 261–272
are from USDA-Economic Research Service 1996 regional farm budgets for the years 1989–1995. Costs
are adjusted to 1997 prices assuming an annual 3 inflation rate. The price of nitrogen incorporated into
the model is the mean price for the period 1989–1995 U.S.D.A.-N.A.S.S., 1996. Five wheat prices varies
slightly by site because of regional differences rep- resenting the historic range of wheat prices for each
wheat class are used.
Weather data are from Environment Canada 1997 and the US Historical Climatological Network East-
erling et al., 1998; Kaiser, 1998. Solar radiation is ap- proximated using a solar radiation generator Richard-
son and Wright, 1984. Missing information regarding precipitation andor temperature data is approximated
by incorporating data from a nearby location or if no location is available by use of a random weather gener-
ator, WGEN Richardson and Wright, 1984. Because winter wheat is planted in the fall and harvested the
next year, 86 years 1910–1985 of weather data are used to represent 85 cropping years.
Two different procedures to classify the weather years are used for the 3P method. The first method
uses the classification of SOI events by the Climate Prediction Center CPC 1998. Because information
that may not be available at planting time for winter wheat is used in making these classifications, a second
method is also used to classify the 3P events. The CPC yearly classification, which extends roughly from Oc-
tober of year t through September of year t+1, may use information on ENSO that is not available until af-
ter the assumed October planting dates. In classifying the years and yields in this study, the CPC classifica-
tion of year t is used to classify year t+1 yields. To clarify, consider 1920. The classification of 1920’s is
used to impact 1921’s yields planting occurs in the fall 1920 but harvest occurs in 1921. Thus, the CPC’s
October through September classification is consistent with the winter wheat growing season. In the second
method, the years are classified based on the average of the July, August, and September standardized SOI.
If the average value is 0.6 or above is classified as La Niña year, whereas a value of −0.6 is classified as an
El Niño event. Years with an average value between −
0.6 and 0.6 are classified as other years. For spring wheat producers, two similar classifica-
tion procedures are used. First, the CPC’s classifica- tion is used. As with winter wheat the classification is
lagged in terms of its impact on spring wheat yields. The lagged procedure is used because the classification
is based primarily on the winter SOI values. ENSO’s main impact on weather is in the October–AprilMay
time frame. The second method uses the −0.6 and 0.6 values to classify years using the average standardized
SOI value for February, March, and April. Under this method, the producer uses updated information on the
SOI value in decision making.
The years are divided into five phases according to the 5P-classification scheme developed by Stone and
Auliciems 1992. Because winter wheat is planted in the fall, classification of the SOI event is based on
the September classification. Missing phase classifi- cations for winter wheat are obtained by interpolating
the first available monthly SOI number before and af- ter September. For spring wheat, April is selected as
the basis of the SOI phase.
Classification of the years in this study is presented in Appendix A. Although the classifications are sim-
ilar, there are differences between the classification. These differences give rise to different expected val-
ues of the SOI-based forecasts.
Maximum and minimum temperatures for January and July, along with mean annual precipitation, are
presented in Table 2. A range of environmental condi- tions is modeled across the different sites. For exam-
ple, mean annual precipitation ranges from 300 mm at the Montana site to 1008 mm at the Illinois site.
2.3. Economic decision model Decision theory Hilton, 1981 is used to derive
the climate forecast’s value for risk neutral wheat producers. It is assumed the producer’s prior knowl-
edge is the climatological historical distribution of climate conditions represented by the 86 years of
daily weather data. Further, it is assumed each year of the weather data is likely equal. Site specific optimal
input combinations for the climatological distribu- tion at a given price are obtained using the following
model:
π
h,i
p = max
n,d
1 T
T
X
j = 1
p y
ij
n, d − r
1
n −
r
2
y
ij
n, d − vc
i
1
H.S.J. Hill et al. Agricultural and Forest Meteorology 100 2000 261–272 265
Table 2 Site climate characteristics for the years 1910–1994
a
Site Monthly mean temperatures Celsius
Mean annual precipitation mm
January July
Minimum Maximum
Minimum Maximum
US winter wheat sites Illinois
− 6.4
3.5 18.6
31.9 1008
Kansas −
3.9 8.1
21.4 34.1
787 Ohio
− 7.9
1.3 16.2
29.5 907
Oklahoma −
1.7 12.0
21.4 35.5
790 Texas
− 0.9
13.8 21.4
36.2 704
Washington −
6.0 0.4
13.2 28.9
406 US spring wheat sites
Montana −
14.0 −
2.5 9.8
26.1 305
North Dakota −
20.0 −
9.2 14.2
28.6 490
South Dakota −
6.4 3.5
18.6 31.9
365 Canadian spring wheat sites
Alberta −
21.4 −
10.6 9.8
23.6 447
Manitoba −
23.9 −
12.8 11.8
26.3 478
Aneroid, Sask. −
19.0 −
7.2 10.6
27.4 356
Watson, Sask. −
23.3 −
12.6 11.5
25.5 368
a
Source: Easterling et al. 1998 and Environment Canada 1997.
where π
h,i
p is the maximum expected net returns per hectare ha for the climatological distribution of
climate given price p, T is the number of years 85, p is expected price per kilogram kg, y
ij
is yield kgha associated with site i and year j, n is applied nitrogen
in kgha, d is planting date, r
1
is nitrogen cost in kg, r
2
is harvest costs in kg, and vc
i
is other variable costs in ha. Eight different nitrogen levels 15, 30,
45, 60, 75, 90, 115, and 120 kgha are modeled for each site. Three different planting dates, representing
a range of dates are modeled per site see Table 1. Planting dates vary by site because of environmental
conditions. Let x
i
represent the optimal combination of planting date and applied nitrogen associated with
use of the decision maker’s prior knowledge. Note, that for each site only one combination one planting
date and one applied nitrogen level is obtained when the producer is using only prior knowledge.
The model used to obtain the expected returns for each forecasted phase is:
π
k,i
p = max
n,d
1 T
k T
k
X
j = 1
p y
ij
n, d −
r
1
n − r
2
y
ij
n, d − vc
i
2 where π
k,i
p is the maximum expected net returns per ha for forecasted event k given price p, T
k
is the number of years represented by each climate forecast.
T
k
corresponds to the subset of years that are associ- ated with the specific climate phase forecasted. In the
3P method, k equals 1, 2, or 3, whereas k=1,. . . , 5 in the 5P method. When perfect forecasts are valued, T
k
equals 1, because with a perfect forecast each years’ climatic variability is known. Let z
k,i
represent the optimal input combination for the climate conditions
following phase k for site i. Note, an optimal input combination is determined by phase for each site. Ei-
ther three or five different input combinations are pos- sible depending on forecast method.
Optimal expected net returns by phase are given by Eq. 2. To obtain the expected net returns for a given
method, net returns for a phase are weighted by the probability of the phase occurring. Mathematically, the
expected net returns are
π
i
p =
M
X
m= 1
w
m
π
m,i
p 3
where w
m
is the probability of phase m occurring and M
equals three or five depending on the system being
266 H.S.J. Hill et al. Agricultural and Forest Meteorology 100 2000 261–272
modeled. The value of the forecast information is dif- ference between the expected net returns with and
without SOI-based forecasts
V
i
p = π
h,i
p − π
i
p 4
where, V
i
p is the expected value additional net re- turns of the forecasting system for site i given price
p . Obtaining the value of the forecast in this man-
ner is consistent with previous value of information studies Hilton, 1981; Mjelde et al., 1997; Hill et al.,
1998. For the information system to have value, the SOI-based forecasts must alter the optimal input com-
bination relative to the prior knowledge case for at least one of the phases. That is, for either SOI-based
method to have value, z
k,i
cannot equal x
i
for all k. Changes in input usage effect returns through wheat
yield changes caused by adjustments in applied nitro- gen level and planting date and costs through changes
in applied nitrogen level. Changes in returns and costs are reflected in the value of the forecasts. This process
is repeated for each of the five price levels.
3. Results and discussion