1. Introduction
If we have a reliable seasonal climate forecast at the beginning of the crop growing season, we can estimate the upcoming season crop yield amount reasonably well using a dynamic crop model. This would be very beneficial in helping farmers and/or crop decision makers to prepare for the crop growing season (Jones et al. 2000; Hansen 2002; Cabrera et al. 2009). Farmers may take certain mitigation measures (e.g., changing planting date, adopting different crop variety) or purchase insurance. However, the temporal and spatial resolutions of most readily available seasonal climate forecasts are too low to use directly in a crop model. A crop model requires a season-long input of daily weather data to simulate the crop yield amount. A skillful seasonal forecast in a monthly or seasonal average sense is necessary, but does not guarantee a good crop yield forecast (Shin et al. 2006; Baigorria et al. 2007). The seasonal climate forecast should capture the high-frequency variability of weather (e.g., wet/dry spell sequences) properly to have a reliable projection from the crop model.
The climate in the southeastern United States has a strong teleconnection to tropical Pacific Ocean sea surface temperatures (e.g., Ropelewski and Halpert 1986; Higgins et al. 2000; Cocke et al. 2007) and therefore has some degree of predictability, especially in the winter. The Southeast Climate Consortium (SECC) developed a climate-based decision support system (http://agroclimate.org) and currently uses El Niño–Southern Oscillation (ENSO)-based historical weather data to implement a probabilistic yield risk forecast for a variety of crops (e.g., cotton, maize, peanut, potato, and tomato). The yield risk forecast is based on location, planting date, soil type, and the ENSO-based climate scenario (Fraisse et al. 2006). For example, if it appears likely that La Niña–type conditions might develop for the upcoming season, weather data from historical La Niña–type years are used to drive a crop model to generate a probabilistic yield risk forecast. However, there is a critical problem in this approach: the ENSO signal is relatively weak during the summer crop season over the southeastern United States. Hence, the categorical projections based on the ENSO index may not provide a useful guideline to stake holders. The limitation of this approach is evaluated in this study.
As an alternative to the above ENSO-based approach, we may use statistical and dynamical downscaling of global climate model (GCM) prediction to generate seasonal forecasts at the station level. While simple statistical methods (e.g., weather generators) have been extensively used (e.g., Dubrovsky et al. 2000; Phillips et al. 1998), recently developed advanced statistical downscaling methods (e.g., Lim et al. 2007; Schoof et al. 2009) have not yet been applied to crop models. A few dynamically downscaled seasonal datasets have been used recently in crop model applications to study the potential predictability of crop yields (e.g., Shin et al. 2006; Baigorria et al. 2007, 2008). However, a comprehensive study has not been performed to intercompare the usefulness of available seasonal forecasts in crop yield simulations.
The main objective of this study is to evaluate alternative crop (maize and peanut) yield forecast techniques, using various seasonal climate projections to drive a dynamical crop model, and compare it to the current ENSO-based yield projection practice. The question we wish to ask is, can a dynamical regional model or statistical downscaling provide sufficiently detailed and reasonably more accurate seasonal climate information for use in crop yield forecasting? To compare with the ENSO-based forecast, this study examines both dynamically downscaled daily data using the Center for Ocean–Atmospheric Prediction Studies (COAPS) regional climate model (RCM; ∼20 km) and statistically downscaled data from both the National Centers for Environmental Prediction (NCEP) Climate Forecast System (CFS) and the COAPS global climate model. Yield sensitivity studies, using the Decision Support System for Agrotechnology Transfer (DSSAT) crop model (Jones et al. 2003), are conducted by using these various daily climate data.
The paper continues in sections 2 and 3 with brief descriptions of the seasonal climate data generated in this study and crop simulation experiments, respectively. In section 4 the current ENSO-based crop yield practice is reviewed and comprehensively compared with the eight different approaches, followed by concluding remarks in section 5.
2. Season-long daily climate data
In addition to the ENSO-based climate data, eight different sets of daily seasonal climate data are produced by using the COAPS global climate model and regional climate model (Cocke and LaRow 2000; Shin et al. 2005), the NCEP CFS (Saha et al. 2006), and a statistical downscaling method (Lim et al. 2007). The COAPS GCM and RCM are run with two different cumulus parameterization schemes [simplified Arakawa–Schubert (SAS; Pan and Wu 1994) and relaxed Arakawa–Schubert (RAS; Rosmond 1992)] to examine the sensitivity to precipitation physics in the model. Table 1 shows the nine-season-long daily climate datasets used in the maize and peanut crop simulations. The abbreviations used in Table 1 are employed to indicate the corresponding climate data.
The ENSO-based daily climate data (D0 in Table 1) are obtained from the National Weather Service cooperative station network. The cooperative observing program has more than 100 yr of observational climate data over the United States. The daily data used in this study cover the period from 1911 to 2006. Since there were too many missing daily values in 1919, that year was excluded. Maximum and minimum temperatures and precipitation are used to drive crop models along with independently observed (or estimated) solar radiation data.
The crop (i.e., maize and peanut) growing season (March–September) simulations are performed for a period of 19 yr (1987–2005) with the COAPS GCM and RCM using observed weekly sea surface temperatures (SSTs). In these simulations, 2 different convection schemes (SAS and RAS) are used along with 10 different atmospheric initial conditions (specifically, 19–28 February) to develop ensembles that characterize uncertainty in the simulations (D1, D2, D3, and D4 in Table 1). The statistical downscaling technique is then applied to these COAPS seasonal datasets to generate new sets of seasonal data (D5 and D6 in Table 1). The CFS ensemble seasonal forecasts are obtained from NCEP and are also downscaled statistically (D7 and D8 in Table 1). More details on the climate models and the statistical method employed are provided in the following subsections.
a. COAPS GCM and RCM with RAS and SAS (D1–D4)
The COAPS GCM and RCM are used in this study to construct 10-member ensemble datasets of season-long daily climate. The COAPS GCM horizontal resolution is T63 (approximately 1.875°) with 17 vertical levels. Prior to use in the crop models, the GCM outputs are converted to regional grids (20 km by 20 km; see Fig. 1) using the Cressman objective analysis scheme. The RCM is nested in the GCM and runs at 20-km horizontal resolution, roughly resolving the county scale (Fig. 1). The RCM can add additional skill by improving the spatial representation of weather systems (Cocke et al. 2007). To improve seasonal surface climate simulations, the COAPS GCM and RCM have been coupled with the National Center for Atmospheric Research Community Land Model version 2 (Shin et al. 2005, 2006).
Since precipitation is a very important input for a crop model, a proper parameterization of moist convection in an atmospheric model is essential to simulate the rainfall frequency and amount accurately. Different convective schemes can produce significantly different results in crop simulations. Two commonly employed convective schemes in the COAPS models are the SAS and RAS schemes. This study will assess their impacts on the crop yields. The main difference between these two convective schemes is that the RAS seeks to relax toward a quasi-equilibrium state rather than adjusting instantaneously to equilibrium as in the SAS scheme. Details of each scheme may be found in the references cited above.
b. Statistical downscaling (D5 and D6)
There is a wide variety of methods for statistical downscaling, ranging from simple interpolation, regression, and analog methods to more complex techniques such as artificial neural networks (e.g., Tolika et al. 2007; Robertson et al. 2007; Schoof et al. 2009). The main technique used in this study for producing downscaled climate data consists of Cyclostationary Empirical Orthogonal Function (CSEOF) analysis, multiple regression, and stochastic time series generation. Lim et al. (2007) showed that CSEOF analysis is a very efficient technique to extract the complete spatiotemporal evolution of significant climate signals over a cyclic period, compared to conventional eigen-techniques. This mode of data decomposition better enables the subsequent regression method to extract GCM evolution patterns that are physically consistent with the evolution of the observed climate modes. Using CSEOF and multiple regression against a network of high-resolution observations, we can identify the statistical relationships between coarse-scale and finescale climate variability, and hence producedownscaled fine-resolution climate data. The complete description of the statistical downscaling technique can be found in Lim et al. (2007). This statistical downscaling method is applied to both COAPS and CFS global model outputs to measure its usability in crop models.
c. NCEP CFS (D7 and D8)
The CFS was developed at NCEP to improve dynamical seasonal forecasts (Saha et al. 2006). It is a fully coupled one-tier model that includes ocean, land, and atmospheric components. The 9-month-long daily CFS reforecast data are available at 2.5° horizontal resolution for the period of 1981–2006 (http://cfs.ncep.noaa.gov). Although the CFS provides a number of atmospheric variables, only daily precipitation data are employed in this study for use in crop models. Since the required maximum–minimum temperature and solar radiation are not available from the CFS output, we used observed values instead. As will be discussed later in the paper, the precipitation accounts for about 80% of the variance of the crop model yield. Therefore, using observed values of temperature and radiation should not contribute significantly to the skill of the model. For the CFS, a 10-member ensemble (i.e., a set of 10 realizations of seasonal climate) is, in this study, prepared for March–September for each year. This ensemble is generated using time-lagged atmospheric initial conditions (specifically, forecasts from 11, 12, 13, 19, 20, 21, 22, 23, 27, and 28 February). Because the CFS data are only available at 2.5° resolution, the statistical downscaling technique (section 2b) is performed to obtain the CFS data on the 20-km regional grid.
3. Crop simulations
Dynamic crop model systems, as decision supporting tools, have been utilized extensively by agricultural scientists to evaluate possible agricultural consequences from interannual climate variability and/or climate change (e.g., Paz et al. 2007; Semenov and Doblas-Reyes 2007; Challinor and Wheeler 2008). The Decision Support System for Agrotechnology Transfer version 4.0 (Jones et al. 2003) is used to perform crop yield simulations. DSSAT integrates the effects of crop genotype, soil profiles, weather data, and management options into a crop model. It includes several process-based crop models with 27 different crops. The crop model uses maximum and minimum surface temperatures, rainfall, and incoming solar radiation from season-long daily weather records. It computes plant growth and development processes on a daily basis in a specific location, from planting to maturity date. As a result, the impact of weather, soils, and management decisions on a crop yield can be well estimated.
Daily seasonal weather data are used as inputs for the DSSAT crop model for the 19-yr period at Tifton, Georgia (Fig. 1). This site is chosen because the weather data are relatively well observed and maintained for a long period. Missing values of incoming solar radiation are estimated using the technique of Richardson and Wright (1984). In the southeastern United States, maize and peanuts are economically important crops. The Crop Estimation through Resource and Environment Synthesis (CERES)-Maize (Ritchie et al. 1998) and “CROPGRO-Peanut” (Boote et al. 1998) models in the DSSAT are well validated and suitable for simulation during the season of interest. Thus, we select these two crop models to be forced with the seasonal data produced in this study. Soil profiles for the dominant agricultural soil types are based on U.S. Soil Conservation Service county data (see details in Baigorria et al. 2008). Rainfed conditions and fixed fertilizer applications are assumed for management conditions. Each crop model setup assumes the same conditions except for the seasonal climate data input. For maize, 250 kg ha−1 of nitrogen is applied as ammonium nitrate divided in five applications. For peanuts, 40 kg ha−1 of nitrogen is applied as ammonium sulfate in one application at planting. Identical initial soil moisture conditions are also used in all simulations (0.183 cm3 cm−3). While 25 April is used as the planting date for peanuts, 1 April is used for maize.
4. Results
a. ENSO climate (the current practice)
The current practice of crop yield simulations based on ENSO is reinvestigated here. While ENSO is a dominant climate signal over the southeastern United States during winter, it is much weaker during the summer months. As a result, the current crop yield projection practice does not always perform well when the crop season extends into summertime.
Figure 2 shows the peanut yields generated using the daily observed weather data (1911–2006) and the corresponding seasonal rainfall amount, categorized by ENSO phase. The ENSO categories, as used by the SECC, are based on the Japan Meteorological Agency index. For the period 1911–2006, there were 20 El Niño years, 21 La Niña years, and 54 neutral years. In terms of total amount of rainfall during the crop growing season, the average amounts do not appear to depend on ENSO phase. However, there is approximately 65% higher rainfall variability within El Niño years, as indicated by the standard deviations shown in Fig. 2 (lower panel). Meanwhile, the simulations show higher average peanut yields during La Niña years (2500 kg ha−1) than in El Niño years (2150 kg ha−1). The difference is statistically significant at the 15% level. While El Niño years suffer from a somewhat consistent dry period during the crop planting dates (April or May), La Niña years do not show much water deficiency during this period over this region (http://agroclimate.org).
In Fig. 3 we show the cumulative probability distribution function of peanut yields based on ENSO phase categorization. Crop yields are higher for La Niña years than El Niño years for all probability levels. A Kolmogorov–Smirnov test indicates that the difference is not statistically significant. The only significant difference is in the crop yields between neutral and La Niña years, significant at the 10% level. A larger sample size may be needed to establish statistical significance. The maize case is not shown because of similar results.
Using the ENSO-based method, maize and peanut yields are projected for the years 1987–2005 and are shown in Fig. 4. Yield projections based on only three different ENSO phases cannot properly capture the observed interannual variability. The temporal correlation coefficients (r) are 0.202 for maize and 0.306 for peanuts. The root-mean-square errors (RMSEs) are shown as well in the figure. While the crop yields during both the El Niño and the neutral years are very similar to each other, the La Niña years produce slightly higher yields. The observed weather-driven yield variability is usually higher during the neutral years. As mentioned previously, this crop yield projection is the current practice. Hence, our question again: Can we make a better projection than this current practice?
b. Precipitation versus yield
It is well known that rainfall is the most important weather data for rainfed crop yield simulations for the southeastern United States (e.g., Baigorria et al. 2008). Hence, it might be interesting to examine the relationship between precipitation and yield. Temporal correlations (for the years 1911–2006) are computed between the observed weather-driven yields and the crop growing season total rainfall amounts: 0.452 for maize, 0.659 for peanuts. Hence, the explained variability of crop yields due to seasonal rainfall total is approximately 20% for maize and 43% for peanuts. Water availability seems to be more important for peanuts than maize. Hence, a perfect seasonal total precipitation forecast could capture 20%–43% of the variability of the crop yields.
Examination of the rainfalls and yields in 1988 (El Niño year) and 1989 (La Niña year) demonstrates the importance of precipitation frequency and amount. For both years, the total rainfall amounts are almost identical (∼550 mm; Fig. 5b). However, there was only an ∼5000 kg ha−1 maize yield in 1988 as compared with ∼10 000 kg ha−1 in 1989 (Fig. 5a). This difference is due to the water stress. For crop yields, water stress is more important than the total water amount. This means that yields simulated by dynamic crop models are highly sensitive to wet/dry spell sequences during the crop growing season. Not only increasing the persistence of wet–dry day occurrences is important, but the timing within the growing season is particularly important (Baigorria et al. 2007). While the maize crop experienced high water stress during the growing season in 1988, it encountered only a brief water stress period in 1989 (Fig. 6). This explains why there was a much larger yield in 1989. Generally, La Niña years experience many less water stress periods than El Niño years and hence produce higher simulated yields. It is important to realize that the effect of timing of rainfall during the crop growing season depends also on the physical characteristics of soil type, organic matter content, and water holding capacity. Crops planted on sandy soils with lower water holding capacity might be more affected by rainfall frequency than those planted on loamy or clay soils or soils with higher organic matter content. Differences among the water availability with these soils are accommodated in several days before crops being affected by water stress.
Water stress is a function of precipitation, evaporation, runoff, soil moisture, and crop physiology. To characterize overall water stress during the season, we define a water stress index. Although there are numerous ways to define a water stress index (e.g., Rizza et al. 2004), the water stress index used in this study is simply defined as the 120-day average of water stress determined by the crop models. Figure 7 shows the relationship between the water stress index and the maize yield for the period of 1987–2005. Not surprisingly, the correlation between them is −0.89. Hence, if the water stress index is suitably defined [e.g., the average stress index during the critical stage for yield (i.e., 30 days centered around flowering) and if it can be accurately estimated ahead of the upcoming crop growing season using predicted atmospheric variables, the yield outcome can be better predicted.
c. Global versus regional models
Figure 8 shows the simulated maize yields in Tifton from 1987 to 2005 and the corresponding total precipitation amounts using D2 and D4 in Table 1. Relative to the ENSO-based approach, the climate models (especially the regional models) have higher interannual fluctuation of the simulated yields. While the direct output from the COAPS global model (D2) provides no improvement (r = 0.128, RMSE = 4299.3 kg ha−1) relative to the ENSO-based approach (r = 0.202, RMSE = 2969.8 kg ha−1), the regional climate model (D4) captures the interannual variability better in the yield simulation (r = 0.405, statistically significant at the 5% level, RMSE = 2753.3 kg ha−1). Generally, the global model produces much lower yield amounts because of more frequent rain days and less daily rainfall amounts. Meanwhile, the regional model produces better yield average estimations. Therefore, the benefit of dynamical downscaling using a regional model is demonstrated in the crop simulations. To make the GCM output useful for crop applications, it is necessary to downscale the model output, using either dynamical or statistical methods. Similar results are also achieved with peanuts. Skills decrease when changing the convective scheme from RAS to SAS (see section 4f).
d. Dynamical versus statistical models
Instead of using dynamical downscaling data, we can use statistically downscaled data (D5 and D6 in Table 1) to drive the crop models. The statistical technique has some advantages (e.g., computationally cheap) as well as some disadvantages (e.g., physical inconsistency among atmospheric variables). The crop yield simulation results are better than the ENSO-based approach (r = 0.251, RMSE = 2959.9 kg ha−1 with D6). This is better than some of the dynamically downscaled results (r = −0.036, RMSE = 3226.5 with D3), but worse than D4 (Fig. 9). Generally, dynamical downscaling methods have the potential to outperform statistical techniques, particularly because the resulting downscaled climate data are physically consistent, as well as consistent with the GCM output from which they are derived. Nevertheless, the statistical method is a very useful tool when regional climate models and/or sufficient computing resources are not available.
e. CFS versus its statistically downscaled data
It is very interesting to assess the capability of an operational climate model in crop simulations. To explore the feasibility of using the CFS model output in determining crop yields in the southeastern United States, a series of crop model experiments are performed using the daily CFS model output and its statistically downscaled data. The CFS model tends to overpredict precipitation by almost 500 mm season−1 for each of the 19 yr (Fig. 10b), which provides no water stress conditions for the crop models. Hence, the maize yield amounts are approximately 1000 kg ha−1 higher when compared with observations (Fig. 10a; r = 0.043, RMSE = 5609.6 kg ha−1 with D7). Meanwhile, the statistical downscaling remedies the CFS precipitation bias problem and therefore improves maize yields (r = 0.295, RMSE = 2872.3 kg ha−1 with D8). This result is not significant at the 5% level. However, a statistically significant result is obtained for peanut yields (r = 0.674 with D8).
Since the CFS data configuration uses observed maximum and minimum temperatures and estimated solar radiation, it might have an advantage over the other datasets used. As shown in section 4b, water stress (precipitation frequency and amount) is the most important parameter and can explain approximately 80% of the variance in crop yields. Therefore, using observed maximum and minimum temperatures and estimated radiation will not affect the results significantly. To test this, another set of seasonal climate data is generated using the D4 precipitation with the observed other variables to drive the crop models. There is not much difference (not shown). This confirms that precipitation is the most dominant variable in determining the crop yield amount.
f. Overall crop yield estimation
The overall performance using each climate dataset in the crop yield simulations is summarized in Fig. 11 for maize and in Fig. 12 for peanuts. Although there are many other ways to evaluate skill, normalized RMSE and temporal correlation coefficient (r) are used as simple evaluation tools in this study to compute the skills for intercomparison. The normalized RMSEs of the ENSO-based yield for maize and for peanuts are 0.466 and 0.576, respectively. As a percentage of the simulated observed-weather yield mean, the RMSE ranges from 43.2% to 88.1% for maize and from 46.7% to 128.1% for peanuts. For maize, D4, D6, and D8 performed better than D0, whereas for peanuts, D4, D5, and D8 performed better than D0. For both crop yield simulations, D7 performed the worst. Similar conclusions can be obtained in terms of the correlation coefficient.
The yield performance strongly depends on the convective scheme used. The simulated maize yields with the SAS scheme (Figs. 11a,c,e) are not better than those with the ENSO-based approach (Fig. 4a). Generally, the RAS scheme substantially improves the yield estimations (Figs. 11d,f) although the performance of the COAPS GCM with the SAS and RAS schemes is not better than the ENSO-based approach. However, the dynamical and the statistical downscaling methods produce better datasets than the COAPS global model leading to improved crop yield projections. The highest skill for maize yield is achieved with the dynamical downscaling using the RAS scheme (Fig. 11d: r = 0.405, normalized RMSE = 0.432). Meanwhile, the best peanut yield skill is obtained using the CFS model combined with the statistical downscaling technique (Fig. 12h: r = 0.674, normalized RMSE = 0.467). Both results are statistically significant at the 5% level.
5. Conclusions
This paper evaluated the sensitivity of two crop models to nine different seasonal climate datasets for maize and peanut yield simulations. The most commonly employed yield prediction method is the ENSO-based approach. ENSO plays an important role in crop yield projections in many regions of the world. However, ENSO-based techniques exhibit limited summertime forecast skill in the southeastern United States because of the weak ENSO influence. Using two global climate models (COAPS GCM and NCEP CFS), two downscaling methods (dynamical and statistical), and two cumulus parameterizations (SAS and RAS), eight different seasonal climate datasets were generated. Each of those climate datasets has 10 ensemble members to show the uncertainty of the simulations.
Instead of assessing the meteorological skill of the COAPS global and regional models and the CFS global model, this study evaluated the skill of crop yield simulations. A simple skill evaluation of meteorological fields (such as precipitation and surface temperature) is sometimes insufficient owing to the nonlinear crop model responses to the seasonal climate data. For crop model yields, the length and timing of dry–wet spells (or frequency of rainfall) during the growing season are more important than the total seasonal rainfall amount. The current ENSO-based crop yield projection practice was improved by using one of several seasonal datasets, depending on the model configuration and the crop type. The type of convective scheme used turned out to be an important parameterization for the crop yield amounts. Generally, the dynamical and the statistical downscaling approaches perform better in maize and peanut yield simulations than the global climate models.
To improve crop yield predictions further, the currently available climate models should be improved to capture the rainfall frequency and amount more accurately. In addition, a reliable a posteriori bias correction method is needed particularly for precipitation. Multimodel (weighted) ensemble approaches (e.g., Shin et al. 2008) might be very helpful in this approach. The authors are also currently developing a two-way crop–atmospheric coupling using the COAPS regional climate model and the DSSAT crop models over the southeastern United States. Until recently, the DSSAT crop models were only applied to specific locations to determine the crop yields. The crop models are now being expanded and assigned to each of the 20-km downscaled grid points in the southeastern United States.
Acknowledgments
COAPS receives its base support from the Applied Research Center, funded by NOAA Climate Program Office. Additional support is provided by the USDA-CSREES. The views expressed in this paper are those of the authors and do not necessarily reflect the views of NOAA, USDA, or any of its subagencies.
REFERENCES
Baigorria, G. A., J. W. Jones, D. W. Shin, A. Mishra, and J. J. O’Brien, 2007: Assessing uncertainties in crop model simulations using daily bias-corrected regional circulation model outputs. Climate Res., 34 , 211–222.
Baigorria, G. A., J. W. Jones, and J. J. O’Brien, 2008: Potential predictability of crop yield using an ensemble climate forecast by a regional circulation model. Agric. For. Meteor., 148 , 1353–1361.
Boote, K. J., W. Jones, G. Hoogenboom, and N. B. Pickering, 1998: The CROPGRO model for rain legumes. Understanding Options for Agricultural Production, G. Y. Tsuji, G. Hoogenboom, and P. K. Thornton, Eds., Kluwer Academic, 79–98.
Cabrera, V. E., D. Solis, G. A. Baigorria, and D. Letson, 2009: Managing climate variability in agricultural analysis. Ocean Circulation and El Niño: New Research, J. A. Long and D. S. Wells, Eds., Nova Science, 163–179.
Challinor, A. J., and T. R. Wheeler, 2008: Crop yield reduction in the tropics under climate change: Processes and uncertainties. Agric. For. Meteor., 148 , 343–356.
Cocke, S., and T. E. LaRow, 2000: Seasonal predictions using a regional spectral model embedded within a coupled ocean–atmosphere model. Mon. Wea. Rev., 128 , 689–708.
Cocke, S., T. E. LaRow, and D. W. Shin, 2007: Seasonal rainfall predictions over the southeast United States using the Florida State University nested regional spectral model. J. Geophys. Res., 112 , D04106. doi:10.1029/2006JD007535.
Dubrovsky, M., Z. Zalud, and M. Stastna, 2000: Sensitivity of CERES-maize yields to statistical structure of daily weather series. Climatic Change, 46 , 447–472.
Fraisse, C. W., and Coauthors, 2006: AgClimate: A climate forecast information system for agricultural risk management in the southeastern USA. Comput. Electron. Agric., 53 , 13–27.
Hansen, J. W., 2002: Realizing the potential benefits of climate prediction to agriculture: Issues, approaches, challenges. Agric. Syst., 74 , 309–330.
Higgins, R. W., A. Leetmaa, Y. Xue, and A. Barnston, 2000: Dominant factors influencing the seasonal predictability of U.S. precipitation and surface air temperature. J. Climate, 13 , 3994–4017.
Jones, J. W., J. W. Hansen, F. S. Royce, and C. D. Messina, 2000: Potential benefits of climate forecasting to agriculture. Agric. Ecosyst. Environ., 82 , 169–184.
Jones, J. W., and Coauthors, 2003: The DSSAT cropping system model. Eur. J. Agron., 18 , 235–265.
Lim, Y-K., D. W. Shin, S. Cocke, T. E. LaRow, J. T. Schoof, J. J. O’Brien, and E. P. Chassignet, 2007: Dynamically and statistically downscaled seasonal simulations of maximum surface air temperature over the southeastern Unites States. J. Geophys. Res., 112 , D24102. doi:10.1029/2007JD008764.
Pan, H-L., and W-S. Wu, 1994: Implementing a mass-flux convective parameterization package for the NMC Medium Range Forecast Model. Preprints, 10th Conf. on Numerical Weather Prediction, Portland, OR, Amer. Meteor. Soc., 96–98.
Paz, J. O., and Coauthors, 2007: Development of an ENSO-based irrigation decision support tool for peanut production in the southeastern US. Comput. Electron. Agric., 55 , 28–35.
Phillips, J. G., M. A. Cane, and C. Rosenzweig, 1998: ENSO, seasonal rainfall patterns, and simulated maize yield variability in Zimbabwe. Agric. For. Meteor., 90 , 39–50.
Richardson, C. W., and D. A. Wright, 1984: WGEN: A model for generating daily weather variables. U.S. Department of Agriculture, Agricultural Research Service Publication ARS-8, 83 pp.
Ritchie, J. T., U. Singh, D. C. Godwin, and W. T. Bowen, 1998: Cereal growth, development and yield. Understanding Options for Agricultural Production, G. Y. Tsuji, G. Hoogenboom, and P. K. Thornton, Eds., Kluwer Academic, 79–98.
Rizza, F., F. W. Badeck, L. Cattivelli, O. Lidestri, N. Di Fonzo, and A. M. Stanca, 2004: Use of a water stress index to identify barley genotypes adapted to rainfed and irrigated conditions. Crop Sci., 44 , 2127–2137.
Robertson, A. W., A. V. M. Ines, and J. W. Hansen, 2007: Downscaling of seasonal precipitation for crop simulation. J. Appl. Meteor. Climatol., 46 , 677–693.
Ropelewski, C. F., and M. S. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev., 114 , 2352–2362.
Rosmond, T. E., 1992: The design and testing of the Navy operational global atmospheric system. Wea. Forecasting, 7 , 262–272.
Saha, S., and Coauthors, 2006: The NCEP Climate Forecast System. J. Climate, 19 , 3483–3517.
Schoof, J. T., D. W. Shin, S. Cocke, T. E. LaRow, Y-K. Lim, and J. J. O’Brien, 2009: Dynamically and statistically downscaled seasonal temperature and precipitation hindcast ensembles for the southeastern USA. Int. J. Climatol., 29 , 243–257.
Semenov, M. A., and F. J. Doblas-Reyes, 2007: Utility of dynamical seasonal forecasts in predicting crop yield. Climate Res., 34 , 71–81.
Shin, D. W., S. Cocke, T. E. LaRow, and J. J. O’Brien, 2005: Seasonal surface air temperature and precipitation in the FSU climate model coupled to the CLM2. J. Climate, 18 , 3217–3228.
Shin, D. W., J. G. Bellow, T. E. LaRow, S. Cocke, and J. J. O’Brien, 2006: The role of an advanced land model in seasonal dynamical downscaling for crop model application. J. Appl. Meteor. Climatol., 45 , 686–701.
Shin, D. W., S-D. Kang, S. Cocke, T-Y. Goo, and H-D. Kim, 2008: Seasonal probability of precipitation forecasts using a weighted ensemble approach. Int. J. Climatol., 28 , 1971–1976.
Tolika, K., P. Maheras, M. Vafiadis, H. A. Flocasc, and A. Arseni-Papadimitriou, 2007: Simulation of seasonal precipitation and raindays over Greece: A statistical downscaling technique based on artificial neural networks (ANNs). Int. J. Climatol., 27 , 861–881.
A summary of seasonal climate data used (March–September 1987–2005) in maize and peanut crop simulations. Except for the ENSO climate, there are 10 ensemble members in each seasonal dataset.