Submitted to: Agronomy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/20/2002
Publication Date: 5/1/2003
Citation: CARBONE, G.J., MEARNS, L.O., MAVROMATIS, T., SADLER, E.J., STOOKSBURY, D. EVALUATING CROPGRO-SOYBEAN PERFORMANCE FOR USE IN CLIMATE IMPACT STUDIES. AGRONOMY JOURNAL. 2003. v. 95. p. 537-544.
Interpretive Summary: Computer simulations have been used by researchers to analyze how environmental variables, such as weather and soils, affect agricultural crops. These tools have also assisted farmers in management decisions and have helped to measure the potential agricultural impacts of future climate variability and change. Of course, crop simulation models must be tested to evaluate their accuracy. Ideally, this is done in individual fields where the details of crop stage and condition can be monitored closely. But since climate impact studies are done over extensive regions, making detailed measurements impractical, it is necessary to evaluate crop models using inputs that are readily available at a coarse spatial scale. Our study evaluated a soybean model at 8 locations in the Southeast United States for 20 years worth of data. The model estimated observed average yield and the magnitude of standard deviation very accurately, but showed weakness in replicating the pattern of year-to-year yield variability. In those individual years where the model performed poorly, extreme weather events or other factors not explicitly considered by the model (e.g. pests and disease) compromised model performance. Our results support the common use of crop simulation models for climate impact research.
Technical Abstract: Researchers frequently use crop simulation models to estimate the impacts of climate variability and change on agricultural production. While most of the models used for this purpose have been validated thoroughly at the field level, few studies have evaluated them for long time series, across multiple sites, and with surrogate inputs commonly used in climate impact studies. Our objective in this paper was to examine how well CROPGRO-Soybean performs across space and time using published genetic coefficients and soil inputs that were estimated from readily-available soil surveys. We evaluated the model at eight agricultural experiment stations in the southeastern United States with respect to its ability to replicate the mean and standard deviation of observed yield. The weighted mean squared deviation across all sites and years was 0.42 (Mg ha-1)2. The model simulated mean yield and the magnitude of interannual yield variability very well. Its inability to capture accurately the pattern of interannual variability contributed most to mean squared deviation. In years when the model performed poorly, extreme weather events or other factors not explicitly considered by the model (e.g. pests and disease), rather than the use of surrogate inputs, compromised model performance. Our results are relevant to climate impact studies that assume crop simulation models can adequately characterize mean yield and the magnitude of interannual variation.