The Effect of Bias in Divisional and State Mean Temperatures on Weather-Crop Yield Model Predictions: A Case Study in Indiana

Abstract

Historical series of mean temperatures for climatological divisions and the state of Indiana contain systematic biases, the greatest being about ?1.0°C in the Central Division in the 1940s. When these data are used in weather-management, crop-yield models they translate into biases in yield simulation and prediction The magnitude and distribution of the yield prediction bias depends on the mean temperature bias and the type of model. The effect of the mean temperature bias on corn (Zea mays L.) yield production was examined for two models: in the first, the historical climatological series was used to fit the regression coefficients, and in the second, a priori regression coefficients were used. For each of the two models, yield simulations and predictions were made with both the original published mean temperature and with those adjusted to the climatological network base for 1976. In the fiat model, the regression-fitting process averaged the effect of the temperature bias over the entire record. The estimates of corn yield trends attributed to management were slightly greater when the adjusted temperature were used in the regression model than when the published temperatures were used. The use of the adjusted temperatures resulted in slightly higher yield predictions but the differences were generally less than 5% of the mean absolute difference between the model predictions and the yields reported by the U.S. Department of Agriculture Statistical Reporting Service. In the second model, the temperature bias was translated directly into the yield simulation, which with adjusted temperature averaged 113 kg ha?1 higher than that stimulated with the original temperature date in the 1950s and 1960s. Although the yield production and simulation errors caused by the mean temperature bias is nontrivial, the temperature bias contributed a relatively small part of the total variance in the yield modeling.