The Testing of a Limited-Data Corn Yield Model for Large-Area Corn Yield Predictions

Abstract

A general philosophy of examining models used in agriculture is outlined in four steps: 1) the examination of the assumptions of the model, 2) sensitivity analysis, 3) tests of reasonableness using available data, and 4) designing of an experiment to rigorously test a model. This philosophy is applied to a regression model developed by Leeper et al. (1974) to predict corn yields in Illinois. The ability of the Leeper model to account for weather-induced variations in statewide corn yields for a historical sequence from 1968 to 1976 is examined. The predictive capabilities of the model are analyzed for 1977 and 1978 utilizing an adjustment equation obtained from the 1968?76 time period. This examination represents step three in the testing procedure. When compared to Crop Reporting Board state yield estimates, the model accurately reflects weather-induced variations for Iowa, Illinois and Indiana during the 9-year historical sequence. The model does not work well for Missouri. The predictive ability of the model for Iowa was very good (?2.4% in 1977, ?1.5% in 1978) for both years. The model did not work as well for Illinois (4.1%, 7.0%) or for Indiana (?9.7%, ?0.7%) but was still within the 4?8% error in Crop Reporting Board estimates in all cases but one. The predicted yields for Missouri (?14.3%, ?8.6%) had considerably more error and were outside the error range of the Crop Reporting Board estimates. This indicates that this model could give some assistance in an early season (mid-August) estimate of the weather-induced changes in corn yields in three of the four states examined.