Predicting Historical Peanut Yields for Anantapur Region in India with Three Crop Models: Calibration, Aggregation, and Bias Correction.

Our objective was to evaluate methods of calibration, aggregation, and bias correction for predicting district-level historical peanut yields in the Anantapur district of India with three crop models: APSIM, DSSAT, and INFOCROP.  Long-term district-level pod yields and weather data were available for the District from 1980 to 2007.  Six years of experimental data over 36 treatments varying in sowing date, sowing density, and irrigation vs. rainfed were available for the TMV-2 cultivar for Anantapur.   Using that site-specific data, cultivar traits for each crop model were calibrated first for prediction of phenology (anthesis and maturity), then for final pod yield, total crop biomass, and to a lesser extent LAI, seed size, and harvest index.  In order to represent the variability within a region contributing to yields within a year, we predicted district-level yields averaged over 81 cases:  9 rainfall stations, 3 soil types, and 3 sowing windows, with sowing within windows conditioned upon receipt of sufficient rainfall to trigger sowing.  We computed a bias-adjustment by regressing simulated regional yield (mean of 81 cases) against observed regional yield for the 28 years.  This confirmed that yield in farmer fields was about 30% less than in research plots.  Also, averaging over the 81 cases gave better bias-adjustment than a single simulation. After bias-adjustment, the simulated and historical regional yields were plotted versus year to see the extent to which model-simulated response to weather accounted for yield variation.  The crop models accounted for a high percentage of the yield variation attributed primarily to rainfall variation.  We will discuss calibration of the different models, the bias adjustments, lack of technology trend in the historical data, the benefits of multi-model assessment, and how to best weight aggregation over predominant weather sites, dominant soils, and primary sowing dates.