Comparison of climate change impact on rainfed maize yield in Kansas using statistical and process-based models

Abstract

Changing climate and the projected increase in variability and frequency of extreme events make an accurate prediction of crop yield critically important for addressing emerging challenges in food security. Precise and timely prediction of crop yield can provide valuable information to agronomists, producers, and decision-makers. Even without considering climate change, several factors including environment, management, and genetics and their complex interactions make the prediction of crop yield challenging. In this study a statistical-based Multiple Linear Regression (MLR) model was developed to predict rainfed maize yield in Kansas and compared with yield predictions of the DSSAT process-based model to assess the impact of synthetic climate change scenarios of 1 and 2 °C temperature rise. Historic weather, soils, and crop management data were collected and converted to model-compatible formats to simulate and compare maize yield using both models. It was found that DSSAT had a large Root Mean Square Error (RMSE) compared to the MLR model whereas the correlation coefficients (r) were 0.93 and 0.70 for MLR and DSSAT, respectively. These results indicated that predicted yields from the MLR model had a stronger association with the observed yields than the simulated yields from DSSAT. Analysis of climate change impact showed that the reduction in rainfed maize yield predicted by DSSAT was 8.7% and 18.3% for the synthetic scenarios of 1 and 2 °C temperature rise respectively. Reduction in rainfed maize yield predicted by the MLR model was nearly 6% in both scenarios. Due to the extreme heat effect, predicted impacts under uniform climate change scenarios were considerably more severe for the process-based model than for the statistical-based model.