Assessing long-term weather variability impacts on annual grain yields using a maize simulation model

Abstract

Process-based model simulation studies using legacy data can be used to expand LTAR (Long-Term Agroecosystem Research) enabling exploration of factors otherwise difficult to measure in the field. Management strategies to improve yield stability in response to long-term weather variability can be readily evaluated. MAIZSIM is a coupled crop and soil simulation model that simulates processes at an hourly time-step. The model was evaluated using 20 years of management and yield data from the ARS Farming Systems Project (FSP) in Beltsville, MD. We also compared model performance relative to previously reported empirical relationships between growing season weather and FSP yield. The model was calibrated using two parameters (staygreen, juvenile leaf number). Model fit was good (Index of Agreement = 0.92, Mean Bias Error = 51 kg ha^-1), but low measured yields were overpredicted and high measured yields were underpredicted. The effect of interannual weather variability was comparable between measured and modeled yields and followed FSP empirical relationships, revealing that MAIZSIM simulated long-term agronomic trends associated with annual weather patterns supporting use of similar model applications when LTAR data aren’t available. Commonality analysis revealed that cumulative precipitation from 9 to 13 weeks and heat stress from 8 to 13 weeks after planting accounted for 62 % of explained (R² = 0.84) annual simulated yield variation. Adapting management strategies (cultivar selection, planting rate, planting date) to avoid critical period water and heat stress could help to minimize yield losses, particularly under future weather scenarios with more variable precipitation patterns and higher growing season temperatures.