A probabilistic framework for forecasting maize yield response to agricultural inputs with sub-seasonal climate predictions

Abstract

Crop yield results from the complex interaction between genotype, management, and environment. While farmers have control over what genotype to plant and how to manage it, their decisions are often sub-optimal due to climate variability. Sub-seasonal climate predictions embrace the great potential to improve risk analysis and decision-making. However, adequate frameworks integrating future weather uncertainty to predict crop outcomes are lacking. Maize (Zea mays L.) yields are highly sensitive to weather anomalies, and very responsive to plant density (plants m ^−2 ). Thus, economic optimal plat density is conditional to the seasonal weather conditions and can be anticipated with seasonal prospects. The aims of this study were to (i) design a model that describes the yield-to-plant density relationship (herein termed as yield–density) as a function of weather variables, and provides probabilistic forecasts for the economic optimum plant density (EOPD), and (ii) analyze the model predictive performance and sources of uncertainty. We present a novel approach to enable decision-making in agriculture using sub-seasonal climate predictions and Bayesian modeling. This model may inform crop management recommendations and accounts for various sources of uncertainty. A Bayesian hierarchical shrinkage model was fitted to the response of maize yield–density trials performed during the 2010–2019 period across seven states in the United States, identifying the relative importance of key weather, crop, and soil variables. Tercile forecasts of precipitation and temperature from the International Research Institute were used to forecast EOPD before the start of the season. The variables with the greatest influence on the yield–density relationship were weather anomalies, especially those variables indicating months with above-normal temperatures. Improvements on climate forecasting may also improve forecasts on yield responses to management, as we found reduced bias and error (by a factor >10), and greater precision (e.g. R ^2 increased from 0.26 to 0.32) for cases where weather forecasts matched observations. This study may contribute to the development of decision-support tools that can trigger discussions between farmers and consultants about management strategies and their associated risks.

Keywords

• corn
• GxExM
• stand density
• plant density
• Bayesian modeling
• shrinkage models