A deep learning crop model for adaptive yield estimation in large areas

Abstract

Estimating crop yield in large areas is essential for ensuring food security and sustainable development. Accounting for variations in the temporal cumulative growth of crops across regions (i.e., spatial heterogeneity of crop growth) can improve the accuracy of yield estimation in large areas. However, current spatial heterogeneity learning methods have limitations such as cutting off inherent correlations among regions, difficulty obtaining accurate prior knowledge, and high subjectivity. Therefore, this study proposed a novel deep learning adaptive crop model (DACM) to accomplish adaptive high-precision yield estimation in large areas, which emphasizes adaptive learning of the spatial heterogeneity of crop growth based on fully extracting crop growth information. Results showed that the DACM achieved an average root mean squared error (RMSE) of 4.406 bushels·acre−1 (296.304 kg ha−1), with an average coefficient of determination (R2) of 0.805. Compared with other state-of-the-art machine learning and deep learning methods, DACM improves the large-area yield estimation accuracy and performs more robustly in space. The analyses on attention values and estimation stability demonstrate that DACM can learn the spatial heterogeneity of crop growth and adopt adaptive strategies to optimize yield estimation. Considering both performance stability and interpretability, DACM provides a practical approach for estimating large-area crop yields by adaptively learning the spatial heterogeneity patterns of crop growth.