Remote sensing estimation and mapping of maize yield in the Huang-Huai-Hai Plain using integrated crop growth models and deep learning

China is the world’s second-largest corn producer, contributing 23% to global production and playing a vital role in stabilizing the global corn supply. Accurate estimation of corn yield is critical for ensuring food security and formulating agricultural policies. This study combines the World Food Studies (WOFOST) model with deep learning to estimate corn yield and has generated a dataset of corn yield per unit area in the Huang-Huai-Hai Plain from 2014 to 2023. Using the WOFOST model, which incorporates meteorological, soil, crop, and management data, the study provides a dynamic simulation dataset of yield formation processes backed by agronomic mechanisms, addressing the issue of sample scarcity in traditional statistical models. Building on this, the deep learning model GRU (Gated Recurrent Unit) captures nonlinear relationships in time-series data and accounts for the cumulative effects of various growth stages during crop development, improving the accuracy of yield estimation. Subsequently, the initial results are corrected using statistical data to ensure closer alignment with actual conditions. Results demonstrate that the estimated accuracy from 2014 to 2023 exceeds 85% overall, with correlation coefficients (r) ranging from 0.47 to 0.76 and RMSE values ranging from 487 to 1 168 kg·hm⁻². This method performs well even without relying on field survey data, facilitating large-scale mapping of corn yield distribution. By integrating crop models based on growth process simulation with deep learning, the method resolves the issue of sample scarcity, strengthens the role of agronomic knowledge in yield estimation, and provides new insights for crop yield estimation, with the potential to be applied to other crops.