Modeling Spatiotemporal Variability in Corn Yield Using AI and Remote Sensing

Abstract Spatiotemporal variability in corn yield provides important information for understanding how in-season crop conditions and weather patterns influence field-level yield variation. Therefore, the objectives of this study were: (a) to understand the spatiotemporal variations of corn yield in South Dakota, USA and (b) to investigate the potential of integrating remote sensing data and AI methods to predict corn yield. This study evaluated the potential of different modeling approaches for predicting corn yield using remote sensing data in South Dakota, USA. Three input scenarios were evaluated: (I) multi-stage vegetation data, (II) spectral indices during the reproduction phase (R1), and (III) the mean of all temporal indices. Using eight years of data (2017–2024), the study assessed the performance of multiple linear regression (MLR), support vector regression (SVR), and deep learning models (LSTM and GRU) under diverse environmental conditions, including dry, normal, and wet years. The results showed that deep learning models generally outperformed multiple linear regression and support vector regression across most years and input scenarios. In addition, Scenario I produced the strongest model performance, highlighting the importance of incorporating multi-stage vegetation information across the growing season. LSTM was most effective in dry (2017, 2021) and normal years (2018), while GRU performed best in wet (2019), very dry (2020, 2022), and normal years (2023, 2024). These findings demonstrate the potential of integrating remote sensing data with AI-based models to improve in-season yield prediction and better capture spatiotemporal variability in corn yield.