In-Season Corn Yield Prediction Using Satellite-Derived Solar-Induced Chlorophyll Fluorescence and Machine Learning Algorithms

Accurate in-season crop yield prediction is critical for timely agricultural decision-making, food security, and climate-resilient farm management. This study presents a framework for forecasting corn yield using only satellite-derived solar-induced chlorophyll fluorescence (SIF), a proxy for photosynthetic activity, as input to machine learning models. Biweekly SIF observations were collected from June to September over five growing seasons (2015–2020) for 210 corn-dominated counties in the U.S. Corn Belt. These series were used to train a feed-forward neural network, with performance evaluated under a nested leave-one-year-out cross-validation scheme simulating real-world forecasting. A key contribution of this study is the systematic evaluation of all possible in-season SIF combinations, revealing a consistent increase in model accuracy as the season progresses. Feature selection using the Boruta algorithm identified mid-to-late season SIF observations as the most predictive, corresponding to critical crop development stages. Across all timeframes, SIF-based models outperformed those using conventional vegetation indices (NDVI and NIRv) in terms of RMSE and R², particularly in later stages of the growing season. These findings demonstrate that SIF alone, when sampled with high temporal resolution and integrated with machine learning, enables reliable in-season yield prediction at regional scales. The proposed approach provides a scalable and data-efficient solution for operational crop monitoring and pre-harvest yield prediction.