Assessing Durum Wheat Yield through Sentinel-2 Imagery: A Machine Learning Approach

Two modeling approaches for the estimation of durum wheat yield based on Sentinel-2 data are presented 66 fields across three growing periods. In first approach, a previously developed multiple linear regression model (VI-MLR) vegetation indices (EVI, NMDI) was used. second reflectance all bands several dates during growth periods were used as input parameters in machine learning algorithms, i.e., random forest (RF), k-nearest neighbors (KNN), and boosting regressions (BR). Modeling results examined against collected by combine harvester equipped with mapping system. VI-MLR showed moderate performance R2 = 0.532 RMSE 847 kg ha−1. All enhanced accuracy when images used, especially RF KNN (R2 > 0.91, < 360 ha−1). Additionally, remained high 0.87, 455 ha−1) from start period until March, months before harvest, indicating suitability early prediction wheat, information considered essential precision agriculture applications.