Crop yield predictions through satellite remote sensing and machine learning: Case studies in cotton and sugarcane

Traditional methods for crop yield prediction, such as manual stalk counting in sugarcane and boll counting in cotton, are labor-intensive and not feasible for large-scale field estimates. This research explores using satellite remote sensing and machine learning (ML) techniques to predict crop yield for cotton and sugarcane. Six machine learning models were used: Random Forest (RF), Artificial Neural Networks (ANN), Support Vector Machine (SVM), Partial Least Squares Regression (PLSR), Xtreme Gradient Boosting (XGB) and Simple linear regression (SLR). Among these models, RF consistently showed the best performance for predicting yield for both crops, showing lower RMSEs. The integration of vegetation indices (VIs) such as NDVI, GNDVI, and NDRE, and multi-temporal satellite data significantly improved model accuracy. Moreover, yield prediction accuracy increased as crops matured further into the growing season, with the RF model performing better in later growth stages showing RMSE values below 4200 kg ha⁻¹ and 259 kg ha⁻¹ for cane yield and cotton yield, respectively. The results prove the effectiveness of combining remote sensing and ML as a field-scale precision agriculture technique for yield prediction in both cotton and sugarcane. This study highlights the potential of advanced data analytics for improving resource management and decision-making for sugarcane and cotton producers.