Yield Forecasting Of Sugarcane Using Machine Learning And UAV-Derived Canopy Height

This study proposes a machine learning approach for early sugarcane yield forecasting using UAV-derived canopy height from Digital Surface Model (DSM) images, tested under varied nitrogen and water treatments from 62 different sugarcane blocks, subjected to nine treatment zones (three water and three nitrogen levels). A 3D-regression analysis of the nine treatment zones showed a strong yield correlation of $\mathrm{R}^{2}=0.825$. To address the limited data, a synthetic dataset of 900 samples was generated and used to train three machine learning models-SVR, Random Forest, and XG-Boost-achieving $\mathrm{R}^{2}$ values of 0.54,0.77, and 0.87, respectively.