Predicting Sugarcane Yield Through Temporal Analysis of Satellite Imagery During the Growth Phase

This research explores the estimation of sugarcane (Saccharum officinarum L.) productivity at harvest by leveraging average satellite image time series collected during the growth phase. The study aims to evaluate the effectiveness of diverse modeling approaches, including a Heteroskedastic Gamma Regression model, Random Forest, and Neural Networks, in predicting sugarcane yield using satellite-derived vegetation indices and environmental variables. Key covariates such as sugarcane varieties, production cycle, accumulated precipitation during the growth phase, and the mean GNDVI vegetation index during the growth phase were analyzed in two different locations for two consecutive growth seasons. The research highlights the integration of satellite data with advanced statistical and machine learning techniques to enhance productivity forecasting in agricultural systems, with a focus on sugarcane cultivation. Additionally, the Heteroskedastic Gamma Regression model demonstrated superior performance in explaining productivity variability, particularly in commercial sugarcane fields, achieving an R2 of 0.89. The findings underscore the potential of these models to support informed decision-making and optimize agricultural practices, offering valuable insights for precision farming.