Spatial Modeling of Sugarcane Yield Using Machine Learning Approaches

Abstract Sustainable precision agriculture and artificial intelligence (AI) are transforming modern farming by enhancing resource efficiency, addressing climate challenges, and fostering sustainability. In Colombia, sugarcane cultivation is pivotal in the agricultural economy, particularly in the Cauca River Valley, a region grappling with climate variability and resource optimization challenges. This study introduces an innovative methodology that combines machine learning (ML) techniques with geospatial data to predict sugarcane yields accurately. The research employs the CatBoost, random forest, and XGBoost ML algorithms, integrating historical data on climate, soil characteristics, and agricultural management practices to develop spatially detailed predictive models. The findings demonstrate that ML-based approaches outperform traditional methods like linear and penalized regressions, offering more precise agricultural planning and resource management predictions. Focused on the municipality of La Candelaria, this work underscores the potential of ML technologies to enhance productivity while promoting sustainable agriculture.