Enhancing crop classification with growing Degree days: bridging classical models and regional generalization

Crop type classification remains a critical challenge in remote sensing, particularly in achieving generalization across climatically diverse regions. Despite advancements in classification models, variability in crop growth timing driven by climatic differences continues to hinder cross-regional performance. In this paper, we demonstrate that incorporating Growing Degree Days (GDD), a biologically relevant thermal time measure, into crop modelling significantly improves generalization across regions and growing seasons. Using data from two climatically distinct regions in Turkey and spanning the 2023 and 2024 growing seasons, our approach consistently enhances classification performance. Numerical results show that integrating GDD yields accuracy improvements of up to 43% in Random Forest, 29% in Gradient Boosting, 32% in XGBoost, 12% in Support Vector Machines (SVM), and 40% in MLP. Across all models, F1-scores also exhibit significant increases, with some crops achieving gains of over 30%. These findings highlight the robustness and scalabilityof GDD for improving cross regional classification.