Impact of Training Set Size and Lead Time on Early Tomato Crop Mapping Accuracy

Estimating key crop parameters (e.g., phenology, yield prediction) is a prerequisite for optimizing agrifood supply chains through the use of satellite imagery, but requires timely and accurate mapping. The moment in season number training sites used are two main drivers classification performance. combined effect these was analysed tomato classification, 125 experiments, using three machine learning (ML) classifiers (neural network, random forest, support vector machine) response surface methodology (RSM). Crop performance between minority (tomato) majority (‘other crops’) classes assessed evaluation metrics: Overall Accuracy (OA) G-Mean (GM), which were calculated on large independent test sets (over 400,000 fields). RSM results demonstrated that lead time interaction most important all ML classifiers. demonstrate feasibility preharvest with high performance, an RSM-based approach enables identification simultaneous effects several factors SVM achieved best grading performances across classifiers, according to both metrics. reached highest accuracy (0.95 OA 0.97 GM) earlier (low time) less than other permitting reduction cost ground truth collection field campaigns.