Integrating Sentinel-2 Imagery and Machine Learning for Crop Damage Evaluation in Thai Agriculture

This research enhances food security and sustainable agriculture by improving crop damage assessments, aligning with SDG 2: Zero Hunger and SDG 13: Climate Action. Accurate damage evaluations can promote growth in Thailand's agribusiness sector, where over 30% of the population works in agriculture, frequently affected by drought. The goal is to create a Crop Damage Assessment Index (CDAI) model for droughtaffected areas with the assumption underlying this research is that vegetation indices derived from satellite imagery can effectively monitor and predict crop damage. We are analyzing over 5 million damaged and non-damaged plots across nine provinces to train the model. Using 9 vegetation indices and the Random Forest machine learning technique, the model distinguished between damaged and undamaged plots with Sentinel-2 satellite imagery at 20 meters resolution. MNDWI stands out as the most influential index, especially for maize and rice. NDVI is notable for sugarcane, while REIP is crucial for cassava. Other indices, such as NDWI, S2REP, and SAVI, show varying contributions. These tailored indices support the model's strong performance, with ROC values of 86% for rice, 91% for maize, 89% for cassava, and 86% for sugarcane. These results highlight the reliability of the CDAI model in distinguishing drought-affected plots. The CDAI model supports farmers and agricultural officers with accurate assessments, essential for drought management and fair compensation. Accessible via the GISTDA platform (cropsdrought.gistda.or.th), it aids decision-making without requiring GIS knowledge, though ongoing validation and data updates are necessary for improved efficiency.