Mapping spatiotemporal agricultural droughts from 2019 to 2024 in Northeast Thailand using multi-temporal and multiple sensor data together with random forest algorithm

Agricultural drought exerts a direct impacts food production and livelihoods, highlighting the critical importance of effective natural resource management in regions frequently affected by these events. Earth observation (EO) satellites offer rich geometric, spectral, and temporal resolution data that are increasingly used to map and monitor drought conditions. However, accurately mapping agricultural droughts in tropical countries like Thailand remains challenging due to the need for extensive training datasets, variable precipitation patterns, persistent cloud cover, and small field sizes. To address these challenges, this study performed a series of analyses: i) investigated the relationship between the Soil Moisture Index (SMI) and the Standardized Precipitation Evapotranspiration Index (SPEI) across wet and dry seasons; ii) developed Random Forest Regression (RFR) models integrating multi-temporal Sentinel-2 (S2) imagery with several vegetation indices and SMI derived reference data to map drought occurrence from 2019 to 2024; and iii) assessed agricultural drought trends for major crops, including rice, sugarcane, cassava, and rubber trees within the Northeast Thailand. The SMI showed the strongest correlation with SPEI-3, with R values ranging from 0.7 to 0.8. The RFR models were highly efficient for all years, with R values exceeding 0.65. Spatiotemporal analysis indicated that the most severe drought events occurred consistently between March and May annually. Regions exhibiting the steepest drought trends were often located in irrigated areas, reflecting changes in water availability and cropping practices over time, despite generally low drought severity. Overall, SMI proved to be a robust reference dataset, while the RFR models showed high reliability for monitoring agricultural droughts in cloud regions. This work offers a valuable approach for generating training data in areas with limited ground observations. The results provide support for agricultural drought mitigation and crop water management. • Mapping droughts remains challenging due to extensive training datasets. • Soil moisture index (SMI) indicator proved to be an excellent reference dataset. • Random Forest Regression (RFR) showed high reliability with R exceeding 0.65. • Areas with the steepest drought trends were primarily irrigated zones.