Identifying drought-affected paddy rice fields using satellite-based temporal vegetation dynamics

This study presents a data-driven framework for identifying drought-affected paddy rice fields associated with actual agricultural drought events in South Korea. The proposed approach examines spatiotemporal patterns of multiple vegetation- and moisture-related indices derived from high-resolution satellite observations to distinguish paddy fields experiencing water stress from normal growing conditions. Spectral–temporal characteristics of paddy fields and barren land are analyzed to detect paddy pixels exhibiting barren-like behavior during drought periods. The framework is demonstrated over Chungcheongnam-do, a major agricultural region where severe water shortages in paddy fields were reported during recent drought events. A Long Short-Term Memory (LSTM) model is employed to capture temporal dependencies in vegetation dynamics. Satellite observations from non-drought years are used for model training and validation, and the trained model is subsequently applied to drought years to identify anomalous paddy field responses. Drought-affected paddy areas are delineated based on the persistence and duration of barren-like conditions relative to the crop phenological cycle. To enhance interpretability, permutation-based feature importance analysis is conducted to assess the contribution of individual indices and to identify those most effective in distinguishing drought-affected conditions. By establishing quantitative criteria for delineating previously ambiguous drought-impacted paddy areas, the proposed framework provides a basis for improved assessment of agricultural drought impacts and supports more robust monitoring of crop stress under variable hydroclimatic conditions.Keywords: Agricultural Drought, Paddy Rice Fields, Vegetation Dynamics, Satellite Remote Sensing, Data-driven FrameworkAcknowledgementThis work was supported by the Korea Environmental Industry & Technology Institute (KEITI) through Water Management Program for Drought, funded by the Korea Ministry of Climate, Energy and Environment (MCEE). (RS-2022-KE002032) and was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (RS-2024-00334564). Also, This research was supported by the Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education(RS-2024-00356439) and was supported by the National Research Foundation of Korea (NRF) (RS-2021-NR060085) funded by the Korea government (MSIT).