ATG-RFM: adaptive thresholding with GMM-OTSU for flooded rice mapping

Timely and accurate detection of flooded crop areas is critical for agricultural disaster mitigation and yield preservation. This study presents an unsupervised method, adaptive thresholding with Gaussian mixture model-OTSU for rice flood monitoring (ATG-RFM). The proposed method mainly includes:(1) Simple Non-Iterative Clustering for superpixel segmentation; (2) a dual-thresholding strategy combining GMM and Otsu to extract the flooded crops. Validation in Yueyang, China, shows that ATG-RFM achieves an overall ac-curacy of 97.3%, higher than conventional GMM (87.0%) and Change Detection and Threshold (94.9%). The proposed method improves subtle flood detection and reduces misclassification within vegetated areas, effectively capturing inundation in dense rice paddies. It further removes reliance on labeled data and manual parameter tuning, providing a practical solution for regional-scale flood monitoring in complex heterogeneous landscapes.