Mapping the Interannual Variability of Irrigated Area Using Categorical Sampling and Machine Learning

Abstract Accurate irrigated area (IA) mapping is essential for hydrological and climate modeling. However, existing IA mapping approaches typically rely on persistently irrigated or non‐irrigated samples, which has reduced sensitivity to year‐to‐year IA variability. Here, we develop a Categorical Triple Collocation (CTC)‐based sampling framework that identifies both continuously and intermittently irrigated pixels, thereby improving the representation of IA temporal dynamics in training samples. Coupled with machine learning, this framework produces annual 500‐m IA maps across China for 2000–2022. Compared with conventional sampling strategies, the proposed approach reduces IA mapping error substantially, with RMSE decreasing from 16.6% to 8.3%. It also captures interannual IA changes driven by large‐scale agricultural policy shifts, which conventional approaches fail to resolve. These results demonstrate the robustness of the CTC‐based sampling framework for IA mapping, which may directly support water management and Earth system modeling in intensively managed agricultural regions.