Mapping winter irrigation areas and timing in arid regions using time series remote sensing data

Winter irrigation (WI) is a vital practice for mitigating soil salinization and replenishing soil water storage in arid agroecosystems. However, accurate spatiotemporal monitoring of WI remains challenging because of the transient nature of flood events and spectral interference from snow and ice, which limit the applicability of traditional threshold-based methods. To address these issues, in this study, an automated, event-driven detection framework was developed by integrating the LandTrendr temporal segmentation algorithm with dense Sentinel-2 and Landsat time series. Instead of relying on static thresholds, the model explicitly identifies the abrupt spectral rise associated with irrigation onset, thereby decoupling irrigation signals from background noise. Additionally, a dynamic dual-index strategy (MNDWI/NDWI), guided by ERA5-Land meteorological data, was employed to minimize snowfall interference. Validated across major oases in southern Xinjiang from 2020 to 2024, the framework demonstrated robust performance, achieving an overall accuracy of > 95 % for spatial extent and > 72 % for irrigation timing within a 7-day tolerance. The results further indicate that the pixel-based sensitivity of the method effectively characterizes intrafield irrigation variability, revealing the fine-scale dynamics of water distribution. Furthermore, the threshold-free nature of the algorithm enhances its potential for transferability to other dryland regions. This study provides a reliable, high-resolution solution for supporting precision water management and salinity control strategies in water-limited environments. • This study detects short-lived winter irrigation events under snow and ice. • An event-based framework detects winter irrigation from abrupt spectral changes. • We mapped spatiotemporal winter irrigation in southern Xinjiang from 2020 to 2024. • The method achieves high accuracy in both irrigation extent and timing detection.