Mapping rodent-induced disturbance of biological soil crusts using PlanetScope imagery and deep-learning-informed graph cut

Biological soil crusts (biocrusts) function as a “living skin” at the soil–atmosphere interface and provide important ecosystem services in drylands. While climate change and anthropogenic disturbances are well-recognized drivers of biocrust degradation, mechanical disturbance by wildlife remains poorly understood. Burrowing rodents can directly damage biocrusts by excavating and displacing soil, creating bright disturbance patches that contrast sharply with the darker surrounding biocrust matrix. Based on this brightness contrast, we developed a mapping framework using high-resolution PlanetScope basemaps from 2017 to 2025 to delineate rodent-induced biocrust disturbance in the Gurbantunggut Desert, China. The framework follows a two-stage strategy. First, a Swin-UPerNet model generates colony probability maps to capture diverse colony patterns. Second, these maps are refined using a morphology-based scoring scheme to enhance small and weakly expressed patches, and then optimized by graph cut using brightness-gradient information to sharpen boundaries. Validation against 125 independent field-supported reference sites showed that the framework achieved an overall accuracy (OA) of 97.52% (97.10–97.92%), with an F1-score of 0.804 (0.778–0.826) and an intersection over union (IoU) of 0.672 (0.636–0.703) for the disturbance class. Rodent-induced disturbance was spatially extensive, reaching 1218.34 km 2 (1157.87–1284.25 km 2 ) in 2019 and comprising approximately 1.11 million mapped disturbance patches. This mapped footprint was equivalent to 10.3% of the estimated fractional biocrust-cover area for that year, indicating that rodent excavation occupied a substantial area that could otherwise support biocrust cover. The extent of mapped disturbance was highly dynamic over time, declining sharply through 2022, reaching a minimum in 2023, and then rebounding rapidly through 2025. Representative trajectories further suggest that rodent-induced disturbance can expand rapidly and persist for years, and that its dynamics may be coupled with hydroclimatic stress in biocrust-covered landscapes. These findings highlight rodent-induced disturbance as an important yet underrecognized component of desert ecosystem dynamics and provide a remote-sensing framework for monitoring such disturbances under a changing climate.