A 10-meter resolution surface water dataset for the Hindu Kush Himalaya during 2016-2022

The Hindu Kush Himalaya (HKH), known as the "Water Tower of Asia" and the source of the ten longest rivers in Asia, provides abundant water resources for the 1.9 billion people in the basin. In recent years, the ecological sustainability of the region has been seriously affected by multiple factors, resulting in changes in water resources. Rapidly obtaining the distribution and range of surface water, monitoring the status of surface water, and timely grasping the dynamic changes of water resources is of great significance for water security, ecological environment protection, and sustainable development in the HKH region. In this study, we produced a 10-meter resolution surface water dataset (HKH-SWD10m) for the HKH region from 2016 to 2022, including an annual-scale surface water product (HKH-SWD10m-Yearly) and a daily-scale surface water product (HKH-SWD10m-Days) extracted from single-scene Sentinel-2 images during the flood season. A surface water extraction deep learning network based on a Vision Transformer architecture is developed by us for quickly and accurately producing the HKH-SWD10m dataset. The Transformer-based surface water extraction network was employed to autonomously extract surface water from Sentinel-2 images with cloud cover less than 20% during the flood season from 2016 to 2022 across the entire HKH region, and thus generated the HKH-SWD10m dataset. Validation conducted with 15,000 randomly selected points across the HKH region demonstrated that the HKH-SWD10m dataset achieved an accuracy with an Overall Accuracy (OA) of 0.9787, a Producer's Accuracy (PA) of 0.9638, an User's Accuracy (UA) of 0.9856, and a Kappa of 0.9476. When compared with the 30-meter resolution Global Surface Water product (GSW), the HKH-SWD10m dataset is generally consistent with the GSW product in the HKH. Additionally, benefiting from the 10-meter resolution remote sensing images, more small water bodies are included in the HKH-SWD10m dataset, and the boundaries of all water bodies in the HKH-SWD10m dataset are more detailed, better describing the morphology of water bodies.