基于多源遥感数据与机器学习的青藏高原雪深数据集（1987-2019）

山区的积雪深度或雪水当量对水文、水资源管理、气象和气候研究至关重要。遥感技术可以用于区域或全球尺度的积雪深度监测。然而，由于传感器的灵敏度和空间分辨率问题，基于卫星的山区积雪深度遥感存在挑战。近年来，时序的Sentinel-1卫星数据被用于山区积雪深度反演，显示较高的精度。本数据集结合光学和被动微波遥感观测估算山区积雪深度。使用2016-2021年的光学和被动微波遥感观测数据以及Sentinel-1反演的积雪深度来训练基于XGBoost机器学习算法的积雪深度反演模型。通过交叉验证方法和独立的地面积雪深度数据进行验证。结果表明，预测的积雪深度与实地测量积雪深度的相关系数为0.61，MAE为0.33米，显示出相较于AMSR-E/AMSR2积雪深度产品更高的精度。

Snow depth or snow water equivalent (SWE) in mountainous areas is critical for hydrology, water resource management, meteorology, and climate research. Remote sensing technologies can be applied to snow depth monitoring at regional or global scales. However, satellite-based remote sensing of mountain snow depth faces challenges due to issues with sensor sensitivity and spatial resolution. In recent years, time-series Sentinel-1 satellite data has been used for mountain snow depth inversion, demonstrating relatively high accuracy. This dataset combines optical and passive microwave remote sensing observations to estimate mountain snow depth. We used optical and passive microwave remote sensing observation data from 2016 to 2021, along with snow depth values inverted from Sentinel-1 data, to train a snow depth inversion model based on the XGBoost machine learning algorithm. Model validation was conducted using cross-validation methods and independent in-situ snow depth measurements. The results show that the predicted snow depth has a correlation coefficient of 0.61 with the in-situ measured snow depth, with a mean absolute error (MAE) of 0.33 meters, demonstrating higher accuracy than the AMSR-E/AMSR2 snow depth products.