MATMS-TP：青藏高原逐日无云MODIS NDSI数据集（2003-2022）

基于MODIS NDSI产品和耦合了mask-aware的Transformer模型，生成了青藏高原逐日无云MODIS NDSI积雪数据集（MATMS-TP，2003-2022）。该数据集不仅利用NDSI的时空背景信息，还结合气象条件、地形特征、地理位置和季节变化等相关辅助信息，填补了因云层造成的数据空缺。特别地，该数据集还集成了mask-aware技术，将云层作为一个独立的输入通道，并结合门控卷积和约束后的loss函数，有效提高了大面积、长时间云层遮挡情况下的填补精度。与传统的时空插值方法相比，该模型的RMSE降低了30%以上；与主流的深度学习模型相比，RMSE降低了9%以上。该时空连续的NDSI数据集对于详细估计积雪覆盖面积（SCA）、积雪覆盖比率（FSC）和雪深（SD）等具有重要意义。A daily cloud-free MODIS NDSI snow cover dataset (MATMS-TP, 2003–2022) over the Qinghai-Tibet Plateau was developed using MODIS NDSI products and a Transformer model integrated with mask-aware technology. This dataset leverages not only the spatio-temporal background information of NDSI, but also incorporates auxiliary information such as meteorological conditions, topographic features, geographic locations and seasonal variations to fill the data gaps caused by cloud cover. Specifically, this dataset integrates mask-aware technology, treats cloud cover as an independent input channel, and combines gated convolutions and constrained loss functions, effectively improving the filling accuracy under large-area and long-duration cloud cover conditions. Compared with traditional spatio-temporal interpolation methods, the RMSE of the proposed model is reduced by more than 30%; compared with mainstream deep learning models, the RMSE is reduced by more than 9%. This spatio-temporally continuous NDSI dataset is of great significance for the detailed estimation of snow cover area (SCA), fraction of snow cover (FSC) and snow depth (SD).

搜集汇总

背景与挑战

背景概述

MATMS-TP是青藏高原2003-2022年的逐日无云MODIS NDSI数据集，采用mask-aware Transformer模型填补云层空缺，显著提高了数据精度。数据集以GeoTIFF格式存储，适用于积雪覆盖面积、比率和雪深等研究。