基于TMI、AMSR-E、AMSR2和GMI的机器学习方法增强时空覆盖的全球日尺度长时序土壤水分数据集（1997-2023）

土壤水分是区域和全球水循环的关键组成部分，在各种生态、水文和大气过程中起着重要作用。精确的全球土壤湿度监测对于加深对这些过程的理解，并为天气和气候预测、农业干旱监测、水资源管理以及全球气候变化研究提供信息至关重要。本研究整合了多个搭载被动微波传感器的卫星数据，以克服传统土壤水分监测中的时间和空间一致性不足的问题。通过结合经过相互校准的倾斜轨道卫星（TMI和GMI）和极地轨道卫星（AMSR-E和AMSR2）的多频亮温数据，将采用多通道协作检索算法（MCCA）精确提取的SMAP的土壤水分数据作为目标，选择LSTM机器学习模型成功生成了四个全球土壤湿度产品：TMI、AMSR-E、AMSR2和GMI。经检验，这四个数据在全球具有广泛的一致性。因此，将这些数据集合并，创建了一个从1997年至2023年、25 KM分辨率的日尺度土壤水分综合产品（MCCA-ML）。MCCA-ML土壤水分数据集的土壤水分反演结果继承了MCCA SMAP数据的优势，包含了融化期的土壤水分含量与冻结期的液态水含量。通过与全球24个地面密集观测网络和其他数据集（MCCA SMAP和ESA CCI）的验证，MCCA-ML在空间分布和季节变化模式上高度吻合全球气候和地理特征。本数据集在24个密集观测网络上整体表现优于传统数据产品。在多卫星协同服务期内，MCCA-ML的全球陆地覆盖率超过80%，明显优于现有产品，有效提升了全球日尺度土壤水分监测能力。

Soil moisture is a critical component of regional and global water cycles, playing a vital role in various ecological, hydrological, and atmospheric processes. Accurate global soil moisture monitoring is essential for deepening the understanding of these processes and informing weather and climate forecasting, agricultural drought monitoring, water resource management, and global climate change research. This study integrated multiple satellite datasets equipped with passive microwave sensors to address the insufficient temporal and spatial consistency issues in traditional soil moisture monitoring. By combining multi-frequency brightness temperature data from cross-calibrated inclined-orbit satellites (TMI and GMI) and polar-orbiting satellites (AMSR-E and AMSR2), and taking the SMAP soil moisture data accurately extracted via the Multi-Channel Collaborative Retrieval Algorithm (MCCA) as the target, the LSTM machine learning model was successfully employed to generate four global soil moisture products: TMI, AMSR-E, AMSR2, and GMI. Validation showed that these four datasets exhibit strong global consistency. Accordingly, these datasets were merged to develop a daily-scale integrated global soil moisture product (MCCA-ML) with a spatial resolution of 25 km spanning from 1997 to 2023. The soil moisture inversion results of the MCCA-ML dataset inherit the advantages of the MCCA SMAP data, encompassing soil moisture content during thawing periods and liquid water content during freezing periods. Validated against 24 global dense ground observation networks and other datasets (MCCA SMAP and ESA CCI), the MCCA-ML product highly aligns with global climatic and geographic characteristics in terms of spatial distribution and seasonal variation patterns. Overall, this dataset outperforms traditional data products across the 24 dense ground observation networks. During the multi-satellite collaborative service period, the MCCA-ML product achieves a global land coverage exceeding 80%, which is significantly superior to existing products and effectively enhances the capability of global daily-scale soil moisture monitoring.

搜集汇总

背景与挑战

背景概述

该数据集是一个全球日尺度长时序土壤水分数据集，覆盖1997年至2023年，基于TMI、AMSR-E、AMSR2和GMI卫星数据，采用LSTM机器学习方法生成，整合为MCCA-ML综合产品。数据集具有25公里空间分辨率，经验证在全球24个地面观测网络上表现优于传统产品，时空覆盖超过80%，显著提升了全球土壤水分监测能力。