Effectiveness of multi-source merged soil moisture products using different triple collocation methods: A comprehensive assessment incorporating spatial similarity evaluation framework

Evaluation of soil moisture products and data merging efforts have commonly focused on their performance in representing temporal dynamics, while overlooking the capacity of the products to capture spatial patterns. In this study, a spatial evaluation framework based on the Structural Similarity Index (SSIM) was for the first time introduced to assess the behavior of soil moisture products. Meanwhile, a modified temporal-spatial triple collocation (TC) merging method that accounts for local spatial errors (referred to as TC_2D_local) was proposed for merging multi-source soil moisture. Three widely used soil moisture products, i.e. Soil Moisture Active Passive (SMAP), Advanced SCATterometer (ASCAT), and ERA5-Land, as well as their merged products based on three TC approaches, including the traditional TC (TC_temporal), a global two-dimensional TC (TC_2D), and the new TC_2D_local, were evaluated and compared from both temporal and spatial perspectives, using conventional metrics (R, RMSE, Bias, ubRMSE), together with SSIM and its component indices, i.e. Similarity in Mean (SIM), Similarity in Variance (SIV) and Similarity in Pattern (SIP). Soil moisture from International Soil Moisture Network (ISMN), agricultural stations and China Meteorological Administration Land Data Assimilation System (CLDAS) were used for validation. The results revealed the distinct behavior of the three parent products in capturing temporal dynamics and spatial patterns of soil moisture over China. SMAP achieved the lowest RMSE and highest SSIM in western China, while ERA5-Land was superior in southeastern China in both temporal and spatial evaluation, with larger values of R and SSIM. Although ASCAT exhibited a less satisfactory temporal performance with lower R and higher RMSE, it better characterized the spatial variability of soil moisture with higher SIV across China. All of the TC merged products demonstrated improvements with a reduced RMSE and Bias, while the two 2D TC approaches that incorporated spatial error information outperformed TC_temporal. However, TC_2D performed better in the eastern part of China, featured by relatively uniform underlying conditions, whereas TC_2D_local yielded lower RMSE and Bias in western China, which has more sophisticated environmental conditions. SSIM-based spatial evaluation further revealed a consistent superiority of TC_2D_local over TC_temporal and TC_2D, achieving SIM > 0.95, SIV > 0.9, SIP >0.95 and SSIM >0.8 across most of the study area, with the most pronounced improvements in western China. This study demonstrated the effectiveness of two-dimensional TC for merging multi-source soil moisture products, particularly the TC_2D_local for environmentally heterogeneous regions. It was highlighted that the spatial similarity evaluation framework could provide a new perspective and additional insights into the performance of soil moisture products, potentially benefiting their applications such as hydrological modelling and agricultural drought monitoring, which requires accurate representation of spatial soil moisture patterns.