The applicability of categorical triple collocation in spatiotemporal two-dimensional merging of soil freeze-thaw products

Abstract

Soil freeze/thaw (FT) cycling is highly sensitive to soil temperature variations. Existing soil FT products often exhibit pronounced temporal and spatial inconsistencies, with considerable classification accuracy variations among different products, leading to insufficient spatiotemporal stability. To improve the spatiotemporal stability of classification accuracy of soil FT products, this study introduces a spatiotemporal two-dimensional (2D) merging approach with Categorical Triple Collocation (CTC) -derived dynamic weights, evaluates the applicability of CTC in this merging approach, and analyzes the factors influencing the merging accuracy. We use three soil FT products, including two passive microwave remote sensing products (SMAP_FT and ESDR_FT), and one reanalysis product (ERA5_FT) as examples, generating one-dimensional (1D) merged products (Tem-merge and Spa-merge) and 2D merged products over China. Validation with six dense in-situ observation networks shows that all merged products outperform at least two of the original products in total accuracy, with the 2D merged product accuracy exhibiting higher stability than the original products and 1D merged products at most stations. Further analysis shows that the dynamic weight adjustment strategy effectively reduces the influence of low-accuracy products while enhancing the contribution of high-accuracy products during the 2D merging process; moreover, the classification accuracy of the merged products is sensitive to the disparity in the original products' ability to correctly identify frozen states versus thawed states. This research provides a robust framework for optimizing the merging of multisource soil FT products and other related categorical variable products, especially on a global scale where ground observation data remains sparse.