A novel soil moisture evaluation framework incorporating brightness temperature and a high-resolution 1 km summer brightness temperature dataset

Abstract

Accurate estimation of soil moisture (SM) is essential for various hydrological, meteorological, agricultural, and ecological applications. However, evaluating SM on a global scale remains challenging due to the limited availability of in-situ observations and the spatial heterogeneity of SM. Coarser resolution SM products, although beneficial for broader area coverage, often struggle to capture fine-scale variations influenced by local hydrological processes, land use, vegetation cover, and microclimates. To address these challenges, this study presents two contributions: a new 1 km brightness temperature (TB) dataset for the summer season and a two-step SM evaluation method. The 1 km TB dataset, developed by integrating SMAP’s 9 km SM product with radiative transfer modeling (RTM) and Mironov model, provides enhanced spatial resolution and is focused on areas where vegetation water content (VWC) is below 3 kg/m2, allowing for a more detailed analysis of SM variations. When validated against SMAP TB data, this dataset showed a solid correlation (R2 = 0.921) and a low root mean square error (RMSE = 4.254 K), making it a useful resource for fine-scale SM monitoring. The two-step evaluation method, which combines physical modeling (RTM and additional models) with machine learning techniques such as non-linear regression and convolutional neural networks (CNN), offers improvements in both temporal and spatial coverage. By transitioning from point-based validation using ISMN to an area-based approach, this method produces SM estimates at the same scale as the evaluated data, addressing the limitations of previous point-scale validations. Comparisons with ISMN data demonstrated the method’s robustness, with key metrics showing improved performance (R2 = 0.749, RMSE = 0.0561 m3/m3) across diverse environmental conditions. Furthermore, the evaluation of the ERA5 dataset using this method revealed a general overestimation of SM, particularly in tropical regions with dense vegetation. These findings are consistent with known ERA5 biases, reinforcing the reliability of the two-step method for global-scale SM evaluations. The results suggest that this approach, which integrates both physical models and machine learning, offers a more comprehensive and reliable framework for SM product evaluation, while the 1 km TB dataset provides valuable support for applications requiring finer spatial resolution.

Topics & Concepts