An advanced error correction methodology for merging in-situ observed and model-based soil moisture.

Highlights • The VIC simulated and in-situ observed SM data were merged. • This was done mainly using CDF matching approach and OI technique. • The CDF was parameterized and then estimated using kriging technique. • A spatial-temporal correlation error function was proposed for OI technique. • The merged SM product was improved compared to the original data. Abstract In order to obtain an improved soil moisture (SM) dataset at large scale, an advanced SM merging methodology based on error correction methods was constructed to merge the model-based and in-situ observed SM data. The SM datasets in a 0–40 cm soil layer were derived from 10 km × 10 km Variable Infiltration Capacity (VIC) model and 797 in-situ stations, respectively. The merging methodology was conducted grid by grid, and mainly included two parts: bias correction and random error correction. Firstly, the bias correction was performed for the VIC simulations by applying the Cumulative Distribution Function (CDF) matching approach combined with the kriging technique. Secondly, the random error of the VIC simulations was corrected using an Optimal Interpolation (OI) technique based on a spatio-temporal correlation function which was proposed and constructed in this study. Through validations against in-situ observations, the merged SM was evaluated, and the merging errors in each step were analyzed and discussed. The results showed that the merged SM product was improved compared to the original SM data, both temporally and spatially. The SM merging methodology is effective and reliable in combining the accurate but sparse in-situ observations and the continuous VIC simulations. In addition, the spatial mismatch impact on the representativeness of in-situ stations was limited, and the merging errors were mainly produced in the CDF estimation process. The random error information in the spatial dimension exhibited a bigger impact on the random error correction comparing to that in the temporal dimension. This study provided strong encouragement for the efficient use of in-situ SM observations and provided valuable methods for combining multi-sources SM datasets. [ABSTRACT FROM AUTHOR]