Objective The aim was to evaluate the applicability of the land assimilation system of China (CLDAS) and the land assimilation system of Inner Mongolia (IMLDAS) in areas with sparse observations, in order to provide a basis for selecting high temporal and spatial resolution soil moisture data for the arid and semi-arid regions of China.

Method This study employed in observations from 111 meteorological stations across Inner Mongolia, measuring 0 - 20 cm soil relative humidity during the 2008 - 2021 (May-September), to validate and compare the CLDAS and IMLDAS. Systematic evaluation of regional accuracy patterns was conducted through multiple statistical metrics, including the index of agreement (IA), correlation coefficients(R), root mean square error (RMSE), Taylor diagram diagnostics, and comparative analyses of temporal variations across three distinct land-cover types: agricultural, pastoral and forest zones.

Result Results demonstrated that IMLDAS achieved substantially higher accuracy than CLDAS in Inner Mongolia. Independent station-based validation revealed that 85% of sites attained daily IA values ≥ 0.5 for IMLDAS, compared to 76% for CLDAS. Furthermore, 97% of stations registered RMSE values were below 30% for IMLDAS, only 68% for CLDAS. The Taylor diagram showed that both the standard deviation and centered root mean square error of IMLDAS were lower than those of CLDAS. Regional assessments highlighted that IMLDAS performance were better in agricultural and forest ecosystems. Specifically, monthly mean deviations in agricultural areas were merely 1% for IMLDAS, whereas CLDAS exhibited systematic positive biases ranging from 11% to 15%. On annual timescales, CLDAS overestimated forest soil moisture by 10% - 39%, while IMLDAS demonstrated remarkable synchronization with observations in both amplitude and phase characteristics. Both products showed overestimation tendencies in pastoral regions. However, IMLDAS monthly and annual mean biases were substantially reduced, representing only 31.1% and 30.1% of the corresponding CLDAS biases, respectively. Notably, both assimilation systems maintained moderate-to-strong correlations with ground observations at over 95% of stations (significant at P < 0.01), indicating that contemporary domestic land surface assimilation systems had achieved operational maturity. The inter-product discrepancies identified stem primarily from systematic biases rather than differences in correlation structure.

Conclusion These findings established that IMLDAS possessed distinct advantages for soil moisture simulation in observation-sparse regions of northern China. The results underscored that refining localization parameters represented the critical pathway for further accuracy enhancements in regional land data assimilation systems. This study provided a scientific guidance for operational drought monitoring in data-scarce environments and offered methodological insights for optimizing initial conditions in land-atmosphere coupling systems.