Abstract:

To address the issues of extensive missing values and limited applicability of ESA−CCI soil moisture data in China, a three−dimensional spatio-temporal gap−filling method was developed by incorporating the K−Nearest Neighbor (KNN) machine learning algorithm, with full consideration given to spatio−temporal data correlations and structural characteristics. A comparative analysis of regional applicability was conducted utilizing ESA−CCI soil moisture products and ground observation data collected from May to October during 2016−2018 across China. Three main findings were obtained: (1) significant improvement in spatio−temporal continuity was achieved in the gap−filled ESA−CCI soil moisture data, with the original spatio−temporal structural characteristics being well preserved. The processed data demonstrated enhanced capacity in representing China's soil moisture variation patterns, though spatial structure and heterogeneity characterization outperformed temporal fluctuation representation. (2) Statistical distribution consistency was observed between the gap−filled ESA−CCI data and ground measurements. The filled data exhibited satisfactory accuracy and consistency nationwide, with average performance metrics recorded as follows: root mean square error (RMSE) of 0.068 m³·m⁻³, bias of 0.008 m³·m⁻³, correlation coefficient (r) of 0.618, and structure similarity index measure (SSIM) of 0.999. (3) Regional comprehensive applicability was evaluated through combined analysis of data missing rates and spatio−temporal metrics. Optimal performance was identified in the Huang−Huai−Hai plain, Loess Plateau and Northeast plain regions. Intermediate applicability was observed in the Yunnan−Guizhou plateau, middle−lower Yangtze river basin and Northern Arid/Semi−arid regions. Although improved spatio−temporal coverage continuity was noted in the Sichuan Basin and South China regions, relatively poorer evaluation metrics were obtained. The most severe data limitations and poorest comprehensive applicability were found in the Qinghai−Tibet Plateau region.

Key words:

ESA-CCI soil moisture, K-Nearest neighbor algorithm, Spatio-temporal data gap-filling, Regional applicability assessment