Abstract:Choosing suitable auxiliary and effective method is the prerequisite to accurately predict the spatial distribution of soil organic matter (SOM) in estuarine wetland. In order to achieve this purpose, a case study was conducted in the Minjiang Estuary Wetland of Fuzhou, southeast China. A total of 23 environmental factors were extracted by ArcGIS geostatistical analyst and remote sensing image analysis technique. Then, stepwise regression model and principal component analysis were used to screen the characteristic variables. At last a hybrid model of the support vector regression Kriging (SVRK) was proposed to analyze the spatial variability of SOM, and compared with BP neural network Kriging (BPNNK) and regression Kriging (RK). The results showed that normalized vegetation index (NDVI), ratio vegetation index (RVI), perpendicular drought index (PDI), flow accumulation (FA) and sediment movement index (SMI) were significantly correlated with SOM, which had the higher coefficient of determination (R2 = 0.446) and the significant probability value (P<0.000 1). Three principal components, explained at least 94% of the total variance, were extracted from these environmental factors by principal component analysis and used as characteristic variables. The spatial variability of SOM was affected by structural factors, showing a trend of "lower in the north and higher in the south". Compared with RK and BPNNK, SVRK had the highest prediction accuracy, and more accurately reflected the spatial variability of SOM, can provide a methodological support for the study of spatial variability of SOM in the same or similar region.