Spatial regression analysis on the variation of soil salinity in the Yellow River Delta

In this paper, spatial autocorrelation analysis, ordinary least square (OLS) and spatial regression models were applied to explore spatial variation of soil salinity based on samples collected from the Yellow River Delta. Generally, spatial data, like soil salinity, elevation height etc., are characterized by spatial effects such as spatial dependence and spatial structure. Inasmuch as these effects exist, the utilization of OLS model may lead to inaccurate inference about predictor variable. Moreover, the traditional regression models used to analyze spatial data often have autocorrelated residuals which violate the assumption of Guess-Markov Theorem. This indicates that conventional regression models cannot be used in analyzing variability of soil salinity directly. To overcome this limitation, spatial regression model was introduced to explore the relationship between soil salinity and environmental factors (including elevation height, pH value and organic matter concentration). By verifying Moran's I scatterplot of residuals, we found no autocorrelation in spatial regression model compared with high significant (p < 0.001) positive autocorrelation in the OLS model; besides, the spatial regression model had a significant (p < 0.01) estimations and good-fit-it in our study. Finally, an approach of specifying optimal spatial weight matrix was also put forward.