Effects Of Different Agricultural Land Use Types On Spatial Distribution Of Soil Organic Carbon And Total Nitrogen In A Typical Small Watershed Of Poyang Lake Plain,China

Soil organic carbon and total nitrogen?SOC and STN,respectively?are closely related to soil productivity in agro-ecosystems,and their spatial distribution is mainly influenced by natural characteristics and human activities.Linear regression models assume that the relationship between soil properties and environmental covariates?including natural characteristics and human activities?is linear.This is problematic in pedometric studies because the relationship is mostly nonlinear.Although improved accuracy of predicting soil properties using natural characteristics has been increasingly recognized in hilly or mountainous areas,few studies consider the effects of human activities,especially in plain areas.Herein,we proposed a radial basis function neural network combined with agricultural land use?RBFNN_ALU?model for mapping soil organic carbon?SOC?and total nitrogen?STN?in a typical small watershed of Poyang Lake Plain?China?.We collected 248 soil samples of different agricultural land use types from a 1×1 km² grid,including 93 forest,98 paddy field,and 57 orchard sites.A radial basis function neural network model combined with agricultural land use?RBFNN_ALU?is used to predict the spatial distributions of SOC and STN,and the prediction results of RBFNN_ALU are related to radial basis neural network model?RBFNN?and geographically weighted regression model?GWR?,multiple linear regression model combined with agricultural land use?MLR_ALU?,multiple linear regression model?MLR?,Ordinary Kriging model combined with agricultural land use?OK_ALU?and Ordinary Kriging?OK?for comparison.The main conclusions of this paper are as follows:?1?Descriptive statistics showed that the SOC content ranged from 2.60 to 38.60 g kg^-1,the average content was 19.85 g kg^-1,and the coefficient of variation was 0.45;the STN content ranged from 0.08 to 3.57 g kg^-1,the average content was 1.78 g kg^-1,and the coefficient of variation was 0.47.The SOC and STN levels corresponding to different agricultural land use types were significantly different,i.e.,paddy field>forest>orchard?p<0.05?.?2?Regression analysis showed that the impact levels of climate on spatial variations of SOC and STN were 2.7%and 2.0%,respectively;the organism factors have no significant effect on the spatial variation of SOC and STN;the impact levels of parent material on spatial variations of SOC and STN were 1.3%and 4.2%,respectively;soil type had no significant effect on spatial variations of SOC and STN;the impact levels of terrain on spatial variations of SOC and STN were 5.8%and 18.5%,respectively.Agricultural land use had the largest influence among all potential factors that could independently explain the spatial variations of SOC and STN,with levels of influence of 38.2%and50.9%,respectively.?3?Several models were used to predict the spatial distribution of SOC and STN.RBFNN_ALU generally exhibited the best performance in terms of prediction accuracy,whereas MLR using natural characteristics achieved the worst accuracy.In addition,the RBFNN_ALU model provided a more detailed and accurate description of the spatial SOC and STN patterns,whereas MLR provided limited information and a more biased description.The results indicate that considering the nonlinear relationship between soil properties and environmental covariates,and the impact of human activities on soil properties are critical for predicting the spatial distribution of soil properties in plain areas.