Predictive Soil Mapping In A Hilly Area Using Terrain Attributes

The spatial distribution of soil property is indispensable for agricultural production and environment modeling. However, the detailed soil information is mainly derived from soil maps produced by conventional soil surveys at present. There are two obvious shortcomings in conventional soil surveys. One is the discrete pattern of the distribution of the soil properties which is contrary to the actual situation and the other is the spatial manifestation of this attribute resolution is incompatible with environmental data derived from digital elevation model (DEM) or from remote sensing imagery, which limits the application area of conventional soil surveys. Predictive soil mapping (PSM) may overcome the defects mentioned above. Predictive soil mapping, emerged with the development of digital technology and the statistical model between environmental variables and soil properties, is a direct and quantitative method that using digital environmental variables to predict soil properties.Together with parent material, climate, biota, and time, topography is one of the five fundamental elements of the soil forming factor theory. Topography is a major factor controlling soil properties at the local scale. Topographic attributes influence the redistribution of soil surface materials and the amount of radiant energy, which in turn result in soil differentiation. Digital elevation models are increasingly available during the past decade. Different topographic attributes could be extracted from digital elevation models using digital terrain analyses (DTA). Therefore, the extracted topographic attributes could be applied to assist in predicting soil properties.This study aims at soil distributed in hilly area, southwestern China, and take Renxian experiment area located in Liangping, Chongqing, for an example. The relationships between terrain attributes and soil properties are discussed at the local scale. Then three different prediction methods (Multiple linear regression (MLR), Ordinary kriging (OK) and Regression kriging (RK)) are applied to predict the spatial distribution of soil properties. The main results are as following:(1) Coefficient of variation (CV%) range from 7.09% to 82.51%. The CV% of P is the largest (82.51%) indicating the contents of P change intensively across the field, while that of pH is the smallest (7.09%) showing the values of pH vary smoothly across the field. According to the rank of CV%, OM, Zn, Cu, N, K, Fe, Mg, Ca, S, Mn, and P show moderate variation, while only pH present weak variation.(2) Soil properties studied in this manuscript show moderate or strong spatial autocorrelation. In which, pH, OM, S, Cu, Mn, Zn, and P show moderate spatial autocorrelation, while Ca, Mg, Fe, N, and K present strong spatial autocorrelation.(3) pH is significantly correlated with all the other soil properties at 0.01 level except P. pH is negatively correlated with OM, N, Mn, Fe, Cu, Zn, and S, the correlation coefficients are -0.52^**, -0.29^**, -0.41^**, -0.58^**, -0.41^**, -0.47^**, and -0.31^**, respectively. While pH is positively correlated with K, Ca, and Mg, the correlation coefficients are 0.29^**, 0.65^**, and 0.55^**, respectively.OM is significantly correlated with all the other soil properties at 0.01 level except P and K. OM is negatively correlated with Ca and Mg, the correlation coefficients are -0.59^** and -0.58^**, respectively. While OM is positively correlated with N, Mn, Fe, Cu, Zn, and S, the correlation coefficients are 0.31^**, 0.28^**, 0.66^**, 0.57^**, 0.35^**, and 0.39^**, respectively.In conclusion, pH and OM are significantly correlated with most soil properties which indicate that pH and OM have an important influence on the other soil properties.(4) Soil properties are significantly correlated with terrain attributes. For elevation, pH, Ca, and Mg are positively correlated with elevation. The correlation coefficients are 0.38, 0.70, and 0.70, respectively. While Mn, Zn, S, OM, Cu, and Fe are negatively correlated with elevation. The correlation coefficients are -0.23^*, -0.29^*, -0.43^**, -0.50^**, -0.61^** and -0.62^**, respectively.For slope, pH, Ca, and Mg are positively correlated with slope. The correlation coefficients are 0.20^*, 0.28^**, and 0.43^**, respectively. While S, Cu, Fe, and OM are negatively correlated with slope. The correlation coefficients are -0.34^**, -0.35^**, -0.36^** and -0.47^**, respectively.For plan curvature, only Cu is negatively correlated with plan curvature. The correlation coefficient is -0.21^*. For profile curvature, only Mg is negatively correlated with profile curvature. The correlation coefficient is -0.28^**.For aspect, S, Mn, Fe, Cu, and Ca are significantly correlated with aspect. The correlation coefficients are 0.19^*, 0.21^*, 0.24^*, 0.25^*, and 0.25, respectively. In conclusion, soil properties studied in this manuscript are significantly correlated with terrain attributes derived from DEM, which demonstrate terrain attributes have an important influence on soil properties as well as a good potential in predicting soil properties.(5) The incorporation of terrain attributes into a regression kriging model is a suitable method for increasing the accuracy of prediction of these soil properties. However, there is a constraint that RK could not be used when correlation coefficients between soil properties and secondary information are less than 0.40. Otherwise, the benefit of RK can become marginal.