Abstract:
Abstract: Soil depth is one of the most important input parameters for hydroecological models in arid and semiarid regions. However, soil depth is highly variable spatially and traditional measures of soil depth are laborious, time consuming and even difficult to practically perform, especially in the complex landscape areas. In these areas, the mapping based on the relationships between soil properties and environmental factors may be useful. However, the approach used to establish their relationships is limited. Therefore, this study proposed an efficient method for obtaining and establishing the soil-environment relationships in complex landscape environments. The method was based on an fuzzy clustering method (fuzzy C-means, FCM) and decision tree (DT). Using this method, the relationships between soil depth distribution and environmental factors in a typical alpine watershed in the Qilian Mountains, northwestern China with easy-to-obtain environmental covariates data was established. The method was based on the assumption that soil was the production of the interaction among its formative environmental factors with time. The environment variables, such as altitude, slope, aspect, plan curvature, profile curvature, topographic wetness index and normalized differential vegetation index, were extracted as auxiliary variables for data analysis. A total of 3626 points obtained by FCM and DT methods was as training sample set, and 31 points collecting from field survey through representative sampling strategy was used as validation sample set. The method consisted of 4 steps: 1) to define the environmental factors playing dominant roles in formation and development of soil depth, then to obtain the environmental niches by running FCM analysis (after correlation analyses altitude, profile curvature and terrain wetness index were selected to carry out FCM analysis); 2) to assign the ranked distribution of soil depth based on the field investigation data and pedogenesis principles; 3) to select the typical areas of fuzzy membership threshold greater than 0.5, and to randomly choose a certain number of points which were proportional to area extent, and to possess an approximate quantity of points, then to extract the locating information of environmental factors so as to build up the training sample set; 4) to obtain the critical thresholds of soil environmental factors and the knowledge about soil-environment relationships by running training sample set through the DT arithmetic. The method was applied in a typical alpine watershed of the Qilian Mountain, the Heihe River basin, and the soil depth distribution map was created. In addition, an independently field sample set was used to validate the effectiveness of the method in establishing the relationships between soil depth and environmental factors. Its overall accuracy and Kappa coefficient reached 74.2% and 0.659 respectively. Therefore, the proposed method is an optional efficient solution for predictive soil depth mapping in the complex landscape environment.