Mapping Soil Properties In Complex Terrain Regions

Soil is the material base for human beings and animals and plants to survive,and the rational use of soil resources is an important content for sustainable development.The traditional data acquisition method of soil through field sampling and laboratory analysis,and scattered historical data to obtain soil,soil property mapping mathematical model is used to infer soil types and soil properties from observation,soil knowledge and related environmental variables,has the advantages of high efficiency,fast and low cost,can make up for the defects of the spatial interpolation of neglect the influence of environmental factors,has been widely studied and applied.Complex terrain area terrain,traffic developed,the field of soil sampling is difficult and risky,soil data update speed,1:100 million soil data in most areas nowadays only the second national soil survey,agriculture,animal husbandry,unable to meet the needs of scientific research and environmental management on soil data.Mapping soil properties provides a new approach for obtaining soil attribute data in complex topographic areas.Located in the southern part of Qinghai Province,Yushu is the hinterland of the Qinghai Tibet Plateau.It has many mountains and rivers in its territory.It is a typical complex terrain area.In order to explore the mapping method of soil properties in complex terrain area,this paper uses ArcGIS software and IBM SPSS Modeler data mining tool to predict soil total carbon in Yushu city.The carbon content of the soil as target variable,DEM,slope,slope,Topographic Wetness Index,profile curvature,plan curvature,vegetation index,weighted by variable construction and soil type,land use types,landform types,geological type,vegetation type data with the digital elevation model and normalized difference vegetation index is common predictive variables,prediction model based on multivariate regression model,regression model and BPNN model to construct the carbon content of the soil.The main conclusions are as follows:(1)The total carbon content of soil in Yushu was in line with the normal distribution,with an average value of 55.80 g/kg and a standard deviation of 19.22 g/kg.The results of the multivariate regression model showed that the content of total carbon was higher in the lower valley area and higher in the mountain area with higher elevation.(2)The regression model built by the weighted variable constructed in this paper is better than the regression model established by the digital elevation model,the slope and the topographic humidity index.BPNN model of weighted variables to establish the verification accuracy is higher than that of other BPNN model,the overall importance of variable weighted BPNN model is constructed by both conventional variables and weighted variables in higher than that of the conventional variable importance and by the importance of geological and soil type,vegetation type and digital elevation model is much higher than the weighted variables digital elevation model itself.Among the many models,the variable multivariate regression model with weighted variables is the best model to predict the total soil carbon in Yushu.Therefore,the construction of weighted variables is a new way to make use of qualitative environment variables effectively.It provides a new method for digital soil mapping to obtain new variables.(3)Qualitative variables do not have significant influence on target variables,and the weighted variables constructed are correlated with the quantitative variables involved in constructing weighted variables,so they are not suitable for the construction of weighted variables.(4)The use of ArcGIS software and SPSS Modeler combined with digital soil attribute mapping,can build a model suitable for the study area according to the methods and experiences of experts,not limited to other modeling software of data format,and methodological limitations.To sum up,this study combined with remote sensing,vegetation,topography and geological and geomorphological data and other multi-source data,modeling and prediction of soil carbon,to achieve the desired effect.The regression model established by weighted variable is superior to the regression model established by conventional variables.The prediction results can provide decision support for the management of agriculture and animal husbandry and urban and rural planning.To verify the accuracy of BPNN model to establish the weighted variables,thus constructing weighted variables can effectively use the qualitative variables,provide new mapping variables for digital soil mapping,but also for other soil lack of reference in the study area,to better service for the regional development and economic construction;provide methods and reference for complex the terrain soil data updating and complex terrain soil attribute mapping.