Spatial Distribution Prediction Of Soil Organic Carbon

Soil is the basic human survival resources. With the lack of soil resources, worsening environmental problems, soil information maps provided by traditional soil mapping based on polygon units, both in scale matching and refelcting variability of soil characteristics, etc. have been unable to meet the need of the precise soil information. How to master the precise soil information is the foundation to achieve precision agriculture and the key prerequisite for environmental simulation.With the high development of GIS technology, and the quantitative expression of a large number of landscape factors, a number of digital soil attribute mapping method have been proposed. These methods are performed through determining the spatial autocorrelation of the soil attribute or its quantitative relationships with the environmental factors. Compared with traditional manual methods of soil mapping, mapping accuracy is improved and mapping cost is reduced. But because they are still mapping methods based on statistics, wit poor generalization of model and requirements of a large number of samples supporting, so that these methods are still in the research stage and not adopt by the government as a general soil mapping techniques for promotion. Therefore, it has certain significance for the digital mapping of soil attributes to begin with the reasonable sampling plan to reduce the sample number, then reveal the autocorrelation of soil attribute or its quantitative relationship with environmental factors, and compare the precision of all kinds of methods to select the most appropriate method.In this thesis, taking the organic carbon content of top soil in Longtan watershed in Shucheng county of Anhui province as the research object, soil samples were collected by using a purposive sampling design method based on typical points. By using Spearman correlation, the correlation between organic carbon content of topsoil and both topographic factors (slope, elevation, aspect, etc.) and vegetation coverage had been analyzed. In ArcGIS, Matlab, GS+ and other software support, six different prediction methods (Multiple linear regression, Regression kriging, Universal kriging, Neural network-kriging, Regression tree, the FCM(Fuzzy C-Means) average weighted were applied to predict the spatial distribution of organic carbon of soil surface. The main results are as following:(1)In this study area, the appropriate parameters of fuzzy c-means clustering are c (the number of categories)=12, m (Weighting Exponent)=1.6.(2)The CV% (Coefficient of Variation) of organic carbon is 30.96%. According to the rank of CV%, it shows a moderate variation. Soil organic carbon studied in this thesis shows a moderate spatial autocorrelation (NSR=47.73%).(3)Soil organic carbon is significantly correlated with terrain factors. It is negatively correlated with elevation and slope (p<0.01), positively correlated with plane curvature, profile curvature and CTI (Compound Terrain Index)(p<0.01). the correlations with aspect, curvature and NDVI (Normalized Different Vegetation Index), are not significant. The surface soil organic carbon content and the spatial locationexists certain correlation, but mainly displays in the horizontal coordinate.(4)The test result of modeling samples shows that the measurements of soil organic carbon are explained in a high degree by all of these six predictive methods, and the hybrid geostatistical methods obviously have more powerful explanation than other methods. The range of semi variance of residuals with drift removed by BP neural network method is shortest (1139.00m). So that BP neural network is the best method to remove drift of soil surface organic carbon content.(5)The test result of test samples shows that the best predictive mapping method of soil surface organic carbon in this thesis is the FCM average weighted method which MAE (5.99) and RMSE (6.69) are the lowest, AC(0.90) and the correlation coefficient (0.58) between predicted and observed values is the highest.(6)The distribution of soil organic carbon is highest in the watershed outlet in the northeast, low in the northern, southwestern and eastern regions with the highest terrain. The change of soil organic carbon in the middle is flat. Soil organic carbon content rises and declines with elevation and slope. The relationship of soil organic carbon content and aspect is not obvious.