Soil Mapping Method Research In Mixing Region Of Plain And Hill

Soil is the most basic resource condition in human living environment.Accurately understanding the spatial distribution of soil resources can make it more reasonable,efficient and sustainable to use the soil resources.Soil census is the basic method to obtain the quantity,quality and spatial distribution of soil.In the wake of the development of science and technology,the traditional soil survey technology has been unable to meet the data needs of various fields because of its shortcomings.Digital soil mapping technology achieves the purpose of reasoning and prediction by establishing the quantitative relationship between soil attribute information and environmental attributes,it shows great potential in updating soil data.The mapping accuracy by terrain factors and parent material information is better in the area of great topographic relief,but the reasoning ability is limited in the low relief area.Therefore,how to improve the reasoning accuracy in the low relief area is a difficult problem.Nowadays,in the wake of the development of remote sensing technology,digital soil mapping is no longer only relying on the parent material information and terrain factors to infer soil attributes.Remote sensing image data contains ground vegetation information,image texture features which can not be reflected by terrain factors,and can improve the accuracy of mapping to a certain extent.A small area of Chengmagang Town,Macheng City,Huanggang City,Hubei Province was selected as the research area,data mining algorithm was used to obtain the soil environment knowledge,and the soil landscape model was used to classify the soil types in the study area and obtain the soil type inference map.It conclude the following processes:(1)Acquisition and screening of soil environmental factors.The environmental factors in this study included parent material information,terrain factors and remote sensing factors.The parent material information and common terrain factors were extracted from the original soil data and contour data,and the normalized difference vegetation index(NDVI),the first principal component and the texture features were obtained from the GF-2 remote sensing image.Clementine software was used to analyze the importance of all environmental factors.After deleting the unimportant factors,the remaining environmental factors participated in the modeling.(2)C5.0 decision tree modeling and reasoning rules acquisition.In this study,the C5.0decision tree model under Clementine software will be used for machine learning of sample data.The environmental factors screened by importance analysis process were used as input variables,and the soil type was used as output variable to generate decision tree rules for prediction and classification.(3)Fuzzy membership degree calculation and inference mapping.The rule-based function of SoLIM was used to calculate the soil similarity of the corresponding combination of environmental factors in each pixel position of the study area,and the fuzzy membership graph of each soil type was obtained.Finally the fuzzy membership map of each soil type was integrated to get the soil type distribution map after reasoning.(4)Accuracy verification of the mapping result.The 144 samples were used to assess the accuracy of the inference soil map,and a series of accuracy evaluation indexes in the confusion matrix were used to comprehensively evaluate the accuracy and effectiveness of the inferential soil map,and the accuracy of different regions were analyzed after dividing the study area.The result shows that: using the principle of frequency distribution to preprocess the sample data can effectively improve the representativeness of the sample data.And the overall accuracy of the inference soil map reaches 60.41%,the spatial distribution information of the inference map is reliable to a certain extent.The kappa coefficient is0.54,indicates that there is moderate consistency between the inference map and the sample points.Therefore,it is feasible to apply the parent material,terrain and remote sensing factors to the soil classification at the soil specie level,this study can provide a development direction and foundation for the follow-up research.