Digital Soil Mapping Research In Transition Region Of Plain And Hill

For human beings,soil is one of the extremely important natural resources,and its spatial distribution information plays a vital role in the management and utilization of natural resources,modern agricultural production,and soil and water conservation.Traditional soil maps are hand-drawn maps obtained by soil census scientists through field surveys and aerial photo interpretation,and their spatial detail and accuracy are low.In today’s practical applications,traditional soil maps can no longer meet the data needs of various fields.Therefore,we urgently need to use new technologies to carry out digital soil mapping to achieve the purpose of updating traditional soil maps.In this paper,the Soil-Land Inference Model(So LIM)is used to conduct research,which correlates soil attribute information with environmental information,and forecast the space distribution of Soil Classification by integrating environment factors.At present,the soil-landscape reasoning model is mainly established by relying on topographic features,and the remote sensing information is seldom used.Moreover,because the mapping accuracy is better in hilly areas with larger terrain fluctuations,the mapping accuracy in flat areas is limited.Therefore,this research decided to take the mixed area of plat and hills as the research area,use different combinations of environmental factors in different terrain areas,use machine learning methods to mine soilenvironmental knowledge,and finally use the soil-landscape reasoning model to analyze soil types in the research area.Inferential mapping for the purpose of updating traditional soil maps.More specifically,the method of this research contains the following steps(1)Construction and screening of environmental factors.The environmental factors used in this research include information of soil parent materials,factors of topographic and factors of remote sensing.The terrain factor includes the elevation data(DEM)extracted from the contour data,and the slope,aspect,curvature and terrain moisture index(TWI)calculated from the DEM.The remote sensing factors include the normalized vegetation index(NDVI)extracted from remote sensing images,the first principal component(FPC)of remote sensing images and eight texture features.In this research,Clementine software was used to calculate factor importance,and environmental factors with low importance were excluded.(2)Acquisition of inference rules.This research uses the C5.0 decision tree algorithm of Clementine software for modeling.After modeling,the relationship between the environment and soil can be cognized,which will be output in the form of text to obtain the soil classification rule set,and then the rules can be filtered and sorted to obtain Final decision tree rules for predicting soil type.(3)So LIM inference mapping.The rule set obtained in the previous process is input into the So LIM software,and the So LIM software is used to calculate the similarity between all pixels and a certain soil type,and express it in the form of a fuzzy membership degree map,taking the membership degree of each pixel.The highest soil type is used as the inferred soil type for this pixel point to obtain the soil type map of the research area.(4)Accuracy verification.In this research,the quantitative accuracy of the inference mapping results was verified through the verification set of 141 field sampling points,and the confusion matrix between the inference soil map and the field sampling points was obtained,the mapping accuracy was calculated,and the mapping effect was analyzed and evaluated.The finding of research revealed accuracy level of overall research area is 53%,while the accuracy level of soil map conformed by inference mapping by terrain area is 71%,the accuracy level has been enhanced by 18%.The research shows that this method can effectively improve Inferring the accuracy of soil mapping provides an effective method for updating traditional soil maps.This shows that this method is feasible for inferential mapping in flat areas,and at the same time expands the research methods and theories on digital soil mapping,especially in the current research bottleneck in this field,that is,the accuracy of inference mapping in flat areas is improved,which is beneficial for the future.Soil census work provides a way forward.