Soil Mapping Based On Remote Sensing Images And C5.0 Data Mining Algorithm

With the rapid development of digital informatization,traditional soil information acquisition methods have not been able to fully meet the new needs of ecological models,land use analysis,water and soil conservation,natural resources management and other related fields.Therefore,how to combine new technologies to improve soil census efficiency and obtain high-precision soil maps has become the focus of research at this stage.Zhu et al.proposed the Soil-Land Inference Model(SoLIM),which maps relatively difficult-to-acquire soil-attribute information to some readily available environmental information,and combines soil-related properties with the combination of landscape factors.The spatial distribution is depicted.At present,the soil-landscape model is mainly based on topographic factors.There are few remote sensing images in this field and remote sensing images have high accuracy and fast mapping.The terrain information has not been fully represented for vegetation information,surface roughness and spectral characteristics.In addition,the validity of soil-landscape knowledge is the key to predictive mapping.It is generally formed by empirical knowledge of soil science experts or is derived from other soil-related information that has been acquired using effective data mining algorithms.As many regions Lack of experienced experts,the spatial data mining based on remote sensing images is a breakthrough point in current research.Under this circumstance,this study took Rushui River Basin of Huajiahe Town,Hong'an City,Hubei Province as an example,comprehensively used remote sensing images and terrain data,combined with effective data mining algorithms to conduct soil prediction mapping in the study area,and studied specific points.For the following sections:(1)Acquisition of environmental factors.Environmental factors include terrain factors and remote sensing factors.The DEM was established based on the contour data to further extract terrain factors such as slope,aspect,curvature,and topographic moisture index.Remote sensing factors were obtained by processing and analyzing high-resolution remote sensing images.This study used the first principal component,normalized vegetation index,and texture features(including eight basic features)as remote sensing factors.(2)Screening of environmental factors.This paper analyzes the importance of environmental factors through Clementine software,removes the factors with lower importance values,and finally retains 10 environmental factors as the final set of environmental factors for decision tree data mining.(3)C5.0 decision tree data mining.The C5.0 decision tree model was built by the Clementine platform.The single model and the Boost model were used for modeling.Compared with the single model,the Boost model was more accurate and more robust.Therefore,the results of the Boost model were used for subsequent forecasting and mapping.(4)Prediction modelling of soil-landscape reasoning model.In this paper,a rule-based reasoning method is adopted.This method used the membership degree curve of landscape factors to generate the fuzzy membership degree map of soil type,and then integrates all the fuzzy membership graphs to generate the final soil prediction map.(5)Accuracy verification.The field verification points were used to verify the accuracy of the original soil map and the inferential soil map,and a confusion matrix was established.The production accuracy,user accuracy,overall accuracy,and Kappa coefficient were used as indicators for evaluation.The results of the study showed that the overall accuracy of the predicted soil map was 88% which higher than that of the existing soil map,and the accuracy of the soil map prediction under the three different sampling methods was 89%,88%,and 86%,respectively.There is a map of the soil.It shows that the predicted soil map can reflect the true spatial differences of the soil types better than the original soil maps,and that the soil maps can also capture changes in soil type and landforms while expressing the overall spatial distribution of soil types.In addition,the detailed level of spatial distribution of soil types has also been greatly improved,and it can meet the requirements of high-precision digital soil mapping,providing a direction for the future soil survey.