Research On Soil Horizon Classification And Identification Based On PXRF Data

Soil classification plays a leading role in the development of soil science.Each horizon in the profile is the basis for the study of soil classification.The soil can be classified by classifying and sampling the in-situ genetic horizon and identifying the physical and chemical analysis of the laboratory diagnostic horizon and diagnostic characteristics.However,the traditional genetic horizon classification is a qualitative classification method,which relies on the investigators' subjective experience and knowledge,lacks quantitative classification standards,and is difficult to obtain consistent horizon classification results.Furthermore,laboratory physical and chemical analysis is time-consuming and labor-intensive,and has high requirements for soil samples,which cannot meet the actual needs of identifying soil diagnostic horizons quickly and efficiently.Therefore,based on the rapid-acquired soil attribute data,this paper conducts soil numerical classification research and attempts to classify and identify soil horizons quantitatively.Multi-element simultaneous determination technology based on portable X-ray fluorescence spectroscopy(PXRF),through the in-situ measurement of soil profile data in Wuxue City,Hubei Province,equal-area quadratic spline function and fuzzy C-means clustering(FCM)are used to study the accuracy of soil genetic horizon classification based on the element content difference.Through the measurement of the soil diagnostic horizon data in Jiangxi Province and Wuxue City in Hubei Province,the random forest(RF),decision tree(DT)and support vector machine(SVM)algorithm are used to model and evaluate the accuracy of the different diagnostic horizons,the ability of PXRF data to identify soil diagnostic horizons are studied.The main conclusions obtained are as follows:(1)In-situ genetic horizon classification of the soil profile.Through the cluster analysis of the soil depth function,the quantitative classification of the soil genetic horizon boundaries can be achieved,and with the increase of the cluster categories,more accurate horizon classification results can be obtained.When the number of genetic horizons in the soil profile is used as the clustering category,the horizon classification results based on the difference in element content and the difference in morphological characteristics are in good agreement,and the accuracy of soil horizon sample collection can reach 89.5%.By calculating the digital gradient value of fuzzy clustering,based on the maximum value of the change peak,the obviousness characteristics of the morphological horizon transition can be quantitatively described;the change range of horizon transition can be used as a basis for establishing a buffer zone to evaluate the accuracy of horizon classification.In terms of results,among the 30 horizon transitions,18 are located in the buffer zone,and 7are within 5 cm depth of the buffer zone.Among the soil types,Aquic Cambosols have higher horizon classification accuracy than Stagnic Anthrosols.In the soil profiles,the transition between A and B horizons and the interior of B horizons,which are less affected by humans,have better horizon classification results.(2)Identify single soil diagnostic horizons.Through feature factors selection,using RF,DT and SVM algorithms,the classification model of each diagnostic horizon is constructed respectively.In the identification of the diagnostic surface horizon and the diagnostic subsurface horizon,RF has obtained the highest accuracy and can better identify the diagnostic subsurface horizon,but the identification ability of the diagnostic surface horizon is weak.Within the diagnostic surface horizons,SVM obtained the highest accuracy in the classification of single diagnostic surface horizons.The order of the identification accuracy of the diagnostic surface horizons is Mollic epipedon,Umbric epipedon>Anthrostagnic epipedon>Ochric epipedon.Within the diagnostic subsurface horizons,RF and SVM achieved the same accuracy in the classification of single diagnostic subsurface horizons.The order of the identification accuracy of the diagnostic subsurface horizons is LAC-ferric horizon>Hydragric horizon>Argic horizon>Cambic horizon.(3)Identify multiple soil diagnostic surface horizons and diagnostic subsurface horizons.In the multi-class classification of the diagnostic surface horizons,SVM obtained the highest accuracy,with a total accuracy of 79.9% and Kappa coefficient of 0.54,which can distinguish Mollic epipedon,Umbric epipedon and Anthrostagnic epipedon.In the multi-label classification of the diagnostic subsurface horizons,SVM obtained the highest accuracy,with a total accuracy of 50.6% and Hamming loss of 0.19,which has a better identification effect on the Hydragric horizon and LAC-ferric horizon.As far as the diagnostic horizon is concerned,PXRF can better ability to identify the diagnostic horizon related to the element content.Compared with the diagnostic surface horizon,the diagnostic subsurface horizon has more complex labels and more diverse classification indexes,thus the classification accuracy is relatively low.