Comparative Study On Modeling Methods Of Total Carbon And Total Nitrogen Based On SVM And RF Algorithm

Soil total carbon and nitrogen are important indicators for measuring soil fertility status,mainly derived from animals and plants,microbial residues and root exudates,etc.,and are the main factors to maintain carbon balance and carbon cycle system of terrestrial ecosystems.In different geographical environments and topography,the soil total carbon and nitrogen factors and spatial distribution are also different.In this paper,the soil of the Heihe source area is used as the research object.With the help of digital terrain analysis methods,geographic information systems,computer technology,and machine learning methods,the vegetation factors,soil factors and terrain factors are systematically discussed.Effects of soil total carbon and nitrogen in different altitude ranges.Support vector machine(SVM)and random forest(RF)are applied to soil total carbon and nitrogen prediction analysis,and the prediction accuracy of the two methods is compared.The following conclusions are obtained:(1)RF is applicable to the prediction of soil total carbon and total nitrogen in the source area of the Heihe River;soil bulk density is the main factor affecting soil total nitrogen and total carbon modeling.(2)SVM is suitable for the prediction of soil total nitrogen at different altitudes,and prediction of soil total carbon in the altitude range of 3757~4257 m,and RF is suitable for the prediction of soil total carbon in the other altitude range;Bulk density is the most important factor affecting soil total carbon and nitrogen.At an altitude of 3757~4257 m,TVI is the main factor affecting soil total nitrogen modeling and Slope is the main factor affecting soil total carbon modeling.(3)SVM is applicable to the prediction of total nitrogen in shrub soil,and RF is applicable to the prediction of total carbon and nitrogen in other soils;soil bulk density is the main factor affecting the modeling of total carbon and nitrogen in shrub and meadow soil,and rainfall is the main factor affecting the total soil nitrogen in grassland.The main factor of carbon total nitrogen modeling.(4)In addition to black felt soil,RF is applicable to the prediction of total soil nitrogen of other soil types,and SVM is applicable to the prediction of total carbon of other soil types;soil bulk density is the main factor affecting the total carbon and total nitrogen of grass felt soil and black felt soil,NDVI It is the main factor affecting the total nitrogen modeling of gray-brown soil;conductivity is the main factor affecting the total carbon modeling,Slope is the main factor affecting the total carbon modeling of cold-calcareous soil,TVI is the main factor affecting the modeling of total nitrogen.