Digital mapping of soil fertility attributes in croplands in Jiangsu,Zhejiang and Shanghai based on recursive feature elimination-random forest model
Obtaining quantitatively characterized spatial information of soil fertility is of great significance for improving soil quality,precision agricultural management and sustainable agricultural development.In this study,we selected croplands in Jiangsu Province,Zhejiang Province and Shanghai as the research areas,cli-mate,topography,vegetation and soil properties were selected as natural environment variables.The total power of agricultural machinery,the amount of agricultural chemical electricity consumption is used as agricul-tural activity variables,combining the above two types of environmental covariates are used as environmental covariates.The original environmental covariates were screened by the recursive feature elimination(RFE)method,and the optimal variables combination after screening was used as the independent variable to estab-lish the spatial distribution prediction model based on random forest(RF)of surface soil pH,organic carbon,total nitrogen,total phosphorus,total potassium and nitrogen,nitrate nitrogen,ammonium nitrogen,available phosphorus,available potassium,exchangeable calcium and exchangeable magnesium in the study area.The importance of environmental covariates was ranked and digital soil mapping was performed,validated with 100 replicates of ten-fold cross-validation.The results show that:1)The types of environmental covariates screened out by the 11 models mainly focus on climate,topography and vegetation variables,while the variables repres-enting human agricultural activities play an important role in the prediction of organic carbon,total phosphorus,total potassium,ammonium nitrogen and available phosphorus.2)The coefficients of determination(R2)of the 11 models were between 0.27 and 0.53,and the coefficients of determination(R2)of the prediction models for pH,available potassium,exchangeable magnesium and exchangeable calcium were all above 0.45.The recurs-ive feature elimination-random forest model(RFE-RF)proposed in this article can be used to map the main soil fertility attributes in croplands and provide the necessary basis for the spatial distribution of soil fertility attrib-utes for agricultural production.
recursive feature eliminationRandom Forestsoil fertility attributescroplandsdigital soil map-pingJiangsu,Zhejiang and Shanghai
NETL
NSTL
万方数据
