Digital Mapping For Soil Nutrients In A Hilly Region

Soil nutrients play an important role in terrestrial ecosystems.A comprehensive understanding the spatial distribution of soil nutrients is crucial for soil fertility improvement,ecological modeling and environment prediction,precision agriculture and land use planning.Previous researches on spatial distribution of soil nutrients relys on the conventional soil cartography.However,the conventional soil cartography is time-and labor-consuming,over-reliance on experience knowledge.Besides,the soil map produced by conventional soil cartography cannot express the continuous and gradually variation of soil nutrient and offer sufficient information to satisfy the applications(e.g.precision agriculture and ecological modeling).Thus,alternative approaches with higher precision and efficiency are required for quantifying the spatial variation of soil nutrient.Digital soil mapping is a quantitative prediction approach for soil nutrient based on environmental vairables and prediction algorithms,which can overcome the weakness of the conventional soil cartography.The study area is located in Jiangjin,Chongqing.The influences of environment variables(e.g.vegetation,climate,topography,and soil types)on soil nutrients and the spatial variation characteristics of soil nutrients were discussed.Then six methods(multiple linear regression(MLR),geographically weighted regression(GWR),support vector machine regression(SVR),gradient boosting decision tree(GBDT),random forest(RF)and random forest regression kriging(RFRK))were applied to produce prediction models of soil nutrients(pH,organic matter(OM),alkali-hydro nitrogen(AVN),effective phosphorus(AVP),effective iron(Fe),effective manganese(Mn),effective copper(Cu)and effective zinc(Zn))based on environmental variables.Finally,the spatial distribution maps of soil nutrients were produced by the best prediction model.The main results were as follows:The range of pH is 3.6~8.5.The contents of OM,AVN and AVP are at the medium level as a whole.The contents of Fe,Mn,Cu and Zn are rich or extremely rich as a whole.Soil nutrients belong to modertate or strong variability.Soil nutrients were closely correlated with vegetation index,climate factors,and terrain factors.Soil nutrients are significantly correlated with vegetation index except AVN,AVP and Fe.Soil nutrients are significantly correlated with climate factors except AVP.Soil nutrients are significantly correlated with terrain factors.Soil microelements were closely correlated with pH,OM,AVN and AVP.pH are siginificantly correlated with soil microelements,OM and AVN were negatively correlated with pH,and positively correlated with soil microelements,AVP were negatively correlation with Fe,Mn,and Cu,and positively correlated with Zn.Soil type and land use type had significant influence on soil nutrient contents(p < 0.05).For soil type,the average value of pH in the calcareous purple soil was the highest(6.23).The average value of OM in the submerged paddy soil was the highest(19.35 g/kg).The average values of AVN,Fe,and Mn in the waterlogged paddy soil were 104.70 mg/kg,95.77 mg/kg,and 88.45 mg/kg,respectively,which were higher than that of the others.The avarege values of AVP and Cu in the yellow soil were 1.38 and 1.88 mg/kg,respectively,which were higher than that of the others.The avarege value of Zn in the acid purplish was the highest(4.77 mg/kg).For land use type,the avarege value of pH in the garden was the highest(5.90).The avarege values of OM,AVN,Fe,Mn,Cu,and Zn in the paddy field were 18.84g/kg,105.10mg/kg,101.19mg/kg,81.45mg/kg,1.50mg/kg,and 3.40mg/kg,respectively,which were higher than that of the others.The avarege value of AVP in the dryland was the highest(11.26 mg/kg).The optiamal combinations of environment variables for soil nutrients were determined by stepwise regression method based on auxiliary variables including vegetation indexes,climate factors,terrain factors,soil types and land usetypes.In the regression equations of soil nutrients,soil types entered all the regression equations of the soil nutrients,land use types entered all the regression equations of the soil nutrients except Fe,vegetation indexes and climate factors entered into the regression equations of pH,OM,Cu and Zn,terrain factors entered into the regression equations of all soil nutrients,pH entered into all the regression equations of soil microelement,OM entered into regression equations of Mn,Cu and Zn.Results showed that soil types,land use types and topography had a widest influence on the variation of soil nutrient content;pH and OM had a large effect on the spatial variation of soil microelement content.For each soil nutrient,the optimal combinations of environment variables from stepwise regressions were used to develop GWR,SVM,GBDT,RF and RFRK model,and the performance of models was tested.Results showed that there are differences in the prediction models were applied to different soil nutrients.The RFRK model had the highest prediction accuracy in the prediction models of pH,AVP,Mn and ZN,and the GBDT model had the highest prediction accuracy in the prediction models of OM,AVN and Fe.The spatial distribution maps with higher precision of soil nutrients could provides a theoretical support for agricultural production,which produced by the best prediction model.The relative importance of environmental variables in prediction model of soil nutrient was calculated.In the current study,the main factors for affecting pH variability were slope,land use,and soil type,the main factors for influencing OM variability were channel network to base level,land use,and slope,the main factors for affecting AVN variability were channel network to base level,slope,and valley depth,the main factors for influencing AVP variability were slope,land use,and relative slope position,the main factors for Fe variability were pH,land use and soil type,the main factors for Mn variability were pH,slope,and soil type,the main factors for affecting Cu variability were OM,soil type,and average annual precipitation,the main factors for influencing Zn variability were pH,vegetation index,and slope.