Model Prediction Of High-resolution Three-Dimensional Forest Soil Nutrients

Forest soil nutrients refer to the mineral nutrients that can be absorbed by plant roots directly or after transformation in forest ecosystem.They are not only important parts of forest soil fertility,but also important indicators to measure forest soil quality.They mainly include soil organic matter(SOM),total nitrogen(TN),total phosphorus(TP),total potassium(TK),alkeline nitrogen(AN),available phosphorus(AP)and available potassium(AK).Figuring out the three-dimensional spatial distribution of forest soil nutrients and determining the levels of soil nutrients play an important role in scientific forest management,reasonable fertilization,improving the efficiency of fertilizer utilization,and promoting the sustainable utilization of forest resources.In this study,the forest soil in Yunfu of Guangdong province was taken as the object.The back propagation artificial neural network(BP-ANN)prediction model of seven soil nutrients(SOM,TN,TP,TK,AN,AP and AK)was established for the soil with depth of 0-100 cm from top to bottom in five layers(D1: 0-20 cm,D2: 20-40 cm,D3: 40-60 cm,D4: 60-80 cm and D5: 80-100cm).The model inputs consisted of two parts: the necessary parameters and candidate parameters.A set of coarse-resolution soil nutrient maps with a scale of 1:1million were the necessary input parameters,including coarse-resolution soil organic matter map(CSOM),coarse-resolution total nitrogen map(CTN),coarse-resolution total phosphorus map(CTP),coarse-resolution total potassium(CTK),coarse-resolution alkeline nitrogen(CAN),coarse-resolution availablephosphorus(CAP)and coarse-resolution available potassium(CAK);The candidate parameters were derived from digital elevation model(DEM)of 10-m resolution by Arc GIS spatial analysis,including 4 topographic parameters,i.e.slope,aspect,topographic position index(TPI),potential solar radiation(PSR),and 5 hydrological parameters,i.e.depth to water(DTW),sediment delivery ratio(SDR),flow length(FL),flow direction(FD)and soil terrain factor(STF).There were a total of 449(385+64)sample data in the study area,including 385 sample data in the modeling area(Luoding and Xinxing in Yunfu),which were used for model building and 10 fold cross validation,and 64 sample data in the independent validation area(Yunan,Yuncheng and Yunan in Yunfu),which were used for model validation of application ability in the independent area.According to the five model evaluation indices of RMSE,R2,ROA±5%,ROA±10% and ROA±20%,nine candidate parameters of topography and hydrology were screened,and the optimal combination of model input parameters was obtained based on the principle that the accuracy of the model was not significantly improved by increasing the number of candidate parameters.The optimal BP-ANN model was used to produce the spatial distribution map of soil nutrients,and the three-dimensional spatial distribution of soil nutrients in Yunfu was evaluated according to the soil nutrient classification standard of the second National Soil Census.The specific conclusions were as follows:(1)In the BP-ANN prediction model of five soil layers for each index of soil nutrients,the optimal model inputs of AN were combinations with 5-6parameters,and the prediction accuracies were significantly better than other indices(RMSE: 431.2mg/kg-866.5mg/kg,R2: 0.86-0.89,ROA±5%: 60%-78%,ROA±10%: 81%-92%,ROA±20%: 87%-95%,respectively).The optimal model inputs of SOM,TP,TK and AK were combinations with 5-8 parameters,the prediction accuracies were lower than AN,higher than TN and AP(R2:0.68-0.88,ROA±5%: 34%-53%,ROA±10%: 58%-72%,ROA±20%: 70%-86%).The optimal model inputs of TN and AP were combinations with 5-7 parameters except a 2-parameter combination in D5 soil layers of TN,and the prediction accuracies were lower than other indices.As mentioned above,the BP-ANN model established by the 10-fold cross validation method could better express the mapping relationship between high-resolution topographical and hydrological parameters and soil nutrient indexes,and reflect the complex variability of soil nutrient indices with terrain changes.(2)Based on the explaination abilities and the statistical frequency of the topographical and hydrological candidate parameters for individual soil nutrient:Slope and TPI had relatively strong ability to predict SOM and AN;FD and Aspect had relatively strong ability to predict TN;SDR and STF had relatively strong ability to predict TP;SDR and TPI had relatively strong ability to predict TK;TPI had relatively strong ability to predict AP;Slope,SDR and DTW had relatively strong ability to predict AK.However,TPI had relatively strong prediction ability to the whole soil nutrient,followed by slope,SDR and aspect;and STF,FD,DTW,PSR and FL had relatively weak prediction ability to the whole soil nutrient.On the whole,the prediction abilities of different candidate parameters in different soil layers and different soil nutrient indices were different.(3)According to soil nutrient classification standard stipulated by the second National Soil Census,the average values of soil nutrient indicators predicted by D1-D5 soil layer of soil nutrient model were classified at horizontal spatial distribution: The average value of SOM in D1 soil layer was 22.66 g kg-1,which belongs to grade III;D2-D4 soil layer was 15.1-16.58 g kg-1,which belonged to grade IV;D5 soil layer was 9.05 ± 3.72 g kg-1,which belonged to grade ?.TN and TK,the average value in D1-D5 soil layer were 1.13-1.44 g kg-1and 15.92-18.83 g kg-1 respectively,belonging to class III.TP,the average value in D1-D5 soil layer was 0.28-0.31 g kg-1,belonging to grade ?.The average value of AN in D1 soil layer was 97.97 mg kg-1,which belonged to grade ?;D2-D5 soil layer was 80.71-87.16 mg kg-1,which belonged to grade ?.The average value of AP in D1-D5 soil layer was 0.72-1.36 mg kg-1,which belonged to grade VI.The average value of AK in D1 soil layer was 51.22 mg kg-1,which belonged to grade IV;in D2-D5 soil layer was 36.77-43.09 mg kg-1,which belonged to grade ?.On the whole,the content of TN and TK were rich;SOM and AN were in the middle level;TP,AP and AK were relatively deficient,which was the main limiting factor of soil quality in Yunfu.(4)The vertical spatial distribution of soil nutrients was described.With the increase of soil depth,the content of SOM,AN,AP and AK decreased gradually.TN content increased in D1-D2 soil layer,and then decreased in D3-D5 soil layer.TP content did not change significantly in D1-D3 soil layer,but increased slightly in D4-D5 soil layer.TK content gradually increased in D1-D4 soil layer,and decreased slightly in D5 soil layer.These results were consistent with the relevant research and reports of soil nutrients in the subtropical region.