Prediction Of Forest Soil Nutrient Spatial Distribution Based On BP Neural Network

The soil provides the necessary living environment and nutrient for production and development in the forest ecosystem,and it is the hub of the transformation of the nutrient elements,which plays a very important role in the development of the forest ecosystem.The soil is uneven and variations in the space continuum,with a high degree of spatial variability in space.It can provide decision support for forest management and realizing the "fine improvement of forest quality".It can provide reference for ecological environment protection,rational use of forest land resources and soil nutrient zoning management.In this study,soil available nitrogen,available phosphorus and available potassium,were selected from the forest soil.Soil types and terrain parameters,hydrological parameters derived from 10 m resolution Digital Elevation Model were used to build predictive BP neural network model.Among them,the terrain parameters include slope,slope,slope length and terrain position index.The hydrological parameters include soil terrain factors,sediment delivery ratio,vertical slope position,flow direction and flow accumulation.Then the best model structure was selected by performance and precision calibration and applied to the prediction and mapping of soil nutrient distribution in the study area.Research specific content and conclusions are as follows.The contents of available N,available P and available potassium in 249 soil samples collected in the study area were statistically analyzed and verified by normality,The results showed that the contents of available N,available P and available K were 6.72~ 508.32 mg/kg,0.11~15.72 mg/kg and 9.15~187.62 mg/kg,respectively.The three nutrient contents were converted by natural logarithm,and the three nutrient contents conformed to the normal distribution after natural logarithmic conversion.The soil samples in the study area were used as model training sets for model training.123 soil samples of Xinxing County were validated as the calibration set.46 models were constructed for the three nutrient contents by using different terrain parameters and hydrological parameters combination and soil type as input sections.Stratified stepwise selection strategy,comparative model accuracy on the validation set to give the best prediction model structure.The three layer network structure of 7:20:1 is the best predictor of available N with the terrain position index,slope,aspect,soil terrain factors,sediment delivery ratio,flow direction and soil type as the input node.The MSE of the model is as low as 1 888.96,the correlation coefficient is as high as 0.787 7,RO A(±10,±15 and ±20)is 45.45%,60.70% and 68.72%.The three layer network structure of 6:18:1 is the best predictor of available P with the terrain position index,slope length,aspect,vertical slope position and soil type as the input node.The MSE of the model is as low as 0.64,the correlation coefficient is 0.854 0,RO A(±0.1,±0.3 and ±0.5)is 21.12%,54.28% and 72.46%.The three layer network structure of 5:18:1 is the best predictor of available K with the slope,slope length,aspect,vertical slope position and soil type as the input node.The MSE of the model is 395.64,the correlation coefficient is 0.762 3,ROA(±5,±10and ±15)is 21.12%,54.28% and 72.46%.The three best models showed excellent performance in independent sample calibration,which indicated that they could effectively predict the corresponding forest soil nutrients and had some universality.Three kinds of soil nutrient distributions were predicted and mapped by using the selected model.The results showed that the distribution of available N and available phosphorus were in the whole study area,and the available potassium was moderate variation.According to the second national nutrient classification criteria for soil nutrient screening,the area of the forest soil lacking and lacking available N,available P and available K accounted for 47%,95.78% and 80.10%,respectively.The terrain parameters have a greater effect on the prediction accuracy of three soil nutrients than the hydrological parameters.The slope has the largest frequency in the best model.The smaller the sediment transport ratio or the longer the slope length corresponds to the high available N content.The area with high available phosphorus content is mostly in the ridge and uphill.The higher the vertical slope position,the higher the content of available P and available K in the corresponding zone.