Estimation And Mapping Of Soil Total Nitrogen,Total Phosphorus And Total Potassium Using Hyperspectral Data

As one of the important components of terrestrial ecosystem,soil provides necessary moisture and nutrients for plant growth.Especially,as main source of plant nutrition,soil nutrients（nitrogen,phosphorus and potassium）play a vital role in plant growth.Thus,it is very important to monitor soil nutrients（nitrogen,phosphorus and potassium）.However,traditional measurement methods are time-consuming and costly for the generation of spatially explicit estimates for regional-scale area,although they can provide accurate estimates of soil nutrient contents.In order to meet the need of modern soil management in regional-scale area,it is urgent to develop new technique for estimating soil nutrients.The development of remote sensing technology,especially hyperspectral remote sensing technology,provides a new way to realize large-scale and rapid monitoring of soil nutrient contents.Due to the low accuracy of the current hyperspectral estimating models for estimating soil nutrients（nitrogen,phosphorus and potassium）,this study acquired a high-accuracy estimating model to estimate model by performing two steps:（1）selecting the optimal algorithm to determine the accurate spectral response variables to soil nutrients;（2）modeling method to obtain the optimal model for more-accurate estimating soil nutrients.This study was conducted in Guangdong of China based on the soil nutrient contents and hyperspectral data collected from 75 soil samples.Firstly,the contents of total nitrogen（TN）,total phosphorus（TP）and total potassium（TK）in soil were determined by indoor chemical analysis method,and the spectral data of soil samples were collected by means of ground object spectrometer.Secondly,the original spectrum was spectrally transformed and pearson correlation coefficient（PCC）,least absolute shrinkage and selection operator（LASSO）and gradient boosting decision tree（GBDT）were used to determine the appropriate spectral variables.Thirdly,the multi linear regression（MLR）,partial least squares regression（PLSR）,back propagation neural network（BPNN）,BPNN with genetic algorithm optimization（GA-BPNN）,support vector machine（SVM）and radial basis function（RBF）methods were used to construct a high-accuracy model to predict the soil nutrient contents,respectively.Finally,the determined high-accuracy model of TK was used to map the soil TK contents at a regional scale using Huan Jing-1A Hyperspectral Imager（HJ-1A HSI）image.The results showed that:1)Based on soil spectral reflectance and various spectral transformation variables,the three algorithm（PCC,LASSO and GBDT）were used extract the appropriate spectral variables of soil TN,TP and TK.Taking the appropriate spectral variables obtained based on PCC as an example,the appropriate spectral variables of soil TN are R342.095,FD562.78,FD1418.56,SD714.116,LR768.1;the appropriate spectral variables of soil TP are R1302.573,FD1009.08,FD613.137,FD356.912,SD905.516,LR_1065.082;and the appropriate spectral variables of soil TK are R_2498.689,FD_442.459,FD_625.25,SD_1043.94 and RL_2461.182.;2)On the basis of the above research,the hyperspectral estimation models of the soil nutrients（TN,TP,and TK）are constructed by MLR,PLSR,BPNN,GA-BPNN,SVM and RBF.The results show that the most accurate estimates of soil TN content is the GBDT-GABP model with the modeling decisive coefficient（R²）,relative root mean square error（RRMSE）and ratio of performance to deviation（RPD）values of 0.9101,12.33%and 3.36,respectively;the testing R²,RRMSE and RPD values of 0.8604,17.08%and 2.52,respectively.The most accurate estimates of soil TP content is the PCC-GABP model with the modeling R²,RRMSE and RPD values of 0.9450,16.53%and 4.21,respectively;the testing R²,RRMSE and RPD values of 0.7855,42.84%and 1.96 respectively.The most accurate estimates of soil TK content is the GBDT-GABP model with modeling R²,RRMSE and RPD of 0.9702,12.53%and 5.82 respectively;the testing R²,RRMSE and RPD values of 0.9067,23.39%and 3.13,respectively;3)The GBDT-GABP method was used to map the distribution of soil TK content using Huan Jing-1A Hyperspectral Imager（HJ-1A HSI）data with a RRMSE value of 22.49%,implying that the GBDT-GABP model provided the potential to map the soil TK content for the large area.