Study On Hyperspectral Inversion Of Soil PH And Total Potassium Content And Influencing Factors

Soil pH and total potassium content were important factors affecting crop growth and yield.Timely and accurate estimation of soil p H and total potassium content could improve existing fertilization technology and meet the needs of precision agriculture.The traditional method of measuring soil p H and total potassium content in laboratories was time-consuming,laborious,costly,and poor in time,which restricted the large-scale application of this method.Therefore,there was an urgent need for a new technology that can satisfy the rapid measurement of soil pH and total potassium content in agricultural production.With the rapid development of hyperspectral remote sensing technology,the use of hyperspectral remote sensing quickly and accurately obtained soil p H and total potassium content,which greatly saved manpower and material resources,and realized real-time monitoring of soil physical and chemical properties for the implementation.This technology provided an important theoretical basis for land information system management and modern agriculture.Spectral preprocessing,characteristic band selection,regression inversion model,and sample set division were important factors that affect the accuracy of hyperspectral inversion of soil properties.Therefore,the influence of hyperspectral inversion of soil pH and total potassium content were specifically studied from four aspects:spectral pretreatment method,characteristic band selection,inversion model,and sample set division method.A total of 111 samples of purple soil and paddy soil in the Wangjiagou small watershed of the Three Gorges Reservoir area were used as research objects,and comparison analysis of different inversion models of soil spectral data was conducted to establish an optimal hyperspectral inversion of soil pH and total potassium Acting model.The main conclusions were as follows:(1)Pretreatment of the soil spectrum can eliminate the noise in the spectrum,which improved the signal-to-noise ratio of the spectral data,maximize the useful information of the spectrum,and enhanced the relationship between the spectrum and the soil pH and total potassium content.Eight kinds of spectral pretreatments were performed on the original spectrum of the soil,including first-order derivative transformation,second-order derivative transformation,multivariate scattering correction transformation,convolution transformation,standard normal variable transformation,de-trend transformation,wavelet transformation,and wavelet de-envelope transformation.The correlation between the spectral data and the soil pH and total potassium content was analyzed.The results showed that the correlation between the spectral data after 8 pretreatments and the soil pH and total potassium content were improved.Among them,the wavelet de-envelope transformation spectrum had the highest correlation with soil pH and total potassium content.The absolute value of the correlation coefficient of the paddy soil pH value appeared at a maximum at the 1 100 nm band,and the value is 0.621;the correlation coefficient range of the purple soil pH value from-0.589 to 0.441,the maximum absolute value of the correlation coefficient appeared at 1 146nm.The absolute value of the correlation coefficient of the total potassium content of paddy soil at the 541nm band was the largest,the value is 0.671;the absolute value of the correlation coefficient of the total potassium content of the purple soil was 0.727 at the 517 nm band.2)The selection of spectral characteristic bands can make the model of soil pH and total potassium content easier,more robust and more accurate.the three characteristic methods of PCA loading method,correlation coefficient method and random forest variable importance method were used to select the characteristic bands of soil pH and total potassium content,and the PLSR method model inversion was performed.The correlation coefficient method had a good effect on the model of characteristic band in soil pH and total potassium content,better than PCA loading method and RF variable importance method.The selected characteristic bands of paddy soil p H were 669 nm,785 nm,1 100 nm,and 2 390 nm,the characteristic bands of purple soil pH were 570 nm,1 147 nm,2 151 nm,and 2 238 nm,and the characteristic band selection of total potassium content in paddy soil At 542 nm,754 nm,1 703 nm,and 2390 nm,the characteristic bands of total potassium content in purple soil were selected from 542 nm,972nm,1 836 nm,and 2 160 nm.(3)The optimal selection of inversion methods can effectively improve the prediction accuracy of soil pH and total potassium content.Three methods,SVR,PLSR,and PCR,were used to perform spectral inversion of soil p H and total potassium content.In the model inversion effect of paddy soil pH,the SVR method was the best,the R~2 of the calibration set is 0.649,the RMSE was 0.748,and the verified R~2 was0.385 and the RMSE was 0.980.In the process of inverting the pH value of purple soil,the inversion effect of the PLSR method was comprehensively compared,and its value is the correction set R~2 was 0.434,and the RMSE was 1.003,the verified R~2 was 0.378,and the RMSE wass 1.067.The hyperspectral inversion of the total potassium content of the soil had the best effect of the SVR model.In the paddy soil,the correction set R~2 was 0.709,and the RMSE was 1.207 g/kg,the verified R~2 was 0.455,and the RMSE was1.564 g/kg;in the purple soil The calibration set R~2 was 0.596,and the RMSE was 1.877 g/kg,the verified R~2 was 0.455,and the RMSE was 2.205 g/kg.(4)The ideal correction set and verification set can improve the model inversion speed and accuracy,making the model more robust in operation.Based on the selection of the optimal pretreatment method,characteristic band and model,the model inversion situation of the four sample set division methods of gradient quality method,random division method,KS division method,and SPXY division method were compared.The samples of the pH value,total potassium content of the paddy soil and the pH value of the purple soil were divided into the sample correction set and the verification set by the SPXY division method,and the model effect was better than the gradient quality method,the random division method,and the KS division method.The correlation coefficients of the measured value and predicted value of the paddy soil p H value,purple soil pH value,and paddy soil total potassium were 0.789,0.668,and 0.827,respectively.The model obtained by dividing the total potassium sample set of purple soil by gradient mass method is the best(R_C~2=0.608,RSME_C=1.963 g/kg,R_v~2=0.572,RSME_v=1.776 g/kg).