Inversion Of Soil Salinity And Organic Matter On Arable Land In The Yellow River Delta Based On Multi-source Remote Sensing

Soil is an important part of the ecosystem,and plays an irreplaceable role in the stability of the ecological environment and the development of agricultural economy.In recent years,soil degradation is getting worse,has become a global issue that constrains sustainable agricultural development and land production.However,arable land is the foundation of agricultural development,and is particularly affected by soil degradation.The soil structure of arable land in the Yellow River Delta(YRD)has been severely damaged due to the factors such as climate change,seawater intrusion,and human production and life.Soil salinity(SS)and soil organic matter(SOM)content are the main parameters about measure arable land quality.Therefore,arable land in the YRD is rapidly and accurately monitored to explore the distribution of SS and SOM,which is of great significance to comprehensive management of soil degradation,ecological environment protection and soil quality improvement.Remote sensing monitoring such as unmanned aerial vehicle(UAV)remote sensing and satellite remote sensing has become the main method for soil composition monitoring.However,a single data source cannot meet the requirements of high precision and multi-scale in the inversion process,so multi-source remote sensing data are gradually used in soil composition inversion research;however,few studies have combined unmanned aerial vehicle(UAV)hyperspectral images and satellite multispectral images for this purpose.In this study,we simultaneously inverted soil salinity(SS)and soil organic matter content(SOM),which are important saline soil quality indicators,on arable land in the Yellow River Delta(YRD)based on ground measurement data,UAV hyperspectral images,and a Landsat-8 image.The SS and SOM hyperspectral inversion models was constructed using the UAV hyperspectral data.The preprocessing of hyperspectral images was carried by Savitzky-Golay filter(S-G)and mathematical transformation,then the modeling parameters were screened by the Pearson correlation analysis.The SS and SOM hyperspectral inversion models were built using the multiple linear regression method,and then inversion was carried at the scale of the test plots based on models and UAV hyperspectral images.The SS and SOM fitted multispectral inversion models were built using the stepwise linear regression methods,and then inversion was carried at the scale of the study areas based on models and reflectance-corrected Landsat-8 image.The main conclusions are as follows:(1)The soil spectral curve generally presents as a sector with gradually increasing reflectance as the wavelength gradually increases.Spectral feature was showed low reflectance and fast rise in the visible range,and was showed higher reflectance and slower growth in the near-infrared wavelength range.(2)The SS and SOM hyperspectral inversion models were built based on UAV hyperspectral images.After S-G filter denoising and mathematical transformation of UAV images,the correlation between the measured values of soil salinity and organic matter and soil reflectance has been significantly improved.The reflectance of SS and hyperspectral reached the maximum value(R=0.633)under transformation of first-order differential of logarithm;the reflectance of SOM and hyperspectral reached the maximum value(R=0.569)under transformation of first-order differential of reciprocal.The best SS hyperspectral inversion model based on the participation of sensitive bands was constructed with the reflectance under transformation of first-order differential of logarithm;the determination coefficient(R~2)and root mean square error(RMSE)are 0.779 and 1.676 g/kg about the model.The verification R~2,RMSE and relative analysis error(RPD)are respectively 0.761,1.799g/kg and 2.004.The best SOM hyperspectral inversion model was constructed with the reflectance under transformation of first-order differential of reciprocal,and R~2and RMSE are0.763 and 4.138 g/kg about the model.The verification R~2,RMSE and RPD are respectively0.757,4.553 g/kg and 2.001.(3)The SS and SOM fitted multispectral inversion models were built based on UAV fitted multispectral images.The correlation between the fitted multispectral reflectance and SS and SOM has been improved to a certain extent after mathematical transformation or combinations.In the correlation with SS,there are 25 and 8 spectral parameters whose absolute value of correlation coefficient(|R|)exceeds 0.4 and 0.55,respectively,and the maximum value is 0.640;about the SOM correlation,there are 23 and 5 spectral parameters whose|R|exceeds 0.4 and 0.55,respectively,and the maximum value is 0.605.The modeling parameters of the best SS and SOM fitted multispectral inversion models were selected from the sensitive parameters with|R|greater than 0.4.The R~2and RMSE of the best SS fitted multispectral model were 0.691 and 1.938 g/kg,respectively;the verification R~2,RMSE and RPD are respectively 0.676,2.202 g/kg and 1.743.The R~2and RMSE of the best SOM fitted multispectral model were 0.684 and 5.105 g/kg,respectively;the verification R~2,RMSE and RPD are respectively 0.663,5.263 g/kg and 1.691.(4)Reflectance correction and multi-scale inversion of SS and SOM were carried.The Landsat-8 image was corrected using the average ratio adjustment based on the analysis of UAV fitted spectral data and Landsat-8 satellite data.The reflectance correlations of the four bands of the UAV fitted multispectral image and the satellite image are R(B_B)=0.7133,R(B_G)=0.6802,R(B_G)=0.756 and R(B_NIR)=0.7273.The reflectance correction coefficients of the four bands(B_B,B_G,B_Rand B_NIR)of the Landsat-8 satellite image are 1.09,1.25,1.19 and1.27,respectively.On the scale of the test plots,the soil salinization degree from high to low is the experimental area C,A,B,while the overall SOM content is opposite to it.At the scale of the study area,SS and SOM were negatively spatially correlated in arable land of the YRD.Areas with high SS and low SOM were mainly distributed in western Lijin County,eastern Kenli County,southern Hekou District,and northern Dongying District.Overall,the degree of soil salinization gradually increased from high to low terrain and from upstream to downstream.Moderately salinized arable land accounted for the largest proportion of area in the YRD(48.44%),with the SOM of most arable land(43.11%)at medium or lower levels.This study provides an approach for rapid and accurate monitoring of SS and SOM,which is of great significance for targeted improvement of arable land in the YRD.