Retrieval Model Of Soil Salinization Based On Radar Satellite Remote Sensing

Soil salt content is an important condition affecting the normal growth of crops and the sustainable development of ecological environment.Timely and effective monitoring of soil salt content is of great significance to guide agricultural production.Satellite radar remote sensing has the ability of all-weather,all-time and certain penetration,which makes up for the deficiency of optical remote sensing affected by cloud and rain weather.Using radar image to monitor soil salt content has certain advantages.In this paper,Shahao canal in Hetao irrigation area of Inner Mongolia is taken as the experimental study area,and the soil salt content at different depths in bare soil period and vegetation coverage period is taken as the research object.Meanwhile,the radar images in bare soil period and vegetation coverage period were collected.By extracting the backscattering coefficient and constructing the polarization combination index,the correlation between the backscattering coefficient,polarization combination index and soil salt content at different depths in two periods was analyzed.The sensitive index and the best combination independent variable were selected by gray correlation and full subset screening to establish the inversion model.PLSR?QR and SVM models were established for soil salt content at bare soil stage combined with two depths;In the vegetation coverage period,the partial PLSR?QR?ELM and SVM were constructed based on the soil salt content at seven depths.Finally,four precision evaluation indexes were used to evaluate the inversion model.The main conclusions are as follows.(1)The sensitive index of soil salt content in different periods was constructed.The sensitive index of soil salt content in different periods was constructed.The indices sensitive to soil salt content at different depths of bare soil stage were screened out by gray correlation analysis,which included 2 backscattering coefficients and 8 polarization combination indices.The indexes sensitive to soil salt content are not the same at different depths of vegetation cover period by using the full subset,and the best combination of independent variables at different depths is selected.Under the condition of vegetation coverage,the sensitivity index is different from that of bare soil at the same depth.There are 5 sensitivity indexes of vegetation cover period at 0-10 cm depth,and 2 sensitivity indexes of vegetation cover period at 10-20 cm depth.(2)Based on radar remote sensing,the inversion model of soil salt content at different depths in bare soil period was established.By combining two groups of backscattering coefficients to construct multiple polarization combination indexes,and then using gray correlation analysis to select the polarization combination index with higher correlation with soil salt content.PLSR?QR and SVM were used to build soil salt content inversion models at 0?10 cm and 10?20 cm soil depths.The results of bare soil period show that the accuracy of SVM is the highest among the three regression methods.At this time,the decision coefficient R_c² of modeling set and R²_p of verification set of SVM model are all above 0.4.The root mean square error R_MSEp of modeling set and validation set are less than 0.3%.QR model is the second,PLSR model is the worst.The prediction results of SVM model in 0?10 cm soil layer are better than other models,R_c² and R_p² are 0.568 and0.686 respectively,R_MSEc and R_MSEp are 0.152%and 0.151%respectively.(3)Based on radar remote sensing,the inversion model of soil salt content at different soil depth in vegetation coverage period was established.Four new indexes were added to the polarization combination index established in the bare soil stage,and the best combination of indexes under 7 soil depths in the vegetation coverage stage were selected by the full subset screening method.Partial least squares regression,quantile regression,extreme learning machine and support vector machine were used to build the inversion model of soil salt content at different depths of 0?10 cm,10?20 cm,0?20 cm,20?40 cm,0?40 cm,40?60 cm and 0?60 cm.Comprehensive analysis shows that under the condition of 7 Sampling depths,the prediction accuracy of SVM model under the condition of 10?20cm depth is the highest,R_p² is 0.67,R_MSEp is 0.12%.The results show that the SVM model of 10?20 cm is suitable for soil salt content inversion of shahao channel in vegetation coverage period.