| Soil salinization is an increasingly severe international land resource and ecological environment problem,which poses significant challenges to agricultural production problems such as land management,farming labor and the sustainable development of wetland ecology.With the increasing population and living needs in coastal areas,it is an urgent problem to accurately extract the soil salinization status in coastal areas and master the temporal and spatial distribution of soil salinization.In recent years,in order to map soil salinity information in all aspects and improve the model’s accuracy,the large increase in model input has brought the problem of complex input data dimensions.Given this problem,a variety of soil salinity inversion factors are comprehensively considered in.this paper,combined with the correlation between the inversion factors and measured soil salinity,and the optimal factor is selected in each type to participate in the modeling.Through Multiple Linear Regression,Partial Least Squares Regression,BP Neural Network,Random Forest,and Support Vector Machine modeled soil salinity,and a soil salinization grade distribution map at different depths is drawn to analyze the temporal and spatial dynamic changes of soil salinization in the Yellow River Delta from 2003 to 2018.The specific work is as follows:(1)Select the optimal soil salinity inversion factor.This paper aims to propose a comprehensive optimal model based on remote sensing to detect soil salinity at different depths in the Yellow River Delta.After image processing of multivariate remote sensing data,26 soil salinity inversion factors are extracted and classified into 5 categories.Measured data on soil salinity at the depth of 30-40 cm and 90-100 cm in the study area are obtained through field exploration.Combined with the correlation between the five types of inversion factors and soil salinity,in order to avoid the collinearity problem between various inversion factors,the optimal factor is selected among each type of factors to participate in the modeling.The experimental results show that the optimal inversion factors are the same under two different soil depths,namely SWIR1,SI9,MSAVI,Albedo,and SDI.(2)Establish a soil salinity inversion model.According to the five optimal soil salinity inversion factors selected,empirical models(MLR,PLSR)and machine learning models(BPNN,RF,SVM)are used to establish soil salinity inversion models in the Yellow River Delta.Exploring the ability of shallow soil and deep soil to resolve salinity.Experimental results show that the accuracy and stability of the PLSR model are higher than those of other models.The R~2 between the measured and inversion values of soil salinity at 30-40 cm is 0.642,RMSE is 0.283,MAE is 0.213,the R~2,RMSE,and MAE at the depth of 90-100 cm is 0.450,0.276,0.220,respectively.(3)Quantitative analysis of soil salinity’s temporal and spatial dynamics from 2003 to 2018.The optimal soil salinity inversion model is selected to draw the distribution map of soil salinization grade at the depth of 30-40 cm and 90-100 cm across the Yellow River Delta in 2003,2008,2013,and 2018.The area and transfer matrix of different salinization grades are analyzed in detail.The results show that soil salinity showed apparent spatial heterogeneity from 2003 to 2018.Soil salinization was higher along the coastal coastline,while soil salinization was lower in the middle of the study area.In the past 15 years,soil salinity at depths of 30-40 cm and 90-100 cm has shown a fluctuating upward trend.In summary,this study proposes a soil salinity inversion model based on the optimal inversion factor of remote sensing.The results show that the soil salinity inversion by the model is consistent with the measured soil salinity. |
PDF Full Text Request