Inversion And Dynamics Analysis Of Soil Salinity Based On Multi-source Data In The Yellow River Delta

As an important part of the ecological environment,soil plays an irreplaceable role in the structural stability and function of the entire ecosystem.In recent years,soil salinization has gradually become a global environmental problem affecting land productivity and sustainable agricultural development.Affected by paticular natural and man-made driving forces such as seawater intrusion,high groundwater level,unreasonable human reclamation,and the construction of Xiaolangdi Reservoir,the soil properties and ecological environment in the Yellow River Delta have been severely damaged,and the salinization situation is particularly severe.Therefore,fastly and accurately monitoring the soil salinity value at the Yellow River Delta and investigating the temporal and spatial pattern and the correlation of the soil salinity and external factors have great meaning to the comprehensive improvement of saline soil,land use planning,the maintenance of a virtuous cycle of the entire ecological environment,and the sustainable development of the region.Traditional ecological parameter monitoring methods based on field surveys and remote sensing monitoring methods based on satellite remote sensing and UAV remote sensing have become the main methods for monitoring regional soil salt content.However,in the dynamic monitoring of soil salinity,it is difficult for a single data to meet the requirements of different regional scales and accuracy.There are problems such as inconsistent band reflectivity and weak data correlation among multi-source sensor data.In this study,the Yellow River Delta Area(Hekou District,Kenli District,Dongying District and Lijin County of Dongying City)was taken as the research area.Two experimental plots located in Kenli District and Hekou District were taken as the experimental area.Based on UAV multi-spectral remote sensing images,Sentinel-2A satellite series images and soil salinity field data.various modeling methods and optimization methods of different spectral parameters were respectively in the construction of the soil salt prediction models and the comparison of the accuracy of the models.The ratio average method was used to complete the reflectance correction of Sentinel-2A satellite images and UAV images,based on which,the reflectance correction results were verified.Then the best prediction model of soil salinity was applied to the UAV multi-spectral imagery and Sentinel-2A satellite imagery to achieve the inversion of the soil salinity in the experimental area and the research area.Combining the data of soil salt content in the study area with the area of land type interest,the spatial and temporal dynamic variation characteristics of soil salt content(SC)and its relationship with land use in the Yellow Triangle region were analyzed by using the methods of dynamic attitude and transfer matrix.Finally,high-precision inversion and spatiotemporal dynamic analysis of soil salinity based on multi-source data were realized.The main research contents and conclusions are as follows:(1)Screening of sensitive bands and spectral parameters.Based on the gray correlation analysis method and the Pearson correlation coefficient method,the correlation analysis of a total of 16 spectral indices including the reflectance of the UAV in each band and the different vegetation index and salinity index was carried out.The results show that:The sensitive bands of soil salinity are green light band(Bg)and red light band(Br);Regarding the correlation between the same band or index and soil salinity,the gray correlation coefficients are higher than the Pearson correlation coefficient,and the gray correlation coefficients of Bg and Br are 0.597 and 0.599,respectively(p>0.01).Considering the comprehensiveness and accuracy of the model parameters,16 spectral parameters with high gray correlation coefficient(P>0.01)were finally selected for later inversion.(2)Construction of soil salt prediction model and optimization of the best prediction model.Four modeling methods were used:Partial Least Squares Regression(PLSR),Multiple Linear Regression(MLR),Back Propagation Neural Network(BPNN)and Support Vector Machine(SVM).The soil salinity inversion model was constructed based on the band reflectivity,vegetation index,salinity index and comprehensive index,and the accuracy was compared.The results show that BPNN has the highest prediction accuracy among the four modeling methods;Among the 4 groups of inversion models,the comprehensive inversion model of soil salinity has the highest overall accuracy;The best prediction model is a comprehensive inversion model of soil salinity based on BPNN with coefficient of determination R~2=0.769 and square error of root mean(RMSE)=2.342 respectively.The verification R~2,RMSE and relative analysis error(RPD)are respectively 0.774,1.475,and1.799.(3)Reflectance correction and verification and inversion of the best prediction model of soil salinity.The relationship between the UAV remote sensing image and the reflectivity of the Sentinel-2A satellite remote sensing image were analyzed.The ratio average method was used to calculate the satellite image reflectivity correction coefficient.The reflectivity correction processing was performed on the Sentinel-2A satellite image and was used to filter out the comprehensive inversion model of soil salt based on BPNN estimates the soil salt content in the experimental area and the research area respectively.The results show that there is a strong correlation between the UAV multi-spectral remote sensing image and the Sentinel-2A satellite image.The correlation coefficients of the reflectance of the two bands are 0.736,0.897,0.863 and 0.821 respectively,which are all higher than 0.7(p>0.01);The inversion results of soil salt content in the experimental area and the study area in different periods(2016 and 2019)have abnormal values within 10%and 15%,respectively,and the distribution of different saline soils are consistent with the actual situation.(4)Analysis of the temporal and spatial dynamics of soil salinity in the Yellow River Delta.According to the spatial distribution map of soil salinity in the experimental area and the research area obtained by the inversion,the temporal dynamics and spatial pattern of different degrees of saline soil are studied through dynamic analysis,area statistics,transfer matrix and other methods with land use classification data.The results show that from 2016 to2019,the overall soil salt content in the study area showed a higher salt content in the central and eastern regions and a lower salt content in the southwest.The interannual fluctuations of the surface soil salt content were small,and the soil salt content in the two periods was moderately variable.In terms of spatial changes,the area of non-saline soil and lightly saline soil in the study area decreased(92.817km~2and 256.91km~2),and the area of moderately-salinized soil and saline soil increased(144.12km~2,117.097km~2and 88.510km~2);Among the ground feature types,the structural stability of cultivated land,construction land,water area and unused land is relatively poor,and it is easy to transform with other ground feature types.In terms of the relationship between land use and soil salinity distribution,the average area of different grades of saline soil in the four land use types changed from large to small in order of unused>grassland>forest land>cultivated land.Among them,the largest change in the area percentage is the heavily saline soil(89.142%)distributed in unused land,and the smallest change in the area percentage is the non-saline soil(-12.149%)distributed in the grassland.