Soil Salinity Inversion In Coastal Area Of Yellow River Delta Based On The Data Fusion Of Satellite And UAV Image

Soil salinization has always posed global resource and ecological problems.General Secretary Xi stressed:"Stabilize China’s grain,saline alkali land has great potential".Rapid and accurate access to soil salt content and its spatial distribution information is the premise of the comprehensive treatment and utilization of saline soil.Remote sensing has become an effective way to quantitatively extract salinization information.However,the accuracy and stability of quantitative analysis of soil salt content based on single sensor need to be improved.The high-precision and fast retrieval of regional soil salinity based on multi-source remote sensing fusion is not only the practical demand of saline soil treatment and utilization,but also the main trend of quantitative remote sensing development of soil salinization.In this article,taking the Kenli District in the Yellow River Delta of China as the study area,a concentrated and continuous saline soil area was selected as the test area,and Sentinel-2A Multi-Spectral Instrument(MSI)image of the study area and unmanned aerial vehicle(UAV)image of the test area were obtained respectively.Then,based on UAV and MSI image data(UAV-MSI),the selection of soil salinity sensitive bands,the optimization of spectral indexes and the construction and optimization of soil salinity inversion models were carried out.Based on the numerical regression method and the average-ratio adjustment,MSI-UAV spectral data were fused.The fusion effects of different fusion methods were compared and analyzed,and the appropriate method was selected.Using the appropriate method,the satellite and UAV image data were fused respectively based on different levels including spectral data fusion,spectral index fusion,and spectral model fusion.The soil salinity spectra response and model accuracy unfused and fused under different levels were compared and analyzed,and the best fusion level and the best fusion model were optimized.Based on the best fusion model applied to MSI image,the soil salinity inversion map of the study area was obtained.The specific research contents and results are as follows:(1)MSI-UAV spectral characteristics and quantitative model of soil salinityThe SSC response bands from MSI-UAV were relatively consistent,R and G bands can be used as soil salinity characteristic bands.The 9 spectral parameters with good response to soil salinity were Int1,Int2,SI1,SI2,BI,Int1 re,Int2 re,SI II re and SI IV re.According to multicollinearity analysis,spectral parameters were divided into groups A and B.Three modeling methods,multiple linear regressions(MLR),support vector machine(SVM)and back propagation neural network(BP),were used to construct the soil salinity inversion models based on MSI-UAV respectively.The results show that whether based on UAV or MSI,SVM model in Group B had the highest accuracy.Therefore,SVM was determined as the optimal modeling method.Int1,Int2,SI1,SI2,BI,Int1 re,Int2 re,SI II re and SI IV re were the preferred MSI-UAV spectral parameters.The best inversion model in test area was the SVM model of soil salinity based on UAV characteristic spectral parameters in Group B.(2)MSI-UAV fusion methodBased on the numerical regression method and the average-ratio adjustment,MSI-UAV spectral data were fused.The results showed that for improving the fitting degree,compared with the average-ratio adjustment,the fitting degree of spectral data based on numerical regression fusion was improved by 0.003～0.009.For enhancing soil salinity spectral response,compared with the average-ratio adjustment,the absolute value of correlation coefficient of spectral data based on numerical regression fusion was increased by 0.031～0.059.For improving the accuracy of the soil salinity model,compared with the average-ratio adjustment,the accuracy of the model based on numerical regression fusion had most improved,with the calibration R~2 improved in the range of 0.047,the validation R~2 improved by 0.011,and the RPD improved by 0.147.Therefore,the numerical regression method was better in improving the fitting degree of MSI-UAV,enhancing the spectral response of soil salinity and improving the accuracy of the inversion models.It was determined that the numerical regression method was the appropriate method for MSI-UAV fusion.(3)MSI-UAV fusion levelBased on the preferred numerical regression method,The MSI and UAV multi-spectral data were fused at the spectral data level,spectral index level,and spectral model level.These results showed that,among the three levels,the fitting of spectral data was the best,followed by that of the spectral index,while that of the spectral model was the lowest.For enhancing soil salinity spectral response,spectral data fusion was better than spectral index fusion.The correlation between spectral index and soil salinity was increased by 0.139～0.167 after spectral data fusion,and increased by 0.128～0.152 after spectral index fusion.For improving the accuracy of the soil salinity model,the accuracy of the soil salinity models based on three fusion levels were all significantly improved.Compared with unfused model of MSI,the accuracy of the model based on fused spectra had most improved,with the calibration R~2improved in the range of 0.107,the calibration RMSE reduced in the range of 0.605,the validation R~2 improved by 0.104,the validation RMSE reduced by 0.901,and the RPD improved by 0.425.Comparing the accuracy of the three fused models,the model after spectral data fusion had the highest accuracy,followed by the model after spectral index fusion,while the third was the model after spectral model fusion.Therefore,spectral data fusion can be used as the preferred fusion level,and the spectral data fusion model was the best soil salinity inversion model of MSI-UAV fusion.(4)Inversion and accuracy analysis of soil salinity spatial distributionBased on the spectral data fusion model and the MSI image in study area,soil salinity inversion analysis was carried out.The result showed that the soil salinization in Kenli district gradually increased from the southwest(agricultural production)area to the eastern(coastal)area.The area of non-saline soil accounted for 13.821%.The saline soil area accounted for86.179%of the total area,of which solonchak accounted for 18.133%,moderately and severely saline soil accounted for 57.076%,and mildly saline soil accounted for 13.282%.Soil salinization is common in this area,which is consistent with the actual situation.Therefore,for the soil salinity inversion in the coastal area of the Yellow River Delta,spectral data fusion of MSI-UAV image has the best effect on improving model accuracy.This paper compared and analyzed the fusion methods and levels in soil salinity inversion based on satellite and UAV data fusion,determined the appropriate fusion methods and levels,and formed a set of high-precision soil salinity inversion technology based on satellite and UAV multi-source remote sensing fusion.The research results can enrich the theory and technology of soil quantitative remote sensing,and provide data support for regional saline soil treatment,utilization,and sustainable development.