Retrieval Of Surface Soil Moisture In Arid Area Based On Sentinel Data And Neural Network

Soil moisture is an important component of the global water cycle.It is of great significance to study the spatial distribution of soil moisture,crop growth and yield,climate change,and spatial and temporal distribution of water resources.Water resources are scarce in the Qaidam Basin,in recent years in Golmud city surrounding,which is mainly composed of Chinese wolfberry and other economic crops developed several farms.Inversion of soil moisture has important scientific and technological support for agricultural production layout,ecological Environment protection and economic development strategy in this region.Large-scale monitoring has been a challenge because of the area’s size and sparsely populated population.In this study,the vegetation covered ground around Golmud City was taken as the research area,and the data of Sentinel-1 radar and Sentinel-2 optical data in the same period combined with the Water Cloud Model and BP and RBF neural network model were used to verify the results of measured soil moisture,and the inversion study of soil moisture around Golmud City was carried out.The main conclusions of this paper are as follows:(1)In the vegetation covered area,We need to consider the influence of vegetation layer on the radar back-scattering signal.It was found that the vegetation water content retrieved by NDWI index was used as the input parameter of the Water Cloud Model,with the purpose of removing the influence of surface vegetation,and a more real soil back-scattering coefficient could be obtained.According to the comparison of radar back-scattering coefficients before and after the removal of vegetation coverage,the VV polarization attenuation value obtained by NDVI index is about 0.01-2.6d B,and the average influence value of vegetation layer on radar VV polarization back-scattering coefficient is 0.39d B.The VH polarization is reduced by0.07-4.06 d B,and the average effect of vegetation layer on the back-scattering coefficient of VH polarization is 0.61 d B.The range of VV polarization attenuation value obtained by NDWI index is about 0.8-3.7d B.The average influence of vegetation layer on the radar VV polarization back-scattering coefficient is 1.58d B,and the VH polarization is reduced by 1.2-4.5d B.The average influence of vegetation layer on the radar VH polarization back-scattering coefficient is 2.57d B;It is known that NDWI index is more suitable as the input parameter of Water Cloud Model.(2)Compared with BP neural network model,RBF neural network model has more obvious advantages in dealing with the complex nonlinear relation Ship among soil back-scattering coefficient,radar incidence angle and soil moisture,and is more suitable for the study of soil moisture inversion.In this study,the soil back-scattering coefficient after removing the influence of vegetation was introduced into the neural network model to construct the soil moisture inversion model which combined the measured soil moisture,VV,VH and radar incidence angle.The verification results show that:The correlation R~2between the predicted water content and the actual water content of the BP neural network model was 0.6159,and the root mean square error was RMSE=8.69;The correlation R~2between the predicted water content and the actual water content of the RBF neural network model was 0.6501,and the root mean square error was RMSE=3.61.The correlation between RBF neural network and BP neural network model is increased by 0.0342,and the root mean square error(RMSE)is reduced by 5.38.It shows that RBF neural network is more suitable for soil moisture inversion in this study area.By comparing the soil water distribution map of the study area obtained by final inversion with the actual ground conditions around Golmud,the water content of Dongxi Farm in Golmud is between 16%-24%.The area with the highest soil water content is concentrated in the area of groundwater overflow,the places with the least soil moisture are located on the right side of Hexi Farm and on both sides of the road from Golmud City to Meishan Factory,which are consistent with the actual situation.(3)Correlation analysis was conducted on the inversion soil moisture distribution map by using the NDVI index map of the study area and the all-weather0.01°soil moisture product of China,The results showed that the correlation index(R~2)between soil moisture and NDVI index in the study area was 0.5549,and the correlation between soil moisture and 0.01°soil moisture of China was 0.5713,both of which had a certain consistency with the spatial distribution of soil moisture.Therefore,it is proved that the soil moisture inversion based on Sentinel-1 SAR data and Sentinel-2 optical data by using Water Cloud Model and RBF neural network has high accuracy,and it is feasible to use this method to carry out large-scale dynamic monitoring of soil moisture in the vegetation-covered surface of the study area.