Soil Moisture Retrieval Model And Method Based On Multi-source Remote Sensing Data

As an important physical quantity reflecting the surface conditions,soil moisture can not only reflect the drought situation of the land,but also the decisive factor of crop water supply.It is regarded as the research basis and focus of attention in many fields such as precision agriculture and water cycle.Therefore,monitoring surface soil moisture in a large area and obtaining real-time and accurate soil moisture information has important scientific value and practical significance.Compared with traditional soil moisture measurement methods,optical remote sensing and microwave remote sensing methods for soil moisture retrieval have the advantages of large-area real-time observation,high efficiency,and low cost,and are widely used in regional and even global soil moisture monitoring.However,the existing inversion models and algorithms generally have low accuracy and reliability,and cannot meet the universal monitoring application requirements of different soil environments in alpine frozen soil,arid area salt desert and temperate vegetation coverage area,especially seasonal The monitoring of soil thawing changes in permafrost regions is the most challenging.Therefore,based on the systematic research of the theory of soil moisture retrieval by optical remote sensing,this paper proposes a universal optical remote sensing soil moisture retrieval data fusion model and algorithm to overcome the accuracy and reliability limitations of only relying on a single optical model for retrieval.Focusing on solving the problem of low sensitivity of soil thawing process monitoring in seasonally frozen soil areas.This paper uses the measured soil moisture time series data in the third Qinghai-Tibet Plateau atmospheric science experiment as data support,and introduces the one-year time series results of the more widely used ATI,PDI and TVDI three single optical models inversion into the improved entropy.The value method measures the uncertainty and evaluates the overall accuracy of each inversion model over a one-year period.Based on this,the weight is reversed,and the three inversion results are weighted and fused to obtain the optimal value,in order to improve the accuracy and reliability of the inversion.This method was used to monitor the seasonal freezing and thawing process in the frozen soil area in Nagqu area.The experimental results show that:(1)The correlation between the inversion results of ATI,PDI and TVDI models and the true value is 0.576,0.647,0.724,respectively.The entropy method has the highest accuracy and the correlation reaches 0.865.(2)Rainfall will cause significant fluctuations in the daily minimum temperature,which will seriously affect the accuracy of traditionally relying on the surface temperature to describe the time window of the freeze-thaw process.(3)The soil moisture inversion results of the entropy fusion method have good trend consistency with the traditional temperature mapping freeze-thaw process,which can effectively monitor the seasonal frozen soil thawing process.This method will provide an effective technical means for research in the field of monitoring the freezing and thawing process of seasonal frozen soil.On the other hand,deep learning-based microwave remote sensing soil moisture inversion models and methods are in-depth research,and a new type of highly reliable and high-precision soil moisture inversion model is proposed for multi-polarized SAR and multiple vegetation indices for wide-area collaborative monitoring applications And algorithm,it has stronger universality and robustness in the application of soil moisture inversion in wide-area temperate zone with complex surface vegetation cover.Considering that the vegetation index for estimating the influence of vegetation is composed of red light and near-infrared bands,and multi-polarization SAR can provide soil roughness information,this article uses Convolutional Neural Network(CNN)to effectively simulate relevant The advantage of feature parameters of physical meaning,combined with SAR image and multispectral data data source,proposes a method to implement wide-area soil moisture inversion combined with an improved convolutional neural network.This method uses 1×1,2×1,3×1 size convolution kernels to perform one-dimensional convolution operations on VV,VH polarized SAR images and original red light and near-infrared band data,and adaptive extraction can reflect the measurement area The high-level feature dimension of the spatial-temporal difference of soil moisture realizes the collaborative inversion of wide-area soil moisture with multi-polarization SAR and multi-spectral data.In addition,in order to avoid the reduction of inversion accuracy caused by the reduction of feature information,and to ensure that all the extracted feature information participates in the inversion,the pooling layer in the traditional CNN is removed from the model.Finally,an experimental area located in the middle of the Sichuan Basin was selected to carry out the model accuracy and applicability verification research.The experimental results showed that:(1)In a wide area with a side length of more than 100 km,the improved CNN model inversion results are comparable to those provided by the China Meteorological Administration.The correlation of the precision soil moisture data reaches 0.934,the error is random deviation,and the RMSE is only 1.45%.(2)Compared with the traditional artificial neural network(Artificial Neural Network,ANN)method,the inversion results of the method in this paper are all the best.The method has higher inversion accuracy,universal applicability and wide area.The generalization ability has stronger universality and robustness in the application of soil moisture retrieval in wide-area temperate zones with complex surface vegetation coverage,and can provide certain technical support for precision agriculture,drought and flood disasters and related research.