Water Area Extraction And Dynamic Monitoring Of Inland Great Lakes Based On Spaceborne Synthetic Aperture Radar Images

Great inland lakes are an important part of the water resources on the earth and playing a pivotal role in earth hydrological and biochemical cycle.It is an effective way to study regional ecosystem by monitoring the lake water because the lake water is sensitive to climate changes and human activities.Thus,a reliable technology for monitoring the changing water dynamically is a key stage to realize harmonious coexistence between human and nature,and can also promote the building of socialist ecological civilization.With the rapid development of remote sensing technologies,the satellite remote sensing technology has been widely used to observe the earth in large area quickly and efficiently.So,monitoring the large lakes by satellite remote sensing can provide real-time data support and guidance for farmland irrigation,regional water resources management and rational planning of the water resources allocation,and it is also of great significance for the prediction,early warning and monitoring of flood and waterlogging disasters.Among the remote sensing technology,the spaceborne Synthetic Aperture Radar(SAR)is an active microwave imaging sensor mounted on a satellite platform.The SAR can be widely applied to observe earth in any kind of weather all day long without affections of the cloud,rain and fog.Therefore,it is of great significance to monitor changes of large lakes based on spaceborne SAR images.In past few years,many researchers have studied monitoring water based on SAR images,and proposed many SAR image water extraction methods based on different technology.Such as the threshold-based method,cluster method,Active Contour Model(ACM)based method and so on.However,the above methods are only suitable for simple scenes,in which case some problems in practical application are not considered by above methods.Such as suppression of the SAR speckle noise,elimination of the effects of complex topography around large lakes,low computational efficiency when processing large scale SAR images,and the limitation of the water extraction results by using Middle-to-Low Resolution(MLR)SAR images.Aforementioned problems make the traditional methods are not satisfied with the practical applications.Aiming at the above problems,this dissertation took the perspective of practical application and proposed several novel improvements based on the traditional methods.Two novel water area extractionmethods were proposed,which worked on pixel-level and subpixel-level,respectively.And based on multi-mode real SAR images,all the proposed methods were verified.In addition,based on the water extraction results with SAR images,a dynamic monitoring method for water changes in large lakes was proposed.The Sentinel-1A/B satellites time-series SAR data for the whole year of 2019 were collected for several large freshwater lakes in China.And the changing trends of water area with seasons in these large lakes were analyzed.The main research contents and innovative work of this dissertation can be summarized in the following three aspects:(1)Based on traditional pixel-level water extraction methods,this paper proposed a novel Strategy of Division in Local Regions based ACM(SDLR-ACM)method for SAR images large lake water extraction.The method mainly includes three steps.Firstly,based on cluster algorithm and a region of interest extraction method,a coarse segmentation method was implemented to extract the initial water boundary.And based on the initial water boundary,the target area containing the real water boundary is established.Secondly,a novel method namely Strategy of Division in Local Regions(SDLR)was given to extract SAR image slices that only included local target area.Secondly,in each image slice,a non-local speckle filter and ACM algorithm were used to suppress the SAR speckle and extract the refined water area.The proposed SDLR-ACM method transferred time-consuming operations such as despeckling and refined water extraction to small image slices only containing the local target area,which can greatly reduce the computation coast and improve the weakness of the ACM algorithm that was susceptible to the initial contour.The quantitative analyses results show a good performance with the accuracy over 97% and the average contour offset distance less than 0.7 pixels.Also,the average running time is much less than traditional methods.Therefore,the SDLR-ACM method greatly improves the overall processing efficiency while ensuring the extraction accuracy.(2)In this dissertation,in order to overcome the limitation of the accuracy of water extraction results by the MLR SAR images,the advantages of the super-resolution restoration technology and the pixel-level water extraction method were combined,and proposed a subpixel-level water extraction method.This research mainly includes two parts.Firstly,based on the convolutional neural network model and residual model,a novel SAR image super resolution restoration model was proposed,which namely Lightweight Resnet based Super-resolution Restoration for SAR(LRSR-SAR).And the LRSR-SAR model was testedwith many real SAR images.The experimental results showed that the LRSR-SAR model had a good efficiency for SAR image super resolution restoration.Secondly,the LRSR-SAR model and the SDLR-ACM method were integrated,and a novel subpixel level water extraction method based on local super-resolution restoration was proposed.This subpixel-method level water extraction method utilized the LRSR-ACM model to super resolution restoration the despeckled target area image slices,and achieved purpose of fine segmentation on a sub-pixel level.The experimental results showed that the accuracy of water extraction results was significantly improved after the introduction of the LRSR-SAR model,and the purpose of extracting high-precision water area with MLR-SAR images was achieved.After introduced the LRSR-SAR into SDLR-ACM,the water extraction performance has been significantly improved with the accuracy over 99% and the average contour offset distance less than 0.2 pixels.(3)For the practical application requirements of the large lakes water extraction with spaceborne SAR images.Firstly,this dissertation proposed a method for monitoring the changing water dynamically based on time-series spaceborne SAR images.Then,for a case study,the time-series SAR data and measured water level data of Danjiangkou reservoir in 2017 were used to verify the validity of the monitoring method.Secondly,the above monitoring method was applied to other study sits that included Danjiangkou reservoir,Poyang lake,Dongting lake,Hongze lake and Gaoyou lake.A total of 145 time-series SAR images obtained by Sentinel-1A/B satellites in five large freshwater lakes in 2019.And the water extraction experiments were carried out on the time-series data set.Based on the water extraction results,the changes of water area of five lakes with seasons were analyzed.The monitoring results show that the annual water area of Poyang Lake and Dongting Lake changed greatly with the seasons,and the plentiful phase of the water source occurred during summer,and the exhausted phase of the water source occurred during winter.For another three lakes,the annual water area of Hongze Lake,Gaoyou Lake and Danjiangkou Reservoir remained stable without great fluctuations.