Landslide Identification And Monitoring With The Fusion Of SAR And Optical Remote Sensing

As one of the geological disasters that caused huge economic losses and casualties,landslides have increasingly attracted the attention of the international community for their impact on social development.China has the largest population and ranks the third land area in the world.Complex terrain has caused frequent occurrence of geological disasters,of which landslides account for more than 2/3.Landslides cause economic losses more than 100 billion and kill more than one thousand people each year.Therefore,the prevention and harness of landslide geological disasters has become a national strategy.With the development of remote sensing and computer technology,the methods for landslide detection and monitoring are improved rapidly.At present,the optical remote sensing technology and Interferometric Synthetic Aperture Radar(In SAR)are two main methods for landslide identification.The spatial resolution of optical remote sensing satellites has been improved to 0.31 m,and the accuracy of radar satellites in detecting surface deformation can reach centimeter or even millimeter levels.Both methods have and unique superiority and limitation in landslide extraction.In order to identify potential landslides more completely and automatically,multi-source data fusion has always been a research hotspot for landslide detection and monitoring.In this paper,two methods for potential landslides identification by integrating SAR images,optical remote sensing imagery and high-resolution digital elevation model(DEM)are proposed.These two methods are implemented at the Wudongde Hydropower Station in Yunnan and the Heifangtai loess terrace in Gansu,respectively.The main work and results include:(1)Analysis the procedure of object-oriented segmentation and classification for landslide detection,and discuss the details of multiresolution segmentation,feature selection,classification and accuracy evaluation.(2)Analysis the principle and procedure of landslide extraction by two-dimensional continuous wavelet transform(2D CWT),and study the selection of feature wavelet scale,calculation of spectral power and translation between an array and landslide map.(3)Automatic extraction of potential landslides method by integrating optical imagery with In SAR-derived deformation map in Wudongde Hydropower Station is executed with Salford Predictive Modeler(SPM)and e Cognition software.(1)At first,deformation map is generated with interferometric point target(IPTA)technology based on ALOS/PALSAR images.Then,initial potential landslide area is extracted by using classification rules established with classification and regression tree(CART)model,in which considering the imagery features,such as spectrum,shape,and texture,and the topographic features like elevation,slope,aspect,hillshade and so on.Next,fuzzy classification is employed to extract omissive landslide area with In SAR-derived deformation rate.Finally,the final potential landslide area is identified by integrating the initial potential landslide area and the omissive area.(2)The proposed method is also verified in large scale catchment of Jinsha River with Sentinel-1A and Sentinel-2A images.The potential landslide area identified by integrating the optical imagery with In SAR-derived deformation map has improved significantly comparing the result extracted by optical data,especially for the complex surface environment.In a conclusion,the proposed fusion methods could increase the accuracy and completeness of potential landslide identification,improve the efficiency of geological disaster investigation and provide valuable information for landslide monitoring and prevention in unknown regions.(4)Automatic extraction of potential landslides method by integrating DEM,optical imagery and In SAR-derived deformation in Heifangtai loess terrace is carried out with programming and e Cognition software.Terra SAR-X images,World View-02 imagery and unmanned aerial vehicle DEM are selected in this part.On the one hand,suspected landslides are extracted by using 2D CWT with high-resolution DEM;on the other hand,suspected landslides are identified by using OBIA method with optical imagery and deformation data.The potential landslide area is extracted by calculating the intersection of two suspected landslide area.The result shows a strong consistency with the landslides inventory mapping,the correctness reaches 97.21%.There is no requirement of rule sets in the proposed method,and the method has high automation and are more realistic results.