INTEGRATION OF RADAR AND OPTICAL REMOTE SENSING FOR LANDSLIDE DETECTION - A CASE STUDY OF MEERIYABEDDA LANDSLIDE IN SRI LANKA

摘要

Landslides are significant geological hazards that can have destructive effectives on human life and property. Mostly people living in mountainous areas, and their properties, face critical danger from landslide disasters. Landslides are triggered due to unsustainable anthropogenic activities like mining, road cutting, urbanization, as well as natural causes like earthquakes and rainfall. Thus, landslide detection is an essential requirement in pre and post disaster hazard analysis. In earlier studies, landslide detection was often achieved through time consuming, cost intensive field surveys and visual orthophoto interpretation. Recent studies show that Earth Observation (EO) data offer new opportunities for fast, reliable and accurate landslide identification at smaller scales. This can contribute for effective landslide monitoring and hazard management. Sir Lanka is a tropical island comprising a heavy mountainous region in the centre. A combination of geology, unsafe land use practices and heavy rainfall from two monsoons have caused irregular landslides throughout this central hilly region. From the 65,000 km2 of land in Sri Lanka, nearly 20,000 km2 is prone to landslides. Hence, the objective of this work is to investigate the potential of detecting landslide areas from EO data. This research is based on the severe landslide that occurred in Meeriyabedda area in Badulla District on 29~(th) October 2014, affecting around 330 people of 57 families and 63 buildings in Ampitikanda tea estate. Basically, two radar and optical images before and after the event are used in order to delineate the landslide area from different processing techniques. Geometrically registered and radiometrically normalized world view II and geoeye optical images are used to implement the change detection techniques. Satellite image pixel intensity changes are extracted by calculating the Normalized Difference Vegetation Index (NDVI) and Principle Component Analysis (PCA). Further thresholding into classes of changes in connection with landslide activity is performed to discriminate the landslide area from surrounding. Two pre and post sentinel-1 images are pre-processed in order to apply pixel based classifications. Backscatter difference and the correlation coefficient between two images are examined to threshold the image to extract the changed pixels or landslide area from non changed pixels. With the inherent characteristics of radar and optical remote sensing, from the post optical image, part of the landslide area is hindered by clouds even though the analysis offers the detailed information about the rest of landslide area. Due to weather independent capability of radar images, all the landslide regions are detected. However radar suffers serious geometrical distortions specially when studying the high relief terrain areas. Moreover when consider the spatial resolution, radar has some limitations to detect small landslides than optical images can detect. Hence this study aims to compare the detection of Meeriyabedda landslide from different change detection techniques applied to the radar and optical images before and after the event with analysing their inherent limitations for studying specially in a high relief terrain areas.