Towards an intelligent and adapted small scale landslides Monitoring system in East Africa: Shyira Landslide Monitoring Using Sentinel - 1 SAR data On Google Earth Engine Cloud Computing

Abstract

Rainfall-induced landslides pose significant threats in the North-Western provinces of Rwanda, contributing to major disasters. This paper addresses technological challenges in disaster response, specifically focusing on soil displacement quantification. The study concentrates on the local scale around Mukungwa River, employing remote sensing techniques and community science. The methodology involves using INSAR polarization and phase measurements, with a focus on Shyira Landslide Monitoring using Google Earth Engine Cloud Computing. A citizen science approach was integrated into the study's framework. The methodology for landslide detection included the careful selection of an Area of Interest (AOI) and distinct time periods Before Event (BEvent) and After Event (AEvent) corresponding to the landslide occurrence. For a comprehensive representation of ground surface properties before and after the landslide event, SAR image stacks were generated. These stacks, calculated as temporal medians of SAR data, were constructed for ascending data, descending data, and a combination of both. Landslide detection involved assessing changes in the backscatter coefficient using the standard SAR intensity log ratio approach. The classification process categorized changes into three classes: stable, subsidence/decrease, and increase/uplift. To gain deeper insights, a CSV file was generated for statistical analysis, providing a quantitative examination of the landslide event dynamics. The study conducted comprehensive statistical analysis and derived meaningful recommendations. This research contributes to the understanding of landslide monitoring through a robust methodology that combines remote sensing technologies, community engagement, and statistical analysis. In a sample of data containing 885,632 points generated from imagery within the study area, using Python script code. This dataset was utilized to create a pair plot, effectively visualizing the before-event and after-event data and these have presented a strong correlation. In a sample of the 5,000 households in the study area, 96 were completely destroyed and 231 heavily damaged .The findings and recommendations have implications for disaster response strategies and underscore the importance of technological advancements in addressing the challenges posed by landslides.