University of Twente Student Theses

Login

Investigating hurricane driven landslides: from physically based to statistically based and from space to space-time

Senthil Nathan, Dayallini (2020) Investigating hurricane driven landslides: from physically based to statistically based and from space to space-time.

Link to full-text:https://ezproxy2.utwente.nl/login?url=https://library.itc.utwente.nl/login/2020/msc/nhr/senthilnathan.pdf
(only accessible for UT students and staff)
Abstract:Hazards are generally defined by three components: Where they occurred? When have they occurred? And How destructive were they? This study focuses on these components for the event of landslides. Initially, where the landslides have occurred is analysed, this concept is commonly termed as landslide susceptibility. Over the years, there had been several techniques via which the susceptibility is estimated. This study specifically researches the quantitative methods namely: statistical and physically-based models. While the physically-based is process driven, the statistical model is data driven. Statistical framework aids in a physical model for the interpolation and parameterisation of the physical parameters but an incorporation of these physical parameters into a statistical model is hardly ever done. Thus, this study aims to visualise the difference in the spatial patterns obtained from statistical analysis done on the different parametric datasets: i) the traditional parameters used in statistical framework for landslide susceptibility; ii) physical parameters which explains the slope instabilities and iii) combination of both in the above-mentioned. In this study the physically-based model is not carried out rather the inputs and the outputs of a previously executed model is utilised. Thereby with this regard, this initial study is done on the region of Grand Bay at Dominica. For this a Generalised Linear Model (GLM) with binomial probability distribution integrated with the Least Absolute Shrinkage and Selection Operator (LASSO) as variable selector is implemented. The traditional parameters were more adept in capturing the spatial characteristics of the landslide susceptibility, this was because of the increased spatial variability of the conditioning factors. The next phase of the study focuses on “When” the landslides have occurred. An attempt to examine whether a statistical framework is capable of recognising a spatial pattern of the temporal dependency on landslide occurrences, is undertaken. In order to scrutinise this, a generalised additive model (GAM) with its temporal counterpart and considering non-linear parameters is executed. Specifically, an autoregressive model acts upon this GAM in order to speculate on capturing a temporal latency effect on the landslide susceptibility. While the previous study was done at pixel level, this was carried out at slope unit level for the whole island of Dominica and the available five landslide inventories for the region was utilised. For this tropical region, no significant temporal latency on susceptibility was observed. This might be due to the fact that there is a spatial variability of the triggering events over the period of years, thus the model is dominated by the spatial trends rather than temporal ones. The final phase of this study was on “How”, and, explicitly approached to model a specific characteristic of the landslide, the percentage of landslide area per mapping unit. A GLM with Gaussian probability distribution was implemented for the whole island for the five time periods. These models were also executed at the slope unit level and though they reflected on the increase/ decrease of the percentages adequately, they were unable to efficiently capture the variance. This was due to the data inadequacy in terms of sample size (a small dimension of input dataset) and spatial invariability (a uniformity in the characteristics of the covariates used). While this study focuses on the components individually, a more established future research would be on ways to integrate all the three components and facilitate the learning ability of the statistical framework which in turn would increase the performance of the susceptibility model.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/85083
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page