GRSG Conference 2022: Orbit to Outcrop
Title: Analysing the posterior predictive capability of landslide susceptibility maps
Author: Tanuj Pareek
Landslide susceptibility maps serve as the basis for hazard and risk assessment, as well as risk-informed landuse planning at various spatial scales. These maps are intended for a variety of purposes, including infrastructure planning and restrictive landuse zoning, by potential end-users such as spatial decision-makers and urban planners. These applications require accurate map information and specific map legends, as decisions based on these maps have the potential to cost lives and cause infrastructure damage.
The usability of the maps depends on whether they provide the required information and whether that information is accurate enough to be utilised for the intended purpose. Therefore, assessing the usability and predictive accuracy of landslide susceptibility maps is of paramount importance. Typically, the accuracy of the maps is evaluated using the same landslide inventory that was used to create the map, which does not actually test the predictive ability of the maps in future situations.
To address these issues, we evaluated three landslide susceptibility maps for an area in Kerala (India) that were generated in the past years by utilising a new landslide inventory created after the maps were generated. This research presents a method for evaluating classified maps intended for use in decision-making and planning.
We assessed (1) the usability of the landslide susceptibility maps by conducting a literature analysis and conducting interviews with the map producers and users in Kerala. The assessment indicated the requirements for a map to be utilised for the intended purpose. We (2) generated a robust (new) landslide inventory using a MsaU-Net deep learning (DL) model, which was (3) used to evaluate the landslide susceptibility maps generated in the past years. We designed a method for evaluating classified maps, with a focus on evaluating and comparing in different scenarios.
A major accomplishment of the research was to generate Unique Conditions Units (UCUs), which were utilised to evaluate classified maps. We propose that these units can also be used to generate landslide susceptibility maps and provide a reasonable topographic representation. Our study has huge significance, particularly in (1) investigating the usability of landslide susceptibility maps and attempting to direct the focus of map producers toward more user-oriented landslide susceptibility mapping, (2) generating landslide inventory of small-sized landslides utilizing open source datasets, (3) designing a method to assess the classified landslide susceptibility maps in multiple evaluation scenarios, and (4) providing a method to generate Unique Conditions Units (UCUs) for evaluation as well as mapping purposes, (5) highlighting the challenges of analysing the importance of landuse and landcover changes on the validity of the landslide susceptibility maps.
We conclude that, although the volume of literature on the best methods for landslide susceptibility assessment is enormous, there is an urgent need to focus more on the forward predictive capability and usability by end-users.