2025 CSDMS meeting-053

Edit registration

Pay registration

(if you haven't already)

Log in (or create account for non-CSDMS members)
Forgot username? Search or email:CSDMSweb@colorado.edu

Browse  abstracts

<<

>>

Back to meeting

Advancing dynamic multi-(hazard-) risk assessment method for a massive refugee camp in Bangladesh

Dewan Mohammad Enamul Haque, Louisiana State University Baton Rouge Louisiana, United States. dhaque1@lsu.edu

Suniti Karunatillake, Louisiana State University Baton Rogue Louisiana, United States. sunitiw@lsu.edu

Juan M Lorenzo, Louisiana State University Baton Rogue Louisiana, United States. gllore@lsu.edu

We are advancing dynamic multi-hazard risk assessment (MHRA) methods for human ecology, using the Kutupalong Rohingya Refugee Camp (KTP) in southeastern Bangladesh as a case study. KTP, home to over 1.1 million refugees within a 15 km² area, represents one of the world’s most densely populated and hazard-prone humanitarian settlements. This research investigates hydro-meteorological risks—primarily shallow landslides and flash floods—before and after refugee settlement, with a focus on landscape changes driven by both anthropogenic and natural processes. We formulated two core hypotheses. The first posits that dynamic hazard modeling, incorporating both geological and anthropogenic factors, more accurately captures the cascading effects of landslides and flash floods than traditional static models. The second hypothesis suggests that hydro-meteorological risk at KRC has declined due to the incremental implementation of slope stabilization and restoration measures.

We began with a landslide hazard assessment using a sloped unit (SU)-based approach, building upon previous grid-based models employed at KTP. Our dynamic, time-lapse assessment, which examines pre- and post-refugee influx scenarios, identifies slope units with increasing, decreasing, or unchanged susceptibility over time. A Generalized Additive Model (GAM) applied at the SU scale outperforms conventional machine learning (ML) methods, providing a robust framework for surface hazard modeling. In parallel, we evaluated landscape degradation and recovery through above-ground biomass (AGB) estimation using Sentinel-2A imagery, NASA GEDI LiDAR, and ESA Biomass products. We estimated AGB for 2017 (pre-influx), 2019 (early restoration), and 2023 (ongoing recovery) using Random Forest, SVM, and XGBoost regression models. This integration of remote sensing and ML demonstrates the utility of multi-source data for tracking dynamic land-use change.

Further fieldwork is required to collect more geotechnical soil samples and detailed information on the geometry of the failure plane in selected large landslides. This will enable us to assess the interaction between slope-forming materials and the underlying bedrock interface, as well as model the velocity and volume of sliding materials in the form of run-out. Like landslides, we will dynamically simulate flash flood inundation to extract critical hydrodynamic parameters, including peak flow height, flow velocity, discharge, and flood arrival time, particularly for the 2017 and 2021 monsoon events at KTP. Multi-temporal DEM generation and land cover mapping will be the key in this regard.

A key contribution of this research lies in the integration of landslide and flash flood risk data to assess their cascading impacts on human ecology. This integrated risk information will be combined with engineering measures and economic modeling to assess the effectiveness and feasibility of the existing mitigation measures. Risk estimation will be conducted under changing hazard scenarios, comparing conditions immediately before the major refugee influx (2018 and earlier) with those in the post-intervention period (2022–2023). A similar modeling framework will also be applied to explore potential future hazard scenarios under evolving landscape and climate conditions.