Khanam et al. (2025) Predictive understanding of socioeconomic flood impact in data-scarce regions based on channel properties and storm characteristics: application in High Mountain Asia (HMA)

Identification

• Journal: Natural hazards and earth system sciences
• Year: 2025
• Date: 2025-10-06
• Authors: Mariam Khanam, Giulia Sofia, Wilmalis Rodriguez, Efthymios I. Nikolopoulos, Binghao Lu, Dongjin Song, Emmanouil N. Anagnostou
• DOI: 10.5194/nhess-25-3759-2025

Research Groups

• Civil & Environmental Engineering, University of Connecticut, Storrs, CT, USA
• Water Resources Science and Engineering, Oak Ridge National Laboratory, Oak Ridge, TN, USA
• Civil & Environmental Engineering, Rutgers University, Piscataway, NJ, USA
• Computer Science and Engineering, University of Connecticut, Storrs, CT, USA

Short Summary

This study introduces a novel geomorphologically guided machine learning method to predict socioeconomic flood impacts in data-scarce regions. Applied to High Mountain Asia (HMA), the model effectively identifies flood susceptibility hotspots and their evolution from 1980 to 2020, demonstrating its versatility for ungauged areas.

Objective

• To develop and apply a geomorphologically guided machine learning (ML) method for mapping spatially distributed socioeconomic flood effects across High Mountain Asia (HMA), leveraging available global datasets to overcome data scarcity challenges.
• To analyze the changes in socioeconomic flood impacts in HMA spanning 1980 to 2020.

Study Configuration

• Spatial Scale: High Mountain Asia (HMA), encompassing approximately 6000 watersheds. Initial training and validation focused on Nepal at both district and watershed scales.
• Temporal Scale: 1980 to 2020 (40 years), with analyses conducted in 5-year intervals.

Methodology and Data

• Models used:
  • XGBoost (eXtreme Gradient Boosting) for machine learning classification.
  • Life Year Index (LYI) for quantifying socioeconomic flood impact.
  • Flood Geomorphic Potential (FGP), a DEM-derived geomorphic index for floodplain delineation.
  • Concentration Index (CI) for characterizing rainfall temporal variability.
• Data sources:
  • Socioeconomic Flood Impacts: Over 6000 flood events (1980-2020) with corresponding 5- and 10-year LYI values. LYI calculated using mortality, affected people, financial damage (from Nepal Disaster Risk Reduction Portal), average life expectancy, median age (from WHO), and income per capita (from The World Bank).
  • Population Density: Nepal's national census for training; Gridded Population of the World (GPW), v4 | SEDAC (2000, 2005, 2010, 2015, 2020) for HMA.
  • Geomorphic Data: NASA National Snow and Ice Data Center (NSIDC DAAC) 8 m digital elevation models (DEMs) for Nepal; Copernicus 30 m DEM (European Space Agency, 2021) for HMA.
  • Rainfall Characteristics: ERA5 hourly rainfall data (1980-2019) for calculating the 5-year Concentration Index (CI).
  • Validation Data: Dartmouth Flood Observatory (DFO) Global Active Archive of Large Flood Events, 1985–Present.

Main Results

• The geomorphologically guided ML model successfully identified flood susceptibility hotspots and their temporal evolution across HMA from 1980 to 2020.
• The model achieved a test accuracy of 63% for Nepal, with a significantly improved recall of 71%, precision of 73%, and F1 score of 72% for the "high" impact category after hyperparameter tuning.
• Population density was identified as the most important variable for predicting socioeconomic flood impact, followed by the climate concentration index.
• Hotspots of high socioeconomic flood impacts (Life Year Index (LYI) > 1000 years) were consistently predicted in regions with higher population exposure.
• Over the entire study period, approximately 57% of watersheds were predicted to have low LYI (1-100 years), 35.9% medium (100-1000 years), and 6% high (>1000 years).
• Watersheds predicted as high risk (LYI > 1000 years per 100,000 people), despite representing only 6% of the total, had a conditional probability of 40% and 10% of experiencing events recorded by the DFO with high severity (e.g., > 1 million people affected).
• An increase in flood vulnerability was observed over time, with the most significant shifts from lower to higher risk categories occurring in 1990-1995 and 2010-2015, particularly in Nepal and China, often correlated with population growth in flood-prone areas.

Contributions

• Introduces a novel, streamlined, and versatile geomorphologically guided machine learning framework for preliminary, large-scale socioeconomic flood vulnerability assessment in data-scarce, ungauged regions.
• Quantifies socioeconomic flood susceptibility using the Life Year Index (LYI) by integrating remotely sensed climate variables, high-resolution terrain information (Flood Geomorphic Potential), and population data.
• Demonstrates the model's effectiveness in identifying flood susceptibility hotspots and their temporal evolution (1980-2020) across the entire High Mountain Asia region.
• Provides a valuable, cost-effective decision-making tool for stakeholders in densely populated, rapidly changing climates, requiring only a small number of input variables.

Funding

• NASA High Mountain Asia program (grant no. 80NSSC20K1300)

Citation

@article{Khanam2025Predictive,
  author = {Khanam, Mariam and Sofia, Giulia and Rodriguez, Wilmalis and Nikolopoulos, Efthymios I. and Lu, Binghao and Song, Dongjin and Anagnostou, Emmanouil N.},
  title = {Predictive understanding of socioeconomic flood impact in data-scarce regions based on channel properties and storm characteristics: application in High Mountain Asia (HMA)},
  journal = {Natural hazards and earth system sciences},
  year = {2025},
  doi = {10.5194/nhess-25-3759-2025},
  url = {https://doi.org/10.5194/nhess-25-3759-2025}
}