Abdel-Mooty et al. (2025) A scalable data driven geospatial framework for climate risk assessment

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-20
• Authors: Moustafa Naiem Abdel-Mooty, Paulin Coulibaly, Wael El‐Dakhakhni
• DOI: 10.1038/s41598-025-32370-7

Research Groups

• Department of Civil Engineering, McMaster University, Hamilton, ON, Canada
• NSERC FloodNet, Department of Civil Engineering, McMaster University, Hamilton, ON, Canada
• School of Computational Science and Engineering, McMaster University, Hamilton, ON, Canada
• RESILIOCS Intelligence, Hamilton, Canada

Short Summary

This study introduces a scalable, data-driven geospatial framework that integrates machine learning and geospatial analysis to dynamically assess climate risks. Applied in Texas, the framework projects a 14% increase in community vulnerability and a 28% rise in economic damages ($1.8 billion per decade by 2050) under the RCP 8.5 emission scenario, emphasizing the urgent need for global climate action.

Objective

• To design a method for integrating geospatial and climate data into a decision-support system for climate risk assessment.
• To test the effectiveness of this framework in a real-world case study (Texas).
• To assess the framework's potential to support climate resilience planning at various levels of governance.

Study Configuration

• Spatial Scale: United States mainland for initial resilience categorization; case study focused on 45 counties within the state of Texas. Data resolution is at the county level.
• Temporal Scale: Historical data from 1996 to 2020; climate projections and analysis extending from 2020 to 2050.

Methodology and Data

• Models used:
  • Unsupervised machine learning (clustering techniques) for community classification.
  • Supervised machine learning models: Bagged Decision Tree (selected as optimum), Gradient-boosted Random Forest, Decision Tree with bagging.
  • Climate models: Bias Correction with Spatial Disaggregation (BCSD) Downscaled Coupled Model Intercomparison Project—Phase 5 (CMIP5) ensemble (16 Global Climate Models).
• Data sources:
  • Historical disaster data records (1996–2020) from the National Weather Service (NWS).
  • Historical climate data from the Global Historical Climatology Network (GHCN-Daily) of the National Center for Environmental Information.
  • Bias-corrected CMIP5 climate projections under Representative Concentration Pathway (RCP) 6.0 and RCP 8.5 emission scenarios.
  • FEMA and NOAA datasets for estimating inflation-adjusted economic damages.
  • Input variables for ML models: monthly average maximum surface air temperature (°C), monthly average minimum surface air temperature (°C), monthly mean wind speed (m/s), average monthly runoff (mm), and monthly precipitation (mm).

Main Results

• Under the RCP 8.5 emission scenario, community vulnerability in the studied Texas counties is projected to increase by 14% by 2050.
• This vulnerability increase is estimated to lead to a 28% rise in economic damages, amounting to $1.8 billion per decade by 2050, alongside heightened socio-economic disruptions like displacement and infrastructure failures.
• Maximum and minimum temperatures are identified as the most influential climatological factors impacting community resilience.
• Flood risk significantly increases when maximum temperatures are between 30 °C and 40 °C, and minimum temperatures are between 15 °C and 30 °C.
• The influence of total precipitation on vulnerability plateaus beyond a 200 mm threshold, suggesting a saturation effect where community infrastructure reaches its capacity.
• The framework provides interpretable insights into ML model behavior, aiding decision-makers in understanding variable influences and interdependencies.

Contributions

• Develops a scalable, data-driven geospatial framework that integrates machine learning and geospatial analysis for dynamic climate risk assessment, offering an alternative to computationally intensive physics-based models.
• Projects the effects of climate change on community resilience under multiple emission scenarios, explicitly incorporating resilience attributes (robustness, rapidity) alongside community vulnerability and hazard exposure.
• Provides a system-wide perspective for assessing cascading climate risks and anticipating vulnerabilities, distinguishing it from localized Digital Twin applications.
• Synthesizes diverse datasets (historical climate records, land-use patterns, hazard exposure, resilience indicators) into a scalable, spatially explicit model for comprehensive resilience planning.
• Enhances model transparency through interpretability techniques (correlation matrices, variable importance, partial dependence plots), enabling decision-makers to derive reliable managerial insights.

Funding

• Vanier Canada Graduate Scholarship (Vanier-CGS)
• Natural Science and Engineering Research Council (NSERC) through the CaNRisk—Collaborative Research and Training Experience (CREATE) program
• INViSiONLab
• INTERFACE Institute at McMaster University

Citation

@article{AbdelMooty2025scalable,
  author = {Abdel-Mooty, Moustafa Naiem and Coulibaly, Paulin and El‐Dakhakhni, Wael},
  title = {A scalable data driven geospatial framework for climate risk assessment},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-32370-7},
  url = {https://doi.org/10.1038/s41598-025-32370-7}
}