RICRAF — the Road Infrastructure Climate Risk Assessment Framework — is an open-source, reproducible workflow for assessing climate-related risks to road networks under current and future global warming levels. It integrates geospatial data fusion, machine learning, and explainable AI to support data-driven resilience planning.
RICRAF comprises three research and code stages, each linked to a corresponding publication:
| Stage | Focus | Journal | Status |
|---|---|---|---|
| 1. Data Fusion | Creation of the fused geospatial dataset linking road and climate hazard data | Scientific Data | Under review |
| 2. Model Development | Development of the XGBoost–SHAP framework and Climate Risk Formula | Climate Services | Under review |
| 3. Model Application | Application of the framework under multiple Global Warming Levels (GWLs) and traffic scenarios | Transportation Research Part D: Transport and Environment | Under review |
RICRAF operationalises the IPCC AR5/AR6 Hazard–Exposure–Vulnerability concept through a transparent, reproducible, and transferable Python workflow.
The framework:
- Links road infrastructure (traffic, road surface, geometry) with climate hazards (heat, rainfall, drought, frost).
- Develops a machine learning-based Climate Risk Formula (CRF) using SHAP-derived weights.
- Produces Climate Risk Scores (CRiskS) at road link scale to support adaptation and resilience investment planning.
- Adheres to FAIR principles — Findable, Accessible, Interoperable, Reusable — through open data, open code, and clear metadata.
The data fusion workflow integrates:
- Road datasets from DataVic — including traffic volume, road surface condition, and road geometry.
- Climate hazard indices from the Australian Climate Service — including precipitation, heat, frost, and drought indicators derived from CMIP6 downscaled models (BARPA & CCAM).
Processing steps include Coordinate Reference System (CRS) standardisation, schema harmonisation, topology checks, spatial joins, and quality validation. The resulting dataset links 7,579 road segments with climate hazard metrics across Global Warming Levels (GWLs of 1.2°C, 1.5°C, 2.0°C, 3.0°C).
Outputs:
data/processed/gdf_road_clim_cln_final_withfuture.geojsondata/processed/variables.csvdata/processed/metadata.json
📘 Reference:
Chin, T.K., Prakash, M., Zheng, N., & Pauwels, V.R.N. (2025). Fused Geospatial Dataset Linking Climate Hazards and Road Infrastructure for Victoria, Australia.
Scientific Data (under review).
This stage develops the XGBoost–SHAP framework for multi-stressor climate risk assessment, integrating:
- Extreme Gradient Boosting (XGBoost) for predictive modelling of road surface distress (roughness, rutting, cracking).
- SHapley Additive exPlanations (SHAP) for interpretable feature attribution and driver importance quantification.
- A Climate Risk Formula (CRF) that computes Climate Risk Scores (CRiskS) using SHAP-derived weights within the IPCC risk structure.
Key Contributions:
- Unified modelling of climate, traffic, and vulnerability stressors.
- SHAP-derived weighting of hazard, exposure, and vulnerability components.
- Network-wide and link-level interpretability for actionable risk mapping.
- Python implementation consistent with national climate services and adaptation frameworks.
Outputs:
- Trained ML models for each distress type.
- SHAP-based driver contribution plots.
- CRiskS maps for current warming (GWL of 1.2°C).
📘 Reference:
Chin, T.K., Prakash, M., Zheng, N., & Pauwels, V.R.N. (2025). Explainable AI for Multi-Stressor Climate Risk Assessment of Road Networks: An XGBoost–SHAP Framework.
Climate Services (under review).
This stage applies the multi-stressor climate risk framework to quantify spatial risk patterns across 23,117 km of Victoria's road network under current climate conditions and projected warming levels up to 3.0°C. It attributes total risk to hazard, exposure, and vulnerability using interpretable Aumann–Shapley decomposition methods, assesses robustness through sensitivity analysis, and identifies emerging network hotspots where interacting stressors amplify climate risk.
Key Contributions:
- Quantification of risk evolution: Currently, 6.4% of the network exhibits high or extreme risk, driven primarily by vulnerability and hazard. Under 3.0°C warming, high/extreme risk expands to 24.4%, with hazard’s contribution increasing.
- Inclusion of traffic growth scenarios, showing risk exceeding 55% in metropolitan corridors.
- Sensitivity tests confirming stable results under ±20% parameter variation.
- Demonstration of framework scalability and transferability for transport resilience planning through modular, open-source design.
Outputs:
- Climate Risk Scores GeoJSON for multiple scenarios and warming levels.
- Confusion matrices, statistical evaluations, and sensitivity analyses.
- Static maps (PNG) and interactive HTML maps (Kepler.gl) for risk visualisation.
📘 Reference:
Chin, T.K., Prakash, M., Zheng, N., & Pauwels, V.R.N. (2025). Climate Risk Assessment of Road Infrastructure under Multi-Stressor Conditions in Victoria, Australia.
Transportation Research Part D: Transport and Environment (under review).
ricraf/
├── README.md
├── requirements.txt
├── CITATION.cff
├── LICENSE
├── data/
│ ├── raw/
│ ├── interim/
│ └── processed/
├── notebooks/
│ ├── ricraf_data_fusion.ipynb
│ ├── ricraf_development.ipynb
│ └── ricraf_application.ipynb
├── outputs/
└── src/
├── ricraf_data_fusion.py
└── ricraf_dev.py
git clone https://github.com/teckkean/ricraf.git
cd ricraf
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txtDownload the raw data from Zenodo (DOI 10.5281/zenodo.17379391) and execute the Jupyter notebook:
jupyter notebook notebooks/ricraf_data_fusion.ipynbRaw data sources:
- Road data: DataVic
- Climate data: Australian Climate Service
Run the model training and explainability workflow:
jupyter notebook notebooks/ricraf_development.ipynbThis notebook reproduces the XGBoost–SHAP pipeline, generates driver attribution plots, and computes Climate Risk Scores (CRiskS).
Outputs are saved to:
data/processed/model_dev/out_crs/Run the application workflow to compute risk scores, generate confusion matrices, and produce maps:
jupyter notebook notebooks/ricraf_application.ipynbThis notebook applies the framework to current and future scenarios, performs statistical evaluations, and saves outputs to
data/processed/model_app/out_cra/| Aspect | Details |
|---|---|
| Spatial Coverage | Victoria, Australia |
| Temporal Coverage | Road data (2020 baseline); Climate projections for GWLs 1.2°C (current), 1.5°C (~2030), 2.0°C (~2050), 3.0°C (~2090). |
| Coordinate Reference System (CRS) | GDA94 / VicGrid projection (EPSG:3111) |
| Format | GeoJSON, CSV |
| Access | Zenodo DOI: 10.5281/zenodo.17379391 |
| License | CC BY 4.0 (data), MIT (code) |
If you use this dataset, code, or methodology, please cite:
Data Descriptor Paper:
Chin, T.K., Prakash, M., Zheng, N., & Pauwels, V.R.N. (2025). Fused Geospatial Dataset Linking Climate Hazards and Road Infrastructure for Victoria, Australia.
Scientific Data (under review).
Dataset (Data Fusion Stage):
Chin, T.K. (2025). Fused Geospatial Dataset of Road Infrastructure and Climate Hazards for Victoria, Australia.
Zenodo. https://doi.org/10.5281/zenodo.17379391
Code Repository:
Chin, T.K. (2025). RICRAF: Road Infrastructure Climate Risk Assessment Framework (Code Repository).
Zenodo. https://doi.org/10.5281/zenodo.17391486
Model Development Paper:
Chin, T.K., Prakash, M., Zheng, N., & Pauwels, V.R.N. (2025). Explainable AI for Multi-Stressor Climate Risk Assessment of Road Networks: An XGBoost–SHAP Framework.
Climate Services (under review).
Model Application Paper:
Chin, T.K., Prakash, M., Zheng, N., & Pauwels, V.R.N. (2025). Climate Risk Assessment of Road Infrastructure under Multi-Stressor Conditions in Victoria, Australia.
Transportation Research Part D: Transport and Environment (under review).
For machine-readable citation metadata, please refer to CITATION.cff in this repository.
- Code: MIT License — see
LICENSE - Dataset: Creative Commons Attribution 4.0 International (CC BY 4.0)
🧩 This repository complies with the FAIR data principles and the open-science standards recommended by Scientific Data, Climate Services, and Transportation Research Part D.
Data sources:
- DataVic — for road infrastructure and traffic datasets
- Australian Climate Service — for climate hazard and projection data
- Zenodo — for open data hosting
- Supported by Monash University and CSIRO Data 61
- Dataset (Zenodo): 10.5281/zenodo.17379391
- Code Snapshot (Zenodo): 10.5281/zenodo.17391486
- GitHub Repository: https://github.com/teckkean/ricraf
- Data Article: (link to final publication DOI once available)
- Model Development Paper: (link to final publication DOI once available)
- Model Application Paper: (link to final publication DOI once available)
Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., Appleton, G., Axton, M., Baak, A., … Mons, B. (2016).
The FAIR Guiding Principles for scientific data management and stewardship.
Scientific Data, 3, 160018. https://doi.org/10.1038/sdata.2016.18
© 2025 Teck Kean Chin