The Risk Analysis Framework for Tropical Cyclones (RAFT) is a unified framework capable of efficiently simulating millions of synthetic tropical cyclones (TCs; also known as hurricanes or typhoons) and their impacts, enabling robust probabilistic risk assessment.
RAFT consists of several different components that are coupled into one cohesive system. It utilizes established physical and statistical methods, while custom machine learning and AI approaches have been developed for more challenging and novel tasks. The summary of capabilities below is intended as a high-level overview; we refer interested readers to the referenced publications for technical details.
Points of Contact:
- Karthik Balaguru (PI): karthik.balaguru@pnnl.gov
- Julian Rice (Lead Developer): julian.rice@pnnl.gov
Table of Contents
RAFT is a collection of models and tools written in Python, which when combined can simulate large numbers of TCs which are both realistic and physically consistent with the large-scale atmospheric forcing conditions.
A sample of 5,000 synthetic storms, generated by RAFT forced with ERA5 historical reanalysis conditions
RAFT is forced with large-scale environmental conditions, which can be extracted either from observations or atmospheric models. Instead of trying to predict exactly what storms would form in a given environment, it simulates plausible storms that are consistent with the characteristics of that environment.
Storm seeds are created by randomly sampling from the smoothed spatio-temporal distribution of historical TC genesis. Although the global rate of genesis seeding is nominally fixed, the intensity model described later may choose to decay these seeds before they develop, effectively modulating genesis frequency based on the favorability of the environment.
These seeds are then propagated with the beta-advection method, which at its most basic means that the storms are blown by the large-scale prevailing winds. The beta-advection technique was initially described by Emanuel et al. (2006), with an improved beta scheme as described in Xu et al. (2024). Note that these tracks are NOT simply perturbations of historical events, but entirely new storms (yet physically realistic) that in some cases have never been seen before.
Storm intensity is modeled with a deep learning neural network developed for this framework. Xu et al. (2021) details the methodology, and finds that the model is competitive with—and even occasionally outperforms—operational TC intensity forecasts.
Average error of our AI intensity model (MLP), compared with a number of established methods including NHC Official Forecasts (OFCL) and NOAA's Hurricane Weather & Research Forecasting model (HWFI). Lower is better. (Xu et al. 2021)
Because RAFT does not require high-resolution forcings, it can be forced with lower-resolution future climate simulations such as those from the CMIP6 project. This enables the projection of TC behavior into the future under various different scenarios. This methodology is described and explored by Balaguru et al. (2023), who find substantially increasing coastal hurricane risk in the U.S. in the SSP5-8.5 scenario.
A sample of 500 RAFT storms representative of a possible future atmosphere, forced by simulated 2015-2050 conditions from the E3SM atmospheric model under the SSP5-8.5
Once an ensemble of TCs have been created, RAFT can assess a variety of impacts. Because of the significant sample size of TCs that RAFT generates (tens-of-thousands to millions), this enables the estimation of very rare impacts (e.g. 1-in-500-year events) that would otherwise be very difficult to model or observe.
There are two rainfall models in this repository:
- TCR is a physics-based rainfall model described in Lu & Lin et al. (2018). Our implementation is described in Xu et al. (2024).
- PHRaMM is a physics-based rainfall model described in Kim et al. (2022). We have found TCR to perform better overall, so this component may be out of date.
A deep learning AI rainfall model is currently in development.
Accumulated TCR rainfall from two sample RAFT storms affecting the New York and Houston regions (Xu et al. 2024)
Storm surge hazards are one of the most dangerous impacts from TCs, yet traditional surge models are often computationally expensive. To handle such large ensembles as generated by RAFT, we developed an AI storm surge model called DeepSurge. The first version of this model is described in Rice et al. (2025), while a second version is in development.
The future change in 1-in-100-year storm surge height along the U.S. Gulf and East coasts, estimated from an ensemble of 900,000 RAFT storms (Rice et al. 2025)
A generalized power outage model, the Electric Power Outages from Cyclone Hazards (EPOCH) model, described in Rice et al. (2025), was developed to assess the risk posed by RAFT-simulated TCs to the electric grid. Development of this model is actively ongoing.
Historical (1980-2015) and future (2065-2100) tropical cyclone-induced electric power outage risk, as projected by RAFT (Rice et al. 2025)
RAFT can also be used to assess the hazard posed by TCs to other energy infrastructure, such as wind turbines. This method was used to quantify future changes in hazard along the U.S. coastline in Lipari et al. (2024).
20, 50, and 100-year return period coastal wind speeds as estimated from an ensemble of 900,000 RAFT storms (Lipari et al. 2024)
Rice, J. R., Balaguru, K., Ticona Rollano, F., Wilson, J., Daniel, B., Judi, D., Sun, N., & Leung, L. R. (2025). Projecting U.S. coastal storm surge risks and impacts with deep learning. Environmental Research Letters, 20(10), 104013. https://doi.org/10.1088/1748-9326/adfd74
Rice, J. R., Balaguru, K., Staid, A., Xu, W., & Judi, D. (2025). Projected increases in tropical cyclone-induced U.S. electric power outage risk. Environmental Research Letters, 20(3), 034030. https://doi.org/10.1088/1748-9326/adad85
Lipari, S., Balaguru, K., Rice, J., Feng, S., Xu, W., K. Berg, L., & Judi, D. (2024). Amplified threat of tropical cyclones to US offshore wind energy in a changing climate. Communications Earth & Environment, 5(1), 1–10. https://doi.org/10.1038/s43247-024-01887-6
Xu, W., Balaguru, K., Judi, D. R., Rice, J., Leung, L. R., & Lipari, S. (2024). A North Atlantic synthetic tropical cyclone track, intensity, and rainfall dataset. Scientific Data, 11(1), 130. https://doi.org/10.1038/s41597-024-02952-7
Balaguru, K., Xu, W., Chang, C.-C., Leung, L. R., Judi, D. R., Hagos, S. M., Wehner, M. F., Kossin, J. P., & Ting, M. (2023). Increased U.S. coastal hurricane risk under climate change. Science Advances, 9(14), eadf0259. https://doi.org/10.1126/sciadv.adf0259
Xu, W., Balaguru, K., August, A., Lalo, N., Hodas, N., DeMaria, M., & Judi, D. (2021). Deep Learning Experiments for Tropical Cyclone Intensity Forecasts. Weather and Forecasting, 36(4), 1453–1470. https://doi.org/10.1175/WAF-D-20-0104.1
Kelly, P., Leung, L. R., Balaguru, K., Xu, W., Mapes, B., & Soden, B. (2018). Shape of Atlantic Tropical Cyclone Tracks and the Indian Monsoon. Geophysical Research Letters, 45(19), 10,746-10,755. https://doi.org/10.1029/2018GL080098
40,000 synthetic TCs in the North Atlantic basin, from RAFT forced by ERA5 reanalysis:
Xu, W., Balaguru, K., Judi, D. R., Rice, J., Leung, L. R., & Lipari, S. (2024). A North Atlantic synthetic tropical cyclone track, intensity, and rainfall dataset. Scientific Data, 11(1), 130. https://doi.org/10.1038/s41597-024-02952-7
Storm surge data for the North Atlantic coastline generated by applying the DeepSurge (v1) storm surge model to 900,000 synthetic tropical cyclones created with RAFT forced by CMIP6 data for the historical (1980-2015) and future (2066-2100) climate under SSP5-8.5:
Rice, J. R., Balaguru, K., Ticona Rollano, F., Wilson, J., Daniel, B., Xu, W., Judi, D., Sun, N., & Leung, R. (2025). DeepSurge storm surge predictions for RAFT-CMIP6 tropical cyclones in the North Atlantic [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.15021868
Power outages at a U.S. county level, modeled for 900,000 RAFT TCs forced by CMIP6 for the historical (1980-2014) and future (2066-2100) period under SSP5-8.5 warming. Outages are modeled with the Electric Power Outages from Cyclone Hazards (EPOCH) model, which was trained on county-level outage data from 23 historical TC events in the EAGLE-I dataset:
Rice, J. R., Balaguru, K., Staid, A., Xu, W., & Judi, D. (2024). Electric power outages from 900k simulated hurricanes in a changing climate, for the United States and Puerto Rico [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.12746675
Hub-height wind and damage probability maps for wind turbines exposed to RAFT-CMIP6 tropical cyclones from historical (1980-2014) and future (2066-2100) climate scenarios along the U.S. Atlantic and Gulf Coasts:
Lipari, S., Balaguru, K., Rice, J., Feng, S., Xu, W., Berg, L., & Judi, D. (2024). Offshore wind turbine damage probability maps and hub height TC wind speeds for U.S. Atlantic and Gulf Coasts exposed to historical and future tropical cyclones [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.13840812
Wind Data Hub. trexo/turbine_damage_probability. Maintained by Wind Data Hub for U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy. DOI: https://doi.org/10.21947/2481035 Accessed: 12 Sept 2025.
Neither the United States Government nor the United States Department of Energy, nor Battelle, nor any of their employees, nor any jurisdiction or organization that has cooperated in the development of these materials, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness or any information, apparatus, product, software, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof, or Battelle Memorial Institute. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.