Hugo Rakotoarimanga
Research Subjects
Research Interests
Hugo Rakotoarimanga has a total of 36 co-authors. Here are the top co-authors by number of shared publications:
Rudy Mustafa
4 shared publicationsRémi Meynadier
4 shared publications- Remi Meynadier3 shared publications
Anna Weisman
2 shared publications- Emmanouil Flaounas
Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland
2 shared publications - Hessel C. Winsemius
Deltares, Delft, The Netherlands
2 shared publications
Latest Publications
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Current and Future Risks of Storm Clustering in Western Europe.
European windstorms drive much of the region’s extreme weather, causing catastrophic winds and flooding.Beyond individual hazards, sequences of windstorms, so-called storm clustering, can make landfall along European coasts and propagate inland, inflicting and compounding socioeconomic impacts. This is directly relevant to local recovery and to understanding how impacts accumulate over short timescales. While several studies have examined how storm intensity may change under future climate conditions, far less attention has been paid to storm clustering, the intensity of clustered storms, and the associated risk.In this study, we use 2,000 years of climate simulations performed with CESM under present-day and future conditions (100 integrations for 1991–2000 and another 100 for 2091–2100, based on the CMIP5 RCP8.5 scenario) to identify and quantify socioeconomic impacts in Western Europe from extreme winds and their clusters. This large sample provides more robust statistics for detecting sub-monthly clustered storms.Our objectives are twofold: first, to analyse the physical characteristics of storm clusters; and second, to quantify their socioeconomic relevance in terms of risk and impacts.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessCurrent and Future Risks of Storm Clustering in Western Europe.
European windstorms drive much of the region’s extreme weather, causing catastrophic winds and flooding.Beyond individual hazards, sequences of windstorms, so-called storm clustering, can make landfall along European coasts and propagate inland, inflicting and compounding socioeconomic impacts. This is directly relevant to local recovery and to understanding how impacts accumulate over short timescales. While several studies have examined how storm intensity may change under future climate conditions, far less attention has been paid to storm clustering, the intensity of clustered storms, and the associated risk.In this study, we use 2,000 years of climate simulations performed with CESM under present-day and future conditions (100 integrations for 1991–2000 and another 100 for 2091–2100, based on the CMIP5 RCP8.5 scenario) to identify and quantify socioeconomic impacts in Western Europe from extreme winds and their clusters. This large sample provides more robust statistics for detecting sub-monthly clustered storms.Our objectives are twofold: first, to analyse the physical characteristics of storm clusters; and second, to quantify their socioeconomic relevance in terms of risk and impacts.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Estimation of extreme tropical cyclone risk using AI-weather models
With its global footprint, AXA is exposed to multiple natural hazards across the globe. Assessing the frequency and intensity of these events, especially unobserved extremes, is crucial to monitor, mitigate and adapt to the risk they pose.Tropical cyclones are one of the most scrutinized natural risks by global (re)insurers. Curated observational records date back to the mid-1800s, with increased reliability from the satellite era onwards (post 1970). They are a global risk, with temporal and spatial dependencies between tropical basins. The extreme damage they cause has been at the root of the development of Natural Catastrophe (NATCAT) modelling capabilities by specialized modelling firms, brokers, and (re)insurers.However, as exposure is increasing and climate is changing, especially in tropical cyclone prone coastal areas globally, the need for robust and accurate estimates of the frequency and intensity of adverse impacts from tropical cyclones is expanding. Observational tropical cyclones datasets like IBTrACS are too short to obtain reliable statistics on rarest and most impactful events.Fine resolution numerical weather models are too computationally expensive to run on extended periods of time.AI-based weather models running on GPU-accelerated compute infrastructure provide the necessary speedup while maintaining physical accuracy, enabling the generation of thousands of synthetic tropical cyclone seasons. Using NVIDIA's Earth-2 platform, we build a pipeline to produce hundreds of downscaled large ensemble predictions.This study investigates the potential of these downscaled runs to generate large sets of tropical cyclones physically consistent in space, time and intensity, yielding robust estimates of their impact probability, especially for the rarest events.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessEstimation of extreme tropical cyclone risk using AI-weather models
With its global footprint, AXA is exposed to multiple natural hazards across the globe. Assessing the frequency and intensity of these events, especially unobserved extremes, is crucial to monitor, mitigate and adapt to the risk they pose.Tropical cyclones are one of the most scrutinized natural risks by global (re)insurers. Curated observational records date back to the mid-1800s, with increased reliability from the satellite era onwards (post 1970). They are a global risk, with temporal and spatial dependencies between tropical basins. The extreme damage they cause has been at the root of the development of Natural Catastrophe (NATCAT) modelling capabilities by specialized modelling firms, brokers, and (re)insurers.However, as exposure is increasing and climate is changing, especially in tropical cyclone prone coastal areas globally, the need for robust and accurate estimates of the frequency and intensity of adverse impacts from tropical cyclones is expanding. Observational tropical cyclones datasets like IBTrACS are too short to obtain reliable statistics on rarest and most impactful events.Fine resolution numerical weather models are too computationally expensive to run on extended periods of time.AI-based weather models running on GPU-accelerated compute infrastructure provide the necessary speedup while maintaining physical accuracy, enabling the generation of thousands of synthetic tropical cyclone seasons. Using NVIDIA's Earth-2 platform, we build a pipeline to produce hundreds of downscaled large ensemble predictions.This study investigates the potential of these downscaled runs to generate large sets of tropical cyclones physically consistent in space, time and intensity, yielding robust estimates of their impact probability, especially for the rarest events.
- Open Access
- 18 March 2025
- Published by Copernicus GmbH
Usage of seasonal forecasts in Tropical Cyclone risk models
Tropical cyclones (TCs) pose significant risks to lives, infrastructure and economies, especially in coastal areas.AXA has been developing stochastic natural hazard models (also called natural catastrophe or NatCat models) to quantify the impact of events such as TCs on its portfolios. However, NatCat models tend to model the average annual risk for a given peril. NatCat models do not consider the present state of the atmosphere and therefore are not conditioned with respect to the current tropical cyclone season.Information about the TC risk in the upcoming weeks or months of a season could be crucial for an insurer, especially regarding its reinsurance coverage, but also for better risk mitigation through reinforced and more efficient prevention systems.Previous studies have demonstrated that ensemble seasonal forecasts have skill in predicting TC occurrence several weeks in advance. We explore the ability of ensemble seasonal forecasts to provided skilled information on the general activity of the season to come for various lead-times (number of occurrences, number of landfalls, ACE…) and how can NatCat models be adapted to provide a more dynamic vision of the TC risk.
- 18 March 2025
- Published by Copernicus GmbH
Open AccessUsage of seasonal forecasts in Tropical Cyclone risk models
Tropical cyclones (TCs) pose significant risks to lives, infrastructure and economies, especially in coastal areas.AXA has been developing stochastic natural hazard models (also called natural catastrophe or NatCat models) to quantify the impact of events such as TCs on its portfolios. However, NatCat models tend to model the average annual risk for a given peril. NatCat models do not consider the present state of the atmosphere and therefore are not conditioned with respect to the current tropical cyclone season.Information about the TC risk in the upcoming weeks or months of a season could be crucial for an insurer, especially regarding its reinsurance coverage, but also for better risk mitigation through reinforced and more efficient prevention systems.Previous studies have demonstrated that ensemble seasonal forecasts have skill in predicting TC occurrence several weeks in advance. We explore the ability of ensemble seasonal forecasts to provided skilled information on the general activity of the season to come for various lead-times (number of occurrences, number of landfalls, ACE…) and how can NatCat models be adapted to provide a more dynamic vision of the TC risk.
- Open Access
- 18 March 2025
- Published by Copernicus GmbH
Use of NVIDIA FourCastNet model to improve tropical cyclones risk modelling.
AXA is developing in-house Natural Hazard models (or Natural Catastrophe models) in order to gain a deeper understanding of, evaluate, and monitor the climate risks underpinning (re)insurance portfolios. Such models simulate large numbers of synthetic weather events to estimate the probability of rare and extreme events, enabling more robust risk management and informed decision-making.AI-driven weather models offer the capability to rapidly produce thousands of unique ensemble scenarios of low-likelihood high-impact weather events such as tropical cyclones. This study specifically utilizes tropical cyclones (TCs) as a primary illustration of the potential of AI-based weather models for risk management.In this study we use FourCastNet SFNO, the global data-driven weather forecasting model developed by NVIDIA available on the NVIDIA Earth-2 platform to simulate historical but also synthetic (i.e. never observed) hurricanes. SFNO trained on ECMWF ERA5 reanalysis data provides short to medium-range global predictions at 0.25° resolution. A large ensemble of hurricane simulations is performed using the HENS method, developed at Berkeley, the NVIDIA leveraging Earth2Studio from NVIDIA’s Earth-2 platform.HENS-SFNO performance is first assessed by evaluating the model's ability to reproduce post-2017 historical hurricanes (intensity, track, landfall location). HENS-SFNO capabilities in simulating synthetic hurricanes are then assessed in a second step by evaluating track density and landfall frequencies by categories of hurricanes against the historical tropical cyclone IBTrACS database.
- 18 March 2025
- Published by Copernicus GmbH
Open AccessUse of NVIDIA FourCastNet model to improve tropical cyclones risk modelling.
AXA is developing in-house Natural Hazard models (or Natural Catastrophe models) in order to gain a deeper understanding of, evaluate, and monitor the climate risks underpinning (re)insurance portfolios. Such models simulate large numbers of synthetic weather events to estimate the probability of rare and extreme events, enabling more robust risk management and informed decision-making.AI-driven weather models offer the capability to rapidly produce thousands of unique ensemble scenarios of low-likelihood high-impact weather events such as tropical cyclones. This study specifically utilizes tropical cyclones (TCs) as a primary illustration of the potential of AI-based weather models for risk management.In this study we use FourCastNet SFNO, the global data-driven weather forecasting model developed by NVIDIA available on the NVIDIA Earth-2 platform to simulate historical but also synthetic (i.e. never observed) hurricanes. SFNO trained on ECMWF ERA5 reanalysis data provides short to medium-range global predictions at 0.25° resolution. A large ensemble of hurricane simulations is performed using the HENS method, developed at Berkeley, the NVIDIA leveraging Earth2Studio from NVIDIA’s Earth-2 platform.HENS-SFNO performance is first assessed by evaluating the model's ability to reproduce post-2017 historical hurricanes (intensity, track, landfall location). HENS-SFNO capabilities in simulating synthetic hurricanes are then assessed in a second step by evaluating track density and landfall frequencies by categories of hurricanes against the historical tropical cyclone IBTrACS database.
- Open Access
- 18 March 2025
- Published by Copernicus GmbH
Multi-model assessment of hazard uncertainties in a European windstorm NatCat model
Extra-tropical winter storms are one of the most impactful natural hazards for the European insurance market causing large socio-economic damages.AXA has been developing stochastic natural hazard models (also called natural catastrophe models) to quantify the impact of such events on its portfolios, including European extra-tropical cyclones. However, the correct representation of windspeeds and their spatial distribution across Europe during a storm is crucial to determine the risk posed by an event. The characterization of uncertainties in natural catastrophe models stemming from the hazard data used and its resolution is crucial to understand their limitations and guide decision-making.We rely on a novel publicly available dataset of 50 extreme European windstorms for the period 1995–2015 (Flynn et al., 2024; doi:10.5194/essd-2024-298) with wind gust footprints derived consistently from four different datasets with different horizontal resolutions. Risk being a function of hazard, vulnerability and exposure, we set constant vulnerability and portfolio, and we quantify the range of uncertainties in the reproduction of historical insured losses stemming from the sole hazard component. We compare the losses derived from AXA’s model to the range of losses derived from this novel extreme windstorms dataset.
- 18 March 2025
- Published by Copernicus GmbH
Open AccessMulti-model assessment of hazard uncertainties in a European windstorm NatCat model
Extra-tropical winter storms are one of the most impactful natural hazards for the European insurance market causing large socio-economic damages.AXA has been developing stochastic natural hazard models (also called natural catastrophe models) to quantify the impact of such events on its portfolios, including European extra-tropical cyclones. However, the correct representation of windspeeds and their spatial distribution across Europe during a storm is crucial to determine the risk posed by an event. The characterization of uncertainties in natural catastrophe models stemming from the hazard data used and its resolution is crucial to understand their limitations and guide decision-making.We rely on a novel publicly available dataset of 50 extreme European windstorms for the period 1995–2015 (Flynn et al., 2024; doi:10.5194/essd-2024-298) with wind gust footprints derived consistently from four different datasets with different horizontal resolutions. Risk being a function of hazard, vulnerability and exposure, we set constant vulnerability and portfolio, and we quantify the range of uncertainties in the reproduction of historical insured losses stemming from the sole hazard component. We compare the losses derived from AXA’s model to the range of losses derived from this novel extreme windstorms dataset.
- Open Access
- 18 March 2025
- Published by Copernicus GmbH
Explicit risk modelling of sting-jet extratropical cyclones.
Extratropical cyclones (ETCs) are a major hazard for Europe as they cause most of the windstorms and floods in the mid-latitudes, resulting in high economic and social costs.Sting jets (SJ) are responsible for windstorm damages well ahead the cyclone center. In this study we employ dedicated diagnostics and modeling approaches that identify -along with cyclone tracks- the spatial extent where actual impacts take place. The fine scales of processes involved in SJ generation demand exceptionally high spatial resolutions and dense vertical levels in model simulations (Rivière et al. 2020).In this study we use the WRF model to simulate 143 historical ETC from 1980 to 2018 that potentially involve SJs. The model simulations use two domains: one parent domain that encompasses the whole cyclone track at a resolution of about 15 km, and another, square-sized domain with each side measuring 1300 km. The nested domain always follows the ETC centers, aiming to resolve explicitly the development of SJs. SJ detection has been achieved through lagrangian modeling, by identifying airstreams that sharply descend ahead of the cloud head and behind the cold front of the cyclones. Historical ETC footprints from ERA-5 and WRF physical downscaling of ERA-5 in convection-permitting resolutions are then used to assess the impact in term of financial losses of an explicit simulation of sting-jets processes.
- 18 March 2025
- Published by Copernicus GmbH
Open AccessExplicit risk modelling of sting-jet extratropical cyclones.
Extratropical cyclones (ETCs) are a major hazard for Europe as they cause most of the windstorms and floods in the mid-latitudes, resulting in high economic and social costs.Sting jets (SJ) are responsible for windstorm damages well ahead the cyclone center. In this study we employ dedicated diagnostics and modeling approaches that identify -along with cyclone tracks- the spatial extent where actual impacts take place. The fine scales of processes involved in SJ generation demand exceptionally high spatial resolutions and dense vertical levels in model simulations (Rivière et al. 2020).In this study we use the WRF model to simulate 143 historical ETC from 1980 to 2018 that potentially involve SJs. The model simulations use two domains: one parent domain that encompasses the whole cyclone track at a resolution of about 15 km, and another, square-sized domain with each side measuring 1300 km. The nested domain always follows the ETC centers, aiming to resolve explicitly the development of SJs. SJ detection has been achieved through lagrangian modeling, by identifying airstreams that sharply descend ahead of the cloud head and behind the cold front of the cyclones. Historical ETC footprints from ERA-5 and WRF physical downscaling of ERA-5 in convection-permitting resolutions are then used to assess the impact in term of financial losses of an explicit simulation of sting-jets processes.
- 31 August 2024
- Published by Springer Nature
- p. 301-316
Flood Compound Modelling: Framework and Application to a Coastal Environment
- 31 August 2024
- Published by Springer Nature
- p. 301-316
Flood Compound Modelling: Framework and Application to a Coastal Environment
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Developing a unified and consistent database for historical natural events, and subsequent losses, within a catastrophe modeling framework
The insurance industry faces highly complex P&C challenges, among which natural catastrophe risk, also labeled as “CAT” risk. Among disasters, climatic and seismic events show large variability in size and frequency, with devastating consequences; not to mention climate change which brings added uncertainty for the future. Global insurance groups, such as AXA, must develop a sound understanding of the frequency, intensity, and impacts of natural hazard events, to protect their economic capital and ensure their solvency.At the AXA Group Risk Management, the CAT modeling process consists in 1) collecting CAT exposure data (geographical, physical, and financial information) on a per-entity (AXA France, AXA Mexico…) and per-location basis (houses, factories, vehicles…) and 2) assessing the risk on a per-entity per-peril per-geography basis (cyclones, earthquakes, floods, hailstorms...) to finally aggregate it at Group level. This process constitutes a technical challenge through the data collection of 50 million of policies, the combination of multiple modeling solutions, and the production of millions of stochastic event losses. Alongside this process, the collection and analysis of “scenarios”, either historical events, or potential future disasters, improves the robustness and understanding of risk assessment. However, there is currently no unified and consistent database recording the characteristics of natural events (a unique identifier, their spatial and temporal extent, their intensities, and the location affected) and their actualized economic and industry impacts. This work aims at developing a database for that would first gather an exhaustive inventory of historical natural events (cyclones, storms, floods, earthquakes…) and, throughout the integration within the existing CAT modeling ecosystem, automatize model validation, back-testing, and risk analysis with respect to market and as-if losses.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessDeveloping a unified and consistent database for historical natural events, and subsequent losses, within a catastrophe modeling framework
The insurance industry faces highly complex P&C challenges, among which natural catastrophe risk, also labeled as “CAT” risk. Among disasters, climatic and seismic events show large variability in size and frequency, with devastating consequences; not to mention climate change which brings added uncertainty for the future. Global insurance groups, such as AXA, must develop a sound understanding of the frequency, intensity, and impacts of natural hazard events, to protect their economic capital and ensure their solvency.At the AXA Group Risk Management, the CAT modeling process consists in 1) collecting CAT exposure data (geographical, physical, and financial information) on a per-entity (AXA France, AXA Mexico…) and per-location basis (houses, factories, vehicles…) and 2) assessing the risk on a per-entity per-peril per-geography basis (cyclones, earthquakes, floods, hailstorms...) to finally aggregate it at Group level. This process constitutes a technical challenge through the data collection of 50 million of policies, the combination of multiple modeling solutions, and the production of millions of stochastic event losses. Alongside this process, the collection and analysis of “scenarios”, either historical events, or potential future disasters, improves the robustness and understanding of risk assessment. However, there is currently no unified and consistent database recording the characteristics of natural events (a unique identifier, their spatial and temporal extent, their intensities, and the location affected) and their actualized economic and industry impacts. This work aims at developing a database for that would first gather an exhaustive inventory of historical natural events (cyclones, storms, floods, earthquakes…) and, throughout the integration within the existing CAT modeling ecosystem, automatize model validation, back-testing, and risk analysis with respect to market and as-if losses.
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
A modular Pan-European Flood Catastrophe model
The flood event “Berndt” in North-Western Europe in July 2021 demonstrated to stakeholders the critical importance of understanding present-day events, their spatial-temporal coherence and magnitude, and the probability of their occurrence. The shear size of multi-country events such as “Berndt” necessitate a large scale and highly automated approach to the modelling of their characteristics.To this end, we (AXA, Fathom and Deltares) developed a modelling framework that can efficiently compute very large event sets and worked on improving its underlying skill by making the framework highly modular. The framework consists of the following modules: a synthetic weather generator, which may sample many thousands of years of continuous weather data at daily and sub-daily time scales; a gridded hydrological model forced by the synthetic weather that produces long-term hydrological timeseries and derived statistics, and a subcatchment-scale fluvial and pluvial flood model archive, produced from large amounts of simulations with the Fathom hydraulic flood model engine. Its modular character allows for exchanging components, improving existing components or modifying parameters to assess sensitivities or uncertainties.In last year’s contribution, we presented our first synthetic simulations of weather and hydrology. In this year’s contribution we have improved stochastic weather generation and have established a full 10,000-year event set and flood catalogue, both under the present climate and a future climate projection that encompasses several assumptions. We will show the statistics and spatial configuration of past events according to our catalogue, including “Berndt 2021”, and how the statistics of extreme events will change according to the defined climate scenario simulation.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessA modular Pan-European Flood Catastrophe model
The flood event “Berndt” in North-Western Europe in July 2021 demonstrated to stakeholders the critical importance of understanding present-day events, their spatial-temporal coherence and magnitude, and the probability of their occurrence. The shear size of multi-country events such as “Berndt” necessitate a large scale and highly automated approach to the modelling of their characteristics.To this end, we (AXA, Fathom and Deltares) developed a modelling framework that can efficiently compute very large event sets and worked on improving its underlying skill by making the framework highly modular. The framework consists of the following modules: a synthetic weather generator, which may sample many thousands of years of continuous weather data at daily and sub-daily time scales; a gridded hydrological model forced by the synthetic weather that produces long-term hydrological timeseries and derived statistics, and a subcatchment-scale fluvial and pluvial flood model archive, produced from large amounts of simulations with the Fathom hydraulic flood model engine. Its modular character allows for exchanging components, improving existing components or modifying parameters to assess sensitivities or uncertainties.In last year’s contribution, we presented our first synthetic simulations of weather and hydrology. In this year’s contribution we have improved stochastic weather generation and have established a full 10,000-year event set and flood catalogue, both under the present climate and a future climate projection that encompasses several assumptions. We will show the statistics and spatial configuration of past events according to our catalogue, including “Berndt 2021”, and how the statistics of extreme events will change according to the defined climate scenario simulation.
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Developing a natural catastrophe model for European winter windstorms, an insurer’s perspective
European windstorms are powerful extratropical cyclones mostly taking place during the winter months, and are one of Europe’s costliest natural disasters. The close study and assessment of this risk has therefore been essential for the insurance industry concerned. Typically, insurers resort to physical natural catastrophe models developed by third-party companies to analyze the risk, as they capture its components of hazard (events frequency and severity), exposure (insured assets values), and vulnerability (assets' damageability to given hazard intensities). AXA proposes a modeling methodology to produce a hazard catalog of synthetic windstorm events, and a vulnerability module, built around publicly available, purchased, or internal data. The hazard catalog is created using a meteorological feature tracking algorithm to extract trajectories and footprints of European windstorms in CMIP6 and ECMWF-ERA5 data. The catalog is then enriched to become a 10,000-year stochastic catalog by physically resampling original events with a perturbation technique, and statistically downscaling them to a 4-km resolution. The vulnerability, that yields damage ratios from local windspeed intensities, predicts the expected probability of claim occurrence and a distribution of conditional damage ratios based on wind gust value and exposure risk drivers. The model shows good backtesting performances at continental scale on market and AXA exposure. It is fully integrated within AXA's modelling ecosytem and is operationnally used to assess one of the major risks faced by the Group.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessDeveloping a natural catastrophe model for European winter windstorms, an insurer’s perspective
European windstorms are powerful extratropical cyclones mostly taking place during the winter months, and are one of Europe’s costliest natural disasters. The close study and assessment of this risk has therefore been essential for the insurance industry concerned. Typically, insurers resort to physical natural catastrophe models developed by third-party companies to analyze the risk, as they capture its components of hazard (events frequency and severity), exposure (insured assets values), and vulnerability (assets' damageability to given hazard intensities). AXA proposes a modeling methodology to produce a hazard catalog of synthetic windstorm events, and a vulnerability module, built around publicly available, purchased, or internal data. The hazard catalog is created using a meteorological feature tracking algorithm to extract trajectories and footprints of European windstorms in CMIP6 and ECMWF-ERA5 data. The catalog is then enriched to become a 10,000-year stochastic catalog by physically resampling original events with a perturbation technique, and statistically downscaling them to a 4-km resolution. The vulnerability, that yields damage ratios from local windspeed intensities, predicts the expected probability of claim occurrence and a distribution of conditional damage ratios based on wind gust value and exposure risk drivers. The model shows good backtesting performances at continental scale on market and AXA exposure. It is fully integrated within AXA's modelling ecosytem and is operationnally used to assess one of the major risks faced by the Group.