Multi-Hazard Entanglement

Globally, natural hazards such as tropical cyclones cause billions of dollars in damages. These hazards rarely occur in isolation. Frequently, one hazard triggers another, such as rainfall triggering a flood. Likewise, the likelihood of a hazardous event can be amplified by the occurrence of a previous event, such as a drought amplifying the likelihood of a wildfire. Traditionally, such extremes have often been modeled independently. However, neglecting the interactions between extremes can underestimate risk, as their combined effects may differ from the sum of individual impacts. A greater understanding of multi-hazard events and the statistical dependencies between hazard types is therefore essential for more holistic multi-hazard risk assessments. A key first step in achieving this understanding is to develop realistic multi-hazard event sets. This thesis explores single-hazard datasets for meteorological, geophysical, hydrological, and climatological events using the MYRIAD-Hazard Event Sets Algorithm (MYRIAD-HESA). MYRIAD-HESA combines single-hazard footprints into multi-hazard event sets based on their spatial and temporal overlap. The algorithm was used to create a global multi-hazard event database that includes eleven hazard types from different classes, covering the period from 2004 to 2017. This database was used to identify global hotspots of multi-hazard events and the most frequent hazard combinations in specific regions. In addition to understanding historical multi-hazard events, it is essential to quantify the statistical dependencies between different hazards. This requires a model capable of capturing conditional dependencies between variables. To address this, the thesis introduced VineCopulas, a Python package designed for conditional bivariate and vine copula modeling. Copulas capture the dependency structure between multiple random variables, while vine copulas extend this method to higher dimensions by combining bivariate copulas. VineCopulas enables users to model complex multivariate dependencies, making it suitable for studying joint hazard behavior, such as how temperature, wind speed, and precipitation interact. The MYRIAD-Stochastic vIne-copula Model (MYRIAD-SIM), a stochastic weather generator (SWG), was developed using VineCopulas. MYRIAD-SIM simulates spatiotemporal dependencies between weather variables. Trained on ERA5-Land data, it generates 256 years of spatiotemporal data for variables like temperature, wind speed, and precipitation over Europe. Validation of the simulated data confirmed that it preserves the statistical properties of ERA5-Land while offering sufficient variability to explore a range of scenarios, making it useful for analyzing alternative multi-hazard events. MYRIAD-SIM was tested by generating stochastic data tailored to evaluate multi-hazard events in Europe. Three case studies were explored: a 2021 false spring in Burgundy, France, where early warmth triggered premature vine growth destroyed by frost; a 2022 sequence of three storms in six days affecting the UK and Ireland, which caused flooding and wind damage; and a 2023 heatwave in Sicily, Italy, that led to wildfires intensified by winds. Using the simulated data, the likelihood of these historic multi-hazard events was estimated by calculating their annual exceedance probabilities and return periods. The triple storm event in the UK and Ireland was the rarest, with a 2% probability and a return period of 51.2 years. The Sicily wildfire event was the most common, occurring 196 times with a 19.1% probability and a return period of 5.2 years. The Burgundy false spring event appeared 64 times, with a 6.3% probability and a 16-year return period. These findings provide valuable insights for multi-risk assessments, improving preparedness for the recurrence of multi-hazard events. This thesis contributes to the advancement of multi-hazard risk assessments by offering open-source tools and datasets. The flexibility of MYRIAD-HESA and MYRIAD-SIM allows them to be adapted to user needs, incorporating higher-resolution hazard data for particular regions. These tools, datasets, and findings support researchers and practitioners in making more informed multi-hazard risk management decisions, improving disaster preparedness and resilience.