Uncertainty in climate change and its impact on earthquake risk

Background

Under climate change, some regions are experiencing changes in precipitation and temperature that lead to changes in the soil characteristics directly (e.g. moisture) or indirectly (e.g., deforestation, water cycle). These alterations can result in enhanced shaking during earthquakes because of the lowered shear wave velocity in the upper soil layers, with potential of a substantial increase in the damage. This impact is a crucial topic to investigate considering the growth of risk in a lot of highly populated cities around the world where precipitation is about to increase substantially. Uncertainties in climate change need to be represented to inform catastrophe modelling and provide a probabilistic view of climate impacts on earthquake risk. The project aims to develop and deploy Machine Learning advances to propagate uncertainties across the hazard and risk chain. This is a new and exciting area of research in partnership with the insurance industry.

PhD opportunity

The main challenge consists of modelling the chain of uncertainty from climate modelling to soil properties, to earthquake shaking, to its impact on buildings. New Machine Learning (ML) models for chains of models using deep learning strategies to replace expensive simulators (climate, earthquake or vulnerability) by surrogate models (also known as emulators) efficiently propagate uncertainties. New ML in climate models accelerate climate predictions that represent key processes (e.g. precipitation) with higher fidelity at a much lower computational cost. The project combines these recent advances in a new context and develops new models to target extreme events.

Aon Impact Forecasting (IF) catastrophe models will be used to evaluate the impact on selected exposures. IF models include a framework that enables the sampling of events, and uncertainty propagation from hazard, vulnerability and exposure characteristics, to compute losses and improve financial planning in mitigation of disaster risk.

There are open avenues for research advances. Gaussian Processes techniques can be used in creating bespoke climate projections representing the local processes that may be key to the changes – say local conditions that influence precipitation. Furthermore, the spatial uncertainty in soil characteristics can be estimated though Gaussian Process and the accuracy of sampling can be investigated and propagated. New ways of probabilistically exploring risk in catastrophe models driven by diverse sources of uncertainties will be a potential research theme.

Another challenge will be to extend the remit of catastrophe modelling to illustrate the benefits of integrating the climate and its uncertainties to risks at the yearly scale. IF will provide support and training for catastrophe models, data on specific sites about exposure and earthquake shaking characteristics.

Other information

Applicant profile: Students from Statistics, with interests in Machine Learning and High Performance Computing.

  • https://arxiv.org/abs/2406.09551
  • https://www.aon.com/en/capabilities/reinsurance/catastrophe-model-insight