Natural hazards such as earthquakes, floods, and landslides pose significant risks to human populations, infrastructure, and ecosystems. These hazards rarely occur in isolation; they often trigger or interact with other processes, leading to multi-hazard effects. For instance, an earthquake may trigger landslides, which in turn can block rivers, causing floods. Accurate prediction and simulation of such multi-hazard events is essential for improving disaster preparedness, risk mitigation, and resilience. However, modeling these interconnected physical processes is extremely complex. Traditional numerical methods like finite element or finite difference models are highly effective in simulating individual hazards, but they face limitations when applied to multi-hazard scenarios, especially at large scales or in real time due to data scarcity and undefined/uncoupled physical models. The interplay between different physical processes involves nonlinearity, high-dimensional parameter spaces, and uncertainty, all of which make conventional methods impractical for multi-hazard systems. Physics-Informed Neural Networks (PINNs) offer a promising new approach. By embedding physical laws directly into neural networks, PINNs can simulate complex systems while ensuring that the predictions remain consistent with the governing equations of the underlying physics. This hybrid approach has shown great potential for reducing computational costs and improving the accuracy of models, even in data-scarce environments.

This PhD position is funded by the Dutch national earth and environmental sector plan, to advance the scientific field of AI on early warning systems. It aims to develop advanced Physics-Informed Neural Networks (PINNs) for modeling the complex interactions between multiple geophysical processes, specifically earthquakes, and landslides. The primary objective is to create a unified, scalable framework that accurately simulates these coupled hazards while addressing computational efficiency and uncertainty quantification. Central to this work is the coupling of partial differential equations (PDEs) governing seismic wave propagation, soil stability, and potentially multi-phase flow, allowing the PINNs to capture the nonlinear, multi-scale interactions between these hazards. This research will also investigate feedback mechanisms and energy transfer between hazards, embedding these principles within the network architecture to simulate how an initial hazard evolves into a multi-hazard event. By optimizing PINN architectures, this research aims to enable real-time, large-scale multi-hazard simulations, offering a powerful tool for early warning systems in data sparse regions.