Douglas Radford
Doug aims to contribute to improving the resilience of social, infrastructure and environmental systems against the complex risks associated with natural hazards and climate variability. Doug’s research applies robust environmental modelling approaches to quantify these risks, with an interest in optimising risk management activities in order to further a communities’ social, economic and environmental values. His interests include natural hazard risk, uncertainty, systems thinking and adaptive planning. Doug is a Westpac Future Leaders scholar, Associate Student at Natural Hazards Research Australia and a recipient of the University Medal at the University of Adelaide. He will complete his PhD studies in 2025.
Innovation title:
Have your cake and eat it too! Using metamodels and machine learning to optimise bushfire risk management
Abstract:
Predicting the probability that a given location will be burnt by a bushfire is an important part of understanding the risks that bushfires pose and how our management actions (e.g., prescribed burning) can reduce this risk. Existing methods to quantify this burn probability involve simulating the spread of many thousands of individual bushfires (e.g., landscape fire simulation models that use a fire simulator like Spark) and are thus highly computationally expensive. To reduce this expense, we propose innovative strategies that enable the development of computationally efficient machine learning assisted metamodels for estimating burn probability. We demonstrate the use of these strategies in a case study in South Australia. Our implementation uses artificial neural networks as the metamodel to emulate the outputs of a landscape fire simulation model. The metamodel can predict burn probabilities with high accuracy (+/- 7.4% error) and only requires 0.6% of the computational time compared to existing landscape fire simulation approaches. Interrogating the metamodel’s structure helps us to better understand how bushfires spread across landscapes and builds confidence in the metamodel’s robustness. The computationally efficient metamodel makes it possible to evaluate many thousands of alternative fuel treatment plans. The power of this capability is shown by using the metamodel to optimise fuel treatment plans to best achieve different bushfire risk management objectives. The tools presented can be used by landscape managers to guide their strategic planning of fuel treatments in a way that best achieves their communities’ specific objectives.
Previous post:
Coming up next: