Introduction
Accurate predictions of natural hazard impacts are essential to reducing disaster risk.
However, leading numerical models used to simulate physical impacts of coastal storms, wildfires, and river floods still struggle to match empirical observations of disaster footprints. New research (by lead supervisor Prof Eli Lazarus) offers a novel explanation for this persistent prediction gap, showing that while deliberate, concurrent, mitigating interventions in the physical processes of a natural hazard event can dramatically alter its outcome, such interventions are missing from high-fidelity numerical models of hazard impacts at the event scale (Lazarus, 2026).
Typically handled by expert practitioners (e.g., vehicle operators, wildland firefighters, specialist crews), these interventions – termed "emergency landscaping" (Lazarus, 2026) – include plowing sand off beachfront roads during a coastal storm, cutting firebreaks at the burning front of a wildfire, stacking sandbags during a river flood, and hosing down a lava flow to slow its advance. But compared to the physics of natural hazards (e.g., material transport, propagation, fluid flow), the human agency of emergency landscaping is poorly constrained. Quantifying how targeted actions by front-line operators affect the dynamics of a hazard event has so far remained beyond the reach of conventional approaches to observing and modelling hazard impacts.
This project therefore will take an unconventional approach to hazard impact observation.
Here, we will characterise and quantify geospatial signatures of emergency landscaping by developing and applying novel AI-driven geospatial analysis of remotely sensed imagery from before and after recurrent hazard events. We will focus on storm impacts on low-lying sandy coastlines, where people and critical infrastructure are disproportionately exposed to damaging hazard events and disaster risk is high. Beyond coastal storms, the results of this studentship will demonstrate the efficacy of this approach for addressing comparable prediction gaps in other natural hazard systems. Our experienced interdisciplinary supervisory team, which bridges physical geography, remote sensing, spatial ecology, and computer science, is uniquely equipped to support a doctoral student in this endeavour.
Methodology
To quantify empirical signatures of coastal emergency landscaping, this project will adapt an emerging data-science approach called causal machine-learning (or causal inference), which links perturbations with effects through statistical analysis beyond correlation (Kaddour et al., 2022; Feuerriegel et al., 2024). Causal inference from imagery has been applied in medicine (Sanchez et al., 2022), but we will extend its application to for physical geography.
The project will involve compiling a catalogue of pre- and post-storm aerial imagery (2D) and InSAR data (3D) from selected coastal sites known to implement emergency landscaping, and from neighbouring natural reaches for comparison. We will then develop a generalisable causal inference workflow to characterise and quantify spatial patterns of storm-driven sediment deposition and redistribution associated with coastal emergency landscaping. Differences in spatial patterns from neighbouring human-altered and natural reaches will reveal an emergency landscaping "signature" that may not be discernible by conventional methods of image classification. The causal machine learning analysis will allow us to quantify emergency landscaping by earth-moving vehicles as a physical mechanism of coastal sediment transport, relative to its storm-driven natural counterpart. Empirical measurement of sediment volume flux by earth-moving vehicles in response to storm-driven sediment deposition is a key constraint to parameterise in high-fidelity numerical modelling of coastal storm impacts (Lazarus & Goldstein, 2019).
References:
Feuerriegel, S. et al. (2024) Causal machine learning for predicting treatment outcomes. Nature Medicine 30(4), 958–968. https://doi.org/10.1038/s41591-024-02902-1
Kaddour, J., Lynch, A., Liu, Q., Kusner, M. J., Silva, R. (2022) Causal machine learning: a survey and open problems. arXiv [preprint]. https://doi.org/10.48550/arXiv.2206.15475
Lazarus, E. D. (2026) Control of natural hazard events through emergency landscaping. EarthArXiv [preprint]. https://doi.org/10.31223/X52N2B
Lazarus, E. D., Goldstein, E. B. (2019) Is there a bulldozer in your model? J. Geophys. Res.: Earth Surf. 124, 696–699. https://doi.org/10.1029/2018JF004957
Sanchez, P., Voisey, J. P., Xia, T., Watson, H. I., O’Neil, A. Q., Tsaftaris, S. A. (2022) Causal machine learning for healthcare and precision medicine. Royal Society Open Science, 9(8). https://doi.org/10.1098/rsos.220638