Jonathan Koh Boon Han | EPFL Graph Search

I am currently a PhD candidate at EPFL under the supervision of Prof. Anthony C. Davison. My broad research focus is on Extreme Value Theory (EVT), a branch of probability and statistics that deals with the analysis of extreme and rare events. My narrower research interest lies in the intersection of EVT, spatial statistics, machine learning and Bayesian statistics, applied to environmental research problems.

Gradient boosting with extreme-value theory for wildfire prediction

Jonathan Koh Boon Han

This paper details the approach of the team Kohrrelation in the 2021 Extreme Value Analysis data challenge, dealing with the prediction of wildfire counts and sizes over the contiguous US. Our approach uses ideas from extreme-value theory in a machine learning context with theoretically justified loss functions for gradient boosting. We devise a spatial cross-validation scheme and show that in our setting it provides a better proxy for test set performance than naive cross-validation. The predictions are benchmarked against boosting approaches with different loss functions, and perform competitively in terms of the score criterion, finally placing second in the competition ranking.

2023

Spatiotemporal wildfire modeling through point processes with moderate and extreme marks

Jonathan Koh Boon Han

Accurate spatiotemporal modeling of conditions leading to moderate and large wildfires provides better understanding of mechanisms driving fire-prone ecosystems and improves risk management. Here, we develop a joint model for the occurrence intensity and the wildfire size distribution, by combining extreme-value theory and point processes within a novel Bayesian hierarchical model, and use it to study daily summer wildfire data for the French Mediterranean basin during 1995–2018. The occurrence component models wildfire ignitions as a spatiotemporal log-Gaussian Cox process. Burnt areas are numerical marks attached to points and are considered as extreme if they exceed a high threshold. The size component is a two-component mixture varying in space and time that jointly models moderate and extreme fires. We capture nonlinear influence of covariates (Fire Weather Index, forest cover) through component-specific smooth functions which may vary with season. We propose estimating shared random effects between model components to reveal and interpret common drivers of different aspects of wildfire activity. This increases parsimony and reduces estimation uncertainty, giving better predictions. Specific stratified subsampling of zero counts is implemented to cope with large observation vectors. We compare and validate models through predictive scores and visual diagnostics. Our methodology provides a holistic approach to explaining and predicting the drivers of wildfire activity and associated uncertainties.

2023

Spatiotemporal modelling of extreme wildfires and severe thunderstorm environments

Jonathan Koh Boon Han

Environmental extreme events can have devastating impacts on society when they interact with vulnerable human and natural systems. Such events can result from natural causes, like phenomena related to the El Ni~no-Southern Oscillation or decadal/multi-decadal climate variations. These causes can follow an increase in human activity, e.g., through land-use changes or anthropogenic climate change, that can influence the frequency, intensity, spatial extent and timing of these events, and spur unprecedented extremes. To accurately understand and quantify the risks associated with these events, it is important to identify trends related to these causes, which may be measured or unmeasured. Fitting models for rare events is inherently difficult because of the paucity of data available. The most destructive extreme events are rarely isolated in space and time, so one must account for their spatial and temporal dependencies. This thesis deals with the parametric modelling of severe thunderstorms and wildfires using models motivated from limiting probabilistic results. The first part of this thesis explores influences on the magnitude and spatial extent of extremes of environments related to severe US thunderstorms. Our results show that the risk from severe thunderstorms in April and May is increasing in parts of the US where it was already high, and that the risk from storms in February increases during La Ni~na years. We also show that these extremes are more localized during spring/summer seasons than in the winter, and find that some of these seasonal differences are more pronounced during El Ni~no years.The second part of the thesis deals with predicting and explaining the spatial extent, frequency, intensity and timing of wildfires using meteorological and land-use covariates. Our first approach uses ideas from extreme-value theory in a machine learning context to give good prediction of the distributional tails of our data. The second approach uses a novel Bayesian hierarchical model designed specifically for extreme wildfires. We show that wildfire risk on the French Mediterranean basin is affected by significant random effects related to land-use and policy changes, and a seasonally-varying fire-weather index.

EPFL2022

Spatiotemporal modelling of extreme wildfires and severe thunderstorm environments

Jonathan Koh Boon Han

EPFL2022