Prediction of regional wildfire activity in the probabilistic Bayesian framework of Firelihood

Abstract

Modeling wildfire activity is crucial for informing science‐based risk management and understanding the spatiotemporal dynamics of fire‐prone ecosystems worldwide. Models help disentangle the relative influences of different factors, understand wildfire predictability, and provide insights into specific events. Here, we develop Firelihood, a two‐component, Bayesian, hierarchically structured, probabilistic model of daily fire activity, which is modeled as the outcome of a marked point process: individual fires are the points (occurrence component), and fire sizes are the marks (size component). The space‐time Poisson model for occurrence is adjusted to gridded fire counts using the integrated nested Laplace approximation (INLA) combined with the stochastic partial differential equation (SPDE) approach. The size model is based on piecewise‐estimated Pareto and generalized Pareto distributions, adjusted with INLA. The Fire Weather Index (FWI) and forest area are the main explanatory variables. Temporal and spatial residuals are included to improve the consistency of the relationship between weather and fire occurrence. The posterior distribution of the Bayesian model provided 1,000 replications of fire activity that were compared with observations at various temporal and spatial scales in Mediterranean France. The number of fires larger than 1 ha across the region was coarsely reproduced at the daily scale, and was more accurately predicted on a weekly basis or longer. The regional weekly total number of larger fires (10–100 ha) was predicted as well, but the accuracy degraded with size, as the model uncertainty increased with event rareness. Local predictions of fire numbers or burned areas also required a longer aggregation period to maintain model accuracy. The estimation of fires larger than 1 ha was also consistent with observations during the extreme fire season of the 2003 unprecedented heat wave, but the model systematically underrepresented large fires and burned areas, which suggests that the FWI does not consistently rate the actual danger of large fire occurrence during heat waves. Firelihood enabled a novel analysis of the stochasticity underlying fire hazard, and offers a variety of applications, including fire hazard predictions for management and projections in the context of climate change.

Spatiotemporal wildfire modeling through point processes with moderate and extreme marks

“…We propose important extensions to Pimont et al (2021) and other works cited above. Since the heavy tails of burnt areas lead to a dominant influence of the most extreme wildfires, we focus on accurate modeling of the distribution of extreme wildfires, and its spatiotemporal variation.…”

“…Joseph et al (2019) estimate separate regression models with random effects for occurrence numbers in areal units and for sizes, and they study posterior predictive distributions for block maxima of wildfire sizes. Pimont et al (2021) developed a marked spatiotemporal log-Gaussian Cox process model, called Firelihood, for daily data using extreme-value techniques for modeling the burnt areas as marks, applying the integrated nested Laplace approximation (INLA, Rue et al 2009) for Bayesian inference. Their distribution of wildfire sizes over positive values is a mixture, of which each component is supported on the interval of a partition of the positive half-line; a GPD is specified for the most extreme interval.…”

“…In the present paper, we develop a more flexible spatiotemporal structure and data subsampling schemes for joint estimation and fully Bayesian inference of all components. We take care to achieve subsample sizes that allow running models on standard desktop computers, in contrast to often highly computer-intensive approaches in the literature (e.g., Joseph et al 2019, Pimont et al 2021, Opitz, Bonneu & Gabriel 2020 requiring high memory resources.…”

Landslide susceptibility maps of Italy: lesson learnt from dealing with multiple landslide classes and the uneven spatial distribution of the national inventory

“…INLA is a popular tool for specifying and inferring Bayesian models, and is used in a wide range of relevant applications (Opitz et al, 2018;Pimont et al, 2021;Titti et al, 2021).…”

Simulating Forest Fire Spread with Cellular Automation Driven by a LSTM Based Speed Model

“…However, LR and RF models are more likely to produce under-fitting. Pimont [29] developed a Firelihood system, which is a two-component, Bayesian, hierarchically structured, probabilistic model of fire, but this approach has higher data requirements. With the development of neural networks, Zheng [30] realized fire spread, which used CA with extreme learning machine (ELM).…”