A combined statistical and machine learning approach for spatial prediction of extreme wildfire frequencies and sizes

Abstract

Motivated by the Extreme Value Analysis 2021 (EVA 2021) data challenge we propose a method based on statistics and machine learning for the spatial prediction of extreme wildfire frequencies and sizes. This method is tailored to handle large datasets, including missing observations. Our approach relies on a four-stage high-dimensional bivariate sparse spatial model for zero-inflated data, which is developed using stochastic partial differential equations (SPDE). In Stage 1, the observations are categorized in zero/nonzero categories and are modeled using a two-layered hierarchical Bayesian sparse spatial model to estimate the probabilities of these two categories. In Stage 2, before modeling the positive observations using a spatially-varying coefficients, smoothed parameter surfaces are obtained from empirical estimates using fixed rank kriging. This approximate Bayesian method inference was employed to avoid the high computational burden of large spatial data modeling using spatially-varying coefficients. In Stage 3, the standardized log-transformed positive observations from the second stage are further modeled using a sparse bivariate spatial Gaussian process. The Gaussian distribution assumption for wildfire counts developed in the third stage, is computationally effective but erroneous. Thus in Stage 4, the predicted values are rectified using Random Forests. Posterior inference is drawn for Stages 1 and 3 using Markov chain Monte Carlo (MCMC) sampling. A cross-validation scheme is then created for the artificially Extreme wildfire frequency and size prediction generated gaps, and the EVA 2021 prediction scores of the proposed model are compared to those obtained using certain natural competitors.

A marginal modelling approach for predicting wildfire extremes across the contiguous United States

“…We also found that for the non-missing CNT and BA observations, the probability of observing a zero observation exceeded 0.999 for both variables when lc (18) i > 0.94 , where lc (18) i denotes the proportion of each location covered by water. Therefore, for any i ∈ CNT val ( i ∈ BA val ) with lc (18)…”

“…For example, [15] show that the Forest Fire Danger Index, typically used in Australia, is inadequate for predicting the behaviour of moderate to high-intensity wildfires. Machine learning techniques have also been adopted for wildfire modelling: [16] and [17] use deep learning techniques; [18] present a four-stage process including a random forest algorithm; and [19] develops a gradient boosting model trained with loss functions appropriate for predicting extreme values. We take a simpler, marginalbased approach.…”