Explainable Global Wildfire Prediction Models using Graph Neural Networks

TL;DR

The paper addresses global wildfire prediction by moving beyond CNN-based approaches that struggle with missing ocean data and long-range dependencies. It introduces a GCN-LSTM framework that represents global climate and fire data as a correlation-based graph, extracting spatial structure with a Graph Convolutional Layer and modeling temporal dynamics with an LSTM. On unseen 30-year JULES-INFERNO ensembles, the approach achieves superior predictive accuracy compared to LSTM, CAE-LSTM, and Conv-LSTM, while providing explainability through community detection and Integrated Gradients. This combination of performance and transparency supports actionable wildfire risk assessment and management, with future work focusing on real-world data assimilation and full climate modelling extensions.

Abstract

Wildfire prediction has become increasingly crucial due to the escalating impacts of climate change. Traditional CNN-based wildfire prediction models struggle with handling missing oceanic data and addressing the long-range dependencies across distant regions in meteorological data. In this paper, we introduce an innovative Graph Neural Network (GNN)-based model for global wildfire prediction. We propose a hybrid model that combines the spatial prowess of Graph Convolutional Networks (GCNs) with the temporal depth of Long Short-Term Memory (LSTM) networks. Our approach uniquely transforms global climate and wildfire data into a graph representation, addressing challenges such as null oceanic data locations and long-range dependencies inherent in traditional models. Benchmarking against established architectures using an unseen ensemble of JULES-INFERNO simulations, our model demonstrates superior predictive accuracy. Furthermore, we emphasise the model's explainability, unveiling potential wildfire correlation clusters through community detection and elucidating feature importance via Integrated Gradient analysis. Our findings not only advance the methodological domain of wildfire prediction but also underscore the importance of model transparency, offering valuable insights for stakeholders in wildfire management.

Figures (10)

• Figure 1: The JULES-INFERNO process overview
• Figure 2: CAE structure with a latent space dimension of 100 zhang2022
• Figure 3: The GCN-LSTM model structure
• Figure 4: Bar chart of the overall performances of the models from 1961-1990.
• Figure 5: Yearly comparison of model performances from 1961-1990.