Deep fire topology: Understanding the role of landscape spatial patterns in wildfire occurrence using artificial intelligence. (September 2021)

• Record Type: Journal Article
• Authors: Pais, Cristobal; Miranda, Alejandro; Carrasco, Jaime; Shen, Zuo-Jun Max
• Is Part Of: Environmental modelling & software. Volume 143(2021)
• Publication Date: 2021-09
• Subjects: Deep learning -- Machine learning -- Landscape topology -- Wildfire ignition risk -- Wildfire management -- Territorial planning; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2021.105122
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Increasing wildfire activity globally has become an urgent issue with enormous ecological and social impacts. In this work, we focus on analyzing and quantifying the influence of landscape topology, understood as the spatial structure and interaction of multiple land-covers in an area, on fire ignition. We propose a deep learning framework, Deep Fire Topology, to estimate and predict wildfire ignition risk. We focus on understanding the impact of these topological attributes and the rationale behind the results to provide interpretable knowledge for territorial planning considering wildfire ignition uncertainty. We demonstrate the high performance and interpretability of the framework in a case study, accurately detecting risky areas by exploiting spatial patterns. This work reveals the strong potential of landscape topology in wildfire occurrence prediction and its implications to develop robust landscape management plans. We discuss potential extensions and applications of the proposed method, available as an open-source software. Highlights: We show the impact of different landscape topologies on wildfire ignitions. We develop a novel and interpretable deep learning framework. The model detects and highlights low/high-risk land-cover topological patterns. We obtain accurate fire occurrence predictions only using land-cover data. Our framework can be applied in any field exploiting computer vision.