AttentionFire_v1.0 - interpretable machine learning fire model for burned-area predictions over tropics

Aug 4, 2026

Upload by Aurora

African and South American wildfires drive over 70% of global burned areas and strongly couple to local climate across sub-seasonal to seasonal dynamics. Yet modeling this wildfire–climate linkage remains difficult because wildfire responses to climate variability and human influence vary across space and time. An interpretable ML fire model (AttentionFire_v1.0) was developed to disentangle these controls and improve burned-area predictions for ASA regions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Abstract

Introduction

UC Irvine UC Irvine Previously Published Works Title AttentionFire_v1.0: interpretable machine learning fire model for burned-area predictions over tropics Permalink [https://escholarship.org/uc/item/5k24d0wj](https://escholarship.org/uc/item/5k24d0wj) Journal Geoscientific Model Development, 16(3) ISSN 1991-959X Authors Li, Fa Zhu, Qing Riley, William Jet al. Publication Date 2023 DOI 10.5194/gmd-16-869-2023 Copyright Information This work is made available under the terms of a Creative Commons Attribution License, available at [https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/) Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California Geosci. Model Dev., 16, 869–884, 2023 [https://doi.org/10.5194/gmd-16-869-2023](https://doi.org/10.5194/gmd-16-869-2023)[ ](https://doi.org/10.5194/gmd-16-869-2023)© Author(s) 2023 . This work is distributed under the Creative Commons Attribution 4 .0 License. AttentionFire_v1.0: interpretable machine learning ﬁre model for burned-area predictions over tropics Fa Li 1,2 , Qing Zhu 1 , William J. Riley 1 , Lei Zhao3 , Li Xu4 , Kunxiaojia Yuan 1,2 , Min Chen5 , Huayi Wu2 , Zhipeng Gui6 , Jianya Gong6 , and James T. Randerson4 1 Climate and Ecosystem Sciences Division, Climate Sciences Department, Lawrence Berkeley National Laboratory, Berkeley, CA, USA 2 State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, China 3Department of Civil and Environmental Engineering, University of Illinois Urbana-Champaign, Champaign, IL, USA 4Department of Earth System Science, University of California Irvine, Irvine, CA, USA 5Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, Madison, WI, USA 6 School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, China Correspondence: Qing Zhu ([qzhu@lbl.gov](qzhu@lbl.gov)) Received: 26 July 2022 – Discussion started: 11 August 2022 Revised: 20 December 2022 – Accepted: 7 January 2023 – Published: 3 February 2023 Abstract. African and South American (ASA) wildﬁres account for more than 70 % of global burned areas and have strong connection to local climate for sub-seasonalto seasonal wildﬁre dynamics. However, representation of the wildﬁre–climate relationship remains challenging due tospatiotemporally heterogenous responses of wildﬁres to climate variability and human inﬂuences. Here, we developed an interpretable machine learning (ML) ﬁre model (AttentionFire_v1.0) to resolve the complex controls of climate and human activities on burned areas and to better predict burned areas over ASA regions. Our ML ﬁre model substantially improved predictability of burned areas for both spatial and temporal dynamics compared with ﬁve commonly used machine learning models. More importantly, the model revealed strong time-lagged control from climate wetness on the burned areas. The model also predicted that, under a highemission future climate scenario, the recently observed declines in burned area will reverse in South America in the near future due to climate changes. Our study provides a reliable and interpretable ﬁre model and highlights the importance of lagged wildﬁre–climate relationships in historical and future predictions. 1 Introduction Wildﬁres modify land surface characteristics, such as vegetation composition, soil carbon, surface runoff, and albedo, with signiﬁcant consequences for regional carbon, water, and energy cycles (Benavides-Solorio and MacDonald, 2001; Shvetsov et al., 2019; Randerson et al., 2006) . Over African and South American (ASA) regions, where more than 70 % of global burned area occurs, wildﬁres emit 􀀘 1.4 PgCyr􀀀1 (􀀘 65 % of global wildﬁre emissions; van Der Werf et al., 2017) and dust and aerosols that can alter regional climate through radiative processes (Etminan et al., 2016;

FAQs

What problem does AttentionFire_v1.0 address?

It addresses the challenge of representing the complex relationship between wildfires and climate, including heterogeneous space-time responses and human influences, to improve burned-area prediction over tropical African and South American regions.

How does AttentionFire_v1.0 perform compared with other ML models?

It substantially improves predictability of burned areas for both spatial and temporal dynamics compared with five commonly used machine learning models.

What key climate mechanism does the model reveal?

The model reveals strong time-lagged control from climate wetness on burned areas.

Aug 4, 2026

Upload by Aurora