Deep Learning Provides Substantial Improvements to County‐Level Fire Weather Forecasting Over the Western United States

Son., R.E. Lambert; Ma, Po‐Lun; Wang, Hailong; Rasch, Philp J.; Wang, Shih‐Yu; Jeong, Jee‐Hoon; Lim, Kyo‐Sun Sunny; Yoon, Jin‐Ho

doi:10.1029/2022ms002995

Cited by 5 publications

(8 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Second, the traditional FWIs have been developed using traditional statistical methods, which have been known to underperform machine learning (ML)-based models (e.g. Son et al 2022). Finally, traditional FWIs are mostly based on a limited number of independent variables; specifically, some of the variables described in the previous paragraph are not reflected in the traditional FWIs (Kondylatos et al 2022).…”

Section: Introductionmentioning

confidence: 99%

Developing novel machine-learning-based fire weather indices

Shmuel

Heifetz²

2023

Mach. Learn.: Sci. Technol.

View full text Add to dashboard Cite

Accurate wildfire risk estimation is an essential yet challenging task. As the frequency of extreme fire weather and wildfires is on the rise, forest managers and firefighters require accurate wildfire risk estimations to successfully implement forest management and firefighting strategies. Wildfire risk depends on non-linear interactions between multiple factors; therefore, the performance of linear models in its estimation is limited. To date, several traditional fire weather indices (FWIs) have been commonly used by weather services, such as the Canadian FWI. Traditional FWIs are primarily based on empirical and statistical analyses. In this paper, we propose a novel FWI that was developed using machine learning (ML) – the MLFWI. We present the performance of the MLFWI and compare it withvarious traditional FWIs. We find that the MLFWI significantly outperforms traditional indices in predicting wildfire occurrence, achieving an AUC score of 0.99 compared to 0.62-0.80. We recommend applying the MLFWI in wildfire warning systems.

show abstract

Section: Introductionmentioning

confidence: 99%

Developing novel machine-learning-based fire weather indices

Shmuel

Heifetz²

2023

Mach. Learn.: Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…Similarly, overestimated durations of burnt fraction and LAI in SEAS coincide with each other (Figs 4l and S7l). To effectively address internal biases of physics-based models, it was suggested to merge deep learning as an external post-processing method (Reichstein et al, 2019;Son et al, 2022). However, this approach is not directly applicable in this study due to dynamical interactions between predictors and DGVMs.…”

Section: Discussionmentioning

confidence: 99%

“…To address the spatiotemporal context for wildfire danger, (Kondylatos et al, 2022) applied a convolutional-LSTM network (Shi et al, 2015) integrating meteorological, environmental, and anthropogenic drivers. Other studies leveraged ML/DL methods to characterize various aspects of fire occurrence, such as fire weather (Son et al, 2022), lightning ignition (Coughlan et al, 2021), fire susceptibility (Zhang et al, 2021) and fuel availability (D'Este et al, 2021).…”

Section: Introductionmentioning

confidence: 99%

Integration of a deep-learning-based fire model into a global land surface model

Son.

Stacke

Gayler

et al. 2023

Preprint

View full text Add to dashboard Cite

Fire is a crucial factor in terrestrial ecosystems playing a role in disturbance for vegetation dynamics. Process-based fire models quantify fire disturbance effects in stand-alone dynamic global vegetation models (DGVMs) and their advances have incorporated both descriptions of natural processes and anthropogenic drivers. Nevertheless, these models show limited skill in modeling fire events at the global scale, due to stochastic characteristics of fire occurrence and behavior as well as the limits in empirical parameterizations in process-based models. As an alternative, machine learning has shown the capability of providing robust diagnostics of fire regimes. Here, we develop a deep-learning-based fire model (DL-fire) to estimate daily burnt area fraction at the global scale and couple it within JSBACH4, the land surface model used in the ICON ESM. The stand-alone DL-fire model forced with meteorological, terrestrial and socio-economic variables is able to simulate global total burnt area, showing 0.8 of monthly correlation (rm) with GFED4 during the evaluation period (2011-15). The performance remains similar with the hybrid modeling approach JSB4-DL-fire (rm=0.79) outperforming the currently used uncalibrated standard fire model in JSBACH4 (rm=-0.07). We further quantify the importance of each predictor by applying layer-wise relevance propagation (LRP). Overall, land properties, such as fuel amount and water content in soil layers, stand out as the major factors determining burnt fraction in DL-fire, paralleled by meteorological conditions over tropical and high latitude regions. Our study demonstrates the potential of hybrid modeling in advancing fire prediction in ESMs by integrating deep learning approaches in physics-based dynamical models.

show abstract

“…Biomass burning affects the structure and dynamics of ecological processes (McLauchlan et al., 2020). Fire emissions alter atmospheric composition of trace gases and aerosol particles (Koppmann et al., 2005; Son, Kim, et al., 2022; Son, Ma, et al., 2022), with subsequent influences on land surface albedo (López‐Saldaña et al., 2015), energy budgets (F. Li et al., 2017), climate (Liu et al., 2019; Voulgarakis & Field, 2015) and global biogeochemical cycles (Carcaillet et al., 2002; Crutzen & Andreae, 1990). Present‐day global carbon emissions due to fire are approximately 1.5–3.0 PgC/yr (van der Werf et al., 2017).…”

Section: Introductionmentioning

confidence: 99%

“…(2022) incorporated meteorological, environmental and anthropogenic drivers into a convolutional‐LSTM to comprehensively address the spatiotemporal context for wildfire danger prediction. Other studies leveraged ML/DL methods to characterize various aspects of fire occurrence, such as fire weather (Son, Kim, et al., 2022; Son, Ma, et al., 2022), lightning ignition (Coughlan et al., 2021), fire susceptibility (Zhang et al., 2021) and fuel availability (D’Este et al., 2021).…”

Section: Introductionmentioning

confidence: 99%

Integration of a Deep‐Learning‐Based Fire Model Into a Global Land Surface Model

Son,

Stacke,

Gayler

et al. 2024

J Adv Model Earth Syst

Self Cite

View full text Add to dashboard Cite

Fire is a crucial factor in terrestrial ecosystems playing a role in disturbance for vegetation dynamics. Process‐based fire models quantify fire disturbance effects in stand‐alone dynamic global vegetation models (DGVMs) and their advances have incorporated both descriptions of natural processes and anthropogenic drivers. Nevertheless, these models show limited skill in modeling fire events at the global scale, due to stochastic characteristics of fire occurrence and behavior as well as the limits in empirical parameterizations in process‐based models. As an alternative, machine learning has shown the capability of providing robust diagnostics of fire regimes. Here, we develop a deep‐learning‐based fire model (DL‐fire) to estimate daily burnt area fraction at the global scale and couple it within JSBACH4, the land surface model used in the ICON‐ESM. The stand‐alone DL‐fire model forced with meteorological, terrestrial and socio‐economic variables is able to simulate global total burnt area, showing 0.8 of monthly correlation (rm) with GFED4 during the evaluation period (2011–2015). The performance remains similar with the hybrid modeling approach JSB4‐DL‐fire (rm = 0.79) outperforming the currently used uncalibrated standard fire model in JSBACH4 (rm = −0.07). We further quantify the importance of each predictor by applying layer‐wise relevance propagation (LRP). Overall, land properties, such as fuel amount and water content in soil layers, stand out as the major factors determining burnt fraction in DL‐fire, paralleled by meteorological conditions over tropical and high latitude regions. Our study demonstrates the potential of hybrid modeling in advancing fire prediction in ESMs by integrating deep learning approaches in physics‐based dynamical models.

show abstract

Deep Learning Provides Substantial Improvements to County‐Level Fire Weather Forecasting Over the Western United States

Cited by 5 publications

References 51 publications

Developing novel machine-learning-based fire weather indices

Developing novel machine-learning-based fire weather indices

Integration of a deep-learning-based fire model into a global land surface model

Integration of a Deep‐Learning‐Based Fire Model Into a Global Land Surface Model

Contact Info

Product

Resources

About