FireBench: A High-fidelity Ensemble Simulation Framework for Exploring Wildfire Behavior and Data-driven Modeling

Abstract: Background. Wildfire research uses ensemble methods to analyze fire behaviors and assess uncertainties. Nonetheless, current research methods are either confined to simple models or complex simulations with limits. Modern computing tools could allow for efficient, high-fidelity ensemble simulations. Aims. This study proposes a high-fidelity ensemble wildfire simulation framework for studying wildfire behavior, ML tasks, fire-risk assessment, and uncertainty analysis. Methods. In this research, we present a simulation framework that integrates the Swirl-Fire large-eddy simulation tool for wildfire predictions with the Vizier optimization platform for automated run-time management of ensemble simulations and large-scale batch processing. All simulations are executed on tensor-processing units to enhance computational efficiency. Key results. A dataset of 117 simulations is created, each with 1.35 billion mesh points. The simulations are compared to existing experimental data and show good agreement in terms of fire rate of spread. Computations are done for fire acceleration, mean rate of spread, and fireline intensity. Conclusions. Strong coupling between these 2 parameters are observed for the fire spread and intermittency. A critical Froude number that delineates fires from plume-driven to convection-driven is identified and confirmed with literature observations. Implications. The ensemble simulation framework is efficient in facilitating parametric wildfire studies.

• The paper presents a high-fidelity ensemble simulation framework that integrates SWIRL-FIRE with VIZIER on TPUs to improve wildfire behavior analysis.
• It employs a structured Cartesian mesh with over 1.35 billion grid points across 117 cases to capture complex wind, slope, and fire dynamics.
• The framework generates the FIREBENCH dataset, enabling robust ML model training and challenging traditional fire-spread models like Rothermel's.

High-Fidelity Ensemble Simulation Framework for Analyzing Wildland-Fire Behavior: A Technical Overview

The research paper introduces an advanced ensemble simulation framework designed to interrogate wildland-fire behavior and facilitate benchmarking for machine learning models. Developed by a collaboration between Google and Stanford University, this framework capitalizes on modern computational advancements to enable high-fidelity simulations that address the complexities and uncertainties inherent in wildfire modeling.

Methodological Innovations

The presented framework integrates the open-source large-eddy simulation tool SWIRL-FIRE with Google's VIZIER optimization platform. The primary aim is to enhance simulations' computational efficiency and accuracy, notably performed on tensor-processing units (TPUs). By adopting a structured Cartesian mesh with over 1.35 billion grid points, the framework achieves a substantial scale and resolution, allowing detailed representations of fire dynamics over complex terrains.

Significantly, the simulations utilize a total of 117 cases, capturing variabilities in wind speed and slope, essential factors influencing wildland fire behavior. The simulations align with experimentally validated data (Morandini & Silvani, 2010), particularly regarding fire rate of spread and the transition between plume-dominated and wind-dominated fires. The research identifies critical parameters, such as the Froude number, which it uses to delineate fire behavior regimes and corroborate observations in extant literature.

Numerical and Statistical Outputs

Quantitatively, the simulations provide a comprehensive dataset (FIREBENCH), which yields insights into fire acceleration, mean rate of spread (ROS), and fireline intensity. A notable conclusion is the observed strong coupling effects between wind and slope on fire dynamics, challenging the assumptions of linear superposition often used in simpler fire-spread models.

The study highlights discrepancies between the predictions of this model and traditional empirical models, such as Rothermel's, particularly under conditions characterized by plume-dominated regimes. This emphasizes the framework's capability to capture the complex interactions of fire-induced turbulence and atmospheric dynamics due to its high-fidelity nature.

Implications and Future Directions

The implications of this study are manifold. Practically, the framework facilitates large-scale parametric studies that are crucial for optimizing fire management strategies and improving predictive models of wildfire spread. Theoretically, the framework's detailed data is invaluable for training sophisticated ML models, potentially advancing the predictive accuracy of these models under varied environmental conditions.

Looking forward, the research opens pathways for extending ensemble simulation frameworks to incorporate additional environmental variables and to test scalability further using different computational infrastructures. Moreover, the FIREBENCH dataset's availability for the research community could expedite development in ML training and evaluation, supporting the broader objective of mitigating wildfire risks through scientific advancement.

In summary, this work represents a critical contribution to wildfire research, addressing existing limitations in predictive modeling and offering a scalable framework for future scientific inquiries.