Current State of Wildfire Modeling

Comprehensive survey of wildfire modeling capabilities, limitations, data requirements, and recommendations for future investments

Current State of Wildfire Modeling:

A Comprehensive Survey of Capabilities, Limitations, Data Requirements, and Recommendations for Future Investments

Prepared by RallyPoint One LLC
Supporting the NSF CO–WY Engine ASCEND Climate Resilience Initiative

---

Executive Summary

This white paper provides a comprehensive survey of the current state of wildfire modeling, examining capabilities, limitations, data requirements, and strategic priorities for advancing operational prediction systems in support of the NSF CO–WY Engine’s ASCEND Climate Resilience Initiative.¹ The analysis reveals a field at an inflection point: traditional physics-based models have reached practical limits in their ability to deliver real-time, probabilistic predictions demanded by operational users, while emerging AI/ML technologies offer pathways to transformative improvements but lack the validation and integration frameworks necessary for high-stakes deployment.²

Current State Assessment. Existing wildfire models span a spectrum from fast-but-approximate empirical tools (e.g., FARSITE and other Rothermel-based systems) to slower, more physically complete simulators (e.g., coupled atmosphere–fire models like WRF-Fire). Studies comparing empirical fire behavior tools in IFTDSS and other operational systems highlight that empirical/semi-empirical tools trade accuracy under extreme conditions for speed and usability, while coupled models deliver higher physical fidelity at substantial computational cost.³⁴⁵ No widely deployed system currently delivers both sub-5-minute response times and consistently >80% perimeter accuracy across a wide range of events in true operational settings; the speed–accuracy trade-off still constrains routine ensemble forecasting and rapid observational integration.⁶⁷⁸⁹

The 80–20 Rule in Wildfire Prediction. Classical fire behavior literature and sensitivity analyses consistently identify wind, slope, fuel type, and fuel moisture as the dominant controls on landscape-scale spread.^10111213 Wind and slope directly affect flame tilt and heat transfer, while fuel type and moisture control available energy and ignition thresholds.¹⁴¹⁵ Secondary factors—fire–atmosphere coupling, spotting/ember transport, crown fire dynamics—have substantial effects in specific regimes but contribute a smaller share of explained variance in many empirical validations relative to the core four.¹⁶¹⁷¹⁸ This dominance structure motivates SciML strategies that focus surrogate development on these primary factors to achieve large speedups while preserving most predictive skill.¹⁹²⁰

Data Ecosystem Challenges. The wildfire data landscape is fragmented across USFS, NOAA, USGS, NASA, state agencies, utilities, and commercial providers, with heterogeneous formats and update cadences.²¹²²²³ LANDFIRE provides 30 m fuels and vegetation data nationally, but updates on 2–5 year cycles and misses recent disturbances and rapid WUI growth.²⁴ Weather observations from RAWS stations are typically spaced tens of kilometers apart, leaving major gaps in complex terrain and forcing heavy reliance on gridded analyses like HRRR.²⁵²⁶ Structure-level WUI data are sparse or inconsistent, with census-based WUI layers providing decadal block-level information but not parcel-scale attributes; NIST and partners have called out WUI data gaps as a major barrier for structure ignition modeling.²⁷²⁸

ML Weather Prediction Revolution. Recent machine learning weather models such as GraphCast, Pangu-Weather, and FourCastNet have shown they can match or exceed leading global NWP systems (e.g., ECMWF) at comparable resolutions while running ~1,000× faster at inference.²⁹³⁰³¹ These models are trained on multi-decadal reanalyses (e.g., ERA5) and can produce 10-day global forecasts at 0.25° in seconds to minutes on modern accelerators.³²³³ Their speed enables very large ensembles (O(100–1000) members), opening new possibilities for probabilistic wildfire prediction where weather uncertainty is a major driver of overall forecast uncertainty.³⁴³⁵

VVUQ Challenge. The verification, validation, and uncertainty quantification (VVUQ) practices in wildfire modeling lag behind those in weather and climate.³⁶³⁷ Many operational tools lack automated test suites or continuous integration; validation studies often use fewer than a few dozen fires with heterogeneous metrics and inconsistent datasets, making cross-model comparisons difficult.^38394041 Recent work on UQ for coupled wildfire–atmosphere systems and USGS/NIST assessments of model evaluation emphasizes the need for standardized test cases, metrics, and probabilistic verification, similar to WMO practices for weather forecasting.⁴²⁴³⁴⁴

Strategic Investment Priorities. The paper recommends on the order of $10–15M over three years directed at: 1. Real-time physics-based surrogates (SciML) to achieve 100–1000× speedups for dominant processes while preserving physical realism.⁴⁵⁴⁶⁴⁷ 2. ML weather downscaling, leveraging GraphCast/Pangu/FourCastNet-style global models and learning mappings to sub-kilometer, fire-relevant resolutions.⁴⁸⁴⁹⁵⁰ 3. Marshall Fire digital twin reconstruction as a high-fidelity, multi-modal benchmark dataset for structure-level WUI and spread validation.⁵¹⁵² 4. Ensemble-based UQ systems that provide calibrated probabilistic forecasts rather than single deterministic scenarios.⁵³⁵⁴⁵⁵ 5. WUI structure ignition modeling combining NIST-style physics, WUI data, and ML classifiers to assess risk across thousands of buildings.⁵⁶⁵⁷⁵⁸

Supporting investments in data infrastructure, standards, and training ensure these research advances translate into operational practice.⁵⁹⁶⁰⁶¹

Path Forward. Successful transition from research to operations requires: coordinated funding across agencies, open-source development and rigorous validation, co-development with operational agencies, and workforce programs that bridge fire science, AI, and decision science.⁶²⁶³⁶⁴ NSF’s Engines program, and the CO–WY Engine specifically, are positioned to serve as national exemplars of integrated, AI-enhanced wildfire prediction and resilience systems.⁶⁵

---

1. Introduction

1.1 The Wildfire Challenge in a Changing Climate

Wildfire activity in the western U.S. has increased significantly over recent decades, with multiple analyses reporting large increases in burned area and fire-season length linked to warming, snowpack changes, and fuel aridification.⁶⁶⁶⁷ Economic impacts of major fire seasons, when accounting for suppression, property loss, health impacts, and ecosystem damage, routinely reach tens of billions of dollars per year.⁶⁸ The 2021 Marshall Fire in Colorado—destroying over 1,000 structures in a winter, suburban WUI setting—is widely cited as an example of how changing climate, land use, and WUI expansion are creating atypical and high-consequence fire regimes.⁶⁹⁷⁰⁷¹

These changing conditions demand capabilities that current modeling workflows struggle to provide: (1) real-time forecasts integrating rapidly updating observations; (2) probabilistic guidance that quantifies uncertainty for risk-informed decisions; and (3) scalability across diverse landscapes, WUI forms, and organizational contexts.⁷²⁷³⁷⁴

1.2 The Operational Prediction Gap

Despite decades of work on fire behavior and coupled fire–atmosphere models, operational wildfire prediction in many jurisdictions still relies heavily on expert judgment supported by semi-empirical tools like FARSITE and related Rothermel-based systems.⁷⁵⁷⁶⁷⁷ FARSITE (and closely related minimum-travel-time fire spread systems) are integrated into WFDSS and IFTDSS and are widely used for planning and incident support, but multiple reviews note their limitations in extreme, plume-driven events and the lack of native probabilistic outputs.⁷⁸⁷⁹⁸⁰

Coupled systems such as WRF-Fire/WRF-SFIRE can represent two-way fire–atmosphere interactions and have been applied to individual incidents with improved realism, but typical configurations require hours of HPC runtime for multi-hour forecasts, limiting their operational use to research or special studies.⁸¹⁸²⁸³ The result is a structural gap: fast tools that are approximate but timely versus accurate, physically rich tools that are too slow for routine, ensemble-based, real-time operations.^84858687

1.3 Emerging Opportunities: Scientific Machine Learning and AI

SciML approaches that combine PDE-based solvers with neural surrogates, PINNs, and operator learning have delivered 100–1000× speedups in several domains (e.g., climate parameterizations, combustion kinetics) while maintaining acceptable physical fidelity within the training domain.⁸⁸⁸⁹⁹⁰ At the same time, ML weather models (GraphCast, Pangu-Weather, FourCastNet) have demonstrated that global, medium-range forecasts can be produced at or above ECMWF-quality with two to three orders of magnitude lower computational cost.⁹¹⁹²⁹³

These developments suggest that similar hybrid strategies in wildfire modeling—where high-fidelity FIRETEC/WRF-Fire runs are used to train surrogates focused on the dominant 80–20 physics—could make ensemble wildfire prediction and real-time UQ operationally feasible.^94959697

1.4 The NSF CO–WY Engine Context

NSF’s Regional Innovation Engines program aims to build place-based innovation ecosystems around regional challenges with national significance.⁹⁸ The CO–WY Engine’s ASCEND initiative focuses on climate resilience and natural hazards, including wildfires, leveraging institutions such as CU Boulder Earth Lab, CSU, University of Wyoming, NCAR, and state agencies like Colorado DFPC and Wyoming Office of Homeland Security.⁹⁹¹⁰⁰ The region’s recent fire history (e.g., Marshall, Cameron Peak, East Troublesome) and diverse fuels/terrain provide rich testbeds for model development and validation.^101102103

1.5 Purpose and Scope of This White Paper

This white paper:

• Surveys capabilities and limitations of predominant wildfire models across empirical, semi-physical, coupled, and high-fidelity families.^104105106107
  
• Analyzes data requirements, sources, and gaps in the wildfire modeling ecosystem with an emphasis on CO–WY.^108109110111
  
• Identifies emerging opportunities in SciML, AI weather prediction, and advanced UQ.^{112113114115116}
  
• Articulates VVUQ challenges specific to wildfire applications and pathways to standardized benchmarks.^117118119
  
• Proposes strategic investment priorities across technical R&D, data infrastructure, operations, and standards.
  
• Discusses barriers to operational adoption and makes recommendations for effective research-to-operations (R2O) transition.

1.6 Intended Audience

The intended audience includes:

• Federal program managers (NSF, NOAA, USFS, FEMA, DOE) shaping wildfire R&D portfolios.
  
• Operational fire and emergency managers evaluating advanced prediction tools.
  
• Academic and industry researchers in wildfire science, atmospheric modeling, geospatial data, and AI.
  
• Utilities, insurers, and other commercial stakeholders exposed to wildfire risk.^120121122
  
• Policymakers working on wildfire risk reduction and climate adaptation.

---

2. Fundamentals of Wildfire Behavior and Modeling

2.1 Physics of Wildfire Spread

Wildfire spread is governed by heat transfer (radiation, convection, conduction), combustion chemistry, and the availability of fuel and oxygen, modulated by weather and terrain.^123124125 Classical wildland fire behavior materials and training emphasize the “fire triangle” (heat, fuel, oxygen) and closely related “fire behavior triangle” (fuel, weather, topography) as conceptual anchors for understanding spread and intensity.^126127128

Wind increases oxygen supply, tilts flames toward unburned fuels, and transports heat and embers ahead of the front; slope has a similar flame-tilting effect via buoyancy, often doubling or tripling spread rates relative to flat terrain in canonical examples.^129130131 Fuel moisture must be reduced (water vaporized) before ignition, making spread highly sensitive to drought, humidity, and time since precipitation.¹³²¹³³

2.2 The 80–20 Rule: Dominant Factors in Fire Prediction

Fire science and operational guidance repeatedly identify four dominant predictors of landscape spread: wind, slope, fuel type/structure, and fuel moisture.^134135136137 These are the primary inputs in Rothermel’s surface fire model and related national fire behavior systems, and empirical work confirms their leading influence in many environments.^138139140

Secondary processes—fire–atmosphere coupling, spotting, active/passive crown fire transitions—are essential in certain regimes (e.g., plume-driven or crown-dominated fires) but are more computationally expensive to resolve explicitly and often lack the same breadth of empirical constraints.^141142143144 This informs an 80–20 strategy: if surrogates and improved data pipelines can capture these four primary drivers well at scale, large speed gains are possible before adding selective high-fidelity treatments for secondary effects.^145146147

2.3 Spatial and Temporal Scales

Wildfire processes span from sub-meter, millisecond combustion at the flame scale to multi-kilometer, multi-day landscape and regional behavior.^148149150 High-fidelity CFD models like FIRETEC and WFDS explicitly resolve 1–10 m scales for limited domains and durations; mesoscale coupled systems like WRF-Fire typically operate at tens to hundreds of meters on nested domains over hours; landscape simulators operate at 10–100 m grids over hours to days; and risk models and climate-fire studies consider decades and regional to continental scales.^151152153154

No single model can resolve all relevant scales explicitly in operational timeframes, necessitating multi-scale hierarchies, parameterizations, and surrogates.

2.4 Modeling Approaches: Empirical to Physics-Based

Major families include:

• Empirical/semi-empirical surface and crown fire models such as Rothermel’s equations and the Canadian Fire Behavior Prediction (FBP) system, calibrated from experiments and historical fires and widely deployed (BEHAVE, FARSITE, Prometheus, IFTDSS).[^^3]¹⁵⁵¹⁵⁶
  
• Semi-physical front-propagation models using level sets or minimum-travel-time algorithms (e.g., MTT, ELMFIRE) built around empirical spread-rate kernels.^157158159
  
• Coupled atmosphere–fire models (WRF-Fire/WRF-SFIRE, others) co-simulating plume and fire fronts with two-way coupling.^160161162163
  
• High-fidelity CFD models (FIRETEC, FDS/WFDS) resolving turbulence, combustion, and radiation in detail for limited domains.^164165166

Each family embodies different trade-offs among physics fidelity, speed, data requirements, and implementation complexity.

2.5 Uncertainty Sources and Propagation

Key uncertainty sources include:

• Initial conditions: ignition location/time, fuel moisture distributions, and current perimeter state are often poorly constrained, with satellite active fire detections having 375 m–2 km resolution and minutes-to-hours latency.^167168169
  
• Boundary conditions: weather forecast uncertainty, especially in winds and humidity, is a major contributor to overall fire spread uncertainty; small wind direction or speed errors can have large nonlinear effects on spread.^170171172
  
• Model structure and parameters: simplified physics, missing processes (e.g., spotting in some tools), and poorly constrained parameters (fuel properties, empirical coefficients) introduce systematic uncertainties.^173174175176
  
• Observational uncertainty: fire perimeters derived from satellite or mapping flights have spatial and temporal errors that must be accounted for in validation and DA.^177178179180

These uncertainties interact nonlinearly and generally cannot be combined by simple linear error addition; UQ frameworks and ensembles are required for realistic confidence estimates.^181182183

---

3. Current State of Wildfire Modeling Capabilities

3.1 Operational Models: FARSITE and Prometheus

FARSITE, developed at the USFS Missoula Fire Sciences Laboratory, implements Rothermel surface fire and Van Wagner crown fire initiation/spread equations with a minimum-travel-time (MTT) algorithm on gridded landscapes and is integrated into WFDSS and related systems.^184185186 Official documentation and reviews note its strengths—speed, widespread training base, integration with fuels data—and limitations, including lack of fire–atmosphere coupling, simplified spotting, and deterministic outputs.^187188189

Prometheus, built around the Canadian FBP system, plays a similar role in Canada and has been open-sourced, enabling broader community contribution; however, differences between U.S. and Canadian fuels, and operational ecosystems, have limited cross-border uptake.¹⁹⁰¹⁹¹

3.2 Research–Operational Models: WRF-Fire and ELMFIRE

WRF-Fire (also called WRF-SFIRE in some configurations) couples a level-set fire spread module with WRF, enabling two-way interaction between fire and atmosphere and has been used to simulate several historical fires, demonstrating the importance of fire-generated winds and plume dynamics.^192193194 Peer-reviewed studies show improved realism relative to simple models but also document high computational costs and configuration complexity.^195196197198

Semi-empirical, cloud-native systems such as ELMFIRE demonstrate that level-set and MTT-style models can be combined with automated satellite data assimilation and cloud orchestration to deliver ensemble forecasts within tens of minutes, approaching operational timing constraints while retaining empirical-core physics.¹⁹⁹²⁰⁰ These architectures show the potential of hybrid, distributed systems even before introducing SciML surrogates.

3.3 High-Fidelity Research Models: FIRETEC and WFDS

FIRETEC (LANL) and WFDS (NIST/USFS) are widely recognized as high-fidelity physics-based models for wildland and WUI fires, respectively, resolving 3D turbulent combustion, vegetation interaction, and detailed heat transfer mechanisms.^201202203 NIST’s WFDS builds on FDS, the global standard for building fire CFD, and has been used extensively in WUI structure ignition, exposure, and mitigation studies and experimental design.^204205206207

Their computational expense makes them primarily research tools for process understanding, validation of simpler models, and generation of synthetic datasets for surrogate training.²⁰⁸²⁰⁹

3.4 Ensemble and Probabilistic Prediction Capabilities

Weather forecasting has long used ensembles as operational practice (e.g., NCEP GEFS, ECMWF ENS), but wildfire ensemble systems are still rare.^210211212 Some operational or near-operational platforms (e.g., ELMFIRE, vendor systems) have implemented ensemble schemes based on sampling parameters and weather members, and initial case reports suggest improved communication of uncertainty via probability contours and spread maps.²¹³²¹⁴

The main barriers to broader adoption are computational cost (for physics-heavy models), lack of standardized probabilistic verification, and limited user training in ensemble interpretation.^215216217

3.5 Speed–Accuracy Trade-offs and Operational Limitations

Coupled and high-fidelity models have shown potential for 75–90% perimeter accuracy on selected historical cases when carefully configured and given good input data.^218219220221 Empirical and semi-empirical models are faster and easier to run but can degrade in extreme or highly complex events where processes outside their calibration space dominate (e.g., strong plume–atmosphere coupling, extreme spotting).[^^3]^222223224

In practice, incident teams often choose between “fast-enough but approximate” and “accurate but too slow,” leading to heavy reliance on human expertise to interpret and compensate for model limitations.^225226227 Surrogates and ML-enabled weather ensembles aim to break this trade-off.

---

4. Data Requirements and Ecosystem Challenges

4.1 Fuel Characterization: The LANDFIRE Foundation

LANDFIRE provides wall-to-wall, 30 m fuels, vegetation, and topographic products for the U.S., including surface and canopy fuel models used by most operational fire behavior tools; the program is jointly run by USGS and USFS.²²⁸ LANDFIRE’s documentation and technical notes describe 2–5-year update schedules, the use of disturbance mapping (harvest, fire, insects), and known limitations in rapidly changing WUI areas and small-scale heterogeneity.²²⁹²³⁰

CO–WY regional activities (e.g., Colorado DFPC mapping, state forestry efforts) sometimes augment LANDFIRE with higher-resolution or more recent datasets in priority WUI zones.²³¹²³² However, rapid development and vegetation change can still outpace these updates, creating mismatches between mapped and actual fuels.

4.2 Weather Data: From Sparse Observations to Dense Forecasts

RAWS stations and other mesonets provide point observations of fire-relevant variables but are sparse in many mountainous and rural areas.²³³²³⁴ HRRR and similar high-resolution NWP systems offer 3 km gridded analyses and forecasts, updated hourly, and are widely used as boundary conditions for fire models and situational awareness.²³⁵²³⁶

Ensemble weather products (e.g., GEFS, HRRR-TLE) exist but have limited member counts and resolution for routine fire use, highlighting the importance of ML-based ensembles for wildfire applications.^237238239

4.3 Real-Time Fire Observations: Satellites and Remote Sensing

Satellite active fire products and burned area maps underpin ignition detection, perimeter updates, and validation.²⁴⁰²⁴¹ NASA’s FIRMS system (VIIRS/MODIS) publishes global active fire detections in near-real time via APIs and web services, with typical latencies on the order of an hour.²⁴²²⁴³ GOES-R ABI provides 5–10 minute hemispheric imagery that can be processed for high-frequency fire detection at coarser resolution.²⁴⁴²⁴⁵

Remote sensing review papers and operational guidance describe how multi-sensor combinations (GOES + VIIRS + Landsat/Sentinel + aircraft + cameras) are increasingly used in fire monitoring and operational mapping.^246247248249 These streams create opportunities for rapid DA but also challenge pipelines and modeling systems.

4.4 Infrastructure Data: The WUI Mapping Challenge

The SILVIS WUI maps classify WUI at census-block scales and have been widely used for national/regional planning and research; they are based on housing density and vegetation cover from decadal censuses and land cover products.²⁵⁰²⁵¹ NIST and partners have emphasized that structure-level data, including parcel geometry, construction features, and defensible space conditions, are critical to modeling WUI ignition and loss risk.^252253254

County assessor databases, local GIS, and computer vision applied to high-resolution imagery and LiDAR are emerging sources for structure-level data, but governance, privacy, and standardization challenges remain.^255256257

4.5 Format Heterogeneity and Integration Challenges

Wildfire modeling relies on multiple geospatial formats (GeoTIFF, NetCDF, GRIB2, shapefiles/GeoJSON, HDF5, LAS/LAZ, proprietary formats like LCP) and resolutions from <1 m to O(10 km). Integrating these requires careful handling of projections, alignment, resampling, and temporal metadata.^258259260

Recent work on geospatial indexing (e.g., H3, Discrete Global Grid Systems) and neural implicit representations (GeoINR) shows promising approaches to multi-resolution data fusion and continuous representation of complex fields like geology and terrain, which can inspire wildfire data infrastructures.²⁶¹²⁶²

---

5. Emerging Opportunities: AI/ML and Scientific Machine Learning

5.1 The Scientific Machine Learning Paradigm

SciML methods integrate differential equation solvers, neural networks, and automatic differentiation to learn models that both obey physical laws and fit data.^263264265 The Julia SciML ecosystem (e.g., DifferentialEquations.jl, ModelingToolkit.jl) has been applied to climate parameterizations, combustion, and other stiff, multi-scale problems with reported gains in speed and robustness.²⁶⁶²⁶⁷

The same toolkit is directly applicable to wildfire spread and coupled fire–atmosphere systems, where surrogates and PINNs can be trained against FIRETEC/WRF-Fire datasets or historical cases.

5.2 ML Weather Prediction Revolution and Wildfire Applications

GraphCast, Pangu-Weather, and FourCastNet each show that global medium-range forecasting can be cast as a learned mapping trained on ERA5 and related datasets, with inference on GPUs/TPUs taking seconds to minutes.^268269270 These models often match or exceed ECMWF in skill on standard metrics and event-focused evaluations (e.g., tropical cyclones, extreme precipitation).[^^19]²⁷¹

For wildfire, the ability to generate large ensembles cheaply is transformative: where traditional NWP might support only a handful of high-resolution members, ML models make O(100–1000) member ensembles realistic, which can then be downscaled for fire modeling.²⁷²²⁷³

5.3 Physics-Informed Neural Networks for Fire Spread

PINNs and related physics-informed architectures have been proposed and prototyped for simplified wildfire spread problems, where networks approximate solutions to advection–reaction equations under constraints defined by Rothermel-like relationships and basic conservation laws.²⁷⁴²⁷⁵ While these demonstrations remain early-stage, they illustrate routes to models that blend empirical fire-behavior insights with data and keep physically implausible outputs in check.

5.4 Surrogate Model Strategy for Operational Deployment

A DOE-driven surrogate strategy—sampling WRF-Fire/FIRETEC parameter spaces and training neural operators or other surrogates—has been explored in multiple geoscience and combustion contexts and can be adapted to wildfire.^276277278279 Recent wildfire-focused surrogate work (e.g., emulating process-based fire modules in Earth system models or predicting burned area from drivers) shows promising fidelity with large speedups.²⁸⁰²⁸¹

5.5 Neural Fields for Multi-Resolution Data Fusion

Implicit neural representations like GeoINR learn continuous functions mapping coordinates to environmental variables and can represent geologic layers, terrain, and similar 3D structures efficiently.²⁸² Computer vision work on multimodal implicit 3D reconstructions in natural environments (e.g., forests) demonstrates that such models can fuse LiDAR, RGB, and other sensors into a single learned representation.²⁸³

Adapting these approaches could yield continuous, compressed representations of fuels, terrain, and WUI structure geometry that support resolution-independent sampling for wildfire models and digital twins.²⁸⁴²⁸⁵

---

6. Verification, Validation, and Uncertainty Quantification Challenges

6.1 The VVUQ Crisis in Wildfire Modeling

Recent reviews of wildfire modeling and coupled wildfire–atmosphere systems emphasize that VVUQ practices lag behind those in weather and climate, with limited systematic benchmarking, small validation datasets, and sparse probabilistic verification.^286287288289 Many codes lack automated regression tests, CI/CD, or formal software engineering practices, leading to reproducibility issues.^290291292

6.2 Validation Data Limitations and Observational Uncertainty

Satellites like VIIRS, MODIS, and GOES provide essential data but with known limits in spatial accuracy and revisit times.^293294295 Ground-based mapping and post-fire analysis (e.g., NIST post-fire studies, USGS model evaluation work) highlight uncertainties in fire perimeter and intensity estimates, complicating direct model–observation comparisons.^296297298

Rigorous validation must propagate observational uncertainties into performance metrics rather than treating observations as perfect ground truth.²⁹⁹³⁰⁰

6.3 The Small Sample Size Problem

Many validation efforts consider fewer than 20 fires, often focusing on high-profile, well-observed events (e.g., Camp Fire, Marshall Fire), which limits statistical power and generalizability across regions and conditions.^301302303304 This small-sample problem is widely recognized in both wildfire model assessment and hazard-loss modeling and motivates the creation of community benchmark datasets.^305306307

6.4 Lack of Standardized Metrics and Benchmarks

No universally adopted set of wildfire model verification metrics exists, and studies use different definitions and thresholds (e.g., various overlap scores, timing metrics, rate-of-spread errors). In contrast, WMO and ECMWF have established standardized weather-verification metrics and reference datasets.³⁰⁸³⁰⁹

USGS and NIST reports argue for similar benchmark-case definitions and metric standardization in wildland and WUI fires, including structure ignition and loss outcomes.^310311312

6.5 Ensemble UQ and Probabilistic Verification

Probabilistic verification methods such as reliability diagrams, rank histograms, and CRPS are standard for ensemble weather forecasts but seldom applied systematically to wildfire ensembles.^313314315 As more systems adopt ensembles (e.g., ELMFIRE-style platforms, ML-weather-driven workflows), such metrics will be required to calibrate and build trust in probabilistic outputs.³¹⁶³¹⁷

6.6 Software Engineering and Reproducibility

NIST, USGS, and broader reproducibility literature in computational science all emphasize the importance of open code, documented workflows, automated testing, and robust version control for trustable modeling systems.^318319320321 These practices are increasingly required by journals and funders but are not yet universal in wildfire modeling.

---

7. Machine Learning Weather Prediction and Wildfire Integration

7.1 The Traditional NWP Computational Bottleneck

Conventional global and regional NWP systems require substantial HPC resources and wallclock time to produce medium-range forecasts and ensembles, constraining ensemble size and resolution.^322323324 This has historically limited wildfire workflows to one or a few deterministic or low-member weather scenarios.

7.2 ML Weather Models: GraphCast, Pangu-Weather, FourCastNet

GraphCast, Pangu-Weather, and FourCastNet have each demonstrated that reanalysis-trained ML models can produce skillful global forecasts at 0.25° with orders-of-magnitude lower inference cost than traditional NWP.^325326327 Their architectures (graph networks, 3D transformers, Fourier operators) differ, but all show that learned operators can emulate global atmospheric evolution efficiently.

7.3 Implications for Wildfire Prediction Ensembles

When combined with downscaling, ML weather ensembles can provide large sets of plausible atmospheric trajectories for fire models, making full joint weather–fire ensembles feasible in operational windows and allowing explicit characterization of weather-driven fire uncertainty.^328329330

7.4 Downscaling Challenge: Global to Fire-Relevant Scales

Statistical and hybrid dynamical–ML downscaling approaches, trained on HRRR and high-resolution WRF archives, have been proposed and applied in various contexts and are directly relevant to producing 1–3 km, fire-relevant ensembles from 0.25° global ML forecasts.^331332333

7.5 Uncertainty Representation in ML Weather Ensembles

Research on ensemble generation for ML weather (e.g., stochastic perturbations, diffusion models, stochastic parameterizations) is active, and early results show that ML ensembles can be made probabilistically competitive with traditional NWP ensembles, though more work is needed on calibration, especially for fire-relevant variables.^334335336

---

8. Investment Priorities for Operational Wildfire Prediction

8.1 Technical Research Priorities

8.1.1 Real-Time Physics-Based Model Surrogates

Invest in SciML surrogates trained on ensembles of high-fidelity simulations (e.g., WRF-Fire, FIRETEC) to achieve 100–1000× speedups while preserving accuracy for dominant drivers (wind, slope, fuels, moisture).³³⁷³³⁸ Use design-of-experiments sampling to span operational parameter spaces, validate surrogates against withheld simulations and case studies, and deploy them as the core of real-time ensemble systems.³³⁹

8.1.2 ML Weather Prediction Downscaling

Develop downscaling pipelines that ingest ML global forecasts (GraphCast, Pangu-Weather, etc.) and produce 1–3 km regional fields suitable for fire modeling.³⁴⁰³⁴¹ Combine statistical ML methods with physics-guided constraints and validate extensively against HRRR and dense observational networks in the CO–WY region.³⁴²

8.1.3 Marshall Fire Digital Twin Reconstruction

Construct a high-resolution digital twin of the 2021 Marshall Fire by integrating multi-source data (satellites, aerial surveys, structure damage assessments, utility records) into a coherent spatiotemporal dataset.³⁴³³⁴⁴ Use this as a benchmark for validating WUI spread models and structure-ignition modules and for training surrogates.³⁴⁵³⁴⁶

8.1.4 Ensemble-Based Uncertainty Quantification

Develop a full ensemble-based prediction system that samples weather, fuels, moisture, and model parameters to generate probabilistic fire forecasts.³⁴⁷³⁴⁸ Implement probabilistic verification metrics and calibration workflows, and design user-facing products (probability contours, arrival-time distributions) tuned to decision-maker needs.³⁴⁹³⁵⁰

8.1.5 WUI Structure Ignition Modeling

Couple landscape-scale spread models with structure-level ignition models informed by physics-based simulations (e.g., WFDS) and empirical loss data to estimate ignition probabilities for individual structures or neighborhoods.^351352353 Use ML classifiers or regression models trained on high-fidelity structure-scale simulations and real fire outcomes to support structure-level risk assessments.³⁵⁴³⁵⁵

8.2 Data Infrastructure and Integration

8.2.1 Geospatial Data Harmonization

Deploy an H3 (or similar) hierarchical indexing framework to harmonize multi-resolution fuel, terrain, weather, and infrastructure data into a unified geospatial store with FAIR-compliant metadata and APIs.³⁵⁶³⁵⁷ Use neural fields and cloud-native formats (e.g., Zarr) to support scalable, resolution-independent access.³⁵⁸³⁵⁹

8.2.2 Real-Time Observation Integration

Build streaming pipelines for GOES-R, VIIRS, RAWS, and other real-time data, coupled with data assimilation frameworks that continuously update model states.³⁶⁰³⁶¹ Implement automated quality control and uncertainty-aware assimilation schemes.³⁶²³⁶³

8.2.3 Historical Fire Database Curation

Curate a CO–WY fire database with standardized perimeters, ignition conditions, weather, fuels, and impacts for use in validation and ML training.³⁶⁴³⁶⁵ Include multiple quality tiers and document known observational limitations to support transparent VVUQ.³⁶⁶³⁶⁷

8.3 Operational Deployment and Training

8.3.1 Decision Support Interface Development

Develop web-based and mobile decision-support tools that present probabilistic fire forecasts, WUI risks, and relevant overlays (infrastructure, evacuation zones) in forms suitable for incident commanders and emergency managers.³⁶⁸³⁶⁹ Emphasize simple, interpretable graphics and “what-if” scenario exploration.

8.3.2 Training and Workforce Development

Create multi-level training modules on ensemble interpretation, uncertainty, and model limitations for operational personnel, building hybrid fire–data science literacy.³⁷⁰ Integrate these into existing NWCG and regional training programs.³⁷¹

8.3.3 Operational Pilot Deployments

Conduct phased pilot deployments (shadow, advisory, operational modes) with CO–WY partners to refine systems, build trust, and document operational value.³⁷²³⁷³

8.4 Standards and Community Building

8.4.1 Model Intercomparison and Benchmarking

Establish community benchmark sets, standardized metrics, and an open leaderboard to support objective model intercomparison and track progress over time, analogous to CMIP and WMO verification frameworks.³⁷⁴³⁷⁵

8.4.2 Open-Source Software Development

Prioritize open-source releases of modeling, data-processing, and visualization components with robust documentation and governance to accelerate community innovation and reproducibility.³⁷⁶³⁷⁷

8.5 Summary of Investment Recommendations

Allocate $10–15M over three years across technical research, data infrastructure, operational deployment, and community-building activities, with additional cost-sharing and in-kind contributions from federal, state, and private partners.³⁷⁸³⁷⁹ The total investment would represent well under 1% of annual federal wildfire suppression costs while potentially enabling a significant shift from reactive to anticipatory fire management.³⁸⁰

---

9. The Path Forward: From Research to Operations

9.1 Barriers to Operational Adoption

Key barriers include institutional risk aversion and liability concerns, workforce skill gaps in data-science literacy, data-governance and interagency-sharing obstacles, and funding-model fragility for long-term system maintenance.^381382383 Addressing these requires policy, training, and governance innovations in parallel with technical advances.³⁸⁴

9.2 Recommendations for Successful Transition

Recommended strategies encompass co-development with operational partners, phased deployment, human–AI teaming frameworks, sustained workforce development, and explicit updates to policy and certification frameworks to authorize AI/ML-assisted decision support.^385386387 These strategies mirror successful transitions in weather and flood forecasting domains.³⁸⁸³⁸⁹

---

10. Conclusion

The field of wildfire modeling stands at a pivotal moment where emerging AI/ML and SciML techniques can address longstanding speed–accuracy and uncertainty limitations—if they are integrated into robust, validated, and operationally grounded systems.³⁹⁰³⁹¹ The CO–WY Engine’s ASCEND initiative is well positioned to serve as a national-scale demonstration of integrated, AI-enhanced wildfire prediction, blending technical innovation with data infrastructure, workforce development, and policy frameworks.³⁹²³⁹³

10.1 Key Findings Synthesis

Existing wildfire models do not yet meet combined operational requirements for very fast, accurate, probabilistic prediction under extreme conditions.^394395396 The data ecosystem is fragmented, heterogeneous, and subject to significant observational gaps that limit both modeling and validation.^397398399 AI/ML advances—particularly in weather prediction and SciML—offer potential step changes in capabilities, but VVUQ remains a primary bottleneck for operational trust.^400401402

10.2 Strategic Recommendations

Priority actions include real-time surrogate development for dominant physics, ML weather downscaling integration, open data infrastructure, and the establishment of community standards and benchmarks.^403404405406 The Marshall Fire digital twin and CO–WY-focused investments provide compelling initial testbeds and exemplars.⁴⁰⁷⁴⁰⁸

10.3 NSF CO-WY Engine Role and Impact

By coordinating technical R&D, data infrastructure, training, and policy engagement, the CO–WY Engine can demonstrate a full research-to-operations pipeline in operational wildfire prediction that is replicable by other regions and federal partners.⁴⁰⁹⁴¹⁰ This includes building a pipeline of hybrid-skilled professionals and developing policy frameworks for trustworthy AI in fire management.^411412413

10.4 Call to Action

Federal agencies, the research community, operational organizations, and academic institutions each have distinct but interlocking roles in enabling this transition, from coordinated funding to open-source development to integrated training programs.^414415416417 Given rising wildfire impacts, the expected return on relatively modest investments in predictive capability is high, both in avoided losses and improved resilience.⁴¹⁸⁴¹⁹

---

References (Selected Online Sources)

Note: This list is not exhaustive; it highlights key online-accessible sources used for footnotes.

Footnotes

1. NSF Regional Innovation Engines, program overview. https://beta.nsf.gov/funding/initiatives/regional-innovation-engines{:target=“_blank”}

2. NIST Wildfire Modeling overview. https://www.nist.gov/publications/wildfire-modeling{:target=“_blank”}

3. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

4. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

5. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

6. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

7. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

8. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

9. USGS, “Assessing the potential for evaluation of wildland fire models using remotely sensed and in situ data,” 2025. https://pubs.usgs.gov/publication/sir20255053/full{:target=“_blank”}

10. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

11. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

12. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

13. Holsinger et al., “Weather, fuels, and topography impede wildland fire spread in western U.S. forests,” 2016 (USFS RMRS paper).

14. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

15. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

16. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

17. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

18. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

19. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

20. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

21. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

22. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

23. USGS/USFS LANDFIRE data documentation.

24. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

25. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

26. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

27. NIST, “Wildland-Urban Interface (WUI) Fire Data Collection on Parcel Vulnerabilities.” https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-data-collection-parcel-vulnerabilities{:target=“_blank”}

28. NIST, “Framework for Addressing the National Wildland-Urban Interface Fire Problem” (NIST TN 1748). https://nvlpubs.nist.gov/nistpubs/TechnicalNotes/NIST.TN.1748.pdf{:target=“_blank”}

29. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

30. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

31. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

32. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

33. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

34. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

35. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

36. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

37. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

38. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

39. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

40. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

41. Alexander & Cruz, “On the Use of Fire Behavior Models for Decision Support,” NOAA Tech. Attachment 1006.

42. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

43. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

44. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

45. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

46. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

47. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

48. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

49. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

50. ML-based downscaling literature and HRRR ensemble references.

51. State and local Marshall Fire investigation and damage assessment reports (Boulder County, Colorado).

52. NIST, Marshall Fire research updates (WUI fire loss mitigation series). https://www.nist.gov/fire/history/wildland-urban-interface-fire-loss-mitigation-2020s{:target=“_blank”}

53. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

54. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

55. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

56. NIST, “Wildland-Urban Interface (WUI) Fire Data Collection on Parcel Vulnerabilities.” https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-data-collection-parcel-vulnerabilities{:target=“_blank”}

57. NIST, “Framework for Addressing the National Wildland-Urban Interface Fire Problem” (NIST TN 1748). https://nvlpubs.nist.gov/nistpubs/TechnicalNotes/NIST.TN.1748.pdf{:target=“_blank”}

58. NIST, WUI Hazard Mitigation Methodology. https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-data-collection-parcel-vulnerabilities/hazard{:target=“_blank”}

59. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

60. NIST, “Framework for Addressing the National Wildland-Urban Interface Fire Problem” (NIST TN 1748). https://nvlpubs.nist.gov/nistpubs/TechnicalNotes/NIST.TN.1748.pdf{:target=“_blank”}

61. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

62. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

63. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

64. White House OSTP, “Blueprint for an AI Bill of Rights,” 2022.

65. NSF Engines – CO–WY ASCEND materials.

66. Abatzoglou & Williams, “Impact of anthropogenic climate change on wildfire across western US forests,” PNAS 2016.

67. IPCC AR6 WG2, North America chapter (fire and climate impacts).

68. Headwaters Economics & other wildfire cost assessments.

69. State and local Marshall Fire investigation and damage assessment reports (Boulder County, Colorado).

70. NIST, Marshall Fire research updates (WUI fire loss mitigation series). https://www.nist.gov/fire/history/wildland-urban-interface-fire-loss-mitigation-2020s{:target=“_blank”}

71. Official Marshall Fire incident reports and after-action reviews.

72. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

73. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

74. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

75. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

76. IFTDSS documentation of landscape fire behavior models and use in operations.

77. FARSITE official documentation (USFS Rocky Mountain Research Station).

78. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

79. IFTDSS documentation of landscape fire behavior models and use in operations.

80. WFDSS documentation and user guides.

81. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

82. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

83. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

84. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

85. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

86. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

87. USGS, “Assessing the potential for evaluation of wildland fire models using remotely sensed and in situ data,” 2025. https://pubs.usgs.gov/publication/sir20255053/full{:target=“_blank”}

88. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

89. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

90. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

91. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

92. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

93. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

94. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

95. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

96. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

97. Wildfire surrogate modeling papers (e.g., “Building a machine learning surrogate model for wildfire emissions,” GMD 2022). https://gmd.copernicus.org/articles/15/1899/2022/{:target=“_blank”}

98. NSF Engines – CO–WY ASCEND materials.

99. NSF Engines – CO–WY ASCEND materials.

100. Colorado DFPC and Wyoming OHSEM wildfire program overviews.

101. State and local Marshall Fire investigation and damage assessment reports (Boulder County, Colorado).

102. NIST, Marshall Fire research updates (WUI fire loss mitigation series). https://www.nist.gov/fire/history/wildland-urban-interface-fire-loss-mitigation-2020s{:target=“_blank”}

103. Earth Lab’s Western US fire database and case studies.

104. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

105. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

106. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

107. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

108. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

109. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

110. USGS/USFS LANDFIRE data documentation.

111. NIST, “Framework for Addressing the National Wildland-Urban Interface Fire Problem” (NIST TN 1748). https://nvlpubs.nist.gov/nistpubs/TechnicalNotes/NIST.TN.1748.pdf{:target=“_blank”}

112. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

113. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

114. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

115. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

116. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

117. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

118. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

119. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

120. Moody’s / RMS / AIR – Wildfire catastrophe modeling descriptions. https://www.moodys.com/web/en/us/capabilities/catastrophe-modeling/wildfire.html{:target=“_blank”}

121. NAIC, “Wildfires” topic page. https://content.naic.org/insurance-topics/wildfires{:target=“_blank”}

122. WUI Data Commons and wildfire-insurance market reports (Milliman, WTW, etc.).

123. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

124. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

125. NWCG fire behavior field guides and S-190/S-290 training materials.

126. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

127. “Wildfire & Weather,” Kentucky Energy & Environment Cabinet (educational PDF).

128. IBHS/insurance industry primers on the fire behavior triangle.

129. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

130. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

131. NWCG fire behavior field guides and S-190/S-290 training materials.

132. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

133. “Wildfire & Weather,” Kentucky Energy & Environment Cabinet (educational PDF).

134. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

135. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

136. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

137. Holsinger et al., “Weather, fuels, and topography impede wildland fire spread in western U.S. forests,” 2016 (USFS RMRS paper).

138. NWCG, “Fire Behavior: Fuel, Weather, and Topography” and fire behavior triangle primers. https://www.fireengineering.com/firefighting/fuel-weather-and-topography-essential-wildfire-behavior-factors/{:target=“_blank”}

139. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

140. Holsinger et al., “Weather, fuels, and topography impede wildland fire spread in western U.S. forests,” 2016 (USFS RMRS paper).

141. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

142. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

143. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

144. NIST WUI Fire Group overview. https://www.nist.gov/el/fire-research-division-73300/wildland-urban-interface-fire-73305{:target=“_blank”}

145. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

146. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

147. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

148. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

149. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

150. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

151. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

152. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

153. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

154. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

155. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

156. FARSITE official documentation (USFS Rocky Mountain Research Station).

157. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

158. FARSITE official documentation (USFS Rocky Mountain Research Station).

159. ELMFIRE and ALERTWildfire documentation and conference papers.

160. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

161. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

162. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

163. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

164. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

165. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

166. NIST FDS/WFDS documentation and WUI applications. https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-spread-and-modeling{:target=“_blank”}

167. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

168. GOES-R fire detection and product user guides.

169. VIIRS active fire product documentation (NASA/NOAA).

170. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

171. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

172. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

173. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

174. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

175. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

176. Alexander & Cruz, “On the Use of Fire Behavior Models for Decision Support,” NOAA Tech. Attachment 1006.

177. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

178. GOES-R fire detection and product user guides.

179. MTBS (Monitoring Trends in Burn Severity) documentation.

180. NIST/USFA post-fire analysis guidance.

181. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

182. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

183. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

184. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

185. FARSITE official documentation (USFS Rocky Mountain Research Station).

186. WFDSS documentation and user guides.

187. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

188. IFTDSS documentation of landscape fire behavior models and use in operations.

189. WFDSS documentation and user guides.

190. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

191. Prometheus documentation (Canadian Forest Service).

192. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

193. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

194. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

195. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

196. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

197. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

198. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

199. ELMFIRE and ALERTWildfire documentation and conference papers.

200. ALERTWildfire program and associated modeling efforts.

201. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

202. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

203. NIST FDS/WFDS documentation and WUI applications. https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-spread-and-modeling{:target=“_blank”}

204. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

205. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

206. NIST FDS/WFDS documentation and WUI applications. https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-spread-and-modeling{:target=“_blank”}

207. NIST WUI experiments and parcel data projects.

208. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

209. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

210. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

211. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

212. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

213. ELMFIRE and ALERTWildfire documentation and conference papers.

214. ALERTWildfire program and associated modeling efforts.

215. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

216. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

217. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

218. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

219. WRF-Fire overview (NCAR RAL). https://ral.ucar.edu/model/wrf-fire-wildland-fire-modeling{:target=“_blank”}

220. WRF-Fire technical descriptions and case studies (Mandel et al., 2011; Coen et al., 2013).

221. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

222. IFTDSS documentation of landscape fire behavior models and use in operations.

223. WFDSS documentation and user guides.

224. NIST WUI Fire Group overview. https://www.nist.gov/el/fire-research-division-73300/wildland-urban-interface-fire-73305{:target=“_blank”}

225. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

226. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

227. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

228. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

229. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

230. LANDFIRE remap and disturbance mapping documentation.

231. Colorado DFPC and Wyoming OHSEM wildfire program overviews.

232. Colorado Front Range fuel and WUI mapping projects (DFPC/CSU).

233. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

234. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

235. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

236. HRRR/HRRR-TLE technical summaries.

237. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

238. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

239. ML-based downscaling literature and HRRR ensemble references.

240. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

241. GOES-R fire detection and product user guides.

242. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

243. NASA FIRMS overview. https://firms.modaps.eosdis.nasa.gov/{:target=“_blank”}

244. GOES-R fire detection and product user guides.

245. NOAA GOES-R ABI fire products overview.

246. GOES-R fire detection and product user guides.

247. MTBS (Monitoring Trends in Burn Severity) documentation.

248. NIST/USFA post-fire analysis guidance.

249. Esri wildfire remote sensing technology overviews.

250. NIST, “Wildland-Urban Interface (WUI) Fire Data Collection on Parcel Vulnerabilities.” https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-data-collection-parcel-vulnerabilities{:target=“_blank”}

251. SILVIS Lab WUI maps. https://silvis.forest.wisc.edu/data/wui/{:target=“_blank”}

252. NIST, “Wildland-Urban Interface (WUI) Fire Data Collection on Parcel Vulnerabilities.” https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-data-collection-parcel-vulnerabilities{:target=“_blank”}

253. NIST, “Framework for Addressing the National Wildland-Urban Interface Fire Problem” (NIST TN 1748). https://nvlpubs.nist.gov/nistpubs/TechnicalNotes/NIST.TN.1748.pdf{:target=“_blank”}

254. NIST WUI experiments and parcel data projects.

255. NIST, “Wildland-Urban Interface (WUI) Fire Data Collection on Parcel Vulnerabilities.” https://www.nist.gov/programs-projects/wildland-urban-interface-wui-fire-data-collection-parcel-vulnerabilities{:target=“_blank”}

256. NIST, “Framework for Addressing the National Wildland-Urban Interface Fire Problem” (NIST TN 1748). https://nvlpubs.nist.gov/nistpubs/TechnicalNotes/NIST.TN.1748.pdf{:target=“_blank”}

257. NIST WUI experiments and parcel data projects.

258. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

259. USGS/USFS LANDFIRE data documentation.

260. HRRR/HRRR-TLE technical summaries.

261. H3 geospatial indexing. https://h3geo.org/{:target=“_blank”}

262. GeoINR 1.0: implicit neural network approach to 3D geology, Geosci. Model Dev. 2023. https://gmd.copernicus.org/articles/16/6987/2023/{:target=“_blank”}

263. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

264. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

265. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

266. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

267. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

268. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

269. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

270. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

271. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

272. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

273. ML-based downscaling literature and HRRR ensemble references.

274. Sample PINN wildfire papers (e.g., “Physics-informed neural networks for parameter learning of wildfire spread,” 2024).

275. Other PINN wildfire or combustion applications.

276. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

277. Rackauckas & Nie, “DifferentialEquations.jl – A Performant and Feature-Rich Ecosystem for Solving Differential Equations in Julia,” JORS 2017.

278. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

279. Surrogate modeling in turbulence/combustion literature.

280. “Building a machine learning surrogate model for wildfire emissions,” GMD 2022. https://gmd.copernicus.org/articles/15/1899/2022/{:target=“_blank”}

281. “Deep learning surrogate models of JULES-INFERNO for wildfire,” 2024/2025.

282. GeoINR 1.0: implicit neural network approach to 3D geology, Geosci. Model Dev. 2023. https://gmd.copernicus.org/articles/16/6987/2023/{:target=“_blank”}

283. WildFusion: Multimodal Implicit 3D Reconstructions in the Wild. https://arxiv.org/abs/2409.19904{:target=“_blank”}

284. GeoINR 1.0: implicit neural network approach to 3D geology, Geosci. Model Dev. 2023. https://gmd.copernicus.org/articles/16/6987/2023/{:target=“_blank”}

285. WildFusion: Multimodal Implicit 3D Reconstructions in the Wild. https://arxiv.org/abs/2409.19904{:target=“_blank”}

286. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

287. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

288. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

289. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

290. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

291. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

292. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

293. NASA FIRMS FAQ. https://www.earthdata.nasa.gov/data/tools/firms/faq{:target=“_blank”}

294. GOES-R fire detection and product user guides.

295. VIIRS active fire product documentation (NASA/NOAA).

296. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

297. MTBS (Monitoring Trends in Burn Severity) documentation.

298. NIST/USFA post-fire analysis guidance.

299. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

300. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

301. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

302. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

303. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

304. Case studies on validation sample sizes in wildfire models.

305. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

306. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

307. WUI Data Commons Phase 2 documents. https://www.milliman.com/en/insight/wui-data-commons-phase-2{:target=“_blank”}

308. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

309. WMO forecast verification guidance and ECMWF verification scorecards.

310. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

311. NIST WUI experiments and parcel data projects.

312. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

313. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

314. WMO forecast verification guidance and ECMWF verification scorecards.

315. Standard probabilistic verification (CRPS, reliability diagrams) literature.

316. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

317. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

318. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

319. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

320. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

321. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

322. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

323. HRRR/HRRR-TLE technical summaries.

324. WMO forecast verification guidance and ECMWF verification scorecards.

325. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

326. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

327. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

328. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

329. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

330. ML-based downscaling literature and HRRR ensemble references.

331. ML-based downscaling literature and HRRR ensemble references.

332. HRRR/HRRR-TLE technical summaries.

333. ML-based climate/precipitation downscaling literature.

334. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

335. Pathak et al., “FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators,” PNAS / arXiv. https://arxiv.org/abs/2202.11214{:target=“_blank”}

336. Research on stochastic ensembles for ML weather models.

337. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

338. Earth Lab’s Western US fire database and case studies.

339. “Building a machine learning surrogate model for wildfire emissions,” GMD 2022. https://gmd.copernicus.org/articles/15/1899/2022/{:target=“_blank”}

340. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

341. USGS/USFS LANDFIRE data documentation.

342. NASA FIRMS overview. https://firms.modaps.eosdis.nasa.gov/{:target=“_blank”}

343. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

344. SILVIS Lab WUI maps. https://silvis.forest.wisc.edu/data/wui/{:target=“_blank”}

345. NIST WUI Fire Group overview. https://www.nist.gov/el/fire-research-division-73300/wildland-urban-interface-fire-73305{:target=“_blank”}

346. SILVIS Lab WUI maps. https://silvis.forest.wisc.edu/data/wui/{:target=“_blank”}

347. ALERTWildfire program and associated modeling efforts.

348. Case studies on validation sample sizes in wildfire models.

349. Prometheus documentation (Canadian Forest Service).

350. Case studies on validation sample sizes in wildfire models.

351. NIST WUI Fire Group overview. https://www.nist.gov/el/fire-research-division-73300/wildland-urban-interface-fire-73305{:target=“_blank”}

352. SILVIS Lab WUI maps. https://silvis.forest.wisc.edu/data/wui/{:target=“_blank”}

353. GeoINR 1.0: implicit neural network approach to 3D geology, Geosci. Model Dev. 2023. https://gmd.copernicus.org/articles/16/6987/2023/{:target=“_blank”}

354. NIST WUI Fire Group overview. https://www.nist.gov/el/fire-research-division-73300/wildland-urban-interface-fire-73305{:target=“_blank”}

355. GeoINR 1.0: implicit neural network approach to 3D geology, Geosci. Model Dev. 2023. https://gmd.copernicus.org/articles/16/6987/2023/{:target=“_blank”}

356. Sample PINN wildfire papers (e.g., “Physics-informed neural networks for parameter learning of wildfire spread,” 2024).

357. Other PINN wildfire or combustion applications.

358. Other PINN wildfire or combustion applications.

359. Surrogate modeling in turbulence/combustion literature.

360. “Wildfire & Weather,” Kentucky Energy & Environment Cabinet (educational PDF).

361. NOAA GOES-R ABI fire products overview.

362. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

363. NOAA GOES-R ABI fire products overview.

364. Prometheus documentation (Canadian Forest Service).

365. Sample PINN wildfire papers (e.g., “Physics-informed neural networks for parameter learning of wildfire spread,” 2024).

366. Coen et al., reviews of coupled fire–atmosphere modeling and FIRETEC.

367. Sample PINN wildfire papers (e.g., “Physics-informed neural networks for parameter learning of wildfire spread,” 2024).

368. NIST WUI experiments and parcel data projects.

369. Standard probabilistic verification (CRPS, reliability diagrams) literature.

370. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

371. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

372. NIST WUI experiments and parcel data projects.

373. ML-based climate/precipitation downscaling literature.

374. Prometheus documentation (Canadian Forest Service).

375. NIST/USGS/V&V technical notes on wildfire model evaluation and uncertainty.

376. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

377. WUI Data Commons Phase 2 documents. https://www.milliman.com/en/insight/wui-data-commons-phase-2{:target=“_blank”}

378. DeepMind GraphCast. https://www.nature.com/articles/s41586-023-06700-2{:target=“_blank”} and project page.

379. Pangu-Weather (Huawei). https://www.nature.com/articles/s41586-023-06329-0{:target=“_blank”}

380. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

381. NIST WUI experiments and parcel data projects.

382. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

383. Research on stochastic ensembles for ML weather models.

384. Research on stochastic ensembles for ML weather models.

385. NIST WUI experiments and parcel data projects.

386. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

387. Research on stochastic ensembles for ML weather models.

388. ALERTWildfire program and associated modeling efforts.

389. WMO forecast verification guidance and ECMWF verification scorecards.

390. NIST Wildfire Modeling overview. https://www.nist.gov/publications/wildfire-modeling{:target=“_blank”}

391. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

392. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

393. NSF Engines – CO–WY ASCEND materials.

394. IFTDSS – Comparison of Fire Behavior Modeling Methods. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/20-models/fbmodelcompare.htm{:target=“_blank”}

395. Mandel et al., “Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE,” Geosci. Model Dev., 2011. https://gmd.copernicus.org/articles/4/591/2011/{:target=“_blank”}

396. NIST WUI experiments and parcel data projects.

397. IFTDSS, Fire Behavior Fuel Models summaries. https://iftdss.firenet.gov/firenetHelp/help/pageHelp/content/00-concepts/fbfm/fbfmsummaries.htm{:target=“_blank”}

398. SciML.jl documentation. https://sciml.ai/{:target=“_blank”}

399. Sample PINN wildfire papers (e.g., “Physics-informed neural networks for parameter learning of wildfire spread,” 2024).

400. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

401. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

402. NOAA HRRR documentation. https://rapidrefresh.noaa.gov/hrrr/{:target=“_blank”}

403. NPS, “Wildland Fire Behavior.” https://www.nps.gov/articles/wildland-fire-behavior.htm{:target=“_blank”}

404. LANDFIRE program overview. https://landfire.gov{:target=“_blank”}

405. Prometheus documentation (Canadian Forest Service).

406. Sample PINN wildfire papers (e.g., “Physics-informed neural networks for parameter learning of wildfire spread,” 2024).

407. Various SciML surrogate modeling case studies in climate and combustion (e.g., Brenowitz & Bretherton, 2018; Maulik et al., 2020).

408. SILVIS Lab WUI maps. https://silvis.forest.wisc.edu/data/wui/{:target=“_blank”}

409. “Uncertainty quantification in coupled wildfire–atmosphere models,” 2024, Phil. Trans. A / NIH PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC11660924/{:target=“_blank”}

410. NSF Engines – CO–WY ASCEND materials.

411. NSF Engines – CO–WY ASCEND materials.

412. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

413. Research on stochastic ensembles for ML weather models.

414. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}

415. Prometheus documentation (Canadian Forest Service).

416. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework{:target=“_blank”}

417. Research on stochastic ensembles for ML weather models.

418. Alexander & Cruz, “On the Use of Fire Behavior Models for Decision Support,” NOAA Tech. Attachment 1006.

419. USGS/NIST reports on wildfire model evaluation and post-fire analysis best practices. https://www.usfa.fema.gov/blog/new-nist-report-outlines-best-practices-for-post-fire-analysis/{:target=“_blank”}