Source: SPATIAL INFORMATICS GROUP LLC submitted to NRP
A FRAMEWORK FOR CONDUCTING, EVALUATING, AND COMMUNICATING PROBABILISTIC WILDLAND FIRE PREDICTION
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
Reporting Frequency
Annual
Accession No.
1028589
Grant No.
2022-33530-37271
Cumulative Award Amt.
$137,881.00
Proposal No.
2022-00846
Multistate No.
(N/A)
Project Start Date
Jul 1, 2022
Project End Date
Feb 29, 2024
Grant Year
2022
Program Code
[8.1]- Forests & Related Resources
Recipient Organization
SPATIAL INFORMATICS GROUP LLC
2529 YOLANDA CT
PLEASANTON,CA 945667513
Performing Department
(N/A)
Non Technical Summary
This proposal addresses the SBIR Phase I topic area, "Forests and Related Resources - Topic Area 8.1." Specifically, it aims to "address the health, diversity and productivity of the Nation's forest and grasslands... [and] improve sustainability of forest resources," where "new technologies are needed to enhance the protection of the Nation's forested lands and forest resources and help to ensure the continued existence of healthy and productive forest ecosystems." It addresses the program's long-term goals of achieving healthy and sustainable forest ecosystems that are more resilient to wildfires and the research priority "Developing Technology that Facilitates the Management of Wildfire on Public Lands" with "research that provides systems for ... managing wildfires; systems for reducing fuel loads in forests; tools and equipment for improving the efficacy and safety of firefighters on the ground and in the air."Forest management today is exceedingly complex, a key difficulty being assessing and managing uncertainty to limit risk. It aims to mitigate the extent and impact of wildfires that are said to be becoming more unpredictable and which are often characterized by complex, uncertain behavior in steep, complex, forested terrain. This is weighed against strategic manipulation of forest fuel amount and structure using tools such as prescribed fire, which is accompanied by its own risks. For example, prescribed fires that unfolded badly include Colorado's Lower North Fork Fire, in which a prescribed fire killed a civilian, and California's Lowden Ranch Fire, which grew to over 800 ha and destroyed 23 homes. Retrospectively, inadequate evaluation of fire and weather conditions is often cited in such cases. Uncertainty about the smoke impacts of prescribed fire also reduces its use. Mitigation now requires quantitative planning, but tools do not currently have the dynamic complexity and means to estimate confidence in predicted fire behavior, smoke, and ecological effects, especially in complex environments. To date, planning has used deterministic kinematic models--and thus ignores the potential for fire-induced winds, dynamic feedbacks, and amplification--and also uses oversimplified meteorological circulations; thus, it can underestimate the range of possible outcomes and misinterpret the impact of management activities. The impacts of better estimating uncertainty extend to improving firefighter safety, as entrapment potential peaks when fire behavior rapidly deviates from an assumed trajectory and becomes extreme (Page et al. 2019).This work is appropriate for a Phase I SBIR because it will demonstrate the technical feasibility of the proposed framework and bring the innovation closer to commercialization by completing the fundamental research and evaluation and demonstrating outcomes in several different types of fires. The innovation involves a high degree of risk, in that our hypothesis that by introducing an uncertainty-based range of primary environmental variables to a modeling system, uncertainty in outcomes can be estimated is untested and the spread of simulations could therefore underestimate natural variability; this is a hurdle that the USDA-funded R&D is intended to overcome.
Animal Health Component
20%
Research Effort Categories
Basic
10%
Applied
20%
Developmental
70%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
12206993100100%
Goals / Objectives
Our goal is to develop, apply, and distribute a framework for conducting, evaluating, and communicating probabilistic fire growth forecasts--an approach that can be used with a wide range of fire behavior modeling systems across several economic sectors. This approach will transmit the uncertainty in key fire environment input variables--notably, weather and fuel loads, which are explored here (the approach can be applied to other variables)--in order to inform stakeholders about the uncertainty in predicted outcomes of fire behavior and effects (e.g., smoke production and fire severity). It will develop new methods to evaluate probabilistic fire behavior forecasts--a different challenge than verifying a single deterministic forecast. An important research component is developing methods to graphically and meaningfully communicate probabilistic fire growth information expressing forecast uncertainty to a wide range of stakeholders, several of which have committed letters of support/collaboration.Objective 1: Develop methods to conduct probabilistic prediction of fire behavior and effects using ensembles of fire behavior simulations--here, coupled weather-fire behavior model simulations using the CAWFE model--where the spread of an input variable represents the uncertainty in its state and spread of outcomes represents the consequent uncertainty in fire behavior and effects (analogous to Whitaker and Loughe [1998]). Demonstrate ensemble design and simulation on select wildfire events.For the key input of weather, use existing ensemble numerical weather prediction forecasts to represent uncertainty in the weather and related dead fuel moisture content.Use new fuel databases that provide distributions of fuel loads and other properties to create an ensemble of simulations for select wildland fire events where the spread in fire behavior and effects outcomes reveals our possible error in fire behavior due to errors in measurement or representation of fuel characteristics.Objective 2: Create and apply improved methods to evaluate and communicate the accuracy of probabilistic fire behavior forecasts. Explore with stakeholders.Objective 3: Improve the dissemination of probabilistic fire growth forecasts by creating new graphical representations and communication methods to express the most likely outcomes, the spread of solutions, occurrence of outliers, and other possible products. Refine with stakeholders
Project Methods
In Phase I, we propose activities to (i) develop a framework to conduct probabilistic fire behavior predictions, (ii) demonstrate it using two primary sources of uncertainty (weather forecast and fuel loads), and (iii) develop and make available better methods to evaluate, display, and communicate how these sources of uncertainty culminate in a range of possible prescribed and wildfire behavior and fire effects. This work will be done with the aim of developing methods and business services products with stakeholders who produce or use fire model forecasts, from the USDA Forest Service, CAL FIRE, and Pyregence.org.Task 1: Select at least four recent wildfire events that represent a range of behavior types (e.g., both wind-driven and plume-driven), have community interest, and are among archived datasets or occur before the project start.In ensemble weather prediction, ensemble spread can depend on the flow and characteristics of topography. Ensemble spread tends to be large in weak forcing scenarios and smaller when forcing is strong. Thus, we will examine the outcomes of simulated fires in different types of environments to see whether certain types/drivers of fires have different spread characteristics (e.g., wind-driven fires vs. plume-driven fires that create winds and propagate themselves).Task 2: Construct ensembles. In this research study, we will use existing probabilistic weather forecast ensembles and probabilistic fuel information to initialize coupled weather-fire behavior simulations.2a: Conduct ensemble simulations of each selected fire event using NOAA ensemble weather forecasts as input, to produce a range of possible fire behavior outcomes.Fire spread is acutely sensitive to wind and, in turn, fire shapes the wind and further fire growth through fire-atmosphere interactions. Thus, weather uncertainty is a primary uncertainty source. Weather uncertainty will be considered by providing CAWFE with varied initial and boundary conditions from one or both ready-made NCEP forecast ensembles including the Short-Range Ensemble Forecast (SREF) (21 members, 16 km grid spacing, hourly output to 39 h and 3-hourly output to 87 h) and the Global Ensemble Forecast System (GEFS) (21 members, 0.5 degree grid spacing, output to 16 days), each of which is initialized 4x per day. We will consider initializing CAWFE meteorology with storm-scale ensembles, such as the 3 km HRRR-E. (Investigators' experience suggests this may or may not be helpful.) As CAWFE interactive grid refinement nests simulations down to convective scales (hundreds of meters), we anticipate the ensemble will produce a range of motions that in turn will result in a range of fire behaviors and effects, the spread of which will grow throughout the length of the forecast. A CAWFE simulation could conceivably continue until the end of the large-scale forecast (to 87 h or 16 days, using SREF or GEFS output, respectively) but the rapid accumulation of model error, particularly when modeling at these fine scales, limits each forecast simulation to a few days. We will draw from our archived SREF forecasts for several past events and will also use significant events of interest to partners that have yet to occur.2b: Conduct ensemble simulations of each selected fire event using NAWFD fuel load probability distributions to construct a range of possible fuel loads as input in order to produce a range of possible fire behavior outcomes.We will investigate the impact of uncertainty in fuels (which arises from measurement or methodological errors in estimating properties--notably fuel load) on modeled fire outcomes. Unlike previous work, in this study the range of inputs will not span all possible natural variability but will be based on NAWFD which assembles data from many sources, its uncertainty estimates, to estimate their subsequent impacts on modeled fire behavior.Task 3: Conduct ensemble simulations of the cases identified in Task 1.To conduct a simulation, in addition to the NCEP large-scale gridded weather forecast and fuel load data for various components from NAWFD, additional data is gathered including terrain data at 30, 3, and 1 arc second, and information on fire ignition time and location or a spatial fire perimeter if already in progress. Fuel moistures will be drawn from USFS databases and/or Remote Automated Weather Stations. (Investigation of uncertainty due to fuel moisture, although among the important factors, is left to Phase II.) The large-scale gridded forecast is projected onto the CAWFE model outer domain grid points (~10 km spacing) to initialize the model, and later periods in the NCEP forecast provide evolving lateral boundary conditions. The simulation telescopes in with finer resolution domains (3.3 km horizontal grid spacing, 1.1 km, and 370 m, with accompanying vertical grid refinement). At the ignition or observation time, the fire ignition or perimeter is inserted into the running weather model and the simulation, with interacting fire and weather, is carried out throughout the forecast period at time steps of approximately a second in the innermost domain, with model and fire state information produced at minute intervals for our analysis.We will demonstrate our developments using the CAWFE model because coupled models have the ability to simulate dynamic interactions and, thus, unanticipated amplification of fire behavior, including the parameter space where outlier events occur; however, the results can be applied to--and would improve--any fire behavior modeling system. The interpretation is that the spread of an input variable represents the uncertainty in its state and that the spread of outcomes represents the consequent uncertainty in fire behavior and effects.Due to limited resources, we will begin with separate ensembles that vary weather and fuel inputs (the weather-varying ensemble using the fuel mean data and the fuel-varying ensemble using the ensemble unperturbed central member). Then conditional probability methods would be a simple approach to combine probability scores from both weather-varying and fuel-varying ensembles.Task 4: Create new graphical representations and communication methods to express the most likely outcomes, the spread of solutions, occurrence of outliers, and other possible products, and explore options with stakeholders.Spatial fire growth probabilities might be expressed a few different ways. Outcomes will be expressed in terms of point probability, for example, whether a particular point will be overrun by fire during the forecast, or one might express a probability regarding how much area is burned.Task 5: Develop and apply improved methods to the ensembles of simulations to evaluate and communicate the accuracy of probabilistic fire behavior forecasts.The probability of whether a fire reaching a point can be determined explicitly from ensemble output, and timing and spatial errors will be objectively determined. Forecasts of fire extent will be compared against available observations of fire extent as measured from satellites, airborne infrared instruments, or incident information. Spatial analysis of the ensemble of CAWFE simulations would first estimate the probability a location is overrun by a fire during the length of the forecast vs. the frequency it occurs in simulations.We will explore methods to estimate ensemble forecast accuracy and seek feedback with stakeholders. It is anticipated that dynamic ensemble forecasts should be able to best FSPro predictions based on climatology, because past weather is not a good predictor of the future.

Progress 07/01/22 to 02/28/24

Outputs
Target Audience:We have identified three primary sectors of audiences (and potential customers) within the US market over the course of the project, including: Electric Utilities Sector - risk managers at utilities require near-term wildfire forecasting capabilities to inform de-energization events (i.e., power shutoffs), and mitigate impacts associated with active fires moving toward critical infrastructure. Emergency Management Sector - including local, state, and federal government emergency management and wildfire agencies. These customers need tools to inform strategies for wildfire suppression and coordinate community evacuation strategies. Insurance Sector -Tools with the PyreCast platform allow for portfolio managers within this sector to understand their potential exposure to wildfire hazards and inform actions to mitigate exposure (e.g., pre-treating insured properties), or otherwise plan actions to best manage their insurance policy portfolios (e.g., apply temporary hold on new policy approvals). Of the sectors we've identified, we believe that the utility and insurance sectors will have the largest budgets to apply the technology develped with SBIR grant funding and ability to support the PyreCast platform, however, the emergency management sector has the potential to be the broadest, most diverse market base. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided? Nothing Reported How have the results been disseminated to communities of interest?We have not disseminated the our results yet beyond sharing with our agreement manager at NIFA. However, we plan to prepare and submita professional journal publication that describes the methods and results produced through this project. Additionally, we will present our results to a group of stakeholders that providedsupport for the project proposal prior to April 15, 2024. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? Objective 1 (Develop Probabilisitc Methods) -Our work produced scientific insights and advances, underlining the advantages of this additional forecast dimension. We conducted CAWFE coupled weather-fire model ensemble simulations of the 2021 Mosquito Fire, the 2017 Tubbs Fire, and 2021 Caldor Fire, variously initialized with physics-varying, probabilistic weather-varying, and probabilistic fuel-varying inputs or settings and used the outputs to develop an analysis and presentation methodology. These model outputs - time of arrival, rate of spread, and a measure of fire intensity - are used to derive key fire management and land management information (Burn Probability, Spread Rate, and a proxy for Fire Severity) and are also the current predictions released on the PyreCast website. We found that ensembles, overall, covered the extent of actual fire growth, but were narrower than expected. This is likely due to the fact that, although nonlinear dynamics may produce rapidly diverging behavior, the forecast period in these simulations ranged from 12 to 34 hours, perhaps too short to see much divergence among members. We found the physics-varying ensemble of the Mosquito Fire created a wider range of outcomes than fuel-varying or weather-varying ensembles of that or other events, where one outlier produced anomalous growth similar to what occurred. We also found that the weather-varying ensembles - initialized with 13 members each from two related weather models - could likely be reduced for computational savings to one 13-member ensemble using the NMMB model. In case-specific results, we saw that variability among members of the Tubbs Fire indicated areas where variability in the wind field altered the complex wind-terrain-slope airflow regime, indicating where fine-scale speeds produce wind "hotspots", potentially electric grid disruptions and wildfire ignitions. Statistically compiling the Mosquito ensembles, we saw that the mean and variability in the rate of spread and heat flux indicated chutes of intense burning and rapid fire spread - a result of operational interest. Fuel-varying ensembles of the Caldor Fire suggest that widespread landscape-scale fuel reduction may have had limited impacts on fire behavior during its run through the Grizzly Flats community, though we cannot rule out higher local effects beneath our simulation scale. Objective 2 (Communication Methods) -We found a significant enhancement of the information interpretability first by simply animating outputs, compared to the static burn probability presentation common in this field, because uncertainty varies in time as well as space. Statistical reduction of the ensemble to companion mean and a variability plots brought out agreed on features (e.g. areas where the fire will spread rapidly or burn intensely) as well as indicating locations of uncertainty (e.g. how widely a fire may swerve). Secondly, offering animations of individual members satisfies the need to inspect particular ensemble members or outliers. Additional statistical slicing (such as distilling the top few percent or quartiles) was conducted and may have specific uses. Objective 3 (Create Graphical Representation of Fire Probabilities) -We discovered surprisingly important but hitherto underemphasized aspects of presenting probabilistic information through including a graphic designer. Fire prediction information has historically been presented in hot-cold color palettes that can exclude users with a range of vision impairments. We developed intuitive color palettes for each type of information and created presentation modes that show clear demarcation between intervals, integrate more than one method for showing gradations (color and transparency, for example), and satisfy the highest levels of web accessibility - a necessity for products serving federal and state users.

Publications


    Progress 07/01/22 to 06/30/23

    Outputs
    Target Audience:During the past period, we have been conducting research and beginning product development. Team members working with LARTA have been developing a business plan that includes building a potential client list. Changes/Problems:In December 2022, we requested and received from USDA SBIR programa 1-year no cost extension due the fact the we needed more time to collect SREF data from significant wildfire event. The extension will allow more time to collect that data. What opportunities for training and professional development has the project provided? Nothing Reported How have the results been disseminated to communities of interest?Not yet - we plan to work with stakeholder in upcoming months to go through frameward, methods and interim results. What do you plan to do during the next reporting period to accomplish the goals?Our next steps include collecting weather and fuels data from at least one more fire event, performing ensemble calculations and demonstrating potential visuallization products for probabilistic error estimates and uncertainty with our initial stakeholders. Once these items are complete, we plan on preparing a draft and final report. We will continue to meet with our project contact (Diomides Zamora) to keep them in the loop on progress. We plan to submit an application for SBIR phase 2 if invited.

    Impacts
    What was accomplished under these goals? The project's goal is to develop, apply, and distribute a framework for conducting, evaluating, and communicating probabilistic fire growth forecasts--an approach that can be used with a wide range of fire behavior modeling systems across several economic sectors. Key tasks are to first estimate the impact from key fire environment input variables--notably, weather and fuel loads -- on fire behavior and effects (e.g., smoke production and fire severity) through ensemble coupled weather fire model simulations of disparate events (Figure 1). Then, develop methods to estimate error and verbally & visually communicate the error & uncertainty to stakeholders (and ultimately, clients). Figure 1. Conceptual diagram of uncertainty propagation in national ensemble weather forecast through CAWFE coupled weather-fire behavior simulations. During this period, we identified key fire behavior event types and examples of each. These are: Wind-driven events (a regional event example of this is Diablo wind events, such as the 2017 Napa fire conflagration). (Our case: the 2017 Tubbs Fire, Santa Rosa, CA.) Plume-driven events (characterized by weak ambient winds, where internally-generated strong fire-induced winds drive fire growth). (Our case: the 2022 Mosquito Fire, the largest CAL FIRE fire of 2022, in central CA.) Fires driven by convective downdrafts. (Our case: the 2022 McKinney Fire, the 2nd largest CAL FIRE fire of 2022, in northern CA.) Another, to be selected from recent high-impact events in 2023 fire season. These present different types of hazards for wildfire management and have different types of uncertainty in regional weather (Figure 2) that cascades into fine-scale weather and fire behavior prediction. Figure 2. Variability in key atmospheric state variables (notably: the third column shows impacts in near surface wind) across input SREF weather ensemble members under low/no precipitation conditions (upper row) and convective precipitation (lower row) conditions. We collected ensemble weather forecasts that will drive ensembles of coupled weather-fire model simulations. We have completed & visualized the baseline (control) simulation for of the Tubbs, McKinney, and Mosquito Fires. Figure 3. Validation of individual simulations can be made against satellite active fire detection products such as the Visible and Infrared Imaging Radiometer Suite (VIIRS). Example is from the CAWFE simulations of the Mosquito Fire. We have conducted team meetings to discuss project timelines and methods to calculate and graphically represent error estimates, compared to various fire mapping data (Figure 3), uncertainty in fire rate of spread, time of arrival, and heat flux/flame length and communication of key stakeholder concerns of the probability of locations being overrun, the distribution (among ensemble members) of time of arrival at each location, and probabilistic estimates of impacts (the intersection of fire & assets) to structures and tailored products for assets of potential business application areas such as utility infrastructure (e.g.) transmission lines.

    Publications