#1: A Machine-Learning Framework for Rapid Adaptive Fire-Propagation Simulation in Complex Environments
Authors: Tarek Zohdi (zohdi@berkeley.edu)
Affiliation: Department of Mechanical Engineering, University of California, Berkeley
Abstract: The objective of this presentation is to illustrate how to integrate Machine-Learning Algorithms (MLA’s) with multistage/multicomponent fire spread models, as well as with rapid data collection by use of multiple Unmanned Aerial Vehicles. In order to tangibly illustrate this process, this monograph develops a framework combining:
1. A meshless discrete element “submodel” that tracks the trajectory of airborne incandescent particles and embers, subject to prevailing wind velocities and updrafts
2. A meshless topographical “submodel” of the ambient combustible material (and possible man-made firebreaks) where embers that make contact are allowed to start secondary fires, if conditions are appropriate, combined with a burn rate “submodel” for generating new embers, new updrafts due to hot air, etc. and
3. A Machine-Learning Algorithm to ascertain the multi-submodel system parameters that force the overall model to match observations.
The submodes compute both ground and airborne hot-ember driven fire propagation, as well as subsequently distribution of soot, which is important for air-quality assessment. The overall framework is designed for use in digital twin technology, which refers to a digital replica of a physical system, where the philosophy is that updates to digital twins are made continuously in near real-time, which necessitates a rapid simulation paradigm that can easily interface with telecommunications, cameras and sensors. Ideally, such a framework should be able to integrate historical data from past model usage, that factors in its real-time digital model and optimizes its operation and maintenance of physical assets, systems and processes. The model is designed to run rapidly on laptops and hand held devices, with the guiding principle being to make it useful for first-responders in real-time.
#2: Next Generation Disaster Intelligence using the Continuum of Computing and Data Technologies
Authors: Ilkay Altintas and Jessica Block
Affiliation: UC San Diego; San Diego Supercomputer Center
Abstract: Modeling of the extent and dynamics of evolving plethora of environmental hazards, and their socio-economic and human impacts, is a data-driven discipline with a vibrant scientific community of computational modeling, remote sensing, data science, technology and social science experts, driven by the urgent societal need to mitigate the rising frequency and severity of hazards. However, there are still challenges and opportunities in integration of the scientific discoveries and data-driven methods for hazards with the advances in technology, data science and computing in a way that provides and enables different modalities of sensing and computing. The WIFIRE Lab at UC San Diego took the first steps to tackle this problem with a goal to create an integrated system and services for wildfire monitoring, simulation, and response. Today, WIFIRE provides an end-to-end management infrastructure from the data sensing and collection to analysis and modeling efforts using a continuum of computing methods that integrate edge, cloud and high-performance computing. Though this cyberinfrastructure, the WIFIRE project provides data driven knowledge for a wide range of public and private sector users enabling scientific, municipal and educational use. This talk will review some of our recent work on building this dynamic data driven cyberinfrastructure. We will discuss WIFIRE’s impactful application solution architecture that integrates a variety of existing technologies, multiple modalities of data, collaborative expertise and automated workflows with the ability to use multiple fire models and data assimilation techniques. The lessons learned on use of edge and cloud computing on top of high-speed networks, open data integrity, reproducibility through containerization, and the role of automated workflows and provenance will also be summarized.
#3: Experiment-Based Wildfire Spread Model
Authors: Mark A. Finney, Jason M. Forthofer, Torben P. Grumstrup, Sara S. McAllister
Affiliation: USDA Forest Service, Missoula Fire Sciences Laboratory, Missoula MT 59808 USA
Abstract: Physical models of wildfire spread are needed to improve training, risk analysis, fuel treatment design, and fire predictions. Operational requirements for practical models include rapid calculation (seconds), simplicity of use, and robustness to environmental uncertainty. With the goal of developing a simple physical model, laboratory and field experiments were designed to understand physical processes of combustion, heat transfer, and ignition. From these findings, a dynamical model was constructed to simulate fire spread through a one-dimensional array of particles. The model demonstrates principal features of flame spread behavior including acceleration, response to time-varying wind, dependency on ignition line width and depth, and spread thresholds caused by fuel heterogeneity, dead fuel moisture, wind, and slope.
#4: Data-Driven Fire Modeling for Wildland and WUI Fires
Authors: Michael Gollner (mgollner@umd.edu) and Arnaud Trouve (atrouve@umd.edu)
Affiliation: University of Maryland
Abstract: Providing accurate predictions of the spread of wildland fires has long been a goal of the fire research community. Whether used as a planning or operational tool to predict the growth of current or potential uncontrolled wildfires, the accuracy of wildland fire spread models and their ability to provide useful information in a timely manner are of paramount importance. Despite the development of a plethora of fire models, their use has been relatively limited operationally. Some of this stems from the fact that most fire models today are simplified versions of reality that are not physically based. Available data to initialize and parametrize these models, such as fuels, topography, weather,etc., are also subject to large uncertainties and limited resolution. A new approach to this problem, “data-driven modeling” is of growing interest. An overview of work between CERFACS in France, the University of Maryland, and the NSF-funded WIFIRE effort at UCSD will be presented with advancements in data-driven fire modeling presented. Further, modeling the transition between fire spread in wildland fuels into WUI communities is of great interest to the fire research community; however, there are many challenges before this goal can be achieved. A review of the mechanisms governing both wildland and WUI fire spread will be presented, including opportunities to move these physics into the modeling space, as is being undertaken in the new NSF-funded WUI-MAPR project.
#5: Open source automated real-time ensemble fire forecasting for the Continental United States
Author: Chris Lautenberger, PhD, PE (lautenberger@reaxengineering.com)
Affiliation: Reax Engineering Inc.
Abstract: In this presentation, an open source modeling system that has been developed to automatically forecast the spread of all named fires in the Continental US will be demonstrated. Wind/weather inputs are obtained from high-resolution operational weather models. Fire positions are initialized from infrared perimeters (GeoMAC) and satellite detections (MODIS, VIIRS). New simulations are automatically initiated as soon as updated fire position data become available. Uncertainty in model inputs and spotting stochasticity are addressed using ensemble forecasts wherein thousands of variations of the same basic simulation are executed simultaneously using a fire model known as ELMFIRETM (Eulerian Level set Model of FIRE spread). Model outputs are disseminated via a publicly accessible webmap (https://fireforecast.com).
#6: Observations from the Fire Front: Understanding the role of Fire-Atmosphere Interactions on Wildfire Behavio
Author: Craig Clements (craig.clements@sjsu.edu)
Affiliation: Department of Meteorology and Climate Science; San José State University
Abstract: Extreme fire behavior has been observed frequently during recent wildfires in the Western US, yet there is still limited understanding of the role of plume dynamics on fire spread. The Rapid Deployments to Wildfires Experiment (RaDFIRE) was the first coordinated meteorological field campaign dedicated to observing fire-atmosphere interactions during large active wildfires to better understand extreme fire behavior. Using a rapidly deployable scanning Doppler lidar, airborne Doppler radar, and a suite of other instruments, the field campaign sampled 25 wildfires from 2014-2017 in California and Idaho. Access to wildfires was accomplished via team members training as wildland firefighters and through integration with wildland fire management agencies. Observations during RaDFIRE include the initiation of vigorous vertical-axis vorticity in a wildfire convective plume, convective plume entrainment processes, newly discovered smoke-induced density currents, and aircraft in-situ observations of a developing pyrocumulus with extreme updraft cores of 58 m s-1. While observations of active wildfires have shed light on processes associated with fire-atmosphere interactions, data collected from a small-scale and comprehensive field experiment (FireFlux2) provide context on the local processes responsible for fire spread that are difficult to observe when sampling large wildfires. These processes include the development of a region of surface low pressure that increases the fire-induced wind into the rear of the fire front causing acceleration of fire spread. Collectively, the RaDFIRE field campaign and FireFlux2 observations highlight the range of phenomena associated with fire-atmosphere interactions, especially plume dynamics, and will provide a valuable data set for the fire behavior modeling communities.
#7: Process-based fire/atmosphere modeling: opportunities and challenges
Authors: Rod Linn, Alexandra Jonko, Eunmo Koo, Scott Goodrick, Carolyn Sieg, Mike Brown, Sara Brambilla
Affiliations: Los Alamos National Laboratory, USDA Forest Service
Abstract: Experiments and observations have demonstrated that the two-way feedbacks between fires and atmosphere play critical roles in determining how fires spread or if they spread. Advancements in computing and numerical modeling have generated new opportunities for the use of models that couple process-based wildfire models to atmospheric hydrodynamics models on high performance computing (HPC) platforms. These process-based coupled fire/atmosphere models, which simulate critical processes such as heat transfer, buoyancy-induced flows and vegetation aerodynamic drag, are not practical for operational faster-than-real-time fire prediction due to their computational and data requirements. However, they do serve critical role as they help increase our understanding of wildfire phenomenology, complement experiments, add perspective to observations, and generate new hypothesis that can be tested experimentally. These HPC-based models can also provide critical insights for the development of faster running coupled fire/atmosphere model that can be used for training, ensemble calculations and eventually operations. One requirement that has been identified for any such future model that is intended for broad wildland fire applications (wildfire and prescribed fire) is that it represents the coupling between the fire and the atmosphere.
#8: High resolution integrated models for operational forest fire behaviour and fire weather prediction
Authors: Jean-Baptiste Filippi
Affiliation: SPE Lab, CNRS- University of Corsica
Abstract: Numerical approaches of wildfire prevision, like most natural disasters, have strong constraints on the availability of information and run time at the time of alert. Data must be ready and compatible to model operators, outputs must be quickly available.
The numerical prediction research system, called FireCaster, has been designed with this constraint, with homogeneous methods, resolutions, data sources and information delivery strategies. Most methods are making use of High Performance Computing such as a coupled high-resolution fire/atmosphere model, data assimilation of fire front positions, and ensemble simulations for uncertainty evaluation.
High-resolution (sub km)/high-frequency (<10 minutes) weather forecasting is in particular among the major enhancements allowed by the availability of this online research platform, as it does reach relevant scales for day-to -day forest fire behavior wind forcing, but also fuel moisture and data assimilation.
Estimated use is to provide fire-specific zoomed view of a high risk area, then, in case of fire, specific input parameters for different kinds of interactive (controlled by an operator) fire behavior simulations (fire propagation, ensemble-based fire probability, fire weather forecast, deterministic H+6 to H+12 front positions, smoke and micro-meteorology. It is tested in the Corsica region since summer 2017.
Estimated use is to provide fire-specific zoomed view of a high risk area, then, in case of fire, specific input parameters for different kinds of interactive (controlled by an operator) fire behavior simulations (fire propagation, ensemble-based fire probability, fire weather forecast, deterministic H+6 to H+12 front positions, smoke and micro-meteorology. It is tested in the Corsica region since summer 2017.
#9: Modeling effects of fire and 21st century climate on vegetation composition, nutrient and carbon cycling, and exchanges with the atmosphere
Authors: William J. Riley (wjriley@lbl.gov), Zelalem A. Mekonnen
Affiliation: Climate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California, USA
Abstract: Ecosystem carbon, nutrients, vegetation composition, and surface energy and greenhouse gas emissions following fire depend on climate, vegetation competitive interactions, and soil thermal and hydrological dynamics. We describe results from recent analyses of boreal Alaska and California fires using the ecosys model, which mechanistically represent these processes and has been rigorously tested against eddy covariance observations in many high-latitude, temperate, and tropical sites across multiple years, and against large-scale remote sensing vegetation observations. Consistent with changes over the Holocene, 21st century climate and fire will alter Alaskan boreal forest composition. Competition for nutrients after fire in early succession and for light later in succession explain modeled forest compositional changes. Over the 21st century, rapid nitrogen mineralization from climate- and fire-induced soil warming enabled deciduous broadleaf trees to sustain rapid nitrogen uptake and CO2 fixation. Our study suggests the relative dominance of boreal deciduous broadleaf trees nearly doubles by 2100, strongly affecting the carbon cycle, surface energy fluxes, and ecosystem function, and thereby feedbacks with climate. We will also present results from ongoing ecosys simulations in the Sierra foothills where we are testing the role of fire frequency and intensity on decadal-scale changes in vegetation composition and carbon storage.
#10: Benefit-cost ratio for Building Structural Mitigation of WUI Fire Risk
Author: Charles Scawthorn
Affiliation: Visiting Researcher, Pacific Earthquake Engineering Research Center (PEER), UC Berkeley Principal, SPA Risk LLC
Abstract: Benefit-cost ratios (BCRs) for the contiguous US were calculated for a typical new and retrofitted single family dwelling complying with the 2015 International Wildland Urban Interface. BCRs for new construction are based on the reduced annualized WUI fire loss divided by the added cost of complying with the IWUIC, while BCRs for retrofitted construction consider the cost of retrofit rather than new construction. Benefits (reduction in losses) were quantified by convolving exposure, hazard and vulnerability for 47,870 census blocks in four US counties representative of low, intermediate and high WUI risk. Losses included not only property but also mortality, morbidity and time element. Results of the four counties were extended to the rest of the nation by regression of BCR versus Burn Probability greater than 0. Of 3188 counties in the contiguous US, 761 counties (24%) have at least a portion with BCR > 1. Similarly, there are 33 of the 48 states (69%) with at least a portion of one county with BCR >1. The investigation was for the Mitigation Saves 2 (MSv2) project conducted by the National Institute of Building Sciences (NIBS). The full report, for WUI as well as other hazards, is available at https://www.nibs.org/page/mitigationsaves
#11: Wildland Fire Spotting by Sparks and Firebrands
Authors: Carlos Fernandez-Pello (ferpello@me.berkeley.edu)
Affiliations: Department of Mechanical Engineering, University of California, Berkeley & Reax Engineering
Abstract: The spot fire ignition of a wildland fire by hot (solid, molten or burning) metal fragments/sparks and firebrands (flaming or glowing embers) is an important fire ignition pathway by which wildfires, WUI fires, and fires in industrial settings are started and may propagate. Power lines, hot work (welding, cutting, grinding), and equipment cause approximately 28,000 wildland fires annually in the United States. Once the wildfire or structural fire has been ignited and grows, it can spread rapidly through ember spotting, where pieces of burning material (e.g. branches, bark, building materials, etc.) are lofted by the plume of the fire and then transported forward by the wind landing where they can start spot fires downwind. The spot fire problem can be separated in several individual processes: the generation of the particles (metal or firebrand) and their thermochemical state; their flight by plume lofting and wind drag and the particle thermo-chemical change during the flight; the onset of ignition (smoldering or flaming) of the fuel after the particle lands on the fuel; and finally, the sustained ignition and burning of the combustible material. By characterizing these distinct individual processes, it is possible to attain the required information to develop predictive, physics-base wildfire spotting models. Such spotting models could be used together with statistical data of weather patterns and vegetation distribution in the development of wildfire hazard maps that could guide identifying high-risk power line runs so that utilities could prioritize fuels treatments. Also, the models together with topographical maps and wind models could be added to existing landscape scale wildfire spread models to improve their predictive capabilities.
#12: Resilient road network during wildfire event by integrating traffic network analysis and communication network analysis at a regional scale
Authors: Kenichi Soga, Louise Comfort, Bingyu Zhao, Millard McElwee
Affiliation: UC Berkeley, Department of Civil and Environmental Engineering
Abstract: The talk will be based on a recent field study of the 2018 Camp Fire in the Butte County, the most recent as well as the deadliest wildfire in California history. To date, investigators have made six field trips to the cities of Sacramento, Oroville, Chico, and the town of Paradise, epicenter of the Camp Fire evacuation, to collect information regarding the evacuation timeline, scale, plans and outcomes. Many observers, including news media, blamed the limited capacity of the highways linking Paradise (evacuated area) with Chico or Oroville (safer areas) as the cause of the traffic gridlocks during the evacuation. However, field interviews revealed that the destination areas were also not prepared for the sudden inflow of evacuation traffic and were unable to respond with optimal traffic measures (guidance, signaling, demand control) to smooth the entrance of the evacuees. The talk will introduce our new project that assesses the interdependencies of traffic and communication networks from an interdisciplinary point of view and determines the performance requirements for them in ensuring viable evacuation strategies under the urgent, dynamic conditions of wildfire.
#13: Organizational Networks in Mobilizing Response to Wildfire
Authors: Louise Comfort, Kenichi Soga, Chiara Ecosse, Millard McElwee, Sae Mi Chang
Affiliations: University of Pittsburgh, and University of California, Berkeley
Abstract: Mobilizing response to the rapidly changing dynamics of wildfire is a critical task in diverse communities with organizations representing people at different levels of vulnerability and capacity. Essential to understanding the scope, extent, interactions, and failures in the evacuation process is modeling the information flow among organizations with different levels of responsibility and capacity for sending and receiving information about a changing, dynamic risk situation. This paper will build on data collected by an interdisciplinary research team during a Quick Response study of the 2018 Camp Fire in Butte County, California. We will draw parameters for organizational interaction and response from operations logs, interviews with operations personnel, and professional reports. We will identify key organizations that were involved in that event from previous study data, amplified by secondary sources, reports, and agency action logs. Using these parameters, we will develop a system dynamics model using the AnyLogic modeling program to model the information flow among organizations, interactions among them, timing, and points of information connection/disconnection that characterized response operations over the 12-hour period of evacuation for the Town of Paradise. The model will design and test dynamic information flow to assist evacuation in other small towns confronting the risk of wildfire.
#14: Coupled fire-atmosphere-fuel moisture modeling driven by satellite data
Authors: Jan Mandel, Martin Vejmelka, Adam K. Kochanski, Angel Farguell, James Haley, Derek V. Mallia, Kirana Bergstrom, Kyle Hilburn
Affiliations: University of Colorado Denver, University of Utah, Colorado State University
Abstract: We present a multiscale coupled weather, fire, fuel moisture, and smoke model, running in an interactive HPC management system. Machine learning and Bayesian data assimilation are used to initialize and drive the model by satellite fire detection data, and to assimilate fuel moisture data. Satellite fire detection products have much lower resolution (375 m at best) than the fire model resolution (30m), there are errors of omission and commission, and data is often missing. For these reasons, we use satellite data to improve the fire simulation in a statistical sense only. Since the atmosphere and the fire become inconsistent when the fire is simply initialized from an advanced fire state or changed by data assimilation, the fire model state is encoded as the fire arrival time, which is then used to spin up the atmosphere model to recover a consistent state.
Setting up a numerical weather prediction model requires a significant expertise, and even more so for a coupled atmosphere-fire model, but the system does not require expert knowledge. It automates the setup of the simulation, download of weather products and satellite data, preprocessing, running multiple jobs on HPC cluster, and visualization.
#15: Improvement of Sub-Modeling in Physical Models of Wildland Fire Spread
Author: Albert Simeoni (asimeoni@wpi.edu)
Affiliation: Department of Fire Protection Engineering, Worcester Polytechnic Institute, Worcester, MA, USA
Abstract: This talk will present an approach that is used to improve CFD fire spread models as tools to better capture and understand the physics that drive fire behavior in wildland fires. First, the CFD modeling approach that we chose will be presented. Then, the different types of sub-models used to close the CFD model will be detailed, along with some of their assumptions and limitations.
The improvement of CFD models will be illustrated through previous research on the representation of gas phase combustion, drag forces in vegetation, convective transfer, and thermal degradation of the vegetative fuel. Different sub-models will be presented with the associated challenge linked to the need of capturing enough physics to enhance the model capacity without making it impossible to implement and validate.
Ongoing work and on heat transfer will also be presented before
concluding with the identification of research bottlenecks and research needs.
#16: Challenges and Opportunities for Using Machine Learning to Identify Mesoscale Fire Behavior Model Sensitivities and Errors
Authors: Daniel Feldman, Jing Li, Ashray Manepalli, Alok Singh, Blake Tickell, Adrian Albert, Brian White, Adam Kochanski, Jan Mandel
Affiliation: Lawrence Berkeley National Laboratory
Abstract: The computational expense of mesoscale fire behavior modeling is extreme. For example, a WRF-SFIRE simulation of a 1000-ha fire requires 70 cpu-hours per hour of simulation. At the same time, there are significant uncertainties in model structure and inputs. Machine learning (ML) methods, while not a substitute for traditional modeling, can enhance traditional modeling capabilities. Through a Generative Adversarial Model (GAN), we can create a fast and efficient emulator of such a mesoscale model, which enables the development of sensitivity calculations to determine the importance, for a given fire, of knowing the temperature, humidity, fuel distribution, fuel moisture, and winds. This machine-learning (ML) model has been trained on hundreds of fires simulated by WRF-SFIRE and incurs between 10 and 100 times less computational expense. We discuss how training, validation, and ensuring that the ML model captures the relevant physics over the parameter space of relevance for a given fire are fundamental, though not insurmountable, challenges for these methods. We also discuss the potential of ML both to shortcut the computational expense associated with WRF_SFIRE, and investigate where ML can, due to reduced computational expense, explore model sensitivities and sources of error.
#17: Flows, Sensing & Wildfires – Data for Modelling and Real-Time Use
Authors: Simo A. Mäkiharju
Affiliation: UC Berkeley, Mechanical Engineering, Berkeley CA 94720
Abstract: Wildfires present an urgent and worsening danger to human life, well-being and property. Spread of fires is influenced by the ambient flow field, and fires themselves can create notable flows. We present quantitative measurement tools suitable both for laboratory and field applications, and a planned experimental effort utilizing multispectral sensing to characterize the spread of a fire up a slope – work in collaboration with those well-versed in wildfires and their modelling. For this effort the FLOW lab, led by Prof. Simo Makiharju, brings to bear modern fluid dynamics research tools for both laboratory and field measurements. Approaches that may be used for this research will be presented, among them a concept for robust real-time flow field measurement to aid in combating wildfires, yielding both local flow field and information on hazardous particles generated by the fire.
#18: Fire-atmosphere coupling: implications for wildfire and air quality modeling
Authors: Adam K. Kochanski, Jan Mandel, Derek V. Mallia, Kyle Hilburn, Steve Krueger, Roger Ottmar, Tim Brown
Affiliations: University of Utah, University of Colorado Denver, Colorado State University, US Forest Service, Desert Research Institute
Abstract: Climate change, in addition to previous land management and fire suppression practices, has resulted in an increase in wildfire activity across the Western U.S. Wildfires across the U.S. are projected to worsen through the 21st Century, thus, their potential to cause damage, deteriorate air quality and modify local weather conditions will continue to increase. Data recently collected during large wildland fires, as well as numerical experiments, indicate that the perception of weather strictly driving the fire behavior may not necessarily be representative of actual wildfire physics. Fires tend to interact with the atmosphere in a complex way. At relatively long timescales, weather impacts fuel moisture, which can affect fire propagation. At shorter timescales, there are dynamical effects associated with the interaction between the rising fire plume and the winds near the fire front. Another fire-atmosphere feedback that will be discussed here is the interaction between wildfire smoke and solar radiation, which can alter atmospheric conditions by blocking incoming solar radiation, which cools the surface and increases atmospheric stability.
In this presentation, we show results from numerical experiments performed using coupled fire-atmosphere model WRF-SFIRE that illustrates the importance of fire-atmosphere coupling when modeling wildfire growth and smoke dispersion.
#19: Fast but not bad: using UVAs and GPUs to build the real time wildfire environment
Authors: Ana Cortés, Carles Carrillo, Antonio Espinosa, Tomás Margalef
Affiliation: Universitat Autònoma de Barcelona
Abstract: The accurate prediction of forest fire propagation is a crucial issue to minimize its effects. In the last years, forest fire spread simulators have proven to be very promising tools in the fight against these disasters. However, to be useful, these tools must deliver accurate results in a very short period of time (minutes if possible). However, the uncertainty of the input data required by wildfire simulators could be a relevant drawback to achieve such a goal. To reduce the impact of the input data uncertainty, different strategies have been developed during the last years going from applying GA to the use of remote sensing to estimate the values of critical variables when the disaster is taking place. However, all strategies lack of obtaining the real-time values of such input data at the required high resolution to forecast wildfire behavior at real-time with high accuracy. The use of new platforms to collect real time data in the same area where the fire is taking place can significantly reduce this input data uncertainty. However, if the site where the fire is taking place has low connectivity, the data cannot be shipped, so it is of the paramount importance to have a platform to perform the forest fire spread simulation in situ using the gathered information. Swarms of UVAs with low consumption GPU on board including also a wide range of embedded sensors to measure almost everything at real-time, are the new forecasting platforms for fighting natural disasters such as extreme forest fire events. The embedded system resulting could exploit the edge computing paradigm distributing the forecasting system into mobiles elements with computing capacities what to would be more efficient to adapt to changing conditions.
#20: UAVs sensing in support of fire modeling
Authors: Maggi Kelly, Sean Hogan, Andy Lyons
Affiliations: Department of Environmental Sciences, Policy and Management, University of California, Berkeley; University of California Division of Agriculture and Natural Resources, Davis, CA
Abstract: Mapping is critical for disaster response, and satellite remote sensing has a long history supporting fire planning and response via modeling fire risk and hazard, mapping vegetation and fuels, and contributing to wildfire behavior modeling. Much of this work utilizes data from moderate-scale sensors such as Landsat (e.g. the USFS’s Rapid Assessment of Vegetation Condition after Wildfire (RAVG) program). However, several characteristics of Unmanned Aerial Vehicles (UAVs): pilot control and on-demand flights, fine spatial resolution, multiple alternative sensor (RGB, multispec, thermal and LiDAR) payloads, and ability to directly sample mean that UAVs will play an increasing role in pre-, during-, and post-fire applications. This talk will give an overview of our work using UAVs across a range of California vegetation types and post-fire vegetation monitoring projects, as well as highlighting important work going on elsewhere. I will also highlight some of the technical limitations to UAVs (battery life, flight time, regulation, metadata standards) that can prohibit more widespread UAV use and sharing of collected data.
#21: Performance-based Fire Design of Bridges Subject to Wildfires
Authors: Mohammadreza Eslami, Khalid M. Mosalam
Affiliations: Department of Civil & Environmental Engineering and Pacific Earthquake Engineering Research (PEER) Center, University of California, Berkeley, USA; Hinman Consulting Engineers, San Francisco, USA
Abstract: Performance-Based Design (PBD) is a modern and efficient framework to conceive and assess complex structural systems (Bontempi 2006), which allows designers to consistently take into account both natural
and man-made hazards. Initially formalized and applied for earthquake engineering applications (FEMA 1997, Krawinkler 2004, ATC 2005), PBD has been recently extended to cope with other design situations, like blast, fire, tsunami and wind scenarios (Hamburger and Whittaker 2003, Rini and Lamont 2008, Riggs et al. 2008, Petrini and Ciampoli 2011).
The 2016 edition of ASCE 7 “Minimum Design Loads and Associated Criteria for Buildings and Other Structures,” for the first time, provided some guidance on the use of performance-based methods to design fire protection. Meanwhile, in 2018 the ASCE/SEI Fire Protection Committee developed its first guideline, “Structural Fire Engineering,” to present best practices for structural engineers working with fire protection engineers. Performance-based fire engineering (PBFE) relies on advanced analysis tools to predict how the structure will behave during a fire. PBFE is an alternative to the traditional prescriptive method of design by qualification testing (DQT), which is used to fireproof almost all buildings in the United States. Given its versatility, PBFE appears to be a viable strategy for a more reliable design of bridges and critical transit infrastructures.
During wildfires, several transit bridges may be subjected to thermal loading. As a matter of fact, these structures play a critical role in the aftermath of wildfires, and therefore their level of structural safety must be rigorously evaluated. Previous experiences show that bridge fires are high-consequence incidents (Kodur 2017). Generally, bridge design codes and standards, in contrast to building codes, do not take into account the concept of fire safety. Fire is not explicitly addressed in the current AASHTO bridge design criteria (AASHTO, 2017), nor is it in most state DOT bridge design manuals. At best, wildfire exposure is briefly mentioned as a consideration in some state DOT documents such as those from Caltrans (2017). However, recent high-profile fire incidents on bridges and in other infrastructure have opened a debate on the need for fire resistance requirements on bridges.
In this study, a review of recent studies related to PBFE of bridges is reported and a proposal for extending the PBFE approach to the case of bridges under wildfire hazards is outlined. Furthermore, using sensor technology, this study seeks to quantify and prioritize risks in a manner that enables smart decision-making to cultivate appropriate emergency planning and/or properly treat wildfire deficient conditions. A strategy to assess and repair fire-damaged bridges is also presented.
#22: Towards Creating an Aerial Fire Hose
Authors: Koushil Sreenath
Affiliation: University of California, Berkeley
Abstract: The first fire hose was created in 1673. Since then, the fire hose has remained relatively unchanged. This talk will propose an idea of enabling a fire hose that is cooperatively carried in the air by a team of UAVs. An aerial fire hose enables quick repositioning of the hose to deliver a continuous supply of water where it is required. We will introduce challenges of creating such a fire hose and discuss potential solutions to some of these challenges. We will present the dynamical of this problem and discuss ideas on addressing some of the issues in planning and control that arise.
#23: Robot manipulation is extreme environments: Applications in infrastructure and disaster response
Author: Hannah Stuart
Affiliation: Department of Mechanical Engineering, University of California, Berkeley
Abstract: Mobile manipulation robots hold great potential to keep people out of harm’s way and perform tasks beyond human capabilities (e.g. deep dives in the ocean, space exploration, entering dangerous structures, etc.). However the current state-of-the-art robotic platforms are not typically deemed both highly robust and highly dexterous at the same time. The Embodied Dexterity Group, led by Prof. Hannah Stuart, works to build new mechanisms that enable these machines to address pressing needs in dangerous environments. One key application is the installation and inspection of infrastructure and the response to events like rapid fire propagation in and around structures. In this talk, an overview of current dexterous mobile robots will be summarized, highlighting technological gaps that must be addressed for future deployment.
#24: UAVs for challenging environments
Author: Mark Mueller
Affiliation: Department of Mechanical Engineering, UC Berkeley
Abstract: Aerial robotics allow for rapid and low cost collection of data over large areas. They are, however, limited by various factors, including especially battery life; ability to operate under large disturbances; and danger to third parties. In this talk I will present some work we have done on creating more capable, more robust aerial vehicles. Specifically, I will discuss how to design aerial systems capable of operating despite large external disturbances by exploiting the addition of angular momentum. I will also present a strategy for increased flight time through in-flight battery replacement, and why simply using larger batteries does not work. This final point is related to the overall mass of the system, which is a primary factor in the relative danger of an aerial robotic system.
#25: Enhanced FUEGO Early Fire Detection System Using Camera Towers and Satellites
Authors: Kinshuk Govil, Carl Pennypacker, and Tim Ball
Affiliation: Fire Urgency Estimator in Geosynchronous Orbit (FUEGO)
Abstract: Over the past few years, the University of Nevada, Reno and the University of California San Diego have established and are growing a network of mountain top cameras for collection/dissemination of intelligence on wildfires. In collaboration with that group, we have developed the ability to detect smoke in the first few minutes after a fire is ignited. We have demonstrated the ability to beat or tie 911 calls on a number of fires and present images to Emergency Command Centers. The images aid dispatchers making certain that they are send forces sufficient to overwhelm the threat.
Simultaneously using several Convolutional Neural Networks trained to detect different features of small smoke plumes in real time, the system analyzes images from static, rotating, and point-tilt-zoom cameras which surveille up to 1000 square kilometers around each installation. At present the system averages less than one false alarm per camera over two days. Further developments in the deep learning system which have already been scoped are expected to reduce false positives by a factor of 10 or more. Where the network is dense enough we are able to confirm fires from more than one camera installation. Our system now regularly polls the more than 300 cameras presently in the network. By the end of 2020 the network is slated to grow to over 1000 installations in California. The network is also expanding to other states. Additional power has been brought to our system by a collaboration with Lawrence Livermore National Laboratory, who are exploring use of existing satellites to detect fires early and plot the movement of on-going fires.
#26: Scientific and Operational needs for modest temporal & spatial resolution mapping at targeted wavelengths
Authors: J. Timothy Ball, Carlton R. Pennypacker, and Kinshuk Govil
Affiliation: Fire Urgency Estimator in Geosynchronous Orbit (FUEGO)
Abstract: It should not be surprising that scientists seeking to model wildland fire behavior and operational firefighters concerned about immediate safety, tactical options, and tomorrow’s strategy want to know very similar things; essentially, how can we know what the fire will do in 15 minutes and tomorrow. Unfortunately, both groups suffer from a near complete lack of even moderate temporal and spatial resolution quantitative empirical data from unconstrained fires. Just from a safety perspective firefighters and the public deserve frequent quantitative updates on fire behavior most helpfully in the form of a map showing: (a) fuel type (b) current perimeter location, (c) present rate of spread, (d) intensity (rate energy release), (e) flame size (area from which energy is being released), (f) spotting distance & frequency, (g) slope, (h) exposure aspect, (i) wind speed & direction, and from a representative weather station air temperature and relative humidity. These and their rates of change are the fundamental parameters for all fire behavior models. Indeed a very useful first approximation of fire spread for tactical and public safety use is a simple forward extrapolation of the measured rate of spread. Models that seek to predict fire behavior over periods of hours need to take changing conditions but it is also critical that the models start from real data. Perhaps the best way to learn and improve all flavors of fire models is data assimilation.
A fire mapping system consisting of shortwave IR (for flame size and location), midwave IR (for energy release), and longwave (for perimeter location and hotspot detection/mop-up) will be described. These systems can be scaled with essentially no change in basic design from something that can be flown on small fixed-wing UAV to a high altitude/wide-swath fast aircraft.
#27: Persistent Autonomous Monitoring for Wildfire Detection
Author: Katia Obraczka
Affiliation: UC Santa Cruz
Abstract: In this talk, we will describe an interdisciplinary effort which brings together engineers and fire scientists, in partnership with Cal Fire (the State of California’s agency responsible for wildland fire protection), aiming atdesigning, developing, deploying, and testing in the field a cyber-physical system based on Internet ofThings (IoT) technology for detecting conditions conducive to rapid fire initiation and growth.The proposed system consists of wireless nodes that integrate an heterogeneous sensor suite as wellas smart sensor processing algorithms aiming at: 1) dramatically reducing the time between dataacquisition and determining whether a wildfire event is imminent; 2) sustaining data collectionoperations for extended periods of time in harsh natural environments with minimal or no humanintervention; 3) collecting data at unprecedented spatial and temporal granularity that will enablenovel algorithms to (i) determine wildfire potential and (ii) inform how to refine and improve theuse of cyber-physical systems in this domain.