EscapeWildFire: Assisting People to Escape Wildfires in Real-Time
Andreas Kamilaris, Jean-Baptiste Filippi, Chirag Padubidri, Jesper Provoost, Savvas Karatsiolis, Ian Cole, Wouter Couwenbergh, Evi Demetriou
EEscapeWildFire: Assisting People to EscapeWildfires in Real-Time
Andreas Kamilaris ∗† , Jesper Provoost † , Jean-Baptiste Filippi ‡ , Chirag Padubidri ∗ , Savvas Karatsiolis ∗ ,Ian Cole ∗ , Wouter Couwenbergh † and Evi Demetriou ∗∗ CYENS Center of Excellence, Nicosia, CyprusEmail: { a.kamilaris, c.padubidri, s.karatsiolis, i.cole, e.demetriou } @cyens.org.cy † Department of Computer Science, University of Twente, Enschede, The NetherlandsEmail: { j.c.provoost, w.couwenbergh } @student.utwente.nl ‡ Centre national de la recherche scientifique (CNRS) SPE, Universit`a di Corsica, FranceEmail: filippi [email protected]
Abstract —Over the past couple of decades, the number ofwildfires and area of land burned around the world has beensteadily increasing, partly due to climatic changes and globalwarming. Therefore, there is a high probability that more peoplewill be exposed to and endangered by forest fires. Hence there isan urgent need to design pervasive systems that effectively assistpeople and guide them to safety during wildfires. This paperpresents EscapeWildFire, a mobile application connected to abackend system which models and predicts wildfire geographicalprogression, assisting citizens to escape wildfires in real-time. Asmall pilot indicates the correctness of the system. The code isopen-source; fire authorities around the world are encouraged toadopt this approach.
Index Terms —Wildfires, Escape routes, Mobile app, Fire prop-agation model.
I. I
NTRODUCTION
Over the past couple of decades the number of wildfires andarea of land burned has been steadily increasing [1], partlydue to climatic changes and global warming [2]. A wildfireis defined as a large, destructive fire that spreads quickly overwoodland or brush. It differs from fires that occur in urbaninfrastructures and buildings. Every year approximately half amillion hectares of land are burned by wildfires in southernEurope [3]. Moreover, the frequency of these fires is projectedto increase by 27% in the coming decade [4], [5]. There is ahigh chance that more people will be endangered by wildfiresin the coming years and this will likely be associated withfurther casualties [5]. Countries tend to deal with increasingwildfire hazards by improving their equipment and capacityin human resources, employing emerging technologies such asdrones [6], [7]. However, the infrastructures of most countriesare lacking in the evacuation of humans from the fire zones,despite evacuation plans being in place. Furthermore, the evac-uation plans in regions where forest fires are not common (e.g.northern Europe such as Netherlands and Nordic countries)have not been fully implemented or tested for safety [8].The contribution of this paper is to address this gap and helptackle the limited preparedness of many countries and regionsaround the world for wildfire evacuation. A mobile applicationand backend, which assist people in danger to safely escapewildfires in real-time, are presented in this work. This software has been created as open-source code , together with completeuser manuals and instructions. The authors wish to encouragefire authorities around the world to embrace and implementthis approach.The rest of the paper is organized as follows: Section IIpresents related work and Section III describes design anddevelopment of the EscapeWildFire system. Then, Section IVexplains our evaluation efforts through a case study at the re-gion of Nicosia, Cyprus while Section V shows the evaluationresults. Finally, Section VI discusses overall implications andSection VII concludes this paper.II. R ELATED W ORK
Related work in the field of fire evacuation has mainly fo-cused on evacuation from inside buildings [9], predominantlypublic buildings. Some authors have worked on evacuatingneighborhoods [10], trains [11], hospitals [12] and road tunnels[13]. Authors have investigated different protocols, escaperoutes, time required to evacuate depending on building’sfloor or position inside the road tunnel, human behavioralpatterns during fires, congestion, etc. Some evacuation modelshave been used to assess and analyze the safety of variousinfrastructures [11], [13] and to forecast and visualize thewildfire in geo-visualizations [14].Regarding wildfires in particular, related scientific worktargeting real-time evacuation assistance is very limited. WFAPocket [15] is a tool targeted for fire fighters, designed tobe used in the field, modelling the progression of the firebased on the user inputs, such as the fuel types, real-timeweather data, etc. WiSE is a tool designed to provide safeseparation distance (SSD) calculations between wildfires andfire fighters. Gaia GPS is a mobile app that has been usedby firefighters to plan escape routes, mark fire lines, and trackprogression. Finally, Emergency is an application developed bythe American Red Cross [16], with the aim to keep people safein severe weather conditions as well as man-made or naturaldisasters. EscapeWildFires software. https://github.com/rise-centre/Escape-Wildfires Wildfire Safety Evaluator, https://wise.wildfireanalyst.com a r X i v : . [ c s . C Y ] F e b ig. 1. Snapshots of the mobile app: Fire illustration (left), turn-by-turn view(center), direction-based view (right). III. D
ESIGN AND D EVELOPMENT
This section discusses the design decisions and implemen-tation details of the fire evacuation software.
A. Mobile App Design
The main idea is to maintain a minimalist design approach(see Figure 1). The fire location and progression indicator isred, with dark red at the center of the fire and increasingtransparency further from the center, indicating time periodsof fire progression (Figure 1, left). Each polygon of increasingtransparency indicates the expected propagation of wildfireafter 15 minutes. For example, in Figure 1 (left), there are fivelayers of the wildfire, at 0, 15, 30, 45 and 60 minutes afterignition. Two different navigational methods were considered:a turn-by-turn view (similar to that used in Google Maps) anda direction-based view (similar to compass navigation). Forthe turn-by-turn view, (Figure 1, center),the top-right cornerindicates the next turn and the turn immediately after, alongwith the distance until action. At the bottom of the screen, theuser has the option to open a complete list of directions. This isa scroll-enabled list, listing turn types alongside the distanceto each turn. Next to the direction list is another button topause navigation, returning the user’s screen to an overview ofthe route. Above these buttons, two zoom buttons are placed,offering the option for fine zoom control. For the direction-based view, an arrow inside a white circle (Figure 1, right)continuously shows the indicated direction of the user, basedon current location and orientation.
B. Backend Implementation
Besides the mobile app, a backend system supports themanagement of vital information, such as where/when the firestarted, how fast it will propagate based on vegetation (fuel)type, wind speed and direction, weather conditions, etc. Thearchitecture of the backend, in relation to the mobile app, isprovided in Figure 2. The architecture is composed of thefollowing components: • Mobile application (see Sections III-A and III-B1) • Fire behaviour model (see Section III-B2) • Web server (see Section III-B3) • Fire management tool (see Section III-B4)Each fire department has its own version of the fire man-agement tool (FMT), which allows them to add ignitionpoints in areas of their responsibility. The fire behaviourmodel (FBM) analyzes the area around the ignition points andcalculates the fire propagation. The Web server communicatesthe information from the FBM to the mobile apps of the endusers and data is presented in the form of polygons, indicatingthe spread of the wildfire in 15-minute intervals, up to 60minutes ahead of time. The mobile app calculate safe escaperoutes for the users for effective evacuation.
1) Mobile application:
Implemented in Android, based onthe design decisions listed in Section III-A. The map provideris HERE maps . The app has been listed on Google Play to facilitate the evaluation process (i.e., for pilot participantsto easily download, see Section V). The most importantalgorithm is the one used for calculating escape routes. Eachpossible route gets a score, depending on the time needed toreach safety , where safety is defined as a one kilometer (km)distance from the wildfire after one hour from its ignition time,plus the angle from the fire’s propagation direction: ∗ r = argmax { r (1) , r (2) , ..., r ( x ) } (1)where ∗ r the best route selected, x the total number of possibleroutes and each route is calculated as: r ( i ) = a ( i ) × angle ( i ) + (1 − a ( i )) × time to safety ( i ) , < a ( i ) < (2)where angle is the angle between direction of wildfire andsuggested route, time to saf ety is the time needed to reacha safe place, while the parameter α suggests the weight ofeach parameter, which differs between transport types (matterof future work). Routes that pass through wildfire zones (seeFigures 1 and 4) during the 15-minute slots when the wildfireis expected to reach those zones are rejected immediately.Travel calculations are considered based on conservative val-ues for user travel speed (Walking: 4.5 km/h, Cycling: 15km/h, Driving car: 50 km/h).
2) Fire behaviour model:
Wildland fire spread simulationsare performed by the numerical solver ForeFire [17]. ForeFirerelies on a front-tracking method where the fire front isrepresented by Lagrangian markers that are linked to eachother via a dynamic mesh. While the tool can theoreticallyuse any formulations, currently the velocity of every point ofthe front is provided by the model of Rothermel [18]. Therate of spread (ROS) is expressed as a function of severalenvironmental properties such as wind speed, terrain slope,fuel moisture content and other fuel parameters characterizingthe vegetation. A simulation mostly consists of the definition HERE Maps. https://mobile.here.com/?x=ep Escape Wildfire mobile app. https://play.google.com/store/apps/details?id=com.ewf.escapewildfireig. 2. Architecture of the EscapeWildFire system. of an initial state of the fire front and then the ROS is computedfor the markers of the fire front based on underlying 2D fieldsfrom which environmental properties are determined. ForeFirerelies on a discrete-event approach where most computationsdeal with the determination of the time at which the markerswill reach their next destination. Destination here is definedby a fixed spatial increment in the direction at the tangent tothe perimeter of the burned area. Real-time wind informationis provided by the Windy online service. The land typedistribution field comes from CORINE Land Cover data [19].The elevation field is extracted from the NASA SRTM dataset,which originally has a 30 m resolution. Fuel parametrizationis performed to assign reference fuel parameters to each typeof vegetation in the land use data for the ROS computations.Data includes 2D fields of wind speed vectors, to account forthe influence of the elevation field.
3) Web server:
The Web server is responsible for linkingthe mobile apps of the end users with the backend (i.e. FMTand FBM). The Web server uses the HERE HYZ databaseand its Maps API in order to properly model fire propagationas polygons, in a language understood by the map providersoftware used by the mobile app. HERE HYZ is a locationdata management cloud service, built around standards likeGeoJSON. Each fire is modelled as a GeoJSON geometryobject. When accessing the service through HERE Maps API,communication is through GeoJSON files.
4) Fire management tool:
This software is intended to beused by fire departments to manage wildfire incidences, i.e.add/remove, ignite/stop and configure wildfires in real-time.New fires can be added via the ”Add new wildfire” button,shown in Figure 3 (left). When the fire officer clicks thisbutton, relevant details can be added (Figure 3, right) such asthe exact location where the wildfire started, time of ignitionand other important information to be shared with stakeholders(i.e. fire fighters and citizens). HERE Studio. https://developer.here.com/products/platform/studio
IV. E
VALUATION
To assess the proposed system, a scenario of a wildfire wassimulated selecting a natural reserve as the landscape type.This landscape is located in the region of Nicosia, Cyprus,illustrated in Figure 4, as visualized by the ForeFire firebehaviour modeller [17] (see Section III-B2). In the figure, firehas been ignited near the center of the map, while red circlesshow the spread of the fire in subsequent one-hour slots. Atthe top of the window, the hectares covered by the fire ateach hour after ignition are displayed, while at the bottom,parameters such as wind speed, humidity and temperature canbe set. Table I shows the characteristics of the scenario understudy.Participants were selected randomly as they were passingby on foot. They were asked to participate in our pilotstudy simulating the scenario of a fire occurring near them.The evaluation exercise took place in the afternoon (around17:00-19:00), coinciding with a high likelihood of free time.Participants were asked to download the mobile app fromGoogle Play and then follow the instructions given by the app.Afterwards, they were asked to fill a questionnaire, elicitinginformation about demographics, plus the questions listed inthe first two columns of Table II. As compensation for theirefforts, participants received a voucher for free coffee at apopular caf´e in the local area.V. R
ESULTS
Seventeen people in total agreed to participate in the pilot.Their demographic information is shown in Table III. Genderwas equally balanced. The large majority of participants werelocals. Most participants were well educated, possessing asdiploma, bachelor or master level degree. The last threecolumns of Table II show the responses of the participantsafter using the mobile app to safely escape the fire. The largemajority found the application easy to use, with little effortrequired to follow the advice given. The vast majority felt theywould be safer using the app when in actual danger and wouldkeep it downloaded on their mobile phones. Two participants ig. 3. Snapshots of the fire management tool.Fig. 4. Evaluation area: Athalassa natural reserve. The ignition of wildfire, participants’ initial positions and suggested escape route.TABLE IS
CENARIO UNDER STUDY AND CHARACTERISTICS / PARAMETERS . Area Type Participants Transport Wind Humidity Temperature Time to evacuate
Athalassa Natural reserve 17 Walking 6 km/h (east-south) 30% ° Celsius 20 minutes (females, 43 and 67 years) argued that during a fire they cannotfeel safe with or without an mobile app as assistance.In all 17 cases, the app helped the users to successfullyevacuate the area. All but one participant felt like the appworked appropriately, following the route suggested by theapp. All participants agreed that the route suggested by theapp was, to their knowledge, the best available for safelyescaping the wildfire. To escape the fire, people needed around minutes to walk the meters to safety ( . km/h,see Figure 4). There was a large standard deviation in time( minutes) for older people (51+) needing considerablymore time to walk than younger people (16-35) ( vs.
13 : 06 minutes). This indicates that the app should considerthe physical condition of the user, partly reflected through agealthough other metrics can be used. VI. D
ISCUSSION
This paper has presented a solution to the problem ofassisting people to safety during wildfires by evacuation.Various technologies have been employed to build a robustsolution. The evaluation focused mostly on the feasibility ofthe proposed system and the correctness of the mobile app,touching upon human factors and aspects such as acceptanceand usefulness. The ForeFire fire propagation prediction modelhas not been evaluated in this paper, but it has been extensivelyevaluated both in simulation [17], [20] and in real-worldwildfire situations [21], [22].The system has worked correctly during the pilot. The firemanagement tool added new wildfires precisely, while the Webserver communicated correctly with the mobile apps of theparticipants to show the fires and suggest escape routes. All
ABLE IIQ
UESTIONNAIRE USED DURING THE EVALUATION PROCESS . Question Type Responses
How easy was the application to understand in general? Likert scale, from very difficult (1) tovery easy (5) Avg. value: 3.9 St. dev.: 0.9How easy was the application to follow using the advicegiven? Likert scale, from very difficult (1) tovery easy (5) Avg. value: 3.6 St. dev.: 0.8Did you feel safer while using the application, althoughthere was a dangerous situation near you? Yes/No Yes (82%), No (18%)Would you keep this app downloaded on your phone? Yes/No Yes (94%), No (6%)Did the application help you escape the fire safely? Yes/No Yes (100%)Did you feel like the application worked the way itshould have? Yes/No Yes (100%)How likely is it to suggest this application to someone? Likert scale, from very unlikely (1) tovery likely (5) Avg. value: 4.1 St. dev.: 0.7How likely is it to use this application in a real lifescenario? Likert scale, from very unlikely (1) tovery likely (5) Avg. value: 4.1 St. dev.: 1.2What improvements would make the app more useful,effective and/or user friendly? Open question See Section VI.How much time did you need to escape the fire? Response in minutes:seconds (for 0.8 km) Avg. value: 8:30 St.dev.: 5:25Did you follow 100% the route suggested by the app?If not, why? Open question Yes (94%), No (6%)Was the route suggested by the app the best one to yourknowledge, in order to safely evacuate? Open question Yes (100%)TABLE IIID
EMOGRAPHIC INFO OF PARTICIPANTS . Area Participants Gender Ages Nationality Education
Athalassa 17 Male (8), Female(9) 16-20 (1), 21-35 (5),36-50 (6), 51+ (5) Cypriot (14), Greek (2),Kurdish (1) Diploma (5), Bachelor(10), Master (2) participants agreed that the routes suggested by the app werethe best available escape routes to their knowledge (see TableII) and also agreed that the app helped them escape the firesafely. Unfortunately, the fire behaviour model, could not beevaluated in this work. However, the model has been evaluatedin related work [17].In terms of acceptance and usefulness, 94% of participantswould keep the app on their mobile phones, while 82% feltsafer using the app during a wildfire event. In a Likert scale of1 (very unlikely) to 5 (very likely), participants gave a scoreof . (likely) to the question of whether they would use theapp in a real-life wildfire situation. It was observed that olderpeople (51+) found it hard to trust mobile apps, especially inemergency situations.Concerning ease of use, participants scored . for thequestion whether the app was easy to understand (Likert scale,1: very difficult to 5: very easy). Participants gave a scoreof . for the question of how easy it was to follow theadvice given by the app. This perhaps reveals some difficultyof navigation outside of areas with designated roads.The system and pilot presented in this paper had some limi-tations, which will be addressed in future work. More detailedland-use datasets and more specific fire velocity models needto be included, while pilots should engage more participantsin a range of different landscapes and modes of transport (e.g.cycling, driving a car, etc.), considering scenarios where usersdo not have much time to evacuate wildfire and need to rush.Finally, an important limitation is that our solution does not embrace or provision solutions for people with disabilities.An important benefit of the evaluation process was thecollection of useful suggestions by the participants. It wasevident that older people required better accessibility featureson the mobile app (e.g. larger buttons, vivid colors, flashingdirections and arrows, verbal/textual directions and visualfeatures, etc.). There was a need to accurately define andassist the orientation of the user. This could be indicatedby the phone compass showing the magnetic north whilstadjusting the map to the user orientation. Showing the historicmovement of the user will also help facilitate orientation. Thebuilt-in compass of many mobile phones could also be usedfor areas without GPS coverage.Some additional requirements recorded by participants inresponse to the question ”what improvements would make theapp more useful and user-friendly” include: • Way-finding using landmarks (e.g. mountain tops, hills,buildings), useful for some pedestrians. • Assisting people in need , i.e. locating humans who areclose to the suggested escape route. • Proactive information about risks in each area . • Alternative routes calculated by monitoring users duringevacuation to avoid congestion that would delay theevacuation process. • A personalized speed of evacuation for each user, de-pending on user’s overall expected fitness, which couldbe partly derived from his/her age. • Automatic notification for nearby wildfires before theyecome life-threatening. • Exploitation of civil defence infrastructures , e.g. includeroutes to nearby shelters [23].
A. Future work
An important action is to validate how the app behaves torescue humans, assuming hypothetical placements of peoplenear the wildfire area, examining escape paths proposed.For wildfires, we plan to consider historical data relating toprior recordings of wildfires. Further, future work will betterconsider the values of α and expected transport speed (seeSection III-B1), while more personalized values depending onsomeone’s fitness or vehicle will be considered.VII. C ONCLUSION
This paper presented EscapeWildFire, a mobile app andbackend for assisting citizens to escape wildfires in real-time.The system provides a complete solution, addressing the issuesof: a) recording wildfire locations, indicating the exact positionand time of ignition; b) modelling and predicting the wildfireprogression, based on the vegetation and fuel type aroundthe fire, and meteorological parameters such as wind speed;and c) providing a mobile application which assists people tosafety by calculating and guiding users through best evacuationroutes. A small pilot study was conducted considering anatural reserve at the island of Cyprus. Results show thatEscapeWildFire constitutes a correct, accepted, easy to use andrealistic solution to the problem under study. Some limitationsof the system and the evaluation process performed havebeen recorded and discussed. Various recommendations forimprovements have also been recorded based on the feedbackof the participants. The code of the project is available as open-source; the authors wish to encourage fire authorities aroundthe world to adopt this approach.A
CKNOWLEDGEMENTS
This project has received funding from the EuropeanUnion’s Horizon 2020 research and innovation programmeunder grant agreement No 739578 complemented by theGovernment of the Republic of Cyprus through the Direc-torate General for European Programmes, Coordination andDevelopment. R
EFERENCES[1] Philip E Dennison, Simon C Brewer, James D Arnold, and Max AMoritz. Large wildfire trends in the western united states, 1984–2011.
Geophysical Research Letters , 41(8):2928–2933, 2014.[2] John T Abatzoglou and A Park Williams. Impact of anthropogenicclimate change on wildfire across western us forests.
Proceedings ofthe National Academy of Sciences , 113(42):11770–11775, 2016.[3] Francisco Moreira, Olga Viedma, Margarita Arianoutsou, Thomas Curt,Nikos Koutsias, Eric Rigolot, Anna Barbati, Piermaria Corona, PedroVaz, Gavriil Xanthopoulos, et al. Landscape–wildfire interactions insouthern europe: implications for landscape management.
Journal ofenvironmental management , 92(10):2389–2402, 2011.[4] Yaoxian Huang, Shiliang Wu, and Jed O Kaplan. Sensitivity of globalwildfire occurrences to various factors in the context of global change.
Atmospheric Environment , 121:86–92, 2015.[5] Giovanni Forzieri et al. Multi-hazard assessment in europe under climatechange.
Climatic Change , 137(1-2):105–119, 2016. [6] Andreas Kamilaris and Francesc X. Prenafeta-Bold ˜Aº. Disaster moni-toring using unmanned aerial vehicles and deep learning. In
DisasterManagement for Resilience and Public Safety Workshop, in Proc. ofEnviroInfo2017 , Luxembourg, September 2017.[7] Andreas Kamilaris, Corjan van den Brink, and Savvas Karatsiolis. Train-ing deep learning models via synthetic data: Application in unmannedaerial vehicles. In
International Conference on Computer Analysis ofImages and Patterns , pages 81–90. Springer, 2019.[8] Erica Kuligowski. Evacuation decision-making and behavior in wild-fires: Past research, current challenges and a future research agenda.
Fire Safety Journal , page 103129, 2020.[9] Jianyong Shi, Aizhu Ren, and Chi Chen. Agent-based evacuationmodel of large public buildings under fire conditions.
Automation inConstruction , 18(3):338–347, 2009.[10] Thomas J Cova and Justin P Johnson. Microsimulation of neighborhoodevacuations in the urban–wildland interface.
Environment and PlanningA , 34(12):2211–2229, 2002.[11] JA Capote, D Alvear, O Abreu, and A Cuesta. Analysis of evacuationprocedures in high speed trains fires.
Fire Safety Journal , 49:35–46,2012.[12] Ernest Sternberg, George C Lee, and Danial Huard. Counting crises:Us hospital evacuations, 1971–1999.
Prehospital and Disaster Medicine ,19(2):150–157, 2004.[13] Daniel Nilsson, Maria Johansson, and H˚akan Frantzich. Evacuationexperiment in a road tunnel: A study of human behaviour and technicalinstallations.
Fire Safety Journal , 44(4):458–468, 2009.[14] M Castrill´on, Pedro A Jorge, Ignacio J L´opez, Adri´an Mac´ıas, D Mart´ın,Rafael J Nebot, Izzat Sabbagh, FM Quintana, Javier S´anchez, Antonio JS´anchez, et al. Forecasting and visualization of wildfires in a 3dgeographical information system.
Computers & geosciences , 37(3):390–396, 2011.[15] Santiago Monedero, Joaquin Ramirez, and Adri´an Cardil. Predicting firespread and behaviour on the fireline. wildfire analyst pocket: A mobileapp for wildland fire prediction.
Ecological Modelling , 392:103–107,2019.[16] Daniel J Bachmann, Nathan K Jamison, Andrew Martin, Jose Delgado,and Nicholas E Kman. Emergency preparedness and disaster response:there’s an app for that.
Prehospital and disaster medicine , 30(5):486–490, 2015.[17] Jean-Baptiste Filippi, Fr´ed´eric Morandini, Jacques Henri Balbi, andDavid RC Hill. Discrete event front-tracking simulation of a physicalfire-spread model.
SIMULATION , 86(10):629–646, 2010.[18] Richard C. Rothermel. A mathematical model for predicting fire spreadin wildland fuels. Res. Pap. INT-115. Ogden, UT: U.S. Department ofAgriculture, Intermountain Forest and Range Experiment Station. 40 p,1972.[19] J. Feranec, T. Soukup, G. Hazeu, and G. Jaffrain.
European landscapedynamics: CORINE Land Cover Data.
CRC Press, Boca Raton, USA,2016.[20] Fr´ed´eric Allaire, Jean-Baptiste Filippi, and Vivien Mallet. Generationand evaluation of an ensemble of wildland fire simulations.
Internationaljournal of wildland fire , 29(2):160–173, 2020.[21] Jean-Baptiste Filippi, Vivien Mallet, and Bahaa Nader. Evaluation offorest fire models on a large observation database.
Natural Hazards andEarth System Sciences , 14(11):3077–3091, 2014.[22] Paul-Antoine Santoni, Andrew Sullivan, Dominique Morvan, andWilliam E Mell. Forest fire research: the latest advances tools forunderstanding and managing wildland fire, 2011.[23] Thomas J Cova, Frank A Drews, Laura K Siebeneck, and AdrianMusters. Protective actions in wildfires: evacuate or shelter-in-place?