A_data_model_for_route_planning_in_the_case_of_forest_fires

A data model for route planning in case of forestfiresZhiyong Wang a,∗,Sisi Zlatanova a,Aitor Moreno b,Peter van Oosterom a,Carlos Toro ba Delft University of Technology,Jaffalaan9,2628BX Delft,The Netherlandsb Vicomtech,Mikeletegi Pasealekua57,20009San Sebastian,SpainAbstractThe ability to guide relief vehicles to safety and quickly pass through environments affected byfires is critical infighting forestfires.In this paper,we focus on route determination in the case of forestfires,and propose a data model that supportsfinding paths among moving obstacles.This data model captures both static information,such as the type of the response team,the topology of the road network,and dynamic information,such as sensor information,changing availabilities of roads during disasters,and the position of the vehicle.We used afire simulation model to calculate thefire evolution.The spread of thefire is represented as movements of obstacles that block the responders’path in the road network.To calculate safe and optimal routes avoiding obstacles,the A*algorithm is extended to consider the predicted availabilities of roads.We prove the optimality of the path calculated by our algorithm and then evaluate it in simulated scenarios.The results show that our model and algorithm are effective in planning routes that avoid one or morefire-affected areas and that the outlook for further investigation is promising.Keywords:Emergency navigation,Fire simulation,Data model,Algorithm1.Introduction1Naturalfires have caused enormous socioeco-2nomic losses and created many victims in the past 3few years.Recently,there has been growing in-4terest in understanding and mitigating the effects 5of these disastrous events.Infighting forestfires, 6a wide range of response activities and emergency 7operations are involved,such as transporting in-8jured persons,distributing supplies,and evacuat-9ing citizens,all of which require navigation aids. 10Because the radiant heat released during burning 11can be considered obstacles that might make some 12roads unsafe and temporarily inaccessible(Taylor 13and Freeman,2010),emergency managers need a 14path planner that is capable offinding a safe and 15optimal route that avoidsfire-affected areas.16Navigation has been thoroughly studied from 17varied theoretical perspectives and across multi-18ple disciplines,such as robotics,geomatics and ap-19plied mathematics(Chabini and Lan,2002;Ge and 20Cui,2002;Huang et al.,2007;Delling et al.,2009). 21∗Corresponding author.Tel.:+31(0)152787934;fax: +31(0)152784422.E-mail addresses:Z.Wang-1@tudelft.nlNevertheless,very few research efforts have been 22devoted specifically to emergency navigation prob-23lems in the context of moving obstacles that dynam-24ically affect the road network(Wang and Zlatanova, 252013b).Although some studies have some relevance 26for route planning in case of disaster events(Mioc 27et al.,2008;Liu et al.,2006),the issues that arise 28in the path planning during disasters have not yet 29been fully addressed.On one hand,the existing 30emergency support systems(Parker et al.,2008; 31Johnson,2008)are capable offinding the short-32est route to a certain location,taking the dam-33ages to the infrastructure into account,but do not 34consider the dynamics of disasters,particularly the 35predicted information on their developments,which 36limits their practical applications in disaster re-37sponse.Some studies of emergency navigation used 38crowdsourced data regarding the state of the road 39to calculate the shortest path(Nedkov and Zla-40tanova,2011;Neis et al.,2010).However,they can 41only cope with static obstacles,and do not offer 42the routing functionality required to avoid moving 43obstacles.On the other hand,most research on dy-44namic obstacles has been centered on robotics(Li 45Preprint submitted to Computers and Geosciences November19,2013et al.,2009;Masehian and Katebi,2007;Gonzalez 46et al.,2012).The results from these studies could 47benefit the navigation offirst responders in certain 48aspects.Nevertheless,the focus of their research 49is mainly on planning obstacle-avoiding paths in a 50given free space,without the constraints of a trans-51portation network.52One of the most critical aspects in emergency 53navigation is information,most of which falls into 54two categories,static and dynamic.Static informa-55tion is relevant to topographic and territorial data 56(e.g.,land use,road network,buildings,and loca-57tions offire hydrants).Most of the static data can 58be obtained through municipality offices and the 59emergency reponse(ER)sectors,as well as pub-60lic resources,such as the location offire hydrants 61on www.openfi and general maps from 62OpenStreetMap().Dy-63namic information is more related to the incident 64description and its impacts,damages,and sensor 65measurements,etc.,and has a highly temporal as-66pect,i.e.,it changes rapidly with time.This infor-67mation consists of historic information,about what 68has happened since the disaster occurred,and pre-69dicted information,about what may happen.Ex-70amples of historical information are the type,scale, 71and affected area of an incident,the number of in-72jured and missing people,etc.This information is 73needed to help emergency managers identify dan-74gerous areas that should be avoided.Examples of 75predicted information are the likelihood offloods 76in a given2.5-dimensional terrain,areas threatened 77by gas plumes,and the forecasted wildfire front, 78etc.Such information is also needed to assist plan-79ners in adjusting original route plans in advance of 80developing disasters.81For the above reasons,a hazard simulation model 82that is capable of providing reliable predicted infor-83mation about disaster changes,is a valuable frame-84work that underlies the solutions for many prob-85lems that arise in the context of advance rescue 86planning.Many disaster models have emerged to 87encourage and facilitate emergency operations in 88the past few years(Hu,2011;Moreno et al.,2012, 892011;Zelle et al.,2013;Lu et al.,2008).For exam-90ple,Zelle et al.(2013)present an integrated system 91for smoke plume and gas cloud forecasts,combining 92a weather model,a smoke plume model and a crisis 93management system.Moreno et al.(2011)present 94a real-timefire simulation algorithm that can be in-95tegrated into interactive virtual simulations where 96firefighters and managers can train their skills. 97These models make it possible for emergency work-98ers to assess the potential impact of a hazard,iden-99tify dangerous areas that should be evacuated,and 100make effective plans to curb damages and protect 101lives.102In our research,a geo-Database Management 103System(geo-DBMS)is selected to manage hazard 104simulation results and dynamic information of geo-105graphic objects.The Geo-DBMS provides efficient 106management of large spatial data sets(often en-107countered in large scale events).In addition,it has 108mechanisms that enable fast update and access to 109geographic information,and functionality for data 110analysis.The geometric model,which has been 111used and implemented in major geo-DBMSs(e.g., 112Oracle Spatial,PostGIS)(Meijers et al.,2005), 113makes the systems capable of handling all types of 114spatial data related to disaster management.Some 115data models haven been developed in geo-DBMSs 116for emergency response(Dilo and Zlatanova,2011; 117Kwan and Lee,2005;Zlatanova and Baharin,2008). 118However,they are not capable of dealing with pre-119dicted information from hazard simulation models 120and can not support routing among moving obsta-121cles.Many researchers have been working on man-122aging moving objects and numerous data manage-123ment techniques have been developed to facilitate 124the collection,organization,and storage of dynamic 125data of moving objects(Wolfson et al.,1998;Merat-126nia,2005;G¨u ting et al.,2006).These studies pro-127vide a rich set of solutions for managing the dy-128namic information produced during disasters,such 129as the locations of the rescue unit,plume move-130ment,and changes in the water level.131In this paper,we focus on the routing process 132in a real road network in the case of forestfires. 133We use afire simulation model to generate datasets 134about the spread of thefire,and obtain information 135about its damage to the infrastructure through spa-136tial data analysis.A spatio-temporal data model 137is proposed to structure dynamic information of 138transportation conditions affected byfires in the 139ing this information,we apply a mod-140ified shortest path algorithm to calculate optimal 141paths avoidingfire-affected areas forfirst respon-142ders.Such an approach is not limited to route plan-143ning during forestfires,but also can be extended to 144assist navigation among moving obstacles brought 145about by other types of disasters.146The organization of the paper is as follows.In 147section2,we describe our system architecture for 148emergency navigation.Section3presents both con-1492Figure1:The overview of the proposed system architectureceptual and logical spatio-temporal data models of 150the dynamic information for routing to avoid ob-151stacles.Section4illustrates the network analysis 152application,including the extended A*algorithm. 153Section6describes the detailed implementation of 154our navigation system.In section7,we test the 155model and the algorithm in different scenarios,and 156detail our results.We draw some conclusions in 157section8and end this paper with proposed future 158work in section9.1592.System architecture160To assistfirefighting in forest areas,a system 161architecture for routing avoidingfire-affected areas 162is designed.The framework of the proposed sys-163tem is depicted infigure1and is composed of the 164following components:data collection,data man-165agement,fire simulation model,agent-based simu-166lation model and visualization of simulation results. 167When afire incident occurs,several measurement 168teams are formed and sent into thefield to per-169form measurements.Real-time sensor information 170(e.g.,wind speed and wind direction)is collected 171from thefield via a communication network and in-172corporated into thefire simulation model(Moreno 173et al.,2012).Thefire model produces dynamic data 174of spatial units about thefire state,from which the 175shape and direction of movement offires are de-176rived.This dynamic information,together with the 177geo-information of the network and the information 178regarding response units(routes,starting point, 179end point,status,etc.)is consistently recorded 180and structured in a geo-DBMS based on the data 181model designed for emergency response(Dilo and 182Zlatanova,2011).We use an agent-based simulator 183with GIS functionalities to predict the availabilities 184of roads in a certain area at a certain time,and to 185display the movement of both thefire and respon-186ders.Thefire simulation results are represented 187as one or more moving polygons crossing a certain 188road network.Thefirst responder is modeled as 189an agent characterized by a set of attributes(e.g., 190speed,type of vehicle)and performs certain actions 191(e.g.,moving,waiting).Using predicted informa-192tion about the status of roads,the path planner, 193within the agent,applies the shortest path algo-194rithm to calculate the safest and fastest route for 195responders.The calculated results are visualized 196to users through a2D view as well as a navigable 1973D view to enhance human situational awareness 198(Schurr et al.,2005).1993.Data model design200A spatial temporal data model is needed to effec-201tively organize all required information and knowl-202edge in the geo-DBMS.This data model should ful-203fill the following requirements:(1)support repre-204sentation of the environment,particularly the net-205work elements and the network topology;(2)sup-206port dynamic simulation,such as the representa-207tions of disaster developments in time,changes in 208the availability of roads,and the movements of re-209lief vehicles;(3)support various analyses,includ-210ing identifying the areas that are most threatened, 211planning paths in the context of moving obstacles, 212etc.;(4)support representation of the calculated re-213sults,e.g.,the navigation route,estimated traveling 214and arrival time;and(5)should be compatible with 215the relevant data models for emergency response 216and existing standards defined by the Open Geospa-217tial Consortium(OGC)or International Standard 218Organization(ISO),e.g.,ISO19107:2003that pro-219vides a formal structure for representation of spatial 220objects.221Using the requirements listed above,we define 222a data model to capture dynamics of the envi-223ronment,using Unified Modeling Language(UML) 224profiles for database design.The proposed model 225is designed adhering to the data model presented 226by Dilo and Zlatanova(2011)as much as possible, 227and is built for the following3groups of data:(1) 228data related to the road network;(2)data relevant 229to disasters;and(3)data on response units.We 230define the topology of the network by ourselves, 231and use the geometric data types specified by ISO 232319107,e.g.,GM Point,GM LineString,GM Polygon, 233and GM MultiSurface,to describe the spatial char-234acteristics of geographic features.Because the data 235we are handling are constantly changing,new data 236types are created to capture this spatio-temporal 237nature.2383.1.Conceptual data model239Figure2is a UML class diagram presenting a 240conceptual model of the data required for naviga-241tion among moving obstacles.The yellow classes 242are created for handling the data related to dis-243asters.The green classes are used to support the 244representation of the road network.The classes in 245light-gray are defined for modeling the data of re-246sponse units.New datatypes are colored in purple. 247The class RoadNetwork is an extended graph,con-248sisting of instances of RoadSegment that contain 249dynamic information produced by disaster events. 250To maintain the topology of the road network,an 251association between RoadSegment and RoadJunc-252tion is established.Both RoadSegment and Road-253Junction have an attribute affected time list used to 254store temporal information regarding the availabil-255ities of the corresponding spatial objects.A new 256data type called AffectedTimePeriod is created for 257these two classes containing the attribute of a dy-258namic nature.A RealIncident is used to record the 259information of the disaster incident.It inherits all 260properties of the abstract class Incident which con-261tains static information of the incident including 262incidentID identifying the incident,the location of 263the incident,the start time,and a text descrip-264tion of the incident.Some additional attributes 265are added to store the dynamic information gener-266ated during the incident,such as the disaster type 267which may change in time,GRIPlevel describing the 268changing severity of the incident,and affected area 269which stores the historic information of affected ar-270eas during the incident.The class SimulatedEvent is 271linked with RealIncident to describe disaster simula-272tions that predict the effect of real incidents within 273a certain period of time.The class Obstacle con-274tains predicted information about the obstacles in 275the form of moving polygons affecting the road net-276work.As soon as a real incident occurs,different 277types of Processes are started.Several teams that 278are sent to address the incident are responsible for 279managing these processes.A team may be com-280posed of one or more vehicles.The class Vehicle 281contains information related to vehicles.The as-282sociation Follow is used to record the routes that 283drivers want to follow.These Routes are calculated 284based on spatio-temporal information in the geo-285DBMS and proposed to the drivers.The stored 286route information will also be used for monitoring 287movement of vehicles during disasters and analysed 288after disaster response.2893.2.Logical data model290The proposed data model has been realized in 291the relational database PostGIS(). 292PostGIS spatial data types and functions are com-293pliant with OGC specifications and ISO19107.Fig-294ure3shows the logical data model for PostGIS. 295Following classical approaches(G¨u ting et al.,2000; 296G¨u ting and Schneider,2005),we create some new 297data types to store the spatio-temporal data,i.e., 298MovingPointInst to store dynamic positions of both 299vehicles and teams;MovingPolygonInst to record 300historic affected regions and identify dangerous ar-301eas in the near future.These data types are de-302fined by adding timestamps as one of attributes to 303capture the temporal aspect.We use the ARRAY 304type,in which the new data types are used as a base 305type of the array elements,to record facts associ-306ated with time.For example,MovingPolygonInst[ 307]is composed of a sequence of pairs of polygons 308and time instances.To represent many-to-many 309associations,an intersection table is created.For 310instance,a table,RoadSegment to Route,is intro-311duced to hold the many-to-many relationship be-312tween RoadSegment and Route,combining the pri-313mary keys from the original tables.The logical 314schema is automatically transformed by a modelling 315tool Enterprise Architect() 316to a collection of Structured Query Language(SQL) 317scripts for creating and dropping tables.These cre-318ated tables are populated with spatial and spatio-319temporal data that are used for analysis and visu-320alization by our navigation application as well as 321traditional GIS tools.322work analysis application considering 323the spread of thefire324In this study,we design and develop a prototype 325network analysis application for forestfire rescue 326planning.The application supports both data pro-327cessing and data analysis,including fetching thefire 328simulation results,formatting them into a general 329representation,calculating the availability of road 330segments,and computing the shortest path while 3314Figure2:Conceptual data model(UML class diagram with ISO19107geometric data types)avoiding predicted inaccessible roads infire-affected 332areas.The shortest path algorithm is extended to 333consider both static information,i.e.,the topologi-334cal and spatial constraints of the network,and dy-335namic information,i.e.,the predicted accessibility 336of roads.3374.1.Intersection of thefire-affected area with the 338road network339For the network analysis application,a cell-based 340fire simulation model developed by Moreno et al. 341(2011)is used to generate datasets offire-affected 342areas.Thefire simulation method divides the to-343pography into a grid of square cells.Each cell con-344tains both static information,such as position,size 345(i.e.,3meters),type,and the burning rate depend-346ing on its type,and the runtime information,such 3475Figure3:Logical data model(UML class diagram with PostGIS geometric data types,note that the ARRAY is used and indicated by square brackets[]after the datatype of the attribute)6as the quantity of combustible,the power intensity 348of thefire,and the state of thefire.Thefire simula-349tion system,integrated with passive data from dif-350ferent sources and dynamic events,including real-351time changes in the weather conditions,calculates 352the spread of the forestfire and updates the run-353time information of forest cells calculated during 354each simulation step.By grouping the cells accord-355ing to the cell state and time step,we create a set 356of moving polygons that overlap a certain road net-357work.Considering that each cell in the simulation 358has a certain width,we introduce a new buffer for 359each road-center line to represent the road network, 360extract all the road segments and junctions inside 361affected areas,and store them with their affected 362time periods in the database according the data 363model described in section3.3644.2.Routing algorithm365Once the state of roads has been updated,the 366application fetches spatio-temporal data of the road 367network from the database and generates a graph 368with affected time of roads.Consider a graph 369G=(N,E)consisting of afinite set of edges E and 370nodes N.Each edge e∈E corresponds to an object 371of class RoadSegment,and each node n∈N corre-372sponds to an object of class RoadSegment.We use w 373to represent the length of each RoadSegment and use 374an interval[t closed,t open]to denote an element of af-375fected time list attached to the corresponding road 376segment and junction.[t closed,t open]is an instance 377of data type AffectedTimeperiod,where t closed is the 378start time of closing,and t open is the end time of 379closing.Here we assume that once the nodes and 380edges are affected by thefire,they will not be avail-381able anymore.Following the above assumption,ev-382ery affected edge and node has only one affected 383time interval,and the opening time,t open,is set 384to inf by default.To calculate routes avoiding ob-385stacles,a special algorithm is needed to handle the 386affected time of roads.387In our application,we have extended the A* 388methodology for shortest path planning among 389moving obstacles.Related research on navigation 390among moving obstacles have been greatly studied 391in the roboticfield.Phillips and Likhachev(2011) 392introduce the concept of safe intervals to compress 393search space and extends the A*algorithm to gen-394erate time-minimal paths in dynamic environments 395with moving obstacles.Similarly,Narayanan et al. 396(2012)use time intervals instead of timesteps and 397develops a variant of A*for anytime path planning 398Figure4:The modified A*algorithmin the presence of dynamic obstacles.However, 399their planners do not take constrains of the real 400road network into consideration and can be only ap-401plied to free space.Our path planner has some sim-402ilarities to the algorithms presented in Visser(2009) 403and Wang and Zlatanova(2013a)which also con-404sider predicted information of the road network and 405introduce waiting options to avoid moving obsta-406cles.Under the above assumptions,waiting would 407not be safe duringfires and the vehicles need to 408move as fast as possible.Therefore,we remove the 409waiting option in the algorithm and do not consider 410the information on the state of nodes.411A*is a well-known algorithm developed to solve 412the one-to-one shortest path problem(Hart et al., 4131968).The A*algorithm uses a heuristic func-414tion to estimate cost from each node to the des-415tination to guide path search.The cost associated 416with a node n is f(n)=g(n)+h(n),where g(n)is 417the actual cost of the path from the start to node 4187n,and h(n)is an estimated coast from node n to 419the destination.The algorithm maintains two sets: 420openSet that stores nodes who are not expanded 421,and closedSet that stores nodes who have been 422expanded.At each iteration,the algorithm selects 423node m with the minimal cost from the openSet 424for expansion.All successors of node m that are 425unexplored will be put in the openSet for further 426expansion.427In our extension of the A*,we take into account 428the affected time of roads and introduce an addi-429tional parameter for the algorithm,the speed of 430vehicles moveRate,to select nodes for expansions. 431The value of moveRate can be obtained in two 432ways:(1)user configuration;(2)real-time calcu-433lation based on the location of vehicles recorded 434in the database.A new parameter departureT ime 435is added to help estimation of arrival time of each 436node.Figure4shows the main structure of the 437modified A*.When a node n is expanded,we com-438pute the estimated arrival time considering the cost 439of the edge w nn and the given speed,moveRate 440(see line15).At line18,we use a condition to de-441cide if the successor n of n should be added to the 442openSet.If the object can safely pass through the 443edge between the expanded node n and the succes-444sor n ,i.e.,the estimated arrival time is earlier than 445the closed time of the edge t closednn ,the successor n446will be added into the openSet for further expan-447sions.If not,it remains un-explored.The same 448condition is also applied on line22,which guaran-449tees that the evaluated node n should be updated 450not only with the faster arrival time but also with 451the safety of passing through the edge nn .4524.3.Theoretical analysis453Here we sketch the proof of the optimality of the 454path calculated by our algorithm.455Theorem1When the modified A*selects the goal 456for expansion,it has found a time-minimal and safe 457path to the goal node d.458Proof Were this not the case,the optimal path, 459P,must have a node n that is not yet expanded 460(If the optimal path has been completely expanded, 461the goal would have been reached along the optimal 462path.).There are then the following two possibil-463ities resulting in the fact that n is not expanded 464to generate successors:(1)f(n)>f(d);(2)all 465successors of n cannot be safely reached,i.e.the 466estimated arrival time is after the closing time of 467the edge between n and its successor.Because f 468is non-decreasing along any path,n would have a 469lower f-cost than d and would have been selected 470first for expansion before the goal node,which con-471tradicts thefirst possibility.We assume n is the 472successor of n along the optimal path,implying that 473g(n)+w nn <t closednn,which eliminates the second 474possibility.In the algorithm,the cost on an edge is 475equal to the time it takes to execute that edge,and 476whenever a g-value is updated(a shorter path is 477found),the time value is also updated to the earlier 478time.Therefore,when the node d is expanded,it 479is the earliest time we can arrive at the goal node. 480This is optimal in terms of time cost.We also know 481that all explored nodes are safely reached,which 482makes the entire path safe,from the start node to 483the goal node.4845.Route safety485To evaluate the safety of the route,we provide a 486method to quantify the safety value of edges and 487routes.Our method is similar to the one pro-488posed by Shastri(2006)that introduces the mar-489gin of safety of nodes,but uses the affected time of 490edges to evaluate the safety of routes.The safety 491of each edge is expressed as difference between the 492time whenfires block the edge and the estimated 493time when the responder arrive at the target node 494of the edge.Mathematically,the safety of an edge 495n i n i+1,S nin i+1,is496S nin i+1=t closedn i n i+1−t ni+1(1)Here t closedn i n i+1is the closed time of edge n i n i+1;t ni+1 497is the estimated time of reaching node n i+1though 498edge n i n i+1.499Because the safety of a route mainly depends on the most unsafe edge along the route,the minimum of safety values of edges is selected as the route safety.Let R={n0,n1,...,n k}be one of routes from s to t,where n0,n1,...,n k are the nodes along the route,n0=s,n k=d.The safety of the en-tire route can be computed by using the following formula(Shastri,2006):S R=min(S nn1,S n1n2,...,S nk−1n k)(2)If S R>0,the route is considered safe;If S R<=0, 500the route is considered not safe.The higher the 501safety value,the more safe the route is.+∞means 502the route is completely safe.5038。