公共公園游憩模式的創新研究.pdf

1、Innovative Modeling of Recreation Visits to Public Parklands Session C2:Big Data for Travel AnalysisPresented by:Mark BradleyJune 5,2023Innovative Modeling of Recreation Trips to Public ParklandsPresentation TopicsGLACIER NATIONAL PARKMONTANA2Project Motivation3 Project Challenges4Model Data5-6Model

2、s Described7-13Model Application14Reflections 15Project MotivationIn some areas of the United States,visits to national,state,and regional parklands generate a great deal of local traffic and vehicle miles traveled(VMT).While some regional and statewide travel forecasting models include a“recreation

3、”purpose,travel for outdoor recreation to and in major parklands has unique characteristics that are not included in existing models.This project provides a new model framework and initial models for potential use by state and regional planning agencies(as well as park planning agencies).Project Cha

4、llengesDifferent Variable NeedsThe type of variables in recreational travel models are different from traditional long-distance models.The variables are about the type of land characteristics:elevation,climate,coastal(or not),and park-specific attributes.Key ChallengesThe types of people attracted d

5、iffer from urban recreation tripsa range of incomes due to the ability to camp,and a range of group types(e.g.,families,retired,digital nomads)A range of trip typesday trips or overnight trips from nearby residents,and longer vacation trips from across the country(and beyond)Creating a generic/gener

6、alized model that can be applied to any major parkland destination in the country4Project Modeling:Data SourcesSome data exist in the form of visitor surveys and counts,but no single source of data exists that is comprehensive and consistent across various types of outdoor destinations.Passively col

7、lected“big data”sources can be used to identify some characteristics of individual visits(e.g.,O/Ds,trip duration,timing),but they do not identify characteristics of the visitors(e.g.,age,income,household type).This was our first attempt at using disaggregate location-based services(LBS)data for mod

8、el estimation(rather than for descriptive studies or calibration/validation of existing models).Project Modeling:Data ChallengesThe quality and completeness of“big data”can be questionable,particularly in parkland areas and trips where people are not using their smartphone apps as often as usual(and

9、 where cell service may be sparse).No clear direction of causality exists between the supply of parkland amenities(e.g.,camping,lodging,trails,food,toilets,parking)and the demand for those amenities(number of visits).Few park sites have the richness of data on visitation and amenities that is availa

10、ble for National Parks and Monuments.Project Model StructureFor a given parkland within a season7Accessibility MeasuresSpatial Distribution of VisitsAuxiliary ModelsDuration of stay:number of nights in and out of the parkArrival&departure time periodsVMT within the park&within the“halo area”(50-mile

11、 radius)Park Visit Generation ModelAccess Mode&Airport ModelVisitor Home Location ModelProject Modeling ApproachPrimary Data SourcesRelied on passively collected“big data”as the main data source for most of the component modelscompiled data for all visits to a variety of selected parklands during 20

12、19(wanted to use pre-COVID data).Used visitation data by park by month from NPS as main data source for the Park Visit Generation Mode.Critical Supporting Data SourcesCensus(ACS)data of sociodemographic characteristics by tractEmployment data by tractClimate data(average temperature and precipitatio

13、n by season)Data on elevation,extent of water coverage,fraction public protected land,coastline,etc.Data on park amenities(e.g.,lodging/camping,trails,beaches)Data from park visitor surveys(for model validation)8Most supporting data sources were created for Census tracts or block groups,and then agg

14、regated to a national zone system with almost 4,900 zones.(Zones are the intersection of PUMAs and counties,the zone system from RSGs FHWA National Long-Distance Passenger Model)Visitor Home Location ModelPassively Collected DataNational LBS data for all of 2019(processed by quarter)Inferred travel

15、behavior for all devices seen in study parks89 national parks and monuments 47 CA state parks29 PA state parksConstructed park-visit data structureVisitor Home Location ModelPassively Collected Data3.5 million park visits observedEach park visit includes inferred attributes associated with the visit

16、or,their travel to the park,and the quality of dataHighest volume during July-Sept.Lowest volume during Oct.-Dec.About 50%of visitors live in“halo area”(within 50 miles of park)After screening for data quality(density of observations in time)and stopping for at least 2 hours within the park,1.5 mill

17、ion park visits used for modeling INDEX COLSDEVICE METRICSVISIT METRICSQUALITY METRICSdevice_idhome_zoneparknum_park_sightingstour_idhome_statelocal_time_depart_home num_non_visit_sightingspark_residentgmt_depart_homenum_geo6park_employee gmt_arrive_homeunique_dayshalo_visitsavg_miles_per_device_day

18、park_vmtvisit_data_densityhalo_vmtarrival_data_densityother_vmtgmt_arrive_halolocal_time_depart_haloin_halo_nightsnum_halo_sightingsdow_depart_halolocal_time_arrive_parklocal_time_depart_parkgmt_depart_parkdow_arrive_parknon_visit_nightsarrival_airportarrival_airport_gmtreturn_airport_gmt10Visitor H

19、ome Location ModelFindingsA logit discrete choice model predicts the home location from among the 4,900 zonesLarger,older national parks attract visitors from greater distancesZones with higher incomes generate more trips.Less influence from other socio-demographic variables.O/D patterns are affecte

20、d by differences in climate and topography between the park zone and the home zone,with different patterns by seasonAccess Mode and Airport ModelAirport ModelTo model trips to the park destination area,it was important to account for cases where the visitor flies to an airport and goes by road from

21、the airport to the park,rather than traveling by road all the way from home to the parkland destination.In application,this model is also used to simulate access airport choice for international visitors,who comprise approximately 5%of visitors across all NPS sitesmuch higher at the most well-known

22、national parks.The same passively collected data that were used to estimate the Visitor Home Location Model was used to estimate this model The choice set of airports used to access each parkland destination was determined from the airport visits observed in the passively collected data The airport

23、choice set for each park was limited to a maximum of 26 possible airports at up to 1,000 miles from the park;this size of choice set included over 99%of the observed airports used in the dataset 12Park Visit Generation ModelA regression model to predict the number of visits to a specific destination

24、 within a specific seasonUsed published visitation data instead of LBS data Uses the expected utilities(logsums)from the Visitor Home Location and Air Mode&Airport models as measures of accessibility to visitorsA major challenge:specifying explanatory variables that are not endogenous;number of faci

25、lities(e.g.,camping spaces,parking spaces,toilets)are highly endogenousModel ApplicationThe auxiliary models to predict timing,duration of stay,and local trips are simpler models to fill in detail on how and when visits to parkland destinations generate traffic in the surrounding area.The model syst

26、em is coded to be run as a stand-alone module or launched from a network software package.Some validation was done against visitor survey data(available for only five of the modeled sites).The project documentation includes a guidebook for using the model code,possibly as“special generator”models in

27、 combination with regional or statewide model systems.14Accessibility MeasuresSpatial Distribution of VisitsAuxiliary ModelsDuration of stay:number of nights in and out of the parkArrival&departure time periodsVMT within the park&within the“halo area”(50-mile radius)Park Visit Generation ModelAccess

28、 Mode&Airport ModelVisitor Home Location ModelThe LBS data were most successful for the long-distance home location&air mode/airport choice modelsLarge number of observations gave good coverage of the zones and seasonsFor longer-distance trips,gaps in device coverage were not a serious issueRequired

29、 using aggregate zonal sociodemographicsCould be merged with visitor survey data if such data are more widely availableThe LBS data were moderately successful for capturing travel timing&local trip patterns&VMTGaps in the data coverage can cause missed trips,inaccurate arrival&departure timesVehicle

30、-based big data may better capture these local trips than LBS data canThe LBS data were not suitable for modeling park visitation levels Evidence of substantial variation in coverage by site and seasonEven published visitation data were difficult to model accuratelyA longitudinal model using several years of“published”visitation and amenities data may be most usefulReflections on Using Big DataCMARK BRADLEYPRINCIPALMark.B