Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Network Based Kernel Density Estimation for Cycling Facilities Optimal Location Applied to Ljubljana

0 views

Published on

ICCSA 2011 Santander, Spain, June

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

Network Based Kernel Density Estimation for Cycling Facilities Optimal Location Applied to Ljubljana

  1. 1. Network Based Kernel Density Estimation for Cycling Facilities Optimal Location Applied to Ljubljana<br />Nicolas Lachance-Bernard1, Timothée Produit1, BibaTominc2, Matej Nikšič2, Barbara Goličnik21 Geographic Information SystemsLaboratory, Ecole polytechnique fédérale de Lausanne2Urban Planning Institute of the Republic of Slovenia, Ljubjlana<br />The International Conference on Computational Science and its Applications – Cities, Technologies and Planning,June 2011, Santander, Spain<br />
  2. 2. Plan<br />Introduction<br />Conceptual background<br />Methodology<br />Ljubljana case study<br />
  3. 3. Plan<br />Introduction<br />Cycling and Urban Planning<br />Challenges and Needs for Optimal Location of Cycling Facilities<br />Conceptual background<br />Methodology<br />Ljubljana case study<br />
  4. 4. Cycling and Urban Planning<br />Cycling?<br />Promoted as one of the most appropriate ways of urban mobility<br />Environmentally friendly, require less space, impacts on health<br />Planning?<br />Importance of cycling facilities provision for cycling development<br />Germany: 12,911km (1976)  31,236km (1996)<br />The Netherlands: 9,282km (1978)  18,948km (1996)<br />Statedpreferencesurveys: Facilitiesdiscontinuities, route attributes<br />Goal?<br />Cyclingfacilities: Right places (O-D), right corridors (Flux)<br />
  5. 5. Optimal Location of CyclingFacilities<br />Opportunities<br />GPS: Portable, lightweight, unobtrusive and low-cost<br />Planners: Insights of current and future behaviors (monitoring)<br />Paststudies<br />Aultman et al. 1997 – Bicycle commuter routes and GIS<br />Dill and Gliebe 2003 – Bicycle and facilities in USA<br />Jensen et al. 2010 – Speed and paths of shared bicycle in Lyon<br />Menghini et al. 2010 – Route choice of cyclists in Zurich<br />Winters et al. 2011 – Motivators and deterrents of bicycling<br />
  6. 6. Optimal Location of CyclingFacilities<br />Challenges and Needs<br />GPS tracking visual presentation: data volume<br />Direct usage of GPS data in the planning practice: lack of methods<br />GVI: free enriched geographic data sources (i.e. OSM)<br />
  7. 7. Plan<br />Introduction<br />Conceptual background<br />Examples of Current GPS Tracking Projects<br />Ljubljana Investigation Background<br />Kerned Density Estimation (KDE)<br />Network Based Kernel Density Estimation (NetKDE)<br />Methodology<br />Ljubljana case study<br />
  8. 8. Examples of Current GPS TrackingProjects<br />San Francisco (USA) – Smart phones<br />Weeklyprizedraw<br />“Developing”facilitiesinstead of “building”them<br />Copenhagen (Denmark) – Web-based GIS portal<br />3,000 trips mappedby citizen VISUM model<br />COWI A/S GPS tracking: before / afterfacilitiesimprovements<br />Barcelona (Spain) – Qualitative / Quantitative<br />Bici_Nprojectrent-a-cycles video/audio<br />Data transfert from station to central DB for furtheranalysis<br />
  9. 9. Ljubljana Investigation Background<br />Statedpreferences (2008)<br />Web-based portal Geae+<br />Cyclist description, trip information<br />Digitalization of trip <br />GPS track transfert fromenableddevice<br />Low-Tech: Paperover mapdrawing<br />Revealedpreferences(2010)<br />GPS trackingdevice<br />User friendly, low-cost, accurate<br />Data transfert by technicians<br />Broader investigation<br />
  10. 10. KDE vs. NetKDE<br />Kernel Density Estimator (KDE*)<br />Operates in euclidean space<br />Weights events by their radial distances from grid centroid<br />Network Based Kernel Density Estimator (NetKDE*)<br />Operates in a network constrained space<br />Weights events by the distance from grid centroid measured along this network<br />* Density estimation function + Epanechnikow kernel function <br />NetKDE and KDE (2009-2011) by TimothéeProduit, Nicolas Lachance-Bernard, Loic Gasser, Dr. StephaneJoost, Prof. Sergio Porta, EmanueleStrano<br />
  11. 11. KDE vs. NetKDE<br />KDE<br />NetKDE<br />
  12. 12. KDE vs. NetKDE<br />KDE<br />NetKDE<br />
  13. 13. Plan<br />Introduction<br />Conceptual background<br />Methodology<br />GPS Tracking<br />Network and Grids<br />Low Resolution KDE, High Resolution NetKDE<br />Ljubljana case study<br />
  14. 14. GPS Tracking<br />Device<br />Sport tracker QSTARZ BT-Q1300s<br />62 x 38 x 7 mm, 10m accuracy<br />One button (On/Off), mini USB port<br />KML, GPX, CVS<br />Tracking: 5 seconds, 15h autonomy<br />Data <br />CSV  SHP (WGS84)  Merge  Projection (UTM33N) [Manifold]<br />
  15. 15. Network and Grids<br />Open Street Map Network<br />Source: Cloudmadewebsite<br />SHP (WGS84)  10km GPS Buffer  Projection (UTM33N)  Places digitalization  Highwaydeleted[Manifold]<br />Topology (0.5m connecting/merging) + attributescleaning[ESRI ArcGIS model builder]<br />Grids<br />100m: Lowresolution multi-bandwidths KDE<br />20m: High resolutionspecific-bandwidthsNetKDE[IDRISI]<br />
  16. 16. Plan<br />Introduction<br />Conceptual background<br />Methodology<br />Ljubljana case study<br />Resources, Data and Calculations<br />Low Resolution Grid KDE Results<br />High Resolution Grid NetKDE Results<br />Discussion<br />
  17. 17. Resources<br />Software / Hardware<br />Postgres/PostGIS/Python/QuantumGIS<br />Windows XP 64<br />Intel® Core™2 Quad CPU Q950 @ 3.GHz 7.83GB of RAM<br />
  18. 18. Data and Calculations<br />Lowresolution KDE 100m  425km213,630 segments, 42,342 gridpoints, 442,260 GPS points<br />KDE bandwidths [200m, 2500m] 24 X 100m steps(2-3h) <br />High NetKDE/KDE 20m  20km28,114 segments, 314,250 gridpoints, 423,748 GPS points<br />NetKDEbandwidths 60m (17h), 100m (19h), 200m (24h), 400m (27h)<br />KDE bandwidths [40m, 100m] 7 X 10m steps [200m, 1000m] 9 X 100m steps (total 18h)<br />
  19. 19. KDE results100m grid<br />Bandwidths:<br /> A)300m B)500m C)1000m<br /> D)2000m<br />*Decilesdistribution<br />
  20. 20. KDE results20m grid<br />Bandwidths:<br /> A)60m B)100m C)200m<br /> D)400m<br />*Decilesdistribution<br />
  21. 21. NetKDEresults20m grid<br />Bandwidths:<br /> A)60m B)100m C)200m<br /> D)400m<br />*Decilesdistribution<br />
  22. 22. Discussion<br />NetKDE 20m (Visual analytics)<br />3:1 ratio - Shows flux corridors (a)<br />5:1 ratio - Smooths corridors only (b)<br />10:1 ratio - Highlights axis and intersections (c)<br />20:1 ratio - Shows cyclist’s main area presence and main axis<br />
  23. 23. Discussion<br />Researchunderrapidevolution…<br />3rd algorithm: Calculationoptimization 90-95% (10h network-indexing, 5 min. for eachsteps)<br />Currentwork on Barcelona, Ljubljana, Geneva, Glasgow, Baghdad<br />Professional uses: Architects, Planners, Criminologs, Biologists<br />Actualprojects…<br />Spatio-temporal and statisticalanalysis<br />Fuzzy-mapcomparison (time, model, resolution, bandwidth)<br />TestingAdaptedLandscapemetrics<br />TestingHPC for calculation and subsequentanalysis<br />Prototyping the integration of NetKDE, KDE, MCA, … into SDI<br />
  24. 24. Thankyou!<br />

×