How to make the most of open data? A travel demand and supply model for regional bicycle paths

Full text

(1)DEGREE PROJECT IN SYSTEMS ANALYSIS AND ECONOMICS, SECOND CYCLE, 30 CREDITS STOCKHOLM, SWEDEN 2021. Where should new bicycle paths be built? Design and test of a demand model based on open data, evaluation of the actual infrastructure. Laurent Cazor. KTH ROYAL INSTITUTE OF TECHNOLOGY ARCHITECTURE AND THE BUILT ENVIRONMENT.

(2)

(3) Abstract This Master Thesis main objective is to answer a problem set by the Swedish Transport Administration: a common regional bicycle planning process would them cheaper and more comparable. They currently offer the planners a model developed by Kågeson in 2007. This model takes the form of a report which advises on when to build a bicycle path between cities or places of a region. Still, it is only used in only 6 of the 21 Swedish counties. Trafikverket requires a new planning support tool, more interactive and complete than the Kågeson model. Some new desired features are the separation of demand per purpose, the inclusion of e-bikes, different trip purposes, and a prioritization of the investments. The Degree Project work is to design and implement this tool, also called Planning Support System (PSS), which compares supply and demand for bicycle path to prioritizing infrastructure improvements. A main constraint for the model is that it needs to be cheap data-wise, but as complete and precise as possible. It bases on several open data providers, such as OpenStreetMap, the Swedish National Road Database (NVDB), or Travel Surveys from Sweden and the Netherlands. The result is a model, disaggregated by trip purpose and type of bicycle. The demand estimation part adapts a classic four-step transportation model to bicycle planning and limited data. For different trip purposes, trips are generated and distributed thanks to an origin-constrained gravity model. Bicycle mode choice is fit to actual travel behaviour through logistic regression with a binary logit model. The trips are then assigned to the network using the "all-or-nothing" assignment method through the Dijkstra algorithm. To evaluate bicycle supply, we used a metric called Level of Traffic Stress (LTS), which estimates the potential use of a network link by different parts of the population as a function of the road network variables. The prioritization ranking is then the ratio between demand and supply metrics. This new tool is implemented with the opensource Geographic Information System (GIS) called QGIS and with Python 3, and it is tested on Södermanland County. Keywords: Planning Support System (PSS), Four-Step Model, Gravity Model, DistanceDecay function, Binary Logit Model, Level of Traffic Stress (LTS), Geographic Information System (GIS). i.

(4) Sammanfattning Detta examensarbete syftar till att svara på ett av Trafikverket fastställt problem: en gemensam regional cykelplanerings process skulle göra dem billigare och mer jämförbara. De erbjuder för närvarande planerarna en modell som utvecklades av Kågeson 2007. Denna modell har formen av en rapport som ger råd om när man ska bygga en cykelväg mellan städer eller platser i en region. Ändå används den bara i endast 6 av de 21 svenska länen. Trafikverket kräver ett nytt planeringsstödverktyg, mer interaktivt och komplett än Kågeson-modellen. Några nya önskade funktioner är separationen av efterfrågan per syfte, införandet av e-cyklar, olika resesyfte och en prioritering av investeringarna. Examensarbetet är att designa och implementera det här verktyget, även kallat Planning Support System (PSS), som syftar till att jämföra utbud och efterfrågan på cykelväg till prioritering av infrastrukturförbättringar. En huvudbegränsning för modellen är att den måste vara billig datavis, men så komplett och exakt som möjligt. Det baseras på flera öppna dataleverantörer, till exempel OpenStreetMap, den svenska nationella vägdatabasen (NVDB) eller reseundersökningar från Sverige och Nederländerna. Resultatet är en modell, uppdelad efter turändamål och typ av cykel. Del för efterfrågeuppskattning anpassar en klassisk fyrsteg transportmodell till cykelplanering och begränsad data. För olika resändamål genereras och distribueras resor tack vare en ursprungs begränsad gravitationsmodell. Valet av cykelläge är anpassat till det faktiska resebeteendet genom logistisk regression med en binär logit-modell. Resorna tilldelas sedan nätverket med tilldelnings metoden "allt-eller-ingenting" genom Dijkstras algoritm. För att utvärdera cykelförsörjningen använde vi ett mått som heter Level of Traffic Stress (LTS), som uppskattar den potentiella användningen av en nätverkslänk för olika delar av befolkningen som en funktion av vägnätvariablerna. Prioriteringsrankningen är då förhållandet mellan mått på efterfrågan och utbud. Detta nya verktyg implementeras med opensource Geographic Information System (GIS) som heter QGIS och med Python 3 och testas i Södermanlands län. Nyckelord: Planeringsstödsystem (PSS), fyrstegsmodell, gravitationsmodell, distansfunktion, binär logitmodell, Level of Traffic Stress (LTS), geografiskt informationssystem (GIS). ii.

(5) Acknowledgements This thesis was performed at Trivector Traffic, under the supervision of Erik Stigell. It was made in collaboration with KTH, Stockholm. I would like to thank Erik Stigell and Christian Dymén from Trivector, foremost for trusting me to work autonomously on this project, and for giving me access to data, interesting contacts, and good advice. I want also to thank Chengxi Liu from VTI, who took the time to help me with some of his data. I would also like to thank my supervisor at KTH, Christer Persson, for his help on some concepts and on the report, and my examinator Anders Karlström for examining this Thesis and helping me to schedule it. Finally, thank you to all the people with who I could share my work time and life in Sweden and France: my family and my friends from Marvejols, LLG, Centrale Lyon, and KTH.. iii.

(6) Contents. 1 Introduction 1.1 Background . . . . . . . . . . 1.1.1 The Kågeson model . . 1.1.2 Needed improvements 1.2 Scope of the degree project . . 1.3 Thesis outline . . . . . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. 2 Theoretical Background: Literature Review 2.1 Planning Support Systems . . . . . . . . . . . . . . . . . 2.2 Estimating and Forecasting Demand . . . . . . . . . . . 2.2.1 Classical multi-steps models . . . . . . . . . . . . 2.2.2 Relative Potential Demand models . . . . . . . . 2.2.3 Include combined commuting . . . . . . . . . . . 2.3 Gravity Models . . . . . . . . . . . . . . . . . . . . . . . 2.4 Mode choice modelling . . . . . . . . . . . . . . . . . . . 2.4.1 Multinomial logit models . . . . . . . . . . . . . . 2.4.2 The Independence of Irrelevant Alternatives (IIA) 2.4.3 Nested logit models . . . . . . . . . . . . . . . . . 2.5 Route assignment . . . . . . . . . . . . . . . . . . . . . . 2.6 Bicycle supply analysis . . . . . . . . . . . . . . . . . . . 2.6.1 Metrics to evaluate a facility . . . . . . . . . . . . 2.6.2 Level of Traffic Stress (LTS) . . . . . . . . . . . . 2.6.3 Total network analysis: connectivity . . . . . . . 2.7 Policy goals of cycling . . . . . . . . . . . . . . . . . . . 2.8 Inclusion of e-bikes in transport models . . . . . . . . . . 3 Model choice and design 3.1 Demand and Supply, is it a match? . . . . . 3.1.1 Demand analysis: A four-step model 3.1.2 Supply analysis . . . . . . . . . . . . 3.2 The different spatial subdivisions . . . . . . 3.3 Step 1: Trip generation and attraction . . . 3.4 Step 2 and 3: Trip distribution, modal split 3.4.1 Step 2: Trip distribution . . . . . . . 3.4.2 Step 3: Mode choice modelling . . . . 3.4.3 Final results . . . . . . . . . . . . . . 3.5 Step 4: Trip Assignment . . . . . . . . . . . iv. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . .. 1 2 2 5 5 6. . . . . . . . . . . . . . . . . .. 7 7 8 8 10 11 12 12 12 14 14 15 16 16 17 18 19 20. . . . . . . . . . .. 21 21 21 23 23 24 25 27 28 30 31.

(7) CONTENTS 3.6 3.7 3.8. Aggregation of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Supply analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Prioritization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34. 4 Data collection 4.1 Statistics and insights . . . . . . . . . . . . . . 4.1.1 Travel surveys . . . . . . . . . . . . . . . 4.2 Spatial data . . . . . . . . . . . . . . . . . . . . 4.2.1 Statistiska centralbyrån (SCB) data . . . 4.2.2 Trafikverket Lastkajen: Road and bicycle 4.2.3 OpenStreetMap: Trip attractors . . . . . 4.2.4 Naturvårdsverket: Natural areas . . . . . 4.3 On data costs and availability . . . . . . . . . . 5 Implementation and Results 5.1 Trip generation . . . . . . . . . . . . 5.2 Trip distribution . . . . . . . . . . . 5.3 Bicycle and e-Bicycle mode choice . . 5.4 Step 4: Assignment and aggregation . 5.5 Analysis of the supply . . . . . . . . 5.6 Prioritization . . . . . . . . . . . . . 6 Discussion 6.1 Model virtues . . 6.2 About simplifying 6.3 Data quality . . . 6.4 Further work . . 6.5 Conclusion . . . .. . . . . . . . . assumptions . . . . . . . . . . . . . . . . . . . . . . . .. . . . . .. . . . . .. . . . . .. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . network . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . .. A Mode choice: comparison between bike and e-bike. v. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . .. . . . . . . . .. 35 35 35 36 37 37 37 38 38. . . . . . .. 40 40 41 42 44 48 49. . . . . .. 52 52 52 53 54 56 63.

(8) List of Tables. 1.1. Decision Support Table for bulding a bicycle path . . . . . . . . . . . . . .. 2.1 2.2 2.3. Bicycle link evaluation indexes . . . . . . . . . . . . . . . . . . . . . . . . . 17 Level of Traffic Stress conditions for bicycle lanes . . . . . . . . . . . . . . 18 Bicycle network analysis methods . . . . . . . . . . . . . . . . . . . . . . . 19. 3.1 3.2 3.3 3.4. Trip purposes according to OViN . . . . . . . . . . . . . . . . . . . . . . . Corresponding trip purposes according to RVU . . . . . . . . . . . . . . . . Recommended infrastructure according to VGU, as a function of Speed limit and ADT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Level of Traffic Stress table as a function of infrastructure, ADT and Speed. 4.1. Demographic datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37. 5.1 5.2 5.3 5.4. Frequency of trips per purpose, according to the Swedish Travel Survey . . 41 Weights for trip generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Distance-decay function parameters . . . . . . . . . . . . . . . . . . . . . . 41 Coefficients of the different explanatory variables estimated from the logit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43. vi. 3. 29 29 32 33.

(9) List of Figures. 1.1 1.2. Cycle paths that should exist according to the model . . . . . . . . . . . . Actual cycle paths in the South of Sweden . . . . . . . . . . . . . . . . . .. 2.1 2.2. Work tour regional model in Sweden [Algers & Beser Hugosson, 2002] . . . 10 Illustration of a two-level nested logit model . . . . . . . . . . . . . . . . . 14. 3.1 3.2 3.3 3.4 3.5. Disaggregation of the models into sub-models . . . . . . . . . . SADT diagram of the model . . . . . . . . . . . . . . . . . . . . DeSO areas (yellow) and Rutor (blue) around Eskilstuna . . . . Nested logit model for non-commuting trips . . . . . . . . . . . Binary logit model for the purposes for which the trip distribution (Work and School commuting) . . . . . . . . . . . . . . . . . . . 3.6 VGU recommendations for infrastructure in Sörmland . . . . . .. . . . . . . . . . . . . . . . . . . . . is known . . . . . . . . . .. . . . .. 4 4. 22 23 24 25. . 26 . 33. 4.1. Bicycle infrastructure by type, NVDB . . . . . . . . . . . . . . . . . . . . . 38. 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 5.14 5.15 5.16. Distance-decay functions for the gravity model . . . . . . . . . . . . . . . . Regular bike choice probability in function of distance . . . . . . . . . . . . Electric bike choice probability in function of distance . . . . . . . . . . . . Work potential daily bicycle flow . . . . . . . . . . . . . . . . . . . . . . . Work potential daily e-bicycle flow . . . . . . . . . . . . . . . . . . . . . . School potential daily bicycle flow . . . . . . . . . . . . . . . . . . . . . . . Services potential daily bicycle flow . . . . . . . . . . . . . . . . . . . . . . Services potential daily e-bicycle flow . . . . . . . . . . . . . . . . . . . . . Shopping potential daily bicycle flow . . . . . . . . . . . . . . . . . . . . . Shopping potential daily e-bicycle flow . . . . . . . . . . . . . . . . . . . . Leisure potential daily bicycle flow . . . . . . . . . . . . . . . . . . . . . . Leisure potential daily e-bicycle flow . . . . . . . . . . . . . . . . . . . . . Touring potential daily bicycle flow . . . . . . . . . . . . . . . . . . . . . . Touring potential daily e-bicycle flow . . . . . . . . . . . . . . . . . . . . . Aggregated potential bicycle flow in Sörmland . . . . . . . . . . . . . . . . Required improvements according to VGU and potential cycle flows calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100km of each kind of improvements that have the highest bicycle flow . . Zoom on Eskilstuna area . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ratio Supply/Demand for Sörmland bicycle infrastructure . . . . . . . . . Zoom on Eskilstuna area . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 5.17 5.18 5.19 5.20. vii. 42 43 44 45 45 45 45 45 46 46 46 46 46 46 47 48 50 50 51 51.

(10) LIST OF FIGURES A.1 A.2 A.3 A.4 A.5 A.6. Mode Mode Mode Mode Mode Mode. choice, choice, choice, choice, choice, choice,. work commuting . school commuting work commuting . school commuting work commuting . school commuting. . . . . . .. . . . . . .. . . . . . .. viii. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. 63 63 63 63 64 64.

(11) Acronyms ADT Average Daily Traffic. vi, 18, 32, 37 BCI Bicycle Compatibility Index. 17 BI Bikeability Index. 19 BLOS Bicycle Level of Service. 17, 19 CBA Cost-Benefit Analysis. 19 CBS Centraal Bureau voor de Statistiek. 36 DeSO Demografiska statistikområden. 24, 25, 41 GIS Geographic Information System. 7, 17, 35 HCM Highway Capacity Manager. 17, 19 IIA Independence of Irrelevant Alternatives. iv, 14, 15 LDS Latent Demand Score. 10 LTS Level of Traffic Stress. 17–19, 34, 49 MNL Multinomial Logit. 12 OD Origin-Destination. 10, 12, 24, 27, 28, 30, 31, 37, 42, 55 OSM OpenStreetMap. 25, 35, 37 OViN Onderzoek Verplaatsingen in Nederland. vi, 29, 36 PBLOS Pedestrian and Bicyclist Level of Service. 17 PCA Principal Component Analysis. 20 PSS Planning Support System. 2, 7, 52, 56 RP Revealed Preference. 16 RTP Regional Transport Plans. 2 ix.

(12) Acronyms RVU Resvaneundersökning. vi, 21, 27–29, 31, 35, 36, 41, 53, 54 SADT Structured analysis and design technique. vii, 23 SCB Statistiska centralbyrån. v, 19, 23, 35–38 SDSS Spatial Decision Support System. 2 VGU Vägar och gators utformning. 3, 18, 23, 32, 34, 48 VTI Väg och transportforskningsinstitutet. 35. x.

(13) SECTION 1. Introduction « Bicycles let people move with greater speed without taking up significant amounts of scarce space, energy, or time. They can spend fewer hours on each mile and still travel more miles in a year. They can get the benefit of technological breakthroughs without putting undue claims on the schedules, energy, or space of others. They become masters of their own movements without blocking those of their fellows. Their new tool creates only those demands which it can also satisfy. Every increase in motorized speed creates new demands on space and time. The use of the bicycle is self-limiting. It allows people to create a new relationship between their life-space and their life-time, between their territory and the pulse of their being, without destroying their inherited balance. » – Ivan Illich, Energy and equity, [Illich, 1974] Sustainable development is an issue that has been identified in the second half of the twentieth century. The Limits to Growth, a report for the Club of Rome in 1972 [Meadows et al., 1972], is one of the first breakthrough. It states that the seventies growth levels would not be possible for more than a century, and give advice to take actions to stabilize demographic, economic growth, and to share wealth better. Recently, climate change and recent crises have fostered debates on sustainable development. Since 2015, international meetings have set roadmaps to reach several goals. That year, Paris agreements, set the goal of limiting global warming to 1.5 to 2 °C until 2100, and the UNO members agreed on seventeen Sustainable Development Goals in the Agenda 2030 [Rosa, 2017]. These goals include erasing poverty, hunger, inequalities between genders and communities, while enhancing health, education, economic growth and climate action. Bicycle is a sustainable and resilient means of transport that only requires human power. According to the World Cycling Alliance and the European Cyclists Federation, it serves eleven of the seventeen Agenda 2030’s goals [ECF, 2016]. Indeed, bicycle is cheap, and strengthens equity in accessibility; it leads to physical activity and contributes to better health. It requires less infrastructure, and does not emit greenhouse gas. It is also likely to reduce gender inequalities. Increased cycling has been identified by countries as a lever of sustainability. Sweden, for instance, set the goal of doubling the proportion of bicycle kilometres traveled by 2025 [Eneroth, 2018], increasing the modal share from 12% to 25 %. This target requires an increased pace in political action, as states the Swedish bicycle organization, Cykelfrämjandet, taking the shape of bicycle plans. Bicycle plans are driven by communities (e.g. counties), and usually include improvements in infrastructure, communication, education and services around cycling. Being a major investment, improving transportation infrastructure requires several stages of planning. Planning goals are to allocate the infrastructure where it will serve the 1.

(14) 1. INTRODUCTION most society, accounting for the goals of sustainability. To do that, information on the potential new users of an improvement is needed. Gathering data on the actual use and future potential use of the transport system, moreover at a regional scale, may be long and expensive. The main solution to this issue are models. Indeed, with sufficient data and several assumptions, model can forecast the impacts and benefits of a change in facilities, and help communities to plan efficiently their works. Planning Support System (PSS), or Spatial Decision Support System (SDSS) give a good framework to plan efficiently infrastructure. After decades of car-oriented planning, bicycle is now receiving more attention. Mostly thriving in urban areas, where short distance between activities fosters its use, it is also part of Regional Transport Plans (RTP), moreover since the electric bike breakthrough increased bicycle’s distance coverage. Urban models are often very complete, trying to include a competition between all modes and tracking individual movements and choices. Bicycle path planning in a regional space needs different planning methods for urban area planning. Indeed, it would be too costly to apply the same methods on a way wider area. The first stages of bicycle planning require to identify the most appealing locations, by forecasting the potential mismatch between bicycle demand and infrastructure. More precise studies are then carried out at a local scale. Until now, Swedish regions have mostly been developing cycle plans on their own, and my project’s goal is to develop a common methodology, that should be integrated into all Swedish planners’ processes. This degree project has been performed in the System Analysis and Economics division at KTH, Stockholm. The supervisor at KTH is Christer Persson, and the examiner is Anders Karlström. It has been carried out with the consultancy Trivector, and Erik Stigell as supervisor. The project to which this work answer is sponsored by Trafikverket, the Swedish Transport Administration.. 1.1. Background. To modernize their planning process, Trafikverket needs a common tool that all regional planners would use, as it would reduce the planning costs and would output more comparable results for budget allocation. It is doomed to replace a model created by Per Kågeson in 2007, which gives insights on where to build a separate bicycle path but without a prioritization methodology.. 1.1.1. The Kågeson model. The actual model that can serve as a common basis for Swedish transport planners has been developed by Per Kågeson in 2007 [Kågeson, 2007]. It takes the shape of a report, in which Kågeson advises on the requirements needed to build future bicycle infrastructure, considering the 2007 Swedish situation. Here is a summary of the report content: • It reports existing main lanes of car-free cycle path with an indication of road width, coating type, and the possible presence of lighting • It determines which additions are needed along state and municipal roads within an urban area to reach the neighbouring urban areas 2.

(15) 1. INTRODUCTION • It indicates which connections between urban areas should be created, including across municipal and county boundaries that meet certain conditions • It gives extra conditions for borderline cases in terms of population, because for instance of cost-reducing synergies • It takes a position on which bathing places, outdoor areas, etc. are located close enough to an urban area have to be provided with a cycle path • It analyzes the potential for recreation trips and cycling tourism, that can deserve the connection of links • It studies the need for safety-enhancing measures at intersections between existing and planned cycle paths and roads with motorized traffic exceeding 1,000 vehicles per annual average day. The model uses the Vägar och gators utformning (VGU) 2004 framework, to give standard costs to investments. VGU is a set of rules for road and street design, which are updated every year. There is a sub-part giving a different kind of bicycle path that can be built along state roads. The model goal is to make an inventory of every link between towns - or between a town and a recreational area - where a separated bicycle path should be created. The only variables considered by the model are distance and population of two areas. Kågeson justifies his model does not advise on a ranking methodology because of the amount of time the sponsor company, Vägverket, had given him to complete the project.. Population in the smaller town. Maximum distance between town boundaries. 500 - 1000 1000 - 2000 2000 - 5000 5000 - 10000 10000 - 20000 > 20000. 3 6 10 12 15 20. Table 1.1: Decision Support Table for bulding a bicycle path The model is then dealing with different exceptions to this table. For instance, if building costs can be reduced, it might be interesting to consider longer distances, for cases where the population in the smallest town is close to the boundary. It also says that constructions along 2+1 roads should be prioritized. It advises also to consider the population living between two cities within the road network, adding the number of inhabitants to the smaller town one. This model has some weaknesses. First, the few amounts of criteria it considers is leading to a lot of exceptions, which are impossible to compute and require a manual check by planners. In his report introduction, Kågeson uses as a justification of the distances he uses that a cyclist can be expected to cycle around 30 minutes. The way it is linked in 3.

(16) 1. INTRODUCTION this table could be a distribution of travel times and the percentage of potential cyclists, but it is nowhere explained in the report. It has resulted that the model is rarely used by planners. In an internal study made by Trivector, discussing with the different regional planners in Sweden and reading their bicycle plans or strategies, it has been highlighted that Kågeson’s model was used by only a few regions. The criteria used by regions to rank their projects are then different from a region to another. The following figure depicts the weaknesses of the Kågeson model: only a minor part of the bicycle paths that should have been built have actually been built.. Figure 1.1: Cycle paths that should exist according to the model. Figure 1.2: Actual cycle paths in the South of Sweden. Moreover, in its socio-economic calculations advises, the Kågeson model does not 4.

(17) 1. INTRODUCTION account for the potential social benefits, benefits for health, and does not account for equitable planning in the inventory process.. 1.1.2. Needed improvements. This subsection reflects Trafikverket project specifications, and have been discussed with Trivector project members. According to the specifications, a new model should firstly define more clearly the notion of time budget, which is vague in the Kågeson model. It should also differentiate the potential time and distance budgets, as a function of the following variables: 1. Travel goal (leisure, commuting, shopping, travel per se or tourism) 2. Socio-demographic variables (age, sex, health capital, immigration) 3. Kind of bicycle (regular or electrical) With these time budgets, the model needs to compute potential demand or relative demand for the different network links. It needs to identify areas where bicycles could contribute to increased sustainability, for instance where car ownership is weak, or where health issues have been identified. It needs also to consider the potential for combined trips (bicycle + public transport). Then, the model needs to match the demand to the different kinds of infrastructure, which can be fast cycle paths, narrow roads, summer cycle paths, roadside cycling (village environmental road), or in cycling in mixed traffic with adapted speed, following Trafikverket’s four-step principle: [Trafikverket, n.d.]: 1. Think about: Consider measures that can affect demand and mode choice, without modifying infrastructure. For instance, here, reducing the speed limit on roads. 2. Optimize: Measures to use more efficiently the actual infrastructure. For instance, creating painted cycle paths. 3. Rebuild: Carry out limited rebuilds. For instance, widening roads or creating separate cycle paths. 4. Build new: Carry out major investments. For instance, separated bicycle highways. By Trafikverket’s interpretation of their four-step principle, building a bicycle path cannot be considered a major investment. The fourth principle above is therefore not relevant in this project. Another project goal, which has been identified later, is to automatize the process through software that would compute and enable visualization of the prioritization results. The interface could be for instance designed for QGIS, taking the form of a plugin.. 1.2. Scope of the degree project. After reviewing the different planning methods, either demand-based or supply-based, we designed a model that had to account for the trade-off between including lot of features 5.

(18) 1. INTRODUCTION and transparency for its future users. Then, based on the freely reachable data, for both bicycle and e-bicycle, and for each trip purpose we could design an aggregated four-step model forecasting the bicycle daily trips. The demand calculated from these models could be then compared to the actual supply, to rank the new required investments. The support system is run as a QGIS 3 plug-in programmed on Python, which will not be part of the Degree Project. This Master Thesis addresses the modelling part and on the testing on Södermanland County. This report will also skip the inclusion of vertical equity variables, and will mostly focus on the adaptation of the four-step models for bicycle demand, and on how they can provide different kinds of results.. 1.3. Thesis outline. The report will present the results of the creation of a new model. It will first draw a literature review (Chapter 2) of the transport models and the specific aspects of bicycle models. The different steps of the created model will be explained in the following part (Chapter 3). It will then present the open data that fed the updated Kågeson model (Chapter 4). The new model will be adapted to a test region: Södermanland and the calculation results will be exposed in the following part (Chapter 5). We will finally be able to criticize the model and draw potential improvements to it (Chapter 6).. 6.

(19) SECTION 2. Theoretical Background: Literature Review Before designing a Planning Support System (PSS), it is important to review the different models that have been or that are currently in use for transport planning, both in a research framework and in practice. We will draw a theoretical background on transport supply and demand modelling, then focus on bicycle peculiarities, and their inclusion with other modes’ planning process. Moreover, one must be aware of the different model virtues and drawbacks to design one that will be used by practitioners. This literature review aims as well to identify some trends and insights on bicycles as a transport mode and to evaluate the potential impact of e-bicycles on the planning process. Finally, the review investigates the inclusion of policy goals in a PSS. This literature review, combined with the data retrieving, will allow a design of the new model and planning support system.. 2.1. Planning Support Systems. A planning support system is a tool, often implemented on a computer, designed to help public authorities and regional planners to analyze the different aspects of a transport system (land use, transport network, environmental and social issues). These tools enable them to exploit large data-sets for decision-making purposes and are often using Geographic Information System (GIS) (e.g. [Page et al., 2020]). Planning Support System (PSS) can be powerful tools, but some barriers make them often under-used by transport planners. Geertman and Stillwell [Geertman & Stillwell, 2009], Dutch researchers, tried to identify the examples and characteristics of successfully used PSS. They identified three bottlenecks to their use: • Instrumental quality: The PSS has to be able to carry out the tasks they have been designed for and has to fit the capabilities and demands of the intended users • Diffusion: The institutions have to begin to use the PSS • Acceptance: The planners have to know that the PSS exists, and for which purpose it can be used, and what benefits it can raise It is important to have in mind these barriers while designing a new support system. It has to deal with its complexity and transparency, as black-box models that are poorly connected to the planning process are less likely to be used by authorities (as states e.g. [te Brömmelstroet, 2010]). 7.

(20) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW To avoid these barriers, it is important to consider end-users as important parts of the PSS designing process. [Saujot et al., 2016] advise a bottom-up approach, involving the stakeholders in the testing of the model on an example that will help them understand its most complex links and interactions. A bottom-up approach, contrarily to the top-down approach, analyses a particular case (here, the test on a region) to understand the global model. It moreover helps to calibrate and validate the model on a real example. Then, stakeholder workshops will be an important part of the designing process.. 2.2. Estimating and Forecasting Demand. Estimating transport demand requires modelling, whose assumptions simplify the transport system complexity. Moreover, when it comes to forecasting the demand evolution after any transport policy, a model is the only solution. Then, there is a broad set of transport models whose goal is to estimate and forecast demand. There exists several models to estimate or forecast demand for the transport system. We can call the transport system the model’s target. Models vary by their simplicity, data needs, computer tractability, but also by their purposes, which can be from the prediction of important network links to traffic flow forecasting hour per hour for each mode. They include assumptions, or simplification, and their number affects the model’s similarity to its target. This section describes some of these different models. Transport models often include supply and demand concerns to match them with infrastructure supply and identify the investments to prioritize. The U.S. department of transportation [U.S. Department of Transportation, 1999] summarized the different transport model used for bicycle paths planning. According to their report, it is possible to separate the demand estimation models into two sub-categories: the models estimating and predicting demand, e.g. in terms of daily flow, crowding; and the ones estimating the relative potential demand, e.g. the relative potential use of a transport alternative. Both these kinds of models will be described in the next subsections.. 2.2.1. Classical multi-steps models. The multi-step transport models are mostly used to predict car and public transport use, but can also be adapted to predict bicycle facility demand. According to van Wee’s classification of transport models [van Wee et al., 2013], we will focus on models that are: • Spatial, as they consider facility demand at different locations and of an area, whereas non-spatial models’ purpose is to describe the transport patterns at a macroscopic scale (country, region...). • Activity-based, i.e. considering transport as derived demand for activities, jobs, services. They are opposed to trip-based models, which consider trips independently of each other. A commonly used transportation model is called the four-step model. The steps consist in: 8.

(21) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW 1. Trip generation: Estimate how many trips depart from a zone, and arrive in the zone on a given period of time (e.g. one day). They are activity-related. The output is a vector of trip origins and of trip destinations 2. Trip distribution: From the vectors of departing and arriving trips, the direction of the trips: or the number of trips between each origin and destination. It is often modeled through a Gravity model, which is explained in Section 2.3 3. Modal split: The distributed trips are assigned to a mode, often using a Discrete Choice Model, explained on Section 2.4 4. Route assignment: The trips are assigned on the road-network, which can be done through several methodologies (e.g. All-or-nothing, or including congestion with the Wardrop principle), explained in Section 2.5 The four-step models can then be separated into two main kids: Aggregated and disaggregated models. Aggregated models: They are models for which the zones of an area are the units of observation. These zones are linked to networks (here, the roads and bicycle paths). The most famous model for demand modelling and forecasting has the shape of a four-step process, first implemented in Chicago in the early 1950s, as states McNally [McNally, 2000]. Disaggregated models: First developed in the seventies [Associates, 1972] in the United States, they are based on individuals and their choices. It can often be seen as a Nested Multinomial Logit model (e.g. [Ben-Akiva et al., 1985]), whose nests represent several steps or choices (see Figure 2.1): choice of doing a trip or not, choice of destination, choice of mode and choice of route The choices are then aggregated by summing every individual behaviour over an area. This is the kind of model that is for instance used in Sweden with the Sampers model [Algers & Beser Hugosson, 2002]. The models can be illustrated with trees whose nodes represent different decisions. The tree below (Figure 2.1) represents the model for mode choice for work tours. Other similar trees also exist for non-work tours, and the results are then aggregated. The nodes represent the three first steps, the assignment part is then done through shortest-path algorithms. These models, when changing the characteristics of the infrastructure network, and thus some variables (as travel time, travel comfort, safety), will allow to compute a new forecast of the volumes for each mode and to derive the potential costs and benefits from it. Mostly used for car and public transport planning, this modelling process has the advantage of highlighting the competition between transport modes. They also have been used for regional non-motorized planning, with a Swedish example run by VTI in Södermanland [Chengxi, 2020]. These models present some technical limitations: if cars and public transport have easily understandable explanatory variables (cost, travel time) for the discrete choice model utilities, it seems that bicycle traffic has a lot of potential explanatory variables that would 9.

(22) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW. Figure 2.1: Work tour regional model in Sweden [Algers & Beser Hugosson, 2002] require a lot of data, as states Krizek et al. [Krizek et al., 2006], like weather, infrastructure quality, intersections, parking... Moreover, they require precise travel surveys, which are not readily available in most rural areas. For instance, the model built by VTI for Södermanland by Chengxi Liu [Chengxi, 2020] builds its model on an Origin-Destination (OD) matrix that has been given by the region. The mode choice part of the model is estimated based on travel survey results from Örebro county, assuming that it shares most of its background variables with Södermanland.. 2.2.2. Relative Potential Demand models. Other simpler models can be data-wise cheaper alternatives to the demand forecasting models. When it comes to prioritizing a project instead of another, is possible to only calculate the relative potential demand for bicycle transportation around network links. Calculating relative potential demand boils down to calculate a demand indicator linked to the density of potential origins and destinations around the road network. These models purpose is not to calculate a flow, but the relative potential use of bicycle links. These models are network-based, as the network links are the units of observation. One famous relative demand potential model is the Latent Demand Score (LDS), developed by Landis [Landis, 1996], whose purpose is to quantify the potential trip interchange between generators and attractors, for four different trip purposes (Work commuting, School, Shopping and Errand, Recreational and social), using a "probabilistic gravity model", computing only the first two stages of the four-step model presented above. These models have the particularity not to account for the existing infrastructure quality, considering the ideal conditions for cycling are actually met throughout the network. Many relative demand indexes have been used for regional models in the United States, with some examples in Charlotte or Twin Cities [CCPRC, 2016, Elmer et al., 2014]. These models consist in creating network scores have been created region-wide, by creating buffers 10.

(23) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW around road networks and counting the number of activities, services, and jobs centres fall within them. These models can also be used for roughly forecasting demand, applying the global travel survey’s bicycle share to the potential trip distribution, as it has been done by McGill University researchers in Québec City, Canada [Grisé & El-Geneidy, 2017].. 2.2.3. Include combined commuting. In Sweden, inter-modal trips were representing 9% of the total trips taken in 2007, while bicycle trip share was around 10% at the time [Hegger, 2007]. Integration of bicycle in the public transport system is a part of most of the transport plans, and is a key of successful bicycle policies (e.g. [Pucher & Buehler, 2008]). Combined commuting or inter-modal trips are trips that use several transport modes. We will focus on the most common combination: bicycle and public transport (bus or rail), where the bicycle can be used for access (sometimes called first-mile) or egress (last-mile). These kinds of trips have a good propensity to replace car trips as they allow greater distance coverage, especially outside urban areas. They combine the flexibility of bicycles and the speed and range of public transport. Still, these considerations are rarely considered in transport planning, and many efforts should be made for integrating these two modes. Include it in bicycle infrastructure planning is one way to do it, [Kager & Harms, 2017], and it has to be coordinated with investments in parking, bike rental schemes, bike-on-board policies, and information for users to be fully efficient. Combined commuting can be included in transport models by different means. For instance, in a nested logit model as presented on Figure 2.1, an additional nest can be added on public transport stations to add another mode choice model [Liu et al., 2020]: the feeder mode, which can be car, walk, bicycle,... It allows to model more precisely the trips between the origin and the public transport stops. Another integration of combined commuting can be through considering trip chains (Bicycle + Public Transport + Walk, Walk + Public Transport + Walk, Car + Public Transport + Bicycle,...) as different competing modes with the classic modes (Car, Bicycle,...) in a discrete choice model. These bicycle trips are taken on shorter distances than full bicycle trips. A study made in the Netherlands [Shelat et al., 2018] estimated that the average bicycle access distance was way higher for faster and far-reaching modes (around 3.8 km for accessing train, while 1.5 km for bus, tramway, or metro). This study also shows that bicycle is most likely to be used for first-mile trips, while last-mile trips are commonly done walking. Now that we developed the different model structures and features, we will explain briefly the mathematical concepts around them.. 11.

(24) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW. 2.3. Gravity Models. Gravity models allow to distribute the trips from the different origins and destinations, and are the main tools of the second step of the four-step models. They illustrate relationships between places and the transport network. First been theorized in the 1850s [Carey, 1867], they are analogous to Newton’s law for gravitation. They began to be used in transport models in the 1950’s (e.g. [Voorhees, 1955]). Knowing the trip generators and attractors of each zone, trip distribution can be computed according to the following equation: Tij = Ai Bj Oi Dj f (cij ) Where Oi , Dj are the origin and destinations sizes, cij is the generalized cost for travelling from i to j, Ai , Bj are the balancing factors, assuring that the origin and destinationPsizes are equal to Pthe sum of the trips beginning from it or arriving to it, i.e. Oi = j Tij and Dj = i Tij . The matrix whose coefficients are Tij is called the Origin-Destination (OD) matrix of the area. The constraints give the expressions: 1 j Bj Dj f (cij ). Ai = P. 1 i Ai Oi f (cij ). and Bj = P. These coefficients are adjusted iteratively with algorithms such as Iterative Proportional Fitting (IFP). The distance-decay function: f is called the distance-decay function, The distancedecay function is an important feature of a gravity model. Its goal is to model the decreasing attractiveness of a destination according to the travel friction. It has been shown in the gravity models that the travel purpose impacts the bicycle travel patterns, and therefore, it is relevant to use a different distance for each travel purpose (e.g. [Iacono et al., 2008]). The classical functions that are fitted to actual travel data are often either power functions or exponential functions: f1 : x −→ x−α f2 : x −→ e−βx Where α and β are positive parameters to be chosen. These functions are commonly used when it comes to evaluating accessibility, but also sometimes for forecasting trip interchange.. 2.4 2.4.1. Mode choice modelling Multinomial logit models. A mode choice model, or more generally a discrete choice model, calculates the probability of using a given transport mode as a function of different parameters, often aggregated as a generalized cost. The most commonly used tool for predicting mode choice is the Multinomial Logit (MNL) model. These models, derived from the utility theory, have been first developed in the late fifties [Luce, 1959]. Their first adaptation to transport models 12.

(25) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW has been done in the early seventies (e.g. [McFadden, 1974]). These models can be based on individuals, or can also be aggregated to zones. They have the following formulation, considering a trip between given origins and destinations: Let C = {1, ..., J} be a set of alternatives, whose attributes are a vector xj for j ∈ J1, JK. Then, we define the utility Uj of alternative j as: Uj (xj ) = Vj (xj ) + j Where Vj is the deterministic utility of alternative j, that can be described by its attributes, and j is the random utility component, linked to background variables that cannot be calculated easily (individual variability, ...). Then, the probability of using an alternative i rather than an alternative j is given by: P(Ui > Uj ) = P(Vi − Vj > j − i ) Assuming i to be Gumbel-distributed for all i, the probability of choosing alternative i rather than any other is given by: eVi P P(Ui ≥ max Uj ) = j∈C eVj j∈C. In transport models, xj is a vector containing the generalized cost components of travelling between the studied zones. Those components are often travel time, cost, comfort... NB: A constrained gravity model with an exponential distance-decay function is a Multinomial logit model, whose choice set is, for each origin i, the different destinations. The deterministic utility of going from origin i to destination j is then given by Vj = log(Bj Dj ) − βcij . The first component is linked to attributes of the destination, and the second is linked to the travel friction between them. The inclusion of a component relies on the available data. For instance, the Propensity to Cycle Tool ([Lovelace et al., 2016], [Goodman et al., 2019]) chose to only include bicycle and any other mode in their forecasts of mode share for work and school trips. They used a binary logit model, whose choice set was {Bicycle, Other}. The probability of using the bicycle is given by the function P(bicycle). Its variables were polynomial terms of distance and elevation. eV (x) 1 + eV (x) V is representing the deterministic utility of using the bicycle, and its parameters are estimated according to the results of two different travel surveys. This model bases on rich English data-sets on mode choice per OD pair. It then forecasts the probability to use a bicycle on different scenarios of cycling uptakes: P(bicycle) : x −→. • If the population had the same bicycle travel behaviour as the Dutch, using the Dutch National Travel Survey (NTS) • If the population was only using electrical bikes, using the ebikes subset of the Dutch NTS 13.

(26) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW This tool, by trying to simulate a change in travel habits is then interesting, and answers to one of Trafikverket’s goals, which is to include the breakthrough of electric bicycles.. 2.4.2. The Independence of Irrelevant Alternatives (IIA) property. A property of the Multinomial Logit Models is the Independence of Irrelevant Alternatives (IIA). The ratio of probabilities of two alternatives i and j in a choice set is given by: P(i) eVi = Vj P(j) e This ratio is independent of all the other choice set alternatives. In some cases, this property can be a problem, as explains McFadden in the Blue bus/Red bus paradox [McFadden, 1974]: • Considering an individual has two possibilities to travel to work, taking the car or taking the bus. The choice set is C = { Car, Bus }. Considering both choice have the same utility, they will both have the same probability of 12 • Now we consider that there exists two choices for the bus, a red one and a blue one, also having the same utility. Then, the probability of each alternative will be of 13 , and the overall probability of using the bus will then be 23 This result seems unrealistic, as the addition of two possibilities for the bus does not seem to decrease the probability of using the car. One of the solutions to this paradox is to arrange similar choices in the same nests, in a so-called Nested logit model.. 2.4.3. Nested logit models. As illustrates Figure 2.1, discrete choice models can include several levels of choice.. .... 1. A1. .... An1. n. B1. .... Bnn. Figure 2.2: Illustration of a two-level nested logit model A straightforward way to describe Nested Logit Models is by decomposing it into two Multinomial logit models (e.g. [Train, 2009]). Let Cj be the choice set of the nest j. The utility Ui of an alternative i in nest j can be written as: Ui = Xj + Yi|j + i Where Xj is the deterministic utility of nest j among all the other nests, and Yi|j is the deterministic utility of the alternative i among nest j. The probabilities of each alternative 14.

(27) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW in the lower nest are calculated through the conditional probability formula. For an alternative i of the nest j, we have: P(i, j) = P(i)P(j|i) eXj +µj Lj P(j) = P n eXk +µk Lk. 1. Y. e µj i|j and P(i|j) = P 1 Y e µj k|j k∈Cj. k=1. Where Lj is called the inclusive utility of nest j, given by the log-sum formula: Ñ é X 1 Vi Lj = ln e µj i∈Cj. We call µj the log-sum parameter varying between 0 and 1, which illustrates to what extent alternatives of a same nest are correlated. The case µj = 1 makes the nest disappear. Nested logit models solve the IIA problem for alternatives that are in different nests. Indeed, the ratio of the probabilities of alternative i from nest k and alternative j from nest l will depend of all the other alternatives in the other nests. The IIA property remains for alternatives in the same nest. In the case of a four-step model, another reason motivates the usage of the Nested Logit Models. Indeed, a model that involves several kinds of choices, which Ben-Akiva called multidimensional choice models ([Ben-Akiva & Koppelman, 1974]), can be represented as a Nested logit model. In this case, separating the different dimensions in nests allows including less variables, thus less data collection and computation times. The value of the log-sum parameter of each nest is given by µ = 1, it indicates that there is no correlation between the alternatives of each nest in terms of unobserved variables.. 2.5. Route assignment. Route or trip assignment consists in assigning paths between each connected origin and destination computed in the trip distribution step, and for all modes computed in the mode choice step. We will focus on this subsection on assignment for private trips (e.g. car, walk, or bicycle trip), as public transport trips use different assignment methods. Route assignment considers the road network as an oriented graph, i.e. nodes that are linked by weighted vertices. Several solutions exist for route assignment, which can be either static or dynamic. All-or-nothing assignment: All-or-nothing assignment can be called a static method. It assigns trips on the route which gives the smallest travel generalized cost for its users. Assignment is then computed through a shortest-path algorithm, such as the Dijkstra algorithm [Dijkstra, 1959]. This method can still be insufficient, mainly for car traffic, as it does not include potential congestion caused by everyone using the same link. Dynamic methods can then be helpful.. 15.

(28) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW Dynamic assignment: Dynamic assignment considers that the generalized cost of travelling on a link depends on the flow of this link. Considering that a road has a fixed capacity, i.e. a maximum flow it can handle without congestion, volume-delay functions link delay to the volume on a road segment. For instance, the U.S. Bureau for Public Roads [BPR, 1964] expressed the travel time ta on a link a as: Ç Å ãβ å va ta (va ) = t0a 1 + α ca Where t0a is the travel time on link a in "free-flow" conditions, ca is the capacity of link a, and va is the volume on link a. α and β are coefficients depending on the road type. Then, assignment can be done following the equilibrium principle, which considers that all the used route between an origin and a destination take the same time, and that no route using less time will be unused. This process allow an update of the generalized travel cost between origins and destinations, and can feed the former steps of the transport model in an iterative process. Still, the dynamic assignment methods are mostly relevant for car traffic, even if they are sometimes used for bicycle in urban areas (e.g. [Johansson, 2018]).. 2.6. Bicycle supply analysis. To match the demand for bicycle paths with the supply of bicycle paths, it is important to know the actual level of service of the network links for bicycles. For instance, on a road with low car traffic flow and a small speed limit, it may not be smart to build a separated cycle path even if the bicycle demand is high. To analyze bicycle infrastructure efficiently, it is important to evaluate it at the segment-scale, but also at the full network scale, to identify where it fails to fulfil accessibility.. 2.6.1. Metrics to evaluate a facility. Supply analysis, when focused on a network segment, allows evaluating the safety and comfort it provides. These metrics may then be used for route choice to calculate the generalized cost of cycling on a link. It might be calculated through different kinds of metrics. Some literature reviews of the different indexes have been recently carried out (e.g. [Kazemzadeh et al., 2020], [Arellana et al., 2020]), and the supply analysis mostly varies by the variables it takes account of. Some of the most used bicycle supply metrics are listed in Table 2.1. The ranking indexes are often fit to Revealed Preference (RP) data through questionnaires about people’s perceived satisfaction with a road segment, or choice between several alternatives with varying attributes. The Level of Service Indexes is then modelled through either linear or logistic regression. It is possible then to select the most significant variables explaining people’s perceived safety and comfort. These data are not linked to demand forecasts.. 16.

(29) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW. Name. Reference. Country. Bicycle Level of Service (BLOS). [Landis et al., 1997]. USA. Bicycle Compatibil[Harkey et al., 1998] ity Index (BCI). USA. Highway Capacity Manager (HCM) [Elefteriadou, 2016] BLOS Pedestrian and Bicyclist Level of Ser- [Jensen, 2007] vice (PBLOS) Level of Traffic [Mekuria et al., 2012] Stress (LTS). USA. Variables accounted Bicycle facility (lane/shoulder), Width, Number/type of traffic lanes, surface Presence of bike lane, width, volume, car speed, presence of parking, residential area, truck volume, right turn factor Presence of facility, nb of lanes, shoulder, surface quality, bike lane width. Average Daily Trafic, infrastructure Netherlands width, average speed, type of land use, facility type Facility type, number of motor lanes, USA bike lane width, speed limit, on-street parking. Table 2.1: Bicycle link evaluation indexes The level of service can then be used for route choice modelling, as it allows giving weights to the network links. A study made in Trondheim, Norway ([Pritchard et al., 2019]) evaluated the results of a weighing scheme through different Level of Service indexes by comparing them with the actual individual route choice. It has been concluded that the more an index is complex (i.e. including a large bunch of variables), the more it could grasp the potential changes in route choice. These indexes can also locally be used to identify a spatial mismatch between demand and supply with some GIS techniques, such as auto-correlation. For instance, Rybarczyk analyzed in Milwaukee municipality the auto-correlation between demand and supply patterns, using Landis’s BLOS and a simple demand index based on neighbourhood land use ([Rybarczyk & Changshan, 2009]). If he identified no spatial autocorrelation for bicycle supply, he identified that adjacent neighbourhoods had similar demands. He concluded that further research had to be made to match efficiently supply and demand in that region.. 2.6.2. Level of Traffic Stress (LTS). This paragraph will focus on one metric that evaluates links: the LTS. Created by researchers from the Mineta Transportation Institute [Mekuria et al., 2012], it is a useroriented index that evaluates bicycle infrastructure. Indeed, it is based on a classification of users that has been created by Geller in Portland city, [Geller, 2006]. Geller separates cyclists in four categories: • The Strong and the Fearless: people who would cycle regardless of the infrastructure • The Enthused and the Confident: are comfortable sharing the roadway with automotive traffic, but they prefer to do so operating on their own facilities 17.

(30) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW • The Interested but Concerned: People who may cycle only under safe conditions, i.e. separated from car traffic, or where traffic is low and slow • The No Way No How: People that do not cycle for transportation, regardless of the conditions. These people may still cycle as an activity per se. The categorization is based on several studies and surveys, including U.S. Census data, American Community survey figures and a smaller scale random phone survey, carried out by professor Dill from Portland University ([Dill, 2005]). These categories are linked to the kind of stress environment a cyclist can be expected to ride on. For instance, "The Strong and The Fearless" can be expected to ride on any kind of road, while “The Interested but Concerned” can be expected to ride only on separated bicycle paths. The LTS ranks the infrastructure according to these cyclists profiles in four categories. The variables are speed limit, ADT, number of lanes, presence of parking lane, presence of separate bicycle path... Furth created tables that give the level of traffic stress in the function of speed and number of lanes for the different configurations. All the tables are available on his website [Furth, n.d.]. Then, separated bicycle paths always have a LTS of 1, then, tables like the following give decision rules on which LTS a road segment should have (with a bicycle lane or in mixed traffic): LTS ≥1. LTS ≥ 2. LTS ≥ 3. LTS ≥ 4. Lanes per direction. 1. 1. 2 or more. 2 or more. Sum of bike lane and parking lane width. 15 ft or more. 14 ft. 13.5 ft or less. N.A.. Speed limit. <25 mph. 30 mph. 35 mph. >40 mph. Bike lane blockage. rare. rare. frequent. N.A.. Table 2.2: Level of Traffic Stress conditions for bicycle lanes This metric is reproducible in Sweden, using the data from Vägar och gators utformning [Trafikverket, 2021], which gives a set of regulations on which kind of infrastructure should be built under which conditions.. 2.6.3. Total network analysis: connectivity. The connectivity of a network [Black, 2001] is its ability to connect well the different origins and destinations, to fulfil accessibility. It often uses a segment-based index to compute a more global index throughout the network. These indexes are associated with accessibility metrics, computing the ability to travel using a bicycle on decent infrastructure. Most of these models are considering that cyclists may do some reasonable detours to avoid unsafe network links. This is important to consider, as some "low-stress" alternative links could replace investments in bicycle paths on "high-stress" infrastructure.. 18.

(31) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW. Name. References. Comments. Bikeability Index (BI). [Lowry et al., 2012]. Low-stress bicycling [Mekuria et al., 2012] Low-stress bicycling con[Ting & Wei, 2019] nectivity. Network-wide HCM BLOS 1 , using a gravity model Network-wide LTS, considering detours Network-wide LTS. Table 2.3: Bicycle network analysis methods. 2.7. Policy goals of cycling. Cycling can lead to benefits for both individuals and society, as it could, for instance, both decrease people’s travel costs and reduce the negative externalities from other modes (e.g. [Sælensminde, 2004, Börjesson & van Wee, 2015]). The main dimensions of benefits are about health, travel time, security, and reduced travel costs (both for individuals, employers, and external costs of motorized transport). Costs are linked to capital costs for building, maintenance costs, and tax-cost factors. If the calculating the costs and benefits linked to a potential investment is not the goal of to-be-designed planning tool, we still aim to include some equity aspects in these improvements allocation. A Cost-Benefit Analysis (CBA) will anyway be run in later stages of the planning process. Equity is also an important feature to include in a transportation model. Indeed, in a CBA framework, not every new bicyclist will provide the same benefit to society. Moreover, some factors like car ownership might have a real impact on the number of new bicyclists a new facility would create. Identifying areas that have more health, social inclusion, or economic issues is one of the dimensions that may help to prioritize investments in bicycle infrastructure. Health: The positive impact of bicycles on health is, for instance, measurable in terms of average sick leave. It has been statistically proven to be reducing sick leave in the Netherlands by more than one day per year [Hendriksen et al., 2010], comparing cyclists and non-cyclists. Then, it can be interesting to identify municipalities with the longest sick-leaves, which can be found on Statistiska centralbyrån (SCB). Social inclusion: Bicycle is often used as a tool to promote social inclusion of different groups. Indeed, it can contribute to improving accessibility for the ones that cannot drive or own a car. As some Swedish areas do not fulfil social inclusion in its transport infrastructure (e.g. [Jennings, 2018]), one policy goal is to improve bicycle infrastructure in zones where minorities live. One of Trafikverket’s goal in cycling investments is to promote social integration of foreign-born people, through infrastructure improvements in clusters where a lot of immigrants live, and through education on how to cycle [Trafikverket, 2019].. 19.

(32) 2. THEORETICAL BACKGROUND: LITERATURE REVIEW Income: As cycling is a relatively inexpensive means of transportation, it is relevant to think they will provide greater social welfare to the poorer households, which often have less access to cars and public transport (e.g. [Di Ciommo & Shiftan, 2017]). Equity is rarely present in a transport modelling process but is often part of assessment studies of actual infrastructures, that may give guidelines for future investments. Anyway, some studies tried to include vertical equity in a gravity model [Manrique et al., 2020], by distorting the origin sizes of a gravity model by a coefficient for each individual average income. Grisé and El-Geneidy [Grisé & El-Geneidy, 2017] evaluated vertical equity in transport planning by calculating a social-disadvantage score for a block, based on income, immigration rate, unemployment and percentage of income spent in rent. Then, the cycle path providing the highest number of socially-disadvantaged people to use them were prioritized. It is important not to consider several variables that are correlated, as we may account several times for the same effect. Principal Component Analysis (PCA) is a good method to reduce the number of attributes to be considered, reducing the number of variables to the ones that explain the highest part of the data variance. For instance, Sanchez-Cantalejo et al. [Sánchez-Cantalejo et al., 2008] calculated a social deprivation index for Spanish cities, reducing six variables (Unemployment, Illiteracy, Manual Labourers rate, Car-less households, Percentage of active foreigners, Percentage of the non-active population) to two aggregate variables, one more correlated to the first three variables, the other to the last three.. 2.8. Inclusion of e-bikes in transport models. Like anywhere in Europe, the use of e-bicycle is fastly increasing in Sweden, and its market share is around 20% in the total bicycle market share, every year since 2018, and the trend is increasing, the first electric bicycles were sold in 2011 [Elm & Strömgren, 2018]. This increase is supported by the Swedish state, which funds 25% of the cost of any new e-bike bought. We can then expect an increase in the share of electric bicycles in the future, even if they’ll probably not replace regular bicycles. Electric bicycles are still under-represented in the infrastructure planning process, and it is one of the to-be-designed model goals to include them. Modelling potential infrastructure demand for electric bicycles will be an output of the distance-decay function. The includes them in a scenario by adding parameters to the distance-decay function reducing the impact of hilliness and distance in the probability of using cycle. A study carried out in Norway [Fyhri & Fearnley, 2015] shows different interesting results on the impact of e-bikes on travel behaviour and mode share: • E-bikes increase the amount of cycling expressed as both the number of trips and as distance cycled • They erase the gender inequalities in terms of bicycle use, having more impact on change in women’s travel habits • E-bikes increase commuting bicycle mode share (but not distance travelled) and increase leisure trip length (but not the amount of trips) 20.

(33) SECTION 3. Model choice and design The goal of this Master Thesis is to create a model that would be used in the first steps of the bicycle path planning process. The budget invested for the bicycle infrastructure investments are not depending on this tool’s output, but it has to advise on where to allocate it the most efficiently. We need to find a trade-off between the model’s simplicity and the amount of assumptions it makes. It has also to be understandable and easily usable by the planners. The main purpose of the model is to identify gaps between potential bicycle demand and actual infrastructure, considering all the trips purposes. It also has to include the potential new trips generated by the breakthrough of electric bikes.. 3.1. Demand and Supply, is it a match?. The main model purpose is the ranking of needed improvements according to the relative demand for each link. Then, supply and demand will be considered separately. It means that the demand forecasts are independent of the actual bicycle infrastructure. Consequently, it will be said to calculate the potential bicycle demand on the regional road networks. The word "potential" is used because of of the model’s assumption of considering the network perfectly suitable for cycling everywhere. Indeed, the goal is to improve the network, the demand must not be biased by the actual network quality. As Landis did for the Latent Demand Score [Landis, 1996], the goal is to evaluate the potential use of the road network by cyclists regardless of its suitability for cycling. Still, this model is designed to give an approximation of the potential daily bicycle flows in a future where the bicycles are more included in the road planning process.. 3.1.1. Demand analysis: A four-step model. The goal is to adapt an aggregated 4-step model as described in the Literature Review to the available data. A separate model would be run for each mode (bike and electric bike) and for each purpose (Work and School commuting, Touring, Service, Leisure and Access to Transit). Depending on the available data, the model will either simulate the trip distribution or use the actual origin-destination relationships. The model will not be disaggregated at the individual model, but will consider sufficiently small areas as origins and destinations so that it is assumed that the travel behaviour of a person will not depend on its location in the zone. The results will be aggregated using the Resvaneundersökning. 21.

(34) 3. MODEL CHOICE AND DESIGN (RVU) statistics on the number of daily trips per person for each purpose, considering they do not depend on the bicycle infrastructure.. Bike. Work/School Commuting. Touring. E-Bike. Service. Work/School Commuting. Leisure. Touring. Service. Leisure. Figure 3.1: Disaggregation of the models into sub-models. Why a separate model for bicycle and e-bicycle? Most of the travel demand models make a competition between the different modes that can be used. In this model, we did not consider bicycle and e-bicycle as competing modes. The reason is that e-bicycle ownership is an important variable to explain e-bicycle trips, as they are still not common. Mode choice is predicted according to Dutch data, which includes ownership in its surveys questions. This data being unavailable in Sweden, we decided to exclude bicycle or electric ownership as a variable and keep only examples of people owning a bicycle or an electric bicycle. Since both models are not calibrated on the same data subsets, they must be separated. 1. Trip generation: First, identify the origins and destinations of potential bicycle trips for the different purposes, and give them size linked to the number of trips they can produce. 2. Trip distribution: Then, we calculate the potential trip interchange between the origins and destinations with an origin-constrained gravity model. The gravity model doesn’t need to be destination-constrained as the destinations modeled (Public Transport Hubs, Leisure and Service centers, Natural areas) are not assumed to have a limited capacity. For the purposes where we already know the origindestination relationships (Work and School commuting), this step is irrelevant. The distance between origins and destinations is calculated with a shortest-path algorithm programmed on Python on QGIS layers. 3. Mode Choice: To model the mode choice, we will model the probability of choosing a bicycle instead of any other mode (resp. electric bicycle instead of any other mode) through a binary logit model whose variables are distance and trip purpose. The road network comfort or safety is neither considered in the distribution nor in the mode choice. Indeed, the goal is to find potential interchange regardless of the infrastructure. 4. Trip assignment: The potential trip interchange is assigned to the network to compute potential flow indexes via a QGIS Python algorithm. As regional bicycle 22.

(35) 3. MODEL CHOICE AND DESIGN path are not likely to reach their capacity, we will use the straightforward "all-ornothing" assignment method.. 3.1.2. Supply analysis. Supply for bicycle is analysed using requirements from Vägar och gators utformning (VGU) and the Level of Traffic Stress metrics. Then, two metrics comparing demand and supply allow a prioritization of the needed improvements. A diagram illustrating the model is drawn on Figure 3.2, which uses a Structured analysis and design technique (SADT) technique. SADT [Dickover et al., 1977] is a commonly used method when it comes to design software or systems. It describes the solution as a hierarchy of functions. Probability of using bike/e-bike as a function of distance and purpose. Travel behaviour data Distance-decay function. Data on trip lengths Weighing scheme for each purpose. Demographic data Trip attractors data. Parameters to classify needed infrastructure. Shortest-path algorithm. Step 1: Compute the origins and destinations as point data. Step 2: Compute the relations between origins and destinations, weighted by their probability. Regional road network. Step 3: Compute the probability of using bike or e-bike for each OD-pair. Step 4: Compute potential ﬂows on the network, for each purpose and mode. Ranking of the most important network links for each recommended infrastructure. Actual bicycle infrastructure Actual bicycle supply evaluation. Trip frequency per purpose. Ranking according to the ratio potential demand/actual supply. Figure 3.2: SADT diagram of the model The model inputs are on the left blue boxes, the model parameters are on the top green boxes, and the model is the content of the white rectangle. The final model outputs are on the right orange boxes. For each trip purpose, a more detailed explanation of the different steps is done in the following subsections.. 3.2. The different spatial subdivisions. Statistiska centralbyrån (SCB), or Statistics Sweden is the main provider of statistics for decision making, debate and research 1 . They produce statistics through surveys to companies, government agencies and private persons. They also supply spatial data on demographics and on the labour market. SCB gives data at different scales, the two that will feed the model are: 1. According to SCB’s website. 23.

No results found