Investigation and Modeling of Urban Street Bikeability from User’s Perspective

Beura, Sambit Kumar (2018) Investigation and Modeling of Urban Street Bikeability from User’s Perspective. PhD thesis.

[img]PDF (26/01/2021)
Restricted to Repository staff only

2537Kb

Abstract

Despite investments in transportation planning and developments, several cities around the
world are facing the problems of poor urban bikeability. Bicycle Level of Service (BLOS)
models play the major roles in identifying the source of problems for this. However, reliable
BLOS models are yet to be developed for street facilities operating under heterogeneous
traffic flow conditions. To this end, the present study has taken an initiative to identify and
model the variables having significant influences on the BLOSs of street segments,
signalized intersection approaches and unsignalized intersection approaches (uncontrolled
type). Efficient strategies are also ascertained for the bikeability improvement at these
facilities. Further, as automobilists are the most dominant road users, effort has also been
made to refine the ascertained strategies in such a way that those would not be at odds with the automobile level of service (ALOS) criteria. For investigation purposes, a wide range of diversified data sets (geometric, traffic and built-environmental details) are collected from as many as 74 road segments, 70 signalized intersection approaches and 70 unsignalized intersection approaches located in various parts of ten Indian cities. The perception surveys of on-site bicyclists and automobilists are also carried out to gather the information on their socio-demographic and travel-related characteristics, and perceived satisfaction scores at respective sites (varying from 1 = extremely satisfied to 6 = extremely dissatisfied).
Spearman’s correlation analyses have been carried out to identify the variables having significant influences on BLOSs of investigated facilities. The set of significant variables for road segments included: effective width of the outermost lane (WOTLEff), peak hour volume per lane (PHV/LSeg), average traffic speed (S), pavement condition index (PCI),
roadside commercial density (CDSeg), interruptions by the stoppages of intermittent public
transits (IIPT), vehicular ingress-egress volume to the on-street parking area (PSeg), and the
frequency of driveways carrying high volume of traffic (DWFreq). Considering these
variables as model inputs, various statistical and artificial intelligence (AI) techniques are
utilized to develop BLOS models for street segments. Here, the statistical tools include
step-wise regression, ordered logit modeling and ordered probit modeling, and the AI tools
include genetic programming (GP), multi-gene genetic programming (MGGP),
associativity functional network (FN) and Bayesian Regularized Neural Network (BRNN).
Statistical analyses are carried out using SPSS, while AI techniques are implemented using
MATLAB. Of all techniques, the regression analysis has produced the least efficient but
simplest model, while FN and MGGP techniques have produced highly precise prediction
results with coefficient of determination (R2) values of around 0.88 and 0.87 respectively.
Sensitivity analysis of the FN-based BLOS model (the most efficient one) has reported that
WOTLEff, PHV/LSeg, and PSeg are by the most important variables, which contribute 38.31%,
21.86% and 12.68% to the prediction of segment BLOS. Thus, the utmost important
strategies for the bikeability improvement at street segments are identified as: the
augmentation of outermost lane width, and the construction of bypass roads, etc. The
remaining variables (S, PCI, IIPT and DWFreq) also have significant roles in the bikeability
improvement and should not be neglected in the planning process. However, of all
ascertained strategies, few strategies like the reduction of traffic speed, etc. seemed to be
contradictory to the automobile users’ service quality. Thus, efforts are made to know which
parameters should not be disturbed at all and which parameters could be disturbed to some
extent for improving the urban bikeability without snooping the existing ALOS. The
Spearman’s correlation analysis revealed that the ALOS of street segments is primarily
influenced by age group of automobile users (AG), spatial stop rate (SR), PHV/LSeg, S,
DWFreq, CDSeg and PSeg. Subsequently, ALOS models are developed by using various
favorable techniques (step-wise regression, GP, MGGP and FN), and the best results are
obtained with the MGGP technique (R2 ≃ 0.89). The sensitivity analysis of this model has
reported that PHV/LSeg and S contribute the highest and second highest amount (i.e., 26.33%
and 17.93% respectively) to the user perceived ALOS. Thus, the important constraints for
bikeability improvement strategies are identified as: (1) the traffic lanes should not be
utilized to construct dedicated bicycle lanes, and (2) the traffic speed should not be reduced
below the desired limit. However, wide outer lanes could be converted to a traffic lane and
a bicycle lane as the number of lanes remains constant, and strict rules could be imposed for
the vehicles moving with speeds higher than the specified limits.
The investigation of 70 signalized and 70 unsignalized intersection approaches has reported that the user perceived BLOSs at these facilities are influenced by seven variables each. Significant variables for signalized intersection approaches included: effective approach width (AWEff, SA), peak hour traffic volume on the approach (PHVSA), crossing pedestrian volume (PVSA), turning vehicular volume across the path of through bicyclists (VTurn, SA), average bicycle delay (DSA), on-street parking turn-over (PSA), and commercial density (CDSA). On the other hand, significant variables for signalized intersection approaches included: effective approach width (AWEff, UA), peak hour traffic volume per lane
(PHV/LUA), conflicting traffic volume (CTVUA), crossing pedestrian volume (PVUA),
commercial density (CDUA), on-street parking turn-over (PUA), and average bicycle delay
(DUA). Various analytical tools (such as step-wise regression, GP, MGGP and FN) are
utilized to model the BLOS of these intersections. The best prediction results for signalized
intersection approaches are obtained with the FN technique (R2 ≃ 0.92), while the best
prediction results for unsignalized intersection approaches are obtained with the MGGP
technique (R2 ≃ 0.94). The sensitivity analyses of these models have reported that the
signalized intersection BLOS is largely determined by DSA and PVSA (26.26% and 26.10%
respectively), while the signalized intersection BLOS is largely determined by DUA and
AWEff, UA (33.12% and 16.90% respectively). Thus, the most important strategies for the
bikeability improvement of road intersections include: the minimization of bicycle delay,
reduction of pedestrian interferences, minimization of bicycle-vehicle interactions through
special provisions (like flyovers, dedicated bicycle paths on subject approaches, etc.), and
the augmentation of approach widths, etc. These strategies are also visibly favorable for the
automobile users and do not need a detailed investigation of ALOS criteria for road
intersections. Field applications of developed BLOS model have indicated that around 95%
of the existing urban and suburban road facilities in India are offering average to very poor
service levels (BLOS C–F). Thus, the traffic planners and engineers could refer the proposed
BLOS models and bikeability improve strategies to identify the source of problems and to
take effective steps for the betterment of road users.

Item Type:Thesis (PhD)
Uncontrolled Keywords:Urban street; Bikeability; Developing Country; Heterogeneous traffic; Bicycle Level of service; Automobile level of service; Prediction modeling; Step-wise regression, Probit modeling; Logit modeling; Artificial intelligence; Genetic programming; Multi-gene genetic programming; Functional Network; Artificial neural network; Bayesian regularized neural network; Sensitivity analysis; Bikeability improvement strategy
Subjects:Engineering and Technology > Civil Engineering > Urban Engineering
Divisions: Engineering and Technology > Department of Civil Engineering
ID Code:9787
Deposited By:IR Staff BPCL
Deposited On:25 Jan 2019 12:42
Last Modified:25 Jan 2019 12:42
Supervisor(s):Bhuyan, P. K.

Repository Staff Only: item control page