Assessing Neighborhood-scale Traffic from Crowd-sensed Traffic Data: Findings from an Environmental Justice Community in New York City

Abstract

Background:

The waterfront in the South Bronx in New York City is used industrially and harbors the Harlem River Yards (HRY). The HRY borders an environmental justice area, which includes a mixed-use area that is separated from a densely populated residential area by interstates. Recently, development of the HRY has expanded including the 2018 opening of a large online store warehouse.

Objective:

The goal of this study was to evaluate trends in traffic congestion nearby the HRY between 2017 to 2019.

Methods:

We analyzed one-hourly time series of crowd-sensed traffic congestion maps, both at the neighborhood scale and the road stretch level. Traffic radar measurements at two locations did not indicate bias in the crowd-sensed data over the study period, i.e., changed mappings between vehicle speed and the reported congestion.

Results:

In the mixed-use areas, traffic congestion increased significantly during all hours of the day, with greatest increases at night and in the morning. Congestion increased close to the entrances of the HRY and along routes used by pedestrians and bicyclists to access a nearby recreational area. In the residential area, congestion increased significantly from midnight to morning and was unchanged for the remainder of the day. On the interstates, congestion decreased during the daytime but increased at night.

Conclusions:

Neighborhood-scale traffic congestion increased in mixed-use and residential areas in an environmental justice community. Our methods can be applied globally as long as crowd-sensed traffic data can be acquired. The data enable communities to advocate for mitigating measures.

INTRODUCTION

Road development and urban planning can have significant impact on traffic volume. The Mott Haven and Port Morris neighborhoods, an environmental justice area in the South Bronx in New York City (NYC) (New York City Climate Policy and Programs 2021) with a concentration of environmental hazards (Zimmerman et al. 2002) overlapping with elevated concentrations of poverty, racial/ethnic minority composition, poor health status (Hinterland et al. 2018), serve as an example. These neighborhoods have a long history of developments that caused traffic to increase, starting perhaps in 1939 with the construction of the initial 1.5-mile long stretch of the 6-lane Major Deegan Expressway (New York Times 1939) which spanned and divided southern to northern parts of the borough. Then in 1962, the first stretch of the 6-lane Bruckner Expressway was completed (New York Times 1962). The early portions of these expressways divided Mott Haven and Port Morris into three pieces, with a limited number of roads connecting them (Figure 1a). Later, these two expressways became part of the federal interstate highway system (I-87 and I-287) carrying not only traffic within the NYC metropolitan area but also traffic along the entire US East coast.

Figure 1:

(a) Map of the Southern portion of the Mott Haven and Port Morris neighborhoods and Randall’s Island to the South. Dashed black boxes delineate road stretches along which we examined traffic changes. (b) The overall study area is comprised of the four shaded subareas, in which we examined neighborhood-scale changes in congestion: two mixed-use areas (I and II), a residential area, and the enclosed interstates and their ramps. Base map from OpenStreetMap.

Interstates are not the only major source of traffic in the South Bronx. The Harlem River Yards (HRY) on the Port Morris waterfront in the South Bronx provides a home to several traffic-intensive operations including a shipping center, the municipal waste management facility for the entire Bronx, and the printing plant of a major newspaper (Zimmerman et al. 2002). In addition, the Hunts Point neighborhood along the South Bronx’ waterfront is home to one of the largest food distribution centers in the US (Zimmerman et al. 2018). In the 2000s, a rare opportunity arose to use a vacant land parcel on the HRY waterfront to create recreational space for the local community (Kochman 2014). Instead, a large warehouse of an online grocery store delivering food to the entire city was built on that parcel and went into operation in 2018. Constructing and opening of the warehouse was supported by an Environmental Assessment (AKRF Inc. 2011) which the community felt did not properly assess the environmental impacts of the warehouse (Ayala 2013, Hu 2013). In the last decade, numerous large residential buildings have also been built in the mixed-use area in Port Morris (South Bronx Unite 2021).

The South Bronx community is concerned about traffic for several reasons: 1) traffic-related air pollution has been associated with asthma exacerbation (Clark et al. 2010, Guarnieri and Balmes 2014, Khreis et al. 2017, Orellano et al. 2017, Lovinsky-Desir et al. 2019) which is of particular concern because of the high pediatric asthma prevalence, with a 17.1% school-based asthma prevalence for children aged 4–5 in the Mott Haven and nearby Hunts Point neighborhoods (Garg et al. 2003); 2) the rate of pedestrian injury hospitalizations almost doubles the rate for NYC as a whole (43 versus 23 hospitalizations per 100,000 people) (Hinterland et al. 2018); 3) the population density is very high, with many people living or children attending school next to the interstates (NYU 2009), thereby amplifying adverse public health impacts of traffic; and 4) access to the waterfront in the South Bronx is extremely limited because the industrial HRY borders the waterfront. Furthermore, traffic-related air pollution and noise have been associated with a number of cardiovascular and respiratory conditions and diseases such as hypertension (Fyhri and Aasvang 2010, Chang et al. 2015), myocardial infarction (Brook et al. 2010), and stroke (Villeneuve et al. 2006, Wellenius et al. 2012), and traffic-related noise disturbs sleep causally and relevantly (Janssen et al. 2011).

To evaluate the impacts of the 2018 grocery store warehouse on traffic, air quality and noise, we conducted a study in which we used traffic radar devices to measure changes in traffic volume at two road segments (Hilpert et al. 2019, Shearston et al. 2020). In that study, we also pioneered the use of “crowd-sourced” traffic congestion maps that are based on the GPS positions of smartphones located in driven vehicles in public health studies (Hilpert et al. 2019). However, smartphone users typically do not actively volunteer their data specifically for research like in many crowd-sourcing projects involving regular citizens as contributors. Instead, application software running on their smartphones shares the data from the device’s sensors with a traffic data provider. We therefore now refer to these data as “crowd-sensed” (Ganti et al. 2011).

We demonstrated that the colors with which road segments are shown in crowd-sensed Google traffic congestion maps are associated with vehicle speed, which we measured with traffic radar devices. We also found that the algorithm for color assignment accounts for the achievable vehicle speed on each road segment. Thus, congestion colors are proxies of vehicle speed, which the traffic data provider determines from the GPS positions of smartphones in driven vehicles (Hilpert et al. 2019). Road segments shown in green indicate free-flowing traffic near some ideal speed for that road segment, while orange, red, and maroon in that order indicate increased congestion. Gray is used to indicate the absence of a sufficient number of smartphones in the moving traffic that share their GPS position with the traffic data provider. The following congestion color code CCC can be assigned to these five colors (Hilpert et al. 2019):

(1)

where the color of the square box indicates the congestion color displayed on traffic maps, and the number in the box indicates CCC. Our previous study, however, did not assess neighborhood-scale traffic conditions, and analysis was limited to about 3 to 4 months of data.

The objective of this paper is to assess potential trends in traffic congestion that may have occurred between 2017 and 2019 in the Mott Haven and Port Morris portion adjacent to the HRY in NYC (see Figure 1b), both at the street level and the neighborhood scale. Our analyses are also based on the premises that traffic in the HRY may affect traffic in the entire area and that some of the potential changes are attributable to the opening of the warehouse. Since it is not realistic to assess neighborhood-scale traffic with traditional physical measurements using, e.g., pneumatic tube counters or radar devices, we instead analyzed hourly time series of crowd-sensed traffic congestion maps we collected from June 2017 to June 2019. Our analyses show that traffic congestion significantly increased within a large portion of our study area.

METHODS

Image acquisition and analysis

From June 8, 2017 to June 3, 2019, we used a Unix “cron” job to acquire automatically a traffic congestion map every hour from a Google traffic server (Hilpert et al. 2019).

Downloaded traffic maps were true color images. As described in Section SI-B, we segmented these images into indexed images with only five possible colors. A pixel was assigned an integer number between 1 and 4 according to Equation (1) if the pixel carried one of the four congestions colors, maroon, red, orange or green. A pixel was assigned a value of 5 when a color was detected that can serve as background in the image without traffic layer (e.g., white or gray for neighborhood streets). Otherwise, pixels were assigned to be NaN (Not a Number) to indicate misclassification.

For a portion of the downloaded traffic maps (delineated by the shaded areas in Figure 1b), we created a network (Hilpert et al. 2000) consisting of nodes and branches where the branches represent road segments shown on the traffic map. The nodes typically represent intersections with more than two branches meeting at a node, or points along roads where CCC can change. In Section SI-C, we describe in detail how we determined the CCC time series for the branches of the network.

Analyzed regions of interest

Temporal changes in congestion were analyzed separately in the following four neighborhood-scale areas (Figure 1b): 1) the mixed-use area I to the South of I-87 and West of I-278, 2) the mixed-use area II to the South of I-278, 3) the residential area to the North of both I-87 and I-278, and 4) the interstates and their ramps. We did not analyze traffic to the North of 138^th St. to obtain about equal sizes of the residential and two mixed-use areas.

We examined street-level changes in traffic for 11 selected stretches of road, which are delineated by black dashed rectangles in Figure 1a. Road Stretches EA1 to EA6 were identified in the Environmental Assessment of the grocery store warehouse as routes of incoming or outgoing traffic into or out of the HRY (AKRF Inc. 2011). However, we also examined traffic along Road Stretches P1 to P4 since we suspected they might also experience warehouse-related traffic. Road Stretch U1 is unrelated to any incoming or outgoing traffic unless drivers miss a turn.

Analysis of congestion time series

We examined changes in traffic over the two-year study period at three scales by analyzing different size portions of the entire road network: 1) in each of the ~300 branches to visualize changes in traffic at the finest scale, 2) the four regions delineated in Figure 1b to examine traffic changes at the neighborhood scale, and 3) the stretches of road shown in Figure 1a to examine traffic along selected traffic routes. For each portion of the network, we determined an average time series $\overline{\text{CCC}}(t)$ weighted by the length of the branches making up the network portion:

$$\overline{\text{CCC}}(t)=\frac{{\sum }_i{L}_i{\text{CCC}}_i(t)}{L}$$ (2)

where CCC_i is the time series for branch i, L_i is the length of branch i, and L = ∑_i L_i is the total length of the digitized portion of the road network.

For each examined portion of the network, we fitted the following model to the two-year long hourly $\overline{\text{CCC}}$ time series:

$$\overline{\mathrm{CCC}}(t)=a_0+{\sum }_{n=1}^2\left[a_n\cos \left({\omega }_{n}t\right)+b_n\sin \left({\omega }_{n}t\right)\right]-\gamma t$$ (3)

where the sum and the a₀ terms represent a truncated Fourier series to capture seasonal effects in traffic, ω_n = 2πn/1yr, and γ represents the trend in traffic. If γ > 0, congestion increases; if γ < 0, congestion decreases. Hence γ with units of yr⁻¹ can be interpreted as an annual change in congestion. Like in our previous study, in which we used traffic radar devices to asses traffic conditions in the same study area, we stratified our analysis by time of day and divided days into eight non-overlapping 3-hour windows (Shearston et al. 2020).

All analyses were performed using Matlab^™ version R2020b.

Traffic data overview

For each of the 292 branches of the road network, we carefully examined the CCC time series for the purposes of quality control. See Section SI-D for more details. Briefly, the hourly time series of traffic maps contained missing data, because map download was not always successful. Moreover, for an about two-month period of time (June 21—August 31, 2018), the left half (42%) of downloaded traffic maps was blank. Overall, the CCC time series of the 114 branches in the left portion of the maps contains 10.3% of missing data while only 0.14% are missing for the 178 branches in the right portion. Among the time series of valid downloaded traffic images or valid right-image halves, our algorithm for identifying CCC of branches failed on average 0.006% and 0.005%, respectively, to identify a CCC between 1 and 5. Failure could, e.g., occur when temporary symbols were displayed by the traffic data provider on top of the branches.

We assumed that the traffic data provider did not change the parametrization of the color assignment algorithm on road segments with no speed limit change. To examine potential changes, we performed two-year long traffic radar measurements at two locations differing substantially in achievable vehicle speed. There was no indication of changes in the algorithm over the course of the study that would bias study results (see Section SI-F for more details). Also, we are not aware of any permanent speed limit changes in our study area over the course of the study that could have potentially affected the algorithm. However, such changes would be of lesser concern, because then congestion colors determined by a reparameterized algorithm would rather represent the effects of traffic control on congestion than introduce a bias.

Diurnal fluctuations in warehouse related traffic

To examine whether changes in congestion were related to warehouse traffic, we analyzed the hourly time series of traffic coming into and going out of the warehouse as projected in the Environmental Assessment from 2011 to compare them to the $\overline{\text{CCC}}$ time series we obtained from the crowd-sensed traffic data. The assessment separately presented incoming and outgoing traffic data for delivery trucks and employee vehicles and stratified the data by weekday/weekend. We added up incoming and outgoing total traffic (the sum of truck and employer vehicle counts) and averaged the hourly data over the three-hour time windows, for which we determined $\overline{\text{CCC}}$ time series. Figure 2 shows the resultant time series.

Figure 2:

Dependence of warehouse-related incoming and outgoing traffic on time of day stratified by weekday and weekend. Data source: Environmental Assessment of the warehouse.

RESULTS

Traffic changes at the neighborhood scale

Table 1 summarizes the neighborhood-scale annual change in congestion, γ. In the two mixed-use areas, overall congestion essentially increased significantly for any time of the day, except for the 9:00–12:00 time window. The greatest increases occurred at night and in the morning (21:00—9:00). In the residential area, congestion increased significantly at night (0:00—6:00), while no significant change occurred during the remainder of the day. On the interstates and their ramps, congestion decreased significantly during the daytime and in the evening (9:00–21:00) but increased significantly during the night-time (21:00–6:00).

Table 1:

Neighborhood-scale annual change in traffic congestion color code, γ (95% confidence interval) [yr⁻¹], in the four regions shown in Figure 1b. Negative and positive values indicate increased and decreased congestion, respectively.

Time window	Mixed-use zone I	Mixed-use zone II	Residential zone	Interstates
0:00–3:00	0.12 (0.10, 0.13)	0.08 (0.07, 0.10)	0.02 (0.01, 0.04)	0.09 (0.08, 0.10)
3:00–6:00	0.18 (0.17, 0.19)	0.08 (0.07, 0.09)	0.06 (0.05, 0.07)	0.10 (0.10, 0.11)
6:00–9:00	0.07 (0.04, 0.11)	0.10 (0.07, 0.14)	0.02 (−0.00, 0.05)	−0.04 (−0.09, 0.00)
9:00–12:00	0.02 (−0.00, 0.04)	0.06 (0.03, 0.09)	−0.00 (−0.02, 0.01)	−0.07 (−0.10, −0.03)
12:00–15:00	0.03 (0.02, 0.05)	0.07 (0.04, 0.09)	−0.01 (−0.02, 0.00)	−0.12 (−0.15, −0.08)
15:00–18:00	0.04 (0.02, 0.06)	0.06 (0.03, 0.08)	−0.01 (−0.03, 0.01)	−0.14 (−0.19, −0.08)
18:00–21:00	0.03 (0.01,0.04)	0.05 (0.03, 0.07)	−0.01 (−0.02, 0.00)	−0.13 (−0.17, −0.09)
21:00–24:00	0.09 (0.08, 0.11)	0.13 (0.11,0.14)	−0.00 (−0.01, 0.01)	0.08 (0.07, 0.10)

Note: the three reddish colors (orange, red, maroon) in this order indicate increased congestion with colors based on the tertiles of the distribution of positive γ values shown in the table. The three greenish/blueish colors (green, turquoise, blue) in this order indicate decreased congestion with colors based on the tertiles of the distribution of negative γ values shown in the table. Black: change not significant.

Figure 3 visualizes congestion changes in the entire road segment network, with branches colored according to their γ value for the midnight (0:00—3:00) and late afternoon (15:00—18:00) time windows. In the Supplementary Information, we show in Figure SI-1 changes for all eight time windows of the day. The figures corroborate findings from Table 1, showing that congestion increased on average in the two mixed-use areas, but more so at night. In contrast, congestion on the interstates improved during the daytime (15:00—18:00); however, at night (0:00—3:00) congestion increased on the interstates. In the residential area, congestion changes are spatially variable, and an overall trend is less clear, although Table 1 shows an overall congestion increase at night.

Figure 3:

Annual change in congestion, γ [yr⁻¹], of all road segments in the road network and for two time windows: 0:00–3:00 and 15:00–18:00. The color of a road segment indicates the value of γ according to the color bar. Greenish/blueish colors indicate significant congestion decreases, and orangish/reddish colors significant increases. The γ thresholds in the color bar represent the tertiles of the two distributions of positive and negative γ values. Road segments with non-significant changes are shown in gray. The four regions of interest were separated for better visualization of interstate traffic. Base map from OpenStreetMap.

To illustrate absolute levels of congestion as well as their changes, we show in Figure SI-2 in Supporting Material the road segment network with branches colored according to the CCC values modeled according to Equation (3) at beginning and the end of the study.

Traffic changes at the road stretch level

Table 2 summarizes congestion changes along the stretches of road delineated in Figure 1a. For the most part, congestion increased along the road stretches identified in the Environmental Assessment of the grocery store warehouse as routes of incoming or outgoing traffic. For all road stretches except EA1, congestion increased significantly in the night time (21:00—6:00). At three stretches (incoming traffic on EA2 and EA3, and outgoing traffic on EA4), congestion increased significantly throughout the day. During the day time, congestion decreased on some road stretches, e.g., for outgoing traffic on EA2 (12:00—15:00) as well as incoming traffic on EA4 (9:00—12:00) and EA5 (9:00—15:00, 18:00—21:00). Road stretch EA1 is the only stretch where congestion decreased during the night time (0:00—3:00).

Table 2:

Street-level annual change in traffic congestion color code, γ [yr⁻¹], for road stretches shown in Figure 1a. “i” and “o” for traffic direction indicate incoming and outgoing traffic with regard to the HRY.

	Routes from Environmental Assessment								Other routes to/from HRY				Other
Road stretch	EA1	EA2	EA2	EA3	EA4	EA4	EA5	EA6	P1	P2	P3	P4	U1
Traffic direction	o	i	o	i	i	o	i	i	i	i	o	o	-
0:00–3:00	−0.02	0.25	0.23	0.27	0.22	0.54	0.11	0.11	0.28	0.17	−0.01	0.22	0.01
3:00–6:00	0.01	0.29	0.30	0.23	0.45	0.86	0.40	0.28	0.56	0.51	−0.01	0.12	−0.00
6:00–9:00	0.08	0.22	0.07	0.26	0.18	0.36	0.03	0.10	0.23	0.09	−0.11	0.18	−0.26
9:00–12:00	0.11	0.10	−0.05	0.30	−0.06	0.15	−0.09	−0.03	0.20	−0.00	−0.08	0.11	−0.46
12:00–15:00	0.12	0.12	−0.07	0.25	0.03	0.23	−0.04	0.02	0.21	0.07	−0.10	0.10	−0.65
15:00–18:00	0.11	0.11	−0.01	0.24	0.01	0.22	−0.01	0.02	0.22	0.16	−0.15	0.15	−0.57
18:00–21:00	0.05	0.08	−0.04	0.21	0.08	0.20	−0.02	−0.00	0.15	0.02	−0.06	0.20	−0.49
21:00–24:00	0.05	0.14	0.07	0.18	0.13	0.29	0.09	0.18	0.20	0.06	−0.01	0.20	−0.05

Note: the three reddish colors (orange, red, maroon) in this order indicate increased congestion with colors based on the tertiles of the distribution of positive γ values shown in the table. The three greenish/blueish colors (green, turquoise, blue) in this order indicate decreased congestion with colors based on the tertiles of the distribution of negative γ values shown in the table. Black: change not significant.

Among all road stretches and time windows examined, the greatest congestion increase occurred on EA4 between 3:00 and 6:00 for outgoing traffic and was γ = 0.86. Figure 4a shows the corresponding time series. While the one-week moving average of $\overline{\text{CCC}}$ was equivalent to a green or gray congestion color until the beginning of 2018, congestion increased sharply around April 2018 as indicated by orange and even red congestion colors. Thus, traffic went from on average free flowing to congested on EA4.

Figure 4:

Sample time series (1-week moving averages) of congestion scolor code CCC for Road Stretches EA4 and U1 and fits according to Equation (2).

On three of the four road stretches the Environmental Assessment did not identify as warehouse-related traffic routes (P1, P2, P4), congestion increased significantly throughout the day (except at P2 between 9:00—12:00 and 18:00—21:00). At P3, however, congestion overall decreased, with significant decreases during the day (6:00—21:00) and non-significant changes at night (21:00—6:00).

On Road Stretch U1, which was likely not associated with HRY traffic, congestion decreased significantly for most of the eight time windows (6:00—24:00). Congestion decrease was greatest between 12:00—15:00 with γ = −0.65. Figure 4b shows the corresponding time series. While in 2017, the 1-week moving average congestion color was red and at times orange, traffic flow dramatically improved at the end of 2017, and congestion colors turned into orange or even green.

DISCUSSION

Changes in traffic

Our study provides strong evidence suggesting that congestion in Mott Haven and Port Morris in the South Bronx, NYC increased from 2017 to 2019. Congestion increased significantly throughout the day in the mixed-use areas, with the greatest increases occurring at night. This observation is consistent with our previous study in which we deployed traffic radar devices in the mixed-use areas (Hilpert et al. 2019). In the residential area, congestion increased significantly during the night (0:00–6:00), potentially resulting in increased traffic-related sleep disturbances. On the interstates cutting through the neighborhood as well as their ramps, however, we identified significant improvements in traffic conditions during the daytime, although congestion increased at night.

The grocery store warehouse which opened in 2018 likely contributed to the overall increase in traffic congestion in the entire study area. However, the effects were more pronounced in the mixed-use areas. Interestingly, the greatest increase in congestion in mixed-use area I (γ = 0.18, see Table 1) occurred in the early morning (3:00–6:00), and total traffic as predicted by the Environmental Assessment also exhibits a local maximum for this time window on both weekdays and weekends (Figure 2). However, the congestion data for neither the mixed-use areas nor the residential zone reflect the absolute maximum in warehouse traffic in the afternoon (12:00–15:00). Even though it is possible that the projection from the Environmental Assessment is outdated (submitted in 2011), we believe that the diurnal dependence of neighborhood-scale changes in traffic congestion cannot be expected to reflect the diurnal dependence of both the warehouse traffic that was projected and that actually occurred, because the warehouse has not been the sole additional traffic source added within the study period (2017–2019).

Nonetheless, warehouse related traffic can be expected to have contributed substantially to congestion. Indeed, the added traffic with, according to the Environmental Assessment, about 1,000 trips per day is not negligible when compared to annual average daily traffic (AADT) counts. For instance, in 2017 the US Department of Transportation (DOT) (NYSDOT 2019) measured on a portion of St. Ann’s Ave., which includes Road Stretch EA4, an AADT of 7,945 counts (both traffic directions). On Willow Ave., which includes Road Stretch EA6, we performed measurements to estimate an AADT of 1,500 (Shearston et al. 2020). However, other sources including residential developments in Mixed-use area I likely also contributed to the congestion (South Bronx Unite 2021).

Environmental Assessment of the warehouse

Our study shows that congestion increased on road stretches that had been identified as routes of incoming or outgoing traffic related to the grocery store warehouse in the Environmental Assessment. However, congestion also increased significantly on Road Stretch P2 along a route not identified as a route for incoming traffic (blue dashed line in Figure 1a), representing Eastbound traffic on I-87 taking Exit 1 to reach the main entrance of the HRY. It does not make much sense to exclude this route, because drivers planning to take Exit 2 via EA3 as anticipated in the Environmental Assessment may favor Exit 1 if Exit 2 is more congested than Exit 1.

The increased congestion we observed on P2 is supported by AADT counts conducted by DOT (NYSDOT 2019). In 2015 and 2018, Eastbound traffic on P2 between Brook Ave. and St. Ann’s Ave. had an AADT of 3,228 and 5,520, respectively. Even though DOT only performed one measurement within our study period (2017—2019), the substantial AADT increase from 2015 to 2018 is consistent with our observation of increased congestion and is additionally indicative of an overall increase of incoming traffic into the HRY.

The incoming traffic route, which was projected to bring Eastbound traffic on I-87 into the secondary entrance of the HRY, does not exist on the 132^nd St. portion (indicated by blue dotted line in Figure 1a). We suspected that the Environmental Assessment omitted the outgoing traffic route indicated by the purple dashed line in Figure 1a. However, congestion decreased significantly on Road Segment P3, supporting that the Environmental Assessment rightfully omitted this route.

Our approach allowed identifying likely inaccuracies in the Environmental Assessment with regard to routes of incoming and outgoing traffic. It is not clear whether one major finding of the Environmental Assessment, that warehouse-related traffic would have no adverse air quality impacts, would have changed with revised traffic routes. That finding, however, conflicts with our previous study, in which we estimated increases in levels of atmospheric black carbon, a tracer for traffic-related air pollution, due to the warehouse opening (Shearston et al. 2020).

Community impacts

Traffic as well as associated traffic-related air pollution and noise have long been a concern to the Mott Haven and Port Morris community, starting perhaps with the construction of the Major Deegan and Bruckner Expressways almost eight decades ago (Caro 1975). Since then, numerous traffic-intensive operations have been added near or at the South Bronx waterfront. In this study, we examined changes in traffic conditions in this South Bronx community between 2017 and 2019, a period of time which included the opening of a large delivery warehouse in the HRY along the South Bronx waterfront.

We found that congestion increased significantly throughout the day in the mixed-use areas next to the HRY, and also significantly at night in a nearby, highly populated residential area. Congestion increased along road stretches used by pedestrians and cyclists to access recreational areas on Randall’s Island, potentially increasing the risk of traffic-related injury.

Some of the overall changes in traffic congestion in the Mott Haven and Port Morris neighborhoods that occurred between 2017 and 2019 can be attributed to the online grocery store warehouse, a finding also supported by our previous study, in which we analyzed time series of traffic radar data (Shearston et al. 2020). Perhaps traffic in the residential area is affected less than in the mixed-use areas, because it is further from the HRY and, according to the Environmental Assessment, only 10% of warehouse-related traffic (for both trucks and employee vehicles) is directed through it while 100% pass the mixed-use areas. Even though the warehouse is not responsible for the overall congestion increase we observed, its construction has drawn particular ire from community members, because it denied the community access to their waterfront. Of further aggravation, the online grocer initially accepted online orders from neither residences in Mott Haven and Port Morris nor recipients of food stamps in an area marked by concentrated poverty and heavy reliance of food-related aid.

Finally, due to the high rate of pedestrian injury hospitalizations in the neighborhood (Hinterland et al. 2018), it would have been important to evaluate how warehouse traffic would affect access to the recreational waterfront areas on Randall’s Island via pedestrian paths along the interstate bridge or the Randall’s Island Connector (see green arrows in Figure 1a). The Environmental Assessment, however, did not consider pedestrian and bicycle safety. Our study found increased congestion on two road stretches that pedestrians/bicyclists would typically use to access the pedestrian/bike paths to Randall’s Island: on EA4 throughout the day and on EA5 at night. Noteworthy there is not a single pedestrian light on EA5 including its intersection with EA4, denying safe access to Randall’s Island.

It is unclear whether the benefits of the warehouse, primarily jobs created in the community, outweigh its harms and the missed opportunity to create recreational space on the waterfront of Mott Haven and Port Morris.

Methods development

In previous work (Hilpert et al. 2019), we used crowd-sensed traffic congestion maps for studying associations with traffic radar data and air pollution levels at two selected road segments. Recently Zalakeviciute et al. (2020) examined associations between Google traffic data, vehicle speed from odometer readings of a driven vehicle, and air quality measurements made with mobile sensors. Here we demonstrate that crowd-sensed traffic data can also be used to evaluate changes in traffic conditions at the neighborhood scale, which is difficult to do with traditional devices such as traffic radar or pneumatic counters because of their expense. We also acquired these traffic data over a rather extensive period of time (two years). This period of time is much longer than the typical 2- to 7-day deployments of traffic radar devices that the DOT uses to derive AADT counts (using day-of-week and seasonal factors) (US Department of Transportation Federal Highway Administration October 2016).

When compared to approaches which assess congestion from Google travel time data between different points (Kim and Kwan 2018, Zafar and Ul Haq 2020), our method has the advantage that a traffic network can be constructed more easily (because routes between points may change depending on traffic conditions).

Limitations

Several factors could potentially affect Google traffic data availability and our conclusions: 1) community demographics (e.g., socioeconomic status or racial/ethnic composition) could affect firstly availability of a smartphone at all and secondly the traffic data provider used (e.g., Google or Apple); 2) not all drivers may be willing to share their GPS positions with Google, and commercial drivers may not be allowed to use their personal smartphones; and 3) the fraction of drivers sharing their GPS position may depend on time of day and day of week, because the ratio of commercial versus private vehicles depends on these factors. Occurrence of the gray congestion color may thus depend on neighborhood characteristics which could have changed over the course of the study. However, we expect this to be of lesser concern, because this only affects the transition between gray and green congestion colors but not the other three transitions ranging from maroon to green.

It is possible that Google changed the parameterizations of algorithms for assigning congestion colors to individual road segments from measured smartphone/vehicle speeds during the course of our 2-year study, potentially impeding interpretation of changes of congestion color over time. To examine this possibility, we analyzed traffic radar data that were collected for a period of two years (much longer than the duration of a typical DOT AADT campaign) at two road segments and found no evidence that Google changed parameterizations. However, we believe that even if Google had changed parameterizations, observed changes in congestion color could still be indicative of real changes in congestion, because a change in parameterization probably occurred for a reason, e.g., a permanent speed limit change, in which case a change in congestion color would indicate a change in congestion, i.e., a difference between actual and achievable vehicle speed.

Conclusions

Innovative use of crowd-sensed traffic congestion maps allowed us to study traffic both at the neighborhood scale and the street level in Mott Haven and Port Morris — environmental justice areas in the South Bronx in NYC. We found overall increases in traffic congestion in mixed-use areas and a highly populated residential zone between 2017 and 2019. This is of concern to the community, because traffic-related air pollution contributes to the high asthma prevalence and because of the high rate of traffic-related pedestrian injuries in the study area. Moreover, the trend of increasing congestion could potentially represent an environmental health disparity as other more affluent areas in NYC are planning interventions to reduce congestion, e.g., the “congestion pricing” policy which would discourage drivers from entering the “Central Business District” in Manhattan, NYC (Wamsley 2019).

Our analysis did not show that neighborhood-scale changes in congestion can solely be attributed to traffic related to the online grocery store warehouse. However, the increased congestion in mixed-use area I we observed during night-time hours is consistent with our previous work, in which we used interrupted time series analysis of traffic radar data to associate increases in traffic at two locations with the opening of the warehouse (Shearston et al. 2020).

The methods used in this study can be applied in many cities globally (Hilpert et al. 2019, Shearston et al. 2021) as long as crowd-sensed traffic data can be acquired. Studies like ours can potentially empower local communities to advocate for measures mitigating the environmental health impacts of Traffic. Moreover, such studies could be used to extend the scope of environmental assessments and impacts statements.

Highlights.

1. We assessed traffic congestion in an environmental justice community over 2 years

2. We analyzed time series of crowd-sensed traffic maps

3. Traffic increased in the South Bronx in New York City between 2017 and 2019

4. The local community will use the data to advocate for mitigating measures

5. Crowd-sensed traffic data allow assessing community impacts of traffic globally