2.0 methodology for vehicle los monitoring los monitoring...2.0 methodology for vehicle los...

Task C4 Memorandum: Auto LOS Monitoring

Iteris, Inc. | 1

To: Bhargava Sana / SFCTA

From: Kavya Sambana / Iteris, Inc. Haley Zhao / Iteris, Inc.

Date: September 27, 2017

Subject: Memorandum #1: 2017 Auto LOS Monitoring Results

1.0 INTRODUCTION This memorandum provides the results for Task C4 – LOS Monitoring Data Analysis. It also describes the methodology for filtering and processing the commercial speed data to obtain the average speed and Level of Service (LOS) for each Congestion Management Program (CMP) segment in San Francisco County. The commercial speed data were supplemented by floating car survey data where commercial speed data were not available. The 2017 monitoring results are presented in the last section.

2.0 METHODOLOGY FOR VEHICLE LOS MONITORING This section describes the methodology for obtaining the average speed and LOS information for CMP segments in San Francisco County.

A. Monitoring Times The LOS analysis was conducted for San Francisco County starting on April 4, 2017 and ending on May 16, 2017. The monitoring was conducted on Tuesdays, Wednesdays and Thursdays. This left 19 days for monitoring. The morning peak period was from 7:00 a.m. till 9:00 a.m. and the afternoon peak period was from 4:30 p.m. to 6:30 p.m.

i. Public holidays and school breaks

While there were some public holidays during the spring of 2017, none occurred on Tuesdays, Wednesdays and Thursdays. Local schools were also in session during this period.

ii. Special Events

Similar to 2015 monitoring, major events in San Francisco County were reviewed to see if they occurred during the Tuesday, Wednesday, and Thursday peak periods. The majority of events did not occur within the monitoring times. SF Giants games were the notable exception. Games started at 12:45 p.m. or at 7:15 p.m. Both of these timeslots were deemed to impact on the afternoon peak period. However, due to the frequency of these events, the data collected from these days were retained in the dataset (Figure 1).


Iteris, Inc. | 2

Figure 1: Planned events in San Francisco County, Spring 2015

iii. Construction Events

Community service announcements were reviewed to identify significant construction impacts during the spring monitoring period. Sources of data included:

• Government websites (including SF Public works); • Specific construction project websites (including Central Subway and the Transbay Center); • Social Media feeds (including 511 SF Bay traffic updates); and • PeMS lane closure database.

Both long term and short term events were investigated. Short term construction or maintenance events include events that had a short duration impact on the CMP segment. INRIX data collected during the work could be identified and excluded from the analysis, and there would still be enough remaining data to successfully record the performance of the CMP segment. In the 2017 analysis, one short term event was identified along Post Street between Franklin Street and Van Ness Avenue. There was a street closure at this location from Thursday, April 6 2017, 12:01 pm to Monday, April 10 2017, 6:00 am (Van Ness Campus Hospital Project). Given the short duration, the data for the affected CMP segments on these days were removed from the monitoring data. It was deemed that enough data remained to gain an appropriate sample size for monitoring. Additionally, the following segments experienced major and ongoing construction throughout the entire monitoring period. In these instances, even on the segment that remained open, there would not be enough alternative days to provide a suitable sample size if all days impacted by construction were removed. Therefore, corresponding data was retained in the analysis. Segments impacted by ongoing construction and maintenance are listed in Table 1.

Table 1: Long-term construction and maintenance projects active during Spring LOS monitoring

Description Corresponding Impacted Roads CMP Segments

Potrero Avenue Roadway Improvement Project

200: Potrero NB from Cesar Chavez to 21st Street 201: Potrero NB from 21st Street to Division


Iteris, Inc. | 3

Description Corresponding Impacted Roads CMP Segments

202: Potrero SB from Division to 21st Street 203: Potrero SB from 21st Street to Cesar Chavez

101 NB and SB between Mission St and Lombard St

222: Van Ness/S. Van Ness NB from 13th to Golden Gate 223: Van Ness/S. Van Ness NB from Golden Gate to Washington 224: Van Ness/S. Van Ness NB from Washington to Lombard 225: Van Ness/S. Van Ness SB from Lombard to Washington 226: Van Ness/S. Van Ness SB from Washington to Golden Gate 227: Van Ness/S. Van Ness SB from Golden Gate to 13th

Transbay Transit Center 136: Howard WB from Embarcadero to S Van Ness

iv. Weather Events

There was one significant weather event observed during the monitoring period. On Thursday April 6, there was light to heavy rain observed during the PM monitoring period (source: Weather Underground Historical Weather Data www.wunderground.com). Therefore, data was removed in the analysis.

B. Processing Similar to 2015, 2017 automobile mode LOS monitoring was conducted using commercial speed data from INRIX where available, and floating car data for all other CMP segments. This section describes the steps performed to obtain automobile LOS for each CMP segment for the morning and afternoon peak periods.

Figure 2: Data Processing Steps

i. Step 1: Review GIS Files

The 2015 GIS files were reviewed to ensure they represented the 2017 CMP road network. Based on the

3a. Process INRIX data Filter and aggregate

1. Review GIS Files

Update for 2015

2. Confirm Mapping

TMCs to CMPs

4. Compute LOS

HCM 1985 and 2000

3b. Collect / Process Field

data Floating Car

Surveys

http://www.wunderground.com/


Iteris, Inc. | 4

review, no changes were identified and the 2015 shape file was used as-is for the 2017 analysis. However, Iteris noted that private cars are restricted for through movement along Market Street in the east bound direction at the following locations:

1. Between 10th and 9th streets; and 2. Between 6th street and 5th street.

This affects CMP segment 156: Market/Portola: Van Ness to Drum. Note that this segment is currently monitored using INRIX data. Iteris requests SFCTA to review this segment closely and provide feedback if any changes need to be made for the CMP segment description/GIS file.

ii. Step 2: Mapping TMC links onto CMP segments

INRIX reports travel speeds on roadway segments called Traffic Message Channel (TMC) links or XD links. TMC links are available through the MTC agreement with INRIX. TMC links are typically short links of roadway averaging 0.23miles in length (range: 17.59 feet to 2.04 miles)1. For this project, it was required that the average speed be reported against a SFCTA CMP segment. CMP segments are typically longer segments of roadway averaging approximately 1 mile in length (range: 0.18 to 4.3 miles). Therefore, TMC links needed to be aligned against or mapped onto the CMP segments. This task was completed in 2013/2015 and updated for the 2017 analysis. Iteris completed the mapping and performed the following quality control processes:

Distance: For each CMP, the length of that CMP was compared to the total length of TMCs assigned

to the CMP. On a small minority of segments, differences were encountered. The reasons for

differences included:

a. The length of the CMP segment recorded by SFCTA was approximate.

b. CMP locations with low percentage of INRIX data points (less than 20%). In these cases, the

floating car runs were completed in May 2017.

Direction: Latitude and longitude coordinates of each TMC were compared with the CMP direction to

confirm that the TMC links chosen were in the correct direction.

Visual inspection: Working maps of each TMC and CMP segment were produced to visually check the

mapping. This ensured that there were no gaps in the TMCs mapped and there were no

typographical errors in choosing TMCs.

It should be noted that for some CMP segments, the ends of the CMP did not align with the ends of the TMCs. Figure 3 shows a schematic example for explanatory purposes. It shows one CMP segment that is made up of four TMC links. However, the end of the last TMC link does not align with the end of the CMP segment. In these instances, Iteris used only the overlapping portion of the TMC length to calculate the average speed.

1 TMC length statistics are based on TMCs used in this monitoring project.


Iteris, Inc. | 5

Figure 3: End points of CMP and TMC do not align

Although the mapping performed for the 2015 analysis was used as a starting point for 2017 analysis, there were significant changes to TMC begin or endpoint locations between 2015 and 2017. The 2017 mapping required a thorough review of TMC links for each CMP segment. For many CMP segments, there were additions or deletions of TMC links compared to the 2015 mapping.

iii. Step 3a: Processing INRIX Data

INRIX reports travel speeds on roadway segments called Traffic Message Channel (TMC) links or XD links. TMC links are available through the MTC agreement with INRIX. This step consists of filtering and aggregating the INRIX data as described below.

Filtering

The raw INRIX data were filtered to remove:

Times outside the morning and afternoon peak periods, as defined above;

Days other than Tuesdays – Thursdays;

Data points impacted by construction and special events, as applicable; and

Data points with lower data quality scores.

INRIX includes a data quality score that accompanies every INRIX data point. The score value is defined as:

Score of 30: INRIX data are exclusively generated from observed real-time sources.

Score of 20: A mix of historical and real-time sources are used.

Score of 10: INRIX data are exclusively generated using historical data.

For this analysis, Iteris removed all data points with a score of 20 or 10; leaving observed data only. The sample size was recorded.

Spatially and temporally aggregate the data - Compute average speeds

This section discusses the methodology of aggregating the data both spatially and temporally. The input to this section was around 4 million data points of INRIX speed data. An example of this data is included in Table 2. The output from this step was the average speed and sample size on each CMP segment. A sample of this data is included in Table 3.

Table 2: Sample INRIX input data

TMC Code Time Stamp Speed Travel Time Score

105P08351 04/09/2017 07:00 27 0.22 30

105+07766 04/09/2017 07:01 49 1.71 20

Table 3: Sample output – Average speed on CMP link

CMP

TMCs 100% 100% 100% 60%


Iteris, Inc. | 6

ID Description Distance (mi)

AM / PM

Sample size

Average Speed (mph)

1 1st St (South): Market to Harrison 0.48 AM 2079 12.8

PM 2050 4.3 The following steps describe how the dataset was manipulated to obtain the results in Table 3. This involved spatial and temporal aggregation.

Spatial aggregation

Using the mapping file and the filtered INRIX data, the TMC data were spatially aggregated on the CMP segments by summing the travel time on each contributing TMC segment. For each TMC, distance-weighting was used to compute the travel time of the overlapping portion. The CMP segment travel time is obtained for each one-minute time period.

𝐶𝑀𝑃 𝑇𝑟𝑎𝑣𝑒𝑙 𝑇𝑖𝑚𝑒 = 𝑇𝑀𝐶1 + 𝑇𝑀𝐶2 + ⋯ + 𝑇𝑀𝐶𝑛

Temporal aggregation

Temporal aggregation involved the translation of the CMP travel time metric for each minute of data into one average speed value corresponding to each CMP segment for the entire monitoring period. The following formula was used for this2:

𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐶𝑀𝑃 𝑆𝑝𝑒𝑒𝑑 =∑ 𝐶𝑀𝑃 𝐿𝑒𝑛𝑔𝑡ℎ

∑ 𝐶𝑀𝑃 𝑇𝑟𝑎𝑣𝑒𝑙 𝑡𝑖𝑚𝑒

Sample size information was retained to assess the confidence in the input data.

Discussion about sample size

The sample size is a measure of the number of aggregated one-minute data points that contributed to the final calculation of average speed. The sample size varied on each TMC through removal of data points during the filtering process and through the processes discussed below. Removal of TMC data points with scores of 20 and 10 eliminated data for particular one-minute time periods from one or more of the TMCs that comprise certain CMP segments. The example shows a longer CMP segment which is comprised of four TMCs. The table shows the data scores for each TMC for each one minute time period. In time periods 1, 2, and 7, one of the TMCs had a data score of 20 and therefore the record from that TMC was excluded for those minutes. In time period 6, two of the TMCs had data scores of 20 and similarly, these TMC records were also excluded for time period 6 (Figure 4).

2 FHWA, Travel Time Data Collection Handbook http://www.fhwa.dot.gov/ohim/tvtw/natmec/00020.pdf

http://www.fhwa.dot.gov/ohim/tvtw/natmec/00020.pdf


Iteris, Inc. | 7

Figure 4: Example of Filtering Process

Iteris performed a check to ensure that any time periods that had too many TMCs removed were not included in the analysis. Where TMC data were available for less than 99% of the total mapped TMCs, that one-minute time period was removed. To extend the above example further, if TMC1 was less than 1% of the CMP segment length, then it would still be possible to use the data in Time periods 1 and 2 (in addition to time periods 3, 4 and 5). This can be justified, because TMC1 does not contribute significantly to the distance-based average speed calculation. In a small minority of cases, using the 99% threshold resulted in too many removed time periods and an inadequate sample size. In these cases, the threshold was lowered to 70% to ensure that the sample size was adequate. A minimum sample size of 180 was chosen, similar to 2015 monitoring cycle. The remainder of this section gives information about the sample sizes observed on all CMPs. Note that there are 245 CMP segments each having a morning and afternoon measurement of average speed. This totals 490 measurements. The following graph shows a frequency plot of the sample sizes obtained for each CMP (morning and afternoon recorded separately). Note that the floating car surveys had a sample size of four and are highlighted in grey below. Refer to Figure 5.

CMP

TMCs


Iteris, Inc. | 8

Figure 5: Frequency of sample sizes for each CMP

iv. Step 3b: Collect and Process Floating Car Survey Data

Floating car surveys were conducted on CMP segments without TMC coverage. The surveys were conducted using conventional methodologies. Drivers were instructed to follow road rules including the speed limit, traffic signals and not blocking intersections. GPS coordinates were recorded as the floating car travels along the CMP segment. The timestamps from the GPS coordinates at the beginning and end of the CMP segment were identified and the CMP travel time was calculated. The temporal aggregation of multiple floating car runs on the same CMP segment was performed in the same manner as for the INRIX data, explained in Step 3a above. CMP segments surveyed with floating car surveys were identified with a note in the results

v. Step 4: Compute and Report Level of Service (LOS)

This section discusses the methodology for assigning a LOS (A to F) to each CMP segment for both of the morning and afternoon peak periods. Firstly, each CMP segment was classified as either an arterial or a freeway. The methodology slightly differs depending on this classification, as follows. The LOS assignments for arterials and freeways are consistent with previous reporting periods and legislative requirements from the California Government Code.

Arterials

LOS for arterial segments was assigned twice using both 1985 and 2000 Highway Capacity Manual (HCM) methodologies. Both of these methods required identifying the class of the street (HCM 1985 Class I, II or III; HCM 2000 Class I, II, III or IV). Class was determined according to the free flow speed of the road. For example, the free flow speed may be considered to be the average speed at 6am when traffic volumes are light and travel speeds are not influenced by interactions with other vehicles. For the HCM 1985 and 2000, the classification of streets was taken from previous LOS monitoring reports. Then, by knowing the average travel speed in the morning and afternoon peak periods and the class of the


Iteris, Inc. | 9

street, the LOS could be assigned according to the HCM 1985 and HCM 2000 methodologies. Refer to Table 4 and 5 for the LOS look up tables.

Table 4: Arterial LOS Assignment, HCM 1985

Arterial Class I II III

Range of Free Flow Speed (mph) 45 to 35 35 to 30 35 to 25

Typical Free Flow Speed (mph) 40 33 27

Level of Service Average Travel Speed (mph)

A ≥ 35 ≥ 30 ≥ 25

B ≥ 28 ≥ 24 ≥ 19

C ≥ 22 ≥ 18 ≥ 13

D ≥ 17 ≥ 14 ≥ 9

E ≥ 13 ≥ 10 ≥ 7

F < 13 < 10 < 7 Source: Table 11-1, Highway Capacity Manual, 1985

Table 5: Urban Street LOS Assignment, HCM 2000

Urban Street Class I II III IV

Range of Free Flow Speed (mph) 55 to 45 45 to 35 35 to 30 35 to 25

Typical Free Flow Speed (mph) 50 40 35 30

Level of Service Average Travel Speed (mph)

A > 42 > 35 > 30 > 25

B > 34-42 > 28-35 > 24-30 > 19-25

C > 27-34 > 22-28 > 18-24 > 13-19

D > 21-27 > 17-22 > 14-18 > 9-13

E > 16-21 > 13-17 > 10-14 > 7-9

F ≤ 16 ≤ 13 ≤ 10 ≤ 7 Source: Exhibit 15-2, Highway Capacity Manual 2000 (U.S. Customary Units)

For example, 1st Street (Market to Harrison) had an average speed during the morning peak period of 12.8 mph. It was classified as HCM 1985 Class III and assigned a LOS D. It was classified as HCM 2000 Class IV and again assigned a LOS D.

Freeways

Freeways followed a similar methodology as arterials, however it was not necessary to assign a class of freeway. The HCM-1985 method was used to calculate LOS for all freeway CMP segments. By knowing the average speed of the freeway in the morning and afternoon peaks, Table 6 was used to assign a LOS in each time period.


Iteris, Inc. | 10

Table 6: Freeway LOS Assignment, SFCTA

Level of Service Density (pc/mi/ln)

Speed (mph)

V/C Ratio Saturation Flow (pcphpl)

A ≤ 12 ≥ 60 0.35 700

B ≤ 20 ≥ 55 0.58 1,000

C ≤ 30 ≥ 49 0.75 1,500

D ≤ 42 ≥ 41 0.90 1,800

E ≤ 67 ≥ 30 1.00 2,000

F > 67 < 30 - - Source: Adapted from Table 4-1, Special Report 209, HCM 1985

For example, the I-280 (Junipero Serra to Weldon) had an average speed during the morning peak period of 24.7 mph. This was correlated to LOS F.

C. Results The comparison between speeds collected in recent years can be used to determine current day variances. Using the length and travel time of each CMP segment computed in step 3a, Table 7 and Figure 6 present the change in average travel speeds, total distance of CMP segments divided by total travel time, in CMP Network between 2009 and 2017. As the table/graph indicates, there was an overall decrease in the average network speeds in 2017 compared to 2015. Furthermore, the statistical analysis shows a significant decrease

in AM Speeds at p < 0.05 (z = -3.9, p = 0.000048, one-tailed), but PM speeds remain similar, as the decrease in PM speeds is not statistically significant (z = -1, p = 0.158655, one-tailed).

Table 7: CMP Network Average Travel Speed comparison across years

Category Peak 2009 2011 2013 2015 2017

Arterial AM 18.6 17.7 17.1 14.6 13.5 PM 16.9 16.6 16.0 12.7 12.1

Freeway AM 48.9 40.6 38.2 37.6 34.8 PM 31.7 31.4 29.5 26.3 26.5

Figure 6: Graph of CMP Network Average Travel Speeds (with % changes compared to 2015)


Iteris, Inc. | 11

Table 8: 2017 CMP Average Travel Speed Results Summary Statistics Number of Segments Average

Speed

Standard Deviation Minimum Speed

Maximum Speed

AM 245 15.90 9.42 6.5 64.8

PM 245 14.60 9.26 3.5 65

Figure 7 presents the change in CMP average speeds between 2015 and 2017. Similar to 2015 CMP Monitoring, a diagonal line from the lower left to the upper right indicating no change has been inserted. Data points in the lower right and upper left triangle are CMPs with significant speed changes, that is, average speeds have decreased or improved from 2015 to 2017, respectively. The pattern of speed changes indicate that data points cluster on the line of no change and most of the changes fall within the range of ±10 mph.

Figure 7: Comparison of 2015 and 2017 CMP Segment Speeds

Few CMP segments along Doyle/ Richardson/ Lombard between Lyon/Francisco Ave and County Line experienced great improvement in average speeds. This improvement is likely because of the completion of Presidio Parkway project in 2016. Significant change in average speeds were also observed for CMP segments along 19th Ave/Park Presidio between US 101 and Lake Ave. The average speed reduced in the SB direction and increased in the NB direction during the PM peak along these segments. While, no construction activities can be associated with this location, it should be noted that there were gaps in INRIX data coverage along the ramps between US 101 and 19th Ave in 2015. These gaps were improved in 2017 with the addition of new TMCs.

CMP 81 & CMP 82: Doyle/ Richardson/ Lombard: Lyon/Francisco to SF Cemetery; and SF Cemetery to County Line

CMP 26: 19th Ave/Park Presidio: Lake to US 101

CMP 27: 19th Ave/Park Presidio: US 101 to Lake

2.0 methodology for vehicle los monitoring los monitoring...2.0 methodology for vehicle los...

Documents