CHAPTER 3: KEY ITS DATA ELEMENTS FOR GOVERNMENT REPORTING SYSTEMS

3.5 USE OF ITS TRAFFIC DATA IN EPA'S MOBILE6 EMISSIONS MODEL

3.5.1 Discussions

With the release of the MOBILE6 emissions model, preparation of transportation inputs will require more detail than previous versions of the MOBILE model. The significant changes that are relevant to transportation activity data are the expansion of the number of vehicle classes from 8 to 28 and decomposing vehicle activity by hour of the day and by four functional highway classes: freeway, arterial, local, and ramp. (Previously, vehicle activity was considered in the aggregate.) MOBILE6 has national default values for all of the transportation activity inputs, but it is doubtful that areas with serious air quality problems would use these. Users may override the defaults by providing:

• VMT percentages by 16 vehicle types (MOBILE6 collapses the 28 classes to 16 for this purpose).
• VMT percentages by 28 vehicle types by hour by four functional classes.
• VMT percentages by hour only.
• VMT percentages by 14 speed ranges by hour for freeways and arterials.

Clearly, the ability of analysts to develop these inputs hinges on the availability of traffic data. The EPA guidance document⁸, however, advises caution when using traditional sources of traffic data:

"Count data and travel demand modeling provide the most widely available information regarding travel activity in urban areas. Nevertheless, because of their primary historical use in evaluating transportation system performance, rather than characterizing regional travel, there are limitations and assumptions associated with their use in emissions estimation and forecasting. For count data, key issues are associated with extrapolation from necessarily limited numbers of locations to regional travel totals. Formalized methods exist based on data from the Highway Performance Monitoring System (HPMS), but the number of HPMS count sites, frequency of sampling, and representativeness of sites are possible sources of uncertainty. For example, in Charlotte, N.C. there are only six HPMS count sites. Selection of either count data or travel demand model outputs for estimating VMT must consider the possibility of hidden flaws which could bias calculated VMT distributions."

A more apropos introduction to the value of ITS-generated traffic data for emission modeling could not be conceived. For the reasons cited earlier, and several more, ITS-generated traffic data overcome most of these limitations because of the intensity of temporal and spatial coverage, at least on the higher functional classes. Additional ways in which ITS-generated traffic data enhance travel activity estimates for emission estimation relate to the use of travel demand forecasting (TDF) models and speed estimation procedures:

• TDF models are geared to providing peak hour estimates of travel. Daily estimates and off-peak travel are factored against the peak hour estimates, usually with data from the handful (4-15) of continuous count locations. Having continuous data in many more locations eliminates the need to estimate volumes at these locations with models.
• Predicted link volume outputs from TDF models are calibrated against short-count (24- or 48-hour) volumes that have been factored to annual averages. Calibration against the actual (rather than factored) counts provided by ITS should improve future forecasted volumes.
• Speed estimates from TDF models are not usually validated against any field data; rather speeds are used as a calibration tool to derive predicted volumes. For air quality planning, the EPA guidance document recommends some degree of post-processing of TDF outputs. Either a modified BPR equation that relates speed to volume-to-capacity ratio or methods found in the Highway Capacity Manual can be used. However, there are several shortcomings with using either of these approaches for estimating speeds:
  • The BPR method does not account for the effect of signal control (density and quality of progression).
  • Neither method accurately captures speeds for oversaturated (i.e., congested) conditions characterized by queuing. As urban areas with the highest need for emissions modeling are also those where congestion is a multi-hour phenomenon, this is a significant factor. ITS-generated data provide direct estimates of speeds for very small time intervals. Currently, these estimates are limited to freeways but research is under way to extend estimates to signalized highways.
  • The effect of factors beyond the base physical capacity of the highway is ignored. Incidents, work zones, special events, day-to-day differences in demand, and weather will all significantly influence speeds. In fact, it is estimated that 40-70 percent of total delay is due to these sources. Continuously measured ITS data imbed all of these factors so that a true estimate of average conditions can be obtained.
  • All modeling procedures - BPR, HCM, or otherwise, all produce estimates of average conditions. However, for air quality planning, the occurrence of extreme "episodes" is important - days or hours where air quality standards are exceeded. In the past, estimating transportation's contribution to these episodes has been limited to anecdotes ("there was a ballgame that day") rather than analysis. Again, the continuous nature of ITS data allows determining transportation activity at specific days and hours.
  • TDF and speed estimation models are still required for doing future year forecasts of air quality. However, the existence of measured speeds over a large portion of the network means that local speed estimation procedures can be developed and validated. Current practice either ignores speed validation altogether or relies on extremely sparse "floating car" travel time estimates taken as little as one run per corridor per peak period per year.

3.5.2 Case Study Using ITS Traffic Data to Provide MOBILE6 Inputs, Louisville, KY

As a demonstration of the ability of ITS traffic data to provide input data for emissions modeling, the Louisville, KY area was contacted. The Louisville MPO, the Kentuckiana Regional Planning and Development Agency (KIPDA), had been in discussions with EPA over updating their methods to supply MOBILE6 inputs. In particular, the previous reliance on modeled (rather than measured) vehicle speeds was viewed as a problem.

The authors of this report obtained data for calendar year 2001 from the TRIMARC system in Louisville, an ITS deployment covering approximately 12.4 miles of the area's freeways, all on the Kentucky side of the Ohio River:

I-64: I-65 Jct to US-ALT-60 - 3.2 miles
I-65: I-64/I-171 to I-264 - 5.4 miles
I-71: I-64 to Indian Hills - 3.8 miles

Several analyses were conducted on the TRIMARC data in support of MOBILE6 input requirements, as discussed below.

Estimation of Free Flow Speeds

Free flow speed (FFS) is not a direct input to MOBILE6 but rather is an input to analytic procedures that in turn produce MOBILE6 inputs. FFS is used in both highway capacity analysis and travel demand forecasting models; these are common techniques used to support the speed input requirements of MOBILE6.

To derive FFSs, the TRIMARC data were analyzed using only daylight hours⁹ where the estimated volume-to-capacity-ratio (V/C) was less than or equal to 0.2. Both average speeds and the 85^th percentile speeds were computed. Following guidance in the Highway Capacity Manual¹⁰ the average speeds were chosen to represent FFSs. As shown in Table 3.3, both the average and 85th percentile speeds vary by route AND direction.

VMT Fraction by Hour (Freeways)

One of the MOBILE6 traffic activity inputs is the fraction of VMT that occurs in each hour of the day by facility type. Presumably, this distribution is supposed to be developed for typical weekdays. However, continuous data allow separate distributions for weekdays and weekend/holidays, if the need arises to model them individually (Table 3.4). These data correspond to the "VMT BY HOUR" input requirement of MOBILE6.

VMT Fractions by Hour and Speed Range

Hourly speed information is a new requirement of emissions models that is difficult to obtain. Transportation planners do not collect this level of information - peak period travel speeds are sometimes collected using sampling methods but the full diurnal pattern of speeds is not used in other planning applications. With ITS-generated traffic data, however, it is possible to generate this information easily. Table 3.5 presents the results for Louisville freeways in a format similar to the "SPEED VMT" input requirement of MOBILE6.

In addition, average hourly speeds by corridor and direction can also be obtained if more detailed emissions analyses must be performed. Figures 3.12 through 3.17 display the average hourly variation in speeds for Louisville freeways.

Development of an Area-Specific Speed Performance Function

Another aspect of speed estimation which was of interest to KIPDA was the re-calibration of the speed performance function used in their travel demand forecasting model. These functions are used to assign traffic to the network and are usually in the form originally specified by the Bureau of Public Roads in the 1960s:

Speed = FFS/(1 + a(V/C)^b)         (1)

Where: a and b are coefficients
FFS = free flow speed
V/C = volume-to-capacity ratio

Speed-flow curves were produced for the freeway corridors in the TRIMARC data (Figures 3.18 to 3.21). The general pattern of these data exhibit the "classic" shape discussed in the Highway Capacity Manual and many other sources of traffic flow theory: speeds are relatively constant over a wide range of volume levels. Near capacity there is a small drop-off in speeds and when demand exceeds capacity, both volumes and speeds fall dramatically. (Volumes fall because once traffic flow has "broken down" due to a bottleneck, capacity itself also drops. The extent of this drop depends on the nature of the bottleneck.) However, the data in the Louisville speed-flow curves also show a large amount of variability, even in the uncongested flow regime. Attempts to fit modified forms of Equation (1) were completely unsuccessful - the goodness of fit parameters were extremely poor. Therefore, it was recommended that for the Louisville travel demand forecasting model, the coefficients contained in NCHRP Report 387 should be used.¹¹

Figure 3.12. I-64 EB, Weekdays 2001

Figure 3.13. I-64 WB, Weekdays 2001

Figure 3.14. I-65 NB, Weekdays 2001

Figure 3.15. I-65 SB, Weekdays 2001

Figure 3.16. I-715 NB, Weekdays 2001

Figure 3.17. I-71 SB, Weekdays 2001

Figure 3.18. Speed-Flow Curve for I-64 Eastbound

Click here for text description of Figure 3.18

Figure 3.19. Speed-Flow Curve for I-64 Westbound

Click here for text description of Figure 3.19

Figure 3.20. Speed-Flow Curve for I-65 Northbound

Click here for text description of Figure 3.20

Figure 3.21. Speed-Flow Curve for I-65 Southbound

Click here for text description of Figure 3.21

⁸SAI and ICF Consulting Group, Development of Methodology for Estimating VMT Weighting by Facility Type, Report No. M6.SPD.003, prepared for EPA, September 1998, http://www.epa.gov/otaq/models/mobile6/m6tech.htm
⁹The authors noted that in Louisville, speeds in the early morning hours (1:00 - 5:00 AM) were lower than in other hours with similar volumes. Similar patterns have been detected in ITS data from other areas.
¹⁰Transportation Research Board, Highway Capacity Manual 2000, Washington D.C., pg. 23-2.
¹¹Dowling, R., Kittleson, W., Zegeer, J., Skabardonis, A., Planning Techniques to Estimate Speeds and Service Volumes for Planning Applications, NCHRP 387, Transportation Research Board, 1997.