Appendix D: Technical Memorandum #3
2.5 Traffic Assignment
In validating traffic assignment, final outcome of a travel demand model, common performance metrics include the following:
• Systemwide volume-to-count ratio;
• Volume-to-count ratios by link group (area type, facility type, and number of lanes);
• Volume-to-count ratios along screenlines;
• Percent difference in model volumes and counts by volume group;
• Systemwide coefficient of determination between assigned volumes and counts;
• Systemwide percent root mean square error; and
• Percent root mean square error by volume group.
Table 7 includes the volume-to-count ratios by facility type, area type, and number of lanes.
Overall, the model is validating at 0.96, within acceptable limits. This is also an improvement in the overall validation since the 2000 model with a volume-to-count ratio of 0.95. Although the validation of expressways improved in the 2007 model, the validation of undivided principal arterials, minor arterials, and one-way facilities got worse when compared to the 2000 model.
This can be attributed to the correction in the network coding in the 2007 model reflecting the 3rd and 4th Street one-way facilities in downtown Chattanooga. Furthermore, additional network detail was included in some portions of the model as a result of further TAZ splits, as well as a significant amount of more traffic counts being provided in the 2007 model. Therefore, although the 2007 model may indicate lower volume-to-count ratios for some facility type categories, it is likely more accurate than the 2000 model due to the additional traffic counts. In addition, no ramp counts were available in the 2000 model and have since been added to the 2007 model validating at a 2007 volume-to-count ratio of 0.99. Four of the six facility type categories that have targets set by TDOT are within acceptable limits.
Chattanooga is validating at 0.74, which has historically validated lower than other categories in the Chattanooga model. With the exception of one lane facilities (by direction), the 2007 model is validating better within each lane category compared to the old 2000 model.
Table 7. Volume Over Count Ratios by Facility Type, Area Type and Number of Lanes
Number of Lanes by Direction
1 0.88 0.93
2 0.99 0.98
3 0.98 0.95
4 1.03 0.92
Total 0.96 0.95
screenline locations is provided in Appendix C. Maps of volume-to-count ratios on each link in the model with a count are included in Appendix D. For eight of the 10 screenlines, the volume-to-count ratios fall within the +/- 10 percent target range of TDOT. Table 8 compares the volume-to-count ratio for each screenline in the 2007 model against the old 2000 model.
Table 8. Comparison of Volume-to-Count Ratios by Screenline
Chattanooga Volume/Count Ratio
Screenline 2007 2000
1 1.00 0.97
2 1.01 0.94
3 0.88 0.97
4 0.99 0.91
5 0.82 0.90
6 0.93 0.99
7 1.10 1.09
8 0.91 1.06
9 1.06 1.02
10 1.00 1.00
Table 9 indicates the percent difference in model volumes and observed counts by volume group. All volume groups within the 2007 model achieve the established TDOT targets.
AADT Chattanooga 2007TransCAD TDOT Targeta
The correlation coefficient of observed counts to model volumes (R2) is performing well at 0.95, exceeding TDOT’s target of 0.88, as demonstrated below in Figure 1.
Figure 1. Correlation of Assigned Versus Observed Volumes
0
0 10000 20000 30000 40000 50000 60000 70000
2007 AADT
2007 Assigned Volumes
As indicated in Table 10, the overall RMSE for the 2007 model is 34.6 percent, within acceptable federal limits. TDOT has established an overall RMSE target of 30 percent. However, based on discussion with TDOT during the June 30, 2009 teleconference, it was determined that 30 percent was just a target and was not required as federal standards indicate a RMSE range of 32-39 percent is within acceptable limits. Although the previous 2000 model indicated an overall RMSE of 23 percent, it is suspected that it was not as accurate due to the reduced
Table 10. Root Mean Square Error (RMSE) by Volume Group percent of VMT that falls within each model facility type compared to TDOT targets. With the exception of minor arterials, all of the facility types fall within or close to TDOT’s targets for percent of VMT by facility type. In addition, the overall ratio of VMT to Highway Performance Monitoring System (HPMS) VMT is 1.06 (excluding external centroid connectors).
Facility Type Chattanooga 2007 Model VMT TDOT Target
Interstate 4,228,041 (31.4%)
Expressway 1,496,818 (11.1%)
33-38%
Principal Arterial Divided 2,067,224 (15.4%)
Principal Arterial Undivided 835,239 (6.2%) 27-33%
Minor Arterial 1,384,961 (10.3%) 18-22%
Collector 1,465,718 (10.9%) 8-12%
Ramp 215,118 n/a
One-Way 16,672 n/a
External Connectors 921,730 n/a
Centroid Connectors 827,412 n/a
Total 13,458,933
2.6 Comparison to Observed Speeds
During November 2008, the RPA conducted travel time runs during the AM and PM peak periods along all routes in their Congestion Management Plan (CMP). The results of these travel time runs were used to calculate average peak period congested speeds. These observed speeds were then compared to the daily congested speeds in the model during the 2007 model validation process. Appendix E includes a map of the observed 2008 AM peak period congested speeds, as well as a map of the daily congested speeds from the 2007 model. As expected, the model speeds compared to the observed speeds are slightly different in some instances since the model speeds are daily and the observed speeds are peak period. However, there was a significant improvement in congested speeds along interstates in the 2007 model as they were too low in the 2000 model.
3.0 Summary of 2007 Model Validation Observations
• The overall 2007 model is validating at 34.6 percent, within acceptable federal limits;
• The 2007 model is validating at a systemwide volume-to-count ratio of 0.96, within acceptable limits and improved since the 2000 model;
• The correlation coefficient of observed counts to model volumes (R2) is performing well at 0.95, exceeding TDOT’s target of 0.88;
• Eight of ten screenlines are validating within acceptable limits;
Home-based Work (HBW) trip lengths did improve;
• The percent of intrazonal trips attributed to Home-based Work (HBW) trips are now within a more reasonable range;
• Congested speeds along the interstate and other higher facilities are improved compared to the 2000 model and observed congested speeds;
• Overall, higher facilities are generally comparable to the 2000 model, however, minor facilities got worse in the 2007 model due to additional traffic count availability and network detail;
• Overall, the 2007 model has more counts with almost 10 percent of the network having counts. As a result, the 2007 model includes improved accuracy and reliability of the model statistics;
• Employment in the 2007 model is lower than in the 2000 model, resulting in lower commercial trips; and
• As with all models, there is always room for improvement. However, this model should be sufficient for updating the 2030 LRTP to 2035.
4.0 Future Model Enhancements
Several enhancements are planned for the next generation of the Chattanooga travel demand multiple data collection and compilation efforts over the next two years. Below are some of the data collection efforts planned:
• Socioeconomic Data – Once the Census 2010 population and household data is released, likely in 2012, the RPA will use the data to develop the base year 2010 model socioeconomic data. In the meantime, the RPA will be providing guidelines to the counties and municipalities in the TPO region stating what socioeconomic data is required to assist with compiling the data. Additionally, proprietary employment datasets for the year 2010 may be evaluated for accuracy to determine the most appropriate data source for the
• Traffic Counts – During calendar year 2010, the RPA will be identifying and collecting traffic counts at key locations throughout the TPO to assist with validating the 2010 base year model as part of the next LRTP update. The traffic counts will be collected by vehicle class and 15-minute intervals in order to validate trips at the truck level and to include a new time-of-day component in the model.
• External Origin and Destination Survey – In 2002, a roadside origin-destination intercept survey designed to obtain characteristics of travelers entering, exiting, and passing through the region from locations outside the CHCNGA study area was conducted. During the last 2030 LRTP Update, it was determined that the results from the 2002 external origin and destination survey were biased. As a result, during the last 2030 LRTP Update, as well as this current 2035 LRTP Update, the 2002 roadside origin-destination intercept survey was not used within the TransCAD model. Therefore, the RPA intends to conduct a new external origin and destination survey during the fall of 2010. In the meantime, the RPA will be designing the sampling plan and survey instrument. The results of the 2010 external origin and destination survey will be used to validate the external model for the 2010 base year validation as part of the next LRTP update.
• On-Board Transit Survey – As mentioned earlier, the RPA intends to conduct an on-board transit survey in order to validate a new mode choice/transit component in the TransCAD travel demand model. The RPA will be designing the sampling plan and survey instrument in the spring/summer of 2010 with the survey being conducted in the fall of 2010.
• Household Travel Diary Survey – In 2002, a household travel diary survey was conducted to determine trip generation rates, average trip lengths, auto occupancy factors, and other characteristics used in model development and validation. Typically, household travel diary surveys are conducted every 10 years to accurately reflect travel characteristics of the region. Pending funding availability, the RPA intends to conduct a new household travel diary survey in calendar year 2011. The results of the 2011 travel diary survey will be incorporated into the new 2010 base year model for the next LRTP update.
• Bicycle and Pedestrian Counts – If a nonmotorized component is added to the travel demand model prior to validating the 2010 base year model for the next LRTP Update, the RPA will collect bicycle and pedestrian counts. Currently, the RPA owns three bicycle and pedestrian counters and has begun collecting counts in key locations.
4.2 Travel Demand Model Enhancements
The above mentioned data collection efforts will be used to incorporate several enhancements to the current CHCNGA TPO TransCAD model. These enhancements could potentially include the following additions to the TransCAD model prior to the next LRTP Update:
• Updated external model based on 2010 external data;
forecast transit ridership;
• New nonmotorized component not previously included in the model to be used to forecast bicycle and pedestrian trips;
• New time-of-day component used to disaggregate daily trips into three to four time periods.
This will improve the accuracy in calculating air quality emissions and is typically necessary for transit models; and
• Validate truck trips in addition to total trips, whereas in the past, only total trips were validated.