A Comparison of CEMDAP Activity-Based Model With DFWRTM 4-Step Model Arash Mirzaei P.E. NCTCOG - Arlington, Texas and Naveen Eluru UT Austin - Austin, Texas for 11 th TRB National Transportation Planning Application Conference Daytona Beach, Florida May 2007
Credit and Acknowledgement Credit CEMDAP Research Team Chandra Bhat, Jessica Guo, Siva Srinivasan (Faculty) Abdul Pinjari, Rachel Copperman, Ipek Sener (Graduate Students) NCTCOG Ken Cervenka, Bin Chen, Arash Mirzaei Acknowledgement TxDOT Janie Bynum, Bill Knowles
Content Introduction The Models Test Method Comparisons Closing Thoughts
Introduction
Regional 4-Step Model Use Long Range Plan, Mobility Plan 2030 Number of lanes Cost estimate Benefit analysis Project selection Air Quality Conformity Analysis Emission - budget analysis Project AQ conformity Project Specific Analysis Alternative analysis in Roadway (including HOV, Toll, and Managed Lane) Transit (system planning, New Starts) TIP and Congestion Management Platform for communication and storage of various data Used as a guide for policy making in various transportation issues
Is 4-Step Model Adequate? For many current uses, 4-step model has been adequate Some areas of shortfall: Peak spreading Treatment of NHB trips Evaluation of demand management strategies Environmental justice
Expectation of ABM At least as good as 4-step in areas 4-step has proven adequate Noticeably better in areas 4-step has not been adequate Focus on person vs. transportation facility (e.g. who parked and why vs. how many parked). A much harder calibration process.
Road From 4-Step to ABM 1. Have a production line 4-step model 2. Create an ABM in parallel of the production line model 3. Clear up uncertainties about random seed, run time, and other operational issues 4. Check validation designed for 4-step vs. ABM 5. Check validation designed for ABM 6. Perform sensitivity tests designed for 4-step vs. ABM 7. Perform sensitivity tests designed for ABM
Validation Designed for 4-Step Production/attraction by purpose Trip length distribution (TLD) by purpose/income Mode choice by market segments Transit ridership by route, stops, route groups, Park-and- Ride, mode of access Traffic volume by functional class, area type, geographic area, and facility type (HOV, Toll, HOT) VMT, VHT, LOS by functional class and time periods Facility and corridors ABM should show at least the same performance in these tests
Validation Designed for ABM Distribution of income by HH employment, size in zonal level Trip production rates by purpose, HH type, time-of-day, Trip attraction rates by type of employment in zonal level Trip chain patterns by time of departure, duration, TLD by purpose/income/hh type/employment type, Transit ridership by type of riders (income) by route (rail, bus) Traffic volumes by type of riders for HOV, Toll, Managed lane More to be determined ABM should show reasonable results on these tests. The tests should be comprehensive enough to help understand the limitation of the ABM model, even though the model structure may support the test.
Sensitivity Designed for 4-Step Controlled supply changes IVTT OVTT Toll/fare/cost Number of lane Controlled demand changes Production/Attraction Population/Employment Project level tests (scenario analysis) Transit New Starts Traffic/Revenue analysis Model setup tests Transit path builder parameters Traffic assignment parameters
Sensitivity Designed for ABM Seed variation Number of iterations at different levels Assumptions in HH synthesis Assumptions in activity generator Assumptions in time of departure More TBD based on different models
Scope of This Comparison Controlled supply changes IVTT, effect of 25% change on aggregate results OVTT Toll/fare/cost Number of lane Controlled demand changes Project level tests (scenario analysis) Model setup tests
The Models
DFWRTM Structure 4874 zones Trip generation cross-classification Trip distribution doubly constrained gravity Mode choice: HBW(6 segments), HNW(6 segments), NHB Time-of-day factors: AM peak(2.5 hours), PM peak(3.5 hours), off peak(18 hours) Transit assignment for daily trips (TransCAD pathfinder) Non-transit UE
Conceptual Overview of CEMDAP Forecast Year Outputs Aggregate sociodemographics (base year) Synthetic population generator (SPG) Link volumes and speeds Dynamic Traffic Assignment (DTA) Activity-travel environment characteristics (base year) Policy actions Detailed individuallevel sociodemographics (base year) Socio-economics, land-use and transportation system characteristics simulator (CEMSELTS) Individual activity-travel patterns Model parameters Base Year Inputs Socio-demographics and activity-travel environment Activity-travel simulator CEMDAP
CEMDAP Features Generic Design Can be applied to any metropolitan area Temporal Resolution Continuous time scale (1 min. for DFW application) Level-of-service data can be provided at any temporal resolution (5 time-of-day periods for DFW application) Spatial Resolution Allows for any number of zones (4874 for DFW application) Graphical User Interface Standard Window-based user interface Allows user to modify model parameters Provides a friendly diagrammatic interface to help the user understand the logic of the system and the underlying models
Test Method
DFWRTM 4-step Process DEMOGRAPHIC INFORMATION ZONE LAYER ROADWAY NETWORK TRIP GENERATION ROADWAY SKIMS TRIP DISTRIBUTION TRANSIT NETWORK TRANSIT SKIMS LOOP MODE CHOICE TRAFFIC ASSIGNMENT NO TRAVEL TIME CONVERGENCE YES INPUT PROCESS TRANSIT ASSIGNMENT DECISION
DFWRTM with CEMDAP DEMOGRAPHIC INFORMATION ZONE LAYER ROADWAY NETWORK TRIP GENERATION ROADWAY SKIMS CEMSELTS TRANSIT NETWORK TRANSIT SKIMS LOOP CEMDAP TRAFFIC ASSIGNMENT NO TRAVEL TIME CONVERGENCE YES INPUT PROCESS TRANSIT ASSIGNMENT DECISION
Warning! We are constraining a Continuous Time model by using three broad time periods for a traditional static traffic assignment. CEMDAP has a lot more input variables for describing households.
Comparisons
Characteristics Of Modeling Area 5,000 Square Miles 4,874 Zones (4,813 Internal + 61 External) Population 4.848 Million in 1999 7.952 Million in 2025 (64% Increase From 1999) Coded Lane Miles 27,000 in 1999 38,000 in 2025 (41% Increase)
% RMSE by Functional Class 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 5 0 Expr essway Pr i nci pal Ar ter i al Mi nor Ar ter i al Col l ector Ramps Fr ontage Al l DFW99 w/kfac DFW99 No KFAC CEMDAP00
% RMSE by Count Range 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 5 0 < 10,000 10,000 to 19,999 20,000 to 29,999 30,000 to 39,999 40,000 to 49,999 50,000 to 59,999 60,000 to 69,999 70,000 to 79,999 80,000 to 89,999 90,000 + ALL DFW99 w/kfac DFW99 No KFAC CEMDAP00
% Volume Error: (Vol-Cnt)/Cnt 15 10 5 0 Expressway Principal Arterial Minor Arterial Collector Ramps Frontage All -5-10 -15-20 DFW99 w/kfac DFW99 No KFAC CEMDAP00
Base Year Vehicle Trips 0 2,000,000 4,000,000 6,000,000 8,000,000 10,000,000 12,000,000 14,000,000 16,000,000 DFW-Drive Alone CEM-Drive Alone DFW-Shared Ride CEM-Shared Ride DFW-Trucks CEM-Trucks DFW-All CEM-All AM Peak Period (2.5 Hours) PM Peak Period (3.5 Hours) Offpeak Period (18 Hours)
Vehicle Miles of Travel (Including Connectors) DFW99 CEMDAP00 CEM/DFW AM Peak (2.5 Hours) 24,090,807 27,863,030 1.16 PM Peak (3.5 Hours) 33,749,427 39,724,904 1.18 Off Peak (18 Hours) 73,724,981 77,623,830 1.05 Weekday 131,565,215 145,211,764 1.10
Forecast Comparisons DFW Model CEMDAP CEM/DFW Vehicle Trips Base (99/00) 14,341,592 14,101,064 0.983 2025 23,056,356 23,123,062 1.003 2025/Base 1.608 1.640 1.020 Vehicle Miles Base (99/00) 131,565,215 145,211,764 1.104 2025 237,717,217 240,807,368 1.013 2025/Base 1.807 1.658 0.918 Trip Length Base (99/00) 9.17 10.30 2025 10.31 10.41
Vehicle Trips to Dallas CBD Drive Alone Shared Ride Total DFW Model 1999 150,550 44,153 194,703 CEMDAP 2000 175,220 55,860 231,080 CEM00/DFW99 1.16 1.27 1.19 DFW Model 2025 195,369 57,813 253,182 CEMDAP 2025 261,340 68,280 329,620 CEM25/DFW25 1.34 1.18 1.30 DFW 2025/1999 1.30 1.31 1.30 CEM 2025/2000 1.49 1.22 1.43 2025 Jobs/1999 Jobs = (156,825/130,286) = 1.20 2025 Pop/1999 Pop = (15,316/1,620) = 9.45 2025 Pop + Emp / 1999 Pop + Emp = 1.31
VMT Impact of 25% Increase in Auto and Transit IVTT AM Peak PM Peak Offpeak Weekday CEMDAP 2000 Assume No Change To Worker Locations Assume Changes To Worker Locations -7.8-14.9-14.9-13.5-12.3-20.1-17.1-17.0 DFW Model 1999 Assume No Change to HBW Trips -2.6-4.6-5.8-4.9 Assume Changes To All Trip Purposes -7.3-7.9-8.1-7.9 DFW Model Forecasts 2010--Assume Changes To All Trip Purposes 2025--Assume Changes To All Trip Purposes -6.9-7.5-7.7-7.5-17.1-14.5-12.6-13.9
VMT Sensitivity to 25% Increase in IVTT - 7.5% = DFW Model (2010) - 7.9% = DFW Model (1999) - 8.2% = Vancouver (Washington) Base Year Model - 8.4% = Puget Sound (Washington) Base Year Model - 13.9% DFW Model (2025) - 17.0% CEMDAP (2000)
CEMDAP Sensitivity Test (with Home-Worker Location fixed) Modes Time Periods Change 25% Increase in IVTT Auto All -13.8 25% Increase in IVTT All All -13.5 25% Increase in IVTT All AM and PM Peak -5.1 25% Increase in IVTT Auto AM and PM Peak -5.0 25% Increase in Auto Costs Auto All -1.3 25% Increase in Auto/Transit Costs All AM and PM Peak -1.0 $2.00 increase in Auto Costs To/From Dallas CBD Auto AM and PM Peak -0.8 25% Increase in Auto Costs All All -0.7 25% Increase in Auto Costs Auto AM and PM Peak -0.5 25% Decrease in IVTT All All 9.3
Closing Thoughts
Concerns in the Method Employment Differences CEMDAP uses census 2000 Jobs DFWRTM uses BEA workers Singly constrained trip-end attraction in CEMDAP Possibly causes more or less attraction to zones with specific number of jobs? Effects of restricting CEMDAP to DFWRTM limitation Limitation on time periods Static traffic assignment
Interim Conclusions Very encouraging 4-step validation results in non-transit VMT Sensitivities for base years change in IVTT is higher in CEMDAP but in the same direction. 4-step sensitivities across different regional models were very close to each other.
Next Steps Understanding of 4-step models needs improvements These models have been in practice for many years but formal validation process and sensitivity tests have not established yet. ABM validation tests These models should be validated beyond 4-step models to deliver the reasons of their creation ABM operational issues (run time, seed, number of feedbacks, more) These issues are not clearly stated or tested yet. Current ABM s performance in projects can be compared with 4-step models Can peak spreading be properly handled? Can NHB trips be analyzed better? Can market segments be looked at in more detail?