Transcript Slide 0
Activity Based Models Review presented to Model Task Force Model Advancement Committee presented by Thomas Rossi Krishnan Viswanathan Cambridge Systematics Inc. Date November 24, 2008 Presentation Overview Study Background and Objectives Models Studied Study Findings Discussion 1 Study Background and Objectives Examine existing activity based models to determine model features, application procedures, and requirements Determine planning analysis needs for which travel models are used Summarize the ability of activity based models to provide accurate information for planning analysis needs 2 Models Studied Urban Models • San Francisco County, CA (2001) • New York, NY (2002) • Columbus, OH (2005) • Sacramento, CA (2007) • Lake Tahoe, NV/CA (2007) • Atlanta, GA • Portland, OR • Denver, CO • San Francisco Urban Area (MTC), CA 3 Models Studied (Cont’d) Statewide Models • Ohio Model (2007) • Oregon Model Research Models • FAMOS (University of South Florida) • CEMDAP (University of Texas) • TASHA (University of Toronto) 4 Models Studied (Cont’d) Atlanta Portland Denver San Francisco (MTC) 2007 2008 (est.) 2008 (est.) 2008 (est.) 2009 (est.) 2000 2000 2005 2000 2035 2030 2035 2030, 2050 2001 1994 1997 SFCTA New York Columbus Sacramento Lake Tahoe Year Completed 2001 2002 2005 2007 Base Year 2000 1996 2000 2005 2020 2030 Forecast Year Ohio Oregon 2007 2008 (est.) 2000 2003 No Survey Survey Data Year 1990 1998 1999 2000 Number of Households in Survey 1,300 11,000 5,600 3,900 1,220 8,100 6,000 4,900 15,000 15,000 No Survey 1,700 (750 in SF) 3,600 1,800 1,500 289 2,000 2,000 2,800 1,454 5,300 3,000 Zones (approximate) Area Size (square meters) 50 (SF only) 150 (est.) 4,000 501 Base Year Population 750,000 (SF only) 1,500,000 2,000,000 63,448 5 4,700,000 500 7,000 1,600,000 6,783,760 Study Findings Model Structure All models estimated from household activity/travel survey • Same type of survey used for four-step model development Individuals in region’s population are simulated • Activity patterns • Locations and times of activities • Modes used to travel between activity locations 6 Study Findings Model Structure (Cont’d) Model structure • Generate daily activity patterns • Location, time and mode made at two levels : Tour and Trip Five to eight activity purposes • Work, school, shop, meal, social/recreation, and personal business Some models consider household interactions • Implications for time of day and mode choice • Is it cost effective to include this to gain accuracy? The “jury is still out.” 7 Study Findings Model Components Population Synthesizer Long Term Choice Models • Auto ownership • Usual workplace location Daily Activity Pattern Models Tour Level Models (primary activity) • Destination choice • Mode choice • Time of day choice 8 Study Findings Model Components (Cont’d) Trip Level Models (intermediate stops) • Destination choice • Mode choice • Time of day choice Trip Assignment • Highway • Transit 9 Study Findings Model Development Process Model development between 1.5 to 8 years (typically 2-3 years) Model development costs – typically $600,000-$800,000 Consultants nearly always used for model development Most models used local household activity survey data along with other sources such as transit on-board, external or visitor surveys Lake Tahoe model was transferred from Columbus 10 Study Findings Model Execution Standard transportation modeling software such as CUBE/Voyager, TransCAD used along with custom programs in C++, Java, or Python Run times range from 10 hours to 2 days • Distributed computing preferable to reduce runtime Models need around 7 to 10 GB of storage per run Most models run only in-house 11 Study Findings Policy Planning Analysis Activity Based Models benefit the following types of analysis • Congestion Management Systems • Toll Feasibility Studies • High-Occupancy Vehicle (HOV) Lane Studies • New Starts/Small Starts Analyses • Hurricane Evacuation Modeling Support • Air Quality Conformity Determinations • Integrated Land Use Model • Incorporate Ability to Test Impact of Gasoline Prices • Freight Studies 12 • Growth Management/Concurrency Applications Study Findings Data Needs No special data needs required to develop activity based models beyond what is used for four-step models Existing household travel surveys can be used to develop data for activity based models Other data sources such as transit on-board surveys, external and visitor surveys are also helpful for activity based models Census data sources such as PUMS useful for population synthesis • ACS disclosure rules can be problematic 13 Conclusions Models use similar approaches • Main differences related to explicit modeling of household interactions Members of population simulated individually • Their activities, locations, times, and mode choices Standard modeling software used along with custom programs Typically 2-3 years, $600,000-$800,000 to develop models 14 Run times typically 0.5 to 2 days Discussion 15