Developing and Validating Regional Travel Forecasting Models with CTPP Data: MAG Experience

Transcription

1 CTPP Webinar and Discussion Thursday, July 17, 1-3pm EDT Developing and Validating Regional Travel Forecasting Models with CTPP Data: MAG Experience Kyunghwi Jeon, MAG Petya Maneva, MAG Vladimir Livshits, MAG Acknowledgements: Ken Cervenka, FTA 1

2 ACS Data Including CTPP Products and PUMS Data Were Used in MAG Model Development and Validation: Trip Generation Model Development Destination Choice Model Development Auto Ownership Model Development Model Validations 2

3 Focus of this Presentation is on Comparing Commuter Flows between the MAG Model and CTPP 5-year estimates CTPP American Community Survey (ACS) 5-year estimates are PERIOD estimates MAG Model Socioeconomic data is based on a single point in time estimate ACS estimates are controlled to the Census Bureau s annual population estimates over the 5 year period 3

4 MAG Modeling Area: Where We Are Area: 16 thousand square miles 211 Population: 4.5 million people 211 Employment: 1.8 million jobs Traffic Analysis Zones: 3,9 4

5 MAG 211 Socio-Economic Projections VS. CTPP 5-year Estimates: Not Directly Comparable but Comparison of Distributions of Commuter Flows on an Aggregate Level is still of Interest Population 211 MAG Model CTPP Difference Perc. Difference Maricopa 4,14,542 3,751, ,97 9% Pinal 395, ,26 66,694 2% Total 4,5,496 4,8,75 419,791 1% Employment 211 MAG Model CTPP Difference Perc. Difference Maricopa 1,744,641 1,442,481 32,16 21% Pinal 59,313 69,879-1,566-15% Total 1,83,954 1,512,36 291,594 19% 5

6 Seasonal and Transient Populations Are Included in MAG Projections but Are not Reflected in CTPP Data 211 MAG Model Population Categories: Residential Group Quarters Seasonal Transient ACS Database Population Categories: Residential Group Quarters Other Differences in Employment Calculations 6

7 Socio-Economic data is not directly comparable between CTPP estimates and MAG annual forecast We can t compare absolute trip numbers and need to look at comparison of proportional distribution of travel between CTPP and MAG model. 7

8 Biggest Population Difference is on the Edge of the Region in Pinal County Proportion of Trips Going To/From this District Does Not Affect Substantially Other Trip Interchanges 8

9 Pinal County is the Only District with Higher CTPP Employment Estimates than MAG Data 9

10 How we Delineated Districts: the Starting Point Initial districts (32) that were used to reweight 28 NHTS - Homogenous median household income for the district s block groups - No less than 1 households surveyed in NHTS 1

11 Income Distribution as a Criterion 11

12 How We Delineated Districts Previous MAG experience showed that in order to facilitate manual analysis of the largest travel patterns in the region 1 to 15 districts are most appropriate. Other factors to consider: - Average trip length for HBW trips: 12.7 miles - Margins of error in CTPP data sets - Land use and socio-economic data 12

13 Analysis of Largest Desire Lines and Spider Networks can Help to Determine Number and Configuration of Districts 13

14 How to Draw Districts Somewhat Similar to Delineating TAZ and Creating Screenlines Starting point can be districts created for another purpose Look for socio-economic homogeneity inside a district but ensure that level of aggregation is sufficient for meaningful comparison Let freeways, jurisdictional limits and major physical barriers serve as boundaries as appropriate Exclude peripheral zones/districts A few iterations might be required 14

15 Sequence of District Delineation 15

16 Defining Districts: Conforming to Jurisdictional Boundaries and Freeways 16

17 Aggregation of TAZs to Districts Model: Select contiguous TAZs in Agency s dataset Aggregate sets of TAZs to districts CTPP: Generate centroids point for CTTP TAZs Overlay CTPP centroids with district polygons Assign whole CTTP TAZs to districts Map both sets of districts for quality assurance 17

18 Final District System Excluded from comparisons because of lack of spatial overlap between the district and the MAG modeling Area Excluded from comparisons because of lack of spatial overlap between the district and the MAG modeling Area 18

19 Creating Tables for Comparison 211 MAG Model PA person trip tables are used pk_hbw_aggregated + op_hbw_aggregated CTPP Database CTPP Flows - TAZ-to-TAZ home-to-work trips, aggregated to district level 19

20 Modeled District-to-District HBW Trips 211 MAG Model District Total 1 6,153 1, ,613 4,417 7,837 3, , ,427 4,68 19,735 21,586 17,64 5, , ,246 15,313 24,79 6,998 1, ,438 4,183 1, , ,57 7,816 9,446 41,687 6, ,718 4,782 3, , ,347 3,132 41,59 91,778 21, ,695 1,44 1, , ,635 1,913 28,39 91, ,19 11,84 2,429 4,974 1,325 1, ,21 7 2,145 7, ,435 15,671 31,498 38,934 4, , , ,452 7,694 11,144 5,658 17, , ,15 15,67 2, ,756 2,953 7, , ,24 27,197 14,64 1, ,129 5,427 27,77 1,83 135, ,481 28,97 21,146 3, ,881 27,785 94,544 1,92 21, ,478 6,98 1, ,49 1,66 23,137 45,778 17,261 Total 1,44 75,275 41, , , ,481 71,749 28,87 16,34 13, ,133 49,847 1,498,11 2

21 CTPP District-to-District Flows CTPP Data District Total 1 5, ,15 3,718 5,47 1, , ,744 3,335 16,4 2,948 17,194 4, ,242 1, , ,848 16,52 14,284 6,264 1, ,49 2,278 2, , ,64 7,23 87,48 38,73 11,234 1, ,891 5,62 5, , ,84 4,152 38,918 92,122 27,521 3,563 1,143 5,426 3,172 3, , ,168 2,971 35,932 85, ,116 16,127 4,793 5,68 5,397 6, , ,533 5, ,998 19,178 28,761 38,17 4,584 1,26 1,25 1, , ,926 7,496 9,189 5,38 15, , ,34 1,16 16,64 15,281 3, ,376 3,51 7,786 1,27 78, ,745 3,592 17,22 13,213 4,872 1, ,835 55,941 21,368 2,82 128, ,217 3,277 26,36 23,349 6,797 1, ,537 25,6 84,988 4,817 21, ,778 5,812 2, ,572 8,576 13,22 55,996 98,526 Total 1,69 8,598 44,127 27, , ,27 78,956 31,625 81, ,419 15,382 67,73 1,49,361 21

22 Difference in Number of Trips 211 MAG Model CTPP Data District Total ,79 1, , ,683 1,345 3, , , ,27 1, , , , ,966 2,957-5,145-1, , , ,2 2, ,961-2, ,768-2, , ,533-1,58-7,542 6, ,287-2, ,72-5, , , ,563-3,57 2, , , , , , , , , , , , ,32-1, ,294-5,514 5,79-1,17 7, , ,737-2,23-3,666-1, ,344 2,779 9,556-2,897 9, ,7 1, ,918 1,49 9,917-1,218 8,735 Total 335-5,323-2,43 22,59 6,429-14,546-7,27-2,818 25,121-8,423 11,751-17,226 7,749 22

23 Things to Consider when Comparing Number of Trips: Differences in population and employment do not directly translate in differences in proportional distribution of commuter trips between trip interchanges. Most of the trips to work in Pinal County are to Maricopa County. Other trip interchanges are not affected by much. Trip rates fluctuate and are different between CTPP period estimates and 211 MAG point estimate. 23

24 D2D Trips MODEL D2D Trips CTPP Total Trips CTPP (211 MAG Model CTPP Trips) / Total CTPP Trips District Total 1.5%.4% -.1%.% -.3%.5%.19%.11% -.1% -.2% -.2%.%.5% 2.1%.18%.9%.25%.4%.3%.8% -.4% -.2% -.4% -.5% -.1%.1% 3.%.3% -.8%.7%.5%.% -.2% -.1%.%.13% -.3%.1%.% 4 -.1% -.21%.4%.2%.2% -.35% -.1% -.3%.6% -.2% -.13% -.3% -.1% 5 -.2% -.5% -.7%.17% -.2% -.4% -.17% -.6%.2% -.12% -.16% -.3% -.2% 6 -.3% -.1% -.7% -.51%.41% -.1% -.29% -.16% -.4% -.27% -.36% -.5% -.3% 7.4%.12% -.1% -.1% -.24%.18%.6%.% -.1% -.7% -.11% -.1%.4% 8.3% -.4% -.2% -.3%.1%.13%.2%.14%.2% -.4% -.3% -.1%.3% 9.% -.6% -.3% -.11%.3% -.11% -.4% -.3%.36% -.4%.% -.6%.% 1 -.1% -.1%.3%.67%.6% -.2% -.8% -.4%.22% -.37%.38% -.7% -.1% % -.13% -.5%.12% -.15% -.25% -.1% -.4%.63%.19%.64% -.19% -.3% 12.% -.5%.1%.11%.8% -.6% -.3% -.3%.46%.1%.67% -.69%.% Total.2% -.36% -.16% 1.48%.43% -.98% -.48% -.19% 1.69% -.57%.79% -1.16%.52% 24

25 Modeled Modeled versus CTTP Commuter Flows 14, 12, 1, y = 1.29x R² = , 6, 4, 2, 2, 4, 6, 8, 1, 12, 14, CTPP %RMSE=27.66 Each data point represents number of trips for a district-to-district trip interchange. 25

26 Distribution of Commuter Trips from District 1 Distribution of Commuter Trips to District Distribution of Commuter Trips from District 2 Distribution of Commuter Trips to District

27 Distribution of Commuter Trips from District Distribution of Commuter Trips to District 3 Distribution of Commuter Trips from District 4 Distribution of Commuter Trips to District

28 .6 Distribution of Commuter Trips from District 5.6 Distribution of Commuter Trips to District Distribution of Commuter Trips from District 6 Distribution of Commuter Trips to District

29 Distribution of Commuter Trips from District 7 Distribution of Commuter Trips to District Distribution of Commuter Trips from District 8 Distribution of Commuter Trips to District

30 Distribution of Commuter Trips from District 9 Distribution of Commuter Trips to District Distribution of Commuter Trips from District 1.5 Distribution of Commuter Trips to District

31 .5 Distribution of Commuter Trips from District 11.5 Distribution of Commuter Trips to District Distribution of Commuter Trips from District 12 Distribution of Commuter Trips to District

32 Takeaways Comparison between models and CTPP travel data is possible despite some apparent differences in population and employment statistics and the way data is produced. Comparison with CTPP data provides valuable insight and can be used as a validation tool for trip distribution purposes, for example. In the scarcity of independent data sets such validations provide a unique additional layer of checking trip distributions on regional level. Careful analysis of modeling and CTPP data should precede comparisons. Districts delineations should provide for sufficient level of aggregation to allow for meaningful comparisons and should be based on network topology, socio-economic and travel characteristics in the region. In large regions proper development of the regional travel demand sub-models and comparisons should ensure relatively close match with aggregated CTPP flows with RMSE below 3% and R 2 exceeding.9. 32

33 Possible Next Steps Analyze possible changes to CTPP products to make annual comparisons more meaningful. Experiment with different district delineations and more detailed districts. Establish an iterative process to delineate districts leading to more homogeneous districts in terms of number of intra-district and inter-district trips. 33

34 Questions? 34