Tiernan Erickson U.S. Census Bureau

Using a Cellular Automata Urban Growth Model to

Estimate the Completeness of an Aggregated Road Dataset

Tiernan Erickson

U.S. Census Bureau

Address Canvassing:Census workers compare what they see on the ground

to what is shown on the Census Bureau's address list. Based on their findings, the census workers will verify,

update, or delete addresses already on the list, and add addresses that are missing from the list.

At the same time, they will also update maps so they accurately reflect what is on the ground.

Housing unit addresses verified: 145 millionCensus workers hired for address canvassing: 140,000

Background

Source: U.S. Census Bureau. Address Canvassing Facts/Statistics. Retrieved June 16, 2012, from http://2010.census.gov/ news/press-kits/one-year-out/address-canvasing/address-canvassing-facts-statistics.html

Geographic Support System (GSS) Initiative:Integrated program in support of the 2020 Census:

Improved address coverageContinual spatial feature updatesEnhanced quality assessment and measurement

A targeted, rather than full, address canvassing operation during 2019 in preparation for the 2020 Census.

Collaboration with federal, state, local, and tribal governments and other key stakeholders to establish an acceptable address list for each geographic entity.

Background

Source: U.S. Census Bureau. Geographic Support System (GSS) Initiative.Retrieved June 16, 2012, from http://www.census.gov/geo/www/gss/index.html

Geographic Support System (GSS) Initiative:Background

Positional Accuracy Thematic Accuracy Temporal Accuracy Logical Consistency

Completeness?

Spatial Data Completeness

Detroit, MI Source: Google Maps

Spatial Data Completeness

South of Austin, TX Source: Google Maps

Spatial Data Completeness

8

Source: Project Gigalopolis http://www.ncgia.ucsb.edu/projects/gig/v2/About/abImages/apps/wash-balt_1792-2100.htm

Urban Growth Forecasting Models

9

Image Source: Cutsinger (2006)

Urban Growth Forecasting Models

Cellular Automata Urban Growth ModelsGenerate realistic urban patternsIntegrate the modeling of the spatial and temporal dimensions of urban processes.-Santé, et al. (2010)

10

Urban Growth Forecasting Models

11

SLEUTH ModelS - SlopeS - Slope

L - LanduseL - Landuse

E - ExclusionE - Exclusion

U - Urban ExtentU - Urban Extent

T - TransportationT - Transportation

H - HillshadeH - Hillshade

12

SLEUTH ModelS - SlopeS - Slope

L - LanduseL - Landuse

E - ExclusionE - Exclusion

U - Urban ExtentU - Urban Extent

T - TransportationT - Transportation

H - HillshadeH - Hillshade

13

SLEUTH ModelS - SlopeS - Slope

L - LanduseL - Landuse

E - ExclusionE - Exclusion

U - Urban ExtentU - Urban Extent

T - TransportationT - Transportation

H - HillshadeH - Hillshade

14

SLEUTH ModelS - SlopeS - Slope

L - LanduseL - Landuse

E - ExclusionE - Exclusion

U - Urban ExtentU - Urban Extent

T - TransportationT - Transportation

H - HillshadeH - Hillshade

15

SLEUTH ModelS - SlopeS - Slope

L - LanduseL - Landuse

E - ExclusionE - Exclusion

U - Urban ExtentU - Urban Extent

T - TransportationT - Transportation

H - HillshadeH - Hillshade

16

SLEUTH ModelS - SlopeS - Slope

L - LanduseL - Landuse

E - ExclusionE - Exclusion

U - Urban ExtentU - Urban Extent

T - TransportationT - Transportation

H - HillshadeH - Hillshade

Model Parameters (“Urban DNA”):DiffusionBreedSpreadSlope ResistanceRoad Gravity

SLEUTH ModelSLEUTH Model

SLEUTH Model

Source: Clarke et al. (1997)

SLEUTH Model

Source: Clarke et al. (1997)

20

Source: Project Gigalopolis http://www.ncgia.ucsb.edu/projects/gig/v2/About/abImages/apps/wash-balt_1792-2100.htm

SLEUTH Model

SLEUTH-3r More efficientJantz et al. (2009)Used to model entire Chesapeake Bay drainage Uses different measures of “fit” to compare prediction with actual for calibration and validationSLEUTH and SLEUTH-3r are free and run on Unix (Cygwin)Requires 1.5G RAM

Use SLEUTH-3r NLCD available for 1992, 2001, and 2006.Calibration:1992 – 2001

Prediction:2006(est.)

Validation:2006(est.) vs. 2006 (actual)

Methods

23

Methods

24

Expected Project Output

SLEUTH's Output: Series of rasters showing percent likelihood of new development for each cell, for each year between 2001 and 2006

Research Product: Aggregate prediction values to the tract level. For each tract last updated more than a year previous to 2006, the percent-likelihood of development will be summed for each year since last updated. If the sum total of unaccounted-for growth is above a threshold, then the tract is in need of updating.

25

Significance

Estimate of completeness of aggregated road dataset (TIGER)

Incomplete in areas where:1) Road growth is occurring rapidly, and 2) Have not been updated recently

Complete (save resources) in areas where:1) Little or no growth, or 2) Have been updated recently

26

LimitationsSLEUTH does not get into the causes of urban growth (as inputs).Instead, focuses on measuring and predicting a region's growth pattern ("Urban DNA") regardless of underlying causes.

DiffusionBreedSpread

SLEUTH does not put constraints on growth, such as:Population Growth ProjectionsEconomic Growth Projections

Extrapolates from previous growth. Uses Self-Modification Rules ("Boom" and "Bust") to produce realistic S-shaped growth curve projections based on recent growth, but not constrained to match other models' projections.

Slope ResistanceRoad Gravity

27

LimitationsAre SLEUTH's predictions of urban growth a satisfactory proxy for predictions of road network growth?

This study will provide an answer to that questionCompare actual 2006 road network growth using same

goodness-of-fit metrics that SLEUTH uses for Validation.

28

PossibilitiesAdapt model to constrain the outputs to match population or economic growth projectionsAdapt the model to make use of demographic inputsNew NLCD data (2011) scheduled for release in December 2013

Updated projections for the rest of this decadeImagery for specific areas could be processed to create more frequent land cover datasets with which to update predictions.

29

PossibilitiesEventually it could be useful to model urban growth for the entire country.

SLEUTH's creator, Keith C. Clarke, has said that he would like to see the model used for the entire United States (Clarke, 2008 and 2011).

Jantz et al. 2009 study of the entire Chesapeake Bay watershed (208 counties) remains the largest application of SLEUTH to date that I found in the literature.

An eventual nation-wide simulation could provide estimates of completeness of coverage for TIGER that could support the Census Bureau's stated goals for targeted update operations.

30

Next StepsDownload and install Cygwin (Unix environment for Windows)Read Cygwin documentationDownload and install SLEUTH softwareRead Cygwin documentationDownload remaining TIGER datasets (Tract, All LInes)Convert TIGER/Line files to shapefilesMerge DEMs for test countyRun SLEUTH on test county, probably my hometown for familiarity:

Pima County, AZ (04019)Estimate reasonable number of counties to processSelect counties for studyStreamline data download, setup, model run processDownload data for additional countiesRun SLEUTH model on countiesValidate output to 2006 land coverValidate 2006 roads to 2006 land coverSimulate TIGER update dates by TractCompare update dates to urban growth predictions by TractCreate percent-likelihood-incomplete estimates by tract

31

Questions? Comments?