A MODEL TO PREDICT IMPERVIOUS SURFACE IMPACTS OF URBANSIM

A MODEL TO PREDICT IMPERVIOUS SURFACE IMPACTS OF
LAND USE AND TRANSPORTATION SYSTEM CHANGE WITH
URBANSIM
ISAAC LAWRENCE
Data Preparation, Methods, and Preliminary Results
Photo: http://www.unce.unr.edu/programs/sites/nemo/photos/index.asp?Photos=Gallery2
UrbanSim: Originally de-
Abstract: The importance of impervious surface area (ISA) as an indicator of
Methods:
veloped at the University of Washington by Paul Waddell and others,
UrbanSim is an agent-based behavioral simulation model of land use built
around a powerful, flexible, opensource modeling environment (Waddell 2011).
human impact on ecosystems and a driver of increases in flooding has been well established. In order to predict impervious surface outcomes for municipal and regional
Master Planning processes, Reilly et al. (2003) developed and tested a model of ISA
based on commonly available planning data. Since publication, adoption of agentbased land use and transportation models by planning authorities and researchers
has increased. UrbanSim, one increasingly popular model, provides a powerful, flexible environment for predicting land use and transportation system change. In order to
leverage UrbanSim towards the management of flooding and stream health with impervious surface as a proxy, I propose and test a model to predict ISA within a Chittenden County, Vermont implementation of UrbanSim. In addition, I intend to compare
the model developed to Reilly et al. as well as a simple factor model commonly utilized in hydrologic modeling.
Data:
Data preparation:
Data development work completed was largely in service of
UrbanSim 2005 base year model and involved preparation of 150m gridcell datasets for
all variables included below, as well as others. Variable distributions were examined and,
where relevant, variables were transformed to more closely match a normal distribution.
Preliminary Model Estimation: Preliminary stepwise linear regression
Efforts are currently underway at the
UVM Transportation Research Center
to implement a 2005 baseyear UrbanSim model for Chittenden County,
Vermont. Data used in this analyis
were initially developed as part of
this project.
was performed with variables listed right using JMP statistical software. Data presented in
table right is based on those prepared for use in the UrbanSim 2005 baseyear and according to findings in Reilly et al. (2003) concerning impervious surface area estimation for
New Jersey towns. Preliminary model performance was considered via the coefficient of determination, R^2, both at the 150m gridcell level and at two levels of aggregation – 1500m
gridcells and Chittenden County town boundaries.
Variable
Name
Percent_ISA
Percent_
floodplain
Percent_Canopy
Employees
ResUnits
Transformed
Variable
lnISA
None
Transforma- Data Source
tion
Natural Log NLCD 2006
None
CCRPC 2012
None
None
NCLD 2006
Natural Log
Natural Log
and Expontential
None
CCRPC 2008
CCRPC 2008
lnEmp
lnResUnits and
ln^2resUnits
NETdist_Air- None
port
Percent_wet- None
land
NETdist_Vil- None
lageCenter
NETdist_
None
Highway
SLdist_AirLnSLdist_Airport
port
SLdist_High- lnSLdist_
way
Highway
Percent_
None
roads
None
None
None
Natural Log
Natural Log
None
Calculated in
ArcGIS
NWI 2012
Calculated in
GIS
Calculated in
GIS
Calculated in
GIS
Calculated in
GIS
Pede 2013
Image: https://trac.urbansim.org/
Preliminary Results/Analysis
Model Estimation
150m Gridcells
Aggregate by Town (18)
Model Variables
Model
RSquare Ad(dependent in
RSquare RSquare Adjusted RSquare
Number
justed
italics)
percent_isa
constant
2
0.281
0.281
0.610
0.586
ln^2resUnits
percent_isa
constant
ln^2resUnits
Employees
Percent_roads
percent_isa
constant
ln^2resUnits
LnEmp
Percent_roads
logSLdist_airport
Aggregate Plots of Actual by Preidcted by
Actual Impervious Surface Area
Aggregate by 1500m Gridcells (723)
Aggregated By Town
Aggregated by 1500m Gridcell
RSquare RSquare Adjusted
0.655
0.654
Conclusions: Stepwise linear regression demonstrated that, not surprisingly,
percent_roads was by far the most important variable, while lnEmp and ln^2ResUnits were
also very significant. These results supported the findings of Reilly et al. (2003), with impervious road surface removed from their total impervious estimation (because, of course,
road surface matches impervious surface on a one to one basis). For the purposes of UrbanSim, however, a model including an estimate of road surface make sense.
Future Work:
3
0.438
0.438
0.769
0.755
0.760
0.759
5
0.612
0.612
0.894
0.887
0.862
0.862
• Coding and Estimation of model in UrbanSim with 2005 baseyear
• Improvement of aggregation and estimation methodologies
• Estimation with additional variables including a binary for whether or not gridcell is at
the Airport
• Spatial examination of model error to identify additional candidate variables
A Note on Variable Selection: Higher RSquared
values were obtained by adding additional available variables
from the data (see table “Data”), but effects were small or, as in
the case of the percent_canopy dataset, derivation from the same
data as the dependent variable rendered the independent variable problematic for use.
Acknowledgements: This research was funded by the U.S DOT through the University Transportation Research Center Program.
Additional thanks to Brian Voigt, Tim Pede, and Austin Troy for work on getting the 2005 baseyear UrbanSim model of Chittenden County, Vermont up and running (or nearly so).
Works Cited:
James Reilly, Patricia Maggio, Steven Karp, (2003) “A model to predict impervious surface for regional and municipal land use planning purposes,” Environmental Impact Assessment Review, Volume 24, Issue 3, April 2004, Pages 363-382, ISSN 0195-9255, 10.1016/jeiar.2003.10.022.
Waddell, Paul(2011) ‘Integrated Land Use and Transportation Planning and Modelling: Addressing Challenges in Research and Practice’, Transport Reviews, 31: 2, 209 — 229
Pede, Tim (2013), ‘Percent_Highway Methods,’ Unpublished.
UNIVERSITY OF VERMONT TRANSPORTATION RESEARCH CENTERBURLINGTON, VERMONT