Thinking spatially: Economic models of urban land use change

1
Thinking spatially Economic models of urban land
use change

• Elena G. Irwin
• Associate Professor
• Department of Agricultural, Environmental and
  Development Economics
• Ohio State University
• Presentation prepared for the conference on
  Spatial Thinking in the Social Sciences,
  University of Illinois, December 17-18, 2006.

2
Key points

• Pattern vs. process-based models of land use
  change
• Traditional geographic models emphasize pattern
  over process
• Traditional economic models emphasize process
  over pattern
• Qualitative changes in land use change patterns
  points out limitations of pattern-based
  geographic models
• Increased availability of fine-scale data points
  out limitations of highly stylized economic
  models
• We need hybrid models that combine process and
  pattern

3
Example pattern-based model of urban land use
change

• Cellular automaton urban growth model
• Non-behavioral model of land use cell transitions
  that are determined by relative geographic
  location of cell (spatial rules)

4
Washington Baltimore historical urban growth
(Urban Growth in American Cities - Glimpses of US
Urbanization, USGS Circular 1252, 2003 Available
online at http//landcover.usgs.gov/LCI/urban/data
.php
Source Clarke and Gaydos, 1998
5
How have economists traditionally represented
space?

• Space is typically represented in economic units
  vs. geographical units, e.g.
• Urban economics transportation costs
• New geographical economics regional economy
• Behavioral (i.e., process-based) models of
  economic agents (households or firms) that
  provide simple explanation and prediction of
  spatial pattern

6
Urban economic model of land use space as
transportation costs

• Monocentric model (or bid-rent model)
• Pre-determined central employment area
• Accessibility to central employment district
  drives firm and household location decisions
• Otherwise space is a featureless plane
• Predicts concentric ring of urban land use
  around central business district and declining
  density gradient

7
Monocentric model land use prediction
Undeveloped
Low density residential
Higher density residential
distance from city
8
Monocentric model land use prediction(distance
via major roads)
Undeveloped
Low density residential
Higher density residential
distance from city
9
Empirical test urban density gradient

• Empirical test of monocentric city model urban
  density gradient (Clark, 1951 Mills, 1972
  Edmonston, 1975)
• Assume negative exponential
• Estimate density gradient

-0.25
x distance from city D population density ?
density gradient
10
How well does this model describe actual patterns
of urban land use?

• Using population density gradient estimates,
  Anas, Arnott and Small (1998) estimate that the
  monocentric model explains approximately 63 of
  urban decentralization between 1950-70 in the US
• To what extent does this conclusion depend on
  spatial scale, geographical extent and type of
  data?

11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
Washington D.C. area population density vs. land
use (2000)
15
Explaining residential land use patterns (Irwin,
Bockstael and Cho, 2006)

• How well does basic monocentric model explain
  finer scale variations in residential pattern?
• Is there structural change across time? across
  urban-rural gradient? are results scale
  dependent?

16
Explaining residential land use patterns (Irwin,
Bockstael and Cho, 2006)

• Regression analysis using Maryland 1973 and 2000
  land use raster data (100 m cell size)
• Dependent variables undeveloped in 1 and 5 sq
  km neighborhoods
• Explanatory variables
• Distance via roads to major urban centers
• Distance via roads to suburban and small city
  centers
• Controls for local spatial heterogeneity (soil
  and topography)

17
Measure of residential pattern undeveloped in
neighborhood
18
Results reported for distance variables only
19
Urban-rural county typology
20
Results reported for distance variables only
21
Results using finer scale land use data

• Distance to city explains some of the variation
  in urban pattern
• Scale dependence distance explains about 30 of
  variation with larger neighborhood size vs. 15
  of variation with smaller neighborhood size
• Spatial heterogeneity in exurban areas, about
  93 of variation is unexplained vs. 49
  unexplained in suburban areas
• Other spatial processes matter, particularly at
  local scale and particularly in exurban areas
• Need explicit representation of geographic space
  to capture these other processes

22
Spatial interactions hypothesis (Irwin and
Bockstael, 2002)

• Can the fragmented pattern of development be
  explained as the result of interactions among
  developed land use parcels?
• Positive spatial externalities ? clustered
  pattern
• Negative spatial externalities ? scattered pattern

23
Data

• Geo-coded land parcel centroids from two Maryland
  exurban counties
• Seven year history of convertible parcels
  (1991-1997)
• Parcel characteristics zoning, network road
  distance to D.C., public sewer, soil, slope, etc.
  
• Neighborhood variable percent of residential
  land within a given buffer of each parcel
  centroid

24
Binary dependent variable 1 if converted in
time period t, 0 otherwise
25
predicted LU change
26
Accounting for multiple spatial processes

• Can spatial interactions be incorporated into
  monocentric model?
• No monocentric model simplifies space to one
  dimension (distance to city)
• Can distance be incorporated into a model of
  spatial interactions?
• Yes explicit representation of geographic space
  allows for consideration of multiple spatial
  processes

27
Hybrid models of process and pattern

• Process-based model agent decision making
• Pattern-based model agents are located in
  geographic space
• As a result, space can matter in multiple ways
• Spatial heterogeneity
• Distance (e.g., to employment, recreation)
• Spatial interactions and externalities
• Spatial scale, scale-dependent effects,
  cross-scale interactions

28
Hybrid models of process and pattern

• Made possible by
• Availability of finer scale land use/cover data
• Geographic data software
• Computational ability and methods

29
Some modeling challenges

• Hybrid models require a combination of
  theoretical, empirical and simulation approaches
• Theoretical challenges
• Identifying relevant spatial and temporal scales
• Accounting for interactions across spatial and
  temporal scales

30
Determinants of Household/Firm Location Land
Use Decisions (Irwin, 2006)
economic restructuring
living costs, agglomeration economics, labor
force, employment,
urban natural amenities
public services, infrastruc-ture, local policies
space
transportation and communications costs
household wealth
neighborhood amenities, zoning, access
land quality, public services, surrounding land
uses
time
31
Some modeling challenges (continued)

• Empirical challenges
• Identifying spatial processes vs. measurement
  error
• Data accuracy, appropriate data for question

32
(No Transcript)
33
Some modeling challenges (continued)

• Simulation challenges
• Specifying parameters and spatial environment
  (e.g., the right amount of spatial heterogeneity)
• Validating model specification
• Testing pattern hypotheses and summarizing model
  results

34
Some modeling methods

• Theoretical
• Complex systems theory
• Behavioral economics
• Empirical
• Pattern detection and metrics using GIS
• Spatial econometrics
• Simulation
• Agent-based (or multiagent) models and geographic
  automata systems
• Object-oriented programming