Integrating Economics and Ecology: A Case Study of LandUse Policies to Reduce Habitat Fragmentation

1
Integrating Economics and Ecology A Case Study
of Land-Use Policies to Reduce Habitat
Fragmentation

• Andrew J. Plantinga
• Department of Agricultural and Resource Economics
• Oregon State University
• IGERT Colloquium
• October 19, 2005

2
Forest Fragmentation by Non-forest Uses
3
Consequences of Forest Fragmentation

• Recent estimates indicate that one-fourth of U.S.
  bird species are declining in population.
• For interior-forest birds, habitat fragmentation
  is a central factor, particularly in forests of
  the eastern U.S.
• Neo-tropical migratory songbirds are particularly
  affected.

4
Effects of Forest Fragmentation on
Interior-Forest Birds

• Bird densities and breeding success lower in
  small patches of forest
• Predators and brood parasites from surrounding
  non-forest lands can penetrate a greater
  proportion of a small forest patch than a large
  forest patch
• Predators include raccoons and house cats from
  urban environments. Brood parasites include the
  brown-headed cowbird from agricultural lands.

5
Measures of Forest Fragmentation Relevant to
Interior-Forest Birds
Patch Size
Core Forest
6
Research Questions

• How can the costs of reducing forest
  fragmentation be quantified?
• How do these costs vary between spatially-uniform
  and spatially-targeted policies?
• How do initial landscape conditions affect the
  costs of different policies?

7
Overview of Methods
Estimate an econometric model of private land-use
decisions with plot-level data
Transition probabilities expressed as functions
of net returns to alternative uses, soil quality,
etc.
Link transition probabilities to GIS data on
actual landscapes
Simulate the effects of market-based policies on
the spatial structure of landscapes
8
Distinguishing Features of this Study

• Simulations rely on probabilistic transition
  rules
• Examine effects of market-based policies
• Major land uses (agriculture, forest, urban)
  considered
• Large geographical area

9
What We Dont Do

• Account for spatial relationships that influence
  private land-use decisions
• Model population dynamics of bird species
• Model the effects of forest management on bird
  habitat

10
Study Region The Coastal Plain of South Carolina

• Privately-owned land dominates the landscape
  (83).
• Composition of private landscape
• Forest 57
• Agriculture 21
• Urban 5
• Forestland is fragmented by both agriculture and
  urban uses.
• Forest fragmentation is widely considered to be a
  threat to the regions migratory bird populations.

11
Forest Land in the Study Region
12
Econometric Model

• Plot-level land-use decisions modeled as a
  discrete-choice problem
• Model explains conversions between forest,
  agriculture, and urban uses on private land.
• Land is assumed to be allocated to the use
  generating the greatest net revenues.

13
Data Sources Econometric Model

• Land-use data
• National Resources Inventory (NRI) plot-level
  data on land-use and land characteristics. Model
  estimated with data on North and South Carolina.
• Economic Returns
• Lubowskis (2002) national data set on net
  returns to land at the county level.
• Parcel-level variation is accounted for with
• Soil quality (NRI)
• Urban Influence (ERS)

14
Econometric Estimates

• Key parameter estimates are significantly
  different from zero and consistent with
  expectations.
• Transition probabilities vary by
• Starting use
• County
• Parcel-level soil quality.
• Parcel-level urban influence.

15
Using Transition Probabilities in Landscape
Simulations

• GIS data needed to match the variables in the
  econometric model.
• GIS data (SCDNR)
• Land use (from SCDNR in conjunction with U.S.
  Fish Wildlife)
• Soil quality (from National Cooperative Soil
  Surveys)
• Public lands
• Urban influence (from ERS)

16
GIS Layer on Land Use (quad level)
17
Landscape Simulations with Probabilistic
Transition Rules

• For a given agricultural parcel, assume
  prob(forest) 0.1, prob(urban) 0.1 and
  prob(agriculture) 0.8
• If the landowner was faced with the same choice
  situation many times, they would convert to
  forest 10 of the time, to urban 10 of the time,
  and remain in agriculture 80 of the time.
• Simulations use a random number generator to
  repeat the choice situation many times for all
  parcels in the landscape.

18
Two Simulated Landscapes
19
Quantifying Spatial Pattern

• We summarize the spatial configuration of forest
  with fragmentation indices computed with Fragstat
  software.
• Two indices of relevance to interior-forest
  birds
• Proportion of the landscape in core forest
• A parcel is core if it is gt200m from nearest
  non-forest edge.
• Average forest patch size
• The total forest area on a landscape divided by
  the number of spatially distinct patches.

20
Distributions Over Fragmentation Indices
21
Market-Based Policies to Reduce Forest
Fragmentation

• Baseline and policy scenarios conducted over a
  25-year horizon
• Spatially uniform per-acre subsidy for conversion
  of agricultural land to forest
• Spatially targeted subsidies to agricultural
  parcels that are adjacent to forest.
• ST1 Subsidy offered if parcel shares a border
  with one or more forest parcels.
• ST3 Subsidy offered if parcel shares a border
  with three or more forest parcels.

22
Tradeoffs Between Policy Types

• The uniform policy selects the least-cost parcels
  for conversion to forest, but may have limited
  effects on the fragmentation metrics
• The ST1 policy is more expensive because it
  selects from a subset of agricultural parcels,
  but all converted parcels increase the average
  patch size.
• The ST3 policy selects from an even more
  restricted set, but may be more likely to
  increase core forest.

23
Effect of a 25 Uniform Subsidy on the Core
Forest Distribution
24
Marginal Cost of Increasing Average Patch Size
(50 initial forest)
25
Marginal Cost of Increasing Core Forest (50
initial forest)
26
Marginal Cost of Increasing Average Patch Size,
by Initial Forest Cover
27
Conclusions

• Spatially-uniform subsidies can have a
  significant effect on forest fragmentation
• A 25 per acre subsidy would increase forest area
  in the region by about 7
• The area of core forest would increase by 3.5
• The average forest patch size would increase by
  65
• For all policies, marginal costs tend to decline,
  except for the heavily forested landscape

28
Conclusions

• Spatially-uniform subsidies perform well relative
  to spatially-targeted policies
• Marginal costs are comparable to or slightly
  above the costs of the ST1 policy for increasing
  average patch size.
• Marginal costs are lower than with the
  spatially-targeted policies for increasing core
  forest
• Largest cost differences not between policy types
  but between initial landscape conditions
• Marginal costs drop significantly as the initial
  amount of forest increases

29

• Thank you

30
Changes in Land Use in North and South Carolina
1982 - 1997

• Acres converted from forest to urban 1,346,700
  (4.86)
• Acres converted from forest to agriculture
  504,500 (1.82)
• Acres converted from agriculture to forest
  1,143,800 (8.99)
• Net decrease in forest of 707,100 acres (2.55)

Source National Resources Inventory
31
Econometric Model

• Specification

Rikt average net returns to use k in the county
where parcel i is located UIi urban influence
on parcel i LCCi land capability class rating
on parcel i
32
Econometric Estimation

• Plot-level data for three transitions
• Panel data estimation of a logit model is
  appropriate only if unobserved components of
  utility are uncorrelated over time (Train, 2003).
  This is unlikely in our case (e.g., distance to
  roads).
• Pooling strategy that provides some of the
  benefits of panel estimation without imposing the
  above restriction
• For parcels that remain in same use for three
  (two, one) periods, randomly select one-third
  (one-half, all) of the parcels from each time
  period.
• Models are estimated with data on parcels all
  beginning in the same use

33
Challenges with Modeling Spatial Dependence

• Data challenges
• Need time-series GIS land-use data on a large
  scale
• Time-series data for changes in land use
• Large scale for variation in net returns
• Methodological challenge of modeling spatial
  correlation in discrete-choice framework
• Recent advances in binary probit (Fleming 2004)
• Simulation approaches needed for multinomial
  models
• Difficult with large number of observations
• Problems with simulation bias

34
Determining the Number of Simulations

• Examined 5 representative quads and 6
  fragmentation indices.
• Considered how the confidence interval lengths
  for the first 3 moments of the distributions
  change with the number of simulations (Ross,
  1997). Lengths change very little beyond 500
  simulations.
• Generate two samples of 500 simulations each and
  test for differences in the first 3 moments of
  the distributions. Fail to reject the null
  hypothesis of no differences at the 1 level.

35
Explaining Differences in Marginal Costs
All agricultural parcels
Parcels with at least one forest neighbor
Parcels with at least three forest neighbors