Selecting Forest Sites for Voluntary Conservation with Robust Portfolio Modeling

1
Selecting Forest Sites for Voluntary Conservation
with Robust Portfolio Modeling

• Antti Punkka, Juuso Liesiö and Ahti Salo
• Systems Analysis Laboratory
• Helsinki University of Technology
• P.O. Box 1100, 02150 TKK, Finland
• http//www.sal.tkk.fi/
• firstname.lastname_at_tkk.fi

2
METSO Program

• Objective is to protect biodiversity in forests
  of Finland
• Southern Finland, Lapland, Province of Oulu
• Lead by Ministry of Agriculture and Forestry in
  cooperation with the Ministry of the Environment
• Subprograms include testing of voluntary
  conservation methods

3
Pilot Projects for Voluntary Conservation (1/4)

• Five pilots
• Forest owners offer their sites for conservation
  against monetary compensation
• In Satakunta pilot, 400000 euros have been spent
  annually since 2003 to preserve a total of some
  2400 ha for 10 years
• Usual process
• The forest owners are informed about voluntary
  conservation methods
• Owners express their interest
• Preliminary assessment of the site together with
  the owner
• The owner makes an offer (help provided)
• Negotiations and selection

4
Pilot Projects for Voluntary Conservation (2/4)

• Multi-criteria methods used to
• Form compensation estimates for forest owners
• Evaluate sites
• Additive scoring models for conservation values
• Value tree analysis
• Value of a site is the sum of its
  criterion-specific values
• or a weighted average of normalized
  criterion-specific values (scores)
• Weights wi represent trade-offs between criteria

5
Pilot Projects for Voluntary Conservation (3/4)

• Example sites value is the sum of its values of
  area, dead wood, distance to other conservation
  sites and burned wood

Vha(x) denotes the value of site x per hectare
6
Pilot Projects for Voluntary Conservation (4/4)

• Limitations of pilot projects models
• Lack of sensitivity analysis
• use of point estimates for scores and weights
  leads to a single overall value for a site
• Piecewise constant criterion-specific value
  functions
• e.g., landscape values are subjective
  evaluations, where especially discontinuous value
  functions may cause big differences among
  experts evaluations
• e.g., 4.6 m3/ha of conifer snags is 150 more
  valuable than 4.4 m3/ha, which is as valuable as
  2.0 m3/ha
• One-by-one selection of sites
• aim of choosing a good portfolio may be missed
• possible inefficient use of budget
• Structural requirements not explicitly accounted
  for
• e.g., the total area of sites selected must be at
  least 250 ha

7
Preference Programming

• Some limitations can be addressed with the use of
  incomplete information
• The relative importance of criteria can be set as
  intervals or as a rank-ordering of the importance
  of criteria
• e.g., increase of 1 m3/ha in dead wood is at
  least as important as increase of 1 m3/ha in
  burned wood
• e.g., dead wood is the most important criterion
• Sites can be evaluated with incomplete
  information about their characteristics
• e.g., the sites landscape values are between 5
  and 10 on scale 0-20
• Set of feasible parameter values (weights,
  scores)
• The overall values become intervals

8
Site Selection Problem

• Which of the m independently evaluated sites
  should be selected, given budget B?
• Subset of sites is a portfolio
• Select a feasible site portfolio p to maximize
  overall value
• Portfolio preferred to another if it has greater
  overall value

9
RPM - Robust Portfolio Modeling (1/4)

• Combines Preference Programming with portfolio
  selection
• Use of incomplete information no precise overall
  values...
• Portfolios compared through dominance relations
• portfolio p is dominated, if there exists another
  portfolio p that has a higher overall value for
  all feasible scores and weights
• Dominated portfolio should not be selected, since
  there is another portfolio that is better for
  every feasible parameter combination
• and thus no unique optimal portfolio
• Non-dominated portfolios are of interest
• For a non-dominated portfolio, there is not
  another feasible portfolio with a greater overall
  value across the feasible weights and scores

10
RPM - Robust Portfolio Modeling (2/4)

• Portfolio-oriented selection
• Consider non-dominated site portfolios as
  decision alternatives
• Decision rules Maximax, Maximin, Central values,
  Minimax regret
• Methods based on exploring the set of
  non-dominated portfolios
• e.g., adjustment of aspiration levels
• Site-oriented selection
• Portfolio is a set of site-specific yes/no
  decisions
• Site compositions of non-dominated portfolios
  typically overlap
• Which sites are incontestably included in a
  non-dominated portfolio?
• Robust decisions on individual sites in the light
  of incomplete information

11
RPM - Robust Portfolio Modeling (3/4)

• Core index of site
• Share of non-dominated portfolios in which a site
  is included (CI0-100)
• Site-specific performance measure in the
  portfolio context
• accounts for competing sites and scarce resources
• Core sites are included in all non-dominated
  portfolios (CI100),
• Exterior sites are not included in any of the
  nd-portfolios (CI0),
• Border line sites are included in some of the
  nd-portfolios (0ltCIlt100),

12
RPM - Robust Portfolio Modeling (4/4)
Decision rules, e.g. minimax regret
Selected
Large numberof sites. Evaluated w.r.t. multiple
criteria.
Core sites Robust zone ? Choose

• Border line sitesuncertain zone
• Focus

Core
Wide intervals Loose weight statements
Narrower intervals Stricter weights
Border
Not selected
Exterior
Exterior sitesRobust zone ? Discard
Negotiation. Manual iteration. Heuristic rules.
Approach to promote robustness through incomplete
information (integrated sensitivity
analysis). Account for group statements
13
Illustrative Example (1/5)

• Real data in form of criterion-specific values
• 27 sites that were selected in Satakunta in 2003
• 227 over 134 million possible portfolios
• Evaluated with regard to 17 criteria
• criteria related to wood value excluded
• irrelevant criteria ( all sites have the same
  value) excluded
• some criteria united (e.g. logs and snags are
  dead wood)
• Here 9 evaluation criteria
• area, dead wood, landscape values, etc.
• Point estimate weights and scores derived from
  the criterion-specific values
• Sum of offers some 300,000 euros
• offers between 130 and 300 euros / ha / year
• Budget 25, 50 or 75 of sum of offers

14
Illustrative Example (2/5)

• Data / values

15
Illustrative example (3/5)

• Perturbation of weight estimates
• Five levels of weight accuracy
• Point estimates (no perturbation)
• 5, 10, 20 relative interval on the point
  estimates
• e.g. with 10 the weight of old aspens is
  allowed to vary within
• 0.9 x 0.120, 1.1 x 0.120 0.108, 0.132
• Incomplete ordinal information (the RICH method,
  Salo and Punkka 2005)
• 6 groups of criteria
• importance-order of the groups known
• no stance is taken on the order of
• importance within the groups
• criteria with same point estimate
• weights form a group

16
Illustrative Example (4/5)

• Core indexes (budget 50)

point estimates a unique solution
5 interval 2 non-d. portfolios
10 interval 6 non-d. portfolios
20 interval 24 non-d. portfolios
incomplete ordinal information 904 non-d.
portfolios
Site
17
Illustrative Example (5/5)

• Variation in budget (incomplete rank-ordering)

25
432 non-d. portfolios
50
904 non-d. portfolios
75
303 non-d. portfolios
Site
18
Conclusions Future Directions

• Robust Portfolio Modeling
• Sensitivity analysis with regard to criterion
  weights and sites characteristics explicitly
  included in the model
• means for subjective evaluation of qualitative
  criteria
• Selection of a full portfolio instead of
  one-by-one selection of sites
• synergies and minimum requirements can be
  explicitly included in the model
• Future task to develop a unified framework for
  selecting site portfolio
• Dedicated decision support system required

19
References

• Liesiö, J., Mild, P., Salo, A., (2005).
  Preference Programming for Robust Portfolio
  Modeling and Project Selection, submitted
  manuscript available at http//www.sal.hut.fi/Publ
  ications/pdf-files/mlie05.pdf
• Memtsas, D., (2003). Multiobjective Programming
  Methods in the Reserve Selection Problem,
  European Journal of Operational Research, Vol.
  150, pp. 640-652.
• Salo, A., Punkka, A., (2005). Rank Inclusion in
  Criteria Hierarchies, European Journal of
  Operational Research, Vol. 163, pp. 338-356.
• Stoneham, G., Chaudhri, V., Ha, A., Strappazzon,
  A., (2003). Auctions for Conservation Contracts
  An Empirical Examination of Victoria's BushTender
  Trial, The Australian Journal of Agricultural and
  Resource Economics, Vol. 47, pp. 477-500.
• Robust Portfolio Modeling site
  http//www.rpm.tkk.fi