HongKong Punkka Salo

Selecting Forest Sites for Voluntary Conservation in Finland :  Selecting Forest Sites for Voluntary Conservation in Finland Antti Punkka and Ahti Salo Systems Analysis Laboratory Helsinki University of Technology P.O. Box 1100, 02015 TKK, Finland http://www.sal.tkk.fi/ [email protected] Outline:  Outline Pilot projects for voluntary forest conservation in Finland Decision analytic observations about pilot projects Site selection procedures Decision support models for sites’ biodiversity How Robust Portfolio Modeling (RPM, Liesiö et al. 2006) can be used in the evaluation and selection of forest sites? Voluntary Conservation in Finland:  Voluntary Conservation in Finland Five pilot projects in METSO program (2003-2007) Objective to protect forest biodiversity in Finland Habitat-oriented instead of species-oriented Led by two ministries in cooperation Voluntary conservation in pilot projects Fixed-term deals (usually 10 years) against monetary compensation Finland: population 5.3M, area 338000 km2 Cf. Hong Kong: population 7M, area 1100 km2 A lot of forest (76 % of area), a lot of private land-owners Our task is to evaluate pilot projects from a decision analytic perspective and give recommendations for future Funding Ministry of Agriculture and Forestry Selection of Conservation Sites:  Selection of Conservation Sites Mix of resource allocation and multicriteria decision-making: How to model the biodiversity of the resulting portfolio (network)? Additivity of value functions Network’s value with regard to sites? Sites’ values with regard to criterion-specific values? Which sites of different costs should be selected with regard to multiple criteria, subject to a limited budget? DA / Optimization Methods in Reserve Site Selection:  DA / Optimization Methods in Reserve Site Selection Several optimization models with one criterion Maximize # of species subject to a limited # of sites Minimize # of sites such that predefined species occur on these sites Potentially optimal networks + SMART/MOP (Memtsas 2003) SMART and multiobjective programming (distance from utopian vector) to compare potentially optimal networks Sensitivity analysis on weights Pareto optimal networks + modified AHP (Moffett et al. 2006) Modified AHP to compare Pareto optimal networks (approximation) Sensitivity analysis on weights Pilot Projects in Finland:  Pilot Projects in Finland Five pilots In the biggest pilot, some 400000 euros have been spent annually since 2003 Average monetary compensation about 200 euros / ha / year Land-owner’s expression of interest some information on the site’s conservation values Evaluation of the site estimation of biodiversity values (compensation estimate) Land-owner’s offer assistance provided (second evaluation) Negotiations, decision examination of one or several sites No deal Deal Selection Procedures in Pilot Projects:  Selection Procedures in Pilot Projects Site-by-site selection: candidates are accepted or discarded soon after evaluation and offer time Portfolio selection: selection is made at a later date from a group of many site candidates expression of interest evaluation specification of offer decision Decision Analysis in Voluntary Conservation:  Decision Analysis in Voluntary Conservation Design of a decision analytic selection procedure: 'Site-by-site' or through portfolio analysis? … or something between these? Evaluation of sites Accuracy of data / evaluations Modeling of sites’ conservation values Decision support Selection of sites Differences between Selection Procedures (1/2):  Differences between Selection Procedures (1/2) Number of evaluations Costly Target of choosing the best site network Spatial aspects Decision delay Information about unselected (but feasible) sites Candidates’ prevailing biodiversity values Differences between Selection Procedures (2/2):  Differences between Selection Procedures (2/2) Portfolio selection tends to be more cost-effective than site-by-site selection if: Site-specific cost of evaluation is not very high The share of infeasible site candidates is not very high The budget is not too small Multi-Criteria Modeling in Pilot Projects:  Multi-Criteria Modeling in Pilot Projects Multi-criteria methods used to Form compensation estimates for forest owners Evaluate site candidates Support selection Additive models based on several conservation values Area, dead wood, distance to other conservation sites, rare species regarded as criteria Weights wi represent relative importance of criteria Deficiencies in Pilot Projects’ Multi-Criteria Models:  Deficiencies in Pilot Projects’ Multi-Criteria Models Lack of sensitivity analysis Use of point estimates for scores and weights leads to a single overall value for a site Piecewise constant value functions Network requirements not explicitly accounted for E.g. the total area of selected sites must be at least 250 ha Figure: valuation of logs Preference Programming: Incomplete Information:  Preference Programming: Incomplete Information Site characteristics The volume of dead wood on site x is between 8 and 11 m3 Relative importance of criteria E.g. Salo and Hämäläinen (2001), Salo and Punkka (2005) Area is more important than landscape values Dead wood is the most important criterion If the maximum value w.r.t. area is 20, max value w.r.t. burned wood is between 80 and 120 Feasible Weights and Scores:  Feasible Weights and Scores In the absence of information feasible criterion weights and scores belong to Incomplete information (linear constraints) leads to subsets Information set Supporting Site Network Selection with RPM:  Supporting Site Network Selection with RPM Incomplete information Subset of sites = a site network = a portfolio p Select a feasible site network p to maximize overall value with budget B Additive, consistent with value tree analysis Comparing Site Networks: Dominance Relation:  Comparing Site Networks: Dominance Relation No unique overall values  no unique optimal portfolio usually Portfolios compared through dominance relation Non-Dominated Portfolios:  Non-Dominated Portfolios Portfolios that are not dominated by any other portfolio Figure: n = 2, fixed scores w1 within the interval [0.4, 0.7] p1 dominates p2 p1 and p3 non-dominated Non-dominated portfolios of interest No other feasible portfolio has greater overall value across the information set Non-dominated portfolios with information S’S are a subset of non-dominated portfolios with S Not necessarily potentially optimal RPM – Site Oriented Analysis:  RPM – Site Oriented Analysis Sites that belong to every non-dominated site network: Core sites If excluded, the selected network is dominated  include Sites that do not belong to any non-dominated site network Exterior sites If included, the selected network is dominated  exclude Borderline sites belong to some but not all non-dominated networks Core index of site Share of non-dominated portfolios in which a site is included (CI=0%-100%) RPM Framework:  Approach to promote robustness through incomplete information (integrated sensitivity analysis). Accounts for group statements RPM Framework Decision rules, e.g. minimax regret •Narrower intervals •Stricter weights • Score intervals • Loose weight statements Large number of site candidates. Evaluated w.r.t. multiple criteria. Border line sites 'uncertain zone' Focus Exterior sites 'Robust zone'  Discard Core sites 'Robust zone'  Choose Core Border Exterior Negotiation. Manual iteration. Heuristic rules. Selected Not selected Example: Sensitivity of Recommendations (1/3):  Example: Sensitivity of Recommendations (1/3) Incomplete ordinal information Importance-order of criteria groups (6) known No stance is taken on the order of importance within the groups Criteria with same w* form a group 20, 15 and 10 % intervals E.g. with 10 % interval the weight of old aspens (0.120) is allowed to vary within [0.9 x 0.120, 1.1 x 0.120] = [0.108, 0.132] Data Real data on 27 selected sites with criterion-specific values (non-normalized) Weights (wi*) and scores derived from criterion-specific values Budget 50 % of sum of offers Example: Sensitivity of Recommendations (2/3):  Example: Sensitivity of Recommendations (2/3) Effect of weight perturbation Example: Sensitivity of Recommendations (3/3):  Example: Sensitivity of Recommendations (3/3) Differences between ND networks with 10 % intervals Examine site candidates in more detail Spatial aspects? Choose sites with highest core index (6/7) ND #3, ND #4 and ND #6 become 'infeasible' Decision rules (Salo and Hämäläinen 2001) recommend network 'ND #6' Precise weights w* lead to solution 'ND #7' Possibilities of RPM in Reserve Site Selection:  Possibilities of RPM in Reserve Site Selection Design of DA selection procedure: 'Site-by-site' or portfolio? Synergies and network requirements can be explicitly included Evaluation of sites Incomplete information on sites’ characteristics Information on how further evalution efforts should be focused effectively Modeling of sites’ conservation values Generic model Additive models widely used and easy to understand Incomplete information on weights Selection of sites A priori sensitivity analysis Several robust decision recommendations References :  References Liesiö, J., Mild, P., Salo, A., (2005). Preference Programming for Robust Portfolio Modeling and Project Selection, European Journal of Operational Research, (to appear). Memtsas, D., (2003). Multiobjective Programming Methods in the Reserve Selection Problem, European Journal of Operational Research, Vol. 150, pp. 640–652. Moffett, A., Dyer, J. S., Sarkar, S. (2006). Integrating Biodiversity Representation with Multiple Criteria in North-Central Namibia Using Non-Dominated Alternatives and a Modified Analytic Hierarchy Process. Biological Conservation, Vol. 129, pp. 181–191. Salo, A., Hämäläinen R. P. (2001). Preference Ratios in Multiattribute Evaluation (PRIME) – Elicitation and Decision Procedures under Incomplete Information. IEEE Transactions on Systems, Man, and Cybernetics – Part A: Systems and Humans, vol. 31, s. 533–545. Salo, A., Punkka, A., (2005). Rank Inclusion in Criteria Hierarchies, European Journal of Operational Research, Vol. 163, pp. 338–356.