Tradeoffs between forestry resource and conservation values under alternate policy regimes: A...

Tp

GMa

b

c

d

a

ARRAA

KPFSNRCF

1

kseetOr

Ff

s(w

0d

Ecological Modelling 221 (2010) 2590–2603

Contents lists available at ScienceDirect

Ecological Modelling

journa l homepage: www.e lsev ier .com/ locate /eco lmodel

radeoffs between forestry resource and conservation values under alternateolicy regimes: A spatial analysis of the western Canadian boreal plains

rant Hauera, Steve Cummingb,∗, Fiona Schmiegelowc, Wiktor Adamowicza,arian Weberd, Robert Jagodzinskia

Department of Rural Economy, 515 General Services Building, University of Alberta, Edmonton, AB T6G 2H1, CanadaDépartement des sciences du bois et de la forêt, Pavillon Abitibi-Price, Bureau 2133, Université Laval, Québec, QC G1V 0A6, CanadaDepartment of Renewable Resources, 751 General Services Building, University of Alberta, Edmonton, AB T6G 2H1, CanadaDepartment of Rural Economy, University of Alberta, Edmonton, AB and Alberta Research Council, 250 Karl Clark Road, Edmonton, AB T6N 1E4, Canada

r t i c l e i n f o

rticle history:eceived 9 February 2010eceived in revised form 12 July 2010ccepted 17 July 2010vailable online 20 August 2010

eywords:roduction possibility frontierorest managementpatial simulation

a b s t r a c t

An important element of resource management and conservation is an understanding of the tradeoffsbetween marketed products, such as timber, and measures of environmental quality, such as biodiver-sity. In this paper, we develop an integrated economic-ecological spatial optimization model that wethen apply to evaluate alternate forest policies on a 560,000 km2 study region of managed boreal for-est in Alberta and British Columbia, Canada. The integrated model incorporates dynamic forest sectorharvesting, current levels of oil and gas sector development, coarse-filter or habitat-based old forestindicators, a set of empirical forest bird abundance models, and statistical models of the natural andcurrent fire regimes. Using our integrated model, economic tradeoff curves, or production possibilityfrontiers, are developed to illustrate the cost of achieving coarse-filter targets by a set time (50 years)

atural disturbance regimeange of natural variationoarse-filter indicatorsine-filter indicators

within a 100-year time horizon. We found levels of ecological indicators and economic returns from thetimber industry could both be increased if spatial constraints imposed by the current policy environmentwere relaxed; other factors being equal, this implies current policy should be revised. We explore theproduction possibility frontier’s relationship to the range of natural variation of old forest habitat, andshow how this range can be used to guide choices of preferred locations along the frontier. We also showthat coarse-filter constraints on the abundance of certain habitat elements are sufficient to satisfy some

resse
fine-filter objectives, exp
. Introduction

A quantitative understanding of the tradeoffs between mar-eted products and environmental quality is fundamental toustainable management and biological conservation in managed
cosystems, such as our study region in the boreal forests of west-rn Canada. Environmental services, such as biodiversity, are notraded in markets so their prices cannot be determined directly.pportunity costs, measured by the monetary value of foregone
esource development, can be used to estimate the costs of policy

Abbreviations: AAC, annual allowable cut; FMA, forest management agreement;MU, forest management unit; NPV, net present value; PPF, production possibilityrontier; RNV, range of natural variation.∗ Corresponding author. Tel.: +1 418 656 2131x2593; fax: +1 418 656 5262.

E-mail addresses: [email protected] (G. Hauer), [email protected],[email protected] (S. Cumming), [email protected]. Schmiegelow), [email protected] (W. Adamowicz),[email protected] (M. Weber), [email protected] (R. Jagodzinski).

304-3800/$ – see front matter © 2010 Elsevier B.V. All rights reserved.oi:10.1016/j.ecolmodel.2010.07.013

d as the predicted abundances of various species of songbirds.© 2010 Elsevier B.V. All rights reserved.

impositions intended to maintain various levels of environmentalservices, biodiversity, or similar non-market goods. The questionsthen arise: What level of environmental services is desirable? Andat what time in the future will such a level be achieved? Answer-ing these questions entails a social choice across the feasible set ofalternatives for joint environmental and economic outcomes. Theproduction possibility frontier (PPF) is that subset of feasible alter-natives defined by the levels of different goods that can be obtainedfrom a system under optimality, such that the level of no individ-ual good can be increased without decreasing the levels of others.While PPFs can represent the tradeoffs between market and non-market goods and services that are feasible, it must be emphasizedthat they do not identify a “socially optimal” choice. Identifying asocially optimal point on the PPF requires an examination of peo-ple’s preferences. We do not take up this question here. We use
a spatial simulation model to estimate the PPF between revenuefrom forest management and indicators of biodiversity. We usedthe resultant PPF to quantify tradeoffs between biodiversity andforest products over a large study region of managed forest, locatedwithin the boreal plains ecozone (Ecological Stratification Working
dx.doi.org/10.1016/j.ecolmodel.2010.07.013

http://www.sciencedirect.com/science/journal/03043800

http://www.elsevier.com/locate/ecolmodel

mailto:[email protected]







dx.doi.org/10.1016/j.ecolmodel.2010.07.013

G. Hauer et al. / Ecological Modelling 221 (2010) 2590–2603 2591

udy re

GC

i(aeoepatc

tdbtaooaoe

Fig. 1. Location of the st

roup, 1996) in northern Alberta and northeast British Columbia,anada (Fig. 1).

Others have applied PPFs to resource management and biolog-cal conservation in forest ecosystems. For example, Nalle et al.2004) develops a PPF for the net present value of timber harvestnd the populations of two species with conflicting habitat prefer-nces. The authors use an unconstrained PPF to illustrate the costsf two management scenarios, the first concentrating conservationffort on public land only, and the second a static reserve selectionroblem. Both scenarios took spatial restrictions on timber harvestss given. In contrast, we focus on quantifying the effects of spa-ial constraints on the feasibility and economic cost of achievingonservation targets.

The development of integrated ecological and economic modelso support policy decisions is critical in the context of governmentecision making on public forest land. As we illustrate below, theoreal forest plains provide a case study for examining the applica-ion of such tools. Several studies have examined a single economicgent (e.g., a forestry firm) that maximizes the net present value
f forestry operations subject to constraints on alternative levelsf the conservation goods. These agents respond to constraints byltering management variables, such as the optimal rotation ager the total area harvested (e.g., Armstrong et al., 2003; Marshallt al., 2000). However, these models lack spatial representation
gion in western Canada.

of both economic and ecological processes. Our study is based ona reformulation of a novel spatial modeling tool (Cumming andArmstrong, 2004) designed to represent ecological processes andeconomic activity over very large areas as well as across jurisdic-tional and administrative boundaries. The ecological componentsof the model include empirical models of the natural and man-aged fire regimes over the study region and models of the relativeabundances of various species of forest songbirds. The economiccomponent of the model represents the behaviors of many individ-ual forest products firms, in particular their responses to changes inforest policy, markets, and the abundance and spatial distributionof resources.

The new contributions of this paper are as follows. First, we esti-mate the range of natural variation (RNV; Landres et al., 1999) inbiodiversity indicators, an ecological concept used in forest policyin Canada (Burton et al., 2006), to identify ecologically defensibleregions of the PPF and levels of the associated management actions.Second, we evaluate how spatial constraints consequent to govern-ment forest policies affect the shape and location of the tradeoff
relationship. Third, we consider how changes in such policies mayaffect the ability to achieve acceptable environmental outcomes.Finally, the paper presents a new, integrated modeling tool we feelhas significant potential for application to large-scale problems ofmanagement and conservation in the circumpolar boreal forest.

2 odelling 221 (2010) 2590–2603

las1ttBrutcdciab

ticoumffrdph

1p

hfouCohobfiedacifi1m1

sretawim

bt

592 G. Hauer et al. / Ecological M

With respect to the RNV criteria, boreal forests are subject toarge natural disturbances which change the spatial habitat andge structure. Within the western boreal plains, it appears that noteady state condition exists at any spatial extent (Cumming et al.,996). Moreover, no fixed target age class structure can be main-ained because the natural disturbance regimes continue to operateo at least some degree in managed forests (Armstrong et al., 2003).oreal species are presumed to be adapted to this stochastic envi-onment in the sense that they can maintain their populationsnder the disturbance regime (Hunter, 1993; Haila, 1994), evenhough the abundance and spatial distribution of preferred habitatshanges markedly over time. Hence, estimates of the range of con-itions that could occur under the natural disturbance regime areonsidered to provide ecological guidance for management targetsn the boreal and elsewhere (e.g. Landres et al., 1999). RNV criteriare explicit in regulatory requirements over much of Canada (seeelow).

On the economics side, a PPF, while providing a picture of theradeoffs, provides little guidance about which point on the curves best. Environmental valuation studies may provide estimates ofonservation benefits in monetary terms that may help to identifyptimal points along the curve. However, the environmental val-es examined in this study (biodiversity in the boreal forest) areostly “passive use values” or values that are cannot be measured

rom observed behavior. These values require structured surveysor measurement (Carson, 2000). In applications such as to theemote boreal forest, where respondents are being asked to makeecisions based on unfamiliar situations, an estimated RNV mayrovide an important complement to valuation studies and mayelp provide context for valuation analysis.

.1. Forest management and conservation in the western boreallains

Boreal landscapes in Alberta and northeast British Columbiaave undergone dramatic changes due to the expansion of the

orest sector since the 1980s and 1990s. The following outlinef the management environment is drawn from the Alberta sit-ation, but also applies in its essentials to northeastern Britisholumbia. Almost all forest lands outside of National Parks aren public lands. By the early 1990s, the most productive areasad been fully allocated to timber production under various formsf tenure. Forest lands are managed on a sustained yield basisy the determination of an annual allowable cut (AAC) to whichrms must closely adhere. Attempts to accommodate public inter-sts in non-timber values, such as wildlife, are made through aetailed planning and approval process, which regulates the size,djacency, location, and timing of harvest blocks. Other less spe-ific and less enforceable policy guidelines, such as those set outn the Alberta Forest Conservation Strategy (1994), recommendrms implement ecosystem-based forest management (Franklin,996), which includes establishing ecological benchmarks andaintaining the abundance of old forest within the RNV (AFCSSC,

997).Schneider (2001) concluded from simulation experiments that

uch broader ecosystem objectives cannot be realized without aeduction in AAC. However, to the extent that forest policies limitfficient joint production of material and ecological goods, ecosys-em objectives might be satisfied without reduction in economicctivity, provided the policies were changed. The determination ofhat combinations of goods can be produced under a given pol-

cy regime is precisely what the estimation and analysis of PPFs iseant to achieve.PPFs will be affected by forest polices that constrain the feasi-

le solutions. Notable characteristics of the forest policy regime inhe study region are: (1) divided landbases and overlapping tenures

Fig. 2. Detail of the study region showing mill locations (red squares) and forestmanagement unit boundaries (blue lines). (For interpretation of the references tocolor in this figure legend, the reader is referred to the web version of this article.)

and (2) appurtenancy. Forest lands are allocated to either deciduousor coniferous production, and rights to these two resources are allo-cated to different firms under different forms of tenure. This leadsto multiple firms operating independently within the same tenurearea or landbase; see Cumming and Armstrong (2001) for details.Appurtenancy is the restriction on wood flows across administra-tive boundaries and mills or between tenure areas (Fig. 2). Bothof these characteristics are traditional elements of Canadian for-est policy (see e.g., Pearse, 2001), even though they are knownto be economically inefficient (Cumming and Armstrong, 2004).Their purpose is to ensure security of supply for mills and supporteconomic development and employment in rural communities;however, this “security” reduces the economic value of wood prod-ucts to the extent that wood resources are not allocated to theirmost valued uses. We will show these constraints also affect theecological-economic tradeoffs between conservation and timberobjectives, which are revealed through the sensitivity of the shapeand location of the PPF to simulated changes in policy.

An additional feature of the existing policy environment is that,as in most of Canada’s forest, the AAC is set without explicit con-sideration of the risk of fire; this may impede the ability to satisfyfuture AAC and conservation objectives (Armstrong, 2004). By inte-grating the management model with a stochastic model of the fireregime, we are able to incorporate spatially variable mean effectsof fire into the optimization solution.

1.2. Overview of study design

The PPF is an efficiency frontier, illustrating the maximum fea-sible combination of outputs given only technological and inputconstraints. We consider two outputs: the area of old forest habitatand the net present value (NPV) of timber production. We estimatepiece-wise approximations of PPFs by solving a spatial optimizationmodel to maximize NPV under progressively increasing levels of theecological indicator target. This estimates the impact of alternativeconservation strategies/targets in terms of the foregone economicactivity. The constraints are interpreted by the model as a minimumlevel of the indicator that must be maintained over the final half ofa simulation run of 100 years. The PPF is a convex set under stan-dard assumptions of diminishing returns to fixed factors, such as
land and other natural capital. In the present context, this impliesthe opportunity cost of increasing biodiversity or other environ-mental services will be low at first, but will increase monotonicallyas more natural capital is allocated to conservation. This led us toquestion whether the modeled system is currently on the ‘flat’ ini-

odell

ta

potecaTs

ofPpg(a1fidpctfiwsmtwpfi

os2udswbti

u2bUarrel

2

aTmauc2s

G. Hauer et al. / Ecological M

ial part of the PPF, such that levels of biodiversity can be increasedt relatively low cost (discussed below in Section 4).

Policy constraints may lead to inefficient combinations of out-uts, represented by points inside the frontier. Thus, constructionf the PPF can identify the costs of alternate policy choices relativeo the optimum output combinations along the frontier. The mainxperimental component of this study is to evaluate the costs of twoentral components of forest policy in western Canada as describedbove: the spatial partition of landbases/tenures and appurtenancy.his is done simply by relaxing the relevant constraints, as repre-ented in the model formulation (see below).

While the choice of the economic objective as the maximizationf net present value from forest products is relatively straight-orward, the choice of indicators for biodiversity is more difficult.ast studies using integrated models have variously chosen wildlifeopulation sizes (Nalle et al., 2004), amount of mature or old-rowth forest (Toth et al., 2006), probability of species persistenceLichtenstein and Montgomery, 2003; Polasky et al., 2005, 2008),nd aggregate measures of habitat suitability (Arthaud and Rose,996). These indicators can be classified as “coarse-filter” or “fine-lter” (Noss and Cooperrider, 1994). Coarse-filter indicators areefined in terms of attributes, such as age class structures and pro-ortional abundances of land cover classes. The degree to whichhoice of filter affects conservation outcomes has been studied inhe context of reserve selection. Juutinen et al. (2004) found coarse-lter objectives under a least cost area selection procedure do notork as well as species-specific approaches for given area con-

traints. However, as Weber (2004) shows, a coarse-filter approachay itself be more cost effective than the direct approach, and

hus result in greater biodiversity for a given cost. In this study,e evaluate instances of both objectives with the secondary pur-ose of evaluating the efficacy of coarse-filter objectives in securingne-filter objectives.

Canadian forestry is tending towards the use of coarse-filterbjectives, at times in combination with fine-filter objectives forpecies of special management concern (Sougavinski and Doyon,005). Our PPFs in effect represent the NPV of timber managementnder varying levels of a coarse-filter objective, namely the abun-ance of old forest habitat. As fine-filter criteria, we embedded auite of forest songbird abundance models into the overall frame-ork and analyzed their predictions ex-post as a proxy for avian

iodiversity (e.g., Hannon and McCallum, 2003). In this way, weested the efficacy of indirect, coarse-filter management in achiev-ng fine-filter or species-level objectives.

In Canada, coarse-filter conservation objectives are set or eval-ated with reference to the RNV (AFCSSC, 1997; Harvey et al.,003; Hunter, 1993) or the distribution of values potentially createdy natural disturbances, such as wildfire (Armstrong et al., 2003).sing the integrated model, we estimated the RNV of the coarse-nd fine-filter indicators by Monte-Carlo simulation of the natu-al disturbance regime in the study region. By exploring the PPFselation to the RNV of both classes of indicators, we illustrate howcological criteria can provide guidance for the choice of preferredocation along the frontier.

. Study region and ecological data

Our study region covers approximately 560,000 km2 (Fig. 1),nd is dissected by both political and administrative boundaries.he policy-relevant boundaries include forest management agree-ent (FMA) areas (known as timber supply areas or TSAs in BC),

nd, in Alberta, smaller forest management units (FMUs) mostlysed as cut-control zones for coniferous species. We are mostlyoncerned with the forested portion of the area, comprising about33,000 km2. In the boreal plains ecozone, sites can be stratified byoil nutrient and moisture regimes. The most productive sites from

ing 221 (2010) 2590–2603 2593

a forest management perspective are called “mesic”, referring toa soil moisture regime intermediate between poorly drained wet-lands with organic soils and very well drained dry sites with coarsesandy soils (Kabzems et al., 1986). To a first approximation, mesicsites are dominated by some combination of trembling aspen, bal-sam poplar, white spruce, and balsam or subalpine fir, which arecommercially the most valuable and most harvested tree speciesin the study region. Non-forested areas are composed mostly ofopen or sparsely treed forested wetlands, and a large inclusion ofareas converted to agricultural land uses. For further details of theregional ecology, see Cumming et al. (1996, 2000) and Cumming(2001b); for a discussion of forest type and tree species in relationto management importance, see Cumming and Armstrong (2001).

The oil and gas sectors are also very active in the study region,and their exploration seismic lines and more permanent infrastruc-ture (e.g., wells and pipelines) remove significant areas of forest andpotentially disrupt larger areas by fragmentation. However, for thisstudy we hold the footprint of the energy sectors constant, so asto isolate the ecological impact of the forest sector. The effects ofthis additional process on the fine-filter indicators are representedby the existing density of drilled wells (Cumming and Cartledge,2004), which enters as a covariate in some of the fine-filter models.It presently has no effect on the simulated dynamics.

2.1. Forest cover data and coarse-filter indicators

The common ecological data used in the management and eco-logical models were forest composition and age structure data.These attributes were obtained from digital and tabular forestinventory data, as described in Cumming and Armstrong (2001,2004) and Cumming and Vernier (2002). The inventories are inter-preted from 1:20,000 aerial photography with a minimum mappingunit of 2 ha. The digital inventories are vectorial. The inventories arespatially organized according to land-survey units called “town-ships” of approximately 10,000 ha in area. We will refer to theseas “landscapes”. Landscapes are the fundamental spatial unit ofmany of the empirical studies described below. The spatial sim-ulation model uses a grid of landscapes to represent a large studyregion. In that context, landscapes are sometimes called “locations”to emphasise the importance of the geographic coordinates of alandscape unit. Mesic sites are not only the most economicallyimportant forests in the study region, as they are also believed tobe the most productive for forest songbirds in terms of both speciesrichness and total abundance. Accordingly, we chose as our coarse-filter indicator the total abundance of old mesic forest; this is our“focal” habitat type. Forest stands with an estimated or projectedcanopy age above 90 years are considered old. This is a compromisebetween the economic rotation ages of trembling aspen (approx-imately 80 years) and white spruce (approximately 110 years). Itis also the age by which structural features characteristic of theold-growth condition begin to appear in aspen stands (Cumminget al., 2000). Old mesic stands can be reliably identified from forestinventory data (Betts et al., 2006; Rettie et al., 1997).

2.2. Avian abundance models

For the ex-post fine-filter analysis, we used the total predictedabundances of a number of forest songbird species in mesic habi-tats. These were based on Poisson Generalized Linear Models ofempirical count data using a number of landscape and stand-levelcovariates. The observational data were collected in 2001 and 2002
by a standard survey protocol (dawn point counts of fixed durationand sampling radius) as part of a landscape-scale study designedto elucidate the effects of focal habitat abundance, configuration,and industrial development on forest songbirds (Schmiegelow andCumming, 2004). Data were collected within 102 landscapes, dis-

2 odell

ttpatihdtieaettcdgaosafftmV

scspCth((twdacwsimumm

2

aolTr(tefatTt


ributed over roughly 100,000 km2 in the southeastern portion ofhe present study region. The landscape design variables were theroportional abundance of focal habitat and a measure of its spatialrrangement within landscapes, derived from the patch size dis-ribution (Cumming and Vernier, 2002). The design variables forndustrial development were the cumulative proportional area ofarvest over the 30 years prior to survey and the total number ofrilled oil and gas wells measured as log well density standardizedo wells per 10,000 ha. This index is a surrogate for the cumulativempact of energy sector activities, especially the densities of lin-ar features such as roads, seismic lines, and pipelines (Cummingnd Cartledge, 2004). Post hoc landscape variables included a yearffect, sampling date within years, and factors for geographic loca-ion. The latter partition the survey region extent into roughly equalhirds from south to north and by halves longitudinally. This wasonsidered sufficient to capture the spatial gradients evident in theata whilst safeguarding out-of-sample prediction over the geo-raphic range of the current simulation study. We also includednumber of stand or patch-level covariates. The most important

f these were two indicators that stratify mesic habitat into fourubclasses by dominant species (deciduous or conifer dominance)nd canopy age (>90 or ≤90 years). See Vernier et al. (2008) forurther details of the underlying avian habitat classification fromorest resource inventory attributes. We also include the mean ofhe log-transformed patch sizes of focal habitat as an elementary

easure of habitat configuration within landscapes (Cumming andernier, 2002; Newbold and Eadie, 2004).

For the present analysis, we developed models of a number ofongbird species known to specialize on various ages and speciesompositions of mesic forest. Canada Warbler (Wilsonia Canaden-is; CAWA) is associated in western Canada with old, relativelyure deciduous stands. This species is listed as threatened inanada (COSEWIC, 2008). Bay-breasted warbler (Dendroica cas-anea; BBWA) is associated with mature mesic forests and attainsighest abundances when the canopy contains roughly 60% coniferVernier et al., 2006). This species is listed as “sensitive” in AlbertaNorton, 2001), which entails that “forest management plans needo ensure retention of breeding habitat”. These are two species forhich fine-filter management plans might be necessary. We alsoeveloped models of American Redstart (Setophaga ruticilla; AMRE)nd Ovenbird (Seiurus aurocapillus; OVEN); these two species areonsidered secure in Alberta, but both are positively associatedith deciduous dominated mesic forest at the patch and landscape

cale, and are affected, positively or negatively, by forest harvest-ng (Vernier et al., 2008). The former species is associated with old

esic forest, while the latter is associated with younger pure decid-ous forest. Thus, these latter two species might be more easilyanageable by coarse-filter techniques. The details of the fittedodels are available from the authors upon request.

.3. Natural disturbance model

A multi-stage stochastic fire model was adapted from Cummingnd Armstrong (2004) to capture spatial variation in the fire regimever the study region. The framework for the fire submodels fol-ows Cumming and Armstrong (2004). The empirical data werehe Canadian Large Fire Database (Stocks et al., 2002), whichecords the location, size, and attributed cause of all large fires>200 ha) from 1959 through 2001. Lightning fires were referencedo a 10,000 km2 hexagonal grid, and the observed fire regime forach location characterized by three estimated parameters: the
requency of large fires (expected counts per unit area and timessuming Poisson errors) and the shape and truncation parame-ers of a truncated Pareto fire size distribution (Cumming, 2001a).he resultant parameter maps were intersected with the simula-ion grid to assign parameters to each landscape. The composition
ing 221 (2010) 2590–2603

of areas burned by simulated fires is calculated from multivari-ate regression models (Cumming, 2001b) that we assume to beapproximately correct over the study region. To represent histori-cal and current fire regimes, we corrected the observed frequenciesto account for changes over the period of record in the probabilitythat a detected fire will become large (see Table B2 in Cumming(2005)); these changes measure increasing effectiveness of wildfiresuppression by initial attack.

2.4. Estimating the range of natural variation of ecologicalindicators

The range of natural variation of ecological indicators over thestudy area was estimated by Monte-Carlo simulation of forestdynamics in the absence of forest management. One thousandreplicates of 200-year simulations of forest development weremade starting with the initial forest conditions from the same forestinventory used with the optimization model. The harvesting sub-model was turned off and the energy sector wells were removedfrom the landscape descriptions so that the only disturbance pro-cess was the simulated pre-suppression fire regime. The state of thesimulated forest at the end of a 200-year simulation was taken tobe a realization of the natural disturbance regime, independent ofinitial conditions. The proportional abundance of old mesic forestand the levels of other indicators were calculated at the end of eachsimulation. The percentiles of the distributions of these indicatorswere used to characterize the natural ranges of variation. A sim-ilar sampling procedure was used to estimate the landscape- andstand-type specific fire risk parameters used in the optimizationmodel, except the model was run for only a single time step undercurrent conditions.

3. Model formulation

The integrated model is a large nonlinear programming modelwith binary integer variables representing forest access decisions.The individual components of the model are described in turnbelow; the solution method is outlined in Section 3.6. The modelitself and all supporting libraries (e.g., graphical display, randomnumber generation) are implemented in C. The code can be com-piled with trivial modifications on any modern version of theUnixTM operating system, including common distributions of Linux.

3.1. Forest dynamics

Harvesting activities are represented spatially at landscape res-olution. As explained in Section 2.1 each landscape has an area ofroughly 10,000 ha. Our study region includes 5600 landscapes (orlocations), indexed by h. The model runs on 10 year time steps,indexed by t. We denote the maximum or end time as period T = 10.

Within each location, forest patches are represented non-spatially. That is, the model tracks the size and attributes of eachforest patch, but not their spatial locations or adjacency relationswithin landscapes (Cumming and Vernier, 2002). Forest patchesare aggregated into classes defined by 44 5-year age classes and 8forest types, which we index by s and i, respectively. Each foresttype has a characteristic tree-species composition. Each forest typeis characterized by two timber types j, hardwood and softwood. Attime t, the land area distribution of forest classes over locations,age classes, and species types is represented by Xhist over all h, i,and s with initial conditions at t = 0. The initial forest land area fora particular combination is then denoted by Xhis0.

Fire is represented deterministically by the constants �hi, theexpected portion of forest land of type i in grid location h that isunburned after a model time-step. These constants were estimatedas part of the Monte-Carlo simulations, discussed above. This deter-ministic approximation of a stochastic fire regime was necessary to

odell

ltfhd

X

waf

X

a

X

wttTE

3

p

w9fptp

a

b

wiioe5waot(po

bTtifPl


imit the size of the dynamic programming problem. At each modelime-step, the area of each forest class Xhist is transferred to a neworest class in up to three ways: growth into the next older age class,arvest, and fire. All three of these mechanisms are shown in theynamic equation:

hi,s+1,t+1 = (Xhist − xhist)(1 − �hi) (1)

here xhist is the area of forest harvested in ha of type i, age class s,nd location h in period t. Areas disturbed by harvesting are trans-erred to younger age classes according to:

hi1t+1 =S∑

s>=mi

xhist + Xhi0t (2)

nd areas disturbed by fire are transferred using:

hi,0,t+1 =mi∑

s=1

Xhist�hi +S∑

s≥mi

(Xhist − xhist)�hi (3)

here mi is the minimum harvest age class and S = 200 years ishe maximum age. Forest area remaining after harvest and fire isransferred to the next oldest age class Xhi,s+1,t+1 in the next period.his construction of forest dynamics is similar to that of Reed andrrico (1986).

.2. Coarse-filter constraints and fine-filter indicators

The coarse-filter habitat objectives are expressed in terms of theroportional area of focal habitat at some chosen period:

1A

∑h

∑s≥so

∑i ∈ Fm

Xhist ≥ Ot (4)

here Ot is the old mesic forest target in period t, so is the age (here,0 years) at which mesic forest is defined as old, and Fm is the set oforest types defined as mesic. This target is varied to estimate theroduction possibility frontiers. The objectives are enforced overhe last 5 planning periods; the constraint is never binding in earliereriods.

Our modeling framework calculates summed expected birdbundances bhist over each aggregation of patches Xhist as:

hist = Xhist

�f (Om

is , Dmis , N1

h , N2h , Wh, OT

ht, P̄, Rht, Eht, ˇ) (5)

here Omis

is an old mesic indicator that equals 1 if stand type is old mesic at age s and 0 otherwise; Dm

isis a deciduous mesic

ndicator that equals 1 if stand type i is deciduous at age s and 0therwise; N1

hand N2

hare indicators that capture geographic gradi-

nts in avian distributions if location h is north of latitude 55.6◦N or6.6◦N, respectively; Wh is an indicator similar to N1

hthat indicates

hether location h is west of longitude 113◦W; OTht

is the percent-ge of old mesic forest in location h; P̄ht is the mean log patch sizef old mesic patches in location h; Rht is the percentage of loca-ion h cut in the last 30 years; Eht is the log of the number of wellsan indicator of energy sector activity); ˇ is a vector of estimatedarameters; f is the model form, in this case exp(z′ˇ); z is the vectorf covariates.

The function f simply generates predicted species abundancesased on the Poisson regression models described in Section 2.2.he constant 1/� scales predictions from 100 m radius circular plots
o expected counts per ha. The predicted values were standard-zed with species-specific constants to account for year effects andor variation in detection probability within the breeding season.redicted patch-level counts can be aggregated up to predictedandscape totals, bht, by summing over i and s, and hence to the
ing 221 (2010) 2590–2603 2595

study region by summing over landscapes h, bt =∑

hisbhist. This fol-lows from the additive property of Poisson random variables. Somecovariates, such as OT

ht, P̄ht and Rht must be computed for each loca-

tion in each time period t, but these are simple summations of areasover each aggregated patch type, Xhist.

3.3. Economic equations, constraints, and objectives

Harvest volumes are computed using mixed-species yieldcurves obtained from forest products companies with licenses inthe study region. Yield curve entry vijs represents the merchantablevolume in m3/ha of timber type j obtainable from forest type i at ageclass s. The total volume of timber type j harvested from location his:

Vjht =∑

is

vijsxhist h ∈ At (6)

where At is a set of all accessible locations at time t (see below).Another important aspect of our model is that demand for timberis spatially represented by multiple mill locations. This allows usto more effectively determine spatial and temporal harvesting pat-terns and the costs of conservation objectives, such as constraintson old forest. Harvested timber is distributed amongst 50 millsdenoted by Mm, each associated with one of the 26 mill locations(Fig. 2). The choice of mill is governed by the objective functiondescribed below. zjmht is then the amount of timber type j shippedto mill m from location h in period t. The amount of wood shippedto each mill m from location h cannot exceed the amount of woodharvested from location h:∑m ∈ Mj

zjmht ≤ Vjht h ∈ At (7)

where Mj is a subset of the M mills that demands wood of type j. Theamount of wood delivered to a particular mill m, or yjmt, is capturedby:

yjmt =∑

h

zjmht h ∈ At (8)

Harvesting activities are driven by demand for wood at each ofthe M mills. Each mill generates benefits from the wood it receivesaccording to a revenue function Rmjt(yjmt). The revenue functionswere estimated as the area under inverse demand curves for woodat each mill location. The demand curves were estimated using thesame methods and data as in Williamson et al. (2004). Demandcurves were estimated for representative mills in the pulp andpaper, waferboard, and sawmill industries. The demand curveswere calibrated to mill capacities and wood-flows using data pro-vided by the Government of Alberta’s Department of SustainableResources Development.

The costs of delivering wood to the mills result from transport,harvesting, regeneration activities, and road building. In this typeof model, if ending inventory has no value then the model will tendto exhaust all resources in the final periods. We solve this problemby introducing ending inventory values, which per ha of forest areaXhisT+1 are denoted Ehis. When added to the revenues generated atall mills, we have the following net benefit equation:

B =∑

jt

∑,m ∈ Mj

Rmjt(yjmt) −∑

jt

∑h ∈ A

∑m ∈ Mj

Cmhtzjmht

−∑∑

H x +∑

X E (9)

h ∈ A ist

ist hist

his

hisT+1 his

where Cmht is the per unit volume transport cost of wood deliv-ered to mill m from location h and Hist is the per ha harvestingcost plus regeneration cost for stand type i, age class s, in period t.

2 odell

Ctct(tictCtl

afsilr

3

tftedlWbllb

lalmIlp∑Fp

V

Acptmdt

A

wbf


ost and revenue components are subscripted with a t to denotehat costs and revenues are different in each period due to dis-ounting of future values, changes in costs related to changes inree size and age distributions, and possible changes in technologyalthough technology is held constant in our model). Logging andransport costs data (Cmht, Hist) used in this study are as describedn Cumming and Armstrong (2001). Net Present Values were cal-ulated at a discount rate of 0.04 and, as noted above, we subsumehe discounting process into the coefficients. For example, Cmht =¯

mh(1/(1 + r)l(t−1)), where C̄mh is the undiscounted per unit volumeransport cost, r is the annual discount rate (0.04 in this case), andis the length of the planning period (l = 10).

One may interpret the objective function specified in Eq. (9) ascompetitive model in which firms compete for timber, thereby

acilitating efficient land use from an economic perspective. Thepatial nature of the model provides information on the impact ofncreasing conservation requirements and indicates forestry firmocations that are most likely to cease activities as conservationequirements are increased.

.4. Forest access

One important aspect of our study area is that substantial por-ions are currently inaccessible by roads. In our model, we accountor the timing and cost of forest access decisions but more simplyhan in studies focused on these specific decisions (e.g., Andalaftt al., 2003; Richards and Gunn, 2000). Integration of forest accessecisions into timber supply models has been done for small prob-

em instances but on a much smaller scale than is considered here.e address the access problem by assuming enough roads are

uilt within an accessed location to provide access to all adjacentocations if necessary. In other words, we do not model roads asinks that connect a set of nodes. This means that wood may flowetween adjacent locations as long as both are accessed.

We define a set A that contains all locations accessed, a set U ofocations that are not accessed previous to our model time horizon,nd AU as a set of locations in A that are adjacent to at least oneocation in U. Major access decisions are captured (i.e., presence of

ajor logging haul roads or highways) through the integer variableht, which indicates the time period t that location h is accessed. Aocation can be accessed only once and we assume the access isermanent. Hence, we require:

t

Iht = 1 (10)

or locations h ∈ U, harvesting cannot take place until access isresent. Hence, we require:

jht ≤t∑

s=0

IhsM h ∈ U (11)

where M may represent the maximum road capacity (seendalaft et al., 2003) or simply a number larger than the largest con-eivable Vjht. For unaccessed locations, the eventual path to accessoints or locations that are permanently accessed cannot be prede-ermined. Hence, the transport costs Chmt, which are based on the

inimum cost route from supply locations h to mills m, will varyepending on the overall access plan. To account for this, we definehe set of valid access plans:

P

{n(U)×T

∣∣∣ T∑ }
= (Iht )∀ht ∈ {0, 1} ∣∣
s

Ihs = 1, if Iht = 1 then h ∈ Ph (12)

here n(U) is the total number of locations not accessed at theeginning of the time horizon and Ph is the set of all feasible pathsrom h to a location in AU, the set of all locations accessed at the

ing 221 (2010) 2590–2603

beginning of the planning horizon. A feasible path is a sequenceof locations h0, h1, . . ., hN such that hi+1 is adjacent to hi, i = 0, . . .,N − 1; hN ∈ A and adjacent to at least one element of U. Let a be onepossible access plan (i.e., a ∈ AP), then define Ca

hmtas the minimum

transport cost from location h to m in period t for access plan a. Theobjective function given a and including access costs is then:

Ba =∑

jt

∑,m ∈ Mj

Rmjt(yjmt) −∑jht

∑m ∈ Mj

Camhtzjmht −

∑hist

Histxhist

−∑

t

∑h/∈A

DtIaht +

∑his

XhisT+1Ehis (13)

Given a, Eqs. (1)–(4), (6)–(8) and (11) form a non-linear pro-gramming problem. An optimal solution, (ya

jmt)∀jmt

, (zajmht

)∀jmht,

(xahist

)∀hist, and (Xa

hist)∀hist

, given a, is obtained by maximizing (13)subject to (1)–(4) and (6)–(8). To obtain an overall optimal solutionwe would need to solve:

maxa ∈ AP

Ba (14)

A solution to this mixed integer-nonlinear problem would be diffi-cult to obtain because of its size and complexity. Hence, we solveEq. (13) for different sets of access plans a, generated through aheuristic algorithm described in Section 3.6.

3.5. Regulations and wood flow constraints

Fig. 2 is a map of administrative boundaries or forest man-agement units and mill locations. Wood quotas arising fromappurtenancy, wood processing constraints, and forest tenurerequirements are defined and enforced according to these bound-aries and mill locations. Let k index forest management units andHk be the set of locations in k. Then quota constraints qjmkt of timbertype j from unit k to mill m ∈ Mj at time t are defined as:

qjmkt ≤∑h ∈ Hk

zjmht (15)

3.6. Solving the model

Maximizing Eq. (13) subject to Eqs. (1)–(4), (6)–(8) and (15) is alarge linear or non-linear programming problem, depending on theform of the revenue functions for each mill location. This problem islarge and time consuming to solve using conventional methods. Weemploy a Lagrangian Relaxation method (see Hauer and Hoganson,1996; Hoganson and Rose, 1984; Nanang and Hauer, 2006) to signif-icantly reduce computation times. The method works by relaxingstrict primal feasibility requirements for selected constraints. Inthis case, the relaxed constraints are the old mesic forest con-straint (Eq. (4)), the market clearing constraint (Eq. (8)), and thequota constraints (Eq. (15)). The method yields optimal solutionsalthough Eqs. (4) and (8) will not be satisfied exactly. We alsoutilize the method described by Larsson et al. (1999) in conjunc-tion with the Lagrangian relaxation method to generate solutionscloser to primal feasibility and optimality and to improve conver-gence. The optimal access plan (Eq. (14)) is a mixed integer problem,which is difficult to solve. Andalaft et al. (2003) formulate a similarbut smaller scale problem and describe and demonstrate a seriesof complex solution procedures. Richards and Gunn (2000) use adifferent method to solve a forest access problem with volume con-
straints, again on a smaller size of problem than we address here.Because the access problems are essentially combinatorial, solu-tion methods quickly become intractable as the number of spatialunits becomes large. We use a simple heuristic for generating alter-native access plans, based on solving a simplified access problem


F aints.f 0th p

fpepia

mt

Wtg

LiTuNdaowowaadmclltwTaaaharo

ig. 3. Production possibilities frontiers with and without forest regulatory constrorest over the entire study region. Also shown are the 0th, 10th, 50th, 90th, and 10

or each un-accessed location. Each un-accessed location may beermanently accessed in one of the 10 periods, and computing anstimate of the maximum net present value of harvest for eachotential access period is simple. Let this maximum be NPVit. If t̄i

s the earliest access time of neighboring locations to i, then theccess time for i can be determined by choosing t to solve:

ax>=t̄i

NPVit (16)

e adopt this simple heuristic to efficiently generate feasible solu-ions to the full access problem. We note the method cannot beuaranteed to produce globally optimum results.

Our model is comparable to those of Nalle et al. (2004),ichtenstein and Montgomery (2003) and Arthaud and Rose (1996)n that it is both spatial and dynamic with multiple time periods.his is contrast to the study of Polasky et al. (2008) which also eval-ated a PPF, but in the context of a static one period model. As withalle et al. (2004) and Lichtenstein and Montgomery (2003), timberemands at mills are downward sloping and delivered wood pricesre endogenous in the model. Most past studies, whether staticr dynamic are formulated as use integer programming problemshich are solved using heuristic methods like Simulated Annealing

r Tabu Search. Where our model differs is that forest harvestingithin spatial units are formulated as continuous decision vari-

bles (e.g. Eq. (1)) while strategic forest access decisions are codeds integer variables (Eqs. (12) and (13)). Therefore our model is bestescribed as a mixed nonlinear-integer programming problem. Ourodel also emphasizes the spatial wood allocation problem and a

ourse filter ecological objective. The spatial wood allocation prob-em is important because of the locations of the mill sites over theandscapes and because delivered wood costs increase with dis-ance from the mills. These factors affect both optimal allocation ofood over space and the location of low opportunity cost habitat.

he modeling of the spatial wood allocation among demand pointslso allows us to explicitly consider regulations (e.g. appurtenancynd other wood flow constraints) that restrict wood allocation
mong demands. The continuous formulation allows us to use aeuristic optimization algorithm which guarantees that solutionsre optimal given that demand and course filter constraints may beelaxed so that they need not be met exactly. This allows us to findptimal near-feasible solutions quickly.
Objective function in dollars plotted against proportional abundance of old mesicercentiles for old mesic forest from the natural fire disturbance regime.

4. Results

Recall that our coarse-filter ecological indicator (or “indicator”,henceforth) was the total abundance of old mesic forest. PPFs werederived by progressively increasing the indicator target levels inEq. (4) while maximizing the objective function (Eq. (13)), subjectto this and other structural and policy constraints. Because highertargets result in lower objective function values, the model esti-mates the impact of alternative conservation strategies in termsof foregone economic activity. Impacts are measured as opportu-nity costs, calculated from the estimated net present values of allmodeled activity over the 200 year time horizon; these values areplotted on the y-axes of the PPFs.

Fig. 3 shows the PPF of the net present value of forest harvest-ing activity and our indicator over the entire study region underthe two management scenarios. The lower curve (“With Quotas”)plots the PPF under the current policy regime, which restricts woodflows across administrative boundaries. The left-most point on thecurve (“Current Policy”) shows the objective function value as plot-ted against the lowest target level of 24% old mesic forest. Thislevel does not in fact constrain the solution; harvest policy withoutany old mesic constraints maintains at least 24% area in old mesicforest over the last planning 5 periods. Thus, the labeled point cor-responds most closely to the current policy regime of overlappingtenures, appurtenancy, and little or no explicit provision for theindicator. The upper curve (labeled “Without Quotas”) is the PPFin the absence of any of the spatial constraints that are featuresof the current policy regime. As such, it represents the solutionsto an unrestricted maximization subject only to the varying con-straints on the levels of old mesic forest. Again, the lowest targetlevel does not constrain the solution. Thus, we conclude the mod-eled system is indeed currently on the ‘flat’ initial part of the PPFand that environmental goods associated with the indicator can beincreased at relatively low cost, with or without spatial constraintson timber harvesting and fiber allocation to mills. The main effectof the spatial constraints is to reduce the NPV of the forest resource.
The absolute magnitude of the reduction is almost independent ofthe level of the indicator (Fig. 3). The relative reduction in NPV isapproximately 8% where mesic forest abundance is unconstrained,and increases to about 27% when the old growth constraint is setto its maximum. This upward shift of the PPF between the two sce-

2598 G. Hauer et al. / Ecological Modelling 221 (2010) 2590–2603

F int ont he tot

ncoaaric

etwattsticerzo

ig. 4. Distribution of old mesic forest at 100 years in simulations with no constraotal mesic forest area (bottom panel). Abundances are mapped as percentages of t

arios is characteristic of technological change or, as in this case,hanges in policy constraints. The existing policy is clearly sub-ptimal in both economic and ecological respects. We note thatny point between the two curves of Fig. 3 is a feasible solution,nd could presumably be achieved under some intermediate policyegime. However, we have not attempted to quantify the relativempact of the two sorts of spatial constraints embodied in the lowerurve.

The flatness of the production possibility curves is due to sev-ral factors. Since the model maximizes net present value it willend to find the ways that satisfy the old mesic forest constraintith least reduction in the objective function value first. This is

ccomplished by choosing areas that are least accessible are far-hest away from mill locations and thus have lower contributionso NPV. There is also an intertemporal dimension. Since the con-traint applies to the last 5 of 10 periods in the model, it is possibleo satisfy the constraint when set at less demanding levels by delay-ng harvest reductions, which is less costly, required to meet the
onstraint to the latter periods. However, when the old mesic for-st constraint is increased, immediate reductions in harvest areequired in order to meet the constraint later in the planning hori-on. Immediate actions have higher opportunity cost terms becausef discounting. The production possibilities curves would begin to
the indicator (top panel) and with the final abundances constrained to be 42% ofal amount of mesic forest in each landscape.

decrease more quickly with respect to increases in old mesic for-est requirements if the constraint was applied earlier, for exampleat 30–100 years instead of 50–100 years. The lower curve (“WithQuotas”) lies inside the upper curve (“Without Quotas”) becausethe boundaries of the forest management units which define thelocations from which wood may flow under the quotas were notcreated with regard to minimizing transportation costs. Thus thequota constraints prevent wood from flowing to its most valuablelocations.

In Fig. 3, we also show the RNV of the indicator as estimated fromthe simulations of unmanaged forests under a natural fire regime.Current policy results in indicator levels below the median and out-side the RNV. However, as the RNV encompasses the flattest part ofthe PPF, attaining the median or even the upper 90th percentile ofthe indicator would be possible at relatively little expense in termsof the net present value of forest harvesting. Indicator levels equiva-lent to the upper RNV can be achieved at minimal cost even if spatialtimber processing restrictions are maintained. Again, the level of
old mesic forest when no coarse-filter constraint is applied is notwithin the RNV, but a position within the RNV may be obtainedwithout substantial cost. Increasing economic benefits and target-ing the upper bound of the RNV of old mesic forest are even possibleby relaxing the spatial timber processing restrictions.

G. Hauer et al. / Ecological Modell

Fig. 5. Relative bird abundance indices for four species exhibiting positive relation-ships between the indicator and predicted landscape counts. The indices are relativetAW

tyftsirpas

fifoictttpoeami

aasnrsWtFiblTia

model are being used to evaluate alternate strategies for wood-

o the species-specific area totals reached in the unconstrained simulation. AMRE:merican Redstart, BBWA: Bay Breasted Warbler, OVEN: Oven Bird, CAWA: Canadaarbler.

In Fig. 4, we illustrate the spatial implications of the coarse fil-er constraint at an intermediate time during the simulation (100ears). We contrast the distribution and abundance of old mesicorest in the unconstrained and maximally constrained case wherehe indicator target is 42%. The location of old forests in the con-trained case is the spatial result of the least cost solution as realizedn the last five periods. The result could be interpreted as a tempo-ary habitat reserve or protected area designed under a least costrotection strategy. We note the contrast between the constrainednd unconstrained cases in the vicinity of mills, especially in theouthern portion of the study region in Alberta.

The main results of the fine-filter analysis are shown in Fig. 5 forour selected species of forest songbirds. The graph plots relativendices of predicted total regional abundances against the objectiveunction values from the spatially unconstrained PPF (upper curvef Fig. 3). For these species, the coarse-filter approach works toncrease the bird counts in the intended direction. That is, use of theoarse-filter requirement to increase old mesic forest also increaseshe abundance of habitat specialists that might be considered forargeting by fine-filter management strategies. These results wereypical of those modeled species where observed abundances wereositively correlated with old mesic forest habitat at the patchr landscape scale. We note the bird abundances are not nec-ssarily optimized for any particular species. Optimizing speciesbundances would require that they be directly constrained in theodel formulation rather than (or in addition to) the coarse-filter

ndicator.In Figs. 6 and 7, we illustrate the effect of the coarse-filter

nd spatial timber harvesting constraints on the total predictedbundances of Canada Warbler and Ovenbird, respectively; theimulated RNVs are also shown. Spatial harvesting constraintsegatively affect total abundances of both species due to the rear-angement of the harvest area under the constraints, which canhift harvesting into or out of the preferred habitats. For the Canada

arbler, the current policy is outside the RNV, but points withinhe range up to the 90th percentile can be achieved at minimal cost.or example, a target level of the coarse-filter indicator of approx-mately 30% would induce end-state Canada Warbler abundancesetween the 10th and 50th percentile of the RNV, which would

ikely be deemed acceptable under the natural disturbance model.hus, by exploring the PPFs relation to the RNV of both classes ofndicators, we illustrate how ecological criteria can provide guid-nce for the choice of preferred location along the frontier.

ing 221 (2010) 2590–2603 2599

For Ovenbird, the RNV does not overlap the tradeoff curves(Fig. 7). Based on the fitted regression models (not reported here),we attribute this result to two main factors. First, the species isassociated with younger, pure deciduous forest, which would havebeen relatively common under the natural fire regime, whereas theareas of older mesic forest maintained under the coarse-filter con-straints includes a conifer component that does not contribute tothe abundance of this species. Second, indirect effects of harvestingat the landscape scale reduce abundance of this species in patchesof its preferred habitat. These appear to be the main reasons whythe RNV of total Ovenbird abundance is shifted to the right of thetradeoff curves; harvesting would have to be severely restrictedor completely excluded from certain areas to ensure this species’abundance is within its natural range. A tentative conclusion to thusbe drawn is that the combination of harvesting and fire manage-ment can shift the forest outside the RNV for some species, and thatthe two processes are therefore not fully compensatory, as mighthave been anticipated.

The spatial distribution of one of the fine-filter indicators(Canada Warbler) is shown in Fig. 8, under the same conditionsas for the coarse-filter indicator (Fig. 4). The predicted abundanceslargely track the distribution of focal habitat, but with some large-scale geographic trends superimposed. This result again shows thecoarse-filter constraint is able to maintain high levels and good geo-graphic distribution of at least some fine-filter indicators under thesimulated management regime.

5. Discussion

The construction of a PPF for ecological and economic outputsrequires an integrated modeling approach. Integrated modelingof forest landscapes is a complex undertaking given the spatialand temporal dynamics affecting both economic and ecologicalsystems. Process based simulation models that identify ecologicaloutcomes, such as species presence and absence, based on land-scape characteristics must be integrated with optimization modelsused to identify efficient solutions (Watzold et al., 2006). Compu-tational demands require modelers themselves to make tradeoffsabout the degree of spatial and temporal resolution in the model,as well as about the representation of the spatial and dynamic rela-tionships within the model. Our understanding of the effects ofthese modeling decisions on tradeoff analysis is increased with eachexploration in this area. This study illustrates the potential benefitsof improving on the spatial and temporal dimensions of integratedforest sector modeling.

The main contributions we feel this study makes are the (1)development of a tradeoff frontier for a large spatial extent anda multi-firm economic environment, thus making it relevant forprovincial-scale policy analysis, (2) incorporation of insights fromthe RNV of our coarse-filter target and species-level indicators intothe tradeoff analysis, and (3) empirical assessment of an existingand controversial policy environment and the impact of appurte-nancy/spatial timber processing constraints in a spatially realisticsimulation of this policy environment.

We applied spatial simulation and economic optimization toa real world management problem over roughly 560,000 km2 ofmanaged boreal forests. We have shown how empirical ecologi-cal data, in this case predictive models of bird abundance and fireregime parameters, can be effectively incorporated into a spatiallyexplicit modeling framework at regional extents. Versions of this

land caribou conservation in Alberta and Québec, and to explore thepotential for forest management to mitigate the predicted effects ofclimate warming on boreal fire regimes (Krawchuck and Cumming,in press). This approach can certainly be applied at much larger

2600 G. Hauer et al. / Ecological Modelling 221 (2010) 2590–2603

F ulatiop

epfpfa

btwooa

Fa

ig. 6. Total predicted regional abundances of Canada Warbler at the end of the simlotted against the NPV obtained by the PPFs of Fig. 3.

xtents, while incorporating many other ecological and economicrocesses. For example, applications to conservation planning andorest management under climate change are now in contem-lation over the entire portion of the Canadian boreal for whichorest resource inventory data exist. Further planned extensionsnd applications are discussed below.

We found that current forest policy will result in indicator levelselow the estimated lower bound of the RNV. However, increasing
he percent of old mesic forest to at least the median of the RNVould be quite inexpensive in terms of opportunity cost. Although
ur results quantify a range of ecologically sound choices basedn comparison of the PPF and the estimated RNV in indicators,dditional normative information is required to determine the loca-

ig. 7. Total predicted regional abundances of Ovenbird at the end of the simulations, withgainst the NPV obtained by the PPFs of Fig. 3.

ns with and without spatial constraints and under the natural disturbance regime,

tion of superior options on the frontier (Toth et al., 2006). Ourpresent analysis only considers ranges of variation evaluated overa large spatial extent. We found that maintaining many indicatorswithin the RNV over very large spatial extents carries little cost.If ecological objectives are to be examined and implemented atsmaller spatial scales (e.g., where each FMA is required to main-tain a proportion of old mesic forest), the tradeoffs are expectedto be quite different. At present, many Detailed Forest Manage-
ment Plans (a regulatory requirement for FMA holders) describeefforts to incorporate RNV in ecological objectives. Future uses ofthis modeling framework will assess the economic and ecologicalimpacts of these smaller spatial scale objectives within a regionalcontext. One question of interest is the how the costs of maintain-
and without spatial constraints and under the natural disturbance regime, plotted


F y regib

iWissta

pnvdrMcoisfOctb

ig. 8. Expected density of Canada Warbler at t = 100 years under the current police at least 42% in the final five periods of the simulation (bottom panel).

ng biodiversity objectives vary with the size of management unit.e conjecture that the aggregate costs of meeting such objectives

ndependently within individual forest management units will beignificantly higher than what would be necessary under regionalolutions, such as presented here. This raises the possibility of spa-ial and temporal tradeoffs between different biodiversity goodsmong regions, a subject we plan to explore in future analyses.

The empirical assessment of the impact of appurtenancy, timberrocessing, and administrative boundary constraints revealed a sig-ificant cost associated with this constraint. Potential net presentalues decreased by nearly 10% with the constraint; achieving bio-iversity targets is much less expensive when the constraint iselaxed. This result is consistent with findings of Lichtenstein andontgomery (2003) and Nalle et al. (2004) where relaxed spatial

onstraints that allowed for more use of public land increased thebjective function. Our analysis builds on their approach, by relax-ng constraints both on how much harvesting may take place onpecific management units and on where wood harvested from dif-
erent forest management units may be delivered (see Eq. (15)).ur exercise provides not only the aggregate costs of the policyonstraint, but can also be used to examine the spatial location ofhe areas and which demand locations are expected to be affectedy a change in the policy. Both pieces of information provide
me (top panel) and when the total abundance of old mesic forest is constrained to

important input into policy analysis of the benefits and costs ofspatial constraints. Historically, spatial constraints were viewedas critical to community stability and resource utilization. Con-sideration of whether these constraints still confer such benefitsand, if so, whether their values outstrip the substantial costs framethe policy debate to which we hope our results may contribute.Our results confirm and substantially expand upon the previousfindings of Cumming and Armstrong (2001, 2004), and we reit-erate their call for a major review of forest policy in the region.Currently, our modeling framework maximizes NPV from forestrysubject to the coarse-filter constraint. The fine-filter objectives(birds) are tracking the landscape outcomes from this optimizationproblem. In future work, we plan to optimize over both NPV andthe fine-filter indicators to assess how optimization over fine-filterindicators differs from the coarse-filter approach. An assessment ofwhether coarse-filter objectives address fine-filter targets, whetherfine-filter objectives may address coarse-filter targets, or if a com-bination of targets is required will inform the on-going debate
on coarse-filter or natural disturbance based forest management.In addition, a likely extension is to include fine-filter indicatorsfor species other than birds, particularly threatened species suchas woodland caribou. Examining the opportunity cost of variousobjectives and management strategies for threatened species will

2 odell

pOeactfr

fiaostceownatwictcs2

eatmtlsgtt([lccattrmwhnrpi

ogbsloiio


rovide valuable information for policy makers and land managers.ur analysis assumed that external demands for and prices of for-st products would remain constant over the time horizon. If thisnalysis were ever to inform actual policy, as we hope, it would ofourse be possible to make the economic component more realis-ic by incorporating information or assumptions about projecteduture market conditions. One could also use whatever discountate policy makers preferred.

The incorporation of a spatially explicit cost structure in theorest management model tends to concentrate the harvest, result-ng in de-facto zoning into areas of greater and lesser harvestingctivity. We observed this in both the simulated distributions ofld mesic forest (Fig. 4), and in the predicted abundances of somepecies (e.g., Canada Warbler; Fig. 8). The spatial structures ofhese distributions suggest that effective reserve design proceduresould be developed within this modeling framework. Further wexpected and observed the locations of the de-facto zones to changever time as areas of younger forest mature. Figs. 4 and 8, revealedhat were, in effect, “floating reserves” or dynamic conservationetworks. This is a strategy (Cumming et al., 1996) whereby highbundances of targeted species (or at least, of their preferred habi-ats) are concentrated temporarily in limited geographic areashose locations change programmatically over time. These float-

ng reserves emerged naturally from the model solution as the mostost effective means of sustaining the coarse-filter conservationarget over time. Such dynamic reserves are also arguably moreonsistent with the natural dynamics of this ecosystem than staticystems of protected areas (Cumming et al., 1996; Bengtsson et al.,003).

The effectiveness of static and dynamic reserves in boreal for-st was examined by Rayfield et al. (2008) on a 430,000 ha studyrea of managed boreal forest in Québec, Canada. Using a protec-ion target of about 10% by area, they found their dynamic system

aintained their fine-filter indicator within reserve networks bet-er than a static system, but that the overall quality of the managedandscape did not differ between the two treatments. There wasome indication that maintaining the dynamic system became pro-ressively more difficult as cumulative harvesting activities alteredhe overall structure of the landscape. Their findings drew atten-ion to the importance of developments in the unprotected matrixLindenmayer et al., 2006), and they concluded that “small, dynamicprotected areas networks] will not necessarily solve the prob-em of our unwillingness to set aside large [protected areas] thatan accommodate natural dynamics within their boundaries”. Weoncur with the importance of the matrix within which protectedreas are embedded. However, we note our study region was 100imes larger than that of Rayfield et al. (2008), and the size ofhe structures we interpreted as possible elements of a dynamiceserve system was also very large compared to the reserve ele-ents of their study. Relatively large reserve elements combinedith higher conservation targets and appropriate constraints onarvest sequencing within the matrix might prove an effective, eco-omically efficient conservation strategy for intensively managedegions, such as the western Boreal Plains, where establishing large,ermanent, and pristine protected areas is no longer possible. We

ntend to explore these issues more fully in a subsequent paper.A significant forest disturbance agent not fully incorporated into

ur model is the energy sector. Energy sector activity (oil, naturalas, and oilsands) has been increasing rapidly in the areas of theoreal forest we study (Schneider et al., 2003). These activities haveignificant cumulative effects in terms of habitat loss, creation of
inear features on the landscape, and disturbance. The present stockf well sites is incorporated in our model. However, we have notncluded projections of energy activities in this paper nor have wencorporated the profits from energy resource extraction into thebjective function. These are important tasks for future research.
ing 221 (2010) 2590–2603

The density of well sites within a landscape has a significant nega-tive effect on the observed counts several bird species, according tothe statistical models developed for this exercise. The contributionsof the energy sector to royalties and provincial revenue is currentlyorders of magnitude greater than the forestry sector. The trade-offs between the two overlapping economic sectors (forestry andenergy) and the variety of ecological indicators will provide a muchmore relevant assessment of the challenges of land use manage-ment in the boreal forest. Incorporating energy into the objectivefunction will be difficult because of the uncertain nature of energyreserves and the complex exploration–extraction behavior of thesector.

Acknowledgements

We thank the following individuals for their contributionsthrough comments, collaborations, or other avenues: Stan Boutin,Fred Bunnell, Carl Walters, Werner Kurz, Chokri Dridi, PierreVernier, Michael Habteyonas, Nancy Bunch (née Holloway), DaveCheyne, Elston Dzus, Shawn Wasel, John Stadt, Keith McClain,Eric Butterworth, Jeff Beal, Craig DeLong, Don Bradshaw, andTed Gooding, and two anonymous reviewers. In addition wethank our industrial and government partners: Alberta Sustain-able Resource Development, Alberta Energy, Alberta Environment,British Columbia (BC) Ministry of Forests, Ducks Unlimited Canada,Alberta-Pacific Forest Industries Inc., Canadian Forest Products(BC), Weyerhaeuser Company, and Millar Western Forest ProductsLtd. This research was supported by a grant from the SustainableForest Management Network.

References

Alberta Forest Conservation Strategy Steering Committee (AFCSSC), 1997. AlbertaForest Conservation Strategy. Alberta Department of Environmental Protection,Edmonton, Alberta.

Alberta Forest Conservation Strategy, 1994. Summary Record Stakeholder AdvisoryGroup Workshop, Edmonton, Alberta, June 1–3.

Andalaft, N., Andalaft, P., Guignard, M., Magendzo, A., Wainer, A., Weintraub, A.,2003. A problem of forest harvesting and road building solved through modelstrengthening and Lagrangean relaxation. Oper. Res. 51, 613–628.

Armstrong, G.W., Adamowicz, W.L., Beck, J.A., Cumming, S.G., Schmiegelow, F.K.A.,2003. Coarse filter ecosystem management in a nonequilibrating forest. ForestSci. 49, 209–223.

Armstrong, G.W., 2004. Sustainability of timber supply considering the risk of wild-fire. Forest Sci. 50, 626–639.

Arthaud, G.J., Rose, D.W., 1996. A methodology for estimating production possibilityfrontiers for wildlife habitat and timber value at the landscape level. Can. J. ForestRes. 26, 2191–2200.

Bengtsson, J., Angelstam, P., Elmqvist, T., Emanuelsson, U., Folke, C., Ihse, M., Moberg,F., Nyström, M., 2003. Reserves, resilience and dynamic landscapes. Ambio 32,389–396.

Betts, M.G., Diamond, A.W., Forbes, G.J., Villard, M.-A., Gunn, J.S., 2006. The impor-tance of spatial autocorrelation, extent and resolution in predicting forest birdoccurrence. Ecol. Modell. 191, 197–224.

Burton, P.J., Messier, C., Adamowicz, W.L., Kuuluvainen, T., 2006. Sustainable man-agement of Canada’s boreal forests: progress and prospects. Ecoscience 13,234–248.

COSEWIC, 2008. COSEWIC Assessment and Status Report on the Canada WarblerWilsonia Canadensis in Canada. Committee on the Status of Endangered Wildlifein Canada, Ottawa, ON, www.sararegistry.gc.ca/status/status e.cfm.

Carson, R.T., 2000. Contingent valuation: a users guide. Environ. Sci. Technol. 34,1413–1418.

Cumming, S.G., 2001a. A parametric model of the fire size distribution. Can. J. ForestRes. 31, 1297–1303.

Cumming, S.G., 2001b. Forest type and wildfire in the Alberta boreal mixedwood:what do fires burn? Ecol. Appl. 11, 97–110.

Cumming, S.G., 2005. Effective fire suppression in boreal forests. Can. J. Forest Res.35, 772–786.

Cumming, S.G., Armstrong, G.W., 2001. Divided land base and overlapping foresttenure in Alberta, Canada: a simulation study exploring costs of forest policy.Forestry Chron. 77, 501–508.

Cumming, S.G., Armstrong, G.W., 2004. Divided landbase, overlapping tenures andfire risk. Forestry Chron. 80, 479–484.

Cumming, S.G., Cartledge, P., 2004. Spatial and Temporal Patterns of the Indus-trial Footprint in Northeast Alberta, 1960–2000. Sustainable Forest ManagementNetwork Research Communication. Sustainable Forest Management Network,Edmonton, Alberta.

http://www.sararegistry.gc.ca/status/status_e.cfm

odell

C

C

C

E

F

H

H

H

H

H

H

J

K

K

L

L

L

L

M

N

N


umming, S.G., Vernier, P., 2002. Statistical models of landscape pattern metrics,with applications to regional scale dynamic forest simulations. Landscape Ecol.17, 433–444.

umming, S.G., Burton, P.J., Klinkenberg, B., 1996. Boreal mixedwood forests mayhave no “representative” areas: some implications for reserve design. Ecography19, 162–180.

umming, S.G., Schmiegelow, F.K.A., Burton, P.J., 2000. Gap dynamics in boreal aspenstands: is the forest older than we think? Ecol. Appl. 10, 744–759.

cological Stratification Working Group, 1996. A National Ecological Frameworkfor Canada. Agriculture and Agri-Food Canada, Minister of Supply and ServicesCanada, Ottawa, ON.

ranklin, J.F., 1996. Ecosystem management: an overview. In: Haney, S., Boyce, M.(Eds.), Ecosystem Management: Applications for Sustainable Forest and WildlifeResources. Yale Univ. Press, New Haven, CT.

aila, Y., 1994. Preserving ecological diversity in boreal forests: ecological back-round, research and management. Ann. Zool. Fenn. 31, 203–217.

annon, S.J., McCallum, C., 2003. Using the Focal Species Approach for ConservingBiodiversity in Landscapes Managed for Forestry. Sustainable Forest Man-agement Network Synthesis paper. Sustainable Forest Management Network,Edmonton, Alberta.

arvey, B.D., Nguyen-Xuan, T., Bergeron, Y., Gauthier, S., Leduc, A., 2003. Forestmanagement planning based on natural disturbance and forest dynamics. In:Burton, P.J., Messier, C., Smith, D.W., Adamowicz, W.L. (Eds.), Towards Sustain-able Management of the Boreal Forest. NRC Research Press, Ottawa, ON, pp.395–432.

auer, G., Hoganson, H., 1996. Tailoring a decomposition method to a large-scale for-est management scheduling problem in northern Ontario. INFOR 34, 209–231.

oganson, H.M., Rose, D.W., 1984. A simulation approach for optimal timber man-agement scheduling. Forest Sci. 30, 220–238.

unter, M.L., 1993. Natural fire regimes as spatial models for managing borealforests. Biol. Conserv. 65, 115–120.

uutinen, A., Mäntymaa, E., Mönkkönen, M., Salmi, J., 2004. A cost-efficient approachto selecting forest stands for conserving species: a case study from northernFennoscandia. Forest Sci. 50, 527–539.

abzems, A., Kosowan, A.L., Hams, W.C., 1986. Mixedwood Section in an Ecologi-cal Perspective. Tech Bull 8, 2nd ed. Forestry Division, Saskatchewan Parks andRenewable Resources, Regina, Saskatchewan.

rawchuck, M.A., Cumming, S.G., in press. Biotic feedback and forest managementact as buffers to climate-driven increases in boreal forest fire activity. Ecol. Appl.

andres, P.B., Morgan, P., Swanson, F.J., 1999. Overview of the use of natural vari-ability concepts in managing ecological systems. Ecol. Appl. 9, 1179–1188.

arsson, T., Patriksson, M., Strömberg, A.-B., 1999. Ergodic, primal convergencein dual subgradient schemes for convex programming. Math. Program. 86,283–312.

ichtenstein, M., Montgomery, C., 2003. Biodiversity and timber in the CoastRange of Oregon: inside the production possibility frontier. Land Econ. 79, 56–73.

indenmayer, D.B., Franklin, J.F., Fischer, J., 2006. General management principlesand a checklist of strategies to guide forest biodiversity conservation. Biol. Con-serv. 131, 433–445.

arshall, E., Homans, F., Haight, R., 2000. Exploring strategies for improving the cost
effectiveness of endangered species management: the Kirtland’s Warbler as acase study. Land Econ. 76, 462–473.
alle, D., Montgomery, C., Arthur, J., Polasky, S., Schumaker, N., 2004. Modeling jointproduction of wildlife and timber. J. Environ. Econ. Manage. 48, 997–1071.

anang, D., Hauer, G.K., 2006. Estimating the costs of overlapping tenure constraints:a case study in Northern Alberta, Canada. Forest Policy Econ. 8, 610–624.

ing 221 (2010) 2590–2603 2603

Newbold, S., Eadie, J.M., 2004. Using species-habitat models to target conservationareas: a case study with breeding mallards. Ecol. Appl. 14, 1384–1393.

Norton, M., 2001. Status of Bay-breasted Warbler (Dendroica castanea) in Alberta.Wildlife Status Report No. 32. Alberta Environment, Fisheries and Wildlife Man-agement Division, and Alberta Conservation Association, Edmonton, Alberta.

Noss, R.F., Cooperrider, A.Y., 1994. Saving Nature’s Legacy. Island Press, Washington,DC.

Pearse, P.H., 2001. Ready for Change: Crisis and Opportunity in the Coast ForestIndustry. BC Ministry of Forests, Victoria, BC.

Polasky, S., Nelson, E., Lonsdorf, E., Fackler, P., Starfield, A., 2005. Conserving speciesin a working landscape: land use with biological and economic objectives. Ecol.Appl. 15, 1387–1401.

Polasky, S., Nelson, E., Camm, J., Csuti, B., Fackler, P., Lonsdorf, E., Montgomery, C.,White, D., Arthur, J., Garber-Yonts, B., Haight, R., Kagan, J., Starfield, A., Tobalske,C., 2008. Where to put things? Spatial land management to sustain biodiversityand economic returns. Biol. Conserv. 141, 1505–1524.

Rayfield, B., James, P.M.A., Fall, A., Fortin, M.-J., 2008. Comparing static versusdynamic protected areas in the Québec boreal forest. Biol. Conserv. 141,438–449.

Reed, W.J., Errico, D., 1986. Optimal harvest scheduling at the forest level in thepresence of the risk of fire. Can. J. Forest Res. 16, 266–278.

Rettie, W.J., Sheard, J.W., Messier, F., 1997. Identification and description of forestedvegetation communities available to woodland caribou: relating wildlife habitatto forest cover data. Forest Ecol. Manage. 93, 245–260.

Richards, E.W., Gunn, E.A., 2000. A model and tabu search method to optimize standharvest and road construction schedules. Forest Sci. 42, 188–203.

Schneider R, 2001. Forest Management in Alberta: A Review. Draft Report. AlbertaCentre for Boreal Research, Edmonton, AB.

Schneider, R.R., Stelfox, J.B., Boutin, S., Wasel, S., 2003. Managing the cumulativeimpacts of land uses in the Western Canadian Sedimentary Basin: a modelingapproach. Conserv. Ecol. 7 (1), 8.

Schmiegelow, F.K.A., Cumming, S.G., 2004. The Remote Areas Project: A Retro-spective Study of Avian Indicators of Forest Change. The Sustainable ForestManagement Network, Edmonton, AB.

Sougavinski, S., Doyon, F., 2005. A Review of Spatial Distribution Guidelines forthe Managed Canadian Boreal Forest. Sustainable Forest Management Network,Edmonton, AB (translated from French; Original title: Directives de répartitionspatiale existantes pour la forêt boréale canadienne aménagée).

Stocks, B.J., Mason, J.A., Todd, J.B., Bosch, E.M., Wotton, B.M., Amiro, B.D., et al., 2002.Large forest fires in Canada, 1959–1997. J. Geophys. Res. 108 (FFR5), 1–12.

Toth, S.F., McDill, M.E., Rebain, S., 2006. Finding the efficient frontier of a bi-criteria,spatially explicit, harvest scheduling problem. Forest Sci. 52, 93–107.

Vernier, P., Cumming, S.G., Schmiegelow, F.K.A., 2006. Application of Remote AreasProject (RAP) Models to Alpac’s DFMP Process. Alberta Pacific Forest IndustriesInc, Boyle, Alberta, 26 pp.

Vernier, P., Schmiegelow, F.K.A., Hannon, S.J., Cumming, S.G., 2008. Generalizabilityof songbird habitat models in boreal mixedwood forests of Alberta. Ecol. Modell.211, 191–201.

Watzold, F., Drechsler, M., Armstrong, C.W., Baumgartner, S., Grimm, V., Huth, A., etal., 2006. Ecological-economic modeling for biodiversity management: poten-tial, pitfalls, and prospects. Conserv. Biol. 20, 1034–1041.

Weber, M., 2004. Assessing the Effectiveness of Tradable Landuse Rights for Bio-diversity Conservation: An Application to Canada’s Boreal Mixedwood Forest.Fondazione Eni Enrico Mattei Working Papers, Milan, Itali.

Williamson, T., Hauer, G., Luckert, M.K., 2004. A restricted Leontief profit functionmodel of the Canadian lumber and chip industry: potential impacts of US coun-tervail and Kyoto ratification. Can. J. Forest Res. 34 (9), 1833–1844.

Tradeoffs between forestry resource and conservation values under alternate policy regimes: A...

Documents

Transcript of Tradeoffs between forestry resource and conservation values under alternate policy regimes: A...