a hierarchical spatial framework for forest landscape planning

Ecological Modelling 182 (2005) 25–48

A hierarchical spatial framework for forest landscape planning

Pete Bettingera,∗, Marie Lennetteb,1, K. Norman Johnsonb,2, Thomas A. Spiesc,3

a Warnell School of Forest Resources, University of Georgia, Athens, GA 30602-2152, USAb Department of Forest Resources, Oregon State University, Corvallis, OR 97331, USAc USDA Forest Service, Pacific Northwest Research Station, Corvallis, OR 97331, USA

Received 26 August 2003; received in revised form 10 June 2004; accepted 6 July 2004

Abstract

A hierarchical spatial framework for large-scale, long-term forest landscape planning is presented along with example policyanalyses for a 560,000 ha area of the Oregon Coast Range. The modeling framework suggests utilizing the detail provided bysatellite imagery to track forest vegetation condition and for representation of fine-scale features, such as riparian areas. Spatialdata are then aggregated up to management units, where forest management decisions are simulated. Management units may alsobe aggregated into harvest blocks to closer emulate management behavior. Land allocations, subdivisions of landowner groups,can be used to represent different levels of management. A management unit may contain multiple land allocations, such asriparian management emphases that vary based on distance from the stream system. The management emphasis required by eachland allocation is retained in the simulation of policies. When applied within a large-scale forest landscape planning context, theimplications of policies that suggest clearcut size restrictions, minimum harvest ages, or the development of interior habitat areasc flow harvestl however,a ry, whichs loser to them©

K s

f

(k(

tainds is

s. Inixedciesroad-icies

0

an be assessed. Simulations indicated that the minimum harvest age constraint has a stronger influence on even-evels than do maximum clearcut size or interior habitat area constraints. Even-flow timber harvest level objectives,lso have an effect on the results: time periods beyond the constraining time period show a build-up of timber inventouggests a possible relaxation or modification of the objective in order to achieve average harvest ages that are cinimum harvest age.2004 Elsevier B.V. All rights reserved.

eywords:Forest management; Forest planning; Simulation; Operations research; Policy analysis; Geographic information system

∗ Corresponding author. Tel.: +1 706 542 1187;ax: +1 706 542 8356.

E-mail addresses:[email protected]. Bettinger), [email protected] (M. Lennette),[email protected] (K.N. Johnson), [email protected]. Spies).

1 Tel.: +1 541 737 2377.2 Tel.: +1 541 737 3090.3 Tel.: +1 541 750 7354.

1. Introduction

The development of forest policies that both sussocial and economic values of forests and watershea major challenge for policy-makers and managerOregon’s Coast Range (USA), for example, the mownership pattern of the region and the major polirecently introduced have suggested a need for bscale forest landscape policy simulations. The pol

304-3800/$ – see front matter © 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.ecolmodel.2004.07.009

26 P. Bettinger et al. / Ecological Modelling 182 (2005) 25–48

now in effect, though, were enacted without examin-ing simulations of likely landscape conditions and as-sociated effects. Rather, they were based on stand-levelanalyses of forest management policies on individualownerships. Policy-makers rarely examine the long-term effects of specific policies or compare policies inboth a spatial and temporal context because they havehistorically lacked the information to do so (Gustafsonand Crow, 1994). Spatial forest landscape planningmodels can assist policy analyses through their abil-ity to evaluate policies at spatial and temporal scalesthat are otherwise difficult or impossible to accomplish(Gustafson et al., 2000).

Forest landscape planning involves the examinationof forest management alternatives or policies acrosslarge areas and long time frames. What sets forestlandscape planning apart from traditional forest plan-ning, or harvest scheduling, is an emphasis on themodeling of management behavior of all landownerswithin a large geographic area (500,000 ha or more).The suite of planning tools available for scheduling orsimulating management activities on landscapes variesfrom the traditional exact techniques (i.e., linear andinteger programming) to the non-traditional, inexacttechniques (i.e., heuristic programming and simula-tion). The non-traditional techniques, however, are of-ten the only alternative for modeling very large sys-tems, with simulation modeling frequently seen as themost appropriate method. Simulation models for for-est landscapes can be developed to provide a spatiala val-u ared amicna sh1

vel-o hav-i n( icale ape.O ow,1 8;G la-t cesa stedl

A number of limitations characterize the currentsuite of forest management simulation models. First,the models generally lack the integration of activitieswithin a spatial hierarchical structure. The allocationof management activities is thus applied either to rasterdatabase pixels or to management units, with limitedrecognition of higher (aggregation of harvest units) orlower (heterogeneity within a management unit) scalesof the system. In addition, only a few variables, suchas transition probabilities or stand ages, are used to in-fluence the allocation of management activities. Otheraspects of the system, such as management intensities,or the spatial position of landscape features, may al-low a model to more closely emulate actual land man-agement processes. Further, in a number of simulationmodels, the spatial allocation of harvest activities uti-lizes random processes that are unrelated to landownerobjectives. And, key landscape features, such as topog-raphy and stream networks, may be ignored.

Many natural resource information managementsystems contain a hierarchy of spatial data (MacMillanet al., 2004), and as with many large-scale planning ef-forts, the analysis process is also hierarchical, rangingfrom large regions with broad goals to small areas withspecific operational aspects (Church et al., 2000). Thesmaller areas, or lower levels in the hierarchy, are gen-erally used to provide forest structural conditions, andthe larger regions are used to provide constraining con-ditions (Burnett and Blaschke, 2003). Each componentof a hierarchy should be structured to reflect their real-w as limitt ulti-sn hichc entso )d vid-u levelg -s s) andt her-l

or-m ob-s .F gi-c nts

nd temporal context in which policy-makers can eate alternatives. Simulation models, in general,eveloped to capture relevant features of the dynature of some “target system” under study (Birtand Ozmizrak, 1996), and their reliability dependighly on how well the models reflect reality (Li et al.,993).

A number of simulation models have been deped in the last two decades to model events or be

ors across forested landscapes.Franklin and Forma1987)were one of the first to simulate the ecologffects of forest management activities on a landscthers (Flamm and Turner, 1994; Gustafson and Cr994, 1996;Wallin et al., 1994; Johnson et al., 199ustafson et al., 2000) have since developed simu

ion models for management activities or disturbant various spatial and temporal scales within fore

andscapes.

orld analog (Harris and Gorley, 2003). Focusing oningle scale, such pixels in a GIS database, mayhe usefulness of a model to represent complex mcale systems (Burnett and Blaschke, 2003). Most plan-ing processes center on a basic unit of analysis, wan either be disaggregated into smaller componr aggregated into larger ones.Bragg et al. (2004escribe a hierarchical system for managing indial forest stands using tree-level data and stand-oals.Burnett and Blaschke (2003)describe a landcape analysis system that utilizes patches (standheir topological relationships to accommodate higevel landscape goals.

Levels of organization in a hierarchy of spatial infation are typically definitional, and defined by the

erver (Burnett and Blaschke, 2003) or planning teamreemark (1995), for example, describes a bioloally driven hierarchy for application in assessme

P. Bettinger et al. / Ecological Modelling 182 (2005) 25–48 27

of ecosystem processes on agricultural land, andHarrisand Gorley (2003)describe a ecologically driven hier-archy for modeling hydrologic systems. Our model is aby-product of the need to adequately represent both theecology of a landscape and the behavior of landown-ers. Our intent is not to review the basis for ecolog-ical (or socio-economical) driven hierarchies, asWuand David (2002)have previously provided a detaileddiscussion of how complex systems can be hierarchi-cally organized using hierarchy theory, and discussedthe need to identify the spatial patterns that are rel-evant to the processes of interest. Our model involvesthe use of both top-down (where broad scale constraintsare important) and bottom-up (where local interactionsare important) approaches described inWu and David(2002) in defining the hierarchy that is important forlandscape planning in western Oregon.

In the work presented here, we try to overcomemany of these problems, to facilitate a close simula-tion of an actual land management system and evaluatechanges to management policies. Toward that end, theLAndscape Management Policy Simulator (LAMPS)(Bettinger and Lennette, 2003) was developed. Our ob-jective here is to detail the mathematical approachestaken in LAMPS to simulate the management of pri-vate industrial and public (state governed) forest ar-eas across a broad, heterogeneous landscape. Theseprocesses are not described inBettinger and Lennette(2003) nor elsewhere. The mathematical approachesrequire a hierarchy of spatial information to representt emenb er ofp iono util-i

e useo ing.F atialh nceb ed inB lel notd andh ntedh l Ivs ap-p and-

scape planning model that can be categorized as bothmodels 5(b) (management models: management of nat-ural resources), and 8(a) (models of terrestrial ecosys-tems: forests) of the delineation of models provided byJørgensen (1997).

2. Methods

The major stakeholders involved in the managementof public and private land in the Oregon Coast Rangeinclude theOregon Board of Forestry (1995), OregonDepartment of Forestry, U.S. Forest Service, Bureau ofLand Management, several industrial timber organiza-tions, and a small woodland owners organization. Eachhas the power to affect land-use decisions at a broad(e.g., Oregon Board of Forestry) or local scale. Eachalso, through numerous meetings, has provided inputand feedback regarding the model structure describedhere, and the utility of the model results in inform-ing the policy debate. The framework described belowaddresses the strategic and tactical objectives of the in-dustrial and state-managed land in the Coast Range. Forexample, as a whole the industrial group behavior hasbeen to produce an even-flow of timber harvest volumefrom the landscape, yet each industrial organizationmay have different tactical goals (e.g., green-up poli-cies, leave tree retention policies). On state-managedland, we have determined that the objective is to alsoproduce an even-flow of timber harvest volume, yetw for-e

twod capep me-w de-v

2

atialf ise trata)a . Allf latedt eredh hi-e yet

he heterogeneous landscape as well as managehavior. We also present the results of a numbolicy simulations applied to the north coastal regf the Oregon Coast Range (USA), to illustrate the

ty of the model.This research presents two advancements in th

f simulation techniques for forest landscape plannirst, it illustrates the usefulness of a large-scale spierarchy for forest landscape planning. The differeetween this approach and the approach describettinger et al. (2003)is the emphasis here on multip

andowners and the ability to model a flexible (i.e.,elineated a priori) spatial arrangement of harvestabitat blocks. Second, the spatial hierarchy preseere allows the modeling of over 5 million Modeariables (Johnson and Scheurman, 1977), a problemize perhaps intractable for integer programmingroaches. The simulation process utilizes a forest l

t hile striving to develop a landscape with certainst structural conditions.

In modeling the behavior of these landowners,istinct areas of concern relevant to forest landslanning efforts are the design of the spatial fraork around which the modeling occurs, and theelopment of the problem formulation.

.1. Hierarchical spatial framework

Most forest planning efforts assume that the spramework is relatively simple: a single ownershipxamined, and management units (or vegetation sre designed to accommodate decision variables

orest conditions and management decisions reo the management units (or strata) are considomogenous. The next few sections describe ararchical framework that builds from a small,


recognizable land unit to the larger landscape, allowingthe recognition and maintenance of vegetation condi-tions at a fine scale, yet facilitating the simulation ofdecisions at larger scales.

2.1.1. Basic simulation unitsThe decision regarding the smallest unit recognized

in a modeling effort is usually made with two con-siderations in mind: the need to use units that ade-quately represent the heterogeneity of the landscape,and the need to develop a set of units that does not taxthe available computer systems. In many university-or agency-sponsored large-scale landscape modelingefforts, raster databases developed from satellite im-agery are the primary data structure around which logicand simulation are based. These basic simulation unitsgenerally range from 30 m (e.g., Landsat imagery) to1.1 km (e.g., AVHRR imagery) square pixels. MostNorth American forestry companies, on the other hand,use a basic simulation unit consisting of irregularlyshaped vector polygons, where the minimum size is2–4 ha. Each of these structures assumes that the forestconditions in each unit are homogeneous. In the indus-trial case, the potential to ignore fine-scale vegetationheterogeneity is relatively high. Alternatively, if onewere to use 30 m pixels as the basic simulation unitsfor a 500,000 ha landscape, one would need to recog-nize over 5.5 million units, since each is approximately0.09 ha in size. If the number of attributes that need tobe tracked for each unit is large, computer memoryc

im-a plan-n lopedb htb ningp thath etail( truc-t us-i s ofs ncef cog-n essw 2)d esult-i cingt mil-

Fig. 1. Basic simulation units: aggregations of similar, contiguousraster grid cells.

lion in a 500,000 ha area. These basic simulation unitscan be viewed as Model I decision variables (Johnsonand Scheurman, 1977), where the vegetation conditionof each is recognized and maintained through time.

2.1.2. Management unitsWhile forest structural attributes may be assigned

and tracked at the basic simulation unit scale, decisionsregarding forest management activities are generallymade at larger scales. Management units consist of ar-eas that encompass terrain and vegetation characteris-tics of a size appropriate for logging systems typical ofa region. Therefore, basic simulation units can be ag-gregated up to management units, where the resultingsize is about 10–20 ha (Fig. 2). The vegetation condi-tion of a management unit is therefore either the sum(e.g., for timber volume) or weighted average (e.g., forage) of the conditions of basic simulation units.

2.1.3. Aggregations of management unitsIn many cases, management units are aggregated

for temporally simultaneous treatment. In the westernU.S., contiguous management units of similar charac-teristics (Fig. 3) are commonly scheduled for simul-taneous treatment (e.g., clearcut, thinning, etc.) dueto the economy of scale of the cumulative manage-ment activities. Thus, to emulate landowner behavior,management units may need to be aggregated usinga blocking process described inBettinger and John-s -m rest ypes

ould be taxed.We assume here that classified Landsat satellite

gery is available for subsequent forest landscapeing and analysis. The vegetation database devey Ohmann and Gregory (2002), as an example, mige considered a crucial part of a landscape planrocess. This raster database consists of pixelsave associated with them fine-scale vegetation dtree lists) that describe the underlying vegetative sure. Given the potential limitations noted above ofng satellite imagery, aggregating contiguous pixelimilar slope class, vegetative condition, and distarom streams can reduce the number of units reized (Fig. 1) to a more manageable size. This procas performed using theOhmann and Gregory (200atabase for the Coast Range of Oregon, and the r

ng size of basic simulation units was 0.12 ha, reduhe number of units to be recognized to about 4.2

on (2003)or Nelson (2001). In addition, the developent of interior habitat blocks on public land requi

he development of contiguous areas of certain t


Fig. 2. Management unit: an aggregation of basic simulation units, defined by topography and dominant vegetation.

Fig. 3. Harvest blocks: aggregations of management units for simultaneous treatment.


of vegetation. Since contiguous management units maybe aggregated for harvest or habitat goals, knowledgeof the adjacency relationships of management units isneeded.

2.1.4. Land allocationsIn some cases, land ownerships are further subdi-

vided and placed into allocations with differing em-phases of management. For example, wilderness areasare generally separated from general forest manage-ment areas. Since basic simulation units are small, theyhave assigned to them a single land allocation in thisspatial hierarchical structure. Management units, on theother hand, may include multiple land allocations. Forexample, areas closer to streams may require the re-tention of more residual leave trees than areas furtheraway from streams, thus two (or more) land allocationsmay be present in a single management unit. In thesecases, when a forest management activity is simulatedin a management unit, the level of activity simulatedmay differ based on the land allocation assigned to eachbasic simulation unit.

2.2. Management prescriptions

The level of detail contained in management pre-scriptions within a forest planning effort varies fromthe rather general (e.g., clearcut or thin entire manage-ment units), to the complex (e.g., clearcut and leavesome residual leave trees, leave undisturbed areas neart tureo ed int an-a avior.H ow-i pro-c ures treep nd-l heri em.

2eral

p restl age-m nceo nt

activities are allowed within a certain distance of thestream system (Fig. 5); (3) limited management activi-ties are allowed within a certain distance of the streamsystem, yet only within certain vegetation types (e.g.,conifer) (Fig. 6). To enable these policies to be mod-eled within management units that encompass areasthat extend well beyond the boundaries of typical ri-parian management areas (30–100 m), these decisionsare made at the basic simulation unit scale, and arebased on the type of activity assigned to a land allo-cation. Our example simulations use the second casenoted above.

2.2.2. Residual leave tree policiesResidual leave tree policies can be modeled at the

management unit scale, where the number of resid-ual leave trees within a simulated clearcut is constant.However, since more than one land allocation mightbe present within a management unit, and since foreststructural conditions are tracked at the basic simulationunit scale, the leave tree policy may also be modeled atthe basic simulation unit scale. For example, within asingle management unit, simulating clearcuts near thestream system may require retention of a large numberof residual leave trees under certain policies, whereasclearcuts further away from the stream system may re-quire retention of a smaller number of residual leavetrees. These two policies can be modeled simultane-ously if performed at the basic simulation unit scale.The level of residual leave trees that we model in oure iredb -i ndm

2ay

r not nt,p alh tionsa rts.S ntedh asics tai-l lan,h de att m a

he stream, etc.). To illustrate the hierarchical naf the forest landscape planning process suggest

his research, we describe a complex system of mgement that closely emulates management behere, with regard to clearcutting decisions, the foll

ng aspects are incorporated into the simulationess: riparian policies, transition probabilities, futtand management intensities, and residual leaveolicies. In addition, a short discussion of the sta

evel forest structure projections is provided to furtllustrate the multi-scale nature of the modeling syst

.2.1. Riparian policiesIn the modeling effort described here, three gen

olicies can be used to describe the behavior of foandowners with respect to riparian areas: (1) man

ent activities are prohibited within a certain distaf the stream system (Fig. 4), (2) limited manageme

xample simulations is consistent with those requy the Oregon Forest Practices Act (Oregon State Leg

slature, 2001) and guidance provided by public laanagers.

.2.3. Management intensitiesThe intensity of forest management, which m

ange from “very low” (e.g., natural regeneration,hinning, final harvest) to “very high” (e.g., plare-commercial thin, fertilize, commercial thin, finarvest), is inherent in the management prescrippplied to units modeled within forest planning effoince, in the hierarchical spatial structure preseere, forest structural information is tracked at the bimulation unit scale, management intensity can beored to this scale. In a typical forest management powever, management intensity decisions are ma

he management unit scale, and when viewed fro


Fig. 4. Riparian policy where no activity is allowed within a certain distance of the stream system.

larger landscape (or ownership) perspective, the rangeof intensity may vary by management unit. For exam-ple, a private forest company may manage 25% of theirland at a high intensity, 25% at a low intensity, and 50%somewhere between the two levels, depending on sil-vicultural budgets. A public agency may manage mostof their land at a medium management intensity (e.g.,plant, final harvest). Therefore, it may seem reason-able to be able to model a range of intensity of man-agement for forest plantations, and choose a level ofintensity to model for an entire management unit whenclearcut activities are scheduled. With a distribution ofmanagement intensities, a simulation model can be de-signed to randomly assign the level of intensity mod-eled as clearcuts are scheduled. All basic simulationunits within the management unit are then assignedprescriptions of similar management intensity, to theextent possible. For example, all conifer basic simu-

lation units may be modeled at a high managementintensity, yet hardwood basic simulation units wouldbe modeled as simply regenerated areas, if intensivehardwood management is not common to a region.

2.2.4. Transition probabilitiesDeciding how clearcut areas, hence regenerated

stands, are to be simulated in periods of time af-ter clearcutting has been scheduled is a key issuewhen evaluating long-term projected conditions oflarge landscapes. Many forest growth and yield mod-els lack the ability to adequately represent the typeof trees regenerating after harvest without some inputfrom the modeler. Most often in modeling processes,a single, or limited transition is assumed to occur afterharvest. Assuming a single transition of forested ar-eas, however, may mis-represent the heterogeneity ofmanagement and ecological processes inherent across


Fig. 5. Riparian policy where limited activity is allowed within a certain distance of the stream system.

a landscape after disturbance. For example, to assumethat all clearcut areas will return as conifer planta-tions on private industrial areas is overly optimistic,and assumes one disregards the problems associatedwith failed planting efforts, competition from residualhardwood species, and other processes that act to sup-press the regeneration success of planted conifers. Tomore adequately model the type of forest that regener-ates after harvest, it would seem reasonable to use a setof transition probabilities that emulate historical pro-gression of forested areas, and subsequently infer thatthe transition is probabilistic. The probabilities mightbe a function of the regeneration management intensityassumed, but perhaps also a function of the distanceto the stream system and the type of vegetation thatresided in the harvested area prior to activity. Withinthe spatial hierarchy described here, transition proba-bilities are applied at the basic simulation unit scale,

where the management intensity of the managementunit can be considered along with the distance eachbasic simulation unit is from the stream system, andthe type of vegetation that was present within the basicsimulation unit prior to harvest. By doing so, the simu-lations can retain some of the landscape heterogeneitythat was present at the initiation of the planning process(as reflected in the initial vegetation database).

2.2.5. Stand-level projections of forest structureIndividual tree forest growth and yield models are

used to facilitate the estimation of forest structurewithin each basic simulation unit. Two growth andyield models were utilized: (1) ORGANON (Hannet al., 1997), a distance independent, individual treemodel based on empirical relationships and validatedfor conifer and some mixed stands up to 100 years ofage, and (2) ZELIG (Urban and Shugart, 1992, Garman


Fig. 6. Riparian policy where limited activity is allowed within a certain distance of the stream system, yet only in certain forest types.

et al., 2003), a distance independent, individual treemodel based on theoretical ecological (gap) relation-ships for use for any stand age and stand composition,but with relatively little empirical validation for anyparticular age and stand condition. Since ORGANONis used by many industrial forestry organizations inwestern Oregon, and thus has credibility with industriallandowners, ORGANON has been employed for man-agement prescriptions that involve regeneration har-vest of stands younger than 100 years of age. We alsowanted to use a growth and yield model that could sim-ulate ecological succession in older forests representa-tive of public reserves. Since ORGANON only mod-els mortality with inter-tree competition relationships(rather than with natural gap disturbances) ZELIG isemployed for management prescriptions on public landwhere regeneration harvests often occur in stands withages in excess of 100 years, or where a regenera-

tion harvest is not considered. ZELIG was calibratedto ORGANON timber volume projections for simu-lations of young, intensively managed stands on pub-lic lands. Both models produce similar projections ofbasal area, quadratic mean diameter, and tree densityfor the first 80 years of a forest rotation. For the projec-tions of landscape condition described here, we utilizethe CLAMS ORGANON data to represent the foreststructural conditions of private industrial lands, and theZELIG data to represent forest structural conditions onpublic lands. In the next generation of landscape sim-ulations, and with further calibration of ZELIG pro-jections to ORGANON projections, we plan to useZELIG to describe the forest structural conditions of alllandowners. With proper manipulation, of course, anygrowth and yield model can be used to represent theforest conditions of all landowners. The trouble withORGANON would be to determine how to represent


gap disturbances typical of older forests, which cur-rently can only be represented by thinning prescrip-tions.

2.3. Problem formulation

Two general types of management behavior are nextdescribed. One relates to the behavior of a large groupof private industrial landowners, where the goal of man-agement may be to maximize individual organizationalobjectives, usually involving wealth. The other relatesto a large public (state) ownership of land, where thegoals of management may be two-fold: to producecommodities, and to provide forest structural condi-tions that might facilitate the achievement of ecologicalgoals.

2.3.1. Modeling of private industrial forestmanagement behavior

The behavior of individual industrial landownerscan generally be described as one that seeks to max-imize the value of the asset (land and timber) to theowners of the company, whether they be stock hold-ers, families, or individuals. However, when viewed asa whole, across a broad landscape and through time,the behavior of the landowner group might be char-acterized differently due to variations in managementobjectives, inventory levels, and other economic, po-litical, and regulatory circumstances. For example, ino lg ablea thei lue.G nerg ounto fort thatm .

w -bt lev

a binary (0,1) variable indicating whether or not basicsimulation unitbwas harvested in time periodt; ab isthe area represented by basic simulation unitb.

To understand the constraints influencing the be-havior of the private industrial landowner group in theCoast Range of Oregon, a number of meetings werearranged over the 1998–2002 time period to discussmodeling approaches and assumptions. The group, asa whole, must manage their forest land within theguidelines specified in the Oregon Forest PracticesAct (Oregon State Legislature, 2001), which limits themaximum size of clearcut areas, requires a green-uptime period of about 5 years between adjacent clearcutareas that might result in a clearcut area greater thanthe maximum area prescribed, and limits the activityallowed in riparian areas. Although there was somevariation in the management behavior derived from dis-cussions with the industrial landowners, a number ofsignificant aspects were deemed important. First, thelandowners stressed that any modeling effort shouldclosely emulate the Forest Practices Act. Second, theact of blocking harvest areas for simultaneous treat-ment was thought important, given their propensity todo so in practice. Third, modeling prescriptions thatincluded intensive management of future forests wasidentified as important, to the extent identified in sur-veys of management behavior (Lettman, 1998). Fi-nally, providing an indication of the impact of policiesthat might further restrict current management behaviorwas, while contentiously debated, thought important.T them umh rosss con-s ibedc

mei arys rys s setf latedb entu de-s sm rests timep duledt uent

ne study (Lettman and Campbell, 1997), the industriaroup as a whole tended to harvest a relatively stmount of timber volume over time, even though

ndividual landowners each sought to maximize vaiven the propensity of the private industrial landowroup, as a whole, to harvest a relatively stable amf timber volume over time, the objective function

his group was developed to provide simulationsaximized an even-flow of timber harvest volume

Objective function

MaximizeB∑

b=1

T∑t=1

vbxb,tab

(1)

hereb is a basic simulation unit;B is the total numer of basic simulation units;t is a time period;T is

he total number of time periods;vb is the harvestabolume per unit area in basic simulation unitb; xb,t is

hese policies included, among others, reducingaximum clearcut size and implementing a minimarvest age. The general flow of information acpatial scales, and related to the objective andtraints of the industry modeling process, is descronceptually inFig. 7.

The achievement of an even-flow of timber volus accomplished with a modified version of binearch.Leuschner (1990)describes the basic binaearch process. Here, a target timber volume ior each time period, and harvests are accumuy blocking together management units for treatmsing a dynamic, deterministic blocking processcribed inBettinger and Johnson (2003), a procesuch different than scheduling harvests by fo

trata. The scheduling process starts with the firsteriod, and once enough harvests have been sche

o exceed the volume target, harvests in subseq


Fig. 7. Conceptual model of the flow of information across spatial scales, and in association with the objective and constraints, for the industriallandowner behavior modeling process.

time period(s) are scheduled. If enough volume can bescheduled in all time periods, the target is increased,and the process begins anew. If there is not enough vol-ume available in one or more time periods, the target isreduced, and the process begins anew. The even-flowconstraint can be described as:

B∑b=1

vbxb,tab ≥ VTt ∀t (2)

where VTt is the target volume to be harvested in timeperiodt.

While the summation of timber volume scheduledfor harvest occurs at the basic simulation unit scale,adjacency restrictions are modeled at the managementunit scale. There are two general approaches to ap-plying adjacency restrictions in forest planning. Thefirst is the application of a technique called the “UnitRestriction Model” (URM), which makes it possibleto restrict harvest activity in management units that


neighbor other management units already scheduledfor clearcutting, disallowing neighboring managementunits from being treated in the same time period (ornear-time periods). Constraints such as the one that fol-lows can be used to control URM problems (Murray,1999).

Xi,t + Xj,t ≤ 1 ∀i, t, j ∈ Ni (3)

wherei is the management unit;Xi ,t is a binary (0,1)variable indicating whether or not management uniti is clearcut during time periodt; Ni is a set of allmanagement units adjacent to management uniti.

Management units (10–20 ha) in our spatial hierar-chy are typically smaller than maximum clearcut sizerestrictions (50 ha), and thus it may be important toschedule adjacent units for harvest to produce a feasiblemanagement plan containing “harvest blocks.” In suchcases, a second technique, called the “Area RestrictionModel” (ARM), can be used to assign simultaneoustreatments to adjacent management units, as long asthe total contiguous area does not exceed the maxi-mum area limit (Murray, 1999). Recursive functionsare generally used to evaluate the resulting spatiallysprawling harvest block of management units. To as-sess ARM problems, constraints such as the one notedbelow are used.

Xi,tCAi,t +∑

j ∈ Ni∪Si

Xj,tCAj,t ≤ MCA (4)

wc aln ns allu nitp ors,e estb

de-sA cks,a ar-v ings timep

umh tionu forc the

age of the vegetation within the basic simulation unitis greater than a minimum age.

If Ageb,t ≥ MHA xb,t ∈ {0, 1}Elsexb,t = 0

(5)

where Ageb,t is average age of the trees in basic sim-ulation unitb during time periodt.

Obviously when using satellite imagery as thedatabase describing the vegetation across a landscape,a considerable amount of heterogeneity in forest struc-ture (i.e., age) can be present within management units.A constraint and some logic is used to control thescheduling of small portions of management units forharvest in any particular time period, if only a small per-centage of basic simulation units have vegetation agesgreater than the MHA. First, a cursory examination ofthe potential harvest opportunities of basic simulationunits is performed by evaluating the percentage area ofeach management unit that could be clearcut harvestedin each time period:

If Ageb,t ≥ MHA andrb = 0 thenyb,t = 1

Elseyb,t = 0(6)

whererb is a binary variable indicating whether (1)or not (0) a riparian restriction has been placed on theclearcutting of basic simulation unitb; yb,t is a binary(0,1) variable indicating whether or not basic simula-t ur-i

asics aren canp an-a acht

[

wap td

achm time

here CAi ,t is the area of management uniti that islearcut during time periodt;Si is the subset of the totumber of harvested management units that containits adjacent to the neighbors of management uilus all units adjacent to the neighbors of the neighbtc.; MCA is maximum permissible area of the harvlock.

The blocking process used in this research iscribed inBettinger and Johnson (2003). It includes anRM technique to evaluate the size of harvest blond a URM technique to prevent two clearcut hest blocks from merging together if they are beimulated for clearcut harvests during the sameeriod.

In addition to adjacency restrictions, a minimarvest age (MHA) constraint allows basic simulanits within a management unit to be consideredlearcut harvest in a particular time period only if

ion unitb could potentially be clearcut harvested dng time periodt.

Here, if the age of the vegetation describing a bimulation unit is greater than the MHA and thereo riparian restrictions, the basic simulation unitotentially be clearcut. The percentage of each mgement unit that potentially could be clearcut in e

ime period is then examined.

∑Bi

b=1yb,tab∑Bi

b=1ab

]= γi,t ∀i, t (7)

hereBi is the subset of basic simulation unitsb thatre contained within management uniti; γ i ,t is theercentage of management uniti that could be clearcuuring time periodt.

With this determination of the percentage of eanagement unit that could be clearcut in each


period, the possible set of values for decision variablesXi ,t andxb,t are known.

IF γi,t ≥ MHP thenXi,t ∈ {0, 1}; xb,t ∈ {0, 1}ElseXi,t = 0; xb,t = 0 ∀b ∪ Bi

(8)

where MHP is the minimum percentage area requiredprior to simulating a management unit for a clearcutactivity.

Three pieces of information are tracked with eachbasic simulation unit, allowing one to understand thestructural characteristics contained within: the vegeta-tion conditions (tree list or other information allow-ing one to compute age and volume), the managementprescription assumed (should it include intermediatetreatments such as thinnings), and the time of the lastregeneration harvest. The assignment of managementintensity to a management unit is made at the time ofclearcutting, and with this information, the vegetativecondition and prescription assigned to the regeneratedbasic simulation units can be defined. The transitionprobabilities key off of the management intensity as-signed to the management unit, the distance each basicsimulation is from the stream system, and the pre-clearcut vegetation conditions. In addition, the leavetree policy for each land allocation indicates the typeof structural legacy remains in the regenerated basicsimulation units. This process allows both basic sim-ulation units and management units to be simulatedfor more than one clearcut activity during the planningh

∑

∑

2b

thea e re-f enta nciesi wedf gicalo is as ities

should be self-sustaining, from economic, ecological,and social perspectives, thus the emphasis on multiple-use of forest resources. To model the behavior of thepublic forest management group in the Coast Rangeof Oregon, a number of meetings were arranged overthe 1998–2002 time period to discuss modeling ap-proaches and assumptions with state forest managers.The group, as a whole, must manage their forestlandwithin the guidelines specified in their respective forestmanagement plans as well as the State Forest PracticesAct (Oregon State Legislature, 2001). These guidelineslimit the maximum size of clearcut areas, require agreen-up time period of about 5 years between adja-cent clearcut areas that might result in a clearcut areagreater than the maximum area prescribed, limit theactivity allowed in riparian areas, suggest the devel-opment of a diverse forest structure, and suggest thedevelopment of interior habitat (older forest) manage-ment areas. As with the private industrial landownergroup, the state forest managers stressed that any mod-eling effort should closely emulate the Forest PracticesAct. In addition, modeling prescriptions that reflecteda medium level of management intensity for futureforests (e.g., plant, commercial thin, final harvest) wasidentified as important. Finally, the managers indicatedthat the simulations should seek to emulate the devel-opment of a diverse forest structure and the develop-ment and maintenance of interior habitat areas, whileproviding a stable timber harvest volume.

In our public forest management example, the sameo in-d max-i hei tede intso dja-c nitsa ringt t isa tageo forc erea ffer-e thes n ofs lderf evel-o (2) a

orizon.

T

t=1

xb,t ≥ 0 ∀b (9)

T

t=1

Xi,t ≥ 0 ∀i (10)

.3.2. Modeling of public forest managementehavior

Public forest management varies according togency charged with managing forest land. Here, w

er to “public” as state (non-federal) land managemgencies. The management behavior of public age

n North America in the past two decades, when vierom the landscape level, has emphasized ecolobjectives along with economic objectives. Theretrong belief that public agency management activ

bjective function used in the modeling of privateustrial forest management behavior is assumed:

mize an even-flow of timber harvest volume. Tmplementation of this objective is achieved, as noarlier, using a binary search criteria. The constraf the analysis include a unit restriction model of aency, where neighboring individual management ure prohibited from being scheduled for clearcut du

he same time period. In addition, a MHA constrainpplied, as noted above, and a minimum percenf area within management units must be availablelearcut prior to scheduling a clearcut activity. Thre two aspects of this modeling process that dintiate it from the private industrial process: (1)cheduling process seeks to achieve a distributiotructural conditions in five structural classes (oorest, layered canopy development, understory dpment, closed canopy, and regeneration), and


distribution of sizes of blocks of interior habitat needto be simulated. The general flow of information acrossspatial scales, and related to the objective and con-straints of the public land (state) modeling process, isdescribed conceptually inFig. 8.

To enable the achievement of a distribution of struc-tural conditions through space and time, the amount ofarea in each class is assessed each time period. Sincemanagement decisions are made the management unitscale, an assessment of forest structural class condi-tion (Table 1) needs to be made at the managementunit scale. The structural class of each basic simula-tion unit is used to determine the class of the man-agement unit. First, the dominant structural conditionwithin each management unit is assessed.

C∑c=1

Bi∑b=1

abhb,c,t = Si,c,t ∀i (11)

C∑c=1

hb,c,t = 1 ∀b, t (12)

wherec is a forest structural condition class;C is the setof forest structural condition classes;hb,c,t is a binaryvariable indicating whether (1) or not (0) habitat classcdescribes the forest structural condition of basic sim-ulation unitb during time periodt; Si ,c,t is the area of

Table 1Characteristics of forest structural condition classes for public man-a

<b

A

≤> py

>

>

>

>

R

forest structural condition classc within managementunit i during time periodt.

EachSi ,c,t is then examined, and the one that rep-resents the highest proportion of area of managementunit i as assumed to represent the structural conditionof the management unit. A binary (0,1) value is thenassigned to variable SCi ,c,t where

C∑c=1

SCi,c,t = 1 ∀i, t (13)

where SCi ,c,t is a binary (0,1) variable indicatingwhether (1) or not (0) management uniti is assumed torepresent forest structural classc during time periodt.

The percentage of the landscape area in each struc-tural class is then assessed.[∑I

i=1AiSCi,c,t

LSA

]= Hc,t ∀c, t (14)

whereAi is the total area of management uniti; LSAis the area of the landscape under consideration;Hc,t

is the proportion of the landscape in forest structuralclassc during time periodt.

Constraints are then imposed on four of the fivestructural classes using sets of logic. First, since theregeneration structural class condition increases witheach clearcut activity scheduled, some control must beplaced on the maximum amount of the landscape int ationc arcuta

w uc-tt truc-t

oali omm hib-i

gement behavior in the Oregon Coast Range

65% hardwoodasal area

≥65% hardwoodbasal area

ge Class Age Class

15 Regeneration ≤15 Regeneration15 and≤30 Closed canopy >15 Closed cano

30 and≤60RD > 45 Closed canopyRD ≤ 45 Understory

60 and≤80RD > 55 Closed canopyRD ≤ 55 and RD > 40 UnderstoryRD ≤ 40 Layered

80 and≤120Thinned LayeredNot thinned Older forest

120 Older forest

D: relative density (total basal area/quadratic mean diameter0.5).

his condition. The percentage area in the regenerlass is assessed with the simulation of each clectivity.

If Hr,t ≤ SGr thenXi,t ∈ {0, 1}ElseXi,t = 0

(15)

hereHr ,t is the proportion of the landscape in strural classr (regeneration) during time periodt; SGr ishe goal (percentage of the landscape desired for sural classr).

Second, if the older forest structural condition gs not met during a time period, harvest activities fr

anagement units that have this condition are proted.

If Ho,t ≥ SGo thenXi,t ∈ {0, 1} ∀i : SCi,o,t = 1

ElseXi,t = 0 ∀i : SCi,o,t = 1(16)


Fig. 8. Conceptual model of the flow of information across spatial scales, and in association with the objective and constraints, for the state landmodeling process.


whereHo,t is the proportion of the landscape in struc-tural classo (older forest) during time periodt; SGo isthe goal (percentage of the landscape desired for struc-tural classo).

Third, if the layered forest structural condition classgoal is not met during a time period, harvest activitiesfrom management units that have this condition areprohibited.

If Hl,t ≥ SGl thenXi,t ∈ {0, 1} ∀i : SCi,l,t = 1

ElseXi,t = 0 ∀i : SCi,l,t = 1(17)

whereHl ,t is the proportion of the landscape in struc-tural classl (layered forest) during time periodt; SGl isthe goal (percentage of the landscape desired for struc-tural classl).

Fourth, and finally, if the understory forest structuralcondition class goal is not met during a time period,harvest activities from management units that have thiscondition are prohibited.

If Hu,t ≥ SGu thenXi,t ∈ {0, 1} ∀i : SCi,u,t = 1

ElseXi,t = 0 ∀i : SCi,u,t = 1(18)

whereHu,t is the proportion of the landscape in struc-tural classu (understory forest) during time periodt;SGu is the goal (percentage of the landscape desiredfor structural classu).

These last three structural class conditions are att n-a mayd andn ut.

eas,a thati estbu es.R ltings ge-m elowa

i

w ge-m igh-

bors of management uniti plus all units adjacent tothe neighbors of the neighbors, etc. Only managementunits where SCi ,o,t = 1, SCi ,l ,t = 1 and SCi ,u,t =1 are available for inclusion into the habitat block;IHA is the minimum area of the interior habitat blockdesired.

The process utilizes an ARM technique to developeach block. The blocking process stops once the min-imum habitat block size has been exceeded. A URMtechnique is then used to prevent two interior habitatblocks from merging together, forming a larger, singleblock. If management units are included in an interiorhabitat block during time periodt, Xi ,t = 0.

3. Case study

To illustrate how the spatial hierarchical modelingframework might be applied, we provide several ex-amples of modeling both private industrial manage-ment behavior and public management behavior on a560,000 ha area of land in the Oregon Coast Range.The area is composed of a mixture of land owner-ships (Fig. 9), the main groups being private indus-trial and public (state). The private industrial areas aremodeled according the methods presented earlier. Anumber of management scenarios were simulated toprovide an indication of the sensitivity of timber har-vest levels to changes in forest policies (Table 2). TheMHA modeled varies from 35 to 50 years, and them ha.I atei rianp restm f thel im-p withh lev-e asis.

TS in then

M zes

3445

heir maximum levels prior to the simulation of magement activities in each time period, thus theyecrease with each clearcut activity simulated,eed to be reassessed with each simulated clearc

To enable the development of interior habitat arprocess of building habitat blocks is utilized, one

s similar of the process of building clearcut harvlocks. Here, the ARM technique (Murray, 1999) issed to build interior habitat blocks of certain sizecursive functions are used to evaluate the resupatially sprawling interior habitat block of manaent units, and constraints such as the one noted bre used.∑∈ Ni∪Si

Ai ≥ IHA (19)

hereSi is a subset of the total number of manaent units that contains all units adjacent to the ne

aximum clearcut size ranges from 24.3 to 48.6n total, 16 simulations are performed using the privndustrial management behavior process. The ripaolicies gathered through surveys of industrial foanagement behavior indicated that on one-half o

and area within riparian areas, partial cutting islemented to the extent possible under state law,arvests reducing basal area to certain minimumls depending on stream type and riparian emph

able 2cenarios modeled for the private industrial management areasorth coast of coastal Oregon

inimum harvest ages Maximum clearcut si

5 24.3 ha (60 acres)0 32.4 ha (80 acres)5 40.5 ha (100 acres)0 48.6 ha (120 acres)


Fig. 9. Land ownership pattern for a 560,000 ha area of the north coastal region of Oregon (USA).

On the other half of the riparian area, harvests are pre-cluded. In our modeling effort, the location of theseareas is randomly defined at the basic simulation unitscale. After clearcut harvest, management prescrip-tions are assigned to basic simulation units based onthe management intensity assigned to each manage-ment unit. We assume here that 20% of all clearcutmanagement units will be assigned a high managementintensity (plant, pre-commercial thin, fertilize), 60%will be assigned a medium-high management intensity

(plant, pre-commercial thin), and 20% are assigned amedium management intensity (plant). These are hy-pothetical assumptions of future management behav-ior; any arrangement of management intensities can beused. Commercial thinnings are simulated in regener-ated management units if over one-half of each man-agement unit returns as a conifer forest type. Transitionprobabilities (Table 3) are used to determine the typeof forest that returns after clearcutting. These probabil-ities are applied at the basic simulation unit scale, and


Table 3Transition probabilities for forest industry land in the coastal ecore-gion of Oregon, when land is managed under a medium managementintensity

Distance from stream (m)

0–50 51–100 101–150 >151

Previous vegetation class: hardwooda

To open/semi-closed 0.10 0.05 0.04 0.02To predominantly hardwood 0.20 0.15 0.15 0.12To predominantly mixed 0.30 0.27 0.21 0.18To predominantly conifer 0.40 0.53 0.60 0.68

Previous vegetation class: mixedb

To open/semi-closed 0.07 0.05 0.03 0.02To predominantly hardwood 0.15 0.12 0.07 0.05To predominantly mixed 0.28 0.25 0.20 0.13To predominantly conifer 0.50 0.58 0.70 0.80

Previous vegetation class: coniferc

To open/semi-closed 0.05 0.02 0.02 0.01To predominantly hardwood 0.10 0.06 0.06 0.04To predominantly mixed 0.22 0.19 0.13 0.10To predominantly conifer 0.63 0.73 0.79 0.85a Areas where hardwood tree species occupy 65% or more of the

basal area of trees.b Areas where hardwood tree species occupy less than 65% and

greater than or equal to 20% of the basal area of trees.c Areas where hardwood tree species occupy less than 20% of the

basal area of trees.

are developed from empirical analyses associated withthe CLAMS project (Spies et al., 2002). A differentset of probabilities is assumed for each level of man-agement intensity, with higher conversion to coniferassumed with higher management intensities.

The state lands within the north coastal region of theOregon Coast Range are simulated as noted earlier inthe public forest management methods section. Here, anumber of policies are modeled to evaluate the sustain-ability and economic goals on public land. The MHA isassumed to vary from 40 to 55 years, and the structuralcondition classes are assumed to be constant, at 20%for each of the five classes. A number of sets of inte-rior habitat blocks (Table 4) are modeled. Therefore,16 simulations are performed using the public man-agement behavior process. The riparian policies weregathered through examinations of state forest plans.In some land allocations, such as reserved or specialforest management areas, either no activity is allowedin the riparian areas or only thinning activities can beimplemented. In other land allocations, as suggestedby state management plans, five residual leave trees

are retained in clearcuts. After clearcut harvest, man-agement prescriptions are assigned to basic simulationunits based on the management intensity assigned toeach management unit. We assume here that all clearcutmanagement units will be assigned a medium manage-ment intensity (plant). Commercial thinnings are sim-ulated in regenerated stands if over one-half of eachmanagement unit returns as a conifer forest type. Tran-sition probabilities similar to those shown inTable 3areused to determine the type of forest that returns afterclearcutting. These probabilities are applied at the basicsimulation unit scale, and are developed from empiricalanalyses associated with the CLAMS project (Spies etal., 2002).

The hierarchical spatial simulation process for bothownership behavior groups requires about 1 hour ona computer equipped with a 2.4 GHz central process-ing unit and 2 Gb of RAM. The simulation model wasdeveloped with the C programming language, and in-cludes a Visual Basic interface to allow the speci-fication of simulation parameters. Depending on theamount of output data desired, up to 1 Gb can be gen-erated to describe the forest structural conditions of thelandscape over the 100-year time horizon.

4. Results

There is considerable debate about the appropriaterotation age for industrial land in the Oregon CoastR heref iesw hep stec flowv re-s cedw lts

TN blicm

I

112

ange, with various groups suggesting ages anywrom 30 to 60 years. When evaluating forest policith variations in MHAs, it is not surprising that tolicy with the lowest MHA would result in the higheven-flow timber volumes (Fig. 10). As the MHA is in-reased from 35 to 40 years, the maximum even-olume decreases about 26%. The 45-year MHAults in less than one-half of the volume levels produith the 35-year MHA, and the 50-year MHA resu

able 4umber and size of interior habitat blocks modeled for the puanagement areas in the north coast of coastal Oregon

nterior habitat block size Number of habitat blocks

A B C D

–100 ha 0 9 27 8101–200 ha 0 6 18 5401–400 ha 0 3 9 27


Fig. 10. Projected annual harvest levels from a simulation requiring an even-flow timber harvest volume with minimum harvest age and maximumclearcut size constraints on private industrial land in the north coastal region of Oregon (USA).

in about a 70% decrease in harvest levels. This all as-sumes, of course, that as policies change, private indus-trial landowners will continue to operate, as a group,to produce even levels of timber volume. As the max-imum clearcut size decreases from 48.6 ha to 24.3 ha,maximum even-flow timber harvest volume levels de-cline by 4% with the 35-year MHA to about 10% withthe 45-year and 50-year MHAs.

What we find with these results is the complex in-teraction of the objective and the constraints. For themost part, the reduction in timber harvest with the in-crease in MHA is the sequential effect of removing (oradjusting) the constraint. These results, however, maybe an artifact of a simulation process. The even-flowharvest method we have demonstrated, for example,is notorious for becoming encumbered by bottlenecks(periods of low available volume), which heavily influ-ence the resulting solutions. These bottlenecks are timeperiods where the harvest volume does not reach thedesired target, which may be influenced by the desireto harvests levels in previous time periods, the initialdescription (i.e., age class distribution and forest struc-tural conditions) of the landscape, or the constraints.The bottlenecks may occur when the simulation pro-

cess is completing the harvest of current (initial) standsand beginning the harvest of regenerated stands (thoseharvested and regenerated by the simulation process).In some cases, however, they occur earlier, suggestingthat one or more of the constraints may be limiting theplacement of activities across the landscape. More thanlikely the combination of the even-flow objective andthe minimum harvest age constraint limits possibilitiesin the first few time periods, given the resultingaverageharvest ages of clearcut blocks (Fig. 11).

The blocking process described inBettinger andJohnson (2003)combines lower valued managementunits (based on net revenue divided by stand age)around higher valued management units to form har-vest blocks. This is consistent with the direction we re-ceived in discussions with forest industry landowners.Thus, while the MHA has a large effect on even-flowvolumes regardless of maximum clearcut size, the neteffect (4–10% reduction of even-flow harvest levels)of smaller maximum clearcut sizes is the delay of theconversion of low valued management units of the ini-tial landscape to regenerated units, since fewer of thelower-valued units would be combined with higher val-ues units. Smaller harvest blocks thus favor harvesting


Fig. 11. Average harvest ages for the four simulations allowing a 35-year minimum harvest age, and the four simulations allowing a 50-yearminimum harvest age, on private industrial land in the north coastal region of Oregon (USA).

more higher-valued management units early in the sim-ulation process.

The MHA constraint also plays the largest role in thepublic management behavior simulation process. Here,as the MHA is increased from 40 years to 45 years, theeven-flow volume levels decline about 7% (Fig. 12).This assumes, as in the private industrial case, that pub-lic forest managers continue to operate as otherwise as-sumed once the minimum harvest age policy changes.As the MHA is increased to 50 and 55 years, respec-tively, even-flow volume levels decline about 13% and31%.

The influence of increasing levels of interior habitatareas has little effect on the achievement of even-flowharvest volumes, mainly because this constraint is com-plementary to the structural stage constraints: both re-quire the reservation of older forest areas. While olderforest areas are required for the interior habitat blocks,they are allowed to move around the landscape throughtime, thus the initial older forest areas are not specif-ically reserved during all time periods. It is, however,the initial condition of the public forest land that is im-portant, as very little of it is considered older or layeredforests, thus much of is it off-limit for harvest for sometime, until the area in each of these classes exceeds

20% of the public forest land. Thus, a heavy relianceis placed on the clearcut harvest of younger stands inthe closed canopy condition, stands which have agesin the 30–50-year range generally. A separate analysisof marginal differences in the percentages of structuralstages required showed similar small influence on theeven-flow harvest levels, except when the maximumpercentage allowed in the regeneration class was de-creased to 10% or less.

5. Discussion

The hierarchical spatial framework we have de-scribed for modeling large-scale, long-term forest poli-cies allows the recognition of fine-scale spatial detailas well as decisions typical of forest landowners in theOregon Coast Range. The framework is, of course, asimplification and synthesis of a more complex systemof land management. Recognition of current and fu-ture management behavior of all landowners increasesthe credibility and realism of simulations. However,achieving reliability in a simulation model is not atrivial task. For example, ecological consequences candiffer dramatically, depending on the pattern of land


Fig. 12. Annual harvest levels from a simulation of even-flow timber harvest volume with minimum harvest age and interior habitat blockconstraints on public (state) land in the north coastal region of Oregon (USA).

use activities imposed on a landscape (Franklin andForman, 1987). Thus, modeling land use activitiesappropriately is quite important. This may require amajor collaboration between scientists, planners, man-agers, and policy-makers to develop the kind of modelthat has widespread application and acceptance at thespatial and temporal scales at which it is used. Withthe hierarchical spatial framework introduced here,fine-scale forest structural conditions are recognizedwhile achieving management goals measured at coarserscales. The recognition of basic simulation units allowsone to track the development of forest structural con-ditions as they may vary with riparian managementemphasis, leave tree policy, and management intensity.This also facilitates further analysis of biological ef-fects of forest management alternatives at a fine scale.For example, riparian conditions are retained, ratherthan lost as a result of assumptions requiring homo-geneity of conditions within management units.

We want to make clear that neither the even-flowtimber volume objective nor the minimum harvest ageconstraint are required by law for either the industryor public management behavior processes. These wereassumptions we made based, in one case (even-flow),

on guidance from the landowners and evidence fromrecent behavior, and in the other case (MHA), on con-cerns about potential future constraints on managementactivity. In the past, the industry has harvested a fairlyeven level of timber volume in western Oregon becausethey had the ability to utilize federal timber harvests tobuffer changes in timber markets. Federal timber saleshave declined dramatically in the past decade, and thusthe ability of industrial landowners to continue to har-vest a relatively even amount of timber each year in thefuture is uncertain.

Results of the simulations also indicate that a higheramount of timber volume can be simulated for har-vest in the future if the even-flow constraint is relaxed.After the constraining time period (the bottleneckwhere even-flow is constrained) has passed, standinginventory volumes build up because regenerated standgrowth exceeds harvest. Relaxing the even-flow con-straint by allowing variable harvest levels could resultin solutions with higher total harvest levels and aver-age harvest ages that are closer to the minimum harvestage assumed. We are currently experimented with sev-eral approaches in this area: (1) if a target harvest isnot obtained in a time period, reduce the target for that


time period only; (2) rather than an even-flow goal, usebinary search to obtain a target harvest level pattern setby the user, where the harvest levels can vary from onetime period to the next, and the entire pattern is shiftedupward or downward depending on whether the targetsare met.

An infinite number of scenarios can be modeled withthe spatial framework, given the continuous nature ofsome of the constraint values (e.g., MHA, maximumclearcut size), and variations in the transition proba-bilities. Our next goal is to formulate reasonable as-sumptions into a “base case” scenario that will be usedto compare against realistic alternatives. The results ofthe base case, specifically timber volume levels andspatial pattern of activities, will be compared againsthistorical data. We plan to model a number of signifi-cant alternative scenarios for the entire Oregon CoastRange to help inform stakeholders and policy makers.These scenarios include additional variations on man-agement intensities, minimum harvest ages, riparianrestrictions, and tree retention policies.

One of the main limitations of this modeling ef-fort relates to the validation of the simulation pro-cess, which is inherently problematic. While projectedharvest levels and spatial patterns of activities can becompared against recent activity, forest landscape plan-ning does not lend itself well to validation processes.In forest landscape planning, we project into the fu-ture a representation of the current condition of thelandscape. When projecting scenarios into the future,a t onf ibu-t ationg tys ot bec scapec e tot andl oft ingh forl

estl y al-l ratep ple’sp ink-i to

compare alternative policies in light of other uncontrol-lable aspects of the future. Scenarios provide a range ofpossibilities, and integrate science with policy, ratherthan waiting for results of further research (Carpenter,2002). Individual scenarios are therefore not subject torigorous statistical validation. Rather, they are testedfor robustness against a set of other alternative sce-narios, and allow policy makers and managers tounderstand possible futures, and how they might beinfluenced by past or present management decisions.

As we have previously noted, with many modelingefforts that address complex behavioral systems andbroad mixed-ownership landscapes, there is consider-able room for improvement of the simulation process.The public forest management behavior process, forexample is quite complex. Through a thorough anal-ysis of the influence of the various constraints on theachievement of management goals, alternative model-ing processes, including both the processes to scheduleactivities and to evaluate constraints, may be devel-oped. In addition, while it appears that the minimumharvest age constraint has the largest impact in boththe public and private modeling processes, a furtherexploration of the interaction among goals and con-straints will improve our understanding of the results,and facilitate better communication of the results tostakeholders and policy makers.

Finally, one of the main limitations of the approachdescribed here is that only the behavior of two of thefour major landowner groups in the region have beend oft s heret hilet ndus-t n-e f theO inarym in ane rs.H stateo av-i g oft theb nersi essa thea eset

number of real system dynamics are contingenactors that may have unknown or uncertain distrions, such as climate change and human populrowth (Carpenter, 2002). Therefore, the uncertainurrounding the projections presented here cannomputed. Some have suggested developing landonditions of 20–50 years ago, then projecting thoshe present to determine whether current patternsevels of activity can be simulated well. The cruxhe problem with this type of analysis is in developistorical databases containing the detail required

arge areas such as those modeled here.Evaluating alternative future scenarios with for

andscape planning models has value, however, bowing one to think through decisions when accuredictions are not possible, by broadening peoerspectives, and by challenging conventional th

ng (Carpenter, 2002). Scenarios make it possible

escribed. We feel confident in our interpretationheir behavior, and have thus presented processeo modeling it in a spatial and temporal manner. Whe area we analyzed was dominated by state and irial ownership, federal and non-industrial landowrs also have a large presence in other portions oregon Coast Range. We have developed prelimodeling processes for these ownerships as well,ffort to fully model the behavior of all landowneowever, federal management policies are in af flux, thus our understanding of their current beh

or continues to be refined (as does the modelinheir behavior). In addition, our understanding ofehavior of the thousands of non-industrial landow

s limited, and currently we use a Monte Carlo proclong with gross probabilities of harvest to projectctivities of this landowner. We continue to refine th

wo processes.


6. Conclusions

The hierarchical spatial framework for forestlandscape planning presented here allows one toexamine alternative forest management policies acrosslarge land areas while recognizing fine-scale foreststructural conditions as well as decisions typical offorest landowners, which operate at larger scales.This framework allows a more reasonable portrayal ofmanagement behavior at the fine-scale while also facil-itating broad-scale analyses of forest policies. Sustain-ability policies encourage maintaining the capacity ofa landscape to provide a wide range of values, services,and products desired by society. Landscape planningmodels that integrate ecological and socio-economicconcerns, while adequately modeling the behavior oflandowners, can help policy-makers, managers, andother stakeholders think through the implications ofpotential sustainability policies from both economicand ecological perspectives. Our simulations, for ex-ample, indicate that a minimum harvest age constrainthas a stronger influence on even-flow harvest levelsthan do maximum clearcut size or interior habitat areaconstraints. Even-flow timber harvest level objectives,however, also have an effect on the results, suggestingthat a possible relaxation or modification of theobjective may be necessary to achieve average harvestages that are closer to the desired minimum harvestage.

Modeling alternative landscape policies may re-q les,t andh patialf Ther arer cale,a onalp ) ist e pic-t es.

A

or-e DAF Sta-t the

College of Forestry at Oregon State University. TheCLAMS (Coastal Landscape Analysis and ModelingStudy) project (http://www.fsl.orst.edu/clams/) pro-vided databases, knowledge, and support for modelinglandscape management policy alternatives.

References

Bettinger, P., Johnson, D.L., Johnson, K.N., 2003. Spatial forest plandevelopment with ecological and economic goals. Ecol. Model.169, 215–236.

Bettinger, P., Johnson, K.N., 2003. Spatial scheduling of forest man-agement activities using a dynamic deterministic harvest blockaggregation process. J. Forest Plan. 9, 29–38.

Bettinger, P., Lennette, M., 2003. LAndscape Management PolicySimulator (LAMPS), version 1.1, user’s guide. Department ofForest Resources, Oregon State University, Corvallis, OR.

Birta, L.G.,Ozmizrak, F.N., 1996. A knowledge-based approach forthe validation of simulation models: the foundation. ACM Trans.Model. Computer Simul. 6, 76–98.

Bragg, D.C., Roberts, D.W., Crow, T.R., 2004. A hierarchical ap-proach for simulating northern forest dynamics. Ecol. Model.173, 31–94.

Burnett, C., Blaschke, T., 2003. A multi-scale segmentation/objectrelationship modelling methodology for landscape analysis.Ecol. Model. 168, 233–249.

Carpenter, S.R., 2002. Ecological futures: building an ecology of thelong now. Ecology 83, 2069–2083.

Church, R.L., Murray, A.T., Figueroa, M.A., Barber, K.H., 2000.Support system development for forest ecosystem management.Eur. J. Operat. Res. 121, 247–258.

Flamm, R.O., Turner, M.G., 1994. Alternative model formulationsol. 9,

F ns byndsc.

F strialPA.

G elop-glas-

orth-eport

G restion of

G He,mber

H er’sState

uire recognizing and utilizing multiple spatial scahus closely emulating both natural processesuman behavior. We have demonstrated here a s

ramework to facilitate these types of analyses.esulting forest structural conditions of each policyecognized, maintained, and reported at a fine sccommodating further analyses. The set of additiotential analyses (e.g., wildlife habitat suitability

herefore broadened, encouraging a more completure of the economic-ecological trade-offs of polici

cknowledgements

Funding for the development of the spatial fst landscape planning model came from the USorest Service – Pacific Northwest Research

ion, the Oregon Department of Forestry, and

for a stochastic simulation of landscape change. Landsc. Ec37–46.

ranklin, J.F., Forman, R.T.T., 1987. Creating landscape patterforest cutting: ecological consequences and principles. LaEcol. 1, 5–18.

reemark, K., 1995. Assessing effects of agriculture on terrewildlife: developing a hierarchical approach for the US ELandsc. Urban Plan. 31, 99–115.

arman, S.L., Cissel, J.H., Mayo, J.H., 2003. Accelerating devment of late-successional conditions in young managed Doufir stands: a simulation study. USDA Forest Service Pacific Nwest Research Station, Portland, OR. General Technical RPNW-GTR-557.

ustafson, E.J., Crow, T.R., 1994. Modeling the effects of foharvesting on landscape structure and the spatial distributcowbird brood parasitism. Landsc. Ecol. 9, 237–248.

ustafson, E.J., Shifley, S.R., Mladenoff, D.J., Nimerfro, K.K.,H.S., 2000. Spatial simulation of forest succession and tiharvesting using LANDIS. Can. J. Forest Res. 30, 32–43.

ann, D.W., Hester, A.S., Olsen, C.L., 1997. ORGANON usmanual: edition 6.0. Forest Resources Department, OregonUniversity, Corvallis.

http://www.fsl.orst.edu/clams/


Harris, J.R.W., Gorley, R.N., 2003. ECoS, a framework for mod-elling hierarchical spatial systems. Sci. Total Environ. 314–316,625–635.

Johnson, K.N., Scheurman, H.L., 1977. Techniques for prescribingoptimal harvest and investment under different objectives – dis-cussion and synthesis. For. Sci. Monogr. 18, 31.

Johnson, K.N., Sessions, J., Franklin, J., Gabriel, J., 1998. Integrat-ing wildfire into strategic planning for Sierra Nevada forests. J.Forest. 96 (1), 42–49.

Jørgensen, S.E., 1997. Ecological modelling by ‘ecological mod-elling’. Ecol. Model. 100, 5–10.

Lettman, G., Campbell, D., 1997. Timber harvesting practices onprivate forest land in western Oregon. Oregon Department ofForestry, Salem, OR.

Leuschner, W.A., 1990. Forest regulation, harvest scheduling, andplanning techniques. John Wiley & Sons, New York.

Li, H., Franklin, J.F., Swanson, F.J., Spies, T.A., 1993. Developing al-ternative forest cutting patterns: a simulation approach. . Landsc.Ecol. 8, 63–75.

MacMillan, R.A., Jones, R.K., McNabb, D.H., 2004. Defining a hier-archy of spatial entities for environmental analysis and modelingusing digital elevation models (DEMs). Comput. Environ. UrbanSyst. 28, 175–200.

Murray, A.T., 1999. Spatial restrictions in harvest scheduling. For.Sci. 45, 45–52.

Nelson, J.D., 2001. Assessment of harvest blocks generatedfrom operational polygons and forest-cover polygons in tac-tical and strategic planning. Can. J. For. Res. 31, 682–693.

Ohmann, J.L., Gregory, M.J., 2002. Predictive mapping of forestcomposition and structure with direct gradient analysis and near-est neighbor imputation in coastal Oregon, USA. Can. J. For. Res.32, 725–741.

Oregon Board of Forestry and Oregon Department of Forestry, 1995.Forestry Program for Oregon. Oregon Department of Forestry,Salem, OR.

Oregon State Legislature, 2001. Oregon revised statutes. Chapter527 – Insect and disease control; Forest practices. Oregon StateLegislature, Salem, OR.http://www.leg.state.or.us/ors/527.html(accessed 01/29/03).

Spies, T.A., Johnson, K.N., Bettinger, P., McComb, W.C., Ohmann,J.L., Reeves, G.H., 2002. Challenges in integrating wood pro-duction, biodiversity, and other socioeconomic values at broadscales: an example from Coastal Oregon. In: Johnson, A.,Haynes, R.W., Monserud, R.A. (Eds.), Congruent managementof multiple resources. Proceedings from the Wood Compatibil-ity Initiative Workshop, USDA Forest Service, Pacific NorthwestResearch Station, Portland, OR. General Technical Report PNW-GTR-563, pp. 13–25.

Urban, D.L., Shugart, H.H., 1992. Individual based models of for-est succession. In: Glenn-Lewin, D.C. (Ed.), Plant succession:theory and prediction. Chapman and Hall, London, pp. 249–292.

Wallin, D.O., Swanson, F.J., Marks, B., 1994. Landscape patternresponse to changes in pattern generation rules: land-use legaciesin forestry. Ecol. Appl. 4, 569–580.

Wu, J., David, J.L., 2002. A spatially explicit hierarchical approach tomodeling complex ecological systems: theory and applications.Ecol. Model. 153, 7–26.

http://www.leg.state.or.us/ors/527.html

a hierarchical spatial framework for forest landscape planning

Documents