LBMF_KennedyFlatsFinalReport2002

50
Assessing Landscape Connectivity in the Kennedy Flats Final Report Andrew Fall, Gowlland Technologies Ltd. Barb Beasley, Long Beach Model Forest May, 2002 Acknowledgments We wish to acknowledge the support of the Habitat Conservation Trust Fund of British Columbia and the Long Beach Model Forest (LBMF). We thank Mike Collyer, LBMF, David White, International Forest Products Ltd., and Adrian Walton, Ministry of Forests Research Branch for data preparation, Laurie Kremsater for helpful comments on this document, and the support of people and organizations that participated in this project and the workshop. We would also like to acknowledge prior support from Micheline Manseau and Parks Canada for development of some of the connectivity methods. Executive Summary The Kennedy Flats Landscape Model was developed to assess the connectivity between late seral forest patches and between wetlands in the Kennedy Flats area of west central Vancouver Island. There are two main components of the model (i) a dynamic landscape model to capture strategic aspects of management to project plausible future forest conditions; and (ii) a set of landscape pattern analysis methods to assess connectivity of habitat patches. The management sub-model was designed to capture the current regime under Interfor’s watershed plan (2001), which implements recommendations from the Clayoquot Sound Scientific Panel (1995), and likely management under a pre-scientific panel regime based on the Forest Practices Code. We used this model to estimate landscape conditions under the two management plans in 50 and 100 years. We applied graph-based methods to examine the structural configuration of habitat patches. Connections between patches can be defined using either straight-line Euclidean distance or a cost surface representing relative barriers to movement under different cover types. We developed movement cost surfaces between wetlands based on general movement data for red-legged frogs, and between old forest patches (>140 years) based on hypothetical movement costs of an old forest dependent organism. We then assessed connectivity of old forests and of wetlands under current and projected conditions. We found that scales of 500m-1100m currently dominate connectivity of old forest across the Flats, while wetland connectivity is strongly influenced from fine scales up to 10km. Reductions in the critical scales under projected conditions for Interfor’s watershed plan indicate a trend of improvement of old forest connectivity. Connectivity of wetlands doesn’t seem to change drastically over time since reductions in dispersal impediment in the intervening forest cover are offset by road development.

description

http://www.modelforest.net/media/k2/attachments/LBMF_KennedyFlatsFinalReport2002.pdf

Transcript of LBMF_KennedyFlatsFinalReport2002

Page 1: LBMF_KennedyFlatsFinalReport2002

Assessing Landscape Connectivity in theKennedy Flats

Final Report

Andrew Fall, Gowlland Technologies Ltd.Barb Beasley, Long Beach Model Forest

May, 2002

AcknowledgmentsWe wish to acknowledge the support of the Habitat Conservation Trust Fund of BritishColumbia and the Long Beach Model Forest (LBMF). We thank Mike Collyer, LBMF,David White, International Forest Products Ltd., and Adrian Walton, Ministry of ForestsResearch Branch for data preparation, Laurie Kremsater for helpful comments on thisdocument, and the support of people and organizations that participated in this projectand the workshop. We would also like to acknowledge prior support from MichelineManseau and Parks Canada for development of some of the connectivity methods.

Executive SummaryThe Kennedy Flats Landscape Model was developed to assess the connectivity betweenlate seral forest patches and between wetlands in the Kennedy Flats area of west centralVancouver Island. There are two main components of the model (i) a dynamic landscapemodel to capture strategic aspects of management to project plausible future forestconditions; and (ii) a set of landscape pattern analysis methods to assess connectivity ofhabitat patches. The management sub-model was designed to capture the current regimeunder Interfor’s watershed plan (2001), which implements recommendations from theClayoquot Sound Scientific Panel (1995), and likely management under a pre-scientificpanel regime based on the Forest Practices Code. We used this model to estimatelandscape conditions under the two management plans in 50 and 100 years.

We applied graph-based methods to examine the structural configuration of habitatpatches. Connections between patches can be defined using either straight-line Euclideandistance or a cost surface representing relative barriers to movement under different covertypes. We developed movement cost surfaces between wetlands based on generalmovement data for red-legged frogs, and between old forest patches (>140 years) basedon hypothetical movement costs of an old forest dependent organism. We then assessedconnectivity of old forests and of wetlands under current and projected conditions.

We found that scales of 500m-1100m currently dominate connectivity of old forest acrossthe Flats, while wetland connectivity is strongly influenced from fine scales up to 10km.Reductions in the critical scales under projected conditions for Interfor’s watershed planindicate a trend of improvement of old forest connectivity. Connectivity of wetlandsdoesn’t seem to change drastically over time since reductions in dispersal impediment inthe intervening forest cover are offset by road development.

Page 2: LBMF_KennedyFlatsFinalReport2002

2

Table of ContentsAcknowledgments 1

EXECUTIVE SUMMARY 1

INTRODUCTION 3

METHODS: LANDSCAPE DYNAMICS MODEL 5

I. Overall Landscape Model Design 5Model State Space 6Stand Aging 8Movement Cost Surfaces 8Harvesting 11

II. Model Outputs 14Forest State Indicators 14Harvest Indicators 14Spatial output 15

III. Scenarios Evaluated 15

METHODS: CONNECTIVITY MODELS 16

I. Overall Connectivity Model Design 16

II. Analysis Methods 18

RESULTS AND DISCUSSION 20

I. Landscape Projection 20

II. Old Forest Connectivity 22Current Conditions 22Projected Conditions 32

III. Wetland Connectivity 37Current Conditions 37Projected Conditions 44

CONCLUSIONS 46

REFERENCES 49

Page 3: LBMF_KennedyFlatsFinalReport2002

3

IntroductionThe Kennedy Flats area of west-central Vancouver Island (theFlats), British Columbia,has a varied history of management, and high ecological, recreational, educational andforestry values. In 1995 the Clayoquot Sound Scientific Panel (CSSP) released a reportwith recommendations based on ecosystem management principles. The implementationdetails of the recommendations are still being developed by the two main licensees in theflats, International Forest Products Ltd. (Interfor) and Iisaak Forest Products Ltd. (Iisaak).Interfor has been developing a comprehensive watershed plan to capture the strategicrecommendations of the CSSP for specific application in the Kennedy Flats.Understanding how the changing management will achieve desired values and objectivesover time is critical for effective landscape management.

The Montreal Process developed a set of guidelines for developing criteria and indicators(C&I) for sustainable forest management (Working Group on Criteria and Indicators forthe Conservation and Sustainable Management of Temperate and Boreal Forests, 2000).The Long Beach Model Forest (LBMF) has been developing a set of Local LevelIndicators (Wright, 1999) that are appropriate for areas in and around Clayoquot Sound.A key component of C&I is monitoring to assess how values of indicators change overtime. However, due the costs of long-term monitoring, it is important to carefully selectspecific indicators and monitoring sites.

Habitat connectivity can be a critical component of maintaining viable populations ofdispersal limited species (Fahrig and Merriam 1985) and hence is a good indicator ofsustainable management. Assessing connectivity of habitat for such species at thelandscape scale is critical for effective landscape management in provincial forests andfor determining ecological integrity of protected areas (Taylor et al. 1993, With 1999). InKennedy Flats, late seral forest patches and wetlands are two types of habitat for whichconnectivity has likely been impacted by past management, and for which connectivitywill likely change under future management.

The main purpose of this study was to explore connectivity of late seral forest andwetland habitat in current conditions and in possible future forest conditions in theKennedy Flats study area. This analysis could be useful both as an assessment of theimpacts and benefits of the current management regime and as a method to assist inidentifying key monitoring sites that either have specific connectivity attributes (e.g.well-connected or poorly-connected sites) or that have certain likelihood of impact fromfuture management (e.g. low or high likelihood of change due to management). Thestudy area consists of the Kennedy Lake watershed unit and adjoining portions of theBeach and Fortune Channel units, an area of approximately 41,000ha of which about31,000ha is productive forest and 21,000ha is classified as timber harvesting landbase.We chose to represent the area using a resolution of 20m by 20m cells.

Specifically, the goal of this project was to address three questions:(i) What is the likely configuration of stand ages and roads in 50 and 100 years

under current management, and under the pre-scientific panel regime?

Page 4: LBMF_KennedyFlatsFinalReport2002

4

(ii) How are late seral forest and wetland habitats currently connected in theKennedy Flats?

(iii) What are the likely changes on connectivity of late seral forest and wetlandhabitats over the next 50 and 100 years under current management and underpre-scientific panel regime?

To achieve the project objectives, we needed to develop and apply modelling methods toevaluate landscape-scale connectivity under a range of conditions. The Kennedy FlatsLandscape Model (KFLM) is a set of spatio-temporal models we developed for thispurpose. All models were implemented using the SELES landscape modelling tool (Falland Fall, 2001), and Excel spreadsheets were used for some of the intermediate and finalanalysis. The KFLM has two main components:(i) We developed a landscape dynamics model to project future forest conditions

under Interfor’s watershed plan and under the likely regime that would have beenin place without the CSSP recommendations. The rules, constraints and strategiesof these two regimes were developed as part of a workshop with invitedparticipants from LBMF, Ministry of Forests, Interfor, Iisaak, Canadian ForestService, First Nations, and others (Beasley and Fall, 2002). The goal of thiscomponent was not to produce an operational model, but rather to capture theessential features of the management regimes in order to project plausibleconditions for stand age and roads at years 50 and 100. Since the model isstochastic, each regime was run 10 times to create a distribution of future forestconditions.

(ii) We refined and adapted a set of pattern analysis models to assess habitatconnectivity. The inputs to this analysis are two spatial layers representing habitatpatches and relative movement cost in the intervening matrix. We applied twotypes of cost measures.Straight-line Euclidean costdefines cost simply bydistance between patches regardless of the intervening cover type. This costmeasure is useful as a generic baseline, providing information on structuralconfiguration of habitat (Keitt et al. 1997). In this case every cell in the costsurface has a relative value of one.Variable costdefines cost as a function ofspatial variables, such as stand age and cover type (e.g. roads, water, forest).Developing a variable cost surface should be done for a selected species or guildof species. For the late seral forest, we chose the guild of species dependent onlate seral forests for which younger forests pose barriers to dispersal. For thewetlands, we chose to focus on red-legged frogs (Rana aurora aurora) anddeveloped a cost surface based on general movement data.

The core of the connectivity methods described in this document were originallydeveloped with support from Parks Canada for prior projects in NorthernColumbia Mountains, British Columbia and in the boreal forest of centralManitoba (Fall, A. 2001a, Fall, A. 2001b). These methods have in turn evolvedfrom the techniques described in Keitt et al. (1997), which help to identify thescales at which habitat patches become connected, as well as the relativeimportance of individual patches at various scales. Incorporating the costmeasures also allows identification of corridors.

Page 5: LBMF_KennedyFlatsFinalReport2002

5

One of the challenges for the applications of these methods in this project is todeal with the large number of conditions. Previously, the methods have beenapplied to only one or two landscape maps in an analysis. In this project, we havetwo habitat types (wetlands and old forest), two cost measures each (straight-lineand variable cost) and three time periods (current, 50 years, 100 years), where thetwo future time periods are each projected under two management regimes(watershed plan, pre-CSSP regime) for 10 replicates. This makes a total of 164landscape conditions to assess.

One objective of the KFLM is to assess connectivity of wetlands and of old forest in theKennedy Flats region using an adaptation of a graph-based analysis approach developedby Keitt et al. (1997). This approach has previously been applied to assess connectivity ofSpotted Owl habitat in New Mexico (Keitt et al. 1997), bottomland forest in NorthCarolina (Bunn et al. 2000), and woodland caribou habitat in B.C. and Manitoba (Fall2001a, 2001b). In the latter, we have developed methods to incorporate a movement costsurface in the analysis to include the degrees to which different landscape elementsimpede or facilitate movement, which we implemented using SELES (Fall and Fall,2001). We contrast the results of applying a cost surface under various conditions(current, and possible future) with the assumption that the only barrier to dispersal isdistance (i.e. where the cost of movement between two habitat patches is simply theEuclidean distance between the patches).

In this report, we describe separately the methods applied to the two main components ofthe KFLM. We then present and discuss the results of the analyses. We conclude with asummary of the main findings and recommendations for monitoring.

Methods: Landscape Dynamics Model

I. Overall Landscape Model DesignThe following describes the design for the landscape dynamics component of theKennedy Flats Landscape Model. This covers a high level description of the model,including the state space and process sub-models.

The general design of KFLM in terms of linkages between model state, landscapeprocesses and output files is shown in Figure 1.

Page 6: LBMF_KennedyFlatsFinalReport2002

6

Figure 1. Linkages between primary components of state (shown in the centre), modelprocesses (shown in ovals) and output files (shown as grey drums).

Model State SpaceAll layers except where noted below, were derived from information from the currentforest inventory on the Flats, in particular vegetation resources inventory (VRI) andTRIM data.

Landscape structure: the landscape biogeographical context and the limits of the studyarea are defined with the following spatial variables:

(i) BEC: biogeoeclimatic classification by variant.(ii) Elevation: elevation in metres(iii) Slope: gradient in percent(iv) LandCover: dominant land cover type

Forest State: the forest is represented by the following layers:

(v) Forest: forested cells(vi) ProductiveForest: cells with productive forest

Age MapsOld Forest

Cost Surfaces

Aging andSuccession

ForestState

HarvestAvailability

THLB

RoadState

Harvest

ManagementRegime

Harvest IndicatorsBlock Size Dist.

LandscapeBiogeography

Page 7: LBMF_KennedyFlatsFinalReport2002

7

(vii) StandAge: age in years (for leading species).(viii) Species: primary leading species

Harvest Availability and History:

(ix) Reserved: During stand aging, the management constraints are assessed,and forest is reserved or recruited as necessary. Remaining forest cellsolder than the minimum harvest age are flagged as available.

(x) BlockType: the type of treatment is recorded so that appropriate re-entriescan be scheduled (e.g. 2-pass high retention old forest treatments).

(xi) TimeSinceLogging: time since last treatment to ensure minimum re-entryintervals are met.

Timber harvesting landbase:

(xii) THLB: the timber harvesting landbase is derived from the productiveoperable forests, ownership information (crown forest) and slope. Slopes70-100% were classified as heli-logging zones and slopes > 100% wereexcluded.

Management Zones: Management zones common to both regimes are:

(xiii) Watershed units: watershed planning units form the basis for applicationof constraints.

(xiv) Riparian zone: estimated using buffers on TRIM streams, lakes andwetlands.

Management zones used by the watershed plan scenario are:

(xv) BioCorridors: biodiversity corridors.

Management zones used by the FPC scenario are:(xvi) Riparian reserve zone: riparian 20m buffers to model areas excluded from

harvest(xvii) Riparian management zone: 100m buffers to model areas with

management restrictions(xviii) VQO: visual quality objective zones

Management Parameters: A range of parameters and tables to set up the harvestingregime, including:

(xix) AAC: annual allowable cut specified as a target area/year.(xx) minimum harvest age: set at 40 years to allow harvest within second

growth stands.(xxi) management constraints(xxii) management preferences.

Roads:

(xxiii) RoadType: type of road (e.g. primary gravel)(xxiv) DistanceToRoads: distance to existing roads in metres.

Page 8: LBMF_KennedyFlatsFinalReport2002

8

Streams and wetlands:

(xxv) StreamClass: stream type classification(xxvi) Wetlands: at the workshop, we decided that the TRIM wetlands

underestimated the number of wetlands on the flats. We chose to add tothe TRIM wetlands an estimate of wetlands based on the digital elevationmodel, Using a hydrological flow model, wetlands were estimated as lowlying areas that were not classified as lakes in the VRI data.

Stand AgingThis event increments stand age each timestep, and updates the age class and seral stageinformation. Old forest is tracked separately within each watershed unit/BEC zonecombination, while young forest is tracked within each watershed unit/VQO zonecombination.

This event determines stand availability and reservation/recruitment. For constraints forwhich recruitment is required (e.g. min. old requirements), cells are recruited in order ofage from the productive forest landbase. The model also tracked the amount of forestlocked up (i.e. unavailable within the THLB) and reserved (i.e. applied towards constraintwithin productive forest) within each watershed unit/BEC zone, as well as the amount ofold forest stratified by watershed unit, BEC zone and THLB and the amount of old forestavailable for harvest.

The only species shifts captured in the model is for regeneration of Douglas-fir forests onthe flats, which are assumed to be planted as cedar. This event also updates the estimateof old forest (defined as stands aged > 140 years) and the old forest and wetland costsurfaces.

Movement Cost SurfacesThere is mounting evidence that dispersal is facilitated or impeded by the configurationof habitat patches and intervening matrix cover types as well as linear features such asroads and railway lines (Bélisle and St. Clair. 2001, Dale et al. 1994). Although in theFlats we do not have direct field data to parameterize a movement cost surface, wedeveloped cost surfaces based on data from elsewhere and general assumptions about thedispersal capabilities of some old forest and wetland dependent species. One advantageof this approach to connectivity is that it is an additive framework, and so analysis usingthe preliminary cost measure we have applies can prioritize further data collection toimprove the graph model (Bunn et al. 2000), and refined results would highlight thesensitivity to changes in the cost surface.

Our goal is to produce cost surfaces in which the value in each 20m by 20m cellrepresents the relative difficulty of movement for dispersing animals across the landscapeconditions of that cell. As the cost surface is non-directional, we are concerned with thedifficulty of movement across each cell independent of its neighbours. That is, unlikeindividual based models, our methods here do not presently include the conditional costof moving between cells of various types. The graph-based analysis described belowaccounts for the cumulative cost of moving between patches.

Page 9: LBMF_KennedyFlatsFinalReport2002

9

The units of the cost surface are values relative to the cost of moving through one cell ofideal habitat. Thus, our basis is a minimum cost of 1 for habitat patches (old forest orwetlands). Alternatively, one may view a cost value as the number of cells of leastresistance that an individual might decide to traverse in order to avoid crossing a singlecell of that type. For example, if we set the movement cost surface for crossing a clearcutto 8, then we are stating that it is as hard for an individual to cross a single clearcut cell asit is to move through 8 wetland cells.

Late Seral Forest Cost Surface

The old forest cost surface was defined for a guild of idealized species that move fine inforest older than 140 years, but have increasing difficulty as stand age declines. The costfunction is defined as 10 up to age 20, and then declines linearly until 1 at age 140 andolder. Note that since cost is a relative function, the minimum value is 1. The costfunction is shown in (Figure 2). At the workshop we had discussed various options forthe cost surface, including one where cost dropped immediately from 10 to 1 at age 20.We decided to apply a cost that gradually declines with age to avoid discontinuous jumpsin costs between stands that differ by one or a few years.

Figure 2. Old forest cost function, which assumes that cost increases for younger stands.

Wetland Cost Surface

For wetland connectivity, we selected the red-legged frog as a focal species to provideguidance to deriving a cost surface. However, our goal is to produce fairly generalresults, and hence we want to keep the cost function relatively simple. Thus it is moreappropriate to consider the inter-wetland movement cost surface as applying to a guild ofdispersal limited amphibian species. We derived the cost surface from an individual-based model developed for red-legged frogs and applied in the Eve River watershed onnortheastern Vancouver Island (Fall and Sutherland, 2000), constructed collaborativelywith Ann Chan-McLeod. A graph-based analysis of red-legged frog habitat iscomplementary to an individual-based approach. The former is more structure-focused,while the latter allows more details of movement and is more functional-focused.

0

2

4

6

8

10

12

0 50 100 150 200

Stand Age

Mov

emen

tCos

t(E

ffect

ive

Dis

tanc

e,km

)

Page 10: LBMF_KennedyFlatsFinalReport2002

10

The cost function is based on land cover type (forest, riparian, water, road, etc.) andslope. It is related to the inverse of the relative probability of selecting a potential cell inthe individual-based model (i.e. where a low probability of selecting a cell represents arelatively high cost).

We compute the cost function as follows:(i) CoverCost: Since the model uses a fairly fine resolution (20m by 20m cells), we

can reasonably classify each cell into one of the following classes, each with anassociated cost. The increasing costs for roads are based on both increasing widthand increasing traffic.

Cover Type Cover Type CostWetland 1

Old forest (> 140 years) 2Immature forest (80-140 years) 4

Pole forest (20-80 years) 8Young forest (0-20 years) 8

Riparian zone 50% of the cost associated with forest ageShrubs (low cover) 8

Natural non-vegetated (e.g. bedrock, rubble) 8Logging branch road 20

Single lane gravel road 40Secondary gravel road 60Primary gravel road 80

Paved double land road 100Buildings, parking, 100

Fresh water (lake, reservoir, rivers) 20Water (ocean) Infinite (i.e. no connectivity)

SlopeCost: is an increasing function of slope, where slope has no influence up togradients of 10% (i.e. a cost value of 1), and then cost increases linearly to a maximum of100 at gradients at and above 90% (Figure 3).

Figure 3. Variable slope cost function for wetland cost surface, which assumes that costincreases with steeper slopes.

0

20

40

60

80

100

120

0 50 100 150 200

Slope (%)

Mov

emen

tCos

t(E

ffect

ive

Dis

tanc

e,km

)

Page 11: LBMF_KennedyFlatsFinalReport2002

11

We didn’t model interactions between cover type and slope (e.g. if a steep road has ahigher cost than either the steepness factor or the road factor).

We compute the final cost as follows, which ranges from 1 to 100:RLFMovementCost = MAX(CoverCost, SlopeCost)

HarvestingThis event performs forest harvesting in available cells. In general, this sub-modelsimulates the allocation of cutblocks (disturbance patches) as they initiate and spreadacross the landscape. The model is driven by an area-based harvest target (ha/yr). Foreach block, a target block size is selected from an input distribution. Cutblocks had tofall on eligible land (determined by the timber-harvesting land base, stand age, access,forest cover rules, and adjacency rules); location also reflected the economic andenvironmental differences (including distance to road, harvestable area in zone, and standage) among eligible stands (table 2). A retention level is also selected for each block,indicating an expected level of randomly selected cells within the block to be left uncut.

Table 2. Steps used to choose cutblocks in the logging sub-model.

Step Description1 Limit harvesting disturbance to eligible land:

• the timber harvesting landbase;• the FPC scenario also excluded areas within riparian zones (20m buffers along rivers and

streams) and steep slopes (> 70% gradient).• eligible zones (not reserved to satisfy forest cover constraints);• for cells that have had a first pass partial cut: minimum re-entry time (25 years) has passed;• areas within 2.5 km of an existing road;• stands older than minimum harvest age (40 years); only old forest (> 140 years) is cut in the

heli-zone2 Select cells in which to initiate treatments with preference based on

• stand age: preference increases linearly with age;• distance to road: no effect up to 200m, and then preference declines linearly to 0 at 2.5km;• harvest history: stands available for 2nd and later passes have an elevated preference;• in the watershed plan (WP) scenario, old forest is not harvested in the biodiversity corridors

and second growth entries have a reduced preference4 For each new cutblock location (first map cell to harvest in block ):

• select a block type;• select a retention level based on block type;• for first entry blocks, pick the size of the cutblock from a block size distribution;• build a spur road from the cell to the nearest existing road cell;• treat the cell – cut with a probability based on retention level;• update tracking variables (e.g. annual vol. harvested; seral distribution in management zones);

5 Continue harvesting cutblock by spreading to adjacent cells based on eligibility until selectedcutblock size is reached (for first entry), until the entire patch is treated (for second and laterentries) or until no more eligible, adjacent cells exist:• update distance to road information• treat the cell (as above)

Page 12: LBMF_KennedyFlatsFinalReport2002

12

Once a harvest block initiates, it attempts to log adjacent cells until a chosen block sizewas reached or until the adjacent eligible area was exhausted (in which case, a smallerblock was created). During harvest, stand availability according to the cover constraintsand adjacency is updated to reflect changing conditions caused by harvest.

The logging sub-model explicitly connects cutblocks to the main road network by addinga link from the first cell harvested in the block to the nearest existing road. It then updatesa map that stored the distance from each cell to the nearest existing road. This featurepermits estimation of the amount of road constructed under a given management regime.

The model was designed to support two main types of scenarios:(i) WP: based on Interfor’s watershed plan under the Scientific Panel’s

Recommendations(ii) FPC: based on basic Forest Practices Code policy

The only fundamental differences (i.e. internal to the model) between these scenarios isthe forest cover rules applied and block types (and hence levels of retention).

Forest Cover Rules: The WP scenario applied an old forest requirement of at least 40%older than 140 years of productive forest in each watershed unit. The FPC scenarioapplies the following seral requirements:

(i) Old forest requirements from the biodiversity guidebook must be met on theproductive forest within each watershed unit and BEC zone:

BEC zone Old age Min. old requirementAtp 250 years 19%CWHvh1 250 years 13%CWHvm1 250 years 13%CHWvm2 250 years 13%MHmm1 250 years 19%

(ii) Visual quality objectives must be met on the productive forest within each VQOtypes and BEC zone combination:

VQO Type Young age Max. young requirementsSmallScaleAlteration 15 years 2%MinimalAlteration 15 years 5%NaturalAppearing 15 years 10%

Page 13: LBMF_KennedyFlatsFinalReport2002

13

Block Type: once a cell is selected to initiate a block, a type is selected. The followingtypes of blocks are modelled in the KFLM. The first four are for treatments in old forestand the latter ones for treatments in the second growth. VR stands for variable retentionand SG for second growth.

Block Type Retention Level Min. Re-entry intervalLowVR 15% n/aMediumVR 30% n/aHighVR 70% 20 yearsHighVR2ndPass 57% (40% net) n/aSGLowVR 15% n/aSGFirstPass 75% 25 yearsSGSecondPass 67% (50% net) 25 yearsSGThirdPass 50% (25% net) 25 yearsSGFourthPass 0% 25 years

The block types for partial harvest re-entries specify the expected retention level ofavailableforest within the treatment patch. The net overall retention level accounting forall prior treatment is shown in parentheses. For the high VR old forest second entry, 57%of the remaining forest after the first entry (70%) is retained, leaving a net of 40%retention after both entries. For the multiple-entry second growth treatments, 25% of theoriginal patch is harvested each entry.

The types of treatments available for a given cell depend on scenario, harvest history andstand age:

(i) First pass in old forest: the FPC scenario checks if this cell is in an adjacencybuffer, defined as cells within 300m of a cell previously harvested with lowretention (15%) that is younger than 15 years. Low retention blocks aredisallowed in adjacency buffers in the FPC scenario.

If this cell is in the heli-zone, or, for the WP scenario, a special management zone(defined as the riparian management zone or slopes over 60%), then the blocktype is set toHighVR

Otherwise the different block types have the following probabilities:

Block Type FPC scenario WP scenarioLowVR 70% 25%MediumVR 30% 50%HighVR 0% 25%

(ii) First pass in second growth forest: the FPC scenario applies blocks of typeSGLowVRwhile the WP scenario appliesSGFirstPass.

Page 14: LBMF_KennedyFlatsFinalReport2002

14

(iii) Second and later passes: these depend on the type of harvesting in the first pass.Note that the cases that do not permit a second pass do not forbid all futureharvest, but indicate that any future entries must be done as first entries (i.e. thedominant age must exceed min. harvest age).

Previous Block Type Block TypeLowVR No second passMediumVR No second passHighVR HighVR2ndPassHighVR2ndPass No second passSGLowVR No second passSGFirstPass SGSecondPassSGSecondPass SGThirdPassSGThirdPass SGLaterPassesSGLaterPasses SGLaterPasses

II. Model OutputsThe primary model outputs relevant to this analysis are described below.

Forest State Indicators

Old Forest: Annual output of the number of hectares old forest, stratified by watershedunit and by watershed unit/BEC zone:

- zone size- amount of old forest in zone, in THLB and in non-contributing productive

forest- amount of forest in THLB locked up to meet constraint- amount of productive forest reserved to meet constraint (contains above

quantity)- amount of old forest available for harvest

Harvest Indicators

Harvest Report: A range of output values that track key aspects of the harvesting process.All are means across the period

- area treated (which equals the area harvested plus the area retained)- area harvested- area retained- area available for harvest in old forest and in second growth- mean age harvested- amount (in km) of spur roads and mainline roads built

Page 15: LBMF_KennedyFlatsFinalReport2002

15

- area harvested by the various treatment types (i.e.LowVR, HighVR, etc.)

Block Size Distribution: Frequency of number of blocks, area, proportion of number andproportion of area in each potential block size between 1 and 40ha. Also output is thefrequency of attempted blocks (actual block harvested may be smaller due to constraints).

Spatial outputThe KFLM stores the following key spatial layers during each model run:(i) OldForest: Cells with age >140 years(ii) OldForestCost: Old forest cost surface(iii) WetlandCost: wetland cost surface(iv) StandAge(v) RoadType:class of road(vi) RoadState: activity state of road

III. Scenarios Evaluated

Both the FPC and WP scenarios were run for 100 years with 10 replicates using a 10-yeartimestep. The key spatial outputs were saved at years 50 and 100 for each replicate.

Besides the fundamental differences between the WP and FPC scenarios describedpreviously, the experiments applied the following parameters differently:

Harvest rate: We applied a harvest rate of 52ha/year for the WP. The rate for WP wasestimated based on the overall AAC for Clayoquot Sound, the expected volume/ha ofharvested stands, and the proportional size of the Kennedy Flats study area (D.MacGregor, pers. comm.) as follows:

- net area of Clayoquot Sound: approx. 175,000ha- net area of forest in study area: approx. 32,000ha (17.5% of sound)- AAC of Interfor and Iisaak for all of Clayoquot Sound: approx. 192,000m3

- Pro-rated AAC for study area: approx. 33,600m3

- Expected volume/ha: approx. 650m3/ha- Expected annual area to harvest: 52ha/year

To keep the analysis simple we doubled the harvest rate to 104ha/year for the FPCscenario. Thus the WP scenario harvests approximately 0.24% of the timber harvestinglandbase forest annually for a mean rotation time of just over 400years. The FPCscenario harvests 0.49% annually for a mean rotation time of just over 200 years. Notethat these rates refer to the amount of forest actually cut; treatment patches will be largerdepending on the level of retention.

Page 16: LBMF_KennedyFlatsFinalReport2002

16

Patch Size Distribution: For first entry blocks, the WP scenario selects block sizesrandomly between 1 and 40ha, while the FP scenario sets all target block sizes to 40ha.

Methods: Connectivity Models

I. Overall Connectivity Model DesignThe basis of our connectivity methods is derived from the methods of Keitt et al. (1997).The approach of Keitt et al. (1997) assumes that the matrix between patches ishomogenous. In many wildlife management situations, the matrix is not uniformlyunsuitable, but rather different conditions represent varying degrees of barriers todispersal. For example, relative to early seral forest, roads and lakes may have highdispersal impedance for terrestrial species, while mid-seral may present less of a barrier.The analysis I developed supports distance as measured in Euclidean (straight-line) spaceor in “movement cost” (least-cost) space where a unit of movement may be higher/lowercost depending on the presence of barriers. Cost can be differentiated according to landcover type, physiography, distance from habitat, etc. Distances are then interpreted as thecost to disperse along a link between two patches. The analysis of Keitt et al. (1997) isthe simple case when cost is constant (e.g. 1 unit per cell).

Habitat patches: Our approach is fundamentally patch-oriented. We first dichotomize thestudy area into habitat/non-habitat. Apatchis defined as a set of contiguous habitat cells(including diagonals). Aclusteris defined as a set of one or more habitat patches thateffectively provide contiguous habitat at a particular scale (cost or distance).

Many raster-based analyses suffer from grid artifacts, in particular tying the definition ofa region of connected habitat with contiguous patches of habitat cells. The main problemis that it is unlikely that species of interest perceive the landscape at the same scale as theraster resolution. Generally, the raster resolution determines the finest spatial scale of ananalysis, but it should not limit analysis at broader scales. One of the main advantages ofthe graph-based approach described below is that it frees analysis to explore multiplescales by linking sets of patches intoconnected clustersof habitat at various scales.

For each habitat map, the first step in this component of the KFLM is to identify thepatches in the patch map, to assign a unique identifier to each patch and to create a list ofelements, one for each patch. Each element tracks the patch id, size, centroid (mean rowand column) and other patch metrics (e.g. perimeter).

Landscape graphs: As in Keitt et al. (1997), I represent habitat configuration as amathematical graph. Nodes in a graph consist of habitat patches, while links representconnections between nodes. Unlike Keitt et al. (1997), I maintain a direct link with thelandscape pattern. That is, links between patches need not be straight lines, but may windand curve through space following the least-cost paths between patches. Both nodes andlinks have a set of attributes. In this experiment, links track the accumulated cost alongthe link, the path distance, and the straight-line distance between the two nodes, thestarting and ending raster position of this vector, and the type of link (see below). I

Page 17: LBMF_KennedyFlatsFinalReport2002

17

derived two types of graphs from the patch layer: a graph containing links between allpairs of neighbours and aplanar graph.

Theall neighbours shortest-distancegraph (neighbour graph; NG) has a link for eachpair of nodes (patches). For a landscape withn patches, it will haven(n-1)/2links. Iproduce this graph by processing patches sequentially. For each patchpi the modelspreads outward across the entire landscape, adding a link the first time the perimeter ofanother patchpj (i < j) is reached. For a large number of patches this may requiresubstantial processing time. However, once the graph is constructed, it is saved in a textfile format and read into other models for further processing. This step is processed oncefor straight-line distances and once using the cost surface. Since the cost surface variesacross space, it is not feasible to track the actual least-cost paths between all pairs ofpatches. During this process, the actually least-cost path can be saved in spatial format,but when post-processing links, can only be drawn using a straight-line between thepoints that are nearest in “cost space.” Links in the neighbour graph are classified asdirect or indirect. A direct link does not cross any intermediate patches.

A planar graph is a graph in which, when drawn on a piece of paper, no two links cross.I define theminimum planar graph(MPG) for a patch map as the graph consisting of alllinks e between nodesni andnj with distanced that satisfy the following criteria:(i) There is at most one link between each pair of nodes(ii) The link cannot cross any other patches(iii) There is no other pair of nodesnk andnl that are closer thand to each other with a

link that crossese(iv) There is no nodenk that is closer to the landscape boundary thand with a link that

crossese

The MPG produces a triangulation of the patches with the exception of some patches nearthe landscape boundary. This triangulation is similar to a Delaunay triangulation, but hassome unique characteristics. Rather than treating patch nodes as points in Euclideanspace, the derivation of the MPG explicitly accounts for patch shape. As a result, thelinks are not necessarily the shortest Euclidean distance links between nodes, asillustrated in the figure below. The graph on the left-hand side shows a patchconfiguration. Nodes 3 and 4 are closer than 1 and 2, precluding the link connectingnodes 1 and 2 between points c and d. However, nodes 1 and 2 can have a link betweenpoints a and b. The graph on the right-hand shows the corresponding mathematical graphwith nodes as points, which clearly illustrates the triangulation.

The minimum planar graph (MPG) has some useful properties. Both thenearestneighbourandminimum spanning treegraphs are sub-graphs of the MPG. A nearest

a b

dc4

321

1 23

4

Page 18: LBMF_KennedyFlatsFinalReport2002

18

neighbour for a nodeni is simply the closest nodenj. The nearest neighbour graph (NNG)contains the set of links between two nodesni andnj such thatni is the nearest neighbourof nj and/or vice versa. The nearest neighbour sub-graph may consist of more than oneconnected component, but each node has at least one link.

A mathematicaltree is a graph in which all nodes are connected in a single componentand there are no loops. The minimum spanning tree (MST) is the tree for which the sumof all link distances is minimum. The MST consists of all nearest neighbour links plussome other links to create a single connected component. A link between nodesni andnj

with distanced is in the MST if there is no path (sequence of links) fromni to nj for whichall links have distance less thand. The MPG adds further links to the MST to link allnodes that are closest without violating planarity of the graph.

II. Analysis MethodsGraph Extraction: I identify the MPG, MST and NNG by spreading from all patchessimultaneously. Initially all patches are considered to be in different connectedcomponents. Each time two patches meet for the first time I add a link to the MPG andjoin the connected components of the patches. If this is the first link for either patch, thenthe link is a nearest neighbour link. If the patches were in different connectedcomponents, then this is an MST link. Spreading continues until every cell in thelandscape has been visited. During this process, the patch closest to each cell isidentified, and the resulting layer is akin to a Voronoi diagram.

In the KFLM, creating the complete neighbour graph NG is one of the mostcomputationally intensive steps, since each distinct landscape state (current and eachreplicate at year 50 & 100 under WP and FPC scenarios), link type (straight-line, cost)and analysis (old forest, wetlands) must be processed.

Base Graph Analysis: I then analyzed the neighbour graphs (NG) using adaptations andextensions of the methods of Keitt et al. (1997). Essentially, for each neighbour graph,the model runs through a sequence of distance/cost thresholdsτ starting at 0. At eachthresholdτ, all links with cost greater thanτ are eliminated. The remaining patches arethen joined into connected clusters (a patch is in a cluster if and only if it has at least onelink joining it to another patch in the cluster). Then I compute a range of cluster metrics:(i) cluster size(ii) number of patches in cluster(iii) radius of gyration, which is defined as:

( ) ( )�=

−−=n

iii yyxx

nR

1

22 *1

where there aren cells in the component and (xi, yi) is the location of celli(iv) cluster centroid (x , y )(v) number of links between patches in cluster; this must be at least(n-1) for a

cluster withn patches

Page 19: LBMF_KennedyFlatsFinalReport2002

19

(vi) number of direct links(vii) link density: number of links divided by the maximum number of links, which is

n * (n-1)/2 for a cluster withn patches.(viii) total perimeter of cluster(ix) cluster perimeter/area ratio

These are summarized for the graph level:(i) number of clusters(ii) mean cluster size(iii) expected cluster size (area-weighted mean cluster size, or the size of cluster

expected from a randomly chosen cell).(iv) correlation length, which is defined as

n

Rn

C

m

jjj

d

�== 1

where there arem connected components in G’,nj is the number of cells incomponentRj, n is the total number of patch cells, andd is the threshold distance.

(v) total number of links in all clusters(vi) total number of direct links in all clusters(vii) total link density(viii) mean link density per cluster(ix) mean perimeter per cluster(x) mean cluster perimeter/area ratio

To determine the maximum threshold to assess, I ran a wide range of thresholds on thecurrent conditions to experimentally find the threshold at which cluster metrics stabilize.I chose to run the model over 26 thresholds (25 equal size increments) ranging from 0 tothis maximum value. I then performed the multi-scale analysis over this range ofthresholds.

Patch Importance Analysis: the purpose of the patch importance analysis is to assess therelative importance of the various habitat patches in term of contribution to connectivity.In both the old forest and wetland analysis, this is a key step in the analysis. However,these two analyses differ in one main respect: the old forest patches change in futureprojections under the WP and FPC scenarios, while wetlands patches remain static.Current old forest patches may be reduced or eliminated by harvesting, and patches maygrow, coalesce and new patches may form due to stand aging. Hence, assessing patchimportance for the future projected conditions does not make sense. However, assessingfuture relative importance, and changes in importance, for the wetland patches is useful.

For each landscape condition, thepatch importancetest consists of individually assessingthe importance of each patch at each threshold distance. For each patch/threshold pairthree types of tests are made:(i) patch removal: remove the patch and all links from the graph.(ii) link removal: remove only the links linked to a patch from the graph.

Page 20: LBMF_KennedyFlatsFinalReport2002

20

(iii) cluster removal:remove the patch and all other patches in the cluster in whichthis patch is a member.

A range of metrics is then recomputed for each resulting graph. For correlation length,removing the patch (as done by Keitt et al. 1997) can lead to some mis-interpretationssince removal of a small patch can actually lead to anincreasein the correlation lengthwhich is usually associated with an increase in connectivity. That is, it does not confoundthe effect of the configuration of a patch in the landscape with the effect of its size andshape on landscape connectivity.

Finally, given a graph level metric valueXd(d) for patchi at threshold distanced, wecompute thepatch importance indexandarea-weighted patch importance indexas:

( )d

dd

XiXX

d iI )(−= andi

dd a

iIiA

)()( =

whereai is the area (in hectares) of patchi. We focus on patch importance for expectedcluster size, which is also correlated with correlation length.

Results and Discussion

I. Landscape ProjectionFigure 4 and Figure 5 show the percent of area of each block type harvested towards theannual harvest target. In both cases the amount of old forest contributing towards thetarget declines over time, but much more markedly in the WP scenario.

Figure 4. Mean percentage of treatments in the FPC scenario.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 2 3 4 5 6 7 8 9 10

Decade

Per

cent

ofar

eaha

rves

ted

SGLowVR

MediumVR

LowVR

Page 21: LBMF_KennedyFlatsFinalReport2002

21

Figure 5. Mean percentage of treatments in the WP scenario.

The amount of road created in the WP scenario was slightly less than in the FPCscenario, with 119km compared to 144km for FPC. This has significant implications forthe wetland cost analysis, since roads are assumed to have a more critical impact thanseral stage. Figure 6 shows a portion of the wetland cost surface under current conditionsand at year 100 for a sample run in the WP scenario. Brighter areas indicate higher cost.The spur roads created during harvest have a more evident effect on the cost layer thanharvesting, which can be seen to generally reduce cost outside of roaded areas in theseimages.

Figure 6. Zoomed in section of wetland cost surface under current conditions and at year100 in a sample run from the WP scenario.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 2 3 4 5 6 7 8 9 10

Decade

Per

cent

ofar

eaha

rves

ted SGThirdPass

SGSecondPass

SGFirstPass

HighVR2ndPass

HighVR

MediumVR

LowVR

Page 22: LBMF_KennedyFlatsFinalReport2002

22

II. Old Forest Connectivity

Current Conditions

Base Graphs

Figure 7 shows the minimum planar graphs that result from applying Euclidean distanceand the cost surface. Overall, they present a similar pattern with fine textured linkagesbridging the main part of the flats. The MPG using the cost surface does not have somelinks along the outside since it is restricted to be contained within the study area. Notethat the MPGs show perimeter-to-perimeter links, not the least cost path between patches.

Figure 7. Minimum planar graph of initial old forest patches using straight-line distance(left) and cost surface (right).

Figure 8 shows the least-cost paths of all inter-patch links using the cost surface. Thisimage highlights corridors of lower cost between patches across the landscape. Inparticular, the connectivity importance of old forest patches in Pacific Rim National Park(PRNP), in Clayoquot Arm Provincial Park (CAPP), and down the peninsula separatingthe two arms of Kennedy Lake stand out. Also of interest is the web of connectionsamong smaller forest patches near Kennedy River, whereas Kennedy Lake and thesouthern portion of the flats show up with relatively low connectivity.

Page 23: LBMF_KennedyFlatsFinalReport2002

23

Figure 8. Least-cost links between initial old forest patches using the cost surface.

Critical Thresholds

The cluster metrics stabilize at about 2.5km for the Euclidean analysis, and so we used athreshold increment of 100m to keep a reasonable balance between the number ofthresholds and threshold increment. The left graph in Figure 9 shows how mean clustersize, expected cluster size and the correlation length change as the threshold costincreases. The right graphs show changes in the number of clusters. These metricshighlight certain thresholds at which rapid changes in values occur (albeit different ones).The main thresholds that we will focus on are at 100m, 300m, 500m, 700m, 900m, and1100m. The critical thresholds are the same for mean cluster size and correlation length,as these two metrics are correlated, although the magnitude of change is different.

When applying the cost surface, the cluster metrics stabilize at a cost threshold of about12,500, and so used a threshold increment of 500 cost units. The graphs in Figure 10show the same metrics as for the Euclidean analysis. The trend for mean cluster size ismarkedly different between the cost and straight-line analyses. When applying the costsurface, there appears to be a significant barrier to movement between thresholds of 6500and 10500. The main thresholds that we focus on are at 500, 1500, 2500, 4500, 5500,6500 and 10500. The key linkages formed at these thresholds correspond to the criticaldistances from Euclidean analysis.

Page 24: LBMF_KennedyFlatsFinalReport2002

24

Figure 9. Mean cluster size, expected cluster size and correlation length (left) and numberof clusters (right) at the different thresholds using straight-line distance.

Figure 10. Mean cluster size, expected cluster size and correlation length (left) andnumber of clusters (right) at the different thresholds using the cost surface.

Linkage Pattern at Critical Thresholds

The emerging pattern of connectivity at the identified thresholds is shown in Figure 11for the Euclidean analysis and in Figure 12 for the cost analysis. Links are drawn betweenpatch centroids since short perimeter-to-perimeter links are obscured. The finer scalethresholds (100-300m) result in forming connectivity among the patches on the southwest(PRNP) and northeast of the study area. At the scale of 500m, the landscape connectsacross the flats along the northwest, and at 900m another connection is made across thecentre of the flats. Broader scales reinforce these connections.

0

10

20

30

40

50

60

70

80

90

100

0 500 1000 1500 2000 2500

Thre s hold

Nu

mb

ero

fC

lust

ers

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

0 500 1000 1500 2000 2500

Thre s hold

Hec

tare

s/M

etre

s

Mean Cluster Size

Expected Cluster Size

Correlation Length

0

10

20

30

40

50

60

70

80

90

100

0 1500 3000 4500 6000 7500 9000 10500 12000

Thres hold

Num

bero

fClu

ster

s

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

0 1500 3000 4500 6000 7500 9000 10500 12000

Thre s hold

Hec

tare

s/M

etre

s

Mean Cluster Size

Expected Cluster Size

Correlation Length

Page 25: LBMF_KennedyFlatsFinalReport2002

25

Figure 11. Linkages between old forest patches at identified critical thresholds on currentforest conditions using the Euclidean distance. For clarity, links are drawn between patchcentroids.

τ = 100m τ = 300m

τ = 500m τ = 700m

τ = 900m τ = 1,100m

Page 26: LBMF_KennedyFlatsFinalReport2002

26

Figure 12. Linkages between old forest patches at identified critical thresholds on currentforest conditions using the cost surface. For clarity, links are drawn between patchcentroids.

These images show that connections formed at the key thresholds correspond between theEuclidean and cost analyses. At fine scales, using the cost surface leads to lower levels ofconnectivity at thresholds where critical links form. In general using the cost surfaceleads to key linkages being formed at larger thresholds. At the fine-scale thresholds, thevalues are 3 times the corresponding straight-line distance, but higher at broader scales.

τ = 500 τ = 1,500

τ = 2,500 τ = 4,500

τ = 6,500 τ = 10,500

Page 27: LBMF_KennedyFlatsFinalReport2002

27

To explore these effects in more detail, we show the perimeter-to-perimeter linkages forthe cost analysis in Figure 13 for the broader scales. This highlights that at a threshold of4500, connectivity across the flats is quite weak, but by 6500 it is strengthened by both asecond corridor across the center as well as additional connections. We also show theconnections for a very broad scale of 21,000 (effective distance of over 20km), whichillustrates the importance of the flats as a wide corridor between PRNP and themountains. This indicates that the flats are effectively connected at a broad scale for old-forest dependent species capable of moving at least 4.5km (effective distance) betweenold forest patches, but disconnected for species with more limited movement abilities.

Figure 13. Perimeter to perimeter linkages for some larger critical scales on current forestconditions using the cost surface.

Patch Importance Analysis

The results of the analysis of patch importance with respect to expected cluster size areshown in Figure 14 and Figure 15 for the Euclidean analysis and in Figure 16 and Figure17 for the cost analysis. The overall importance measure (shown in the left images)highlights the key large patches in PRNP, CAPP and elsewhere in the mountains to theeast. The area-weighted importance highlights small stepping stone patches thatcontributed significantly to connectivity for their size. At fine scales, the analysis

τ = 4,500 τ = 6,500

τ = 10,500 τ = 21,000

Page 28: LBMF_KennedyFlatsFinalReport2002

28

identifies small bridging patches near larger patches. At intermediate scales, stepping-stone patches across the flats are emphasized, while at broad scales patches key to thefinal linkages are identified.

Figure 14. Patch importance (left images) and area-weighted importance (right images)assessed at the finer critical thresholds on the current old forest conditions usingEuclidean distance. Darker patches have higher importance.

τ = 100m τ = 100m

τ = 300m τ = 300m

τ = 500m τ = 500m

Page 29: LBMF_KennedyFlatsFinalReport2002

29

Figure 15. Patch importance (left images) and area-weighted importance (right images)assessed at the broader critical thresholds on the current old forest conditions usingEuclidean distance. Darker patches have higher importance.

τ = 700m τ = 700m

τ = 900m τ = 900m

τ = 1,100 τ = 1,100

Page 30: LBMF_KennedyFlatsFinalReport2002

30

Figure 16. Patch importance (left images) and area-weighted importance (right images)assessed at the finer critical thresholds on the current old forest conditions using the costsurface. Darker patches have higher importance.

τ = 500 τ = 500

τ = 1,500 τ = 1,500

τ = 2,500 τ = 2,500

Page 31: LBMF_KennedyFlatsFinalReport2002

31

Figure 17. Patch importance (left images) and area-weighted importance (right images)assessed at the broader critical thresholds on the current old forest conditions using thecost surface. Darker patches have higher importance.

τ = 4,500 τ = 4,500

τ = 6,500 τ = 6,500

τ = 10,500 τ = 10,500

Page 32: LBMF_KennedyFlatsFinalReport2002

32

Projected Conditions

Base Graphs

Figure 18 and Figure 19 show the least-cost path network at years 50 and 100 for arandomly selected scenario under the FPC and WP scenarios. Both figures indicate anincrease in connectivity over time. However, the finer texture in the WP scenarioindicates a higher degree of connectivity than the FPC scenario.

Figure 18. Least-cost paths at years 50 and 100 for a sample FPC run using the costsurface.

Figure 19. Least-cost paths at years 50 and 100 for a sample WP run using the costsurface.

Critical Thresholds

Differences between the mean cluster size, number of clusters, expected cluster size andcorrelation length metrics of the current conditions with conditions at years 50 and 100under the FPC and WP scenarios are shown in Figure 20 and Figure 21 for Euclidean

Year 50 Year 100

Year 50 Year 100

Page 33: LBMF_KennedyFlatsFinalReport2002

33

distance and in Figure 22 and Figure 23 for the cost analysis. Due to the higher harvestlevel, the FPC scenario has a lower maximum mean cluster size (which represents thetotal area of old forest) while the WP scenario maintains and even slightly increases theamount of old forest. The critical scales of connectivity are distinctly different betweenthe WP, FPC and current conditions. In the FPC scenario when using Euclidean distance,there appears to be a decline in the landscape connectivity as indicated by the largerthresholds at which changes occur. Conversely, the WP scenario indicates maintenanceand slight enhancement of connectivity. However, when applying the cost surface, bothFPC and WP scenarios indicate increasing connectivity at broader scales as there appearsto be a decline in the thresholds as which changes occur. In both analyses, at year 100,the WP scenario also increases connectivity at finer scales, as shown by the more rapiddecline in number of patches and increase in cluster size at small thresholds.

Figure 20. Mean size and number of clusters for the current conditions and at years 50and 100 under the two management scenarios using Euclidean distance.

Figure 21. Mean expected cluster size and correlation length for the current conditionsand at years 50 and 100 using Euclidean distance.

0

20

40

60

80

100

120

140

0 500 1000 1500 2000 2500

Thres hold

Nu

mbe

rof

Clu

ster

s

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

0 500 1000 1500 2000 2500

Thres hold

Mea

nC

lust

erS

ize

(ha)

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

2000

4000

6000

8000

10000

12000

0 500 1000 1500 2000 2500

Thre shold

Cor

rela

tion

Len

gth

(m)

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

0 500 1000 1500 2000 2500

Thre shold

Exp

ecte

dC

lust

erS

ize

(ha)

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

Page 34: LBMF_KennedyFlatsFinalReport2002

34

Figure 22. Mean cluster size and number of clusters for the current conditions and atyears 50 and 100 under the FPC and WP management scenarios using the cost surfacedistance.

Figure 23. Mean expected cluster size and correlation length for the current conditionsand at years 50 and 100 under the FPC and WP management scenarios using the costsurface distance.

Linkage Pattern at Critical Thresholds

Figure 24 and Figure 25 show the linkages for several key thresholds in a single run atyears 50 and 100 under the two scenarios. Care must be taken when comparing linkagesfor different underlying patch maps due to patch formation and attrition. For example,consider the bottom right image in the two figures (Years 50 and 100 for WP scenario ata cost threshold of 2500). It may appear that connectivity is higher in the former as thereare more links. However, there are fewer, but larger patches in the latter. Comparisonsare more easily made on the summarized graph information. For the two imagesmentioned, the condition at year 100 has a higher expected cluster size and correlationlength (Figure 23).

0

20

40

60

80

100

120

140

0 1500 3000 4500 6000 7500 9000 10500 12000

Threshold

Num

bero

fC

lust

ers

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

0 1500 3000 4500 6000 7500 9000 10500 12000

Thres hold

Mea

nC

lust

erS

ize

(ha)

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

2000

4000

6000

8000

10000

12000

0 1500 3000 4500 6000 7500 9000 10500 12000

Thres hold

Co

rrel

atio

nLe

ngth Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

0 1500 3000 4500 6000 7500 9000 10500 12000

Thre shold

Exp

ecte

dC

lust

erS

ize

(ha)

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

Page 35: LBMF_KennedyFlatsFinalReport2002

35

Figure 24. Comparison of the linkages between old forest patches at year 50 for a samplerun under the FPC (left) and WP (right) scenarios. Two results from the Euclideananalysis (500m and 700m) and one from the cost analysis (2500) are shown.

FPC, Yr 50,τ = 500m,Euclidean

WP, Yr 50,τ = 500m,Euclidean

FPC, Yr 50,τ = 700m,Euclidean

WP, Yr 50,τ = 700m,Euclidean

FPC, Yr 50,τ = 2500,Cost

WP, Yr 50,τ = 2500,Cost

Page 36: LBMF_KennedyFlatsFinalReport2002

36

Figure 25. Comparison of the linkages between old forest patches at year 100 for asample run under the FPC (left) and WP (right) scenarios. Two results from theEuclidean analysis (500m and 700m) and one from the cost analysis (2500) are shown.

FPC, Yr 100,τ = 500m,Euclidean

WP, Yr 100,τ = 500m,Euclidean

FPC, Yr 100,τ = 700m,Euclidean

WP, Yr 100,τ = 700m,Euclidean

FPC, Yr 100,τ = 2500,Cost

WP, Yr 100,τ = 2500,Cost

Page 37: LBMF_KennedyFlatsFinalReport2002

37

III. Wetland Connectivity

Current Conditions

Base Graphs

Figure 26 shows the minimum planar graphs that result from applying Euclidean distanceand the cost surface. Overall, they present a similar pattern with fine textured linkages inthe western portion of the flats. As before, the MPG using the cost surface does not havesome links along the outside since it is restricted to be contained within the study area.Note that the MPGs show perimeter-to-perimeter links, not the least cost path betweenpatches.

Figure 26. Minimum planar graph of wetlands using straight-line distance (left) and costsurface on initial conditions (right).

Figure 27 shows the least-cost paths of inter-patch links using the cost surface. Thisimage highlights the concentration of wetlands and connectivity along the western part ofthe study area and several key corridors across the flats. We displayed the paths on top ofInterfor’s biodiversity corridors to show the fairly high match between the corridors andthe least-cost path locations. The isolated clusters of wetlands to the west arise becausewe imposed an upper inter-patch cost limit during the graph extraction process toimprove efficiency.

Critical Thresholds

The cluster metrics stabilize at about 5km for the Euclidean analysis, and so we used athreshold increment of 200m. The left graph in Figure 28 shows how mean cluster size,expected cluster size and the correlation length change as the threshold cost increases.These differ dramatically, since mean cluster size is strongly influenced by very smallpatches. The right graphs show changes in the number of clusters. The critical thresholdsthat we will focus on are at 200m, 400m, 600m, 800m, 1000m, and 1600m. The criticalthresholds are the same for mean cluster size and correlation length, as these two metricsare correlated, although the magnitude of change is different.

Page 38: LBMF_KennedyFlatsFinalReport2002

38

Figure 27. Least-cost links between initial patches using the cost surface displayed on topof some of Interfor’s biodiversity corridors.

When applying the cost surface, the cluster metrics stabilize at a cost threshold of about12,500, and so we used a threshold increment of 1000 cost units (effective increment of1km). The graphs in Figure 29 show the same metrics as for the Euclidean analysis. Thetrend for mean cluster size is markedly different between the cost and straight-lineanalyses. Several small wetlands remain isolated even at large thresholds, keeping themean, but not the expected, cluster size low. The main thresholds that we focus on are at1000, 2000, 3000, 4000, 5000, and 8000.

Figure 28. Mean and expected cluster size (left) and number of clusters (right) at thedifferent thresholds using straight-line distance.

0

50

100

150

200

250

300

0 1000 2000 3000 4000 5000

Threshold

Nu

mbe

rofC

lust

ers

0

100

200

300

400

500

600

700

800

900

0 1000 2000 3000 4000 5000

Thre shold

Hec

tare

s

Mean Cluster Size

Expected Clusters Size

Page 39: LBMF_KennedyFlatsFinalReport2002

39

Figure 29. Mean and expected cluster size (left) and number of clusters (right) at thedifferent thresholds using the cost surface.

Figure 30. Correlation length using straight-line distance (left) and the cost surface (right)at the different thresholds.

Linkage Pattern at Critical Thresholds

The pattern of connectivity at the identified thresholds is shown in Figure 31 for theEuclidean analysis and in Figure 32 for the cost analysis. Links are drawn between patchcentroids since short perimeter-to-perimeter links are obscured. The finer scale thresholds(200-600m) result in forming connectivity among the small groups of wetlands. Thesecoalesce at larger thresholds and at the scale of 1,000m, the landscape connects across theflats along the northwest, and at 1,600m connections are made across the northern portionof the Flats. Broader scales reinforce connections. In this case, the patterns formed at thedifferent thresholds do not correspond between the Euclidean and cost analyses. Usingthe cost surface leads to the formation of more tightly connected clusters.

0

50

100

150

200

250

300

0 5000 10000 15000 20000 25000

Thres hold

Num

ber

ofC

lust

ers

0

100

200

300

400

500

600

700

800

900

0 5000 10000 15000 20000 25000

Thres hold

Hec

tare

s

pMeanSize

pExpectedSize

0

1000

2000

3000

4000

5000

6000

7000

8000

0 1000 2000 3000 4000 5000

Thres hold

Cor

rela

tion

Len

gth

(m)

0

1

2

3

4

5

6

7

8

0 5000 10000 15000 20000 25000

Thre s hold

Co

rrela

tion

Len

gth

Page 40: LBMF_KennedyFlatsFinalReport2002

40

Figure 31. Linkages between wetland patches at identified critical thresholds on currentconditions using Euclidean distance. For clarity, links are drawn between patch centroids.

τ = 200m τ = 400m

τ = 600m τ = 800m

τ = 1,000m τ = 1,600m

Page 41: LBMF_KennedyFlatsFinalReport2002

41

Figure 32. Linkages between wetland patches at identified critical thresholds on currentconditions using the cost surface. For clarity, links are drawn between patch centroids.

τ = 1,000 τ = 2,000

τ = 3,000 τ = 4,000

τ = 5,000 τ = 8,000

Page 42: LBMF_KennedyFlatsFinalReport2002

42

Patch Importance Analysis

The results of the analysis of patch importance with respect to expected cluster size areshown in Figure 33 and Figure 34 for the cost analysis. We don’t show the area-weighted importance, since small wetlands are hard to see in the images.

Figure 33. Wetland patch importance at the critical thresholds on the current conditionsusing Euclidean distance. Darker patches have higher importance.

τ = 200m τ = 400m

τ = 600m τ = 800m

τ = 1,000m τ = 1,600m

Page 43: LBMF_KennedyFlatsFinalReport2002

43

Figure 34. Wetland patch importance assessed at the critical thresholds on the currentconditions using the cost surface. Darker patches have higher importance.

τ = 200m τ = 400m

τ = 600m τ = 800m

τ = 1,000 τ = 1,600

Page 44: LBMF_KennedyFlatsFinalReport2002

44

Projected ConditionsAssessing changes under projected conditions only applies to the cost surface analysis,since we do not projected any changes in the wetland patches themselves. Hence, all thefollowing results refer to the cost analysis.

Base Graphs

Figure 35 and Figure 36 show the least-cost path network at years 50 and 100 for arandomly selected scenario under the FPC and WP scenarios. Both figures show almostidentical patterns as in the current conditions, indicating the importance of relativelystatic (e.g. slope and large roads) features in influencing connections.

Figure 35. Least-cost paths at years 50 and 100 for a sample FPC run using the costsurface.

Figure 36. Least-cost paths at years 50 and 100 for a sample WP run using the costsurface.

Year 50 Year 100

Year 50 Year 100

Page 45: LBMF_KennedyFlatsFinalReport2002

45

Critical Thresholds

Differences between the mean cluster size, number of clusters, expected cluster size andcorrelation length metrics of the current conditions with conditions at years 50 and 100under the FPC and WP scenarios are shown in Figure 37 and Figure 38. Both the FPCand WP scenarios appear to result in an increase in connectivity between thresholds 5,000and 7,000, while the WP scenario also increases connectivity at a threshold of 20,000.Figure 39 shows how the number of direct links changes, which shows a relative increasein connectivity under the WP scenario.

Figure 37. Mean cluster size and number of clusters for the current conditions and atyears 50 and 100 under the FPC and WP management scenarios using the cost surfacedistance.

Figure 38. Mean expected cluster size and correlation length for the current conditionsand at years 50 and 100 under the FPC and WP management scenarios using the costsurface distance.

0

50

100

150

200

250

300

0 5000 10000 15000 20000 25000

Thre shold

Nu

mbe

rof

Clu

ster

s

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

20

40

60

80

100

120

0 5000 10000 15000 20000 25000

Thres hold

Mea

nC

lust

erS

ize

(ha)

Current

FPC Year 50

WP Year 50

FPC Year 100

WP Year 100

0

1000

2000

3000

4000

5000

6000

7000

8000

0 5000 10000 15000 20000 25000

Threshold

Cor

rela

tion

Leng

th(m

)

Current

FPC Y ear 50

WP Y ear 50

FPC Y ear 100

WP Y ear 100

0

100

200

300

400

500

600

700

800

900

0 5000 10000 15000 20000 25000

Threshold

Exp

ecte

dC

lust

erS

ize

(ha)

Current

FPC Y ear 50

WP Y ear 50

FPC Y ear 100

WP Y ear 100

Page 46: LBMF_KennedyFlatsFinalReport2002

46

Figure 39. Number of direct links for the current conditions and at years 50 and 100under the FPC and WP management scenarios using the cost surface distance.

Linkage Pattern at Critical Thresholds

Figure 40 and Figure 41 show the linkages for several key thresholds in a single run atyears 50 and 100 under the WP scenario. The connectivity linkages are very similar at allthree time periods. Comparing with the results for current conditions (Figure 32) we cansee that there is some increase in connectivity at thresholds 2,000-5,000. At year 100,connectivity is achieved across the flats at a lower threshold of 4,000.

ConclusionsOur analysis of spatial habitat patterns in the Kennedy Flats has highlighted a number ofinteresting issues, for both current conditions and projected future conditions. On thecurrent landscape, two main clusters of large old forest patches in Pacific Rim NationalPark and in Clayoquot Arm Provincial Park are separated by residual fragments of oldforest from past management activities. Our detailed analysis of straight-line connectivityin current conditions highlighted that critical scales of connectivity are presently in therange of 500m to just over a kilometer, while applying a cost surface highlightsconnectivity at effective distances of just over 1km to 10km. At the identified criticalscales, we mapped important patches, which may be useful to assist in selection ofmonitoring sites.

0

100

200

300

400

500

600

700

800

900

1000

0 5000 10000 15000 20000 25000

Threshold

Num

bero

fDire

ctLi

nks

Current

FPC Year 50

WP Y ear 50

FPC Year 100

WP Y ear 100

Page 47: LBMF_KennedyFlatsFinalReport2002

47

Figure 40. Comparison of the linkages between wetland patches at year 50 for a samplerun under the WP scenario. Results from the cost analysis at the indicated thresholds areshown.

WP, Yr 50,τ = 1,000 WP, Yr 50,τ = 2,000

WP, Yr 50,τ = 3,000 WP, Yr 50,τ = 4,000

WP, Yr 50,τ = 5,000 WP, Yr 50,τ = 8,000

Page 48: LBMF_KennedyFlatsFinalReport2002

48

Figure 41. Comparison of the linkages between wetland patches at year 100 for a samplerun under the WP scenario. Results from the cost analysis at the indicated thresholds areshown.

Connectivity changes under the two management regimes, which were designed tocapture the essence of Interfor’s watershed plan and a plausible regime that might have

WP, Yr 100,τ = 1,000 WP, Yr 100,τ =2,000

WP, Yr 100,τ = 3,000 WP, Yr 100,τ = 4,000

WP, Yr 100,τ = 5,000 WP, Yr 100,τ = 8,000

Page 49: LBMF_KennedyFlatsFinalReport2002

49

been in place without the Clayoquot Sound Scientific Panel recommendations. Bothhave relatively low harvest rates, and while the Forest Practices Code scenario maintainssome old forest, the watershed plan leads to an increase. Differences in harvest rate,retention levels and management rules between the two scenarios result in differenteffects on connectivity. Our analysis shows that Interfor’s watershed plan will likelyincrease connectivity of old forest across the flats over the next century. This increase isprojected to occur at a range of scales, with the largest benefit at intermediate to longerdistance thresholds. Incorporating an estimate of movement cost through the matrixseparating old forest patches, served to differentiate the scenarios more fully than usingonly Euclidean distance.

Changes seem to be less significant for wetland connectivity. This is due to the stronginfluence of topography and roads on the connectivity cost measure. Under the twoscenarios, increasing stand ages are offset by road development. Nonetheless, under theWP scenario, our analysis indicates that connectivity will improve slightly. The wetlandanalysis identifies some wetlands as being more crucial than others to maintainingconnectivity. Those have been identified with high importance values could be goodcandidates for assessing as monitoring sites.

Our analysis may be helpful in assisting monitoring programs to select both old forestpatches and wetlands for assessing population and habitat trends. Patches with highimportance values may be key components of connectivity in the flats, and so may begood candidates. These may be contrasted with patches that are less central toconnectivity. Likewise, patches that are expected to undergo significant change under theprojected conditions may be useful to monitor.

ReferencesB.C. Min. for Forests and B.C. Min. of Environment, Lands and Parks. 1999.

Biodiversity Guidebook.Beasley, B. and Fall, A. 2002. Kennedy Flats Ecosystem Dynamics Modeling Workshop

Proceedings. Long Beach Model Forest Society Report.Bélisle, M. and C. C. St. Clair. 2001. Cumulative effects of barriers on the movements of

forest birds. Conservation Ecology 5(2): 9. [online] URL:http://www.consecol.org/vol5/iss2/art9

Bunn, A.G., Urban, D.L., and Keitt, T.H. 2000. Landscape Connectivity: A ConservationApplication of Graph Theory.Journal of Environmental Management59(4):247-263.

Dale, V.H., Pearson, S.M., Oferman, H.L., and O’Neill, R.V. 1994. Relating patterns ofland-use change to faunal biodiversity in the central Amazon. ConservationBiology 8: 1027-1036.

Fahrig, L. and G. Merriam. 1985. Habitat patch connectivity and population survival.Ecology 66:1762-1769.

Fall, A. and J. Fall. 2001. A Domain-Specific Language for Models of LandscapeDynamics.Ecological Modelling. 141(1-3): 1-18.

Fall, A. 2001a. Assessing Critical Scales of Late Seral Forest Connectivity in theNorthern Columbia Mountains. Internal Ministry of Forests Report.

Page 50: LBMF_KennedyFlatsFinalReport2002

50

Fall, A. 2001b. Assessing Critical Scales of Woodland Caribou Connectivity in theInterlakes Region of Manitoba. Unpublished report.

Fall, A., and Sutherland G. 2000. Red-legged Frog Individual-based Model. ModelDesign and Usage. Unpublished report.

Keitt, T.H., D.L. Urban, and B.T. Milne. 1997. Detecting Critical Scales in FragmentedLandscapes.Conservation Ecology[online] 1(1): 4. Available from the Internet.URL: http://www.consecol.org/vol1/iss1/art4

Taylor, P.D., L. Fahrig, K. Henein, and G. Merriam. 1993. Connectivity is a vital elementof landscape structure. Oikos, 68:571-573.

With, K.A. 1999. Is Landscape Connectivity Necessary and Sufficient for WildlifeManagement? InForest Fragmentation: wildlife and management implications.J.A. Rochelle, L.A. Lehmann, and J. Wisniewski, eds. Pp. 97-115. Brill, Leiden,The Netherlands.

Working Group on Criteria and Indicators for the Conservation and SustainableManagement of Temperate and Boreal Forests. 2000. Montreal Process Year 2000Progress Report – Progress and Innovation in Implementing Criteria andIndicators for the Conservation and Sustainable Management of Temperate andBoreal Forests. Montreal Process Liason Office, Ottawa, ON.

Wright, P.A. 1999. Work Plan for the Monitoring Program of the Scientific Panel forSustainable Ecosystem Management in Clayoquot Sound. Long Beach ModelForest Society, Ucluelet, B.C.