Land Ownership and Land-Cover Change in the Southern Appalachian Highlands and the Olympic Peninsula

Ecological Applications. 6(4). 1996. pp. I 150-l 1720 1996 by the Ecological Society of America

LAND OWNERSHIP AND LAND-COVER CHANGE IN THE SOUTHERNAPPALACHIAN HIGHLANDS AND THE OLYMPIC PENINSULA’

M ONICA G. TURNERSEnvironmentul Sciences Division, Oak Ridge Narionul Laboratory, P.0. Box 2008, Oak Ridge, Tennessee 3783]-6038 UsA

DAVID N . WEARUSDA Forest Service, Forestry Sciences Laboratory, P.O. Box 12254. Research Triangle Park, North Carolina 27709 UsA

RICHARD 0. FLAMM~Environmental Sciences Division, Oak Ridge National Laboratory, p.0. Box ZOO& Oak Ridge, Tenne.r,ree 37831-6038 UsA

Abstract. Social and economic considerations are among the most important driversof landscape change, yet few studies have addressed economic and environmental influenceson landscape structure, and how land ownership may affect landscape dynamics. Watershedsin the Olympic Peninsula, Washington, and the southern Appalachian highlands of westernNorth Carolina were studied to address two questions: (1) Does landscape pattern varyamong federal, state, and private lands? (2) Do land-cover changes differ among owners,and if so, what variables explain the propensity of land to undergo change on federal, state,and private lands? Landscape changes were studied between 1975 and 1991 by using spatialdatabases and a time series of remotely sensed imagery. Differences in landscape patternwere observed between the two study regions and between different categories of landownership. The proportion of the landscape in forest cover was greatest in the southernAppalachians for both U.S. National Forest and private lands, compared to any land-ownership category on the Olympic Peninsula. Greater variability in landscape structurethrough time and between ownership categories was observed on the Olympic Peninsula.On the Olympic Peninsula, landscape patterns did not differ substantially between com-mercial forest and state Department of Natural Resources lands, both of which are managedfor timber, but differed between U.S. National Forest and noncommercial private landownerships. In both regions, private lands contained less forest cover but a greater numberof small forest patches than did public lands.

Analyses of land-cover change based on multinomial logit models revealed differencesin land-cover transitions through time, between ownerships, and between the two studyregions. Differences in land-cover transitions between time intervals suggested that addi-tional factors (e.g., changes in wood products or agricultural prices, or changes in laws orpolicies) cause individuals or institutions to change land management. The importance ofindependent variables (slope, elevation, distance to roads and markets, and populationdensity) in explaining land-cover change varied between ownerships. This methodologyfor analyzing land-cover dynamics across land units that encompass multiple owner typesshould be widely applicable to other landscapes.

Key words: land use; land-cover change; landscape ecology; Olympic Peninsula; remote sensing,southern Appalachians; spuiiol cmalysis.

INTRODUCTION

Landscapes are dynamic mosaics of natural and hu-man-created patches that vary in size, shape, and ar-rangement. Although considerable attention has beengiven to describing changes in landscapes through time(e.g., Johnson and Sharpe 1976, Whitney and Somerlot1985, Iverson 1988, Turner and Ruscher 1988, Turner1990a, Hall et al. 1991, Kienast 1993, LaGro andDeGloria 1992, and many others), few studies have

’ Manuscript received 29 August 1994; revised 6 April1995; accepted 25 June 1995.

z Present address: Department of Zoology, Birge Hall, Uni-versity of Wisconsin, Madison, Wisconsin 53706 USA.

3 Present address: Florida Marine Research Institute, 1008th Avenue SE, St. Petersburg, Florida 33701-5095 USA.

attempted to understand economic and ecological in-fluences on landscape structure (Turner 1987, Parks andAlig 1988, Parks 1991) and land ownership (Lee et al.1992, Spies et al. 1994, Wear and Flamm 1993).

Several authors have noted the critical need forknowledge about why landscape changes occur andhow environmental factors and market processes in-teract (e.g., Turner 1987, Baker 1989). In the southernAppalachian highlands, Wear and Flamm (1993) foundthat the likelihood of forest cover being disturbed wasa function of (1) the type of owner; (2) environmentalattributes such as slope, aspect, and elevation; and (3)locational variables, such as distance to roads or marketcenters, which related to the economics of forest har-vest and residential development. In Rondonia, Brazil,Dale et al. (1993) demonstrated that land use and land-

1150

November 1996 OWNERSHIP AND LAND-COVER CHANGE

TAXE 1. Percentages of land by ownership category in the Little Tennessee (LTRB), Hoh(HORB), and Dungeness (DURB) River Basins.

Ownership LTRB DURB HORBU.S. Forest Service (USFS) 35 22 <IPrivate (includes commercial and small private owners) 65 24 1 7National Park Service (NPS) 32 58Washington Department of Natural Resources (DNR) <l 24Wilderness 1 8

1151

cover changes were a function of individual parcel sizesand shapes, attributes of individual land owners, sitecharacteristics such as soils and agricultural suitability,and the distances to the road network. In the CascadeRange in Oregon, Spies et al. (1994) found the greatestdecline in coniferous forest on private lands, least inwilderness, and intermediate in public nonwildernessareas. Clearly, social and economic considerations areamong the most important drivers of landscape change,and a broad-based understanding of landscape structureand function is essential for promoting integrated man-agement where human sustenance and environmentalintegrity are considered part of the same system (Leeet al. 1992).

The developing paradigms of sustainability (e.g.,Lubchenco et al. 1991) and ecosystem management(e.g., Agee and Johnson 1988, Overbay 1992, Slocombe1993, Grumbine 1994) require understanding of land-scape dynamics and ecological processes across co-herent land units, which typically encompass multipleowner types. The units for which land management andplanning are undertaken frequently bear little resem-blance to the ecological systems within which they re-side or to their connections to economic and socialprocesses (Slocombe 1993). Rivers and streams, forexample, frequently are used as jurisdictional bound-aries, bisecting watersheds. Protected lands often arenot sufficiently large to encompass natural disturbancedynamics and are embedded within a matrix of landsof mixed ownerships. Understanding landscape dynam-ics in areas of mixed ownerships is an important com-ponent of ecosystem management, yet managers andscientists have only begun to develop strategies thatrecognize ecological conditions outside jurisdictionalboundaries.

We examined the influence of land-ownership pat-terns on landscape structure in two forest-dominatedlandscapes: the Olympic Peninsula, Washington, andthe southern Appalachian highlands of western NorthCarolina. The Olympic Peninsula is dominated by ex-tensive tracts of land controlled by few owners, where-as the southern Appalachian highlands region containssmaller tracts of land controlled by many different own-ers (Table 1). We recognize, as did Mladenoff et al.(1993), that comparison between landscapes, or be-tween ownerships within the same landscape, limits theinferences that can be applied to the general class offorested landscapes. Nonetheless, comparisons be-

tween landscapes and within watersheds can provideuseful and necessary information on the complex dy-namics that occur in landscapes of mixed ownership.We are aware of only one other study that analyzedlandscape patterns on adjacent public and private own-erships (Spies et al. 1994). Spies et al. (1994) focusedon forest-cutting patterns, addressing the spatial pat-terns and rates of change of coniferous forest between1972 and 1988 in and around the Willamette NationalForest. They observed differences in cutting patternsbetween the public and private lands, with lower ele-vation portions of the landscape on private lands havinghigher rates of disturbance and a greater proportion ofearly successional habitats. In this study, we analyzeland-cover patterns and dynamics on public and privatelands in two different forested regions, and test forsignificant differences in how land-cover transitionsvary between owners.

Public lands may be managed in various ways bydifferent institutions. We recognize that the true ownersof the public lands are the citizens of the country, state,or other political jurisdiction, and that agencies simplymanage the lands for the public. For simplicity of pre-sentation in this paper, however, we will use “landownership” to refer to the entity managing public lands(e.g., National Forests, State Department of NaturalResources, etc.).

We hypothesized that land-ownership patterns, i.e.,the abundance of land in different ownership classes,such as public and private, would have a discernibleand predictable influence on landscape structure andland-cover change. This general hypothesis was ex-amined by comparing landscape changes in our studyregions during a 16-yr time period (1975 to 199 1). Twospecific questions were addressed: (1) Does landscapepattern vary between federal, state, and private lands?(2) Do land-cover changes differ among owners, and,if so, what variables explain the propensity of land toundergo change on federal, state, and private lands?Landscape pattern was analyzed by ownership classwithin each watershed to address the first question. Therelationship between land-cover changes and severalexplanatory variables by ownership class was exam-ined to address the second question. Our analysis as-sumed that land-use choices are based on comparisonsof revenues and costs of various land uses at each site.Observed changes in land cover (or the lack of a

1152 MONICA G. TURNER ET AL. Ecological ApplicationsVol. 6, No. 4

change) reflect these choices about land use and theranking of land values.

The value of land in any particular use is defined bythe prices for its associated goods and services and aset of cost factors. Our study areas are relatively small(in a market sense), so that we can assume that prices(e.g., the prices of delivered logs and agricultural prod-ucts) and unit costs (e.g., the wage rate and costs ofcapital and energy) are constant throughout each wa-tershed. Accordingly, these factors should not affectland use specialization within a watershed for a specifictime period. However, there is a set of factors that arevariable within each watershed and that define costdifferentials between locations within the area. Thesefactors include: (1) steepness of the site, measured asits slope; (2) elevation of the site, which serves as aproxy for vegetation and climatic differences; (3) dis-tance between the site and the nearest road, definingaccess costs; and (4) distance between the site and thenearest market for its goods and services, measuredalong the road network. Distance to market defines aset of transportation costs for goods and services. Ad-ditionally, (5) population density in the neighborhoodof a site should influence its comparative advantage indifferent uses. These variables define either the qualityof the site (slope, elevation, population density) or thelocation of the site within a physical/human landscape(access distance and distance to market), and shouldtherefore influence land rents and land uses (Katzman1974). As part of the second question, we examinewhether or not these explanatory variables influenceland-cover change within the study areas by developingstatistical models and testing three hypotheses:

Hypothesis 1. Tempo& change in transition mod-els.--Factors not included in the mode1 (e.g., changesin prices for wood and agricultural products, or changesin policies and laws) may vary in time and may shiftthe probability of land-cover transitions between pe-riods (e.g., land-cover changes may differ between197.51980 and 1980-1986). Thus, we test the nullhypothesis that the relationship between the explana-tory variables and the probabilities of land-coverchange did not change between periods.

Hypothesis 2. Effects of ownership on transitionmodels.-We may similarly test for identical transitionmodels between the ownerships represented in eachwatershed. Here, the null hypothesis is that the rela-tionship between explanatory variables and the prob-abilities of land-cover change did not differ betweenowners.

Hypothesis 3. Effects oj’spatial variables on trun-sition models.-We hypothesized that the five explan-atory factors would be positively related to the costsof productive activities, in general, within the water-shed (summarized in Table 2). A negative relationshipwas hypothesized between slope, elevation, distance toroads, and distance to market and the probability offorest conversion (e.g., land-use conversion or forest

TABLE 2. Hypothesized direction of marginal effects of independent variables on specific land-cover transitions. Aplus indicates an expected positive relationship, and a mi-nus indicates an expected negative relationship.

Land-cover transition

Forest Grass Unve- Unve-Forest to un- Grass to un- getat- getat-

t o vege- to for- vege- ed to- ed toVariable grass tated est tated forest grass

Elevation + - + +Slope - - + - + +Distance to road - - + - + +Distance to market - - + - + +Population - + - -I- - -

management), as well as the probability of agricultural .land conversion to developed uses (transition fromgrassy to unvegetated cover). In contrast, we expectedthat increased population density would create moredevelopment pressure (e.g., conversion from forest orgrassy covers to unvegetated cover), but would de-crease the probability of forest harvesting. Conversely,population density would be negatively related to tran-sitions from grassy or unvegetated cover to forest cov-er.

S T U D Y AR E A S

We studied two forested landscapes: the OlympicPeninsula, Washington, and the Southern AppalachianMan and Biosphere (SAMAB) region, a multistate zoneof cooperation within the U.S. Man and the BiosphereProgram. These landscapes were selected because theyreflect vastly different land-ownership patterns andmay serve as microcosms for many land-cover changesobserved in forested regions of temperate North Amer-ica.

Southern Appulachiun Highlands

The SAMAB region encompasses the southern Ap-palachian highlands and extends approximately fromChattanooga, Tennessee, northeast to Roanoke, Virgin-ia, crossing four states. Approximately 57% of the SA-MAB region is held in small private ownerships, andU.S. Forest Service (USFS) lands account for another20% of land ownership. Forested lands in the SAMABregion have experienced increasing demands for non-market services, and associated pressures to decreasetimber harvests. The Great Smoky Mountains NationalPark is the most visited national park in the U.S. be-cause of the tremendous human population within al-d drive, and this recreation demand also affects ad-jacent national forests and private lands. The relativelysmall holdings of the national forests in the southernAppalachians are interspersed among many land own-ers and must be managed in the context of a regionalmixed-ownership landscape.

Within the SAMAB region, we selected the LittleTennessee River Basin (LTRB) for intensive study. The

November 1996 OWNERSHIP AND LAND-COVER CHANGE 1153

/ OLYMPIC \

SAMAB\

L i t t l e T e n n e s s e e

F IG. 1 . Maps of the Olympic Peninsula and southern Appalachians, with locations of the Hoh, Dungeness, and LittleTennessee River Basins indicated.

116 090-ha LTRB is located primarily in western NorthCarolina, extending approximately from the Georgia-North Carolina border to Fontana Dam, just south ofthe Great Smoky Mountains National Park (Fig. 1).Although = 10% the LTRB is located in north Georgia,we considered only the 103,635 ha located within NorthCarolina, because of limited availability of digital spa-tial data for the Georgia area. The LTRB is character-ized by rugged topography and species-rich eastern de-ciduous forest. Franklin, North Carolina, the major de-veloped area in the LTRB, is experiencing an influx ofnew residents. Tourism in Franklin, now a $50 mil-lion/yr business, is growing. Forest products remain animportant industry in the LTRB, and the USFS is amajor landholder, owning 35% of the watershed, pri-marily at the higher elevations (Table 1, Fig. 2). Therotation of forest cutting on the national forest landsranges from 80 to 120 yr; harvest is primarily cove andupland hardwoods for saw timber. The USFS CoweetaHydrological Laboratory, a Long-term Ecological Re-search (LTER) site, also is located within the LTRB.

palachians. The controversy over the harvest of old-growth timber and conservation efforts focused on theNorthern Spotted Owl (Strix occidentalis) in the PacificNorthwest have underscored the importance of under-standing landscape dynamics on the Olympic Penin-sula.

Olympic Peninsula

Two watersheds (Hoh and Dungeness River Basins)on the Olympic Peninsula were selected for intensivestudy (Fig. I), because a representative range of land-ownership classes did not occur in a single watershed.Both basins originate in the high elevations of theOlympic National Park, centrally located on the Pen-insula. The 58 876-ha Dungeness River Basin (DURB)extends north from the Park to the town of Sequim.Major land-ownership classes in the DURB are the Na-tional Park Service, USFS, and small private owner-ships in the Sequim area (Table 1, Fig. 2). The78 007-ha Hoh River Basin (HORB) extends west fromthe Park to the Pacific Ocean. Major land-ownershipclasses in the HORB are the National Park Service, theWashington DNR, and large commercial private own-erships (Table 1, Fig. 2).

The Olympic Peninsula, Washington, encompasses-1.6 X lo6 ha, with the Olympic National Forest andOlympic National Park comprising nearly one-third ofthe land area. The pattern of land ownership on theOlympic Peninsula is quite different from that in theSAMAB region. Both public and private lands gener-ally are held in large blocks, and the majority of thenonfederal lands are managed for timber production bythe state of Washington’s Department of Natural Re-sources (DNR) and by large private corporations. Smallprivate ownerships comprise only ~21% of the Olym-

M E T H O D S

Datubase development

Land-cover interpretation.-Land-cover patternswere interpreted from Landsat Multispectral Scanner(MSS) and Thematic Mapper (TM) imagery for fourtime periods in each region. In the LTRB, MSS imagerywas dated 25 August 1975; 7 August 1980; 21 July1986; and 7 May 1991. Dates of MSS imagery for theOlympic Peninsula, encompassing both the HORB andDURB, were 31 May 1975; 5 August 1980; and 3 Au-

pit Peninsula, compared to ~57% in the southern Ap- gust 1986. TM imagery from 16 September 1991 was

MONICA G. TURNER ET AL. Ecological ApplicationsVol. 6, No. 4

Hoh

SCALE 1:640,000KILOMETRES__.-.---.-L--S_-_~--..-..

16 0 16 3 2

Little Tennessee

I Wilderness

Private

DNR

Park Service

F IG . 2 . General patterns of land ownership in the Hoh, Dungeness, and Little Tennessee River Basins.

used for the most recent time period on the OlympicPeninsula. Because the resolution of MSS imagery is-90 m, the landscape was represented as a 90-m gridand other data layers were resampled to a 90-m reso-lution, as necessary, within the Geographic ResourcesAnalysis Support System (GRASS) geographic infor-mation system (GIS) (USA CERL 1991). Image inter-pretation was done similarly for each region withinERDAS (ERDAS 1994).

MSS data were classified prior to rectification andresampling. An iterative self-organizing algorithm inERDAS (ISODATA) was run on MSS bands 2, 3, and4 (the first band had heavy striping that could not becorrected). This resulted in 100 spectral signatures thatwere used by MAXCLAS, a maximum-likelihood clas-sifier, to generate a final single-layer coverage basedon spectral similarities. For the Olympic Peninsula,MSS land-cover layers were registered to the 199 1 TMscene using the image-to-image registration feature

within ERDAS. Each image was rectified with a rootmean square (RMS) error of less than 1 pixel (90 m).

For classification on the Olympic Peninsula, the TMimage was cut to encompass the two study areas. Thesix bands used (l-5 and 7) were transformed using theTM Tasseled Cap into “brightness,” “greenness,” and“wetness” spectral indices (Crist and Cicone 1984),which is useful in determining structural characteristicsin western hemlock and Douglas-fir forests (Cohen andSpies 1992). The “wetness” layer was minimally in-fluenced by topographic shadowing and had the highestcorrelation to stand structure. Cohen and Spies (I 992)proposed that “wetness” be renamed to “maturity” toreflect the relationship to structural attributes of closed-canopy coniferous forests. The “maturity” layer is es-sentially a contrast between bands 5 and 7 in the mid-infrared range, whereas “greenness” expresses the dif-ference between the first three visible bands (1, 2, 3)and the near-infrared band 4. “Brightness” is a weight-

November 1996 OWNERSHIP AND LAND-COVER CHANGE 11.55

ed sum of all six bands together, weighing bands 3, 4,and 5 twice as much as the others.

A relative sun-incidence layer was developed fromavailable 7.5’ (I:24 000) digital elevation models(DEMs) to help reduce the effects of topographic shad-owing on classification (Eby 1987). Where there weregaps in these larger scale data, smaller scale 1:250,000DEMs, resampled to 25 X 25 m, were inserted. TheERDAS RELIEF program calculates the relative sunincidence using the sun’s azimuth and elevation at thetime of image acquisition (provided in the header data),resulting in a shaded relief layer that reflects conditionsat the time of the satellite’s overpass.

The brightness, greenness, maturity, and sun-inci-dence layers were combined in a separate image file,and an unsupervised classification algorithm (ISO-DATA) in ERDAS was used to generate 150 class sig-natures. A maximum-likelihood classifier (MAX-CLAS) was run on the data using the ISODATA sig-natures as input. Based on field experience in the studyareas and available aerial photos, the original 150 TMclasses were condensed to 12 land-cover classes, whichwere used to generate random sample sites for accuracyassessment. A stratified random sample was generatedusing the ERDAS RANDCAT program. In total, 241points were located in the Hoh River drainage, 179were located in the Dungeness watershed, and 157 inthe Little Tennessee watershed. These points were plot-ted on color prints of the original TM imagery and thenwere taken to the field for closer assessment. Elevenpercent of the plots were visited in the field, and theremaining plots were evaluated using the most recentaerial photography available from DNR, NPS, andUSFS. Results of the accuracy assessment were alsoused to further lump the land-cover classes to increasethe overall accuracy. Despite the use of the sun-inci-dence layer, topographic effects were still apparent inthe classified images. Final accuracies of the inter-preted maps were >90%.

The final land-cover classes used in the study wereas follows. In the LTRB, analyses were conducted onthree classes from the final map layer (Fig. 3): (1) for-est, which was primarily mixed hardwoods with oc-casional stands of pine; (2) grassy cover, including ag-ricultural fields, pasture, lawns, and old fields; and (3)unvegetated, which included exposed soil, pavement,and developed areas. In the HORB and DURB (Fig.3). the grassy and unvegetated classes were as de-scribed for the LTRB. For forest cover, however, co-niferous forest was distinguished from deciduous/mixed forest (primarily alder regeneration plus someareas of cottonwood and big-leaf maple), resulting infour classes. Vegetation classes in the Olympics weresimilar to those developed by the Wilderness Society(Morrison 1992).

Other spatial c&a.-In addition to land cover, a setof spatial data layers including slope, aspect, elevation,land ownership, roads, and population density was as-

sembled for each region and stored in the GRASS,Slope, aspect, and elevation were derived from 7.5’digital elevation model (DEM) data obtained from theU.S. Geological Survey for both regions. The DEMdata were imported into GRASS at 25-m resolution;slope, aspect, and elevation layers were created withinGRASS and resampled to a 90-m cell size. Land-own-ership maps for part of the LTRB were obtained indigital form from the USFS (for the Wayah RangerDistrict), and the remainder of the ownership patternin the LTRB was digitized manually from 1.24000maps. For the Olympic Peninsula, data on general own-ership were obtained from the Puget Sound River BasinTeam, Washington DNR’s GIS database, the USFS, Na-tional Park Service, and commercial owners. Primaryand secondary road data were obtained for a single timeperiod only from the 1990 TIGER lines, which werereceived as ARC/Info coverages then converted toGRASS. The North Carolina Center for GeographicalInformation and Analysis (NCGIA) provided the roaddata for the LTRB. Road data were used within GRASSto derive two additional data layers: (1) distance fromeach grid cell to the nearest road, and (2) distance fromeach grid cell, by road, to the nearest market center.Market centers included Franklin, North Carolina inthe LTRB; Sequim, Washington in the DURB; andHighway 101 in the HORB. Finally, population densitydata were obtained from the 1990 census at the censustract level (irregular polygons of varying size) fromTIGER/Line Census files from the NCGIA and theWashington Geographic Redistricting System. Gridcells occurring within a given census tract received thepopulation density for that tract.

Lnndscupe pattern analysis

Land-cover patterns were analyzed by computing in-dices that describe both overall landscape pattern andthat of each land-cover class. Analyses for all four timeperiods were conducted separately for private and pub-lic ownership classes within each watershed by usingthe SPAN program (Turner 1990a, b). The proportionof the landscape area, p, occupied by each cover typewas calculated. Nearest neighbor probabilities, q,,,,which represent the probability of cells of land coveri being adjacent to cells of land cover j, were calculatedby dividing the number of cells of type i that are ad-jacent to typej by the total number of cells of type i.The q,,, values were used in the contagion index (Eq.2) and as a fine-scale measure of the degree of clumpingin any cover type.

Two overall landscape indices adapted from O’Neillet al. (1988) were calculated. The first index, D, i s ameasure of dominance, calculated as the deviation fromthe maximum possible landscape or habitat diversity:

D = I Hmax + 2 V’,)logV’,) I/ H,,,, (1)1 ,=, I/

il.56

N

W E

s

Kilometres

i - - ~ -------- I0 10

MONICA G. TURNER ET AL.

a) LTRB

Forest

Grass/brush

Unvegetated

Ecologicnl ApplicationsVol. 6, No. 4

FIG. 3. Land-cover patterns interpreted from Landsat Multispectral Scanner (MSS) imagery for each of four time periodsin (a) the Little Tennessee, (b) the Hoh, and (c) the Dungeness River Basins.

November 1996 OWNERSHIP AND LAND-COVER CHANGE

b) HORB

1980 c) DURB

Conifer

Deciduous/mixed

.2-..><0i:~.. Grass/brush

Unvegetated

Water

0 Snow/ice

N

W E

S

Kilometresr--- - -~;

0 10

1157

FIG. 3 . Cont inued

1158 MONICA G. TURNER ET AL. Ecological ApplicationsVol. 6. No. 4

where m = number of land-use types observed on themap, P, is the proportion of the landscape in land usei, and FL,,, = log(m), the maximum diversity when allland uses are present in equal proportions. Values ofD approaching one indicate a landscape that is domi-nated by one or a few land uses, and lower valuesnearing zero indicate a landscape with land uses rep-resented in approximately equal proportions.

The second overall index, C, measures contagion, orthe adjacency of land-cover types (O’Neill et al. 1988,Li and Reynolds 1993). The index is calculated froman adjacency matrix, Q, in which q,,, is the proportionof cells of type i that are adjacent to cells of type j,such that:

C = Km,,, + 2 2 (q,.,)lWq,,,)I/

Lx, (2),=, ,=I

where K,,,,,= m log(m) and is the absolute value of thesummation of (q,,,)log(q,,,) when all possible adjacen-ties between land-cover types occur with equal prob-abilities. The index C will be zero when the pattern isdispersed and all possible adjacencies occur with equalprobability. Values of C approaching one indicate alandscape with a clumped pattern of land-cover types.

The remaining metrics were computed separately foreach land-cover category. For each cover type, everypatch in the landscape map was identified and its areaand perimeter recorded. A patch was defined as con-tiguous, adjacent (horizontally or vertically) cells ofthe same land cover; diagonal cells were not consideredto be contiguous. The total number of patches, arith-metic mean patch size, area-weighted mean patch size,size of the largest patch, and number of single-cellpatches were recorded for each cover type. The arith-metic mean patch size was calculated by simple divi-sion of the summation of the patch sizes, XS,, by thenumber of patches, N,. The area-weighted mean patchsize was computed by dividing the summation of thesquared patch sizes, c(S,2) by CS,. Finally, an edge-to-area ratio was computed for each cover type by dividingthe total number of edges by the number of grid cellsof that cover type.

Models of lurid-cover change

Land-cover change was estimated on a per-grid-cellbasis by using the time series of remote imagery. Aregular grid was used to sample individual grid cellswithin each watershed, and the land-cover category ateach of the four time periods plus the values of the fiveindependent variables and the land-ownership classwere recorded for each sampled grid cell. Sample sizesfor each watershed were 4008, 8298, and 10459 gridcells for the LTRB, DURB, and HORB, respectively.The watersheds and ownership patterns were of dif-fering size, and the spacing between individual pointsin the sample grids varied among watersheds (spacing:LTRB, six grid cells north-south and five grid cells

east-west; DURB and HORB, three grid cells north-south and east-west). Because we focused on areassubject to human-influenced land-cover change, NPSlands and wilderness areas were excluded from thesample. Although including the “background” land-cover changes occurring in the absence (mostly) ofanthropogenic factors might yield an interesting com-parison, there was also substantial between-year vari-ation for ice and snow cover in Olympic National Park,limiting its utility for this purpose.

We posited that the probability of land cover chang-ing was related to the five spatially explicit variables:slope, elevation, distance to the nearest road, distanceto the nearest market, and population density. DefineXk as the 6 X m matrix of these five independent vari-ables plus a unitary constant for each of m sites withan initial land-cover class k. In addition, define Y,l asa discrete variable that defines land-cover change. Ifthe cover class did not change, then YF = 0; Y: = 1or 2 if the cover class k changes to one of the two othercover classes. The probability of cover changing fromk to j is defined as a function of site attributes:

Pr(Y,k = j) = P,: = F,k(X,k) for all j, (3)

where Pr is the probability operator and F is a cumu-lative distribution function. The relationship in Eq. 3can be estimated by specifying the functional form forF and fitting the equations to observed land-coverchanges. We chose to examine this process using amultinomial logit formulation (e.g., Maddala 1983):

P,” = F&V,)eX’R

p,” = -sems ’

where e is the base of the natural logarithms and 6 isa 6 X 1 vector of estimated coefficients. A multinomialchoice model is appropriate because the dependentvariable is discrete and takes on more than two values.Because the dependent variable is categorical (i.e., itsvalues have meaning only as labels), either a probitmodel (based on a joint normal distribution) or a logit(based on the logistic distribution) may be applied.Monte Carlo analysis indicates that the logistic closelyapproximates the normal (Judge et al. 198.5:778). Fur-thermore, the logistic has a closed-form solution thatprovides a practical advantage for predicting proba-bilities of land-cover change.

The multinomial logit defines F as a logistic cu-mulative-distribution function (CDF), which closelyapproximates the normal. Because the probabilities oftransition must sum to one, one of the equations in Eq.4 is redundant, so one equation was dropped from theestimation exercise. For our analysis, we estimated theequations for land-cover change (where Y,l = 1 or 2)and dropped the no-change alternative. The effects ofvariables on the “null transition” were then estimated

using the estimated coefficients and the “summing-up” to the length of the period. The lengths of our periodsrule. are: (1) 60 mo for 1975-1980, (2) 72 mo for 1980-

We defined models for three initial cover classes (in- 1986, and (3) 58 mo for 1986-1991. We treated thedexed by k): forest, grassy, and unvegetated (coniferous periods 1975-1980 and 1986-1991 as essentiallyand mixed forest were modeled separately in the HORB equivalent. To make the period 1980-1986 comparableand DURB). Each of three different blocks of two equa- to the others, we evaluated each observed transitiontions defined by Eq. 4 were estimated separately by with a draw from a uniform distribution (between 0maximizing their respective likelihood functions using and 1). If the number was less than 60/72 = 0.8333,a nonlinear optimization algorithm implemented in the then the transition was recorded. Otherwise, it was dis-software package LIMDEP (Greene 1992). The three carded.hypotheses were tested as follows by using estimation Effects of ownership on transition models-We sim-results. ilarly tested for identical transition models between the

Temporal change in transition models.-Land-cover two ownerships. The model for the null hypothesis oftransition models (Eq. 4) defined the probability of a identical transition models constrains all elements oftransition as a function of five spatially variable factors. p to be equal between ownerships. For the alternativeWe hypothesized that the influence of other unmea- hypothesis, all elements of p were allowed to varysured factors (e.g., wood and agricultural prices, or between ownerships. The likelihood ratio test was usedchanges in policies) would shift the average probability to construct the test.of transition, but would not affect the marginal effects Effects of spatial variubles on transition modek-of spatial variables on probabilities. That is, they would The general hypothesis that the spatial factors (eleva-affect only the intercept terms in the vector 6. This tion, slope, distance to a road, distance to market, andhypothesis was tested by determining whether or not population density) explain cover transitions was ex-the relationships between the various explanatory vari- amined by testing for the significance of the estimatedables and transition probabilities remained constant be- models (we test for a significant difference from thetween periods. The test was implemented by assuming null model defined by constraining all elements of /3,that, although the constant term might shift between except the intercept, to zero) and defining the associ-periods to reflect changes in unmeasured variables that ated likelihood ratio test. The significance of individualvaried over time, all other coefficients would be con- variables in explaining specific land-cover transitionsstant between periods. This required constructing (1) was also tested (hypotheses are summarized in Tablea transition model for the null hypothesis, where all p 2). With a linear regression, we could simply test thecoefficients except the intercept were held constant for significance of the estimated coefficients (6). However,the two periods, and (2) a transition model for the al- with the multinomial logit model, the estimated coef-ternative hypothesis, where all p coefficients were al- ficients and their variances do not necessarily corre-lowed to vary between periods. The null model was, spond to the sign, relative magnitude, or significancetherefore, a constrained version of the alternative. Ac- of the referenced transition probability. Marginal ef-cordingly, we can test the hypothesis by comparing fects of individual variables and their variances werelikelihood function values for the null and alternative calculated from the estimated CDF (e.g., ME, = SF/&X,)models (see Judge et al. 1985: 182). The log likelihood and variance-covariance matrix for p. These estimatesratio test was constructed as: depend on the value of the independent variables (X),

LR = -2 ln(L,IL,), (5)and we set all X to mean values. This generates a testthat is conservative: i.e., a marginal effect may prove

where L,. and L, are likelihood values for the con- to be insignificant with independent variables set atstrained and unconstrained versions of the model. LR mean values but significant for some other plausiblehas a chi-squared distribution, with degrees of freedom combination of values. We tested for significance at theequal to the number of constraints imposed to form the P 5 0.05 level but, in light of the conservative naturenull hypothesis. The model for the alternative hypoth- of the test, also report the results for the P 5 0.20 level.esis is defined by using a dummy variable (D) that isequal to one for one period and equal to zero for the RESULTS

other: Land ownership and landscape patternF( ) = (X’P + DX’y). Little Tennessee River Basin.-Landscape patterns

Accordingly, the dummy variable allows for different differed subtly between USFS and private lands in therelationships between probabilities and site attributes LTRB. Forest was the dominant land cover in bothfor the different periods. The null hypothesis is con- ownerships, although forest was in lower proportionsstructed by constraining all y to zero. on private lands (0.78-0.86) than on USFS lands (0.96-

To construct comparisons between periods of un- 0.98) (Table 3). Dominance and contagion were alwaysequal length, we assumed that the probability of any greater on the USFS lands than on private lands (D =transition occurring within a period was proportional 0.91, 0.92, 0.86, 0.96 on USFS, and 0.66, 0.59, 0.54,

November 1996 OWNERSHIP AND LAND-COVER CHANGE 1 1 5 9


TABLE 3. Indices of landscape pattern for USFS and private land ownerships in the Little Tennessee River Basin, NorthCarolina, from maps derived from Landsat MSS imagery, where p, is the proportion of the watershed in the specified covertype; q,, is the probability of adjacency for grid cells of the same cover type being adjacent in the horizontal or verticaldirection; and N,.,,,, is the number of single-cell patches.

USFS Private

Index 1975 1980 1986 1991 1975 1980 1986 1991

a) Forest cover

P,YuNo. patches (N)Normalized NtAverage natch sizer

0.980.99

107107404

0.850.98 0.96 0.960.99 0.97 0.98

1 1 1 134 1191 1 1 134 119391 317 350

7763 7686 763237 55 420.29 0.34 0.34

14273 14226 14072

0.860.93

0.830.91

549297124

39460156

0.780.90 0.90

409221173

42868117

734397

462250145

4492488

1.5923215

0.5328351

Weigh<ed average s’ize$ 7670Normalized N,.,,,,? 36 143

0.5254477

--..Edge/area 0.30Largest patch size* 14114

b) Unvegetated cover

P, 0.012a . . 0.41

0.42 0.4954525 51189

0.006 0.0080.27 0.19

145 228145 228

1.8 1.53.2 2.3

0.0020.23

48481.63.0353.119

0.0300.37

940508

88 1572.98 3.27

12 7

2.611.3

2792.54

71

0.014 0.031 0.036 0.1050.31 0.31 0.33 0.45

307 651 711 2196307 651 711 1187

2.0 2.1 2.2 3.95.1 5.3 4.8 22.9

190 414 420 5612.88 2.84 2.76 2.2227 28 21 150

0.0550.41

1472796

3.120.8

4162.37

212

0.048 0.0170.32

6310.41

1301,No. patches (N) 183Normalized Nt 183Average patch size* 2.9Weighted average sizef 7.0

7033.0

21.7390

2.38

3412.1

16.9Normalized N,,,,t 89Edge/area 2.44Largest patch size* 29

c) Grassy/brushy coverP, 0.01142, 0.32No. patches (N) 237Normalized Nt 237Average patch size$ 2.1Weighted average size$ 5.9Normalized N,.,,,,? 150Edge/area 2.84Largest patch size* 29

2302.72

193 125

0.1150.43

2673

0.1670.51

25451376

5.494.9

7381.98

529

0.1290.37

36521974

2.816.0

11062.54

273

14443.5

14.9765

2.3191

7 Number of patches was normalized for differences in area of each ownership class by dividing the actual number ofpatches on private lands by the ratio of private : USFS lands (0.65/0.35 = 1.85); this permits the number of patches to becompared between the ownerships.

$ Units are 90 X 90 m grid cells.

0 .65 on pr iva te lands ; C = 0 .67 , 0 .64 , 0 .51 , 0 .48 onUSFS, and 0.38, 0.39, 0.36, 0.36 on private lands for1975, 1980, 1986, and 1991, respect ively) . These in-dices indicate a more even distr ibution and lessclumped spatial pattern of land-cover classes on privatelands than on USFS lands.

The spatial arrangement of land-cover classes onUSFS lands remained remarkably constant from 1975to 1991. Although the number of forest patches fluc-tuated somewhat, there was little change in the area-weighted average patch size, size of the largest patch,and number of single-grid-cell patches for forest cover(Table 3). Unvegetated and grassy/brushy cover typescomprised a total of 2-4% of the USFS lands between1975 and 1991, and the most notable change was anincrease in the number of grassy/brushy patches and adecrease in the number of unvegetated patches. Patchsizes of these two cover types, both average and area-weighted average, showed little change (Table 3).

More variabi l i ty in landscape pat tern through t imewas observed on private lands (Table 3). Forest coverdeclined by 8% between 1975 and 1986, then increased

in 1991 to levels comparable to those in 1975. Thenumber of forest patches increased between 1975 and1986, with an almost twofold increase in the numberof single-cell forest patches. Average, area-weightedaverage, and largest patch size for forest cover de-creased during the 1975-1986 period, and then in-creased in 1991. Unvegetated areas accounted for 2-5% of the private land cover between 1975 and 1991,and grassy/brushy cover accounted for IO-17%. Area-weighted average sizes of patches of nonforest covervaried substantially through time (e.g., sixfold forgrassy/brushy cover) (Table 3).

Differences between the USFS and private lands re-flect the greater abundance of nonforest cover classesand larger relative number of small patches on privatelands (Table 3). Forest edge-to-area ratios were greateron private lands compared to USFS lands. The averagesize, area-weighted average size, and number of patch-es of nonforest cover were substantially larger on pri-vate lands compared to USFS lands. It is interesting tonote, however, that the size of the largest patch of con-t iguous forest was greatest on private lands, reflecting


TABLE 4. Indices of landscape patterns for USFS and private ownership types in the Hoh River Basin, Washington, basedon Landsat MSS imagery. Indices are defined as in Table 3.

Index 1975

a) Coniferous forest coverP, 0.564tr 0.86No. patches (N) 281Normalized Nt 199Average patch size* 46Weighted average sizej: I63 1Normalized N,.,,,,t 113Edge/area 0.65Largest patch size+ 2969

b) Deciduous/mixed forest coverP, 0.08qtz 0.53No. patches (N) 460Normalized NT 326Average patch sizei. 4.4Weighted average size+ 17Normalized N,.crllt 159Edge/area 2.06Largest patch size$ 56

c) Grassy/brushy cover1’1 0.090.. 0 .5 11 I .

No. patches (IV) 522Normalized Nt 370Average patch size$ 4.2Weighted average size$ 1 8Normalized N,.,,,?Edge/areaLargest patch size$

d) Unvegetated coverPIr/uNo. patches (N)Normalized Nt

1932.08

72

0.250.72

518367

Average patch sizes 11.4Weighted average size$ 87Normalized N,.,.,t 189Edge/area 1.22Largest patch size$ 309

DNR Private

1980 1986 1 9 9 1 1975 1980 1986 1 9 9 1

0.54 0.40 0.50 0.26 0.270.82 0.77 0.76 0.66 0.64

388 484 619 616 628275 343 439 616 62833 1 9 1 9 7.0 7 . 1

850 480 1357 93 98145 176 238 320 311

0.79 1.01 1 . 0 7 1.54 1.581762 1348 3531 372 444

0.10 0.26 0.14 0.42 0.400.51 0.65 0.49 0.74 0.71

544 694 877 407 464386 492 622 407 464

4.4 8.7 3.7 1 7 . 5 14.422 121 26 580 322

199 228 372 205 2352.00 1.49 2.15 1 . 1 8 1 . 2 5

122 466 139 1750 1 0 4 1

0.14 0.19 0.16 0.17 0.160.52 0.48 0.35 0.58 0.54

718 1029 1556 503 547509 730 1103 503 547

4.7 4.4 2.5 5.7 4.822 25 I O 64 53

274 396 688 268 3192.00 2 . 1 1 2.66 1.82 1 . 9 5

99 169 58 252 250

0.21 0.15 0.20 0.14 0.180.66 0.63 0.67 0.66 0.67

583 574 634 368 397413 407 450 368 697

8.4 6 . 1 7.4 6.3 7.482 54 64 129 1 3 8

218 238 264 205 2271.74 1 . 5 8 1.44 1 . 5 5 1.48

387 287 220 414 437

0.22 0.390.57 0.69

635 655635 655

5.9 10.15 1 320

319 3761 . 8 3 1.37

1 8 7 746

0.49 0.190.74 0.49

399 944399 944

20.7 3.4486 621 7 5 589

1.14 2.171267 332

0.16 0.210.47 0.42

811 11248 1 1 1124

3.4 3 . 126 1 5

501 6672.26 2.41

122 73

0.11 0.200.62 0.66

328 426328 426

5.5 7.939 53

159 2171.74 1 . 5 3

171 163

t Number of patches was normalized for differences in area of each ownership class by dividing the actual number ofpatches on private lands by the ratio of private : USFS lands (0.65/0.35 = 1.85); this permits the number of patches to becompared between the ownerships.

$ Units are 90 X 90 m grid cells.

in part, the extent and spatial distribution of the own-erships themselves (Fig. 2).

Hoh River Basin.-Both DNR lands and privatelands (with primarily commercial owners) in the HORBshowed low-to-moderate levels of Auctuation in land-scape pattern through time, but some differences inlandscape structure between ownerships were observed(Table 4). In general, the dominance index was greateron DNR lands than on private lands (e.g., D = 0.36and 0.25, respectively, in 1975), but contagion wassimilar between ownerships at each time period, rang-ing between 0.37 and 0.48. Coniferous forest landsgenerally occupied 250% of the DNR lands but only22-40% of the private lands (Table 4). Coniferous for-est was more aggregated on the DNR lands than on

private lands, as indicated by both the nearest-neighborprobabilities and the edge-to-area ratios. Average patchsizes of coniferous forest were consistently greater onDNR lands than on private lands. The area of decid-uous/mixed forest increased on DNR lands but de-creased on private lands (Table 4), with private landshaving larger patches. The proportion of the landscapeoccupied by grassy/brushy cover increased on bothDNR and private lands, and the spatial pattern of thiscover type was similar between the two ownerships.The spatial pattern of unvegetated cover also showedfew differences between ownerships, with patch char-acteristics being similar, and fluctuations through timerelatively small.

Dungeness River B&n.-Although the USFS and


TABLE 5. Indices of landscape pattern for USFS and private ownership types in the Dungeness River Basin, Washington,based on Landsat MSS imagery. Indices are defined as in Table 3.

Index 1975

USFS Private

1980 1986 1991 1975 1980 1986 1991

a) Coniferous forest coverP, 0.73Yzr 0.88No. patches (N) 89Average patch size? 130Weighted average sizet 6759No. I -cell patches 66Edge/area 0.56Largest patch sizet 8352

b) Deciduous/mixed forest coverPC 0.10qtt 0.30No. patches (N) 716Average Datch sizet 2.2Weigh;ed average sizet 6.3No. l-cell patches 448Edge/area .Largest patch sizet

c) Grassy/brushy cover

Pa0..

2.8236

0.0050.32

1 I I

No. patches (N)Average uatch sizetWeighTed average sizetNo. l-cell patchesEdge/area -Largest patch sizei

d) Unvegetated cover

Yt4rrNo. patches (N)Average patch sizetWeighted average sizetNo. l-cell patchesEdge/areaLargest patch sizet

392.15.7

242.78

1 7

0.15 0.06 0.09 0.14 0.210.63 0.50 0.62 0.60 0.61

321 250 211 387 5017.4 3.9 6.7 6.0 7.275 36 102 47 143177 132 112 217 246

1.58 2.12 1.64 1.70 1.5980 706 544 228 627

0.83 0.82 0.73 0.17 0.23 0.180.92 0.92 0.88 0.64 0.73 0.62

66 78 107 434 301 395200 167 110 6.7 11.3 8.07551 7202 6229 196 216 80

45 44 70 257 138 1910.40 0.40 0.58 1.59 1.21 1.59

9211 8871 7746 649 551 356

0.06 0.06 0.01 0.31 0.16 0.170.43 0.48 0.22 0.58 0.56 0.54

298 220 128 819 460 5523.4 4.2 1.6 6.6 5.8 5.2

11.0 14.0 3.4 88 38 461. 5 8 107 99 442 215 272

2.29 2.11 3.11 1.72 1.77 1.8638 45 13 330 154 190

0.04 0.03 0.11 0.29 0.29 0.300.38 0.40 0.31 0.59 0.51 0.55

245 183 772 587 877 7452.9 3.0 2.2 8.4 5.7 6.96.7 8.4 6.6 120 40 60

126 105 520 292 400 3752.50 2.43 2.78 1.68 1.99 1.83

22 27 34 462 220 200

0.30 0.330.69 0.70

393 40713.2 13.9

554 627195 198

1.24 1.231469 1770 10105

0.110.56

4044.7

69239

1.86185

0.020.35

1862.15.9

1292.73

20

0.160.45

7933.634

4772.22

216

0.680.87

19759.6

8717120

0.55

t Units are 90 X 90 m grid ceils.

small private ownerships accounted for similar pro-portions of the DURB (22% and 24%, respectively),the landscape patterns observed in these ownershipsdiffered dramatically (Table 5). The USFS lands weredominated by coniferous forest cover, ranging between0.73 and 0.83 of the landscape between 1975 and 1991.Private lands had a much lower proportion of conif-erous forest land, ranging between 0.11 and 0.23, witha net decrease through time. The dominance index wasgreater on USFS lands than on private lands for alltime periods (D = 0.52,0.62,0.62,0.55 for USFS, andD = 0.20, 0.21, 0.21, 0.46 for private lands for 1975,1980, 1986, and 1991). Contagion increased throughtime and was similar between ownerships, ranging be-tween 0.42 and 0.65.

The USFS lands were characterized by moderatechanges through time and no net loss of forest cover,whereas private lands exhibited substantial loss of for-est. The abundance and spatial distribution of conif-erous forest cover were relatively stable on USFS lands(Table 5). Deciduous/mixed forest cover decreased on

USFS lands, with associated decreases in patch sizes;grassy/brushy cover increased; and unvegetated covershowed moderate fluctuation through time (Table 5).

Private lands, however, changed dramatically. Theproportions of coniferous forest and deciduous/mixedforest on private lands declined substantially between1975 and 1991, with concomitant large declines in av-erage and area-weighted average patch sizes (Table 5).Patch sizes of conifers were much smaller on privatelands than on USFS lands, and private lands generallyhad four to six times the number of conifer patches(Table 5). The largest patch of contiguous coniferousforest cover was an order of magnitude greater on USFSthan on private lands. The proportion of the landscapein grassy/brushy cover on private lands also declinedby about half during the time interval. In contrast tothese declines, the unvegetated cover on private landsincreased between 1975 and 1991 from 0.21 to 0.68(Table 5). Increased connectivity of the unvegetatedcover is evident in the increase of the q,, values from0.61 to 0.87, order-of-magnitude increases in average


TABLE 6. Tests for temporal change in transition models. Each entry is the log likelihoodratio for the test of identical transition models between the referenced periods, by initialCover class. ** indicates rejection of identical transition models at the P 5 0.01 level. Blankcells indicate that the test could not be constructed, due to a limited sample size relative tothe number of independent variables in the alternative model.

Lands, periods dfConifer

forest

Initial cover class

Forest? CklSSY Unvegetateda) Little Tennessee River Basin

Private lands1975-1980 vs. 1980-1986 1 01980-1986 vs. 1986-1991 10

U.S. Forest Service1975-1980 vs. 1980-1986 101980-1986 vs . 1986-1991 10

b) Hoh River BasinPrivate lands (commercial)

1975-1980 vs. 1980-1986 15 50.01**1980-1986 vs. 1986-1991 15 33.15**

Washington Department of Natural Resources1975-1980 vs. 1980-1986 15 73.40**1980-1986 vs . 1986-1991 15 19.54

c) Dungeness River BasinPrivate lands

1975-1980 vs. 1980-1986 18 33.151980-1986 vs . 1986-1991 18 20.02

U.S. Forest Service1975-1980 vs . 1980-1986 18 19.871980-1986 vs . 1986-1991 18 27.31

19.38 19.4048.76"* 20.34

33.62**20.62

16.6527.49**

37.94**39.58**

29.0137.8-l**

48.47**28.25

54.74**50.16**

81.01**38.39**

67.91**68.22**

34.1839.04**

74.14**16.98**

18.8347.67**

24.38 18.72 48.64**31.76 23.57 34.61

+ Deciduous forest with occasional nine in Little Tennessee River Basin; mixed deciduousand forest in Hoh and Dungeness Basins.

and area-weighted average patch sizes, decreases in thenumber of single-cell patches, and a decrease in edge-to-area ratio.

Land ownership and landscape change:Little Tennessee River Basin

Temporal change in transition models.--7-he hy-pothesis of identical transitions for forest cover be-tween 1980-1986 and 1986-1991 on private lands wasrejected (Table 6a). However, we do not reject identicaltransitions for forest cover between 1975-1980 and1980-1986. There was no significant difference forgrassy cover in both cases, but for unvegetated coverwe found a significant difference between 1975-1980and 1980-l 986 but not between 1980-I 986 and 1986-1991.

A similar result for forest cover was observed on theUSFS lands (Table 6b). That is, the overall relationshipbetween site features and transition probability for for-est land was not significantly different between 197%1980 and 1980- 1986, but shifted between 1980-I 986and 1986-1991. Tests for temporal change in grassyand unvegetated cover types could not be constructedfor USFS lands because of limited degrees of freedom(i.e., there were very few observations relative to es-timated parameters in the alternative model).

Effects of ownership on transition models.-We test-ed for differences in transition models between own-erships by comparing pooled and separate-effects mod-els based on the tests of temporal change in transitionmodels. Accordingly, we pooled data for the 1975%1980 and 1980-I 986 periods for forest and grassy cov-er, as they did not differ.

In both 1975-1986 and 1986-1991, the hypothesisof identical transition models for the private and publicownerships was rejected (Table 7), indicating structuraldissimilarities in the spatial relationships for forestcover changes between USFS and private lands. How-ever, there were no significant differences between thetransition models for grassy and unvegetated cover onpublic and private lands. This may reflect the relativelysmall sample size for these types of cover on nationalforests.

Effects of spatial variables on transition models.-All models estimated for the LTRB were significant.On private lands, nine of the 30 marginal effects weresignificantly different from zero for the period 197%1980 (Table 8). Slope was especially important in ex-plaining these transitions. For the transition from forestto grassy cover, slope was negative, consistent with ourexpectations (see Table 2) of the effect of cost on timberharvest or development. Slope was also a significant

1164 MONICA G. TURNER ET AL. Ecological Applications

TABLE 7. Tests for differences in transition models betweenownership classes in each watershed. Each entry is the loglikelihood ratio (with a chi-squared distribution) for thetest of identical transition models between ownerships. Thedegrees of freedom for each test differ for pooling periods1975-1980 and 1980-1986 in some cases. ** indicates re-jection of identical transition models at the P S 0.01 level.Blank cells indicate that the test could not be constructeddue to a limited sample size relative to the number of in-dependent variables in the alternative model.

Initial cover class

Conifer Unveg-Lands, periods df forest Forest? G r a s s y etated

a) Little Tennessee River Basin

U.S. Forest Service vs. private lands1975-1986 14 48.23** 2 4 . 6 61986-1991 12 50.63** 22 .1 I

b) Hoh River Basin

7.50

Washington Department of Natural Resources vs. privatelands (commercial)

1975-1980 15 42.05*” 42.12** 44.60** 75.31**1980-1986 IS 47.68** 2 5 . 2 6 26.7 1 56.95**1986-1991 1 5 35.28** 47.81*” 2 4 . 4 6 19.46

c) Dungeness River Basin

U.S. Forest Service vs. private lands1975-1980 15 34.63 51.39** 55.66**1980-1986 15 40.43** 46.02** 67.92”* 84.61**1986-1991 15 45.56** 49.22** 44.60** 38.71**

t Deciduous forest with occasional pine in Little TennesseeRiver Basin; mixed deciduous forest in Hoh and DungenessBasins.

factor for the transition from forest to unvegetated cov-er. For grassy cover on private lands in the period 1975-1980, transition back to forest was positively relatedto slope at the P C 0.20 but not the P 5 0.05 level(Table 8), consistent with our expectations reflectingincreasing costs of development. The same relationshipwas reflected in the negative effect of slope and dis-

Vol. 6, No. 4

tance to market on the grassy-to-unvegetated transition.However, the relationship between the grassy-unve-getated transition and elevation was significantly pos-itive, whereas we expected a negative marginal effect.Transition from unvegetated to forest land was posi-tively related to slope, consistent with expectations.Transition from unvegetated cover was positively re-lated to elevation but negatively related to slope. Thelatter result is counterintuitive.

In contrast to results for the period 1975-1980, onlythree of 30 effects on private lands were significantlydifferent from zero during the period 1986-l 991 (Table8). None of the spatially explicit variables had a sig-nificant effect on transitions from forest cover or ontransitions from grassy and unvegetated cover to forest.For transitions from grassy to unvegetated cover, wefound a negative relationship with elevation and slopeand a positive relationship with population (the latterat only the P 5 0.20 level). All three results wereconsistent with our expectations. The only significantrelationship for the unvegetated-to-grassy transitionswas reflected in a negative marginal effect for popu-lation.

Changes in marginal effects between periods mayprovide insights into the structural change observed inforest cover transitions between 1975-l 986 and 1986-1991. In general, fewer spatial variables produced sig-nificant marginal ‘effects in the period 1986-1991. Toexamine the resulting implications for forest-coverchange, we plotted the probabilities of a forest-grassytransition for the three periods, relative to distance toroads, slope, and elevation (three variables producingsignificant marginal effects in 197.5-1980). Probabili-ties were calculated with all variables held at samplemeans and the referenced variable varied over its ob-served range (Fig. 4). All relationships had the antic-ipated downward-sloping effect of the variable on a

TABLE 8. Tests of hypotheses regarding marginal effects of independent variables on specificland-cover transitions for private lands in the Little Tennessee River Basin. For each entry,the tirst sign represents the expected sign of the marginal effect on the referenced transition(from Table 2). The sign in parentheses is the marginal effect, calculated at the means ofthe other independent variables. * indicates that the calculated marginal effect is significantat the P 5 0.05 confidence level; absence of asterisk indicates significance at the P 5 0.20level. NS indicates the result was not significant.

Forest to Grass to Unveg- Unveg-Forest unveg- Grass to unvegetated to etated to

Variable to grass etated forest etated forest grass

1975-1986Elevation

1;:; ’-“;t,*NS -(+)* NS +c+j*

Slope * +(+) -c-1* +(+I* +c-j*Distance to road

-‘;S)NS NS NS

Distance to market NS NS -“;$ NS -.;: )*Population NS +(+I NS NS NS NS

1986-1991Elevation NS NS NSSlope NS NS NS ?T NS

NSNS NS

Distance to road NS NS NS NS NS NSDistance to market NS NS NS NS NSPopulation NS NS NS +(+I NS -‘;“,*


s 0.2022 0.16

a. Distance to roads-0J!? 0. ’ 2 F..,,g 0.08 “Y .._,,_

'X.,g . .._. . . . . . . . 'y..0.04 'Ct~.:,~,---.--------...............................~..~~~~~~

8 0.00‘.‘.“.-.........-...-..,-... -....-....-_._ .,,.._______,__,,,_, ” ,,,_ * ,,___,,_,_ _,

rfI 0 5 10 15 20 25 30 35Distance (m)

b. Slope

0 10 20 30 40 50 60 70Slope (“)

0.20 I I

c. Elevation

0.12 1 I

f

600 750 go0 1050 1200 1350 1500 1650Elevation (m)

FIG. 4. Comparison of the probability of transition fromforest cover to grassy cover in the Little Tennessee RiverBasin, as related to independent variables between time pe-riods.

forest-to-grassy transition. However, these relation-ships changed substantially between periods. Proba-bility of transition was concentrated much closer toroads for the period 1986-1991 (Fig. 4a). In contrast,forest-to-grassy transitions were less influenced byslope and elevation in this later period (Figs. 4b andc). Accordingly, the probability of disturbance wasgreater at higher elevations and on steeper slopes in1986-1991 than in the previous two periods.

Effects of spatial variables on transition models weresubstantially different for the USFS lands. In contrastto the private lands, no spatial variable produced asignificant marginal effect on any transition probabil-ity, even at the P 5 0.20 level. Furthermore, only eightof 30 coefficients were significant in constructing thecumulative distribution function for transitions.

Land ownership and landscape chunge:Hoh River Basin

Temporal chunge in transition models-Tests fortemporal change in transition models indicate rejectionof stable models on private lands in the HORB (Table6b). With the exception of grassy cover between theperiods 1975-l 980 and 1980-l 986, all transition mod-els for private lands shifted significantly through time.

On DNR lands, transitions for the coniferous foresttype (which applies to the largest share of the land-scape) were constant between 1980-1986 and 1986-1991. However, all other temporal comparisons ofmodels indicate dynamic transition relationships forDNR lands.

Effects of ownership on transition models.-Differ-ences between private and DNR transition models weretested for all periods and for all initial cover types inthe Hoh Basin (i.e., no data were pooled between pe-riods). For coniferous forest, there were significant dif-ferences between owners in the transition models forall three periods (Table 7b). Identical transition modelsfor all other cover types in 1975-1980 were also re-jected. For 1980-1986, we could not distinguish be-tween private and DNR transition models on mixedforest and grassy cover types. For 1986-l 99 1, we couldnot distinguish between private and DNR transitionmodels for grassy and unvegetated cover types (Table7b).

Effects of spatial variables on transition models.-In contrast to the LTRB, individual spatial variablesprovided little explanation of private cover transitionsin the HORB. Only one of the marginal effects coef-ficients (elevation, Table 9a) was significant at the P5 0.20 level for private lands for 1975-I 980, showingthe expected positive relationship. Only four of 24 mar-ginal effects coefficients were significant at the P 50.05 level for 1986-1991, primarily within the grass-to-forest and grass-to-unvegetated transitions. Popu-lation was not included because the reported values didnot vary across this sparsely populated drainage.

Effects of spatial variables on the DNR lands in theHORB were quite different from those on the privatelands (Table 9b). For the period 1975-1980, three of24 marginal effects coefficients were significant at theP 5 0.05 level (Table 9b); 10 of 24 were significant atthe P 5 0.20 level. Five of 24 were significant for1986-1991 (Table 9b) (eight at P 5 0.20), but severalof these significant factors had signs opposite to ourexpectations. The strongest set of spatial relationshipswas found for transition from coniferous forest tograssy cover in 1975-1980, where all four of the spatialvariables had significant marginal effects. The signsfor the effects of elevation and distance to roads onharvest probability were consistent with our expecta-tions (i.e., that increasing costs are inversely related toharvest probability). However, the signs of the marginaleffects for slope and distance to market were not con-sistent with our expectations, indicating positive re-lationships with harvest probability.

Land ownership and landscape change:Dungeness River Basin

Temporal change in transition models.-In contrastto findings for the HORB, transition models for theDURB were relatively stable. Transition models for theconifer land-cover type did not differ significantly on

1 1 6 6 MONICA G. TURNER ET AL. Eco l o g i c a l App l i c a t i onsV o l . 6 , N o . 4

TA B L E 9 . Tests of hypotheses regarding marginal effects of independent variables on specificland-cover transitions for the Hoh River Basin, by ownership class. For each entry, the firstsign represents the expected sign of the marginal effect on the referenced transition. Thesign in parentheses is the marginal effect calculated at the means of the independent variables.Significance levels indicated are as in Table 8.

Variable

a) Private lands

Forestto grass

Forest Grass Unveg- Unveg-to unveg- Grass to unvegetated etated

etated to forest etated to forest to grass

1975-1980Elevation N SSlope N SDistance to road N SDistance to market N S

1986-1991Elevation N SSlope N SDistance to road N SDistance to market N S

b) Department of Natural Resources

N S N S N S +c+j N SN S N S N S N S N SN S N S N S N S N SN S N S N S N S N S

N S +c+j* N S N SN S +(-I* -“;s) N S N SN S +(+) -(-)* NS N SN S N S -(+)* NS N S

1975-1980ElevationSlopeDistance to roadDistance to market

1986-1991ElevationSlopeDistance to roadDistance to market

1;;;N SN S

ri;;* NSN S

N S -(-)N S N SN S -(+I*N S N S

N S -(-I N S N SN S +(+I N S

N S+y: j*

1:;; * N S -T:)

+c+j* N S N S N SN S N S N S

+(+I* +c+j N S+(-I* * N S N S

either ownership type between any periods (Table 6~).On private lands, some shifts in transition relationshipsfor other cover types were observed, suggestingchanges in lands dedicated to agriculture and other de-veloped uses. On USFS lands, however, the transitionmodels were generally stable, with only unvegetatedcover showing significant change between 19751980and 1980-I 986.

IZflects of ownership on transition moctels.-Al-though t ransi t ion re la t ionships were more stablethrough time in the DURB than in the HORB, differ-ences between ownership types were pronounced (Ta-ble 7~). With the exception of coniferous forests in1975-1980, all transition models for public and privatelands in all periods were significantly different. As inthe HORB and LTRB, there were structural dissimi-larities in land-cover dynamics between public and pri-vate lands.

Effects of sputial variables on trunsition models.-On private lands in the DURB, the significant temporalchanges in transition models were reflected in the ef-fects of the spatial variables. For the period 197% 1980(Table IOa), no spatial variables influenced transitionsfrom coniferous forest cover. However, for the period1986-l 99 I, nine of 10 marginal effects coefficients forthese transitions were significant. Population densityhad the strongest influence in the transition models onprivate lands. For the period 19751980, three of thesix transition models displayed in Table IOa were sig-

nificantly influenced by population density at the P “=0.20 level, and two were significant at the P Cr 0.05level. For the period 1986-1991, all six models indi-cated significant effects of population on transitionprobabilities.

Population density also had a significant influenceon USFS transitions (Table lob), indicating that, forexample, timber harvesting was less likely where, cet-eris paribus, population density was higher. In general,spatial factors had much more influence on USFS tran-sition probabilities in the DURB than in the LTRB.While no spatial variable yielded a significant marginaleffect on USFS lands in the LTRB, 11 of the 40 mar-ginal effects (Table lob) were significant at the P 50.05 level, and 22 were significant at the P 5 0.20level. There were differences in the marginal effectson transitions from forest cover between periods, es-pecially in the effects of slope: both were insignificantfor 19751980 but negative for 1986-1991.

D I S C U S S I O N

Land ownership and landscape pattern

Land ownership clearly influenced landscape pattern,despite differences between the two study regions. Pri-vate lands contained less forest cover but a greaternumber of small forest patches than did public lands,indicating greater forest fragmentation. Lands that wereactively managed for timber harvest, however, showed


TABLE 10. Tests of hypotheses regarding marginal effects of independent variables on specificland-cover transitions for the Dungeness River Basin, by ownership class. For each entry, thefirst sign represents the expected sign of the marginal effect on the referenced transition. Thesign in parentheses is the marginal effect calculated at the means of the independent variables.Significance levels indicated are as in Table 8.

Variable

Unveg- Unveg-Forest to Forest to Grass to Grass to etated etated

grass unvegetated forest unvegetated to forest to grass

a) Private lands

1975-1980ElevationSlopeDistance to roadDistance to marketPopulation


b) U.S. Forest Service



NSNSNSNSNS

-(+)*.-(-)*-c-j*-C-F+(+I*

-(-)-7:)

NSNS

*NSNS

NSNSNSNSNS

-(+)NSNSNSNS

-(-I-c-j*-(+)-c-1*-c-j*

+(+)+(-I*-+-)+(-I*-c-1*

+c-)*+(+)*+(--I‘VP-c-1*

NS

cc+)NS

+(-I*-c-j*

N SN SN SN SN S

+(+)*+(+It-c+)--‘;t,*

NSNSNS

-7)

+(+I

+“;‘,*NS

-c-j*

NSNSNS

-‘;s,

NSNSNS

-“;“)

+(-)NSNS

+(+)+(+I*

little difference in landscape pattern between owner-ships (e.g., private and DNR lands in the HORB). Dif-ferences in landscape structure between ownerships re-flect the management and land-use decisions of theowners. When owners share similar objectives (e.g.,commercial forests and state DNR both emphasize tim-ber production), then landscape patterns may be sim-ilar.

Differences in landscape pattern observed betweenthe two regions probably reflect differences in the im-portance of forest harvesting. Forest cover was greatestin the LTRB on both ownership classes as comparedto any land-ownership category in the HORB andDURB. Although the LTRB was almost completely cutover during the early decades of the 20th century, tim-ber extraction is no longer dominant. Private lands inthe LTRB were characterized by only 10% less forestcover than USFS lands, and connectivity of the foresttbroughout the watershed was high. Small forest patch-es occurred primarily near the town of Franklin, andnot in the higher, less accessible areas.

Greater variability in landscape structure throughtime and between ownership categories was observedon the Olympic Peninsula. The proportion of the own-ership in coniferous forest followed a consistent rankorder: USFS > DNR > HORB private > DURB pri-

vate. In contrast, the proportion of the landscape inunvegetated cover was greatest in DURB private landsand lowest in USFS lands; private commercial andDNR lands had comparable intermediate levels of un-vegetated cover. These trends reflect land use withinthe watersheds. The DNR and private commercial landsin the HORB are both managed for timber production;spatial patterns differed between these ownerships pri-marily due to the relative abundance of coniferous(higher on DNR lands) vs. deciduous/mixed forest cov-er (higher on private commercial lands). Patterns onthe small private ownerships that occurred in the DURBreflected rapid development in and around the town ofSequim, resulting in substantial increases in unvege-tated cover. In contrast, USFS lands in the DURBshowed relatively small changes through time, al-though changes were greater than those observed inthe LTRB.

Land ownership und lundscape change

Temporal change in transition models.-The anal-yses of landscape change revealed differences in land-cover change through time. In the LTRB, the rate offorest transition decreased between the 1975-1986 pe-riod and the 1986-1991 period for both the USFS andprivate ownership classes. In the HORB, transition


models were also dynamic in time for both ownershiptypes. These results suggest the need to account foradditional factors that cause individuals or institutionsto change land management strategies. For example,shifts in the LTRB models may reflect a transition fromland use focused on forest management to one focusedon expanding residential development, along with anemphasis on public lands of in situ value (e.g., waterquality maintenance and biodiversity). In the HORB,where active timber management is the dominant use,land-cover transitions may be strongly influenced bytimber prices.

Temporal stability in the transition relationships forconiferous forest on USFS lands in the DURB wassomewhat surprising, given the dynamics of woodproducts markets during this period. Between 1975 and1980, timber markets were especially strong, with tim-ber prices for West Coast species increasing at un-precedented rates (e.g., Mattey 1990). In contrast,stumpage prices dropped substantially in the early1980s and then recovered beginning in 1986. For ex-ample, the average price of Douglas-fir sawtimber soldfrom the national forests in Washington and Oregonpeaked in 1980 at $432/mbf (thousand board feet), andhad fallen to $11 B/mbf by 1982. Prices remained lowthrough the early 1980s but began to climb in 1986.By 1991, Douglas-fir prices were $395/mbf (Warren1992). Our focus on the details of a smaller area forthree periods allows us to examine spatially variablefactors, but does not support direct analysis of the ef-fects of prices and other temporally variable factors. Amore frequent, perhaps annual, sampling over a largerarea could allow a direct analysis of price effects onharvest behavior.

Harvesting behavior and landscape dynamics on pri-vate forest lands in the Dungeness apparently were notsubstantially intluenced by these strong market forces,and remained relatively stable. In contrast, forest-coverdynamics changed substantially between these periodson the large commercial private lands in the HORB.In this situation, land-cover dynamics on commercialprivate lands were more volatile than dynamics onsmall private holdings. Transition models were alsostable between periods on USFS lands in the Dunge-ness. However, the scale of our analysis may not besufficiently large to accurately address the USFS re-sponse to markets. That is, timber harvesting may shiftamong drainages over time, so that the better unit ofobservation may be an entire national forest.

Effects of ownership on transition modeZs.-Differ-ences between ownerships in the models of forest-coverchange were observed in all three watersheds. In theLTRB, little commercial forestry is practiced on privatelands, but residential development has increased duringthe study period. The HORB is dominated by forestryuses, and significant differences in transition modelsfor coniferous forest cover probably reflect differencesin the forestry practiced by the Washington DNR and

the commercial private owners. The similarity of tran-sition models for grassy and unvegetated cover typesin the HORB may reAect similar patterns of forest standregeneration and regrowth. In the DURB, coniferousforest transition models did not differ between theUSFS and private lands in the 1975-1980 period, butal1 other models differed between public and privatelands. As in the LTRB, this difference probably reflectsdifferent influences on forestry practices, or multiple-use management on the USFS lands and increased res-idential development on the private lands.

.

Effects of spatial variables on transition models.-The importance of independent variables in explainingland-cover change generally varied between owner-ships in each watershed. In the LTRB, spatia1 variablesdid influence land-cover change on private lands, al-though effects were stronger during the 1975-1986 pe-riod than in the 1986-1991 period. Most of the mar-ginal effects were in the hypothesized directions (Table2). However, the positive relationship between eleva-tion and grassy-to-unvegetated transitions on privatelands in the LTRB was counter to the hypothesizedrelationship. Increasing development at higher eleva-tions is inconsistent with the hypothesis derived froman argument based on cost. This finding may reflectpreferences for scenic views, which would encourageresidential development at higher elevations, consistentwith anecdotal observations on recent developments inthe Southern Appalachians. Overall, land-coverchanges on private lands in the LTRB were consistentwith a shift from land use focused on forest manage-ment to land use focused on expanding residential de-velopment. Indeed, population has grown steadily inthe LTRB, and timber production has declined. In sum,a structural shift in the pattern of disturbance was in-dicated on private lands, and forest disturbance on pri-vate Iands was more strongly influenced by locationrelative to the road network than by other site factors,such as elevation and slope.

On USFS lands in the LTRB, no spatial variable hada significant marginal effect on any transition proba-bility, and the average probability of change appliedacross the USFS lands was the best predictor of change.Apparently, rules that are not correlated with the spatialvariables defined here have guided the management ofUSFS lands during the study periods, although privateowners were strongly influenced by cost factors asso-ciated with development, timber harvest, or transpor-tation. Results for USFS lands may be consistent withmultiple-use management that mitigates the negativeeffects of timber sales on wildlife habitat and scenicviews by spreading harvest activities over broad areas.

In the HORB, individual spatial variables providedlittle explanation of land-cover transitions on privatelands, but did explain cover transitions on DNR lands.The lack of effects on private lands in the HORB sug-gests that factors other than those represented by thespatial variables measured here explain the probability

November 1996 OWNERSHIP AND LAND-COVER CHANGE 1 1 6 9

of harvesting timber. It may be that, in an area such asthe Olympic Peninsula, where both timber volume peracre (per 0.405 ha) and timber values are very high,variable costs of logging and transport have relativelylittle impact on harvesting decisions. That is, the highvalue of timber may make the spatially variable costsof timber extraction unimportant for timbering deci-sions. Rather, the harvesting plan may be more sen-sitive to temporal change in the relative prices of woodproducts, optimal depletion schedules, or forestry pol-icies.

Although the DNR lands are also managed for timberproduction in the HORB, the sign of the effects of slopeand distance on harvest probability was not consistentwith our expectations. The difference in the relation-ship between slope and distance to market for conif-erous forest transitions on DNR lands in the HORBmight be consistent with the aggregation of harvestunits. That is, as the basin has been developed for tim-ber production, initial harvests may have been con-ducted on level and accessible sites. Accordingly, largeopenings in these areas may constrain further harvest,leading to a subsequent bias towards more remote andsteeper sites. The availability of harvestable timbermay constrain timbering choices, with the less acces-sible lands being harvested while the more accessiblelands regenerate from past harvest, but this issue clear-ly needs additional investigation.

The findings for these large ownerships in the HORBagain raise the issue of the appropriate scale of analysisfor the types of questions addressed here. Unlike smallprivate landowners, whose holdings are focused withina small area, large firms or public institutions may re-spond to regional, national, or international factors. Forexample, a timber corporation may alter harvest plansin response to capital requirements for milling facili-ties. A large government agency may focus on even-How harvesting from a broad region. This suggests thatlocal conditions may hold less influence over the land-use choices of larger owners.

In many ways, the social settings of the DURB andLTRB are very similar. Both areas have experiencedpopulation growth and expansion in residential devel-opment in recent years. However, public and privatelands are much less intermingled in the DURB, andprivate lands tend to be concentrated in areas that areless steep and less remote than public lands. The dif-ferences between ownerships observed in the LTRBwere not found in the DURB. No significant relation-ships between spatial variables and land-cover changeon USFS lands were observed in the LTRB, but severalspatial variables had a significant influence on forest-cover transitions on USFS lands in the DURB.

For private lands in the DURB, no spatial variableinfluenced transitions between 1975 and 1980, butmany had significant effects during the period 1986-1991. One possible explanation relates to differencesin timber markets. That is, when prices are temporarily

high, marginal cost factors become less critical as forestowners attempt to capture ephemeral revenues. Anotherpotential explanation is that the basin has become moreinfluenced by population growth pressures and less bytimber harvesting on private lands. This is supportedby the significant effects of population on nine of the12 transition relationships displayed in Table 10.

Conclusions

The analysis presented here demonstrated that dif-ferent broad ownership groups produce qualitativelydistinct landscape patterns. Furthermore, it demonstrat-ed that different types of owners interact with similarlands in distinct ways (Table 7). The way in whichhuman endeavors are organized through the institutionsand scale of land ownership significantly influences thedynamics of land cover. Land ownership produces dis-tinct signatures on landscapes, creating patterns that,in turn, will influence a variety of ecological processes.Thus, understanding and predicting land cover requiresknowledge about land ownership. Purely biophysicalmodels will provide limited insight into land-cover dy-namics, as some explanatory variables are likely to besocioeconomic and political (Lee et al. 1992, Machlis1992). There remains a tremendous need for work thatintegrates ecological and socioeconomic dynamics atlandscape scales.

We know of only one other study in which landscapepattern and rates of change in forest cover were eval-uated as a function of land ownership: Spies et al.(1994) examined an area including part of the Willam-ette National Forest, Oregon, from 1972 to 1988, sim-ilar to the 1975-1991 period of our study. Ownershippatterns in the Willamette study area were most similarto those in the HORB in this study. Public land-own-ership classes occupied -70% of the study area andincluded USFS, Bureau of Land Management, and theState of Oregon. Private lands consisted primarily ofindustrial land ownerships. As in our three watersheds,Spies et al. (1994) also reported that a greater propor-tion of forest cover in the Willamette study area oc-curred from 1984 to 1988 on public lands and from1981 to 1984 on private lands. In the HORB, publicand private lands both experienced the most rapid lossof coniferous forest cover between 1980 and 1986 (Ta-ble 4), similar to the Willamette. In contrast to theWillamette study area, however, coniferous forest coveron the public lands in HORB increased subsequently.Greater amounts of edge on private vs. public landswere reported for both the Willamette and HORB studyareas. Although we did not compute interior forest hab-itat, the 8 to 16-fold difference in weighted-averagepatch size and lower edge-to-area ratio for coniferousforest on public vs. private lands in the HORB (Table4) is consistent with a greater abundance of interiorhabitat on public lands, as reported by Spies et al.(1994). The amount of interior coniferous forest in theHORB (based on weighted-average patch siz.e and


edge-to-area ratio) probably reached a low in 1986 onboth public and private lands, and then increased be-tween 1986 and 1991.

Patterns of change in coniferous forest were similarat low and high elevations in the Willamette study area(Spies et al. 1994), consistent with the lack of a sig-nificant effect of elevation for private lands in theHORB during any time period. However, on DNR landsin the HORB, we observed a negative relationship be-tween elevation and forest transition to grassy coverduring the 1975-1980 period, and to unvegetated coverduring the 1980-1986 period, suggesting that forest-cutting rates were higher at lower elevations. In sum,forest-cover patterns on public and private lands in theWillamette study area and HORB were generally sim-ilar, but the Willamette did not show the increasingtrend in coniferous forest cover observed during thelattermost time period in the HORB.

Spatially referenced physical (slope, elevation) andcultural (distance to roads and markets, as well as pop-ulation density) features had measurable influencesover the probability of land-cover change on publicand private lands. Transition models for all river basins,ownerships, and cover types provided significant ex-planation of observed transitions. Thus, spatially ref-erenced data, in comparison to simple averages, canimprove the estimation of transition probabilities. Spa-tially explicit estimates of land-cover change can in-dicate to land managers what portions of the landscapemay be most subject to rapid change. For example, thestatistical models developed in this study can be ex-trapolated spatially across the landscape to map theprobability of any land-cover change as a function ofthe attributes of each grid cell (e.g., Wear and Flamm1993). Such maps can provide a graphical summary ofwhere either desirable or undesirable land-cover changesare most likely to occur.

Transition probabilities generally were not stablethrough time, suggesting that simple Markovian mod-els of land-cover change are not likely to representfuture landscape conditions in anthropogenic land-scapes. Rather, transition probabilities are likely to varythrough time, as the responses of individuals and in-stitutions to social and economic conditions change.Shifts in conditions (e.g., timber prices), trends in rec-reational preferences in the population, and variablerates of residential development may all lead, individ-ually or in concert, to substantial shifts in rates of land-cover transition.

The length of time over which the imprint of land-ownership patterns will remain on the landscape is notknown. Wallin et al. (1994) demonstrated that land-scape patterns created by dispersed disturbances aredifficult to erase, and time lags may be considerable,even with substantial reductions in disturbance rates.Thus, the imprint of current land-ownership patternson the landscape is likely to persist for some time, evenif land ownership changes. As noted by Spies et al.

(I 994), existing conditions are important for the designof future landscape patterns geared toward maintenanceof particular species or ecosystem functions. Existingpatterns will constrain future conditions for some time,and managers will continue to face the challenge ofintegrating present patterns with desired future con-ditions.

The results obtained in this study were affected bythe spatial scale of the data and the land-cover cate-gories selected for analysis. Direct comparisons of thenumerical results (e.g., landscape metrics or estimatedcoefficients) with other studies must be done with care.Measures of landscape pattern are strongly influencedby both the grain (e.g., spatial resolution, or grid-cellsize) and extent (total area considered) of the data (e.g.,Allen and Starr 1982, Turner et al. 1989). In addition,selection of the categories used in the analysis con-strains the results. For example, stand age was not in-cluded in this study as a modifier of forest cover; there-fore, the results reported here do not distinguish be-tween old- and secondary-growth forest cover. In ad-dition, the use of land-cover classes based on canopycharacteristics cannot be used to infer land use in theabsence of additional data sources. For example, low-density residential development that does not result incanopy breakup is not likely to be detected.

Extrapolation of these analyses to other locations canbe considered in two ways. First, it is of interest todetermine whether or not the qualitative differencesbetween ownership classes observed in this study areapplicable to other river basins within the same regions(i.e., southern Appalachian highlands and OlympicPeninsula), or perhaps even to other river basins inforested landscapes. The results for the LTRB, HORB,and DURB should be compared with other river basinsto search for generalities that might be broadly appli-cable. Second, the methodology demonstrated herecould be applied in other systems. Data availability isoften the primary limiting factor, but this constraint isdiminishing rapidly with the widespread developmentof GIS databases for many regions.

The strong inlluence of land ownership on both land-scape pattern and land-cover change has important im-plications for the future landscape mosaic. Ownershipclass must be considered when potential changes withina river basin or landscape are predicted. If transitionprobabilities are to be used to simulate future condi-tions (e.g., Flamm and Turner l994a, b), separate mod-els should be developed for different ownership class-es. The transition models developed in this study mayprove especially useful in a simulation framework thatcan forecast the effects of various ownership scenarios.In such an exercise, the potential implications of, e.g.,changes in USFS policy, can be examined at a whole-landscape scale. For example, the models describedhere for the LTRB were applied in a factorial simulationexperiment to project both the effects of extrapolatingobserved rates of change into the future, and of im-


posing some constraints , or rule changes, on futuretransitions (Wear et al. 1996).

Land ownership may have strong effects on numer-ous ecological processes because the intluence of own-ership on landscape pattern is so strong. Natural dis-turbance dynamics (e .g. , Turner 1987, Roland 1993),s tabi l i ty of the vegetat ion mosaic (e .g. , Turner et ai.1993), persistence of species (e.g. , Lamberson et al .1992, Pulliam et a l . 1992, Turner e t a l . 1994), andquality of water resources (e.g. , Peterjohn and Correll1984, Kesner and Meentemeyer 1989) can all be af-fected by landscape structure. Indeed, this study couldbe expanded to identify l inkages with addit ional se-lected ecological end points, such as species persis-tence or water quali ty. Our results suggest that Iand-ownership effects must be considered explicitly whenthe future vital i ty of ecological systems is considered.If paradigms such as sustainability and ecosystem man-agement are truly to inform land management, effortsto seek generali ty and to understand and predict thedynamics of mixed-ownership landscapes must contin-ue.

ACKNOWLEDGMENTSThis study especially benefited from discussions with Rob-

ert J. Naiman, Robert G. Lee, and Robin Gottfried, and themanuscript was improved by comments from Robert G. Lee,Robert V. O’Neill, Robert J. Naimnn, Scott M. Pearson, andtwo anonymous reviewers. We thank Michae) Daly and DavidLeversee for interpreting the remote imagery, and Penny Eck-ert for technical assistance. This work was supported by agrant from the U.S. Man and Biosphere Program; the ap-pointment of R. 0. Flamm to the U.S. Department of EnergyLaboratory Cooperative Postgraduate Research Training Pro-gram administered by Oak Ridge Institute for Science andEducation; and by the Ecological Research Division, Ofhceof Health and Environmental Research, U.S. Department ofEnergy, under Contract No. DE-AC05840R21400 with Mar-tin Marietta Energy Systems. Publication No. 4432, Envi-ronmental Sciences Division, Oak Ridge National Lahora-tory.

LITERATURE CITEDAgee. J. K., and D. R. Johnson, editors. 1988. Ecosystem

management for parks and wilderness. University of Wash-ington Press, Seattle, Washington, USA.

Allen, T. E H., and T. B. Starr. 1982. Hierarchy. Universityof Chicago Press, Chicago, Illinois, USA.

Baker, W. L. 1989. A review of models of landscape change.Landscape Ecology 2:1 11-133.

Cohen, W. B., and T. A. Spies. 1992. Estimating attributesof Douglas-fir/western hemlock forest stands from satelliteimagery. Remote Sensing of the Environment 4: l-1 8.

Crist, E. P, and R. C. Cicone. 1984. A physically-basedtransformation of the Thematic Mapper: the TM Tasseledcap. IEEE Transactions on Geoscience and Remote SensingGE 22(3):256-263.

Dale, V. H., R. V. O’Neill, M. Pedlowski, and E Southworth.1993. Causes and effects of land-use change in centralRondonia, Brazil. Photogrammetric Engineering and Re-mote Sensing 59:997-1005.

Eby. J. R. 1987. The use of sun incidence angle and infraredreflectance levels in mapping old-growth coniferous forest,Pages 36-44 in Proceedings of the American Society forPhotogrammetry and Remote Sensing, Rem), Nevada,IJSA.

ERDAS. 1994. ERDAS field guide. Third edition. ERDAS,Atlanta, Georgia, USA.

Flamm, R. O., and M. G. Turner. 1994~. Alternative modelformulations of a stochastic model of landscape change.Landscape Ecology 9:47-46.

Flamm, R. O., and M. G. Turner. 1994b. Multidisciplinarymodeling and GIS for landscape management. Pages 201-2 12 in V. A. Sample, editor. Forest ecosystem managementat the landscape level: the role of remote sensing and in-tegrated GIS in resource management planning, analysis,and decision making. Island Press, Covelo, California,USA.

Greene, W. H. 1992. LINDMEP: user’s manual and referenceguide. Version 6.0. Econometric Software, Inc., Bellport,NY, USA.

Grumbine, R. E. 1994. What is ecosystem management?Conservation Biology 8:27-38.

Hail, E G., D. B. Botkin, D. E. Strebei, K. D. Woods, andS. J. Goetz. 1991. Large-scale patterns of forest successionas determined by remote sensing. Ecology 72:628-640.

Iverson, L. R. 1988. Land-use changes in Illinois, USA: theinfluence of landscape attributes on current and historicland use. Landscape Ecology 2:45-62.

Johnson, W. C., and D. M. Sharpe. 1976. An analysis offorest dynamics in the north Georgia Piedmont. ForestryScience 221307-322.

Judge, G. G., W. E. Griftiths, R. C. Hill, W. E. Lutkepolh,and T. C. Lee. 1985. The theory and practice of econo-metrics. John Wiley, New York, New York, USA.

Katzman, M. T 1974. The Von Thunen paradigm, the in-dustrial-urban hypothesis, and the spatial structure of ag-riculture. American Journal of Agricultural Economics 56:683-696.

Kesner, B. T., and V. B. Meentemeyer. 1989. A regionalanalysis of total nitrogen in an agricultural landscape.Landscape Ecology 2: 15 I-1 64.

Kienast, E 1993. Analysis of historic landscape patterns witha Geographic Information System: a methodological out-line. Landscape Ecology 8: 103-l 18.

LaGro, J. A., Jr., and S. D. DeGloria. 1992. Land-use dy-namics within an urbanizing nonmetropolitan county inNew York State (USA). Landscape Ecology 7:275-289.

Lamberson, R. H., K. McKelvey, B. R. Noon, and C. Voss.1992. A dynamic analysis of Northern Spotted Owl via-bility in a fragmented forest landscape. Conservation Bi-ology 6:505-5 12.

Lee, R. G., R. 0. Flamm, M. G. Turner, C. Bledsoe, P Chan-dler, C. DeFerrari, R. Gottfried, R. J. Naiman, N. Schu-maker, and D. Wear. 1992. Integrating sustainable devel-opment and environmental vitality. Pages 499-521 in R.J. Naiman, editor. New perspectives in watershed manage-ment. Springer-Verlag, New York, New York, USA.

Li, H., and J. E Reynolds. 1993. A new contagion index toquantify spatial patterns of landscapes. Landscape Ecology8:155-162.

Lubchenco, J., A. M. Olson, L. B. Brubaker, S. R. Carpenter,M. M. Holland, S. P Hubbell, S. A. Levin, J. A. McMahon,I? A. Matson, J. M. Mellillo, H. A. Mooney, C. H. Peterson,H. R. Pulliam, L. A. Real, P J. Regal, and P. G. Risser.199 1. The Sustainable Biosphere Initiative: an ecologicalresearch agenda. Ecology 72:37 I-412.

Machlis, G. E. 1992. The contribution of sociology to bio-diversity research and management. Biological Conserva-tion 62:161-170.

Maddala, G. S. 1983. Limited dependent and qualitative vari-ables in econometrics. Cambi-idge University Press, NewYork, New York, USA.

Mattey, J. P. 1990. The timber bubble that burst: governmentpolicy and the bailout of 1984. Oxford University Press,New York.


Mladenoff, D. J., M. A. White, J. Pastor, and T. R. Crow.1993. Comparing spatial pattern in unaltered old-growthand disturbed forest landscapes. Ecological Applications 3:294-306.

Morrison, P M. 1992. Ancient forests of the Pacific North-west. The Wilderness Society, Washington, D.C, USA.

O’Neill, R. V., J. R. Krummel, R. H. Gardner, Cl. Sugihara,B. Jackson, D. L. DeAngelis, B. T Milne, M. Cl. Turner,B. Zygmunt, S. Christensen, V. H. Dale, and R. L Graham.1988. Indices of landscape pattern. Landscape Ecology 1:153-162.

Overbay, J . C. 1992. Ecosystem management . Nat ionalWorkshop on Taking an Ecological Approach to Manage-ment, Salt Lake City, Utah. USDA Forest Service, Wash-ington, D.C., USA.

Parks, P J. 1991. Models of forested and agricultural land-scapes: integrating economics. Pages 309-322 in M. G.Turner and R. H. Gardner, editors. Quantitative methods inlandscape ecology. Springer-Verlag. New York, New York,USA.

Parks, P J., and R. J. Alig. 1988. Land-base models for forestresource supply anaIysis: a critical review. Canadian Jour-nal of Forest Research 18:965-973.

Peterjohn, W. T., and D. L. Correll. 1984. Nutrient dynamicsin an agricultural watershed: observations on the role of ariparian forest. Ecology 65:1466-1475.

Pulliam, H. R., J. B. Dunning, and J. Liu. 1992. Populationdynamics in complex landscapes: a case study. EcologicalApplications 2:165-177.

Roland, J. 1993. Large-scale forest fragmentation increasesthe duration of tent caterpillar outbreak. Oecologia 93:25-30.

Slocombe, D. S. 1993. Implementing ecosystem-based man-agement. Bioscience 43:612-622.

Spies, T. A., W. J. Ripple, and G. A. Bradshaw. 1994. Dy-namics and pattern of a managed coniferous forest land-scape in Oregon. Ecological Applications 4:555-568.

Turner, M. G., editor. 1987. Landscape heterogeneity anddisturbance. Springer-Verlag, New York, New York, USA.

-. 1990~. Landscape changes in nine rural counties inGeorgia, USA. Photogrammetric Engineering and RemoteSensing 56:379-386.

----. 1990b. Spatial and temporal analysis of landscapepatterns. Landscape Ecology 4:21-30.

Turner, M. G., R. V. O’Neill, R. H. Gardner, and B. T. Milne.1989. Effects of changing spatial scale on the analysis oflandscape pattern. Landscape Ecology 3:153-162.

Turner, M. G., W. H. Romme, R. H. Gardner, R. V. O’Neill,and T. K. Kratz. 1993. A revised concept of landscapeequilibrium: disturbance and stability on scaled landscapes.Landscape Ecology 8: 2 13-227.

Turner, M. G., and C. L. Ruscher. 1988. Changes in landscapepatterns in Georgia, USA. Landscape Ecology 1:241-251.

Turner, M. G., Y. Wu, W. H. Romme, L. L. Wallace, and A.Brenkert. 1994. Simulating winter interactions betweenungulates, vegetation, and fire in northern YellowstonePark. Ecological Applications 4:472-496.

USA CERL. 1991. Geographic resources analysis supportsystem, version 4.0. U.S. Army Corps of Engineers, Con-struction Engineering Research Laboratory, Champaign, II-linois, USA.

Wallin, D. O., E J. Swanson, and B. Marks. 1994. Landscapepattern response to changes in pattern generation rules:land-use legacies in forestry. Ecological Applications 4:569-580.

Warren, D. D. 1992. Production, prices, employment, andtrade in Northwest forest industries, fourth quarter 1991.USDA Forest Service, Resource Bulletin PNW RB-192,Portland, Oregon, USA.

Wear, D. N., and R. 0. Flamm. 1993. Public and privatedisturbance regimes in the Southern Appalachians. NaturalResource Modeling 7~379-397.

Wear, D. N., M. G. Turner, and R. 0. Flamm. 1996. Eco-system management with multiple owners: landscape dy-namics in a southern Appalachian watershed. EcologicalApplications 6: 1 173-1 188.

Whitney, G. T., and W. J. Somerlot. 1985. A case study ofwoodland continuity and change in the American midwest.Biological Conservation 31:265-287.

.

Land Ownership and Land-Cover Change in the Southern Appalachian Highlands and the Olympic Peninsula

Documents

Transcript of Land Ownership and Land-Cover Change in the Southern Appalachian Highlands and the Olympic Peninsula