Anticipating Forest and Range Land Development in CentralOregon (USA) for Landscape Analysis with an ExampleApplication Involving Mule Deer

Jeffrey D Klinemiddot Alissa Mosesmiddot Theresa Burcsu

period to 2040 The example application illustrates a sim-ple way for policy makers and public lands managers tocombine existing data and preliminary model outputs tobegin to consider the potential effects of development onfuture landscape conditions

Keywords Spatial land-use modelsmiddot Landscape changemiddotWildland-urban interface Mule deer

Introduction

Forest policymakers public lands managers and scientistsin the Pacific Northwest (USA) seek ways to evaluatethe landscape-level effects of policies and managementthrough the multidisciplinary development and applicationof spatially explicit analytical methods and models (egBarbour and others 2007 Spies and others 2007) Policy-makers and managers desire ways to display and predictsocioeconomic and ecological outcomes of policy andmanagement alternatives on public and private lands Sci-entists seek ways to work across disciplines to examineinteractions among socioeconomic and ecological phe-nomenon that occur at different temporal and spatial scalesDespite ambitions conducting multidisciplinary landscapeanalysis in a cost-effective and timely manner sufficient tobe of practical use to policy makers and managers is apersistent challenge Multidisciplinary landscape studiescan be costly to initiate and sustain They can involvepainstaking effort on the part of collaborating scientists toagree upon research objectives appropriate spatial andtemporal scales procure data develop and integrate spa-tially explicit methods and models and then deliver resultsto the policy makers and managers who presumably can usethem (Kline and others 2010) The observers quip that

Abstract Forest policymakers public lands managersand scientists in the Pacific Northwest (USA) seek ways toevaluate the landscape-level effects of policies and man-agement through the multidisciplinary development andapplication of spatially explicit methods and models TheInteragency Mapping and Analysis Project (IMAP) is anongoing effort to generate landscape-wide vegetation dataand models to evaluate the integrated effects of distur-bances and management activities on natural resourceconditions in Oregon and Washington (USA) In this initialanalysis we characterized the spatial distribution of forestand range land development in a four-county pilot studyregion in central Oregon The empirical model describesthe spatial distribution of buildings and new buildingconstruction as a function of population growth existingdevelopment topography land-use zoning and other fac-tors We used the model to create geographic informationsystem maps of likely future development based on humanpopulation projections to inform complementary landscapeanalyses underway involving vegetation habitat andwildfire interactions In an example application we use themodel and resulting maps to show the potential impacts offuture forest and range land development on mule deer(Odocoileus hemionus) winter range Results indicate sig-nificant development encroachment and habitat lossalready in 2000 with development located along keymigration routes and increasing through the projection

With a crayon and a sheet of paper I could have done inan afternoon what they did with $5 million in 10 yearssuggests the degree to which perceptions of the practice ofmultidisciplinary landscape analysis can be improved

In the Pacific Northwest landscape studies typically aredesigned to address the long-term effects of policy andmanagement on vegetation wildfire and terrestrial andaquatic habitat while accounting for exogenous socioeco-nomic changes that affect landscapes Spies and others(2007) for example sought to examine the effects of forestpolicies intended to protect spotted owls (Strix occidentaliscaurina) and coho salmon (Oncorhynchus kisutch) andpredict future outcomes for the western Oregon CoastRange Spanning 1996 to 2005 and involving landscapeecologists aquatic and wildlife biologists hydrologistseconomists and several geographic information systemsanalysts and research assistants the project resulted in highresolution (30-meter pixel) spatial models of biophysicalconditions (eg vegetation topography streams) andprojected future conditions over 100 years Several linkedmodels addressed habitat suitability for select terrestrialand aquatic species landslide and debris flow and geo-morphic dynamics and other factors In another exampleBarbour and others (2007) aspired to develop a morestreamlined approach to landscape analysis that wouldenable public land managers to evaluate interactionsbetween management forest succession and wildfire innortheastern Oregons Blue Mountains Spanning fiveyears and involving up to 20 researchers the project pro-duced a coarser scaled vegetation modeling system linkedwhere feasible to other resource models describing habitatquality insect activity ungulate grazing timber manage-ment and wood utilization Management outcomes wereprojected over 100- and 200-year horizons

Although such efforts produce data and information ofinterest to policymakers and managers they often fail tofulfill expectations that resulting models and model outputswill be immediately useful in policy and managementdecision-making Multidisciplinary landscape analyses tendto progress slowly Difficulties and delays arise fromincomplete data and the need to adapt and develop spatialmethods and models Difficulties developing one studycomponent can delay other component applications BothSpies and others (2007) and Barbour and others (2007) forexample expended significant time developing vegetationsimulation methods even as other study components nearedcompletion Once complete landscape models often are toocomplex or cumbersome to be accessible to the policy-makers and managers expected to use model outputs Thepractice of spatially-explicit multidisciplinary landscapeanalysis is still evolving and so slowness and complexity asdefining characteristics arguably are par for the courseUntil practices advance analysts might best satisfy policy

and management expectations through early and earnesttechnical transfer efforts that address vital resource con-cerns using study components most readily available andapplicable to the task Additional and more comprehensivemodel applications can follow iteratively as other studycomponents are completed Policy makers and managersoften are more than willing to overlook imperfections innew information when any new information is especiallytimely

Following this approach we report on a land use modelapplication developed as part of an ongoing multidisci-plinary landscape study in Oregon In this initial analysiswe characterize the spatial distribution past and potentialfuture forest and range land development in a four-countyvicinity of Bend Oregon Rapid development there is aprimary concern of State resource policymakers and man-agers for its potential impact on resource industries such asforestry and agriculture (eg Lettman 2004) and alsodeclining habitat for mule deer (Odocoileus hemionus)(eg Oregon Department of Fish and Wildlife 2009b) Theempirical model describes the spatial distribution ofbuildings and new building construction as a function ofpopulation growth existing development topographyland-use zoning and other factors We used the empiricalmodels to create geographic information system maps ofpotential future forest and range land development basedon published human population projections Maps of futuredevelopment eventually will be used to inform comple-mentary landscape study components addressing vegeta-tion habitat and wildfire interactions-all of which arestill in preparation In this preliminary application we usemaps to show the degree to which future forest and rangeland development might reduce mule deer winter range infuture years

Study Region

Our land use modeling is part of a pilot landscape analysisof the Interagency Mapping and Analysis Project (IMAP)IMAP is a partnership of federal and state agencies andnon-government organizations whose goal is to generatelandscape-wide vegetation data landscape models andrelated information with which to evaluate the integratedeffects of natural disturbances and management activitieson natural resource conditions in the Pacific Northwest(Kline and others 2010) Key concerns of policymakers andmanagers involved in IMAP are reducing wildfire risksmaintaining and enhancing wildlife habitats and main-taining timber outputs despite ongoing socioeconomicchange Plans are for IMAP methods and models toeventually include all of Oregon and Washington How-ever current efforts focus on a 275187 ha pilot study area

include loss of forest and range land to developmentincreased traffic congestion increased water demand andpotential adverse habitat effects for some species Of spe-cial concern is the desire to maintain viable mule deerpopulations sufficient to permit continued hunting Muledeer have declined across the western US as a result ofhabitat loss and other factors Mule deer winter range oftencoincides with new development (eg Stein and others200712) In mountainous forest areas new developmenttends to be located on relatively flatter and lower elevationvalley bottoms where for mule deer the relative absence ofpersistent snow cover makes winter movement easier andfood more abundant In eastern Oregon (east of the crest ofthe Cascades) declining mule deer populations owing todevelopment and habitat loss is the single most pressingissue for the Oregon Department of Fish and Wildlife andhunting enthusiasts (Oregon Department of Fish andWildlife 200) Potential disruption of migration patternsowing to increased development density along key routes isof primary concern Mule deer hunting exceeds 74000participants and generates $22 million in economic activityannually (Oregon Department of Fish and Wildlife 2009a)The sale of hunting licenses is the major source of fundingfor the Oregon Department of Fish and Wildlife

Methods

Building Count Data

Land-use data describing historical building counts werecompiled by the Oregon Department of Forestry and theUSDA Forest Service Pacific Northwest Research Sta-tions Forest Inventory and Analysis program The datawere designed to examine historical forest and range landdevelopment and evaluate wildfire risks to homes andother issues of concern The data consist of aerial photoobservations of building (or structure) counts-the number

west of Bend Oregon consisting of 216103 ha of federalforest reserves and wilderness and 59084 ha of privatelands For land use modeling purposes we focus on a largerfour-county study region surrounding Bend Although manyIMAP study components remain under development wereport forest and range land development results here alongwith a model application of immediate relevance to resourcepolicymakers and managers in Oregon involving the main-tenance of habitat for mule deer

Our central Oregon study region includes Crook Des-chutes and Jefferson Counties and the northern third ofKlamath County (Fig 1) The area has experienced fairlyrapid population growth from 1970 to 2000 ranging from28 in northern most Klamath County to 283 in Des-chutes County for a region wide average of 121 (USDCCensus Bureau 2000) The region somewhat epitomizes thenew West in Oregon by experiencing recent declines innatural resource extractive industries in favor of increasedtourism outdoor recreation and amenity-based in-migra-tion (Judson and others 1999) The study region is borderedon the west by the Deschutes National Forest and includesthe scenic towns of Bend and Sisters which are noted asdesirable travel destinations in national media (eg Laskin2004 Preusch 2004) The region comprises roughly 234million ha land of which 137 million ha is public-ownedand 097 ha is private-owed Major land-cover classesidentified on private lands show a mix of forest range andagriculture with developed areas extending south north-east and north of Bend-eastern Oregons largest (82280persons) and most rapidly growing city (Lettman 2004)Other cities include Madras (6650) Prineville (10370persons) and Redmond (25800 persons) (Proehl 2009)which also have grown in population partly as a result ofBend spillover

Rapid housing growth an influx of new residents bothpermanent and seasonal and their potential environmentaland natural resource impacts are of particular interest topolicymakers and land managers in the region Concerns

of buildings of any size or type-within 32-hectare (ha)circles surrounding sample points located on aerial photosof non-federal land in eastern Oregon (Lettman 2004)With 13000 sample points and three observation periodsthe data comprise almost 40000 observations of buildingcounts in eastern Oregon varying in space and time

The 13000 sample points for the study region weredrawn from the primary sample of points used for thestratification of secondary sample points that are measuredin the periodic forest inventories conducted in easternOregon by the Forest Inventory and Analysis Program(Azuma and others 20041) The primary sample consistsof a grid of nearly 70000 points established from aerialphotos taken in 1982 The sampling was implemented toproduce an even geographic distribution of points acrosseastern Oregon Details about Forest Inventory and Anal-ysis Program sampling in eastern Oregon can be found inAzuma and others (2004) Details about how buildingsactually were counted on aerial photos are described inLettman (2004)

The building count data do not distinguish the specificuses of counted buildings such as residential commercialindustrial or public infrastructure because specific usescould not be identified from the aerial photos alone Alsothe limited availability of historical aerial photos for east-ern Oregon necessitated that data collection draw uponaerial photos taken at varying dates spanning 1968 to 2001(Lettman 2004) Photos were selected to provide threetemporal observations of building counts for each samplepoint From these three temporal observations buildingcount values for 1975 1985 and 1995 were derived using acombination of interpolation and extrapolation

Modeling Approach

Our purpose was to describe potential future forest andrange land development within the four-county studyregion in terms of the spatial distribution and rate of changein new buildings Our approach was to describe changes inbuilding counts observed between subsequent sample pointobservations based on socioeconomic and topographicfactors hypothesized to influence forest and range landdevelopment This approach involves estimating aneconometric model of building count change as a functionof explanatory variables that represent relevant socioeco-nomic and topographic factors and then using the esti-mated model coefficients to predict (or compute) futurechanges in building counts based on anticipated changes inexplanatory variable values Predicted changes in buildingcounts can then be added to existing building densities toforecast future building densities Tracking building countson individual sample points at each of the three points in

time yielded two observations of building count change(number of new buildings built) for each sample point Weomitted building count observations that already exceededthe development threshold of eight buildings per 32 hect-ares-roughly equivalent to 25 buildings per square km-to focus our modeling effort on relatively undevelopedforest and range lands Combining the resulting buildingcounts with data describing explanatory variables yielded6131 observations

Explanatory Variables

Following previous econometric approaches to fine-scaledland use modeling (eg Bockstael 1996 Kline and oth-ers 2007) we expect that landowners are more likely todevelop forest and range land to residential housing orother more intensive uses once the present value of futurereturns earned by land in development less conversioncosts equal or exceed returns earned by land remaining inforest or range Spatial economic data describing potentialreturns to forestry and range generally are not available andso proxy variables must be found to permit fine-scaledspatial modeling Within the relatively localized studyregion forest and range lands tend to be rather uniform inthe landscape characteristics that influence the economicreturns to forestry and grazing As a result spatial vari-ability in rates of development is more likely to arise fromvariation in the potential value of those lands in developeduses than from variation in forestry and farming returns Inthe absence of historical data describing developed landvalues we used several proxy variables shown in past landuse analyses to provide a reasonable accounting of devel-opment opportunities faced by landowners in the PacificNorthwest (Kline and others 2003 2007) Those variablesinclude regional population growth the driving accessi-bility of land to cities and other developed areas viaexisting roads and topographic characteristics (eg slope)that influence the feasibility of developed land uses(Table 1) We also included information describing landuse zoning adopted under Oregons statewide system ofland use planning based on evidence of past zoning influ-ence (Kline 2005)

Econometric Model

Following previous spatial land use analysis using similardata (Kline 2005) we constructed a dependent variable-^BUILDINGS-as a non-continuous count describingchanges in building counts observed between subsequentphoto dates Assuming ^BUILDINGS is distributed as aPoisson leads to the negative binomial model

land use planning was implemented during the mid-1980susing two time-steps worth of observations for model esti-mation enables us to include several zoning explanatoryvariables that show land use zoning effeets on development

A final modeling issue is potential spatial autocorrelationamong observations of building count changes Spatialautocorrelation can result from omitted spatial variablessuch as location that influence the development decisions oflandowners and spatial behavioral relationships such ascommon ownership of sampled land parcels The first leadsto inefficient but asymptotically unbiased estimated coeffi-cients the second can lead to inefficient and biased esti-mated coefficients (eg Nelson and Hellerstein 1997)Spatial autocorrelation often is addressed in model estima-tion by including a spatial lag (or neighbor) variable in theregression equation However a difficulty in applied work isthe lack of simple and universally accepted methods fordealing with spatial lag variables when using estimatedmodel coefficients to compute predicted (or forecasted)values Previous analysis has suggested that spatial auto-correlation in the building count data likely are minimal andthat the inclusion of spatial lag variables in model estimationimproved overall predictive accuracy only slightly (Klineand others 2007326) We suspect that any spatial behavioralrelationships unaccounted for by our spatial explanatoryvariables are minimal and proceeded with model estimationleaving any spatial autocorrelation unaddressed

Evaluating Prediction Accuracy

One way to evaluate the prediction accuracy of econo-metric land-use models is to reserve a portion of sample

where y is a random variable and exp(y) has a gammadistribution with mean 1 and variance X Xi is a vector ofindependent variables and B is a vector of coefficients tobe estimated (Greene 1997) The negative binomial modelis a general form of the Poisson model relaxing the Poissonassumption that the dependent variables mean equals itsvariance (Wear and Bolstad 1998)

The panel nature of the data-generally two temporalobservations of building count change per sample point-creates a potential for correlation among pairs of time-seriesobservations for individual sample points to deflate standarderrors and bias estimated coefficients These potential cor-relations can be accounted for using a random effectsnegative binomial model (Greene 1998629-634) Becausegroup effects are conditioned out (not computed) projectedvalues cannot be computed using the random effects model(Greene 1998630) but the estimated coefficients can becompared to those of the model estimated without randomeffects An alternative approach would be to combine thetwo temporal observations of building counts per observa-tion into a single observation of change over a singlecombined time period which would remove the necessityfor a random effects model However a disadvantage of thisalternative approach would be the significant loss of infor-mation owing to reducing by half the number of observa-tions used to estimate the model Also because statewide

data for computing predicted values based on estimatedmodel coefficients and then comparing these to their actualvalues We chose not to use this method because thebuilding count data included relatively few observations ofboth higher building counts and building count changesWe were hesitant to reduce these observations further byreserving any portion of the data sample from model esti-mation As an alternative we graphically examinedpotential spatial patterns in prediction accuracy by plottingresiduals (Yi - Y^i against select explanatory variable val-ues describing key landscape characteristics Mappingresiduals is not permitted by Forest Inventory and AnalysisProgram confidentiality rules concerning the display ofsample point locations We also used the estimated modelcoefficients to compute within sample changes (t - 1 to t)in building counts These predicted changes were added toinitial building counts (observed at t - 1) to estimate anending building count (observed at t) for each observationi The percentages of correct building counts predicted bythe model are reported for three broad building densitycategories lt7 buildings per square km (relatively undev-eloped) 7 to 25 buildings per square km (moderatelydeveloped) and gt25 buildings per square km (relativelydeveloped) We evaluated the prediction accuracy byexamining the percentage of correct predictions withinbuilding count categories and observing the chance-cor-rected agreement between the actual and predicted valuesusing a Kappa statistic (Cohen 1960)

Results

The general regression equation describing changes inbuilding counts on sample points from one photo date t - 1to the next t was

Model coefficients were estimated using LIMDEP (Greene1998) The negative binomial model is highly statisticallysignificant based on log-likelihood ratio tests (Table 2) andthe signs and statistical significance of the estimated coef-ficients for explanatory variables generally are consistentwith previous analyses of forest and range land develop-ment positive for ^POPULATION DENSITY negative forMARKET CENTER positive for BUILDINGSt-1 bandnegative for SLOPE (Kline 2005 Kline and others 2007)Estimated coefficients for land-use zoning variables sug-gest that zoning has focused new building constructionwithin urban growth boundaries rural-residential or other

developable zones relative to lands in forest range andagricultural zones consistent with land use planning effectsfound in Oregon by previous studies (eg Kline 2005) Therandom effects version of the estimated model yieldedsimilar results

Model predicted residuals (Yi - Y^i) plotted against esti-mated travel times to the nearest market center (MARKETCENTER) indicate a fairly even balance between under-prediction (Yi gtY^i) and over-prediction (YiltY^i) (Fig 2)Residuals plotted against initial building counts (BUILD-INGSt-1) indicate that a core group of observations also arefairly evenly balanced between under- and over-predictionAlthough several outlier observations are under-predictedat BUILDINGSt_1 values of 4 and below and over-pre-dicted at BUILDINGSt-1 values of 5 and above theseobservations represent relatively few of the 6131 obser-vations examined Residuals plotted against SLOPE alsoindicate a fairly even balance between under-prediction andover-prediction on those slopes most feasible for con-struction-generally less than 35 percent (Fig 2) Takentogether the residual plots do not indicate significant spatialpatterns in predicted value errors They do however sug-gest a smoother pattern of predicted development than isevident in the data when viewed relative to outlier values

Within-sample prediction accuracy indicates that thepercentages of correct predictions within each of threebuilding density categories are 963 (lt7 buildings persquare km) 667 (7 to 25) and 632 (gt25) for an overallprediction accuracy of 933 and a chance-corrected pre-diction accuracy of 659 (Table 3) Our comparison of thedistributions of actual predicted values among the threebuilding density categories indicates that the model tends toover-predict development The distribution of actual valuesis 902 (lt7)76 (7 to 25) and 22 (gt25) the distri-bution of predicted values is 883 (lt7) 90 (7 to 25)and 27 (gt25) Because the method used to compute eachsuccessive future building count depends on the previousperiods building count prediction errors will magnify withsuccessive prediction iterations for future time periods Theproblem potentially becomes multiplied when developmentmodel predictions are combined with the predicted outputsof other models describing other study componentsAlthough error propagation often is par for the course withpredictive models particularly in multidisciplinary researchwhere numerous models are combined analysts will want toconsider how error propagation may influence landscapeanalysis results and research outcomes

Example Model Application Involving Mule Deer

The estimated model coefficients can be used to informother IMAP components describing ecological conditions

and processes One way is to characterize development isto compute predicted values of ^BUILDINGS as

The ^BUILDINGS values can then be added to a base mapof existing building counts to create maps depicting futurebuilding counts In this way anticipated future populationdensities-such as would be derived from officialpopulation projections-could be used as a basis fordescribing future development scenarios while controllingfor topography and zoning Alternatively some landscapeanalysis applications call for a probabilistic treatment ofpotential future development A second approach then is tocompute the probability of a specific building countincrease using a set of recursive equations FollowingGreene (1998607) the probability that ^BUILDINGSequals zero (y = 0) for observation i is

Applying these equations at the pixel level enables analyststo create maps describing the probability of specific

building count increases to facilitate probabilistic land-scape simulations at fine spatial scales

To illustrate example model predictions we took thefirst approach and used the estimated model coefficients topredict future increases in building counts based on pro-jected county population density changes to 2040 (OregonOffice of Economic Analysis 2004) Following proceduresdescribed in Kline and others (2003357-358) a base year2000 map of building counts was developed from 2000sample point data by interpolating between sample-pointbuilding count values The estimated negative binomialregression coefficients (Table 2) were combined withprojected population densities for study region countiesand other explanatory variable data to compute predictedchanges in building counts at 10-year time intervals Pre-dicted changes in building counts for each l0-year timeinterval were added to the beginning (t - 1) building countvalue for that interval to obtain the ending building countFor example the predicted changes occurring between the2000 base year and 2010 were added to the 2000 base yearbuilding count map to create a 2010 building count mapThe 2010 map was combined with 2010 to 2020 predictedchanges in building counts to create a 2020 map Projectedpopulation growth in the central region ranges from a lowof 14 for the portion located in northern Klamath Countyto a high of 84 for Deschutes County for a region-widearea-weighted average of 73 The resulting maps (Fig 3)suggest noticeable expansion of development on lands nearexisting cities with notable increases in building countsalong major transportation corridors and in select locationsbetween existing cities

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

include loss of forest and range land to developmentincreased traffic congestion increased water demand andpotential adverse habitat effects for some species Of spe-cial concern is the desire to maintain viable mule deerpopulations sufficient to permit continued hunting Muledeer have declined across the western US as a result ofhabitat loss and other factors Mule deer winter range oftencoincides with new development (eg Stein and others200712) In mountainous forest areas new developmenttends to be located on relatively flatter and lower elevationvalley bottoms where for mule deer the relative absence ofpersistent snow cover makes winter movement easier andfood more abundant In eastern Oregon (east of the crest ofthe Cascades) declining mule deer populations owing todevelopment and habitat loss is the single most pressingissue for the Oregon Department of Fish and Wildlife andhunting enthusiasts (Oregon Department of Fish andWildlife 200) Potential disruption of migration patternsowing to increased development density along key routes isof primary concern Mule deer hunting exceeds 74000participants and generates $22 million in economic activityannually (Oregon Department of Fish and Wildlife 2009a)The sale of hunting licenses is the major source of fundingfor the Oregon Department of Fish and Wildlife

Methods

Building Count Data

Land-use data describing historical building counts werecompiled by the Oregon Department of Forestry and theUSDA Forest Service Pacific Northwest Research Sta-tions Forest Inventory and Analysis program The datawere designed to examine historical forest and range landdevelopment and evaluate wildfire risks to homes andother issues of concern The data consist of aerial photoobservations of building (or structure) counts-the number

west of Bend Oregon consisting of 216103 ha of federalforest reserves and wilderness and 59084 ha of privatelands For land use modeling purposes we focus on a largerfour-county study region surrounding Bend Although manyIMAP study components remain under development wereport forest and range land development results here alongwith a model application of immediate relevance to resourcepolicymakers and managers in Oregon involving the main-tenance of habitat for mule deer

Our central Oregon study region includes Crook Des-chutes and Jefferson Counties and the northern third ofKlamath County (Fig 1) The area has experienced fairlyrapid population growth from 1970 to 2000 ranging from28 in northern most Klamath County to 283 in Des-chutes County for a region wide average of 121 (USDCCensus Bureau 2000) The region somewhat epitomizes thenew West in Oregon by experiencing recent declines innatural resource extractive industries in favor of increasedtourism outdoor recreation and amenity-based in-migra-tion (Judson and others 1999) The study region is borderedon the west by the Deschutes National Forest and includesthe scenic towns of Bend and Sisters which are noted asdesirable travel destinations in national media (eg Laskin2004 Preusch 2004) The region comprises roughly 234million ha land of which 137 million ha is public-ownedand 097 ha is private-owed Major land-cover classesidentified on private lands show a mix of forest range andagriculture with developed areas extending south north-east and north of Bend-eastern Oregons largest (82280persons) and most rapidly growing city (Lettman 2004)Other cities include Madras (6650) Prineville (10370persons) and Redmond (25800 persons) (Proehl 2009)which also have grown in population partly as a result ofBend spillover

Rapid housing growth an influx of new residents bothpermanent and seasonal and their potential environmentaland natural resource impacts are of particular interest topolicymakers and land managers in the region Concerns

of buildings of any size or type-within 32-hectare (ha)circles surrounding sample points located on aerial photosof non-federal land in eastern Oregon (Lettman 2004)With 13000 sample points and three observation periodsthe data comprise almost 40000 observations of buildingcounts in eastern Oregon varying in space and time

The 13000 sample points for the study region weredrawn from the primary sample of points used for thestratification of secondary sample points that are measuredin the periodic forest inventories conducted in easternOregon by the Forest Inventory and Analysis Program(Azuma and others 20041) The primary sample consistsof a grid of nearly 70000 points established from aerialphotos taken in 1982 The sampling was implemented toproduce an even geographic distribution of points acrosseastern Oregon Details about Forest Inventory and Anal-ysis Program sampling in eastern Oregon can be found inAzuma and others (2004) Details about how buildingsactually were counted on aerial photos are described inLettman (2004)

The building count data do not distinguish the specificuses of counted buildings such as residential commercialindustrial or public infrastructure because specific usescould not be identified from the aerial photos alone Alsothe limited availability of historical aerial photos for east-ern Oregon necessitated that data collection draw uponaerial photos taken at varying dates spanning 1968 to 2001(Lettman 2004) Photos were selected to provide threetemporal observations of building counts for each samplepoint From these three temporal observations buildingcount values for 1975 1985 and 1995 were derived using acombination of interpolation and extrapolation

Modeling Approach

Our purpose was to describe potential future forest andrange land development within the four-county studyregion in terms of the spatial distribution and rate of changein new buildings Our approach was to describe changes inbuilding counts observed between subsequent sample pointobservations based on socioeconomic and topographicfactors hypothesized to influence forest and range landdevelopment This approach involves estimating aneconometric model of building count change as a functionof explanatory variables that represent relevant socioeco-nomic and topographic factors and then using the esti-mated model coefficients to predict (or compute) futurechanges in building counts based on anticipated changes inexplanatory variable values Predicted changes in buildingcounts can then be added to existing building densities toforecast future building densities Tracking building countson individual sample points at each of the three points in

time yielded two observations of building count change(number of new buildings built) for each sample point Weomitted building count observations that already exceededthe development threshold of eight buildings per 32 hect-ares-roughly equivalent to 25 buildings per square km-to focus our modeling effort on relatively undevelopedforest and range lands Combining the resulting buildingcounts with data describing explanatory variables yielded6131 observations

Explanatory Variables

Following previous econometric approaches to fine-scaledland use modeling (eg Bockstael 1996 Kline and oth-ers 2007) we expect that landowners are more likely todevelop forest and range land to residential housing orother more intensive uses once the present value of futurereturns earned by land in development less conversioncosts equal or exceed returns earned by land remaining inforest or range Spatial economic data describing potentialreturns to forestry and range generally are not available andso proxy variables must be found to permit fine-scaledspatial modeling Within the relatively localized studyregion forest and range lands tend to be rather uniform inthe landscape characteristics that influence the economicreturns to forestry and grazing As a result spatial vari-ability in rates of development is more likely to arise fromvariation in the potential value of those lands in developeduses than from variation in forestry and farming returns Inthe absence of historical data describing developed landvalues we used several proxy variables shown in past landuse analyses to provide a reasonable accounting of devel-opment opportunities faced by landowners in the PacificNorthwest (Kline and others 2003 2007) Those variablesinclude regional population growth the driving accessi-bility of land to cities and other developed areas viaexisting roads and topographic characteristics (eg slope)that influence the feasibility of developed land uses(Table 1) We also included information describing landuse zoning adopted under Oregons statewide system ofland use planning based on evidence of past zoning influ-ence (Kline 2005)

Econometric Model

Following previous spatial land use analysis using similardata (Kline 2005) we constructed a dependent variable-^BUILDINGS-as a non-continuous count describingchanges in building counts observed between subsequentphoto dates Assuming ^BUILDINGS is distributed as aPoisson leads to the negative binomial model

land use planning was implemented during the mid-1980susing two time-steps worth of observations for model esti-mation enables us to include several zoning explanatoryvariables that show land use zoning effeets on development

A final modeling issue is potential spatial autocorrelationamong observations of building count changes Spatialautocorrelation can result from omitted spatial variablessuch as location that influence the development decisions oflandowners and spatial behavioral relationships such ascommon ownership of sampled land parcels The first leadsto inefficient but asymptotically unbiased estimated coeffi-cients the second can lead to inefficient and biased esti-mated coefficients (eg Nelson and Hellerstein 1997)Spatial autocorrelation often is addressed in model estima-tion by including a spatial lag (or neighbor) variable in theregression equation However a difficulty in applied work isthe lack of simple and universally accepted methods fordealing with spatial lag variables when using estimatedmodel coefficients to compute predicted (or forecasted)values Previous analysis has suggested that spatial auto-correlation in the building count data likely are minimal andthat the inclusion of spatial lag variables in model estimationimproved overall predictive accuracy only slightly (Klineand others 2007326) We suspect that any spatial behavioralrelationships unaccounted for by our spatial explanatoryvariables are minimal and proceeded with model estimationleaving any spatial autocorrelation unaddressed

Evaluating Prediction Accuracy

One way to evaluate the prediction accuracy of econo-metric land-use models is to reserve a portion of sample

where y is a random variable and exp(y) has a gammadistribution with mean 1 and variance X Xi is a vector ofindependent variables and B is a vector of coefficients tobe estimated (Greene 1997) The negative binomial modelis a general form of the Poisson model relaxing the Poissonassumption that the dependent variables mean equals itsvariance (Wear and Bolstad 1998)

The panel nature of the data-generally two temporalobservations of building count change per sample point-creates a potential for correlation among pairs of time-seriesobservations for individual sample points to deflate standarderrors and bias estimated coefficients These potential cor-relations can be accounted for using a random effectsnegative binomial model (Greene 1998629-634) Becausegroup effects are conditioned out (not computed) projectedvalues cannot be computed using the random effects model(Greene 1998630) but the estimated coefficients can becompared to those of the model estimated without randomeffects An alternative approach would be to combine thetwo temporal observations of building counts per observa-tion into a single observation of change over a singlecombined time period which would remove the necessityfor a random effects model However a disadvantage of thisalternative approach would be the significant loss of infor-mation owing to reducing by half the number of observa-tions used to estimate the model Also because statewide

data for computing predicted values based on estimatedmodel coefficients and then comparing these to their actualvalues We chose not to use this method because thebuilding count data included relatively few observations ofboth higher building counts and building count changesWe were hesitant to reduce these observations further byreserving any portion of the data sample from model esti-mation As an alternative we graphically examinedpotential spatial patterns in prediction accuracy by plottingresiduals (Yi - Y^i against select explanatory variable val-ues describing key landscape characteristics Mappingresiduals is not permitted by Forest Inventory and AnalysisProgram confidentiality rules concerning the display ofsample point locations We also used the estimated modelcoefficients to compute within sample changes (t - 1 to t)in building counts These predicted changes were added toinitial building counts (observed at t - 1) to estimate anending building count (observed at t) for each observationi The percentages of correct building counts predicted bythe model are reported for three broad building densitycategories lt7 buildings per square km (relatively undev-eloped) 7 to 25 buildings per square km (moderatelydeveloped) and gt25 buildings per square km (relativelydeveloped) We evaluated the prediction accuracy byexamining the percentage of correct predictions withinbuilding count categories and observing the chance-cor-rected agreement between the actual and predicted valuesusing a Kappa statistic (Cohen 1960)

Results

The general regression equation describing changes inbuilding counts on sample points from one photo date t - 1to the next t was

Model coefficients were estimated using LIMDEP (Greene1998) The negative binomial model is highly statisticallysignificant based on log-likelihood ratio tests (Table 2) andthe signs and statistical significance of the estimated coef-ficients for explanatory variables generally are consistentwith previous analyses of forest and range land develop-ment positive for ^POPULATION DENSITY negative forMARKET CENTER positive for BUILDINGSt-1 bandnegative for SLOPE (Kline 2005 Kline and others 2007)Estimated coefficients for land-use zoning variables sug-gest that zoning has focused new building constructionwithin urban growth boundaries rural-residential or other

developable zones relative to lands in forest range andagricultural zones consistent with land use planning effectsfound in Oregon by previous studies (eg Kline 2005) Therandom effects version of the estimated model yieldedsimilar results

Model predicted residuals (Yi - Y^i) plotted against esti-mated travel times to the nearest market center (MARKETCENTER) indicate a fairly even balance between under-prediction (Yi gtY^i) and over-prediction (YiltY^i) (Fig 2)Residuals plotted against initial building counts (BUILD-INGSt-1) indicate that a core group of observations also arefairly evenly balanced between under- and over-predictionAlthough several outlier observations are under-predictedat BUILDINGSt_1 values of 4 and below and over-pre-dicted at BUILDINGSt-1 values of 5 and above theseobservations represent relatively few of the 6131 obser-vations examined Residuals plotted against SLOPE alsoindicate a fairly even balance between under-prediction andover-prediction on those slopes most feasible for con-struction-generally less than 35 percent (Fig 2) Takentogether the residual plots do not indicate significant spatialpatterns in predicted value errors They do however sug-gest a smoother pattern of predicted development than isevident in the data when viewed relative to outlier values

Within-sample prediction accuracy indicates that thepercentages of correct predictions within each of threebuilding density categories are 963 (lt7 buildings persquare km) 667 (7 to 25) and 632 (gt25) for an overallprediction accuracy of 933 and a chance-corrected pre-diction accuracy of 659 (Table 3) Our comparison of thedistributions of actual predicted values among the threebuilding density categories indicates that the model tends toover-predict development The distribution of actual valuesis 902 (lt7)76 (7 to 25) and 22 (gt25) the distri-bution of predicted values is 883 (lt7) 90 (7 to 25)and 27 (gt25) Because the method used to compute eachsuccessive future building count depends on the previousperiods building count prediction errors will magnify withsuccessive prediction iterations for future time periods Theproblem potentially becomes multiplied when developmentmodel predictions are combined with the predicted outputsof other models describing other study componentsAlthough error propagation often is par for the course withpredictive models particularly in multidisciplinary researchwhere numerous models are combined analysts will want toconsider how error propagation may influence landscapeanalysis results and research outcomes

Example Model Application Involving Mule Deer

The estimated model coefficients can be used to informother IMAP components describing ecological conditions

and processes One way is to characterize development isto compute predicted values of ^BUILDINGS as

The ^BUILDINGS values can then be added to a base mapof existing building counts to create maps depicting futurebuilding counts In this way anticipated future populationdensities-such as would be derived from officialpopulation projections-could be used as a basis fordescribing future development scenarios while controllingfor topography and zoning Alternatively some landscapeanalysis applications call for a probabilistic treatment ofpotential future development A second approach then is tocompute the probability of a specific building countincrease using a set of recursive equations FollowingGreene (1998607) the probability that ^BUILDINGSequals zero (y = 0) for observation i is

Applying these equations at the pixel level enables analyststo create maps describing the probability of specific

building count increases to facilitate probabilistic land-scape simulations at fine spatial scales

To illustrate example model predictions we took thefirst approach and used the estimated model coefficients topredict future increases in building counts based on pro-jected county population density changes to 2040 (OregonOffice of Economic Analysis 2004) Following proceduresdescribed in Kline and others (2003357-358) a base year2000 map of building counts was developed from 2000sample point data by interpolating between sample-pointbuilding count values The estimated negative binomialregression coefficients (Table 2) were combined withprojected population densities for study region countiesand other explanatory variable data to compute predictedchanges in building counts at 10-year time intervals Pre-dicted changes in building counts for each l0-year timeinterval were added to the beginning (t - 1) building countvalue for that interval to obtain the ending building countFor example the predicted changes occurring between the2000 base year and 2010 were added to the 2000 base yearbuilding count map to create a 2010 building count mapThe 2010 map was combined with 2010 to 2020 predictedchanges in building counts to create a 2020 map Projectedpopulation growth in the central region ranges from a lowof 14 for the portion located in northern Klamath Countyto a high of 84 for Deschutes County for a region-widearea-weighted average of 73 The resulting maps (Fig 3)suggest noticeable expansion of development on lands nearexisting cities with notable increases in building countsalong major transportation corridors and in select locationsbetween existing cities

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

of buildings of any size or type-within 32-hectare (ha)circles surrounding sample points located on aerial photosof non-federal land in eastern Oregon (Lettman 2004)With 13000 sample points and three observation periodsthe data comprise almost 40000 observations of buildingcounts in eastern Oregon varying in space and time

The 13000 sample points for the study region weredrawn from the primary sample of points used for thestratification of secondary sample points that are measuredin the periodic forest inventories conducted in easternOregon by the Forest Inventory and Analysis Program(Azuma and others 20041) The primary sample consistsof a grid of nearly 70000 points established from aerialphotos taken in 1982 The sampling was implemented toproduce an even geographic distribution of points acrosseastern Oregon Details about Forest Inventory and Anal-ysis Program sampling in eastern Oregon can be found inAzuma and others (2004) Details about how buildingsactually were counted on aerial photos are described inLettman (2004)

The building count data do not distinguish the specificuses of counted buildings such as residential commercialindustrial or public infrastructure because specific usescould not be identified from the aerial photos alone Alsothe limited availability of historical aerial photos for east-ern Oregon necessitated that data collection draw uponaerial photos taken at varying dates spanning 1968 to 2001(Lettman 2004) Photos were selected to provide threetemporal observations of building counts for each samplepoint From these three temporal observations buildingcount values for 1975 1985 and 1995 were derived using acombination of interpolation and extrapolation

Modeling Approach

Our purpose was to describe potential future forest andrange land development within the four-county studyregion in terms of the spatial distribution and rate of changein new buildings Our approach was to describe changes inbuilding counts observed between subsequent sample pointobservations based on socioeconomic and topographicfactors hypothesized to influence forest and range landdevelopment This approach involves estimating aneconometric model of building count change as a functionof explanatory variables that represent relevant socioeco-nomic and topographic factors and then using the esti-mated model coefficients to predict (or compute) futurechanges in building counts based on anticipated changes inexplanatory variable values Predicted changes in buildingcounts can then be added to existing building densities toforecast future building densities Tracking building countson individual sample points at each of the three points in

time yielded two observations of building count change(number of new buildings built) for each sample point Weomitted building count observations that already exceededthe development threshold of eight buildings per 32 hect-ares-roughly equivalent to 25 buildings per square km-to focus our modeling effort on relatively undevelopedforest and range lands Combining the resulting buildingcounts with data describing explanatory variables yielded6131 observations

Explanatory Variables

Following previous econometric approaches to fine-scaledland use modeling (eg Bockstael 1996 Kline and oth-ers 2007) we expect that landowners are more likely todevelop forest and range land to residential housing orother more intensive uses once the present value of futurereturns earned by land in development less conversioncosts equal or exceed returns earned by land remaining inforest or range Spatial economic data describing potentialreturns to forestry and range generally are not available andso proxy variables must be found to permit fine-scaledspatial modeling Within the relatively localized studyregion forest and range lands tend to be rather uniform inthe landscape characteristics that influence the economicreturns to forestry and grazing As a result spatial vari-ability in rates of development is more likely to arise fromvariation in the potential value of those lands in developeduses than from variation in forestry and farming returns Inthe absence of historical data describing developed landvalues we used several proxy variables shown in past landuse analyses to provide a reasonable accounting of devel-opment opportunities faced by landowners in the PacificNorthwest (Kline and others 2003 2007) Those variablesinclude regional population growth the driving accessi-bility of land to cities and other developed areas viaexisting roads and topographic characteristics (eg slope)that influence the feasibility of developed land uses(Table 1) We also included information describing landuse zoning adopted under Oregons statewide system ofland use planning based on evidence of past zoning influ-ence (Kline 2005)

Econometric Model

Following previous spatial land use analysis using similardata (Kline 2005) we constructed a dependent variable-^BUILDINGS-as a non-continuous count describingchanges in building counts observed between subsequentphoto dates Assuming ^BUILDINGS is distributed as aPoisson leads to the negative binomial model

land use planning was implemented during the mid-1980susing two time-steps worth of observations for model esti-mation enables us to include several zoning explanatoryvariables that show land use zoning effeets on development

A final modeling issue is potential spatial autocorrelationamong observations of building count changes Spatialautocorrelation can result from omitted spatial variablessuch as location that influence the development decisions oflandowners and spatial behavioral relationships such ascommon ownership of sampled land parcels The first leadsto inefficient but asymptotically unbiased estimated coeffi-cients the second can lead to inefficient and biased esti-mated coefficients (eg Nelson and Hellerstein 1997)Spatial autocorrelation often is addressed in model estima-tion by including a spatial lag (or neighbor) variable in theregression equation However a difficulty in applied work isthe lack of simple and universally accepted methods fordealing with spatial lag variables when using estimatedmodel coefficients to compute predicted (or forecasted)values Previous analysis has suggested that spatial auto-correlation in the building count data likely are minimal andthat the inclusion of spatial lag variables in model estimationimproved overall predictive accuracy only slightly (Klineand others 2007326) We suspect that any spatial behavioralrelationships unaccounted for by our spatial explanatoryvariables are minimal and proceeded with model estimationleaving any spatial autocorrelation unaddressed

Evaluating Prediction Accuracy

One way to evaluate the prediction accuracy of econo-metric land-use models is to reserve a portion of sample

where y is a random variable and exp(y) has a gammadistribution with mean 1 and variance X Xi is a vector ofindependent variables and B is a vector of coefficients tobe estimated (Greene 1997) The negative binomial modelis a general form of the Poisson model relaxing the Poissonassumption that the dependent variables mean equals itsvariance (Wear and Bolstad 1998)

The panel nature of the data-generally two temporalobservations of building count change per sample point-creates a potential for correlation among pairs of time-seriesobservations for individual sample points to deflate standarderrors and bias estimated coefficients These potential cor-relations can be accounted for using a random effectsnegative binomial model (Greene 1998629-634) Becausegroup effects are conditioned out (not computed) projectedvalues cannot be computed using the random effects model(Greene 1998630) but the estimated coefficients can becompared to those of the model estimated without randomeffects An alternative approach would be to combine thetwo temporal observations of building counts per observa-tion into a single observation of change over a singlecombined time period which would remove the necessityfor a random effects model However a disadvantage of thisalternative approach would be the significant loss of infor-mation owing to reducing by half the number of observa-tions used to estimate the model Also because statewide

data for computing predicted values based on estimatedmodel coefficients and then comparing these to their actualvalues We chose not to use this method because thebuilding count data included relatively few observations ofboth higher building counts and building count changesWe were hesitant to reduce these observations further byreserving any portion of the data sample from model esti-mation As an alternative we graphically examinedpotential spatial patterns in prediction accuracy by plottingresiduals (Yi - Y^i against select explanatory variable val-ues describing key landscape characteristics Mappingresiduals is not permitted by Forest Inventory and AnalysisProgram confidentiality rules concerning the display ofsample point locations We also used the estimated modelcoefficients to compute within sample changes (t - 1 to t)in building counts These predicted changes were added toinitial building counts (observed at t - 1) to estimate anending building count (observed at t) for each observationi The percentages of correct building counts predicted bythe model are reported for three broad building densitycategories lt7 buildings per square km (relatively undev-eloped) 7 to 25 buildings per square km (moderatelydeveloped) and gt25 buildings per square km (relativelydeveloped) We evaluated the prediction accuracy byexamining the percentage of correct predictions withinbuilding count categories and observing the chance-cor-rected agreement between the actual and predicted valuesusing a Kappa statistic (Cohen 1960)

Results

The general regression equation describing changes inbuilding counts on sample points from one photo date t - 1to the next t was

Model coefficients were estimated using LIMDEP (Greene1998) The negative binomial model is highly statisticallysignificant based on log-likelihood ratio tests (Table 2) andthe signs and statistical significance of the estimated coef-ficients for explanatory variables generally are consistentwith previous analyses of forest and range land develop-ment positive for ^POPULATION DENSITY negative forMARKET CENTER positive for BUILDINGSt-1 bandnegative for SLOPE (Kline 2005 Kline and others 2007)Estimated coefficients for land-use zoning variables sug-gest that zoning has focused new building constructionwithin urban growth boundaries rural-residential or other

developable zones relative to lands in forest range andagricultural zones consistent with land use planning effectsfound in Oregon by previous studies (eg Kline 2005) Therandom effects version of the estimated model yieldedsimilar results

Model predicted residuals (Yi - Y^i) plotted against esti-mated travel times to the nearest market center (MARKETCENTER) indicate a fairly even balance between under-prediction (Yi gtY^i) and over-prediction (YiltY^i) (Fig 2)Residuals plotted against initial building counts (BUILD-INGSt-1) indicate that a core group of observations also arefairly evenly balanced between under- and over-predictionAlthough several outlier observations are under-predictedat BUILDINGSt_1 values of 4 and below and over-pre-dicted at BUILDINGSt-1 values of 5 and above theseobservations represent relatively few of the 6131 obser-vations examined Residuals plotted against SLOPE alsoindicate a fairly even balance between under-prediction andover-prediction on those slopes most feasible for con-struction-generally less than 35 percent (Fig 2) Takentogether the residual plots do not indicate significant spatialpatterns in predicted value errors They do however sug-gest a smoother pattern of predicted development than isevident in the data when viewed relative to outlier values

Within-sample prediction accuracy indicates that thepercentages of correct predictions within each of threebuilding density categories are 963 (lt7 buildings persquare km) 667 (7 to 25) and 632 (gt25) for an overallprediction accuracy of 933 and a chance-corrected pre-diction accuracy of 659 (Table 3) Our comparison of thedistributions of actual predicted values among the threebuilding density categories indicates that the model tends toover-predict development The distribution of actual valuesis 902 (lt7)76 (7 to 25) and 22 (gt25) the distri-bution of predicted values is 883 (lt7) 90 (7 to 25)and 27 (gt25) Because the method used to compute eachsuccessive future building count depends on the previousperiods building count prediction errors will magnify withsuccessive prediction iterations for future time periods Theproblem potentially becomes multiplied when developmentmodel predictions are combined with the predicted outputsof other models describing other study componentsAlthough error propagation often is par for the course withpredictive models particularly in multidisciplinary researchwhere numerous models are combined analysts will want toconsider how error propagation may influence landscapeanalysis results and research outcomes

Example Model Application Involving Mule Deer

The estimated model coefficients can be used to informother IMAP components describing ecological conditions

and processes One way is to characterize development isto compute predicted values of ^BUILDINGS as

The ^BUILDINGS values can then be added to a base mapof existing building counts to create maps depicting futurebuilding counts In this way anticipated future populationdensities-such as would be derived from officialpopulation projections-could be used as a basis fordescribing future development scenarios while controllingfor topography and zoning Alternatively some landscapeanalysis applications call for a probabilistic treatment ofpotential future development A second approach then is tocompute the probability of a specific building countincrease using a set of recursive equations FollowingGreene (1998607) the probability that ^BUILDINGSequals zero (y = 0) for observation i is

Applying these equations at the pixel level enables analyststo create maps describing the probability of specific

building count increases to facilitate probabilistic land-scape simulations at fine spatial scales

To illustrate example model predictions we took thefirst approach and used the estimated model coefficients topredict future increases in building counts based on pro-jected county population density changes to 2040 (OregonOffice of Economic Analysis 2004) Following proceduresdescribed in Kline and others (2003357-358) a base year2000 map of building counts was developed from 2000sample point data by interpolating between sample-pointbuilding count values The estimated negative binomialregression coefficients (Table 2) were combined withprojected population densities for study region countiesand other explanatory variable data to compute predictedchanges in building counts at 10-year time intervals Pre-dicted changes in building counts for each l0-year timeinterval were added to the beginning (t - 1) building countvalue for that interval to obtain the ending building countFor example the predicted changes occurring between the2000 base year and 2010 were added to the 2000 base yearbuilding count map to create a 2010 building count mapThe 2010 map was combined with 2010 to 2020 predictedchanges in building counts to create a 2020 map Projectedpopulation growth in the central region ranges from a lowof 14 for the portion located in northern Klamath Countyto a high of 84 for Deschutes County for a region-widearea-weighted average of 73 The resulting maps (Fig 3)suggest noticeable expansion of development on lands nearexisting cities with notable increases in building countsalong major transportation corridors and in select locationsbetween existing cities

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

land use planning was implemented during the mid-1980susing two time-steps worth of observations for model esti-mation enables us to include several zoning explanatoryvariables that show land use zoning effeets on development

A final modeling issue is potential spatial autocorrelationamong observations of building count changes Spatialautocorrelation can result from omitted spatial variablessuch as location that influence the development decisions oflandowners and spatial behavioral relationships such ascommon ownership of sampled land parcels The first leadsto inefficient but asymptotically unbiased estimated coeffi-cients the second can lead to inefficient and biased esti-mated coefficients (eg Nelson and Hellerstein 1997)Spatial autocorrelation often is addressed in model estima-tion by including a spatial lag (or neighbor) variable in theregression equation However a difficulty in applied work isthe lack of simple and universally accepted methods fordealing with spatial lag variables when using estimatedmodel coefficients to compute predicted (or forecasted)values Previous analysis has suggested that spatial auto-correlation in the building count data likely are minimal andthat the inclusion of spatial lag variables in model estimationimproved overall predictive accuracy only slightly (Klineand others 2007326) We suspect that any spatial behavioralrelationships unaccounted for by our spatial explanatoryvariables are minimal and proceeded with model estimationleaving any spatial autocorrelation unaddressed

Evaluating Prediction Accuracy

One way to evaluate the prediction accuracy of econo-metric land-use models is to reserve a portion of sample

where y is a random variable and exp(y) has a gammadistribution with mean 1 and variance X Xi is a vector ofindependent variables and B is a vector of coefficients tobe estimated (Greene 1997) The negative binomial modelis a general form of the Poisson model relaxing the Poissonassumption that the dependent variables mean equals itsvariance (Wear and Bolstad 1998)

The panel nature of the data-generally two temporalobservations of building count change per sample point-creates a potential for correlation among pairs of time-seriesobservations for individual sample points to deflate standarderrors and bias estimated coefficients These potential cor-relations can be accounted for using a random effectsnegative binomial model (Greene 1998629-634) Becausegroup effects are conditioned out (not computed) projectedvalues cannot be computed using the random effects model(Greene 1998630) but the estimated coefficients can becompared to those of the model estimated without randomeffects An alternative approach would be to combine thetwo temporal observations of building counts per observa-tion into a single observation of change over a singlecombined time period which would remove the necessityfor a random effects model However a disadvantage of thisalternative approach would be the significant loss of infor-mation owing to reducing by half the number of observa-tions used to estimate the model Also because statewide

data for computing predicted values based on estimatedmodel coefficients and then comparing these to their actualvalues We chose not to use this method because thebuilding count data included relatively few observations ofboth higher building counts and building count changesWe were hesitant to reduce these observations further byreserving any portion of the data sample from model esti-mation As an alternative we graphically examinedpotential spatial patterns in prediction accuracy by plottingresiduals (Yi - Y^i against select explanatory variable val-ues describing key landscape characteristics Mappingresiduals is not permitted by Forest Inventory and AnalysisProgram confidentiality rules concerning the display ofsample point locations We also used the estimated modelcoefficients to compute within sample changes (t - 1 to t)in building counts These predicted changes were added toinitial building counts (observed at t - 1) to estimate anending building count (observed at t) for each observationi The percentages of correct building counts predicted bythe model are reported for three broad building densitycategories lt7 buildings per square km (relatively undev-eloped) 7 to 25 buildings per square km (moderatelydeveloped) and gt25 buildings per square km (relativelydeveloped) We evaluated the prediction accuracy byexamining the percentage of correct predictions withinbuilding count categories and observing the chance-cor-rected agreement between the actual and predicted valuesusing a Kappa statistic (Cohen 1960)

Results

The general regression equation describing changes inbuilding counts on sample points from one photo date t - 1to the next t was

Model coefficients were estimated using LIMDEP (Greene1998) The negative binomial model is highly statisticallysignificant based on log-likelihood ratio tests (Table 2) andthe signs and statistical significance of the estimated coef-ficients for explanatory variables generally are consistentwith previous analyses of forest and range land develop-ment positive for ^POPULATION DENSITY negative forMARKET CENTER positive for BUILDINGSt-1 bandnegative for SLOPE (Kline 2005 Kline and others 2007)Estimated coefficients for land-use zoning variables sug-gest that zoning has focused new building constructionwithin urban growth boundaries rural-residential or other

developable zones relative to lands in forest range andagricultural zones consistent with land use planning effectsfound in Oregon by previous studies (eg Kline 2005) Therandom effects version of the estimated model yieldedsimilar results

Model predicted residuals (Yi - Y^i) plotted against esti-mated travel times to the nearest market center (MARKETCENTER) indicate a fairly even balance between under-prediction (Yi gtY^i) and over-prediction (YiltY^i) (Fig 2)Residuals plotted against initial building counts (BUILD-INGSt-1) indicate that a core group of observations also arefairly evenly balanced between under- and over-predictionAlthough several outlier observations are under-predictedat BUILDINGSt_1 values of 4 and below and over-pre-dicted at BUILDINGSt-1 values of 5 and above theseobservations represent relatively few of the 6131 obser-vations examined Residuals plotted against SLOPE alsoindicate a fairly even balance between under-prediction andover-prediction on those slopes most feasible for con-struction-generally less than 35 percent (Fig 2) Takentogether the residual plots do not indicate significant spatialpatterns in predicted value errors They do however sug-gest a smoother pattern of predicted development than isevident in the data when viewed relative to outlier values

Within-sample prediction accuracy indicates that thepercentages of correct predictions within each of threebuilding density categories are 963 (lt7 buildings persquare km) 667 (7 to 25) and 632 (gt25) for an overallprediction accuracy of 933 and a chance-corrected pre-diction accuracy of 659 (Table 3) Our comparison of thedistributions of actual predicted values among the threebuilding density categories indicates that the model tends toover-predict development The distribution of actual valuesis 902 (lt7)76 (7 to 25) and 22 (gt25) the distri-bution of predicted values is 883 (lt7) 90 (7 to 25)and 27 (gt25) Because the method used to compute eachsuccessive future building count depends on the previousperiods building count prediction errors will magnify withsuccessive prediction iterations for future time periods Theproblem potentially becomes multiplied when developmentmodel predictions are combined with the predicted outputsof other models describing other study componentsAlthough error propagation often is par for the course withpredictive models particularly in multidisciplinary researchwhere numerous models are combined analysts will want toconsider how error propagation may influence landscapeanalysis results and research outcomes

Example Model Application Involving Mule Deer

The estimated model coefficients can be used to informother IMAP components describing ecological conditions

and processes One way is to characterize development isto compute predicted values of ^BUILDINGS as

The ^BUILDINGS values can then be added to a base mapof existing building counts to create maps depicting futurebuilding counts In this way anticipated future populationdensities-such as would be derived from officialpopulation projections-could be used as a basis fordescribing future development scenarios while controllingfor topography and zoning Alternatively some landscapeanalysis applications call for a probabilistic treatment ofpotential future development A second approach then is tocompute the probability of a specific building countincrease using a set of recursive equations FollowingGreene (1998607) the probability that ^BUILDINGSequals zero (y = 0) for observation i is

Applying these equations at the pixel level enables analyststo create maps describing the probability of specific

building count increases to facilitate probabilistic land-scape simulations at fine spatial scales

To illustrate example model predictions we took thefirst approach and used the estimated model coefficients topredict future increases in building counts based on pro-jected county population density changes to 2040 (OregonOffice of Economic Analysis 2004) Following proceduresdescribed in Kline and others (2003357-358) a base year2000 map of building counts was developed from 2000sample point data by interpolating between sample-pointbuilding count values The estimated negative binomialregression coefficients (Table 2) were combined withprojected population densities for study region countiesand other explanatory variable data to compute predictedchanges in building counts at 10-year time intervals Pre-dicted changes in building counts for each l0-year timeinterval were added to the beginning (t - 1) building countvalue for that interval to obtain the ending building countFor example the predicted changes occurring between the2000 base year and 2010 were added to the 2000 base yearbuilding count map to create a 2010 building count mapThe 2010 map was combined with 2010 to 2020 predictedchanges in building counts to create a 2020 map Projectedpopulation growth in the central region ranges from a lowof 14 for the portion located in northern Klamath Countyto a high of 84 for Deschutes County for a region-widearea-weighted average of 73 The resulting maps (Fig 3)suggest noticeable expansion of development on lands nearexisting cities with notable increases in building countsalong major transportation corridors and in select locationsbetween existing cities

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

data for computing predicted values based on estimatedmodel coefficients and then comparing these to their actualvalues We chose not to use this method because thebuilding count data included relatively few observations ofboth higher building counts and building count changesWe were hesitant to reduce these observations further byreserving any portion of the data sample from model esti-mation As an alternative we graphically examinedpotential spatial patterns in prediction accuracy by plottingresiduals (Yi - Y^i against select explanatory variable val-ues describing key landscape characteristics Mappingresiduals is not permitted by Forest Inventory and AnalysisProgram confidentiality rules concerning the display ofsample point locations We also used the estimated modelcoefficients to compute within sample changes (t - 1 to t)in building counts These predicted changes were added toinitial building counts (observed at t - 1) to estimate anending building count (observed at t) for each observationi The percentages of correct building counts predicted bythe model are reported for three broad building densitycategories lt7 buildings per square km (relatively undev-eloped) 7 to 25 buildings per square km (moderatelydeveloped) and gt25 buildings per square km (relativelydeveloped) We evaluated the prediction accuracy byexamining the percentage of correct predictions withinbuilding count categories and observing the chance-cor-rected agreement between the actual and predicted valuesusing a Kappa statistic (Cohen 1960)

Results

The general regression equation describing changes inbuilding counts on sample points from one photo date t - 1to the next t was

Model coefficients were estimated using LIMDEP (Greene1998) The negative binomial model is highly statisticallysignificant based on log-likelihood ratio tests (Table 2) andthe signs and statistical significance of the estimated coef-ficients for explanatory variables generally are consistentwith previous analyses of forest and range land develop-ment positive for ^POPULATION DENSITY negative forMARKET CENTER positive for BUILDINGSt-1 bandnegative for SLOPE (Kline 2005 Kline and others 2007)Estimated coefficients for land-use zoning variables sug-gest that zoning has focused new building constructionwithin urban growth boundaries rural-residential or other

developable zones relative to lands in forest range andagricultural zones consistent with land use planning effectsfound in Oregon by previous studies (eg Kline 2005) Therandom effects version of the estimated model yieldedsimilar results

Model predicted residuals (Yi - Y^i) plotted against esti-mated travel times to the nearest market center (MARKETCENTER) indicate a fairly even balance between under-prediction (Yi gtY^i) and over-prediction (YiltY^i) (Fig 2)Residuals plotted against initial building counts (BUILD-INGSt-1) indicate that a core group of observations also arefairly evenly balanced between under- and over-predictionAlthough several outlier observations are under-predictedat BUILDINGSt_1 values of 4 and below and over-pre-dicted at BUILDINGSt-1 values of 5 and above theseobservations represent relatively few of the 6131 obser-vations examined Residuals plotted against SLOPE alsoindicate a fairly even balance between under-prediction andover-prediction on those slopes most feasible for con-struction-generally less than 35 percent (Fig 2) Takentogether the residual plots do not indicate significant spatialpatterns in predicted value errors They do however sug-gest a smoother pattern of predicted development than isevident in the data when viewed relative to outlier values

Within-sample prediction accuracy indicates that thepercentages of correct predictions within each of threebuilding density categories are 963 (lt7 buildings persquare km) 667 (7 to 25) and 632 (gt25) for an overallprediction accuracy of 933 and a chance-corrected pre-diction accuracy of 659 (Table 3) Our comparison of thedistributions of actual predicted values among the threebuilding density categories indicates that the model tends toover-predict development The distribution of actual valuesis 902 (lt7)76 (7 to 25) and 22 (gt25) the distri-bution of predicted values is 883 (lt7) 90 (7 to 25)and 27 (gt25) Because the method used to compute eachsuccessive future building count depends on the previousperiods building count prediction errors will magnify withsuccessive prediction iterations for future time periods Theproblem potentially becomes multiplied when developmentmodel predictions are combined with the predicted outputsof other models describing other study componentsAlthough error propagation often is par for the course withpredictive models particularly in multidisciplinary researchwhere numerous models are combined analysts will want toconsider how error propagation may influence landscapeanalysis results and research outcomes

Example Model Application Involving Mule Deer

The estimated model coefficients can be used to informother IMAP components describing ecological conditions

and processes One way is to characterize development isto compute predicted values of ^BUILDINGS as

The ^BUILDINGS values can then be added to a base mapof existing building counts to create maps depicting futurebuilding counts In this way anticipated future populationdensities-such as would be derived from officialpopulation projections-could be used as a basis fordescribing future development scenarios while controllingfor topography and zoning Alternatively some landscapeanalysis applications call for a probabilistic treatment ofpotential future development A second approach then is tocompute the probability of a specific building countincrease using a set of recursive equations FollowingGreene (1998607) the probability that ^BUILDINGSequals zero (y = 0) for observation i is

Applying these equations at the pixel level enables analyststo create maps describing the probability of specific

building count increases to facilitate probabilistic land-scape simulations at fine spatial scales

To illustrate example model predictions we took thefirst approach and used the estimated model coefficients topredict future increases in building counts based on pro-jected county population density changes to 2040 (OregonOffice of Economic Analysis 2004) Following proceduresdescribed in Kline and others (2003357-358) a base year2000 map of building counts was developed from 2000sample point data by interpolating between sample-pointbuilding count values The estimated negative binomialregression coefficients (Table 2) were combined withprojected population densities for study region countiesand other explanatory variable data to compute predictedchanges in building counts at 10-year time intervals Pre-dicted changes in building counts for each l0-year timeinterval were added to the beginning (t - 1) building countvalue for that interval to obtain the ending building countFor example the predicted changes occurring between the2000 base year and 2010 were added to the 2000 base yearbuilding count map to create a 2010 building count mapThe 2010 map was combined with 2010 to 2020 predictedchanges in building counts to create a 2020 map Projectedpopulation growth in the central region ranges from a lowof 14 for the portion located in northern Klamath Countyto a high of 84 for Deschutes County for a region-widearea-weighted average of 73 The resulting maps (Fig 3)suggest noticeable expansion of development on lands nearexisting cities with notable increases in building countsalong major transportation corridors and in select locationsbetween existing cities

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

and processes One way is to characterize development isto compute predicted values of ^BUILDINGS as

The ^BUILDINGS values can then be added to a base mapof existing building counts to create maps depicting futurebuilding counts In this way anticipated future populationdensities-such as would be derived from officialpopulation projections-could be used as a basis fordescribing future development scenarios while controllingfor topography and zoning Alternatively some landscapeanalysis applications call for a probabilistic treatment ofpotential future development A second approach then is tocompute the probability of a specific building countincrease using a set of recursive equations FollowingGreene (1998607) the probability that ^BUILDINGSequals zero (y = 0) for observation i is

Applying these equations at the pixel level enables analyststo create maps describing the probability of specific

building count increases to facilitate probabilistic land-scape simulations at fine spatial scales

To illustrate example model predictions we took thefirst approach and used the estimated model coefficients topredict future increases in building counts based on pro-jected county population density changes to 2040 (OregonOffice of Economic Analysis 2004) Following proceduresdescribed in Kline and others (2003357-358) a base year2000 map of building counts was developed from 2000sample point data by interpolating between sample-pointbuilding count values The estimated negative binomialregression coefficients (Table 2) were combined withprojected population densities for study region countiesand other explanatory variable data to compute predictedchanges in building counts at 10-year time intervals Pre-dicted changes in building counts for each l0-year timeinterval were added to the beginning (t - 1) building countvalue for that interval to obtain the ending building countFor example the predicted changes occurring between the2000 base year and 2010 were added to the 2000 base yearbuilding count map to create a 2010 building count mapThe 2010 map was combined with 2010 to 2020 predictedchanges in building counts to create a 2020 map Projectedpopulation growth in the central region ranges from a lowof 14 for the portion located in northern Klamath Countyto a high of 84 for Deschutes County for a region-widearea-weighted average of 73 The resulting maps (Fig 3)suggest noticeable expansion of development on lands nearexisting cities with notable increases in building countsalong major transportation corridors and in select locationsbetween existing cities

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

range largely is a function of elevation which constrainswinter range extent on the west with significant wintersnow cover at higher elevations of the eastern slope of theCascades range and in sporadic central locations to the eastof urban areas with generally drier conditions at lowerelevations (Fig 3) The map overlay shows that by 2000development was already present in many locations withinmule deer winter range-some of it at sufficiently highdensities to adversely affect animal movement from oneportion of range to another Notable examples are the largearea of development at densities of from 7 to 25 to greaterthan 25 buildings per square km northwest of Bend as wellas the smaller area of development at greater than 25buildings per square km filling a narrow strip of winterrange just south of Bend

Projections suggest greater development in the futureespecially in western portions of winter range with con-tinued infill of buildings northwest of Bend (Fig 3)Development projections suggest that the proportion ofwinter range falling into the 7 to 25 and gt25 buildingdensity categories collectively will rise from 0045(0028 + 0017) in 2000 to 0067 (0017 + 0050) by 2040(Table 4) Conversely the proportion of winter rangefalling into the public land and lt7 building density cate-gories collectively will decline from 0955 (0493 + 0462)in 2000 to 0933 (0493 + 0440) by 2040 Although themagnitudes of these shifts do not appear all that significantrelative to the total amount of winter range presentexpected development could result in policy-relevantimpacts to winter range if it occurs as anticipated at keychoke points where it could hinder movement betweendifferent portions of winter range The notable examplesare again the area northwest of Bend which shifts from 7 to25 buildings per square km or less in 2000 to gt25 build-ings per square km by 2020 to add to the already developednarrow strip of winter range just south of Bend Evenat low densities development could adversely affect muledeer migration if new housing is accompanied bythe installation of fencing to accommodate horses andother livestock as is often the case in central OregonManagement challenges can be exacerbated if housing

The maps of future development can be combined withother information to describe the intersection of develop-ment with ecological conditions and processes of interestIn a simple application we combined projected buildingcounts with a map of current mule deer winter rangeavailable from the Oregon Department of Fish and Wild-life In our four-county study region mule deer winter

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

encroachment leads to increased property damage fromwildlife-browsing landscaping for example-which cancause tension between landowner interests and State muledeer population targets

The extent and nature of development within mule deerwinter range in central Oregon would seem to call for morecomprehensive investigation of resulting habitat impactsToward that end another team of scientists has initiated acompanion landscape pattern analysis to examine habitatfragmentation forage quality hiding and thermal coverimpacts by combining development predictions withdetailed vegetation data (Duncan and Burcsu 2010) TheOregon Department of Fish and Wildlife also has initiatedradio-collar tracking of mule deer to better understandmigration patterns and disturbance impacts Developmentpredictions provided in this paper enable other researchersto identify favorable locations for more geographicallyfocused studies of these fine-scaled habitat impactsAdditional analyses could examine the extent of distur-bance zones associated with building densities (egVogel 1989 Theobald and others 1997) and incorporateempirical simulation of changes in winter range extent overtime based on dynamic modeling of vegetation wildfireand other factors (eg Hemstrom and others 2007)Although together these efforts can provide a richer bodyof information with which to evaluate developmentimpacts and define appropriate policy and managementresponses they are unlikely to change the basic result thatdevelopment is leading to habitat loss

Conclusions and Research Implications

We have presented a relatively simple way to characterizethe spatial distribution of forest and range land development

in a four-county region of central Oregon (USA) using datadescribing building densities population growth topogra-phy and other factors The estimated empirical model andresulting development forecasts can be combined with otherinformation and models characterizing ecological conditionsand processes and wildfire to inform ongoing multidisci-plinary landscape analyses in the region Because thebuilding count data and econometric approach used in theanalysis enable development to be characterized at relativelyfine spatial scales the method is sufficiently flexible toaccommodate integration with other models at a variety ofdevelopment thresholds and spatial scales For examplemuch of the IMAP analysis likely will aggregate modeloutput-including and use model output--at the hydro-logical unit code (or HUC) four level However the ability todescribe finer degrees of development in terms of actual andpredicted building densities also is useful if analysts are toexamine ecological conditions and processes across devel-opment gradients (eg Wimberly and Ohmann 2004) Ourexample application shows how model outputs can becombined with existing habitat or other resource maps toprovide policymakers and managers with initial and timelyinformation about development and its effects regardinghabitat and other resource issues as they await the comple-tion of other study components

Our analysis indicates that continued developmentencroachment onto central Oregon mule deer winter rangeis likely through 2040 with expansion of new developmentout from existing urban areas as well as infill developmentespecially along major transportation corridors Relativelysimple applications such as these can help policy makers andmanagers begin to anticipate and respond to potential futuredevelopment even as more comprehensive analysis ofhabitat fragmentation effects may be unavailable Forexample resource managers may want to initiate or expandefforts to work with landowners local land use planningofficials and nonprofit conservation organizations to con-sider what combination of planning and programmaticresponses are warranted given anticipated developmentimpacts on winter range Modifications to existing land usezoning and the targeted purchase of conservation easementsand land for preservation and management are just a fewactions that could be taken now to help to maintain existingmigration corridors and minimize the extent of disturbancezones associated with new development Policy makersmight consider providing more consistent or increasedfunding to existing state programs that protect and enhancehabitat (eg Deer Enhancement and Restoration ProgramHabitat Improvement Program) and assist landowners whoexperience damage caused by wildlife (eg Green ForageProgram) (Oregon Department of Fish and Wildlife 2003)It is through the cost-effective and timely application ofscience focused on examining critical natural resource

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646

issues where multidisciplinary landscape studies might bestmeet policy and management expectations

Acknowledgments Funding for this article was provided by theInteragency Mapping and Analysis Project (IMAP) USDA ForestService Pacific Northwest Research Station Portland Oregon Wethank Dave Azuma Miles Hemstrom Gary Lettman and two anon-ymous reviewers for helpful comments

References

Azuma DL Dunham PA Hiserote BA Veneklase CF (2004) Timberresource statistics for eastern Oregon 1999 Resource bulletinPNW-RB-238 USDA Forest Service Pacific NorthwestResearch Station Portland OR p 42 hltplwwwfsfeduspnwpubspnw_rb238pdf

Barbour RJ Hayes JL Hemstrom MA (2007) The Interior NorthwestLandscape Analysis System a step toward understandingintegrated landscape analysis Landscape and Urban Planning80333-344

Bockstael NE (1996) Modeling economics and ecology the impor-tance of a spatial perspective American Journal of AgriculturalEconomics 78 1168-1180

Cohen JA (1960) A coefficient of agreement for nominal scalesEducational and Psychological Measurement 2037-46

Duncan J Burcsu T (2010) Impacts of future residential developmenton the spatial pattern of mule deer habitat in central OregonManuscript in preparation On file with T Burcsu PacificNorthwest Research Station 620 SW Main Street Suite 400Portland OR

Greene WH (1997) Econometric analysis Prentice Hall UpperSaddle River New York

Greene WH (1998) LIMDEP version 70 users manual revised ednEconometric Software Inc Plainview NY

Hemstrom MA Merzenich J Reger A Wales B (2007) Integratedanalysis of landscape management scenarios using state andtransition models in the upper Grande Ronde River SubbasinOregon USA Landscape and Urban Planning 80(3) 198-211

Judson DH Reynolds-Scanlon S Popoff CL (1999) Migrants toOregon in the 1990s working age near-retirees and retireesmake different destination choices Rural Development Perspec-tives 1424-31

Kline JD (2005) Forest and farmland conservation effects of Oregons(USA) land use planning program Environmental Management35(4)368-380

Kline JD Azuma DL Moses A (2003) Modeling the spatiallydynamic distribution of humans in the Oregon (USA) CoastRange Landscape Ecology 18(4)347-361

Kline JDMoses A Lettman G Azuma DL (2007) Modeling forest andrangeland development in rural locations with examples fromeastern Oregon Landscape and Urban Planning 80(3)320-332

Kline JD Lettman GJ Hemstrom MA (2010) Lessons for landscapeplanning and ecological assessments In Brouwer F Goetz SJ(eds) The dynamics of land use and ecosystem services atransatlantic multidisciplinary and comparative approachSpringer New York NY

Land Conservation and Development Committee (LCDC) (1992)Acknowledgement scoreboard May 12 1992 Oregon Depart-ment of Land Conservation and Development Salem Oregon

Laskin D (2004) A town thats more than a pretty face New YorkTimes March 7

Lettman GJ (2004) Land-use change on non-federal land in easternOregon 1975-2001 Oregon Department of Forestry SalemOregon p 42 httpwwworegovODFSTATE_FORESTSFRPdocsEOR DZpdf

Nelson GC Hellerstein D (1997) Do roads cause deforestation Usingsatellite images in econometric analysis of land use AmericanJournal of Agricultural Economics 7980-88

Oregon Department of Fish and Wildlife (2003) Oregons Mule DeerManagement Plan Oregon Department of Fish and Wildlife SalemOR p 29 httpwwwdfwstateoruswildlifemanagement_plansdocsMuleDeerPlanFinalPDF

Oregon Department of Fish and Wildlife (2009a) Mule Deer InitiativePlanning Committee meets May 28 in Prineville News releaseMay 22 Oregon Department of Fish and Wildlife Salem ORhttpwwwdfworusnews2009may052209c asp

Oregon Department of Fish and Wildlife (2009b) Oregon Mule DeerInitiative Oregon Department of Fish and Wildlife Salem ORhttpwwwdfwstateoruswildlifehot_topicsmule_deer_initiativeasp

Oregon Office of Economic Analysis (2004) Forecasts of Oregonscounty populations and components of change 2000-2040Department of Administrative Services Salem Oregon

Proehl RS (2009) Certified population estimates for Oregon andOregon Counties Population Research Center College of Urbanand Public Affairs Portland State University Portland OR p 3

Preusch M (2004) Journeys 36 hours bend ore New York TimesOctober 15

Spies TA Johnson KN Burnett KM Ohmann JL McComb BCReeves GH Bettinger P Kline JD Garber-Yonts B (2007)Cumulative ecological and socio-economic effects of forestpolicies in coastal Oregon Ecological Applications 17(1)5-17

Stein SM Alig RJ White EM Comas SJ Carr M Eley M ElverumK ODonnell M Theobald DM Cordell K Haber J BeauvaisTW (2007) National forests on the edge development pressureson Americas national forests and grasslands General TechnicalReport PNW-GTR-728 USDA Forest Service Pacific North-west Research Station Portland OR p 26

Theobald OM Miller JR Hobbs NT (1997) Estimating the cumula-tive effects of development on wildlife habitat Landscape andUrban Planning 3925-36

US Department of Commerce Census Bureau (2000) Populationestimates for states counties places and minor civil divisionsannual time series US Department of Commerce WashingtonDC

Vogel WO (1989) Response of deer to density and distribution ofhousing in Montana Wildlife Society Bulletin 17406-413

Wear DN Bolstad P (1998) Land-use changes in southern Appala-chian landscapes spatial analysis and forecast evaluationEcosystems 1575-594

Wimberly MC Ohmann JL (2004) A multi-scale assessment ofhuman and environmental constraints on forest land coverchange on the Oregon (USA) coast range Landscape Ecology19631-646