using a gis model to assess terrestrial salamander

Journal of Environmental Management (2001) 63, 281–292doi:10.1006/jema.2001.0482, available online at http://www.idealibrary.com on

Using a GIS model to assess terrestrialsalamander response to alternative forestmanagement plans

Eric J. Gustafson†*, Nathan L. Murphy‡ and Thomas R. Crow§

†USDA Forest Service, North Central Research Station, 5985 Highway K, Rhinelander,WI 54501, USA‡Dept. of Biology, Jordan Hall 142, 1001 E. Third St., Bloomington IN 47405, USA§USDA Forest Service, North Central Research Station, 1831 Highway 169 E. Grand Rapids,MN 55744, USA

Received 19 February 2000; accepted 8 June 2001

A GIS model predicting the spatial distribution of terrestrial salamander abundance based on topography and forest agewas developed using parameters derived from the literature. The model was tested by sampling salamander abundanceacross the full range of site conditions used in the model. A regression of the predictions of our GIS model against thesesample data showed that the model has a modest but significant ability to predict both salamander abundance andmass per unit area. The model was used to assess the impacts of alternative management plans for the Hoosier NationalForest (Indiana, USA) on salamanders. These plans differed in the spatial delineation of management areas where timberharvest was permitted, and the intensity of timber harvest within those management areas. The spatial pattern of forestopenings produced by alternative forest management scenarios based on these plans was projected over 150 years usinga timber-harvest simulator (HARVEST). We generated a predictive map of salamander abundance for each scenario overtime, and summarized each map by calculating mean salamander abundance and the mean colonization distance (averagedistance from map cells with low predicted abundance to those with relatively high abundance). Projected salamanderabundance was affected more by harvest rate (area harvested each decade) than by the management area boundaries.The alternatives had a varying effect on the mean distance salamanders would have to travel to colonize regeneratingstands. Our GIS modeling approach is an example of a spatial analytical tool that could help resource managementplanners to evaluate the potential ecological impact of management alternatives. 2001 Academic Press

Keywords: spatial models, GIS, forest management, risk assessment, terrestrial salamanders, timberharvesting, fragmentation, biological diversity, landscape change, land use planning.

Introduction

The relationship between pattern and process isthe focus of landscape ecology, but the practi-cal application of landscape ecology principles byland managers has been difficult due to a lack ofanalytical and spatial tools. Evaluation of alter-native management practices to provide multiplevalues and benefits requires both spatial and tem-poral information. For example, timber harvesting

Ł Corresponding author. Email: [email protected]

profoundly affects the composition and structureof forested landscapes in both time and spaceby modifying vegetation and habitat (Franklinand Forman, 1987). These patterns have ecolog-ical consequences for species and ecological com-munities. Evaluating the consequences of timbermanagement is difficult without large-scale contex-tual (spatial) and long-term (temporal) informationabout the patterns expected under managementalternatives.

Computer simulation offers a practical approachfor generating patterns expected under specificmanagement strategies (Li et al., 1993; Gustafson

0301–4797/01/110281C12 $35.00/0 2001 Academic Press

282 E. J. Gustafson et al.

and Crow, 1994; Spies et al., 1994). These expectedpatterns can be evaluated using other models thatpredict the biological response of organisms to spa-tial patterns. To test this premise, we developedand tested a GIS model of the spatial distribu-tion of terrestrial salamanders based on vegetationcomposition, forest age, and topography. The modelcan generate predictions over very large areas, anduses spatial data that are readily available formany managed forests. We studied salamandersbecause they represent physiologically sensitivespecies that are responsive to changes in micro-climate and forest floor debris caused by timbermanagement activities. We used a timber har-vest simulation model (HARVEST; Gustafson andCrow, 1996; Gustafson, 1999) to generate stand-age maps of future forest landscapes under alter-native management strategies developed for theHoosier National Forest in southern Indiana (USA)(Figure 1). We then applied the salamander GISmodel to these landscape patterns to assess thepotential impacts of the management alternativeson terrestrial salamanders.

Situation

The National Forest Management Act of 1976requires the US Department of Agriculture For-est Service to complete a plan for each NationalForest that constitutes a comprehensive statementof management direction. The Hoosier NationalForest (HNF) published its first Land and ResourcePlan in 1985, and it included even-age management(mostly by clearcutting) over most of the Forest(USDA Forest Service, 1985). Because of strongpublic opposition, the 1985 Plan was replaced by

a 1991 Forest Plan Amendment (USDA ForestService, 1991) that placed greater emphasis on pro-tecting and managing ecosystems, and on providinga visually pleasing landscape. In contrast to the1985 Plan, the 1991 Amendment called for primar-ily uneven-age management using group selection(removal of small (�0Ð4 ha) patches of trees) over amuch smaller portion of the land base. Gustafsonand Crow (1996) used a timber harvest simulator(HARVEST) to compare the landscape structureresulting from these two alternative managementstrategies. Although realistic changes in landscapestructure can be simulated using a harvest simula-tor such as HARVEST, interpreting the ecologicalconsequences of these spatial patterns remains dif-ficult. Habitat suitability for many species may bedetermined by multiple characteristics of vegeta-tion and site conditions, and these may interact incomplex ways to determine population abundance.

Amphibian species associated with forests arethought to be sensitive to forest management(Blaustein and Wake, 1990). In the case of terres-trial salamanders there is sufficient informationavailable in the literature (e.g. Heatwole, 1962;Bennett et al., 1980; Duellman and Trueb, 1986;Pough et al., 1987; deMaynadier and Hunter, 1995)to construct spatially explicit hypotheses aboutabundance, although this information was devel-oped from studies conducted outside Indiana. Inthis study, we used published data to formulatea hypothesis about how vegetation and topog-raphy interact to determine site moisture andthereby control terrestrial salamander abundance.We formalized this hypothesis mathematically ina GIS model that generates a map of the spatialdistribution of predicted salamander abundanceacross a managed landscape. Concern has also

TESTDATA

INDIANA

PRUN

LRIV

TELL

UNITED STATES

Figure 1. Map showing location of the study areas within the Hoosier National Forest, and the area where salamanderswere sampled to test the GIS model. LRIV, PRUN, and TELL are study areas.

Salamander response to forest management 283

been raised about the limited ability of salaman-ders to recolonize sites disturbed by clearcutting(Blaustein et al., 1994; Petranka, 1994). Our modeloutput can also be used to assess the averagedistance colonizers must travel to reach young,disturbed sites as they mature.

The HNF consists primarily of the centralhardwood forest type dominated by maple (Acersaccharum), beech (Fagus grandifolia), oak (Quer-cus spp) and hickory (Carya spp) with embed-ded pine (Pinus) plantations. Pine occurs onlyin plantations. This area is an unglaciated, den-dritic network of hills and ravines (Schneider,1966), forming two major ecological subsections,the Brown County Hills and the Crawford Uplands.Ravines are typically about 60 meters deep.

Objectives

Our objectives in this study were to (1) construct aGIS model to predict salamander abundance on theHNF, (2) test the model by sampling salamanderabundance across the full range of site conditionsused in the model, and (3) use the model to comparethe predicted responses of terrestrial salamandersto different timber management strategies on theHNF, as simulated by the HARVEST model.

Methods

Predicting salamander abundance

Predictive model

We constructed a theoretical GIS model predictingsalamander abundance from site conditions impor-tant for terrestrial salamanders, related primarilyto physiographic features and stand managementhistory. These features are commonly representedwithin a GIS as digital elevation models and standage maps. Salamanders require moisture to main-tain respiration, as do most amphibians, and theyare easily desiccated (Duellman and Trueb, 1986).The risk of desiccation limits the surface activityof salamanders and therefore their foraging oppor-tunities. Areas with lower moisture levels shouldsupport lower densities of salamanders. Site mois-ture is therefore a critical factor to predict habitatsuitability for salamanders. We assumed site mois-ture was linearly related to slope position (seebelow) (J. Van Kley, 1993, pers. comm.). Site mois-ture is further affected by slope aspect, and the

relationship between aspect and moisture varieswith physiographic province (Van Kley, 1993).We assumed that site moisture was 40% less onsouthwest-facing slopes, and 20% less on CrawfordUpland sites, than on northeast-facing slopes onBrown County Hills sites (J. Van Kley, 1993, pers.comm.).

Slope position was defined relative to the bottom(D1) and the top (D100) of the slope and wasderived from a 30-m digital elevation model asfollows. We developed GIS functions to identifyridge tops and valley bottoms that were then storedas separate GIS layers. Using a proximity function,we created layers representing the distance of eachpixel from the nearest ridge (r) or valley bottom (b),respectively. Slope position (s) for each pixel wasthen calculated by

sD[b/.rCb/]Ł100.

We applied a cosine transformation of aspect (a)so that aD0Ð0 for northeast slopes (azimuthD45°)and aD2Ð0 for southwest slopes (Beers et al., 1966).Predicted relative (rangeD1�100) site moisture(m) for each pixel is

mDjs�101j�.aŁp/

where pD20 on Brown County Hills sites and pD10on Crawford Upland sites.

Site disturbance by clearcutting has also beenshown to affect salamander abundance in easternforests (Pough et al., 1987; Petranka et al., 1993),and we included the effects of stand age onsalamander abundance in our model. We relatedstand age (x) to salamander numbers (y) by fitting athird order polynomial (Figure 2) to data on mean

100

20

Stand age (y)

Mea

n n

o. s

alam

ande

rs/p

lot

10

0 80604020

y = 3.17 + 1.29e-2x + 7.41e-3x2 – 6.25e-5x3

Figure 2. Polynomial used to relate salamander abun-dance to stand age. Data points were derived from Figure 3of Petranka et al., 1993. Stands aged>80 yr were assumedto have the salamander abundance represented by thedashed line.


number of salamanders (all species) in westernNorth Carolina study plots published by Petrankaet al. (1993).

yD3Ð17C0Ð0129xC0Ð0074x2�0Ð00006x3

These data are also consistent with the resultsobtained by Pough et al. (1987) in central NewYork. However, we constrained the function sothat stands aged >80 yr (Figure 2) were calculatedas 80-yr-old stands because these studies suggestthat the observed decline in abundance in stands>80 yr is probably due to sampling variation, andnot biologically significant.

Field surveys on the HNF have documented apaucity of all amphibians in pine stands (Ewert,M. A., J. N. Barron, C. R. Etchberger, C. L. Grimes,E. Levri, M. J. Lodato, B. Nelson and C. E. Nel-son unpublished report), which is consistent withother studies (Bennett et al., 1980; Pough et al.,1987; deMaynadier and Hunter, 1995). Pine standstend to be drier and have less understory veg-etation than deciduous areas (Van Kley et al.,1995). Based on this information, we assumedthat pine stands support only 15% of the salaman-der abundance predicted for 80-yr-old deciduousstands (Figure 2). Finally, we related salaman-der abundance to moisture (m) by assuming thatrelative salamander abundance increased linearlyas site wetness increased. The slope and inter-cept of this relationship were estimated using dataon salamander abundance in California and Ore-gon (Welsh and Lind, 1988). The model predictsrelative salamander abundance (sa) using valuesderived from stand age (y) and the digital elevationmodel (m):

saDyŁ.0Ð983C0Ð017m/.

sa was normalized to range between 1 and 100,and output to a GIS layer. These relationshipsare shown graphically in Figure 3. The modelconsidered only forested habitat, so non-foresthabitats were assigned zero. The model wasimplemented using PC-ERDAS (v.7Ð5).

Predictive model testing

To verify the predictive ability of the model forthe HNF we collected empirical data on terres-trial salamander abundance and compared themto the predictions of the model. We sampled ter-restrial salamander abundance and mass in Aprilof 1997 and 1998 in the Pleasant Run Unit of the

100

Stand age (y)

80

60

40

20

100

1102030405060708090

1009080706050403020101Relative moistu

re (%)

100

80

60

40

20

sa

Figure 3. Relationship between relative salamanderabundance (sa) and relative site moisture (m) and standage (x), as formulated in the salamander model.

HNF (Figure 1). Four species of salamanders werecaptured in the study area (Plethodon cinereus, P.dorsalis, P. glutinosus and Eurycea longicauda),and all were captured at least once. E. longicaudais a stream salamander that ventures a little fur-ther from streams than other stream salamandersin the study area. We did not study the pond-breeding salamanders because the Ambystomaspp. live in burrows that could not be sampled,and the terrestrial stage of Notopthalmus viri-descens was uncommon in the study area. Samplingoccurred along transects within randomly selectedforested stands. Transects were stratified byslope position (top, middle, bottom), aspect (north(316–45 degrees azimuth), east (46–135 degrees),south (136–225 degrees), west (226–315 degrees)),and stand type. Stand types included pine planta-tions and three deciduous types–young (15–25 yrafter clear cutting), intermediate (26–60 yr sincecutting), and mature (>60 yr since cutting). Tran-sects were 100 m long and followed topographicalcontour lines. This transect length was sufficient todisperse the plots and sample the range of condi-tions, but short enough to stay within a stand type.Transects were located randomly within an areahaving a given slope position and aspect. Twentyrandomly spaced 2-m2 plots (i.e. 1Ð4 mð1Ð4 m) weresampled along each transect. In 1997, 460 plotswere sampled on 23 transects, and in 1998, 661plots were sampled on 27 different transects, givinga total of 1121 plots (2242 m2 of forest).

In each plot, cover objects were overturned toexpose salamanders. Next, all litter within the plot


was scraped down to the mineral soil and piledin the middle of the plot. The pile of litter wasthen sifted to collect all salamanders. Salamanderswere identified to species, and weighed using anOhaus spring scale (š0Ð1 g). All salamanders werereleased immediately and litter and cover objectswere replaced.

Abundance was quantified as the mean numberof individuals and the mean salamander massper plot for each transect (ND20 plots). Wecompared the results of these surveys to themodel by plotting the abundance measures againstthe model prediction at the center point of thetransect. Because our test data did not use thesame units as the model predictions (numbers andmass vs. relative abundance), we could not testfor an expected slope between empirical and modelresults (e.g. slopeD1Ð0). Instead, after verifyingthe Gaussian nature of the variables, we fitted aregression line to these plots and tested the nullhypothesis that the slopeD0.

Simulating alternative timberharvest strategies

Timber harvest simulation model

HARVEST is a timber harvest simulator, linkedto a GIS, that simulates the allocation of standsfor timber harvest in a spatially explicit man-ner through time (Gustafson and Crow, 1996;Gustafson, 1999). The model allows flexible simula-tion of harvest activity using parameters commonlyfound in National Forest Plan standards and guide-lines. Only four input maps are required: stand age,forest type, management area boundaries (spatialzones with specific management objectives), andstand identifier value (for model bookkeeping pur-poses). It produces landscape patterns that havespatial attributes resulting from the initial land-scape conditions and the planned managementactions. The model does not attempt to optimizetimber production or quality, and it ignores manyof the site attributes considered by forest planners,such as visual objectives and road access. Instead,it stochastically mimics the allocation of standsfor harvest by forest planners, using the parame-ters of the broad management strategies. Modelingthis process allows us to link harvest activitieswith landscape patterns. In our study, we consid-ered only harvests that produce openings >0Ð1 hawithin the forest (clearcutting, shelterwood andgroup selection). The effect of single-tree selection

on salamander abundance appears to depend onthe species (Pough et al., 1987; Pais et al., 1988),and we assumed that single-tree selection wouldproduce little change in the abundance of salaman-ders as a group (N. Murphy, pers. observ.).

HARVEST allows control of the size distributionof harvests, the total area of forest to be harvested,the rotation length (by specifying the minimumage on the input stand map where harvests maybe allocated), and the width of buffers left aroundharvest openings. Group selection is implementedon the HNF such that one-sixth of a stand is cut oneach entry, and entries occur every 20–30 yr (USDAForest Service, 1991; T. Thake, pers. comm.).The model selects stands for group selectionfrom those stands with an age greater than theprescribed rotation length; it then tracks standsmanaged by group selection, ensuring re-entry at30-yr intervals, and prevents other treatmentsin those stands. Within group-selected stands,openings are placed randomly, with at least 30 m (1pixel) between openings. HARVEST is describedin detail in Gustafson and Crow (1996) andGustafson (1999), and is available on the Internet(http://www.ncrs.fs.fed.us/harvest/harvhome.htm).

Simulation study areas

Study areas were selected in three of thefour administrative units of the HNF [PleasantRun (PRUN) study area (34 053 ha), Lost River(LRIV, 38 822 ha), and Tell City (TELL, 49 515 ha)](Figure 1). Stand age maps of National Forest landwithin the three study areas were digitized frompaper-based and mylar planning maps and grid-ded to 30 m cells, and ages were calculated as of1988. We used the management area (MA) bound-aries of the two published management plans. EachMA has a specific management objective, and theMA boundaries encompass tracts to be managedto meet that objective. Several disjunct polygonsof a particular MA may be designated within anadministrative unit. The objectives of each MA andthe arbitrary decimal designations (e.g. 3Ð1, seeTable 1) are used consistently among all NationalForests. The MA boundaries within each studyarea were manually transferred to 1:100 000 USGeological Survey maps, digitized, and griddedto 30-m cells. Land use on non-Forest Serviceland was derived from Landsat Thematic Map-per (TM) imagery collected in 1988, as describedin Gustafson and Crow (1994). Classes distin-guished were forest, grazed pasture/young winter


Table 1. Harvest intensities as derived from the 1985 Hoosier National Forest Plan and the 1991 Forest PlanAmendment

Model parameter 1985 MA1 1991 MA1

2Ð1 3Ð1 3Ð2 6Ð1 Total 2Ð8Mean clearcut opening size (ha) – 7Ð0 4Ð9 4Ð0 – 2Ð8Mean group opening size (ha) 0Ð4 – – – – 0Ð2Maximum opening size (ha) 0Ð7 10Ð8 7Ð2 5Ð4 – 4Ð0Total harvested/decade2 (ha) 96Ð0 2360Ð6 2890Ð0 363Ð0 5709Ð6 2504Ð8Harvest rate/decade3 (%) 5Ð4 11Ð5 7Ð8 7Ð5 9Ð5 6Ð4Timber yield/decade2 (Mbf) 180 4928 6006 767 11 881 4120Rotation length (years) 80 80 120 120 – 80Buffer width4 (m) 30 30 30 30 – 30

1Forest Plan management areas. The (arbitrary) decimal designation is that used by the Forest Service.2Represents harvest activity by management area across the entire Forest.3Represents percent of forest within the management area that is harvested each decade.4The width of buffers left between harvest allocations and other harvests, streams, and openings.

wheat, meadow/hay fields, row crops, water, androad/developed.

Simulation experiments

We simulated five specific management alterna-tives on each study area: (1) the 1985 Forest Plan(USDA Forest Service, 1985), (2) the 1991 For-est Plan Amendment (USDA Forest Service, 1991),(3) the management area (MA) boundaries of the1985 Plan with the harvest intensity of the 1991Amended Plan (85 Intensity-91 Map), (4) the MAboundaries of the 1991 Plan Amendment with theharvest intensity of the 1985 Plan (91 Map-85Intensity), and (5) no harvest (i.e. no openingsproduced or maintained on National Forest land).The 91 Map-85 intensity scenario used the param-eters of MA 3Ð1, the most intense of the 1985Plan harvest scenarios (Table 1), to provide thegreatest contrast with the 1991 plan harvest inten-sity (MA 2Ð8). We simulated 150 yr of managementunder each alternative; given the low variabilityof the spatial patterns produced by the simula-tions (see Results), three replicates were adequateto ensure robust results. Wildlife openings weremaintained throughout the simulations except inthe no-harvest scenario, where conversion to for-est was simulated and closed canopy forest wasreached in two decades. All harvested stands regen-erated to forest, and pine stands harvested wereassumed to convert to hardwood, as proposed inboth management plans. We did not simulate har-vest activity on privately owned inholdings to avoidconfounding our assessment of the effects of publicland management strategies. The existing patternof forest openings on private land was derived

from the TM imagery, and these conditions weremaintained throughout the simulations.

Quantification of experimental results

We applied the predictive salamander model toeach of the modified stand age maps produced foreach management alternative using HARVEST,producing a map showing predicted salamanderabundance for each. These maps were quantifiedby calculating the mean predicted abundance valueof all forest (non-zero) pixels, and the modelpredictions were plotted as a function of time.To establish the pre-simulation trend, the 1988stand age maps were successively recoded bydecrementing by 10 the age value (A) at time tfor each pixel j to produce the stand age mapof the previous decade (i.e. Aj.t�10/DAj.t/�10). Werepeated this process to establish the pattern offorest openings since 1948. Stands that reached anage of zero during this process were assumed tohave been mature forest (>80 yr) at the time theywere cut.

To assess the fragmenting effect of timber har-vest on salamander habitat quality, we calculatedthe average distance between cells where sala-mander abundance was reduced by timber harvest(sa�50%), and those likely to contain potential col-onizers (sa>50%); sa is a relative value, and wearbitrarily chose the 50% threshold. Some of thecells with low sa values were pine plantations, butthese remained constant among alternatives. Weused the GIS to calculate the distance from pixelsrepresenting low salamander abundance (sa�50%)to the nearest pixel of relatively high salamanderabundance (sa>50%). We then calculated the meandistance value of these pixels.


Statistical analysis

To evaluate the relative effects of harvest inten-sity and the zonation of harvest activity (restrictedspatially by MA boundaries) on salamander abun-dance and colonization distance, we used arepeated measures ANOVA to test for treatmenteffects reflecting harvest intensities (INTENSITY),management area boundaries (MAP), and time(DECADE). The time-periods were included in theanalysis to account for potential autocorrelationamong spatial pattern measures across successivedecades.

Results

Predictive model testing

Over 500 salamanders were captured during sam-pling over a 2 year period (243 P. cinereus, 245P. dorsalis, 24 P. glutinosus and 4 E. longicauda).The regression of the predictions of our GIS modelagainst these sample data showed that the modelhas a modest, but significant ability to predictboth salamander abundance (FD12Ð2546,Pr>FD0Ð0010) and mass per unit area (FD9Ð2424,Pr>FD0Ð0038) (Figure 4). The analysis of varianceindicated that the model explained about 20% ofthe total variance for salamander abundance, andabout 16% for salamander mass (Table 2). Thenull hypothesis that the model provides no pre-dictive ability (i.e. slopeD0) was rejected for bothsalamander abundance (TD3Ð5,Pr>TD0Ð001) andsalamander mass (TD3Ð04,Pr>TD0Ð004).

Simulation experiments

The replications of the simulations for each sce-nario produced little variability in mean salaman-der abundance. Harvest levels in all MAs werehigh enough that most stands > rotation age wereharvested each decade. Consequently, the opportu-nity for stochastic variation in the spatial patternof harvest openings was limited. Even the 1991Plan had a high number of stands harvestedbecause the reliance on group selection requiredharvest in many stands to achieve the specifiedtimber production. Although harvest intensity washigh within MA 2Ð8 (1991 Plan), much less ofthe land base was dedicated to timber production(MA 2Ð8) than under the 1985 Plan. Alternativeswith lower harvest intensities within MAs would

100

2.0

Model prediction (relative value)

Mea

n m

ass/

plot

(gr

ams)

1.5

806040200

1.0

0.5

R2 = 0.16

0

2

Mea

n n

o. o

f in

divi

dual

s/pl

ot

1

R2 = 0.20

Figure 4. Plot of empirical test data against modelpredictions. The solid is a regression line fit to the data,and the dotted line represents the 95% confidence intervalfor the slope of the regression line. Relative abundance(x-axis) is scaled as a percent of the maximum possibleabundance prediction.

result in greater spatial pattern variability. Thevariability in our results was too low to showclearly with error bars on line graphs, so errorbars are not shown. The standard deviation fromthe mean salamander abundance never exceeded0Ð02.

Table 2. Analysis of variance comparing empirical mea-sures of salamander abundance and the relative abun-dance predictions of the salamander model

Source df SS F Prob>F R2

NUMBER OF SALAMANDERSSalamander model 1 1Ð263 12Ð28 0Ð001Error 48 4Ð936

Total 49 6Ð199 0Ð204

SALAMANDER MASS (G)Salamander model 1 0Ð738 9Ð21 0Ð004Error 48 3Ð848

Total 49 4Ð586 0Ð161


Mean salamander abundance was affected moreby harvest intensity than by the management areaboundaries in our simulations. Examination of thesums of squares of the main effects shows thatharvest intensity (INTENS) accounted for 68–76%of the variability in predicted mean salamanderabundance, while management area boundaries(MAP) accounted for only 4–16% of the variabilityand DECADE accounted for only 3–17% (Table 3).The only factor affecting salamander abundancethat varied among simulations was stand age. Thetotal area harvested each decade determined thefrequency distribution of stand age classes, andthe spatial distribution of age classes across thelandscape was determined by the managementarea boundaries.

Mean salamander abundance was consistentlyhigher for scenarios using the 1991 Plan har-vest intensities than those using the 1985 Planintensities (Figure 5) because less area in youngage classes was produced by the 1991 Plan. The1985 Plan did not reduce salamander abundancebelow levels observed between 1948 and 1988on the PRUN and LRIV study areas, and itreduced salamander abundance only slightly below

Table 3. Analysis of variance comparing the effects ofharvest intensity (INTENSITY), the zonation of harvestactivity by management area boundaries (MAP), and thetime period simulated (DECADE) on relative salamanderabundance across the landscape. Analysis includes threereplicates of simulations conducted for 15 decades on thethree study areas within the HNF

Source Salamander abundance (sa)

df SS F Prob>F R2

LRIV1

MAP 1 41Ð6931 218Ð09 0Ð0001INTENSITY 1 332Ð3570 1738Ð51 0Ð0001DECADE 14 31Ð0209 11Ð59 0Ð0001Error 163 31Ð1613

Total 179 436Ð2324 0Ð93

PRUN1


Total 179 1017Ð9460 0Ð91

TELL1


Total 179 1267Ð4118 0Ð92

1LRIV, PRUN, and TELL are study areas.

522140

62

Year

TELL

Mea

n s

alam

ande

r ab

un

dan

ce (

%)

1940 2100206020201980

60

58

56

54

66 LRIV

56

64

62

60

58

62PRUN

52

60

58

56

54

No Harvest91 Plan85 Plan

91 Map-85 Intensity85 Map-91 Intensity

50

Figure 5. Change in the predicted mean relative abun-dance of salamanders over time resulting from simulationof management strategies. Relative abundance is scaledas a percent of the maximum possible abundance pre-diction. Simulations began using 1988 stand data. PRUN,LRIV, and TELL are study areas.

pre-simulation levels on the TELL study area. Theearly (1988–98) decline of the no-harvest scenariowas due to the closure of wildlife openings, a changefrom non-forest to young forest, which initiallydecreased the mean age of all forest. Differencesamong study areas were generally caused by differ-ences in the frequency distribution of age classeson the study areas (Gustafson and Crow, 1996),because the distributions of slope aspects and slopepositions were similar among study areas.

Areas of relatively high salamander abundancewere highly interconnected on all study areas(Figure 6). Increases in mean colonization distances(Figure 7) were related to increased area of


Figure 6. Maps of the spatial distribution of predicted mean relative salamander abundance at the end of simulation(150 yr) of the 1985 and 1991 Plans. Relative abundance is scaled as a percent of the maximum possible abundanceprediction. PRUN and TELL are study areas. The LRIV study area is not shown.

young stands with relatively low salamander abun-dance (see Figure 5). These young stands oftenenlarged existing areas of low salamander abun-dance resulting in increased mean colonizationdistance. Declines before simulation were probablydue to aging of young stands and a decrease in theaverage size of harvest treatments.

Discussion

Our salamander model provides predictive insightinto the effect on salamander abundance of featuresthat vary at a coarse scale across a landscape(i.e. topography and forest stand age). Our modelpredicts the expected salamander abundance as


742140

88

Year

TELL

Mea

n c

olon

izat

ion

dis

tan

ce (

m)

1940

76

2100206020201980

86

84

82

80

70LRIV

40

60

50

64PRUN

54

62

60

58

78

52

56

No Harvest91 Plan85 Plan

91 Map-85 Intensity85 Map-91 Intensity

Figure 7. Change in the predicted mean colonization dis-tance over time resulting from simulation of managementstrategies. This distance is the mean distance of pixelsrepresenting relatively poor salamander habitat (sa�50)from the nearest pixel of relatively good salamander habi-tat (sa½50). Simulations began using 1988 stand data.PRUN, LRIV, and TELL are study areas.

a function of topography (i.e. slope position andaspect) and forest age. It generates predictionsover very large areas, and uses spatial data thatare readily available for many managed forests.Although our model was constructed using datafrom sites in various parts of the United States,its ability to predict salamander abundance inIndiana suggests that the model incorporatessome of the important features of salamanderbiology that make it broadly applicable. Micrositecharacteristics (e.g. abundance of coarse woodydebris and litter depth) and weather (Crump,1994) may be a source of much of the unexplainedvariation in our model (N. Murphy, unpublished

data). Incorporating microclimate edge effects (e.g.Daolan and Chen, 2000) may result in improvedprediction of spatial effects. For better site-specificprediction, additional information is required thatmay not be readily available from existing datasources. The model may also be relevant forother moisture sensitive species such as frogs.Conversely, it may also be useful for predictinghabitat quality for species that favor dry habitats,such as reptiles. This modeling approach providesa feasible and cost-effective means to assess thepotential impacts on terrestrial salamanders offorest management alternatives formulated duringthe strategic planning process. Predicting theimpact of site-specific, tactical plans, or the effectsof management on a specific salamander speciesmay require more detailed models.

The two Forest Plans provided contrasting land-scape patterns during 150 yr of simulated manage-ment. A strategy of using small, dispersed harvestunits greatly increases fragmentation (Gustafson,1998). However, the mean abundance of salaman-ders was affected more by harvest rates than bythe spatial configuration of harvests because thescale at which salamanders perceive habitat isrelatively small. Abundance was related to forestage, and lower harvest rates resulted in increasedmean age of the forest. The size and proximityof harvests did not appear to exert a non-lineareffect on mean salamander abundance as they dothe existence of forest interior habitat (Gustafsonand Crow, 1996), because no edge effects (affectingareas around the disturbance) were incorporatedinto our model of salamander abundance. The max-imum difference in salamander abundance amongalternatives was relatively low (5–10%, Figure 5).None of the alternatives may represent a threatto the long-term viability of salamander popula-tions, although not enough is known about thesepopulations to draw definitive conclusions. How-ever, as minimum viable population guidelines aredeveloped, these models will provide a useful toolto evaluate whether management alternatives willsustain those population levels, and how those pop-ulations will be distributed across the landscape.

Areas of relatively high salamander abundancewere highly interconnected in all study areas(Figures 6, 7). Site moisture conditions are relatedto topography, and the more moist sites arefound near bottoms and drainage, which areinterconnected because of the stream patternsproduced by geophysical erosion. Areas of goodsalamander habitat (sa>50) almost always formeda matrix in each study area, with areas of poorerhabitat (mostly caused by timber harvesting)


embedded within it (see Figures 6, 7). However,at some point (i.e. higher harvest intensity),this interconnected matrix would likely becomefragmented.

Conclusions

The relationship between pattern and process hasbeen the focus of landscape ecology, but the prac-tical application of landscape ecology principlesby land managers has been difficult due to alack of analytical and spatial tools. Forest man-agement has a profound effect on both terrestrialand aquatic habitats. Forest managers have longconsidered the impacts of management activitieson mammals and birds, but recently have focusedon overall biodiversity, including amphibians. Thequality of terrestrial habitats is critical for main-taining amphibian populations (Marsh and Tren-ham, 2001). The modeling tools we have describedprovide a way to evaluate the ecological conse-quences of management alternatives on specieswhose habitat needs are determined primarily bysite moisture. For example, a manager could ensurethat good habitat will not become disconnected bya management strategy, or evaluate how variousproportions of clearcut and group selection silvi-culture treatments impact salamander abundance.Furthermore, the long temporal scale of forest man-agement effects can be appropriately considered bylinking the two models.

We believe that the approach presented in thisstudy could be adapted to investigate other man-agement questions. Application of these modelsby planners and managers is feasible because themodels use commonly available GIS data, and oper-ate on commonly used computer platforms (e.g.Windows). The user can easily modify the salaman-der model to incorporate other important variablesand new empirical data about the relationshipbetween salamander abundance, stand age, andtopography. We have demonstrated that the out-puts from these models are themselves hypothesesthat can be tested relatively quickly and easily, andthey may be used to provide guidance and feedbackto adaptive management practices.

Spatially explicit simulations models such asHARVEST are needed to evaluate managementalternatives in both time and space, to aid in thedesign of landscapes for multiple benefits. We haveillustrated the utility of combining a simulationmodel with a GIS predictive model based on thebiology of a taxon, for risk assessment. Such ana-lytical tools are needed to help resource planners

develop management plans that will reliably con-serve biological diversity and accommodate soci-ety’s increasing demands for commodity resources.

Acknowledgments

We thank Monica Schwalbach, Ellen Jacquart, LibbyRice, Tom Thake, Mary Stafford, Steve Olsen andDeb Albright of the Hoosier National Forest for theirassistance in providing data and information. We thankKyle Forbes for digitizing maps and John Wright for GISassistance. Bill Hargrove, Rick Howard, Patrick Zollner,Jim Petranka, Vicky Meretsky, Doug Karpa-Wilson, andthree anonymous reviewers provided critical commentson earlier drafts of the manuscript. Lucy Burde gaveeditorial assistance.

References

Beers, T. W., Dress, P. E. and Wensel, L. C. (1966).Aspect transformation in site productivity research.Journal of Forestry 64, 691–692.

Bennett, S. H., Gibbons, J. W. and Glanville, J.(1980). Terrestrial activity, abundance and diversityof amphibians in differently managed forest types.American Midland Naturalist 103, 412–416.

Blaustein, A. R. and Wake, D. B. (1990). Decliningamphibian populations: a global phenomenon? Trendsin Ecology and Evolution 5, 203–204.

Blaustein, A. R., Wake, D. B. and Sousa, W. P. (1994).Amphibian declines: judging stability, persistence,and susceptibility of populations to local and globalextinctions. Conservation Biology 8, 60–71.

Crump, M. L. (1994). Climate and environment. In Mea-suring and monitoring biodiversity: standard methodsfor amphibians (R. W. Heyer, M. Donnelly, R. W. McDi-armid, L. C. Hayek and M. S. Foster, eds), pp. 42–46.Washington DC: Smithsonian Institute Press.

Daolan, Z. and Chen J. (2000). Edge effects in fragmentedlandscapes: a generic model for delineating area ofedge influences (D-AEI). Ecological Modelling 132,175–190.

DeMaynadier, P. G. and Hunter Jr, M. L. (1995).The relationship between forest management andamphibian ecology: a review of the North Americanliterature. Environmental Review 3, 230–261.

Duellman, W. E. and Trueb, L. (1986). Biology ofamphibians. New York: McGraw-Hill.

Franklin, J. F. and Forman, R. T. T. (1987). Creatinglandscape patterns by forest cutting: ecological conse-quences and principles. Landscape Ecology 1, 5–18.

Gustafson, E. J. (1998). Quantifying landscape spatialpattern: what is the state of the art? Ecosystems 1,143–156.

Gustafson, E. J. (1999). HARVEST: a timber harvest allo-cation model for simulating management alternatives.In Landscape Ecological Analysis: Issues and Applica-tions (J. Klopatek, R. H. Gardner, eds), pp. 109–124.New York: Springer-Verlag.

Gustafson, E. J. and Crow, T. R. (1994). Modeling theeffects of forest harvesting on landscape structure and


the spatial distribution of cowbird brood parasitism.Landscape Ecology 9, 237–248.

Gustafson, E. J. and Crow, T. R. (1996). Simulating theeffects of alternative forest management strategieson landscape structure. Journal of EnvironmentalManagement 46, 77–94.

Heatwole, H. (1962). Environmental factors influencinglocal distribution and abundance of the salamanderPlethodon cinereus. Ecology 43, 460–474.

Li, H., Franklin, J. F., Swanson, F. J. and Spies, T. A.(1993). Developing alternative forest cutting patterns:a simulation approach. Landscape Ecology 8, 63–75.

Marsh, D. M. and P. C.Trenham. (2001). Metapopulationdynamics and amphibian conservation. ConservationBiology 15, 40–49.

Pais, R. C., Bonney, S. A. and McComb, W. C. (1988).Herpetofaunal species richness and habitat associa-tions in an eastern Kentucky forest. Proceedings of theAnnual Conference of the Southeast Association of Fishand Wildlife Agencies 42, 448–455.

Petranka, J. W. (1994). Response to impact of timberharvesting on salamanders. Conservation Biology 8,302–304.

Petranka, J. W., Eldridge, M. E. and Haley, K. E. (1993).Effects of timber harvesting on southern Appalachiansalamanders. Conservation Biology 7, 363–370.

Pough, F. H, Smith, E. M., Rhodes, D. H. and Collazo, A.(1987). The abundance of salamanders in forest standswith different histories of disturbance. Forest Ecologyand Management 20, 1–9.

Schneider, A. F. (1966). Physiography. In NaturalFeatures of Indiana (A. A. Lindsey, ed.), pp. 40–56.Indianapolis: Indiana Academy of Science.

Spies, T. A., Ripple, W. J. and Bradshaw, G. A.(1994). Dynamics and pattern of a managed coniferousforest landscape in Oregon. Ecological Applications 4,555–568.

USDA Forest Service (1985). Land and ResourceManagement Plan, Hoosier National Forest. USDepartment of Agriculture, Forest Service, EasternRegion, Hoosier National Forest, Bedford, Indiana,USA.

USDA Forest Service (1991). Plan Amendment, Landand Resource Management Plan, Hoosier NationalForest. US Department of Agriculture, Forest Service,Eastern Region, Hoosier National Forest, Bedford,Indiana, USA.

Van Kley, J. E. (1993). An ecological classification sys-tem for the Central Hardwoods Region: the HoosierNational Forest. Ph.D. Dissertation, Purdue Univer-sity, West Lafayette, Indiana, USA.

Van Kley, J. E., Parker, G. R., Franzmeier, D. P.and Randolph, J. C. (1995). Field guide: ecologicalclassification of the Hoosier National Forest andsurrounding areas of Indiana. USDA Forest Service,Eastern Region, Hoosier National Forest, Bedford,Indiana, USA.

Welsh, Jr, H. H. and Lind, A. J. (1988). Old growth forestsand the distribution of the terrestrial herpetofauna.In Proceedings of the Symposium on management ofamphibians, reptiles and small mammals in NorthAmerica (R. C. Szaro, K. E. Severson and D. R. Patton,eds), pp. 439–455. General Technical Report RM-166. Fort Collins, Colorado, USA: US Department ofAgriculture, Forest Service, Rocky Mountain Forestand Range Experiment Station.

using a gis model to assess terrestrial salamander

Documents