Vegetation characteristics of forest stands used by woodland caribou and those disturbed by fire or logging in Manitoba

This study examined woodland caribou (Rangifer tarandus caribou) in an area known as the Kississing-Naosap caribou range in west central Manitoba. The vegetation characteristics of areas used by caribou and areas disturbed by fire or logging were measured in order to develop a model to estimate habitat quality from parameters collected during stan¬ dard resource inventories. There was evidence that habitat index values calculated using a visual score-sheet index could be used as the basis to relate parameters commonly collected during resource inventories to habitat suitability. Use of this model to select long and short-term leave areas during forest management planning could potentially mitigate some of the negative impacts of forest harvesting. Abundance of arboreal lichen and wind-fallen trees were important predictor variables in the suitability model, but their inclusion did not explain more variance in habitat suitability than models that did not include them. Extreme post-fire deadfall abundance may play a role in predator-prey dynamics by creating habitat that is equally unsuitable for all ungulates, and thus keeping both moose and caribou densities low.


Introduction
Woodland caribou (Rangifer tarandus caribou) are believed to be the late-successional forest ungulate in the boreal forest biome of North America.Numerous researchers have investigated the microhabitat characteristics of areas used by woodland caribou in boreal forest ecosystems over an entire year, as well as during specific seasonal periods, and across various disturbance regimes (e.g.Schaeffer & Pruitt, 1991;Racey et al., 1996;Rettie et al., 1997;Antoniak & Cumming, 1998;Martinez, 1998;Hirai, 1998;Wilson, 2000).
Studies have generally emphasized the importance of the abundance of arboreal and terrestrial lichen as a primary factor associated with microhabitat used by caribou during winter (Antoniak & Cumming, 1998;Martinez, 1998;Wilson, 2001).Snow condi¬ tions have also been considered important during winter, particularly factors contributing to lower Rangifer, Special Issue No. 14: 255-266 levels of snow accumulation (Schaeffer & Pruitt, 1991;Wilson, 2000).The proximity of escape habi¬ tat in the form of water or open bogs and wetlands is thought to be important during the calving and post-calving periods in the spring and summer (Bergerud, 1985;Cumming & Beange, 1987;Bergerud et al., 1990).Spring and summer habitat characteristics also reflect changes in diet from ter¬ restrial and arboreal lichens in winter to emergent herbs, grasses, and deciduous shrubs during the snow-free period (Bergerud, 1972;Darby & Pruitt, 1984).Characteristics of habitats used during the breeding season (fall rut) are less known, although it is thought that sparsely treed and open upland areas or open muskeg swamplands are the primary habitat types used (Fuller & Keith, 1981;Cumming, 1992).Two other factors are thought to be important at all times of year.Deadfall can be a barrier to movement, potentially making habitat unsuitable for caribou, other ungulates, or predators depending upon its abundance.Shrub density can affect the ability of caribou to see and escape from predators, and can also change the suitability of the habitat to favour other ungulate species (e.g.moose, Van Dyke et al., 1995;Courtois et al., 1998).
Natural and human habitat disturbance can cause a number of changes in microhabitat characteristics.Natural disturbance (fire) is thought to reduce the abundance of lichens and increase accumulations of snow and deadfall, which reduce residual forage availability (Klein, 1982;Schaeffer & Pruitt, 1991).Human disturbance (logging) is thought to alter the structure of floral communities to favour deciduous early successional forest species (Carleton & MacClennan, 1994).These habitat types are general¬ ly thought to be unsuitable for caribou, as arboreal and terrestrial lichen forage species abundance is lower and the density of competing ungulates and predators is higher (Rettie & Messier, 2000).The tendency for caribou to segregate themselves spatial¬ ly from other ungulates (particularly moose) in spring and summer is well known (Bergerud et al., 1984;Bergerud, 1985;Poole et al., 1999).This abil¬ ity is potentially compromised by habitat distur¬ bance (Smith et al., 2000).
It is believed that woodland caribou habitat selec¬ tion varies with spatial scale in response to variations in impact of limiting factors at different spatial scales (Bergerud et al., 1990;Rettie & Messier, 2000).Johnson (1980) describes four scales or orders of habitat selection extending from the selection of geographic ranges (first order) down to the selection of components within a daily area (fourth order).Models of habitat suitability for caribou exist at var¬ ious orders in this hierarchy.The majority are based on the general format for habitat suitability models (United States Fish and Wildlife Service, 1981), where a number of habitat variables are weighted in a mathematical relationship to give a defined habitat area a suitability score that ranges from 0 to 1.Such models are generally developed as decision support tools for resource management agencies.The source data are generally derived from forest resource inven¬ tory databases, as these are the standard vegetation databases used for forest management planning.Examples of such models have been developed in Manitoba (Palidwor & Schindler, 1995), British Columbia (Apps & Kinley, 1998), and Ontario (Antoniak & Cumming, 1998).
Most habitat suitability models apply at the level of the forest stand, and relate to the third order of habitat selection -the characteristics of habitats within an individual's home range (Johnson, 1980).A potentially useful model of habitat suitability that uses a visual score sheet assessment technique and relates to the third and fourth (the microhabitat characteristics within forest stands) order scales was previously developed in this study area (Storey & Storey, 1980).This method has been proposed in the past as a tool to evaluate forests in Manitoba for their suitability for woodland caribou, but has not been widely implemented (Hristienko, 1985).This visual score sheet method assesses seven habitat index com¬ ponents (stand type; tree size class and canopy clo¬ sure; cover; food plant diversity; food plant percent ground cover; deadfall; and area diversity) on a scale of 1 to 10.These components are then weighted to derive an aggregate habitat index ranging from 1 to 10 for that area.The visual assessment is based on how closely a site's characteristics for each of the habitat index components resembles ideal condi¬ tions, which in this case was defined as ideal winter habitat conditions (Storey & Storey, 1980).

Study area
This investigation was conducted in west central Manitoba and studied woodland caribou in the Kississing-Naosap range (approx.4500 km 2 ), locat¬ ed northeast of the towns of Flin Flon and The Pas (Fig. 1).Though precise range definitions have changed, a population estimate for the general area of the Kississing-Naosap range in 1992 varied from 150 to 300 individuals (Johnson, 1993).An aerial survey of the general area conducted in the winters of 1993 and 1994 observed between 90 and 164 ani¬ mals (Cross & Smith, 1995).Though it could be debated, the data available indicate that the popula¬ tion in this region remained fairly stable or declined slightly throughout the last fifteen years.
The Kississing-Naosap range intersects the boundary of the Churchill River upland and the mid-boreal lowland eco-regions of the boreal shield and boreal plains eco-zones.The boreal shield land¬ scape consists of rolling uplands and lowlands with many bedrock outcrops.This contrasts with the boreal plains landscape, which is topographically level to gently rolling, consisting of lacustrine or organic parent materials.Tree species include black spruce (Picea mariana), white spruce (Picea glauca), jack pine (Pinus banksiana), tamarack (Larix laricina), trembling aspen (Populus tremuloides), and white birch (Betula papyrifera).The climate is continental, characterized by short warm summers and cold snowy winters.Mean daily temperatures in the study area range from +17.7 °C in July to -21.4 °C in January.Mean annual rainfall and snowfall range from 323.3 mm and 170.2 cm in The Pas to 345.3 mm and 143.9 cm in Flin Flon.Snowfall accumula¬ tion is typically present from mid-November to early April, with maximum mean depths of 40 cm to 45 cm occurring in January and February.
Logging began in the area in the early 1970s (Anon., 1996).The Kississing-Naosap caribou range is within the provincial Forest Management License Area No. 2 and is currently actively managed by Tolko Industries, Ltd. under a Forest Management Plan approved until 2009 (Anon., 1996).Fires are a natural component of ecosystems in the area, although they are currently actively suppressed.The most significant recent burn was the Webb Lake fire in 1989, which burned approximately 125 000 hectares in the central portion of the study area and a significant portion (approx.25%) of the range of the Kississing-Naosap caribou (Anon., 1996).Caribou in the Kississing-Naosap range are also affected by highway and rail transportation corri¬ dors, road development associated with forestry operations, hydro transmission line construction, and various recreational activities.Because of the potential cumulative impacts of all of these distur¬ bances and the potential for an increase in develop¬ ment pressure in the near future, the Kississing-Naosap range is being considered for designation as a high-risk range under a provincial woodland cari¬ bou conservation policy.

Study objectives
To mitigate the potential impacts of logging on woodland caribou, natural resource development agencies have proposed a number of management techniques.In the Kississing-Naosap range in westcentral Manitoba, Canada, managers have estab¬ lished short and long term leave areas and undis¬ turbed travel corridors that allow access to known calving sites and other important habitats (Anon., 1999).Qualitative information on the presence of caribou sign is sometimes collected as part of preharvest forest investigation surveys and operational timber cruises for forest management (Anon., 1996).However, quantitative relationships between param¬ eters collected during these surveys and microhabitat suitability for woodland caribou have not been estab¬ lished in the area.
In order to establish such relationships, microhab¬ itat characteristics thought to be important to wood¬ land caribou were investigated in undisturbed areas used by this subspecies and in areas disturbed by fire or logging.Quantitative ecological and Forest Resource Inventory (FRI) parameters were related to habitat index scores calculated from the visual score sheet method previously developed in the study area (Storey & Storey, 1980).A visual arboreal lichen abundance index to quantify the relative biomass of arboreal lichen at the sites sampled was also devel¬ oped.The assumption that the visual score sheet index could distinguish habitats during different calendar seasons and disturbance regimes was tested.It was also hypothesized that a small number of eas¬ ily measured quantitative FRI and ecological vari¬ ables could be related to the visual index and predict habitat suitability for woodland caribou at the micro-site level.

Data collection
Sampling areas were located in undisturbed seasonal ranges known to be used by individual animals, as well as human disturbed (logged) sites, and natural¬ ly disturbed (burned) sites.Seasonal ranges (100% MCP) were defined from radio-telemetry locations collected from February 1996 to January 2000 as part of a concurrent study of range use and habitat selection.Six home ranges in each of the four calen¬ dar seasons (spring, summer, fall, and winter) were selected for use as sampling areas.These were chosen using a random number generator.The first six radio transmitter frequencies to be generated were select¬ ed.Fourteen disturbed sites of varying ages and dis¬ turbance types were also sampled.A random number generator was used to select ten by ten kilometre township squares as sampling areas for transects located at disturbed sites.
Line transects, a minimum of 500 m in length with plots located at 50 m intervals, were conducted within each sampling area.Transect start points within individual sampling areas were randomly selected using GIS, but limited to areas within 500 m of roads or lakes in order to provide access.Transect bearings were selected to intersect at least one location where the animal was known to have been present.When transportation resources were not available to access a randomly selected home range during the data collection period, an alternate site was selected.A similar procedure was used to select the starting point for transects located at logged and fire disturbed sites.Data collection occurred between July 29 th and August 4 th , 1999 and June 5 th and July 26 th , 2000.A total of 38 areas were sampled, consisting of 393 individual sample plots.
Data collected at each sample plot included scores for each habitat index component, Forest Resource Inventory (FRI) characteristics (species composition, age, height, canopy closure, diameter), and ecologi¬ cal characteristics (shrub/herb species composition), including stand attributes thought to be important to caribou (deadfall density, visual density, and arbo¬ real lichen abundance).
Data were collected hierarchically, using a number of nested plot sizes.Habitat index components were calculated by observing the characteristics of a 50 m x 50 m plot.The seven habitat index components (stand type; tree size class and canopy closure; cover; food plant diversity; food plant percent ground cover; deadfall; and area diversity) were each assessed on a scale of 1-10 ( Storey & Storey, 1980).The indi¬ vidual habitat index components were also weighted and averaged (Storey & Storey, 1980) to give an aggregated habitat index score ranging from 1-10 for each plot.A brief description of how the habitat index components were scored is provided in the appendix.
Forest Resource Inventory and ecological charac¬ teristics were observed in a 10 m x 10 m plot locat¬ ed at the centre of the 50 m x 50 m visual index plot.Tree species composition, density, and diameter dis¬ tribution at each plot was measured using the pointcentred quarter method (Cottam & Curtis, 1956).Height and age were measured from a single repre¬ sentative tree selected at each plot.Heights were measured using a Suunto clinometer.Cores were extracted using an increment borer and collected for aging with the aid of a dissecting microscope.Crown closure was estimated visually to the nearest ten per¬ cent.
Shrub species composition was recorded from the same 10 m x 10 m plot as the FRI characteristics, while herb species composition was recorded on a randomly selected 2 m x 2 m plot within the 10 m x 10 m plot.Shrub and herb species composition was determined by visually estimating the percent cover (Daubenmire, 1959).Cover percentages were record¬ ed to the nearest five percent.Shrubs were consid¬ ered to be any woody plants between 1 and 5m in height and less than 10cm in diameter.Herbs included all non-woody plant species and woody shrubs less than 1m in height.Deadfall amounts were determined by counting the number of fallen trees greater than 0.5 m from the ground along the 50 m line walked between each plot.Visibility was quantified by measuring the distance at which an individual walking along the transect between plots was no longer visible to an observer located at the previous plot.
Arboreal lichen abundance was assessed using a four level abundance index ranging from zero to three.To quantify this index, arboreal lichens were collected from sites throughout the study area.Nine sites representing each of the levels of abundance, with the exception of sites given an index value of 0, were sampled using a modification of the procedure first described by Van Daele & Johnson (1980).
Lichens were collected from heights of 0-1 m and 1¬ 2 m at each sample site, as maximum snow accumu¬ lation in this study area averaged less then 50 cm (Anon., 1998).For sampling purposes, each of the sample heights were divided into quarters.One quarter was then randomly selected and all arboreal lichens were stripped and collected.Samples were air dried for 48 h and oven-dried at 70 °C for one hour and measured to the nearest 0.01 g.This was then multiplied by four times the tree density at the plot to determine arboreal lichen abundance in kg ha-1 in each height stratum.

Data analysis
The hypothesis that the total amount of arboreal lichen in each abundance index class did not differ was tested using a one-way analysis of variance and a Tukey's HSD multiple range test.Lichen biomass values were log-transformed to correct for skewness and stabilize variance.One-way analysis of variance and Tukey's HSD multiple range tests were used to explore the hypothesis that each of the individual visual index components, as well as the final habitat index did not differ among seasons and disturbance regimes.Where variance homogeneity assumptions were not met, the ANOVA results were confirmed with non-parametric (Kruskal-Wallis) tests.
Relationships between the habitat index score and measured habitat variables were modeled using stepwise multiple linear regression.Two regression mod¬ els were developed.The first used standard FRI data as predictors (Model 1).The second used a combina¬ tion of FRI data and ecological data as predictors (Model 2).Plant species percent cover estimates were reduced to presence/ absence data to simplify the data requirements for applying the models.Only plants present in a minimum of 15% of plots were used to eliminate skewness.
Approximately half of the 393 plots sampled were randomly selected to develop the regression models.To examine the internal relia¬ bility of the regression mod¬ els, the regression predicted habitat values were correlated with measured habitat values at the sites not used to devel¬ op the regression model.The predicted habitat values were also regressed onto the meas¬ ured habitat values.A one standard error of the estimate prediction interval was then applied to the measured/pre¬ dicted relationship starting from the lowest possible habitat value that the regression models (1 or 2) could estimate in order to determine three statistical habitat quality categories.
All statistical analyses were done with SPSS 9.0 for Windows and results were considered statistically significant with P<0.05.

Results
Arboreal lichen abundance index Significant differences in the relative biomass of arboreal lichen were found among the four abun¬ dance index classes (Fig. 2).This was true for total lichen biomass (F(2,24)=9.93,P<0.01), lichen biomass from 0-1 m above ground (F(2,24)=8.36,P<0.01), and lichen biomass from 1-2 metres above ground (F(2,24)=7.31,P<0.01).Post-hoc testing (Tukey, P<0.05) showed that, in all cases, lichen bio¬ mass in each index class differed significantly from the others.Mean total lichen biomass was 15.7, 41.1, and 100.1 kg ha -1 for abundance classes one, two, and three respectively (Fig. 2).Since the abun¬ dance index classes as defined reflected true differ¬ ences in lichen biomass, this variable was considered in the rest of the analysis.

Habitat index component values
Each of the components in the habitat index was analyzed individually.Means and standard devia¬ tions for the habitat index components by season and disturbance type are listed in Table 1.In each case, there were significant differences (F(5,387)=9.81 to 58.82 , P<0.01) between seasonal use areas and dis¬ turbance regimes.Using post-hoc testing (Tukey P<0.05), it was found that used sites scored higher than disturbed sites in most cases, although there were some exceptions (Table 1).Burned sites were more likely than logged sites to not be different from used sites.An exception to this was deadfall, where burned sites scored much lower (indicating higher deadfall accumulation) than any of the used sites or logged sites, which did not differ significantly from each other.
For each habitat index component, sites used in the winter were found in the highest value subset, with the exception of food plant diversity, where it scored less than fall sites.Both fall and winter sites had significantly higher food diversity than spring and summer sites.Summer sites also had significant¬ ly lower food plant ground cover than all the other seasons, but did not differ significantly from burned sites for this parameter (Table 1).Besides these vari¬ ables, there was homogeneity in the habitat index component scores among seasons.All sites tended to be either closed or open stands of mature softwoods or mixed softwoods with more than 75% food plant ground cover and some (rather than no) deadfall, usually within 100 m of at least one or more other distinct habitat types.

Aggregate habitat index values
After analyzing each of the seven components indi¬ vidually, the individual scores were weighted to determine an aggregate habitat index score ranging from 0 to 10 for each plot.Significant differences were again noted between season and disturbance classes (F(5,390)=86.2,P<0.01) (Fig. 3).Post-hoc testing (Tukey, P<0.05) found that within seasonal use areas, winter sites scored significantly higher than summer sites, neither of which differed from spring or fall sites.All used sites as a group scored significantly higher than disturbed sites of either type.Within disturbed sites, burned areas scored significantly higher than logged areas (Fig. 3).At used sites, mean habitat index scores were 8.0, 7.5, 7.2, and 7.5 for winter, spring, summer, and fall areas respectively.At disturbed sites, mean habitat index scores were 5.7 and 5.0 for burned and logged areas, respectively.

Regression models to predict aggregated habitat index values
Habitat index scores were related to FRI variables only (Model 1, Table 2) and a both FRI and ecological variables (Model 2, Table 3) using multiple lin¬ ear regression to create a statistically significant rela¬ tionship (F(4,193)=67.79and 78.20 for Models 1 and 2, respectively, P<0.01).Presence of spruce trees, age, and mean tree diameter were positive pre¬ dictors, while the presence of trembling aspen trees was a negative predictor for Model 1 (Table 2).Presence of spruce trees and arboreal lichen index were positive predictors, while presence of trembling aspen shrubs and deadfall density were negative pre¬ dictors for Model 2 (Table 3).Both models explained approximately 60% of the variance in habitat index  1 ) 9 ) 3 ) 2 ) 1 ) 8  0 5 9 5 3 8 9 .9 .8 .9 .7 .8 .pa u Q < pa <<uC

< < pau
2 ) 2 ) 8 ) 0 ) 8 ) 1 ) .8 .3.7 .9.9.6 9 .9 .9 .9 .7 .6 .scores (r 2 =0.58 and 0.62, P<0.01, for Models 1 and 2, respectively).Measured habitat index scores were significantly correlated with the regression model predicted habi¬ tat index scores for plots not used to calculate the models (r 2 =0.58 and 0.51, P<0.05 for Models 1 and 2, respectively.A regression relationship between measured scores and model predicted scores was developed by regressing predicted habitat index val¬ ues onto measured habitat index values (Table 4).A prediction interval of +/-one standard error of the estimate was then applied to the relationship between actual and predicted values to calculate three statistical habitat quality categories for both models (Table 4).The bottom limit for the low qual¬ ity habitat category was set to the lowest possible predicted value from the Model 1 and Model 2 rela¬ tionships (4.6 and 5.0 for Models 1 and 2 respec¬ tively (Table 2 and 3)).Using the Model 1 and Model 2 equations to predict habitat index values, any sites scoring more than 8.7 for Model 1 or 8.4 for Model 2 represent high quality woodland caribou habitat.particularly winter and summer areas.The relation¬ ship was not as clear for spring and fall areas, which did not differ from either winter or summer use areas.Migrations between winter and summer ranges are a factor influencing where animals are located during the spring and fall, resulting in the sampling of areas that animals may have only been temporarily associated with.A closer look at the spa¬ tial distribution of study animals revealed that, in many cases, areas sampled as spring or fall use areas were often used at other times of the year, usually summer or winter.It is also possible that there are no differences in the vegetation characteristics of habi¬ tat used by this species in the spring and fall relative to other times of the year.Both of these possibilities suggest that summer and winter are the only two times of year that woodland caribou select habitats with specific characteristics at the micro-site scale in this area.At larger spatial scales, such a conclusion is supported by habitat and movement studies else¬ where in Canada that have shown a bimodal pattern of range use, with distinct summer and winter use areas (Edmonds, 1988).However, such a pattern is not constant, with some populations showing much overlap between seasonal ranges (Ouellet et al., 1996;Stuart-Smith et al., 1997).

i-lC\]i-i-rr^rTN
Considering all factors, the visual score sheet worked as it was intended.Numerically the trend was appropriate.Within use areas it generated the highest scores for winter sites and the lowest scores for summer sites, with intermediate scores for spring and fall (Fig. 3).Scores for disturbed areas were lower than for all used areas.Statistically, it was able to generate significantly different aggregate habitat index scores for the two seasons (summer and winter) that caribou seem to recognize in this area.It was

Habitat index prediction relation¬ ships
The results support the conclu¬ sion that the habitat index could successfully be related to both standard FRI data and FRI data in combination with ecological data, to distinguish three habitat quality classes.In each case, approximately 60% of the vari¬ ance in habitat value could be explained using the quantitative data set under consideration.Although the addition of ecolog¬ ical data created a different model than the FRI data alone, it did not explain any more variance and cre¬ ated similar habitat quality categories.This analysis provides evidence that the relative abundance of arboreal lichen and deadfall density are important ecological parameters influencing micro-habitat suitability for woodland caribou.However, it does not necessarily support the need to measure these parameters during pre-harvest forest assessments or during FRI data collection since they did not help to create a better assessment of habitat suitability in this case.
With the exception of arboreal lichen, no other individual forage plant could be considered a predic¬ tor of habitat suitability.This is not surprising con¬ sidering the variety and seasonal fluctuation in the forage plants that woodland caribou are known to consume (Bergerud, 1972).Generally, arboreal lichen is most important as a forage plant during winter (Bergerud, 1972).Mean scores for the arbore¬ al index in this study were 1.4 (s=0.9, n=60) for winter sites, 1.3 (s=0.9, n=191) for other seasons, and 0.2 (s=0.5, n=141) for disturbed sites (J.Metsaranta, unpubl.data).Thus, in this study area, arboreal lichens appear to be equally abundant at sites used during all seasons, and much less abundant at dis¬ turbed sites.
Many studies have postulated the importance of arboreal or terrestrial lichen in determining habitat suitability for woodland caribou.However, few stud¬ ies have considered deadfall density.The only habi¬ tat index component that did not differ between used sites and logged sites was the abundance of deadfall (Table 1).Measured deadfall density at burned sites was 17.3 stems m -1 (s=16.5, n=50), while measured deadfall density at all other sites was 5.2 stems m-1 (s=6.4 n=325) (J.Metsaranta, unpubl.data).Schaeffer & Pruitt (1991) also examined the influence of deadfall on woodland caribou, and found that deadfall densities were much higher in burned areas than in corresponding unburned areas that con¬ tinued to be used by caribou.They found continued use of unburned or crown burned residual stands and unburned lowlands during summer, with an appar¬ ent lack of use only in winter (Schaeffer & Pruitt, 1991).Similarly, in this study, a number of animals continued to use unburned residual portions and unburned islands in lakes within the extensive area burned by the Webb Lake fire in 1989.
Previous telemetry studies in the Reed Lake area southeast of this fire documented traditional caribou use of the burned area prior to 1989, but generally during summer only (Shoesmith & Storey, 1977;Benoit, 1996).Although data for determining if the current level of use is less than before the fire is lack¬ ing, the results would still suggest that a certain level of habitat suitability remains within the burned area since it has not been completely aban¬ doned.Previous studies have suggested that the reduction in caribou habitat suitability following fire occurs as a result of lichen combustion or increased snow accumulation reducing residual for¬ age availability (Klein, 1982;Schaeffer & Pruitt, 1991).The results of this study suggest that deadfall accumulation may also be an important factor, with the degree of impediment to travel imposed by the accumulation of deadfall possibly limiting the use of burned uplands at all times of year.
Lichens begin to re-establish in both logged and burned areas anywhere from 10 to 50 years after dis¬ turbance, with regeneration potentially occurring sooner after logging than after fire (Harris, 1996;Webb, 1998).Although disturbance by fire may have short-term detrimental effects on caribou, the long-term effects tend to be positive in terms of for¬ age productivity (Klein, 1982;Schaeffer & Pruitt, 1991).However, in the interim period, the post-fire accumulation of deadfall might concurrently dis¬ courage use of this habitat by all ungulates, by cre¬ ating areas that are virtually impassable to travel.
The literature has generally emphasized the posi¬ tive impacts of fire on moose habitat by its tendency to create more abundant forage for this species (e.g.Schwarz & Franzmann, 1989;Loranger et al., 1991).However, it is also acknowledged that this may not universally be the case, depending on such factors as fire intensity and pre-fire moose density (e.g.Peek, 1974;Gasaway et al., 1989).Some have suggested that moose appear to avoid blowdown areas, where presumably deadfall densities would be high (Cumming, 1980).If this was in fact true, and con-Rangifer, Special Issue No. 14, 2003 tinued to be the case during the 15 to 40 year postfire period when moose populations are thought to be highest (Schwarz & Franzmann, 1989), then changes in the composition of the faunal communi¬ ty (i.e.increases in moose populations, and as a con¬ sequence increases in wolf populations (e.g.Bergerud & Elliot, 1986;Seip, 1992)) that potentially have a negative effect on woodland caribou might not occur.
The proposed mechanism for preventing this faunal shift is the post-fire accumulation of deadfall that could occur under certain fire intensity and vegeta¬ tion combinations.Such circumstances may create habitat that, for a certain period of time, is equally unsuitable for both woodland caribou and moose.In the 12 years after the Webb Like fire in 1989, two surveys of moose populations in this area have been conducted (Cross, 1991;Cross, 2000).An initial sur¬ vey in 1990 predicted that moose density in this burned area should increase.However, a second sur¬ vey in 2000 concluded that these increases had not yet occurred, and that areas where deadfall accumu¬ lations were high were nearly devoid of moose activ¬ ity (Cross, 2000).

Conclusions
The analysis provides evidence for the efficacy of the visual score sheet method (Storey & Storey, 1980) for assessing microhabitat suitability for woodland cari¬ bou.It also shows that habitat index values calculat¬ ed using the visual score-sheet index could be used as the basis to relate FRI and ecological parameters to microhabitat suitability.Any of the three methods presented here (i.e. the score sheet method or Model 1 and Model 2) could be used to select forest stands for inclusion into short and long-term leave areas and travel corridors with more confidence that they contain habitat that will be suitable for use by wood¬ land caribou.This would potentially, at least at the microhabitat scale, mitigate some of the negative impacts that forestry and other development has had on this species.However, it is known that different selective pres¬ sures act at different spatial scales to limit woodland caribou populations (Bergerud et al., 1990;Rettie & Messier, 2000).It is also known that there are no simple solutions to natural resource management problems, and that unless an ecosystem approach is applied, actions can have unwanted and unpredicted consequences (Thompson & Welsh, 1993).Management actions to conserve woodland caribou populations need to consider all the factors that potentially act to limit this species (Cumming, 1992).Ecologically, it is known that these act at both small and large spatial scales.Applying the models presented here could conserve habitat characteristics needed by woodland caribou at small spatial scales.However, a further investigation quantifying the pattern and arrangement of suitable habitat types at a larger spatial scale would further increase the confidence in the ability to mitigate the impacts of development on this species.

Fig. 1 .
Fig. 1.Location of the Kississing-Naosap caribou range in west central Manitoba.The study area intersects the boundary of the Boreal Shield and Boreal Plains ecozones.

Fig. 3 .
Fig. 3. Comparison of the aggregate habitat index scores among sampling categories.Scores are presented as means (+/-1 s).Letters represent homogeneous subsets calculated using Tukey's HSD post-hoc testing.Subset A represents the highest ranking subset and D the lowest ranking subset.

Table 4 .
Regression relationships between actual and predicted values for Models 1 (FRI parameters only) and 2 (FRI and ecological parameters in combination), and the statistical habitat quality classes calculated from these relationships based on +/-standard error of the estimate prediction intervals.The bottom limits of the low quality category are based on the smallest possible value that either the Model 1 or 2 rela¬ tionships could predict.