Printer Friendly

Visibility of moose in a temperate rainforest.

ABSTRACT: Aerial surveys are the principal methods used to estimate populations of moose (Alces alces gigas) in Alaska. Accounting for missed animals during aerial surveys is problematical, especially in forested habitats; incorporation of a visibility correction factor to account for the proportion of animals missed is known to improve accuracy of population estimates. Our purpose was to study factors affecting visibility of radio-collared moose during aerial surveys in a temperate rainforest on the Yakutat Foreland, Alaska, USA. Wildlife managers in the area typically assume they observe only 50% of moose during surveys regardless of widely varying conditions. We used logistic regression to examine factors that influenced visibility including vegetation, light conditions, snow cover, and sex, age, and group size of moose. We then used logistic regression to develop a simpler model that only contained variables easily measured during aerial surveys: forest cover, snow cover, light, open versus vegetated habitat, and group size. We used that model to estimate a visibility correction factor. The mean correction factor was 1.304, ranging from 1.005-2.138, yielding a population estimate of 699 (90% CI = 671-724) moose from a survey count of 595 animals. Our correction factor was within the range reported for other populations of moose, and lower than the correction factor (2.0) currently used in this area. We conclude that application of site and time-specific visibility models is critical when estimating populations of large ungulates, especially in forested habitats.

Key words: aerial surveys, Alaska, Alces alces gigas, GIS, moose, population estimate, radio-telemetry, visibility.

**********

Population estimates of ungulates based on aerial surveys are subject to error associated with the inability to detect animals that are present (visibility bias; Timmerman 1993, Anderson and Lindzey 1996). Environmental factors such as rugged terrain or dense cover may obscure visibility of animals, and differences in habitat selection and morphology by sex and age groups may make some animals more difficult to observe, thereby biasing their visibility (Peek et al. 1974, Thompson and Veukelich 1981, Bowyer et al. 2002, Bowyer 2004). Grouping behavior, activity of individuals (i.e., lying or standing), weather, and ground conditions (e.g., snow cover) can measurably affect visibility of animals. Many of these problems are manifest in aerial surveys of moose (Alces alces gigas) in temperate rainforests on the Yakutat Foreland of southeast Alaska, USA where snow conditions that facilitate detecting moose can be intermittent, weather conditions for flying are frequently poor, and forest cover is dense and widespread. Ideally, population surveys should be conducted during the mating season when moose are more active and sexes aggregated (Miquelle et al. 1992, Oehlers et al. 2011). Because Yakutat does not generally receive sufficient snowfall to enhance visibility before sexes spatially segregate after mating and males cast antlers, identification of sex is difficult.

Sightability (also referred to as detectability or visibility) is the probability that an animal within the field of search for an observer will be seen by that observer (Caughley 1974). That probability can be expressed as a scalar, or correction factor for visibility bias, which is then multiplied by the number of moose observed to obtain a more accurate population estimate than an uncorrected count (Steinhorst and Samuel 1989). Correction factors for visibility bias (commonly referred to as Sightability Correction Factors or SCFs, and hereafter referred to as correction factors) that account for the proportion of animals undetected during aerial surveys are known to improve the accuracy of population estimates (Timmerman 1993), particularly for areas with extensive forest cover and variable weather conditions that occur on the Yakutat Foreland. Survey precision incorporates both the variance of total moose sighted and the variance of the correction factor (Timmerman 1993). Logistic regression is commonly used to develop correction factors for ungulates (McCorquodale 2001, Quayle et al. 2001, McIntosh et al. 2009); this method is designed for use with binomial dependent variables (observed or not), and can accommodate continuous and categorical independent variables (Hosmer and Lemeshow 2000).

We studied factors affecting visibility of moose on the Yakutat Foreland to improve population estimates from aerial surveys. We derived a series of models predicting correction factors using data from visibility trials from aerial surveys involving radio-collared moose. We examined the influence of temporal and weather-related variables such as month, time of day, cloud cover, light intensity, precipitation, and wind speed on visibility. We considered effects of environmental variables such as snow, forest, and vegetation cover on visibility of moose. In addition, we investigated the influence of sex, age, group size, sex and age composition of groups, activity, and intensity of site use on visibility. Logically, we expected that forest cover and lack of snow cover would reduce the probability of moose being observed, and that visibility would increase with increasing snow cover. We also hypothesized that visibility would decline with smaller group size or if moose were bedded. Further, we postulated that age or sex would affect visibility, because of morphological differences or if age groups and sexes used different habitats.

We derived a model containing all of the covariates that we determined were important predictors of visibility, and a second model that included only those variables for which information could be obtained from routine aerial surveys. The full model was needed to consider all variables, including life-history characteristics such as sex and age and their potential influence on visibility, and would be useful in areas where sex and age composition is known or could be determined during surveys. We considered the second model to be more appropriate for management purposes in our study area, because it did not require data that could only be obtained reliably from radio-collared animals, and is more appropriate for late-winter surveys when sex cannot be accurately determined. Finally, we applied the management model to a sample data set to estimate the density of moose within our study area. Ours is one of few studies to examine factors influencing visibility of ungulates in a northern temperate rainforest, and our results should be useful to biologists managing ungulates throughout the northern coastal forests of the Pacific Northwest.

STUDY AREA

We conducted research on the Yakutat Foreland of the Tongass National Forest, located along the southeast coastline of Alaska (Fig. 1). Our study area of approximately 1,280 [km.sup.2] encompassed most of the Foreland, and included ~80 km of coastline extending from Yakutat Bay to Dry Bay. Distance between the coast and mountain ranges varies from 8-24 km. There are several large rivers as well as numerous smaller streams distributed throughout the study area (Fig 1).

[FIGURE 1 OMITTED]

The Yakutat Foreland (Lat. 59[degrees] 20, N, Long. 139[degrees]0' W) falls within the humid temperate domain, characterized by year-round cloudy, cool, and wet conditions (Shephard 1995). The mean annual temperature was 4.1[degrees]C and the mean total precipitation was 381 cm (combined snow and rain) from 19712000 (NOAA 2005). The mean temperamre during this same time period was -3.4[degrees] C during January (the coldest month) and 12[degrees] C during July (the warmest month). Total snowfall during the study was 345 cm; mean daily snowfall was 3.0 cru and the mean snow depth was 20 cm.

Other than a few rolling bedrock hills, most of the Foreland is of low relief (average elevation 20 m; Shephard 1995), and is a mosaic of forests, wetlands, and shrublands (Shephard 1995). Forested areas are dominated by Sitka spruce (Piceasitchensis), and a small percentage of the upper canopy is composed of black cottonwood (Populus trichocarpa), western hemlock (Tsuga heterophylla), and mountain hemlock (T. mertensiana). Shephard (1995) documented 20 different forest communities on the Foreland, with canopy cover ranging from 1-80% and averaging 60% for the common forest communities, with stand heights ranging from 15-47 m. Nonforested areas include wetlands and shrublands composed primarily of graminoids, forbs, and shrubs including several species of tall and low willow (Salix spp.) ranging from 1-6 m in height, and Sitka alder (Alnus sinuata) up to 4 m. Nonforested areas dominate the coastal areas on the western half of the study area, with patches of spruce dispersed on the heaths and adjacent to some riparian zones; contiguous forested stands predominate the remainder.

The most recent aerial surveys (2002) conducted in the Forelands by Alaska Department of Fish and Game (ADFG) estimated a density of 0.5 moose/[km.sup.2] with a composition ratio of 19 males:100 females:14 young (N. L. Barten, ADFG, pers. comm.). Total count surveys by parallel transects set approximately 0.4-0.5 km apart are conducted in the nonforested portions of the Foreland by ADFG in late autumn as soon as snow covers most of the ground, but often those conditions do not occur until well into winter. The ADFG assumes that 50% of moose along transects are detected; consequently, the observed number of moose is doubled to estimate population size. That correction factor, however, has never been empirically evaluated or assessed. In addition, ADFG does not survey forested portions of the Forelands because of low (unknown) visibility, which constitutes about one-half of the study area.

Other large mammals that occur on the Forelands include brown bear (Ursus arctos), black bear (U. americanus), and gray wolves (Canis lupus). Sitka black-tailed deer (Odocoileus hemionus sitkensis) occupy some of the islands offshore but are uncommon on the mainland. In addition, moose are an important part of the subsistence economy (Ballew et al. 2006, Schmidt et al. 2007).

METHODS

Capture and handling

Twenty-two female and 16 male moose were darted from a helicopter by ADFG personnel with Palmer CAP-CHUR equipment with the immobilizing drugs carfentanil and xylazine (Roffe et al. 2001) during March and November 2002, and March and December 2003. Dosages ranged from 3.0-5.0 mg of carfentanil and 100-130 mg of xylazine depending on time of year, sex, and animal condition. All capture and handling methods followed guidelines established by the American Society of Mammalogists Animal Care and Use Committee (1998) for research on wild mammals. Our protocols were approved by independent Institutional Animal Care and Use committees at the University of Alaska Fairbanks (protocol # 04-26) and the ADFG (protocol # 03-0001).

We fitted moose with GPS radio-collars (Model 4000, Lotek Wireless, Ontario, Canada) that recorded locations 4 times daily, or standard VHF radio-collars (Model MP2-MPP4, AVM, Colfax, California and Model 600NH, Telonics, Mesa, Arizona). We programmed both types of collars to release remotely relative to time of deployment (typically 1.5 yr). A lower incisor was removed from each moose to determine age from cementum annuli (Gasaway et al. 1978). Naltrexone (350-1300 mg) and tolazoline (400-800 mg) were subsequently administered and moose were monitored until they recovered from the immediate effects of immobilization. We also monitored each moose by aerial survey for 1 month post-capture to assess capture-related mortality. Three females died or their collars malfunctioned within 1 month of capture, and were not included in the visibility trials.

Visibility trials

We flew surveys to locate collared moose between 24 November 2003 and 18 March 2004 using a Cessna[R] 185 fixed-wing aircraft. Timing of sampling and type of aircraft were the same used by ADFG when conducting moose surveys. We defined a visibility trial as the effort by the survey crew to count all moose within a 5 [km.sup.2] sampling quadrat (square survey block) that included a radio-collared moose on a particular day. The aerial-survey crew was composed of the pilot, the primary observer in the front seat, and a secondary observer in the back seat behind the pilot. We attempted to control as many factors as possible, such as using the same aircraft, pilot, and primary observers for all trials. One pilot and 2 primary observers with > 150 h of moose survey experience were used in the trials, with 8 secondary observers ranging in initial experience level of 6-8 (40-150 h of moose survey experience) on a Lickart scale of 1-10.

Our trial procedure was similar to that of Quayle et al. (2001). Trials to locate individual radio-collared moose were separated by [greater than or equal to] 3 days to reduce autocorrelation among locations. The extremely large home ranges of moose on the Foreland (mean seasonal home ranges varied from 24.3-86.3 [km.sup.2]; Oehlers et al. 2011) made this intervala reasonable choice for attempting to achieve independence among locations. Frequencies for the subset of moose to be sampled during a flight were programmed into a receiver (model R4000, ATS, Isanti, Minnesota) and scanned while flying at an altitude of 245-300 m above ground level. Once a signal was received, the primary observer obtained the general position of the moose without identifying an exact location. We used a laptop computer equipped with Baker Geolink Sketchmapping software (Michael Baker Corporation, Moon Township, PA) to record our location and flight path so that the telemetry operator (primary observer) could identify the approximate location of the collared moose on the map without viewing the ground, thereby minimizing observer bias. The pilot and secondary observer also avoided scanning the ground in the immediate survey area to prevent detection of the target animal before beginning the survey. The survey crew noted if the collared moose was accidentally spotted by either observer while obtaining the general location; those observations were eliminated from analyses.

The primary observer then delineated a 5 [km.sup.2] (2.23 km x 2.23 km) quadrat (Quayle et al. 2001) centered around the general location of the identified moose on the laptop computer using a 0.4 km grid overlay on the screen. Because the location of the moose was inexact, the actual location of the moose was not centered within the quadrat. Consequently, observer bias was minimized because none of the observers knew where in the quadrat to expect to find the radio-collared moose. The pilot then flew over the quadrat along transects spaced 0.4 km apart, which were delineated by the grid overlaid on the screen. The laptop screen displayed our flight path, allowing the pilot to navigate and follow the specified transect lines. The pilot flew the aircraft at an altitude of approximately 185 m and speed of about 130 km/h, resulting in a search intensity of approximately 1.0 min/ [km.sup.2]. We circled the location of each moose sighted in the quadrat to identify and record information on all of the variables included in the Appendix, and recorded the location of the moose using the Sketchmapper software. If the targeted moose was not sighted during the survey, we located that animal via telemetry and recorded the same information.

Forest cover was measured at 2 scales and recorded as "0" if the predominant vegetation within both a 10 m and 250 m radius of the radio-collared moose was nonforested, and "1" if this same area was predominantly forested (including a range of canopy covers). Vegetation cover was defined as "0" if the predominant vegetation was open habitat such as muskeg, meadow, sand, or gravel bar, or "1" if there was vegetation such as tall shrubs or forest that could obscure visibility of the moose. Percent vegetation was recorded as a categorical variable (1-3) representing percentage of vegetative cover (shrubs or trees) within a 10 m and 250 m radius of the observed moose that could obscure visibility of that moose.

We defined a "group" as 1 or more moose within 50 m of each other (Siegfried 1979, Molvar and Bowyer 1994, Bowyer et al. 2001) to encompass the complete range of sociality for this species (Monteith et al. 2007). We categorized age of non-collared animals as young (< 1 year) or adult ([greater than or equal to] 1 years old) through visual observation. We expected a high pregnancy rate of yearlings (Boer 1992), because preliminary data indicated a predator-limited population (Bowyer et al. 2005, Oehlers et al. 2011). Consequently, we considered yearling females as adults (Monteith et al. 2007). Moreover, distinguishing between yearlings and adults during aerial surveys in winter was difficult, and further distinguishing of ages beyond yearling or adult was not possible during aerial surveys.

We used Arc View 3.2 geographic information software (ESRI, Redlands, CA, USA) to plot GPS locations for moose and determine elevation and distance to the coast for each moose. Elevation was extracted from a raster data layer provided by the U.S. Forest Service (USFS), which was based on USGS digital elevation model with 20-m resolution. Distance from shore was calculated with the USFS shoreline polygon layer for the study area.

Statistical analyses

Detection of a radio-collared moose during visibility trials was coded 1 if detected and 0 if not observed. We used SAS 9.1 (SAS Institute, Cary, NC) for all statistical tests, and adopted an [alpha] = 0.05. We used multivariate logistic regression to model visibility. Our suite of potential predictors of detection included parameters such as sex, age, group size, forest cover, snow cover, light conditions, aircraft speed, and experience of observers (Appendix). Group size was squared because the untransformed covariate was not linear in the logit. We included the identification of individual moose as a coded variable to control for making repeated measures of individual moose. We reduced potential multicollinearity among independent variables by testing for strong correlations between pairs of covariates ([absolute value of r] [greater than or equal to] 0.7) and preventing their simultaneous inclusion in logistic regression models. During initial model screening, we also examined variance inflation factors (VIF) and tolerance (Tol) of independent continuous and discrete variables to identify intercorrelated variables. Values of VIF <10 and Tol >0.40 were considered acceptable (Neter et al. 1996, Allison 2001). We ultimately considered 16 variables from the initial set of 27 candidate predictor variables. We then screened these remaining covariates using forward step-wise logistic regression (PROC LOGISTIC; Agresti 1990) with an alpha to enter of 0.15 (Hosmer and Lemeshow 2000, p. 118) and alpha to remove of 0.3, and backward logistic regression with alpha to remove of 0.3, to define a broad initial set of candidate models. We restricted the number of covariates within any candidate model to [less than or equal to] 8, because our sample size of visibility trials was 88; our sample size precluded a global model. Our sample size also precluded an all possible regressions approach. We used Hosmer and Lemeshow tests for goodness-of-fit (Hosmer and Lemeshow 2000) to determine the appropriateness of the logistic models.

Once we had established a large set of candidate models, we used Akaike's Information Criterion (AICc) (Burnham and Anderson 2002) to select model variables. Age and sex were included in most of the top candidate models. Classifying moose into discrete age classes (i.e., beyond yearling or adult) is not possible from aerial surveys, and correct classification of sex is difficult once males have cast their antlers, so we repeated this same process omitting age and sex to allow development of models that did not rely on data from captured moose. Accordingly, we developed overall explanatory models that included life-history characteristics, as well as management models which included variables that could be measured easily during aerial surveys alone. We used model-averaging procedures to derive composite explanatory and management models (Burnham and Anderson 2002, Giudice et al. 2012). We only considered candidate models with AICc A values <4 for inclusion in composite models. We calculated relative effects (risk ratios) for covariates included in our composite models (Farmer et al. 2006). Relative effects estimate the change in relative probability of detection for an incremental change in magnitude of a predictor variable (Riggs and Pollock 1992). We evaluated relative effects to determine the comparative importance of independent variables in affecting the probability of detection. In general, relative effects >2.0 or <0.5 indicated large effects of covariates on detectability (Riggs and Pollock 1992).

For demonstrative purposes, we applied our composite management model to existing surveys of the moose population that were conducted by ADFG on the Yakutat Foreland from 30 November-4 December 2005 using their survey methodology previously described in Study Area. Model variables were assessed for each individual or group ofmoose observed during these surveys, and then the corresponding correction factor was calculated for each observation and multiplied by the number of animals in that observation. These corrected estimates were then totaled to derive a mean population estimate and the range of population estimates using the upper and lower correction factor based on the 90% CI (Becker and Reed 1990, Anderson and Lindzey 1996, White 2005). These data included 262 observations of single moose or groups and 595 total moose observed.

RESULTS

The median age for both females (n = 22, range = 3-13 yr) and males (n = 16, range = 1-10yr) was 6 years. We conducted 88 trials involving 55 radio-collared females and 33 males; each was surveyed 1-4 times (x = 2.3, SD = 0.70). Snow conditions were generally adequate for aerial surveys from November-January and during the last 20 surveys conducted in Match, but comparatively poor during February. We observed 254 groups of moose.

Radio-collared animals were sighted in 71% of the surveys; males were observed in 76% and females in 66% of the trials. Radio-collared animals were detected in 82% of trials in nonforested areas, and in 27% of trials in forested cover. Animals 1-3, 4-6, 7-10, and 11-13 years old were detected in 89, 55, 75, and 100% of trials, respectively. Mean ([+ or -] SE) group size of collared animals was 3.7 [+ or -] 0.4. Radio-collared animals were observed in open (31%), shrub (52%), and forested (17%) habitat during the trials. The location of females and males in nonforested and forested habitat was similar; 82 and 85% and 18 and 15%, respectively.

Logistic regressions

Forest cover, vegetation cover, and percent coverwere each correlated ([absolute value of r] [greater than or equal to] 0.7) between the 2 scales of measurement (10 m and 250 m). We considered the 10-m scale more easily estimated and likely to be consistent between observers; consequently, we chose to include the 10-m scale for each of these variables for consideration in our models. Following tests for collinearity, variance inflation factors, and tolerance, candidate models for overall visibility included the parameters age, group size, forest cover, light, snow cover, experience secondary, and wind speed start (Table 1). Age, group size (2), forest cover, and snow cover were included in each of the top 3 candidate models. Visibility increased by 38% for each additional year of the moose aged, and by 75% for each additional (increasing) experience level of the secondary observer (Table 2). Overcast skies (versus sun) increased visibility by 175%. Visibility increased with group size (2) and speed of the plane (flight speed ranged from 129-145 km/h), but effects were small. Visibility declined under forested cover (94%), snow cover of 0-33% (76%) or 34-66% (82%), and for females (23%).

Candidate models derived for management purposes (omitting sex and age) included group size (2), forest cover, snow cover, light conditions, and vegetation cover (Table 3). Similar to the overall model, detectability increased with group size, nonforested and open habitat, overcast skies, and higher snow cover in the composite management model (Table 4). Application of the composite management model to our sample data yielded a range of correction factors from 1.005-2.138 for each observation. The mean correction factor was 1.304, and mean upper and lower (90% CI) correction factors were 1.215 and 1.390, yielding a population estimate of 671-724 animals (x = 699 moose) from an uncorrected count of 595 animals.

DISCUSSION

Both our overall and management models included group size, forest cover, and snow cover as covariates of visibility. Lack of snow cover strongly reduced visibility of moose and confirmed our hypothesis that visibility would be higher as snow cover increased. Nonetheless, that relationship was not linear because visibility was similar between snow cover of 0-33% and 34-66% (57% and 54%, respectively). We believe that snow cover of 34-66% did not improve visibility because snow was still sufficiently patchy to obscure many moose against a dark background. We hypothesize that no snow cover actually may be preferable to patchy snow because patchy snow conditions may fatigue observers more quickly than uniform coverage.

Forest cover has been included in visibility models for both North American elk (Cervus elaphus; Samuel et al. 1987, Bleich et al. 2001) and moose (Peterson and Page 1993, Anderson and Lindzey 1996, Drummer and Aho 1998, Quayle et al. 2001). In our study area, coniferous tree species predominate in the forested areas, obstructing visibility of moose year-round, whereas vegetation in non-forested areas included alders and willows that do not retain leaves during winter and have less effect on visibility.

Group size was less influential on visibility than either forest or snow cover. Group size affects visibility of elk (Samuel et al. 1987, Bleich et al. 2001, McCorquodale 2001), feral horses (Equus caballus; Ransom 2012), and mule deer (Odoeoileus hemionus; Ackerman 1988); logically, larger groups are generally more visible. Moose tend to aggregate in open areas in Alaska during rut (Miquelle et al. 1992, Molvar and Bowyer 1994); therefore, if snow conditions are adequate, visibility would be highest during the peak of rut. Visibility did not differ when moose were standing or bedded. Light condition also was an important predictor of visibility as moose were more visible in overcast conditions when glare and shadows were minimized. Fox (1977) noted similar issues with glare from snowfields during mountain goat (Oreamnos americana) surveys conducted in clear weather in southeast Alaska.

Visibility increased with increasing age of moose, and was higher for males than for females, although the relative effect of sex was small. Greater visibility of males could distort male:female ratios and result in the underestimation of the female population, unless a correction for differential visibility is incorporated. Although several other studies of ungulate visibility reported that sex or group composition was accounted for in multivariate models because of correlation with other covariates such as group size or vegetation (Anderson and Lindzey 1996, Bleich et al. 2001, McCorquodale 2001), sex in our model was not correlated with any other variable. The effect of sex on visibility probably occurred because of physical differences between the sexes; larger body size, darker color, and presence of antlers in early winter likely explain the higher visibility of males. Solberg et al. (2010) also reported that male moose were observed by hunters with a 1.26 higher probability than females during the hunting season, and suggested that this difference was reflective of fundamental differences in antipredator behavior, including risk taking (such as use of open habitat), activity level, and space use. Although age was not significantly correlated with other variables, all observations of the oldest animals were in large groups in non-forested habitat; therefore, other covariates besides age were likely more influential on visibility. We did not detect an influence of age and sex composition of groups on visibility.

We attempted to standardize flight speed during surveys, and weather conditions resulted in a minimal range of speeds (130-145 km/h). Remarkably, visibility of moose increased with speed of the plane. Nonetheless, flight speed was in only 1 of the top 4 candidate overall explanatory models, its relative effect was small, and the 90% CI included 0; within the range of speeds we flew, this variable was likely of minimal importance. Experience level (1-10) of the second observer increased visibility by 75% in the explanatory model; however, the effect was highly variable, and was not included in the management model. Although experienced observers have developed a search image, and therefore may be more likely to observe moose, observer experience is difficult to quantify, and experience level changes over the course of visibility trials. Previous studies have documented differences in visibility related to observer experience (LeResche and Rausch 1974, Caughley et al. 1976); however, recent studies have noted little effect on visibility when observers were experienced (Ackerman 1998) or when observer experience correlated with other variables in the model (Samuel et al. 1987, Anderson and Lindzey 1996). All second observers in our study were experienced in moose surveys (i.e., 40-150 h of moose survey experience); consequently, our model will be most effectively applied when using experienced observers, a conclusion also reached by Quayle et al. (2001).

Our overall visibility of moose was 70.5% and similar to that in Quebec (Crete et al. 1986; 73%), Alberta (Rolley and Keith 1980; 64%), and Isle Royale, Michigan (Peterson and Page 1993; 78%), and higher than in Minnesota (Giudice et al. 2012; 38-56%), Michigan (Drummer and Aho 1998; 39%), Wyoming (Anderson and Lindzey 1996; 59%), and Alaska (LeResche and Rausch 1974; 43-68%). Correction factors for moose range from 1.03-3.2 (Oosenberg and Ferguson 1992, Timmerman and Buss 1998) and are generally higher in areas of denser cover and higher moose density (Gasaway et al. 1986, Peterson and Page 1993). Comparisons of visibility rates may be tenuous, however, because of differences in aircraft type (Crete et al. 1986), number of observers, search intensity, and habitat (Anderson and Lindzey 1996). Our results are within the range of correction factors reported for moose, but emphasize the variability in visibility and the need to develop correction factors specific to a particular area and time frame.

The use of a dynamic correction factor, such as that developed with a visibility model, is superior to the use of a static correction factor. Our modeled correction factor is offered as an alternative to the use ofboth a calculated SCF ([SCF.sub.c]) and an observed SCF ([SCF.sub.o]) as described by Gasaway et al. (1986). Observed SCFs must be calculated for each survey (preferable daily), and are cost prohibitive in areas dominated by dense coniferous forests and areas of low moose density (Gasaway et al. 1986), both of which occur in our study area. Our results confirm that visibility of moose from aircraft varies with environmental factors and group size. Therefore, application of the visibility model, combined with an appropriate sampling strategy, and with sophisticated analytical methods such as machine learning ('non-linear statistics'; Breiman 2001), may improve the accuracy and precision of population estimates over the use of a static correction factor.

Our method could be extended to other areas of similar environmental conditions such as the remainder of coastal Alaska and British Columbia (and could be tested for applicability to interior Alaska) if protocols associated with the chosen model are followed (McCorquodale 2001). Because visibility may differ among types of aircraft used (Crete et al 1986), surveys should be conducted using a Cessna[R] 185 or similar fixed-wing aircraft at approximately 185 m above ground elevation, as used in model development (Samuel et al. 1987, Anderson and Lindzey 1996). Additionally, observers should be experienced and their observation skills constantly calibrated in aerial surveys of moose. Conducting surveys when moose are likely to be most visible (i.e., with nearly continuous snow cover and overcast light conditions) will provide the most precise population estimates. Improved population estimates will allow for more knowledge-based and effective management decisions by state and federal managers.

ACKNOWLEDGEMENTS

This research was funded primarily by the U.S. Forest Service, with additional funding from the Department of Interior Bureau of Indian Affairs, and in-kind support from the ADFG. The Institute of Arctic Biology, Department of Biology and Wildlife, and the Alaska Fish and Wildlife Cooperative Research Unit of the University of Alaska Fairbanks were all instrumental in the educational and funding portion ofthe project. W. Eastland contributed technical support. Many thanks to T. O'Connor, C. Grove, and E. Campbell of the U.S. Forest Service for project support. Special thanks to ADFG biologists J. Crouse, S. Jenkins, N. Barten, and K. White for their support in capture and handling of moose. Pilots D. Russel, L. Hartley, B. Bingham, and J. Liston contributed to captures of moose, and our aerial survey efforts. Thanks also to helicopter pilots of Temsco Helicopter and U.S. Forest Service helicopter managers A. Stearns, J. Schlee, and D. Andreason. We acknowledge U.S. Forest Service personnel N. Catterson, K. Schaberg, B. Lucey, D. Gillikin, S. Mehalick, M. Moran, and C. Wiseman for field and logistical support.

Appendix
Candidate predictor variables considered during initial modeling for
visibility of moose on the Yakutat Foreland, Alaska, 2003-2004.

Variable              Type          Description

Month                 Discrete      Month of visibility trial

Age                   Discrete      Age of collared moose

Sex                   Discrete      Sex of collared moose

Group                 Indicator     0 = single moose, 1 = > m moose

Group size            Discrete      Total number of moose seen within
                                    50 m of collared moose

Composition           Indicator     0 = single-sex group; 1 = both
                                    sexes in group

Males                 Discrete      Number of adult males in group

Females               Discrete      Number of adult females in group

Calves                Discrete      Number of calves in group

Unknown Sex           Discrete      Number of unknown sex adults in
                                    group

Forest Cover          Indicator     0 = nonforested, I = forested,
10 m                                within 10 m of moose

Forest Cover          Indicator     0 = nonforested, I = forested,
250 m                               within 250 m of moose

Vegetation Cover      Indicator     0 = open habitat such as muskeg, 1
10 m                                = shrub or forested habitat within
                                    10 m of moose

Vegetation Cover      Indicator     0 = open habitat such as muskeg, 1
250 m                               = shrub or forested habitat within
                                    250 m of moose

Percent Vegetation    Indicator     1 = 0-33%, 2 = 34-66%, 3 = 67 -
10 m                                100% vegetative cover within 10 m
                                    of moose

Percent Vegetation    Indicator     1 = 0-33%, 2 = 34-66%,3 = 67-100%
250 m                               vegetative cover within 250 m of
                                    moose

Elevation             Continuous    Elevation above sea level in
                                    meters

Distance from         Continuous    Straight-line distance from
coast                               coastline to center of moose group
                                    in meters

Activity              Indicator     0 = bedded, 1 = active (any moose
                                    in group)

Site use              Indicator     0 = no beds, few tracks, 1 = beds
                                    and multiple tracks

Cloud cover           Indicator     0 = clear, I = partly cloudy, 2 =
                                    overcast

Precipitation         Indicator     0 = none, I = mist, 2 = light
                                    rain, 3 = hard rain, 4 = snow

Snow cover            Indicator     1 = 0-33%, 2 = 34-66%,3 = 67-100%

Wind speed start      Continuous    Wind speed (km/h) at beginning of
                                    survey

Wind speed end        Continuous    Wind speed (km/h) at end of survey

Flight speed          Continuous    Average flight speed (km/h) during
                                    survey (excludes circling)

Temperature           Continuous    Average temperature (Celsius)
                                    during survey

Start Time            Discrete      Survey start time; military time
                                    rounded to hour

Light                 Indicator     0 = sunny, 2 = flat light/even
                                    shadows

Experience            Continuous    Previous experience level of
primary                             primary observer, scale of 1-10

Number flights        Discrete      Number of previous visibility
primary                             trials by primary observer

Experience            Continuous    Previous experience level of
secondary                           secondary observer, scale of 1-10

Number flights        Discrete      Number of previous visibility
secondary                           trials by secondary observer

Variable              Method/Time of
                      Collection
Month                 Aerial Survey

Age                   Capture

Sex                   Capture

Group                 Aerial Survey

Group size            Aerial Survey

Composition           Aerial Survey

Males                 Aerial Survey

Females               Aerial Survey

Calves                Aerial Survey

Unknown Sex           Aerial Survey

Forest Cover          Aerial Survey
10 m

Forest Cover          Aerial Survey
250 m

Vegetation Cover      Aerial Survey
10 m

Vegetation Cover      Aerial Survey
250 m

Percent Vegetation    Aerial Survey
10 m

Percent Vegetation    Aerial Survey
250 m

Elevation             GIS

Distance from         GIS
coast

Activity              Aerial Survey

Site use              Aerial Survey

Cloud cover           Aerial Survey

Precipitation         Aerial Survey

Snow cover            Aerial Survey

Wind speed start      Aerial Survey

Wind speed end        Aerial Survey

Flight speed          Aerial Survey
                      (plane
                      instrumentation)

Temperature           Aerial Survey

Start Time            Aerial Survey

Light                 Aerial Survey

Experience            Collected from
primary               each surveyor prior
                      to visibility trials

Number flights        Collected from
primary               each surveyor prior
                      to visibility trials

Experience            Collected from
secondary             each surveyor prior
                      to visibility trials

Number flights        Tabulated
secondary             throughout
                      visibility trials


REFERENCES

ACKERMAN, B. R. 1988. Visibilitybias ofmule deer aerial census procedures in southeast Idaho. Ph. D. Dissertation, University of Idaho, Moscow, USA.

AGRESTI, A. 1990. Categorical Data Analysis. John Wiley & Sons, New York, New York, USA.

ALLISON, P.D. 2001. Logistic Regression Using the SAS system: Theory and Application. SAS Institute Inc., Cary, North Carolina, USA.

ANDERSON, C. R., and F. G. LINDZEY. 1996. Moose sightability model developed from helicopter surveys. Wildlife Society Bulletin 24: 247-259.

ANIMAL CAPE AND USE COMMITTEE. 1998. Guidelines for the capture, handling, and care of mammals as approved by the American Society of Mammalogists. Journal of Mammalogy 74:1416-1431.

BALLEW, C., A. R. TZILKOWSKI, K. HAMRICK, and E. D. NOBMANN. 2006. The contribution of subsistence foods to the total diet of Alaska natives in 13 rural communities. Ecology of Food and Nutrition 45: 1-26.

BECKER, E. F., and D. J. REED. 1990. A modification of a moose population estimator. Alces 26: 73-79.

BLEICH, V. C., C. S. Y CHUN, R. W. ANTHES, T. E. EVANS, and J. K. FISCHER. 2001. Visibility bias and development of a sightability model for tule elk. Alces 37:315-327.

BOER, A. H. 1992. Fecundity of North American moose (Alces alces): a review. Alces Supplement 1: 1-10.

BOWYER, R. T. 2004. Sexual segregation in mminants: definitions, hypotheses, and implications for conservation and management. Journal of Mammalogy 85: 1039-1052.

--, D. R. MCCULLOUGH, and G. E. BELOVSKY. 2001. Causes and consequences of sociality in mule deer. Alces 37: 371-402.

--, D. K. PERSON, and B. M. PIERCE. 2005. Detecting top-down versus bottom-up regulation of ungulates by large carnivores: implications for conservation of biodiversity. Pages 342-261 in J. C. Ray, K. H. Redford, R. S. Steneck, and J. Berger, editors. Large Camivores and Biodiversity Conservation. Island Press, Covelo, Califomia, USA.

--, K. M. STEWART, S. A. WOLFE, G. M. BLUNDELL, K. L. LEHMKUHL, P. J. JOY, T. J. MCDONOUGH, and J. G. KIE. 2002. Assessing sexual segregation in deer. Journal of Wildlife Management 66: 536-544.

BREIMAN, L. 2001. Statistical modeling: the two cultures. Statistical Science 16: 199-231.

BURNHAM, K. P., and D. R. ANDERSON 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Second edition. Springer-Verlag Inc., New York, New York, USA.

CAUGHLEY, G. 1974. Bias in aerial survey. Journal of Wildlife Management 38: 921-933.

--, R. SINCLAIR, and D. SCOTT-KEMMIS. 1976. Experiments in aerial survey. Journal of Wildlife Management 40: 290-300.

CRETE, M., L. RIVEST, H. JOLICOEUR, J. BRASSARD, and F. MESSIER. 1986. Predicting and correcting helicopter counts ofmoose with observations made from fixed-wing aircraft in southern Quebec. Journal of Applied Ecology 23:751-761.

DRUMMER, T. D., and R. W. Ano. 1998. A sightability model for moose in Upper Michigan. Alces 34: 15-19.

FARMER, C. F., D. K. PERSON, and R. T. BOWYER. 2006. Risk factors and survivorship of black-tailed deer in a managed forest landscape. Journal of Wildlife Management 70: 1403-1415.

FOX, J. L. 1977. Summer mountain goat activity and habitat preference in coastal Alaska as a basis for the assessment of survey technique. Pages 190-199 in W. Samuel and W. G. Macgregor, editors. Proceedings First Intemational Mountain Goat Symposium, 19 February, 1977, Kalispell, Montana, USA.

GASAWAY, W. C., S. D. DUBO1S, D. J. REED, and S. J. HARBO. 1986. Estimating moose population parameters from aerial surveys. Biological Paper of the University of Alaska Fairbanks 22: 1-108.

--, D. B. HARKNESS, and R. A. RAUSCH. 1978. Accuracy of moose age determinants from incisor cementum layers. Journal of Wildlife Management 42: 558-563.

GUIDICE, J. H., J. R. FIEBERG, and M. S. LENARZ. 2012. Spending degrees of freedom in a poor economy: a case study of building a sightability model for moose in northeastern Minnesota. Journal of Wildlife Management 76: 75-87.

HOSMER, D. W., and S. LEMESHOW. 2000. Applied Logistic Regression. Second edition. John Wiley and Sons, New York, New York, USA.

LERESCHE, R. E., and R. A. RAUSCH. 1974. Accuracy and precision of aerial moose censusing. Journal of Wildlife Management 38: 175-182.

MCCORQUODALE, S. M. 2001. Sex-specific bias in helicopter surveys of elk: sightability and dispersion effects. Journal of Wildlife Management 65: 216-225.

MCINTOSH, T. E., R. C. ROSATTE, J. HAMR, and D. L. MURRAY. 2009. Development of a sightability model for low-density elk populations in Ontario, Canada. Journal of Wildlife Management 73: 580-585.

MIQUELLE, D. G., J. M. PEEK, andV. VAN BALLENBERGHE. 1992. Sexual segregation in Alaskan moose. Wildlife Monographs 122: 1-57.

MOLVAR, E. M., and R. T. BOWYER. 1994. Costs and benefits of group living in a recently social ungulate: the Alaskan moose. Journal of Mammalogy 75:621-630.

MONTEITH, K. L., C. L. SEXTON, J. A. JENKS, and R. T. BOWYER. 2007. Evaluation of techniques for categorizing group membership of white-tailed deer. Journal of Wildlife Management 71: 1712-1716.

NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION (NOAA). 2005. Local climatological data. <http://cdo.ncdc.noaa.gov> (accessed May 2007).

NETER, J. W., M. H. KUTNER, C. J. NACHTSHEIM, and W. WASSERMAN. 1996. Applied Linear Statistical Models. Foul-th edition. The McGraw-Hill Companies, Inc., NewYork, New York, USA.

OEHLERS, S. A., R. T. BOWYER, F. HUETTMANN, D. K. PERSON, and W. B. KESSLER. 2011. Sex and scale: implications for habitat selection by Alaska moose. Wildlife Biology 17: 67-84.

OOSENBERG, S. M., and S. H. FERGUSON. 1992. Moose mark-recapture survey in Newfoundland. Alces 28: 21-29.

PEEK, J. M., R. E. LERESCHE, and D. R. STEVENS. 1974. Dynamics of moose aggregations in Alaska, Minnesota, and Montana. Journal of Mammalogy 55: 126-137.

PETERSON, R. O., and R. E. PAGE. 1993. Detection of moose in midwinter from fixed-wing aircraft over dense forest cover. Wildlife Society Bulletin 21: 80-86.

QUAYLE, J. F., A. G. MACHUTCHON, and D. N. JURY. 2001. Modeling moose sightability in south-central British Columbia. Alces 37: 43-54.

RANSOM, J. I. 2012. Detection probability in aerial surveys of feral horses. Journal of Wildlife Management 76: 299-307.

RIGGS, M. R., and K. H. POLLOCK. 1992. A risk ratio approach to multivariable analysis of survival in longitudinal studies of wildlife populations. Pages 74-89 in D. R. McCullough and R. H. Barrett, editors. Wildlife 2001: Populations. Elsevier Applied Science, New York, New York, USA.

ROFFE, T. J., K. COFFIN, and J. BERGER. 2001. Survival and immobilizing moose with carfentanil and xylazine. Wildlife Society Bulletin 29:1140-1146.

ROLLEY, R. E., and L. B. KEITH. 1980. Moose population dynamics and winter habitat use at Rochester, Alberta, 1965-1979. Canadian Field-Naturalist 94: 9-18.

SAMUEL, M. D., E. O. GARTON, M. W. SCHLEGEL, and R. G. CARSON. 1987. Visibility bias during aerial surveys of elk in north-central Idaho. Journal of Wildlife Management 51: 622-630.

SCHMIDT, J. I., J. M. VER HOEF, and R. T. BOWYER. 2007. Antler size of Alaskan moose: effects of population density, harvest intensity, and use of guides. Wildlife Biology 13: 53-65.

SHEPHARD, M. E. 1995. Plant community ecology and classification of the Yakutat Foreland, Alaska. US Forest Service, R10-TP-56. Alaska Region, Juneau, Alaska, USA.

SIEGFRIED, W. R. 1979. Vigilance and group size in springbok. Madoqua 12: 151-154.

SOLBERG, E. J., C. M. ROLANDSEN, M. HEIM, J. D. C. LINNELL, I. HERFINDAL, and B. SEATHER. 2010. Age and sex-specific variatiDn in detectability of moose (Alces alces) during the hunting season: implications for population monitoring. European Journal of Wildlife Research 56:871-881.

STEINHORST, R. K., and M. D. SAMUEL. 1989. Sightability adjustment methods for aerial surveys of wildlife populations. Biometrics 45: 414-425.

THOMPSON, I. D., and M. F. VEUKELICH. 1981. Use of logged habitats in winter by moose with calves in northeastern Ontario. Canadian Journal of Zoology 59: 2103-2114.

TIMMERMAN, H. R. 1993. Use of aerial surveys for estimating and monitoring moose populations--a review. Alces 29: 35-46.

--, and M. E. Buss. 1998. Population and harvest management. Pages 559-615 in A.W. Franzmann and C.C. Schwartz, editors. Ecology and Management of the North American moose. Smithsonian Institute Press, Washington, D. C., USA.

WHITE, G. C. 2005. Correcting wildlife counts using detection probabilities. Wildlife Research 32: 211-216.

John R. Fieberg (1) and Mark S. Lenarz (2)

(1) Minnesota Department of Natural Resources, Biometrics Unit, 5463-C West Broadway, Forest Lake, Minnesota 55025; (2) Minnesota Department of Natural Resources, Forest Wildlife Populations and Research Group, 1201 East Highway 2, Grand Rapids, Minnesota 55744, USA.

Susan A. Oehlers (1,2), R. Terry Bowyer (3), Falk Huettmann (2), David K. Person (4), and Winifred B. Kessler (5)

(1) Yakutat Ranger District, Tongass National Forest, 712 Ocean Cape Drive, Yakutat, Alaska 99689 USA; (2) EWHALE lab, Institute of Arctic Biology, and Department of Biology and Wildlife, University of Alaska Fairbanks, Alaska 99775 USA; (3) Department of Biological Sciences, 921 South 8th Avenue, Stop 8007, Idaho State University, Pocatello, Idaho 83209 USA; (4) Division of Wildlife Conservation, Alaska Department of Fish and Game, 2030 Sea Level Drive, Ketchikan, Alaska 99901 USA; (5) 26700 West Fork Road, Prince George, British Columbia, V2K 5L6 Canada.
Table 1 Number of model parameters (k), differences in Akaike's
Information Criterion ([AIC.sub.c]) scores ([DELTA]) and [AIC.sub.c]
weights (wi) for candidate visibility models for moose on the Yakutat
Foreland, Alaska, 2003-2004.

Model    k    Parameters                [AIC.sub.c]      [AIC.sub.c]
                                      [[DELTA].sub.i]     [w.sub.i]

A        5    Age, group (2a),             0.0000           0.4428
              forest cover,
              light, snow

B        7    Age, sex,                    0.7520           0.3041
              observer2 (b),
              speed (c), group (2),
              forest cover,
              snow

C        6    Age, sex,                    1.8083           0.1793
              observer2,
              group (2), forest
              cover, snow

D        4    Group (2), forest            3.5851           0.0738
              cover, snow,
              light

E        16   Saturated (2)               18.8200           0.0000

(a) Group size Z.

(b) Experience secondary.

(c) Wind speed start.

(d) Includes survey start time, temperature, group, sex, age,
experience primary, experience secondary, wind speed start, flight
speed, group size 2, forest cover, vegetation cover, percent cover,
activity, light, snow cover, and elevation.

Table 2. Regression coefficients and risk ratios (RR) for selected
composite overall explanatory model forvisibility of moose on
the Yakutat Foreland, Alaska, 2003-2004. Confidence intervals did not
overlap 1 in the individual models.

Variable           [beta]        SE        RR       RR 90% CI

Intercept          -14.284    16.844      n/a             n/a

Age                  0.325     0.169     1.384    1.047-1.829

Group size (2)       0.074     0.075     1.077    0.951-1.219

Forest cover        -2.849     0.945     0.058   0.012- 0.275

Light                1.010     1.157     2.746   0.407-18.524

Snow cover          -1.419     1.201     0.242    0.033-1.755
1 (0-33%) (a)

Snow cover          -1.661     1.059     0.190    0.033-1.090
2 (34-66%) (a)

Sex (b)             -0.256     0.352     0.774    0.433-1.384

Flight Speed         0.063     0.075     1.065    0.941-1.205

Experience           0.562     0.693     1.754    0.559-5.504
Secondary

(a) Snow cover is relative to the reference variable of level 3, 67-
100%.

(b) Sex is relative to the reference variable of male.

Table 3. Number of model parameters (k), differ-ences in Akaike's
Information Criterion ([AIC.sub.c]) scores ([DELTA]), and AICc weights
(wi) for candidate visibility management models for moose on the
Yakutat Foreland, Alaska, 2003-2004.

                                     [AIC.sub.c]      [AIC.sub.c]
Model    k    Parameters           [[DELTA].sub.i]     [w.sub.i]

A        4    Group (2a),                     0.000          0.540
              forest cover,
              snow, light

B        5    Group (2),                      1.061          0.318
              forest cover,
              vegetation
              cover, snow,
              light

C        3    Group (2), forest               2.684          0.144
              cover, snow

D        14   Saturated (b)                  14.480         0.0004

(a) Group size z.

(b) Includes survey start time, temperature, group, experience
primary, experience secondary, wind speed start, flight speed, group
[size.sup.2], forest cover, vegetation cover, percent cover, activity,
light, snow cover, and elevation.

Table 4. Regression coefficients and risk ratios (RR) for selected
composite management model for visibility of moose on the Yakutat
Foreland, Alaska, 2003-2004. Confidence intervals did not overlap 1 in
the individual models.

Variable          [beta]      SE        RR        RR 90% CI

Intercept           0.048     0.905      n/a             n/a

Group size (2)      0.070     0.038     1.073    1.007-1.142

Forest cover       -2.551     2.190     0.078    0.002-2.894

Light               1.441     0.920     4.225   0.926-19.279

Snow cover 1       -1.028     0.935     0.358    0.076-1.673
(0-33%) (a)

Snow cover 2       -1.377     0.897     0.252    0.057-1.109
(34-66%) (a)

Vegetation         -0.284     0.383     0.753    0.400-1.416
cover

(a) Snow cover is relative to the reference variable of level 3.
67-100%.
COPYRIGHT 2012 Alces
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2012 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Oehlers, Susan A.; Bowyer, R. Terry; Huettmann, Falk; Person, David K.; Kessler, Winifred B.
Publication:Alces
Article Type:Abstract
Geographic Code:1USA
Date:Jan 1, 2012
Words:7971
Previous Article:Comparing stratification schemes for aerial moose surveys.
Next Article:Age, sex, and seasonal differences of carcass weights Of moose from the Central Interior of British Columbia: a comparative analysis.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters