https://cpw.cvlcollections.org/files/original/32bec22c77f6933bdd7e40a5281b4832.pdf bb0d3a8d8ba4eda7b78ffe841cdaa6f0 PDF Text Text Dan Prenzlow, Director, Colorado Parks and Wildlife • Parks and Wildlife Commission: Marvin McDaniel, Chair • Carrie Besnette Hauser, Vice-Chair Marie Haskett, Secretary • Taishya Adams • Betsy Blecha • Charles Garcia • Dallas May • Duke Phillips, IV • Luke B. Schafer • James Jay Tutchton • Eden Vardy �Anderson, Charles R. J r . , A Siahtabilitv Model for Moose Developed From Helicopter Surveys in Western W y o m i n g . M.S., Department of Zoology & Physiology, May, 1994. I conducted helicopter surveys of radiocollared moose (Alces alces) to determine factors that influence moose sightability from the air and to develop a predictive model for future surveys. Variables measured were time of day, sex/age composition of groups, % vegetative cover and vegetative cover type, % snow cover, topography, moose activity, light intensity, group size, perpendicular distance to the group, study area, and primary observer. I determined significant variables using logistic regression analyses. Multivariate analyses indicated that % vegetative cover was the only variable influencing sightability. Further analyses, however, suggested an interaction between group size and topography may also be important. My final model selection was based on compromises between statistical significance and biological interpretation. I selected the model that included only % vegetative cover over the more complicated models. The model correctly classified 83% of 104 observations. Estimator precision is best maintained when moose are using open cover types (<50% vegetative cover). 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �A SIGHTABILITY MODEL FOR MOOSE DEVELOPED FROM HELICOPTER SURVEYS IN WESTERN WYOMING by Charles R. Anderson Jr. A thesis submitted to the Department of Zoology & Physiology and The Graduate School of The University of Wyoming in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE in ZOOLOGY & PHYSIOLOGY Laramie, Wyoming May, 1994 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �UMI N um ber: E P 17480 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignm ent can adversely affect reproduction. In the unlikely event that the author did not send a com plete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. UMI ® UMI Microform EP17480 Copyright 2007 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �To The Graduate School: The members of the Committee approve the thesis of Charles R. Anderson Jr. presented on February 1, 1994. rederick G^"-ii±hclz Michael P./G i M i n g h a m APPROVED: William A. Gern, Head, Department of Zoology & Physiology Thomas G. Dunn, Dean, The Graduate School Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �ACKNOWLEDGEMENTS The involvement of several Wyoming Game and Fish personnel was critical in obtaining the objectives of this project. I thank Dave Moody and Joe Bohne for their much needed assistance through most aspects of this study. Doug McWhirter, Bill Long, Steve Kilpatrick, Garvice Roby, Ron Lockwood, Neil Hymas, and Glen Stout were invaluable as aerial observers. Don Kwiatkowski and Tom Thorne provided valuable veterinary assistance. I thank Fred Reed, Merlin Hare, and Bob Robertson of Western Air Research for their fixed-wing expertise. Helicopter pilots Claud Tyrell and Dan Williams of Hawkins & Powers Aviation and Dave Savage of Savage Air Services were gratefully appreciated. Special thanks to Steve Kerr, Colorado State University Veterinary Teaching Hospital, for lending his veterinary skills at a critical time in this project. I also thank Bill Lance of Wildlife Pharmaceuticals, Inc. for supplying the moose sedatives. Thanks to Dave Leptich, Idaho Dept, of Fish & Game, for demonstrating this survey technique, and Oz Garton, University of Idaho, for explaining the statistics. Global Position System technology was explained by Eric ii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Winthers, Bridger-Teton National Forest, USFS. Richard Anderson-Sprecher, University of Wyoming, gave valuable statistical advice. Don Whittaker and Loren Ayers provided helpful suggestions when I was office bound, and Mike Hooker provided valuable assistance throughout the field stages of this study. My graduate committee members, Dr. Fred Lindzey, Dr. Mike Gillingham, and Dr. Doug Bonett, were very helpful throughout all phases of my graduate experience. I also thank the Wyoming Game and Fish Department for funding this project. Finally, I thank my wife Tracy for making long wintery treks across Wyoming and putting up with my absence for the past 2 years, and my parents Chuck and Denise for their support and encouragement. iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �TABLE OF CONTENTS INTRODUCTION............................................ 1 DESCRIPTION OF STUDY AREAS............................ Greys River Ar e a ................................. Snake River A r e a ................................. Hams Fork Ar e a .................................... 12 12 14 14 METHODS................................................. Moose Capture..................................... Sightability Trials................................ Sightability Model................................ Model Selection.................................. Model Precision.................................. 15 15 16 23 26 27 RESULTS................................................. Sightability Analysis............................ Sightability Model Selection..................... Predictive Sightability Model.................... Sightability Model Precision..................... 29 37 39 43 45 DISCUSSION.............................................. Model Selection................................... Factors Influencing Moose Sightability.......... Comparison to Other Sightability Model Research. Model Precision................................... i Management Summary................................ 48 48 50 57 59 61 LITERATURE CITED....................................... 64 APPENDIX A. 70 Sightability Data Sheet................. iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �INTRODUCTION Aerial surveys are the only practical way to estimate ungulate numbers in most of North America (LeResche and Rausch 1974, Timmerman 1974, Gassaway and Dubois 1987). These surveys, however, often provide biased estimates and only under specific conditions do they allow detection of even large population changes (Caughley 1974, Gassaway et al. 1985). Ideally, aerial survey estimators should be accurate, precise, cost effective (Gassaway et al. 1986), and repeatable to provide timely management decisions. The 2 major sources of error encountered in aerial surveys are caused by the spatial variability of animals (sample variance; Steinhorst and Samuel 1989) and failure to observe all animals during surveys (visibility bias; Caughley 1974, LeResche and Rausch 1974, Samuel and Pollock 1981, Steinhorst and Samuel 1989). Recent advances in aerial survey techniques (Cook and Jacobson 1979, Crete et al. 1986, Gassaway et al. 1986, Bartmann et al. 1987, Samuel et al. 1987, White et al. 1989) have improved precision by incorporating estimates for most sources of variation and decreased estimator bias so that repeated estimates may center near the true value (Gassaway and Dubois 1987). 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Seldom, if ever, are all animals seen during aerial surveys thus causing biased estimates of population size and composition. This obvious bias in aerial surveys led to the development of methods that correct for unseen animals. Techniques recently developed to correct for animals missed during aerial surveys include mark-resight (Bartmann et al. 1987), resurvey (Gassaway et al. 1986), line transect (White et al. 1989, Johnson et al. 1991), and logistic regression models (Samuel et al. 1987, Ackerman 1988, Otten et al. 1993). Application of techniques that estimate sightability correction factors has greatly enhanced population statistics. The level of accuracy and precision from these techniques, however, is limited by the ability to meet their respective assumptions. Mark-resight estimates of population size (Bartmann et al. 1987, Neal et al. 1993) are derived from at least 1 survey where correction factors are estimated from the proportion of marked animals seen in the population. Unbiased population estimates are obtained from this method if: 1) the population is geographically and demographically closed, 2) animals do not loose their marks, 3) marked animals are correctly identified, counted, and recorded, and 4) each animal has the same sighting probability during each sighting occasion (Otis Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �et al. 1978). Although most of these assumptions can be met with adequate survey design, assumptions of equal sightability may be difficult to meet if factors causing sightability bias are variable (e.g., group size, habitat type). Neal et al. (1993) reported that mark-resight estimates of mountain sheep (Ovis canadensis) were very sensitive when sightability was heterogeneous. Monte Carlo simulations suggested that although bias was small, estimates were less precise than confidence intervals suggested. Additionally, Bartmann et al. (1987) recommended that >45% of small populations should be marked to provide reliable mark-resight estimates of mule deer (Odocoileus hemionus) . This marking intensity may be cost prohibitive. The resurvey estimator (Gassaway et al. 1986) is the most commonly used correction technique for moose (Alces alces) surveys in North America (Gassaway and Dubois 1987). Application of this method requires that a standard search intensity (1.5-2.4 min/km2) be conducted over a random sample of search units for each strata, that a portion of search units be resurveyed at a higher search intensity (4.6 min/km2) to correct for animals missed during the standard search, and that a previously determined correction factor from radiocollared animals missed during intense searches be applied. Gassaway et Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �al. (1986) concluded it was not feasible or economical to estimate correction factors in low density strata and therefore pooled correction factor information over all strata. Becker and Reed (1990) criticized this approach and proposed that sightability was not identical for all strata. They suggested that correction factor information from intense searches be gathered in each strata to account for heterogeneous sightability between strata. This would alleviate some of the bias associated with heterogeneous sightability. Sightability, however, may not be constant within strata if canopy structure or other factors influencing sightability are variable. Also, the resurvey technique must rely on correction factors derived from radiocollared animals because moose are missed even under a high search intensity (Gassaway et al. 1986). Population estimates will be biased and estimator precision variable if this correction factor is applied as a constant and not adjusted for variable sighting conditions. Estimating abundance using line transect sampling has been applied to a variety of species (Buckland et al. 1993). Correction factors for line transect estimates are derived from a detection function that decreases with distance from the transect line. must be observed. All animals on the line Aerial line transect surveys have been Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �evaluated for moose (Thompson 1979), mule deer (White et al. 1989), and pronghorn (Antilocapra americana. Johnson et al. 1991). Johnson et al. (1991) reported that the assumptions of line transect sampling could be met when applied to pronghorn aerial surveys. White et al. (1989) and Thompson (1979), however, reported that line transect estimates of mule deer and moose numbers, respectively, were underestimated in some cases. They felt that this was largely attributed to animals being missed on the transect line during surveys. Logistic regression models have recently been developed for elk (Cervus elaphus: Samuel et al. 1987, Otten et al. 1993) and mule deer (Ackerman 1988). These sightability models predict the probability of occurrence for a dichotomous dependent variable (a group seen or missed) from aerial surveys via logistic regression analysis. Correction factors are developed from variables that influence sightability (e.g., group size, canopy cover) during initial model development using radiocollared animals and subsequently applied on a group by group basis during surveys. The application of logistic regression models to estimate population parameters assumes: 1) population closure, 2) animal groups are independently observed, 3) groups are completely counted and counted only once, 4) the survey Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �design for land units can be specified, and 5) the detection probability of a group can be estimated (Steinhorst and Samuel 1989). If a segment of the population has no chance of being detected then assumption 5 cannot be met (Samuel et al. 1987). Proper survey design, however, should greatly reduce or eliminate this situation in most cases. The sightability model approach is closely related to the mark-recapture and resurvey methods discussed previously. distinct difference. There is, however, 1 Estimating sightability on a group by group basis based on the influence of environmental and/or behavioral factors does not require that all animals have an equal and independent probability of being sighted. Additionally, marked animals are only needed during model development. Visibility bias during aerial surveys is exacerbated by a variety of factors (Pollock and Kendall 1987). Experience and number of observers, length of surveys, aircraft type, flight pattern, air speed, height above ground, and timing of surveys are factors that can be standardized to improve accuracy and precision of estimates and permit comparison of data among years (Unsworth et al. 1991). Group size (Cook and Jacobson 1979, Samuel and Pollock 1981, Gassaway et al 1985, Samuel et al. 1987, Ackerman 1988), canopy cover (Samuel et al. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �1987, otten et al. 1993), cover type (LeResche and Rausch 1974, Gassaway et al. 1985, Ackerman 1988), show conditions (LeResche and Rausch 1974, Gassaway et al. 1986), light intensity (LeResche and Rausch 1974, Bisset and Rempel 1991), time of day (LeResche and Rausch, Timmerman 1974, and Crete et al. 1986) and animal activity level (Gassaway et al. 1985, Ackerman 1988) may also contribute bias to aerial survey estimates but are difficult or impossible to control. Sightability may also decrease as distance from observers increases (Burnham and Anderson 1984). Important uncontrollable variables can be incorporated into a logistic regression model to improve population estimates and estimator precision over other correction methods when perfect visibility along the flight path or homogeneous sightability can't be achieved. Steinhorst and Samuel (1989) developed a modified Horvitz-Thompson estimator to estimate variance components associated with the logistic regression estimator. They concluded that survey sampling error, sightability error, and the error associated with estimating the sighting probabilities (model error), needed to be addressed to account for the spatial variability of animals and visibility bias of animal groups. Survey sampling error can be controlled by employing a proper sampling scheme (e.g., stratified random sample of land units), however, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �the spatial distribution of animals and the degree of sampling effort will determine the amount of variance associated with this component. The variance associated with sightability error is dependant on the magnitude of the correction factor applied to each observed group. If small groups were more difficult to detect than large groups, for example, the correction factor applied to small groups would be greater and the resulting sightability variance would increase with the number of small groups seen and corrected. The third variance component, model error, is derived from the variation not explained by the model parameters in predicting sightability (e.g., "noise" from observer variation, changes in snow conditions). Most population correction methods disregard the potential variation from extraneous factors and assume that factors such as observers, pilots, lighting conditions are fixed and do not contribute variability to the population parameters being estimated. When these factors are disregarded, estimates may be overly precise. The relative influence of each variance component can change markedly depending on survey design, visibility bias, and the sample size used to estimate sighting probabilities (Steinhorst and Samuel 1989). Estimates of population composition may be similarly biased if all sex and age groups are not seen with equal Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �probability (Samuel et al. 1992). Habitat preferences for moose differed among sex and age groups in Ontario (Thompson 1979, Novak 1981) and Alaska (Gassaway et al. 1985) causing some classes to be more difficult to see than others during aerial surveys. Similarly, biologists in Wyoming observed higher bull:cow ratios during severe winters than during mild winters (D'. Moody, Wyo. Game and Fish Dep., pers. commun.) indicating that habitat selection by bulls during mild winters may have made them more difficult to see than cows during traditional trend surveys. Applying correction factors to each group during surveys should account for this bias and provide reliable sex and age composition estimates. Annual winter trend counts of moose are currently used in Wyoming as a measure of population model performance. These surveys are traditionally applied only to areas of concentrated use. Sampling theory for finite populations, however, requires the establishment of a sampling frame where sampling units (i.e., land units and/or animals) must have non-zero probabilities of being sampled (Samuel et al. 1992). While it is widely accepted that animals are missed during aerial surveys (LeResche and Rausch 1974, Caughley 1977, Cook and Jacobson 1979, Pollock and Kendall 1987), trend counts are not adjusted for moose missed during surveys. Consequently, winter Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �trend counts for moose are not accompanied by estimates of variance or sightability bias leaving managers with no measure of the quality of these data. Sightability models developed for elk in Idaho (Samuel et al. 1987) and Michigan (Otten et al. 1993) were compared against elk drive counts in Montana (Unsworth et al. 1990) and Michigan (Otten et al. 1993), respectively, and in both cases provided abundance estimates comparable to drive counts. Sightability models have not been developed and applied to estimate moose population parameters. My objectives were to standardize most controllable factors during aerial moose surveys, measure the bias associated with the factors that were not controllable, and develop a logistic regression model to estimate the sightability of moose. The influences of cover type, percent vegetation cover, percent snow cover, group size, topography, sex/age segregation, moose activity, perpendicular distance, light intensity, time of day, observers, and study areas were examined to test the hypotheses that these factors do not significantly influence (P < 0.05) aerial moose sightability. Individual factors that influenced sightability were examined collectively with multivariate analysis to test the hypothesis that these factors are not interrelated. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �11 These hypotheses were tested to assist in the development and evaluation of a moose sightability model. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �STUDY AREAS AND METHODS STUDY AREAS Three study areas were chosen in western Wyoming (Fig. 1). These areas afforded a variety of habitat types that supported wintering moose, and are representative of moose winter range throughout the state. Grevs River Area. The Greys River Area (GRA), predominantly southeast of Alpine, WY (Sublette Moose Herd Unit, Moose Hunt Area 23), was bounded by the Salt River Range on the west, the South Fork of Indian Creek on the north, the Wyoming range on the east, and Moffat and Deadhorse Creeks on the south. Primary drainages in GRA were the Greys River and the Little Greys River. Elevations ranged from about 1829 m (6000 ft) to 2499 m (8200 ft). GRA was topographically rugged with steep-sided valleys and narrow riparian corridors. Riparian vegetation was dominated by willow (Salix spp.) interspersed with narrowleaf cottonwood (Populus anaustifoliaf and Engelmann spruce (Picea enoelmannii) . Upland sites supported stands of aspen (Populus tremuloides), lodgepole pine (Pinus contorta), and Douglas-fir (Pseudotsuga menziesii) sagebrush (Artemisia tridentata) . interspersed with big Higher elevations were 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �13 22 JACKSON r \~ x i SRA r ^ l GRA HFA 189 RIVER ALPINE O PINEDALE 191 89 x < o 189 30 KEMMERER Fig. 1. Location of the Snake River (SRA), Greys River (GRA), and Hams Fork (HFA) moose sightability study areas in western Wyoming. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �14 vegetated primarily by subalpine fir (Abies lasiocarpa) and spruce. Upland shrubs common throughout the area included rose (Rosa wopdsii), mallow ninebark (Phvsocarpus malvaceus), Saskatoon serviceberry (Amelanchier alnifolia), and common snowberry (Svmphoricarpos albus: Rudd and Irwin 1985). Snake River Area. The Snake River Area (SRA), southwest of Jackson, WY (Sublette Herd Unit, Moose Hunt Area 20), was bounded by the Snake River Range on the west, State Highway 22 on the north, U. S. Highway 191 on the east, and Dog Creek on the south. Elevations ranged from about 1829 m (6000 ft) to 2103 m (6900 ft). Topography in SRA was flat to moderately steep with a broad flood plane formed by the Snake River to the east and several narrow drainages emanating from the mountains to the west. The Snake River floodplain was vegetated primarily by narrowleaf cottonwood and willow with infrequent pockets of spruce. Douglas-fir and quaking aspen were dominant in upland areas with big sagebrush occurring occasionally. Hams Fork Area. The Hams Fork Area (HFA), northwest of Kemmerer, WY (Lincoln Moose Herd Unit, Moose Hunt Area 26), was bounded by U. S. Highway 30 and the Smith's Fork River on the west, an east-west line along the southern edge of Lake Alice on the north, Commissary Ridge on the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �15 east, and U. S. Highway 30 on the south. Elevations ranged from about 2164 m (7100 ft) to 2499 m (8200 ft). HFA was a complex riparian system with flat to rolling topography within several broad river systems. A variety of willow species dominated riparian areas. Upland areas at lower elevations were vegetated primarily by big sagebrush and aspen. Subalpine fir and lodgepole pine were the dominant species at higher elevations with aspen occurring in some areas. Big sagebrush, Saskatoon serviceberry, and antelope bitterbrush (Purshia tridentataf were interspersed throughout higher elevations. METHODS Moose Capture. Moose were approached on foot and immobilized with drugs administered in a dart fired from a rifle (model 171c, Williamsport, P A ) . .50 cal., Pneu Dart, Inc., We used 20 mg Medetomidine with 600 mg Ketamine or 3 mg Carfentanil with 50 mg Xylazine (Wildlife Pharmaceuticals, Inc., Ft. Collins, CO). The effects of Medetomidine and Carfentanil were antagonized with 80 mg Atipamezole (20 mg intravenous and 60 mg subcutaneous) and 300 mg Naltrexone (0.75 mg intravenous and 2.25 mg subcutaneous; Wildlife Pharmaceuticals, Inc., Ft. Collins, CO), respectively. Moose were fitted with radio Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �16 transmitter collars with activity sensing options (MOD500, 148 MHz, Telonics, Inc., Mesa, A Z ) , blood samples were taken, and ear tags were attached. Capture location, sex, estimated age, transmitter frequency, collar symbol, and ear tag numbers were recorded. Ritchey plastic livestock collars (Ritchey Sales, Brighton, CO; white with black symbols, 11.4 cm wide) were attached to each radiocollar to. aid aerial observers in identifying moose once located. Sightability Trials. Pre-survey flights were conducted using a fixed-wing aircraft (Maule M5) to locate radiocollared moose. The fixed-wing pilot then randomly located a 4.6 km2 (1.8 mi2) circular plot over each moose location and radioed the coordinates to the helicopter survey crew. We used a GPS Apollo 820™ receiver (II Morrow, Inc., Salem, OR) to determine moose locations during pre-survey flights, and a GPS Pathfinder Basic™ receiver (Trimble Navigation, Ltd., Sunnyvale, CA) to navigate the helicopter and delineate search unit boundaries. Survey crews were directed to search units that contained 0-2 radiocollared moose groups (>1 moose/group). Sightability surveys were flown in a turbo-charged Bell-47, a Hiller-Soloy, or a Hiller-12E helicopter. All 3 helicopters were structurally similar except for engine Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �type and afforded good visibility with 3 seats abreast. The helicopter crew consisted of a pilot, a primary observer (experienced in aerial moose surveys and familiar with the survey area), and a secondary observer (at least some aerial survey experience). The pilot was seated in the middle with observers on either side for both Hillers; the pilot sat on the left and observers on the right in the Bell. All 3 crew members assisted in spotting and classifying moose. The primary observer recorded data, and the secondary observer operated the GPS receiver and located moose with a telemetry receiver (TR-2, 148/150 MHz, Telonics, Inc., Mesa, AZ) after the survey. Observers were limited to a maximum of 4 hours/day in the helicopter to minimize observer fatigue. Search units were typically surveyed flying 150-250-m (492-820 ft) wide strip transects in areas of minimal topographic relief, or contour intervals in broken topography. Distance between transects was least in areas with greatest canopy cover. Surveys were flown at 30-46 m (100-150 ft) above the ground at a speed of 56-72 km/hr (35-45 m p h ) . Surveys typically began at low elevations and progressed upslope. When the survey block was approached the specific radiocollar frequency was selected and audio output fed directly to a tape recorder. and observers could not hear the signal. Pilot Radiocollared Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �18 moose missed during surveys were located using helicopter mounted telemetry eguipment immediately after a search unit was surveyed. Group size, topography (flat, moderate, or steep), vegetation type, percent vegetative cover, percent snow cover, light intensity (flat or bright), sex and age (calf or adult) of each group member, perpendicular distance from the group to the flight line, and moose activity were recorded for radiocollared moose seen or missed during surveys (Appendix A ) . Observers, study area, and time required to survey each search unit were also recorded. Vegetation type was classified into categories based on dominant species and structure where the group was first seen (Appendix A ) . Percent snow and vegetative cover (recorded in 5% increments) were estimated within a 9-m (30 ft) perimeter around a group where it was first observed. Percent vegetative cover was determined by flying a complete circle at an oblique angle around the point where moose were first seen, and estimating the proportion of that area that was obscured from view by vegetation (Fig. 2). Any vegetation that blocked the view of the animals was considered vegetation cover (Unsworth et al. 1991:7, Appendix B ) . Helicopter locations were recorded every 3 seconds using a GPS Pathfinder Basic™ receiver to delineate Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Fig. 2. Examples of percent vegetative cover as estimated during helicopter sightability trials of moose in western Wyoming, winter 1993 (After Unsworth et al. 1991). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �transect paths during the survey. Moose locations were determined with the GPS receiver by flying over the position where the group was first observed and recording that location. Transect paths and moose locations were differentially corrected using the computer software package PFINDER™ (version 2.14, Trimble Navigation, Ltd., Sunnyvale, CA) with GPS base data obtained from USFS base stations in Jackson, WY, McCall, ID, or Missoula, MT to maintain position accuracy within 5 m (Trimble Navigation, Ltd. 1992:39). Some elevations associated with 2- dimensional GPS positions appeared incorrect and may have resulted in inaccurate locations. Consequently, if the altitude of a 2-dimensional GPS position differed >50 m from the corresponding topographic map elevation, the position was adjusted to reflect the map elevation. Thirty meters was added to each map elevation to account for the height of the helicopter. Distance from the transect path to moose group locations was measured in perpendicular distance from the closest transect path (for groups missed during surveys) or the path from which the group was seen (for groups seen during surveys) using the measure command in PFINDER™ (Fig. 3). I observed radiocollared moose from the ground and documented their activity while simultaneously tape recording signal characteristics to develop a signal Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �21 Diet ( N ) ; 57.615422, Fare fen : 24B.B63174. Back fen : (B.ffiZTO Fig. 3. Map depicting flight pattern (top) and perpendicular distance from transect path to moose location (bottom) during a helicopter sightability trial of moose in western Wyoming, winter 1993. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �profile to estimate moose activity from signal characteristics alone. Activity for radiocollared moose, seen during surveys, was recorded for the first group member observed. I determined activity of radiocollared moose missed during surveys by comparing their activity when located with telemetry equipment after the survey to the tape-recorded characteristics of their radio signal during the survey flight. Moose activity was initially recorded as bedded, standing, or moving when observed. I did not feel that I could reliably document whether a moose that was missed was standing or moving during surveys based on observations after surveys or recorded signal characteristics. Therefore, all moose activity data were reclassified as simply bedded or active (standing or moving). I examined recorded pulse rates for moose missed during surveys for the 4 minutes corresponding to the transect path near moose locations (from GPS data) to determine if the recorded signal characteristics agreed with the observed activity of the moose when it was located after the survey. I assumed a slow constant pulse rate (about 65 beats/min, head up) indicated a bedded moose and a variable or fast pulse rate (about 100 beats/min, head down) indicated an active moose. I also considered a moose bedded if the pulse rate changed from slow to variable or fast for a single episode Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �23 <15 seconds during the time period examined. If the recorded radiocollar frequency indicated a different behavior during the survey than was observed when the moose was first seen after the survey, activity adjusted from the tape-recording was used in analysis. Siahtabilitv Model. I used univariate analyses to determine individual relationships of the independent variables with the dichotomous dependant variable (moose groups seen or missed), and to determine appropriate groupings for the discrete variables. Categorical independent variables (time of day, sex/age groupings, topography, moose activity, light intensity, cover type, study area, and primary observer) were examined using x2 contingency analysis (Zar 1984, SAS Institute, Inc., 1988; Proc Freq). Continuous independent variables (% vegetative cover, % snow cover, group size, and perpendicular distance) were examined using univariate logistic regression (Hosmer and Lemeshow 1989, SAS Institute, Inc., 1990; Proc Logist). The likelihood ratio X2 test from a contingency table is identical to the value of the likelihood ratio test of a univariate logistic regression model where the dependant variable is dichotomous (y = 0, 1; Hosmer and Lemeshow 1989). I combined categories for discrete independent variables if the x2 score improved and biological interpretation Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �24 remained intact. Additionally, a t-test was used to determine if bull groups and cow and cow/calf groups selected different vegetative cover densities based on percent vegetation cover observed for each group during sightability trials. Multivariate analysis of data from moose groups seen or missed during helicopter sightability trials was conducted using forward stepwise logistic regression (Dixon et al. 1988; BMDP-LR, Hosmer and Lemeshow 1989, SAS Institute, Inc., 1990; Proc Logist) where the dependant variable (groups seen or missed) is dichotomous with values 1 (probability 9) and 0 (probability 1 - 9 ; Kleinbaum et al. 1988). I used maximum likelihood estimation to predict model parameters. The dichotomous classification of moose groups seen or missed served as the dependent variable, and all other variables served as the independent variables. To guard against the inadequacies of stepwise analysis (Jones and McCulloch, 1990), I also included significant variables (P < 0.05) from univariate analyses in the logistic model (1 at a time) with variables selected after multivariate stepwise analysis to determine if their significance became important. I used significant (P < 0.05) independent variables to develop the sightability model to predict the probability of observing moose groups under a variety of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �25 environmental factors and behavioral conditions. The logistic regression is: exp" 7T = -------------- ' 1 + expu where re is the sighting probability and u = 60 + BiXj + R 2x 2 + ... + 6kxk is the linear regression equation of variables (xlf x2, ..., xk) significantly influencing sightability. inverse of n The is the correction factor applied to each group observed during surveys. Non-linear transformations of continuous variables were tested for significance to test the assumption that these variables in the logit are linear. Reciprocal, natural log, square root, squared, and cubed transformations of all continuous variables and arcsine square root transformations of percent vegetative cover and percent snow cover were also examined. Two-way interactions of all independent variables were considered as well as the 3-way interaction between percent vegetative cover, group size, and topography. Problems were encountered with independent categorical variables in which moose were always observed Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �and never missed in a given category (i.e., cover type and observers), and when the product of an interaction term resulted in categories where all moose were seen or missed. The asymptotic x2 distribution may not hold for highly imbalanced data causing x2 values to be biased high from Wald, likelihood ratio, and scores x2 tests (Zar 1984:49, 70, Mehta and Patel 1993:Appendix A). To examine the influence of imbalanced categorical variables, I used the exact scores statistic (Mehta and Patel 1993) when possible. However, when some variables containing 0 cell frequencies (i.e., all seen or missed) were forced into the logistic regression model I could not derive x2 statistics because the maximum likelihood estimates were not obtainable and computations were too complex to determine the exact scores. When this occurred, I examined the stepwise significance of the variable outside the model containing previously selected variables. Model Selection. I developed a number of models during the model building phase of analysis. Model selection was based on compromises between: 1) significant (P > 0.05) improvement in the x2 score from the Wald test for variable addition, 2) biological soundness, 3) the change in predictability (i.e., percent of moose observations correctly classified as seen or missed by the given model) as model complexity decreased, and 4) the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �27 relative change in coefficients already in the model by removal of potentially important variables (Hosmer and Lemeshow 1989:88). I examined model fit using Hosmer- Lemeshow and Brown's goodness-of-fit x2 tests (Hosmer and Lemeshow 1989). Model Precision. To illustrate the influence of percent vegetative cover on precision of population estimates and the relative contribution of the variance components, I constructed 2 data sets from moose groups that were seen during model development. The first data set was developed to illustrate survey results expected in dense canopy habitats (e.g., late winter surveys) and the second to illustrate results expected in relatively open habitats (e.g., Hams Fork Area). Number of moose (n = 181) and group size distribution was identical in the 2 constructed data sets, but percent vegetative cover was <70% for the first (moose were not seen in >70% vegetation cover during model development) and <50% for the second. Moose seen in >50% vegetative cover in the first constructed data set were simply reassigned 50% vegetative cover for the second data set. To incorporate sampling variance in this comparison, I assumed both data sets were sampled without replacement and with similar intensity (i.e., high density strata = 91%, low density strata = 50%). I then removed the same randomly selected search Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �28 units from each constructed data set to examine the influence of reduced sampling intensity on variance estimates (i.e., high density strata = 82%, low density strata = 40%). These data sets were constructed to assist in the development of future sampling strategies. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �RESULTS Seventeen moose (2 yearling and 5 adult bulls, and 1 yearling and 9 adult cows) were radiocollared in GRA; 1 bull lost his collar and 1 cow died before sightability trials. Nine moose (1 yearling and 3 adult bulls, and 1 yearling and 4 adult cows) were radiocollared in SRA. Moose sightability trials were conducted in GRA on February 28, March 1, 3, 21, 23, and 29, 1993 (n = 44), and in SRA on February 8, 26, March 2, 4, 22, 24, and 26, 1993 (n = 52). Sightability trials were also conducted on April 14 and 15, 1993 in HFA sampling 5 moose (5 adult cows, n = 9) radiocollared during another study (Lockwood 1991). Ninety, 9, and 6 moose groups were sampled in the Hiller-Soloy, Hiller-12E, and Bell-47 helicopter, respectively. Nine, 91, and 7 of the units surveyed contained 0, 1, and 2 radiocollared moose, respectively. Average survey time for each search unit completely searched was 21.2 minutes (SD = 5.4 mi n ) . I collected 105 observations of 29 radiocollared moose from 6 February to 15 April 1993. On 1 occasion a radiocollared moose missed during the survey was later found standing in open water. The dark background of the water blended well with the dark coloration of the moose. The influence of percent snow cover (0%) was strongly 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �30 weighted by inclusion of this observation since this moose was likely missed because it blended with its surroundings and not necessarily because of lack of snow cover. Additionally, I believe that our search image changed once we were aware of the possibility of missing moose in open water, and therefore we increased our search intensity over water during subsequent surveys. Two other moose were later seen in water during surveys. Consequently, I deleted the first observation of a moose standing in water from further analysis. Of the 104 remaining moose groups, 61 were seen (x = 1.52 moose/group) and 43 were missed (x = 1.28 moose/group) during surveys. I was unable to calculate distance from the transect path to moose on 18 occasions due to problems with the GPS receiver. Ocular estimates of distance to moose seen proved inaccurate. Therefore, the influence of perpendicular distance on aerial moose sightability was analyzed using the remaining 86 observations. Fifty were from moose seen during surveys and 3 6 were from moose missed. All 86 transect path and moose location files were differentially corrected from GPS base station data. Distance measurements from 67 of the transect path and corresponding location files were from 3-dimensional positions. Distance measurements from the remaining 19 files were from 2-dimensional transect path, location, or Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �31 transect path and location files. Elevations for 2- dimensional positions were adjusted for 8 of the 19 distance measurements because elevations differed >50 m from corresponding topographic map elevations. Perpendicular distance from the transect path was greater for moose that were seen than moose that were missed (Table 1). method. This difference was due to our sampling Perpendicular distance was measured to the point where the radiocollared moose group was first seen regardless of whether the distance exceeded transect spacing (<548 m ) . Perpendicular distances for moose that were missed, however, were restricted to being no greater than 1/2 the transect width (<154 m ) . Twelve moose groups were encountered at distances >125 m. surveys. Two groups were missed and 10 were seen during Nine of the 10 groups seen were in open habitat types (<25% vegetation cover). The tenth group was initially missed at 65 m, and observed from the next transect path at 261 m in 70% cover. To retain comparability in the data (i.e., group long distances for moose seen in open habitat with a distance category where moose could be potentially missed), I grouped perpendicular distances into 25-m categories and pooled observations beyond 125 m into a single category. Therefore, distance values from 1-6 were used in analyses. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1. Results of univariate and multivariate analyses of independent variables measured during helicopter sightability trials of moose in western Wyoming during winter, 1993. Variable n % Seen Stepwise15 significance x2 (univariate) Univariate® significance x2 (multivariate) 0.962 0.916 1.480 0.830 2.718 0.257 0.960 0.620 0.418 0.518 0.500 0.478 0.067 0.795 0 i850 0.356 DISCRETE Time of day 0800-1000 1000-1200 1200-1400 1400-1600 1600-1800 11 28 24 30 11 64 61 58 60 45 Sex/age Bull(s) Cow(s) Cow/Calf 41 34 29 49 65 66 45 62 59 56 28 76 61 58 Animal activity Bedded Standing/ Moving Light intensity Flat Bright u to �Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1. Continued. Variable n % Seen x2 x2 Stepwiseb significance (univariate) Univariate® significance 5.533 0.019 3.070 0.080 37.331 <0.001 4.390 0.222 7.908 0.019 2.210 0.332 (multivariate) DISCRETE Topography Flat Moderate/ Steep 48 71 56 48 6 100 Cover Type Open/Water Deciduous Shrub Deciduous Timber Conifer 17 100 23 58 83 33 Study areas0 SRA GRA HFA 51 44 9 65 45 89 u> u> �Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1. Continued Variable n % Seen x2 x2 (univariate) Univariate® significance (multivariate) 19.455 0.007 13.220 0.067 2.623 0.105 0.240 0.625 5.596 0.018 3.020 0.082 Stepwiseb significance DISCRETE Observers'1 1 2 3 4 5 6 7 8 28 26 15 11 8 7 5 4 46 62 87 55 50 14 100 75 Distance® 0-25m 26-50m 51-75m 76-100m 101-125m 126m+ 24 15 19 11 5 12 58 60 26 73 80 83 Group size 1 2 3 62 40 2 50 70 100 CONTINUOUS OJ �Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1. Continued. Variable n % Seen x2 (univariate) Univariate® significance (multivariate) x2 Stepwiseb significance 57.639 <0.001 57.639 <0.001 1.987 0.159 1.180 0.276 Continuous % vegetative cover 5-15 26 20-35 29 40-50 22 55-70 23 75-80 4 100 79 36 17 0 % snow cover 0-20 25-40 45-60 65-80 85-100 100 100 67 44 58 3 1 6 9 85 “Univariate results from x2 contingency analysis of discrete independent variables and logistic regression analysis of continuous independent variables. bFinal significance of independent variables after stepwise logistic regression analysis with only percent vegetation cover included in the model (P < 0.05). CSRA = Snake River Area, GRA = Greys River Area, HFA = Hams Fork Area. dNumber represents each primary observer present during moose sightability trials. “Analysis of perpendicular distance based on 86 observations due to 18 missing data points during sightability trials (all other variables are based on 104 observations). u> ( ji �36 Four moose were observed from the ground for a total of 87 minutes while recording characteristics of their radio signals. These moose were active for 59 minutes and bedded for 28 minutes. Pulse rates varied from constant slow, variable, and constant fast while active. Standing moose exhibited a constant slow pulse rate for no longer than 3.23 minutes, while feeding moose generally exhibited pulse rates that switched from slow to fast over relatively short time periods (<1 m i n ) . Bedded moose, however, exhibited only a constant slow pulse rate while observed. Signal characteristics were recorded for 35 of the 43 moose missed during surveys. I assumed that the activity observed after the survey was the same as during the survey for the 8 occasions that signals were not recorded. Recorded signal characteristics suggested that on 27 occasions a moose's activity (bedded or active) during the survey was not different than its activity when observed directly after the survey. On 8 occasions moose activity appeared to differ, however, and I reclassified activity from active to bedded for 6 observations and from bedded to active for 2 observations based on signal characteristics when the moose was initially passed during the survey. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �I reduced the number of topography and cover type categories to improve their fit to the dependant variable observed or missed moose. Improvement in the likelihood ratio x2 f°r topography (from x2 = 5.967, df = 2, P = 0.051, to x2 = 5.533, df = 1, P = 0.019) indicated that moderate and steep terrain affected moose sightability similarly and only flat and moderate/steep categories were needed. The exact scores x2 was slightly improved (from X2 = 35.458, df = 5, P < 0.001 to x2 = 37.331, df = 3, P < 0.001) when cover type categories were combined to deciduous shrub, open/water, conifer, and deciduous timber. Percent vegetative cover was significantly greater (t = 2.295, df = 1, P = 0.024) for bull groups (x = 42%, SD = 23%) than cow and cow/calf groups (x = 33%, SD = 19%) surveyed during sightability trials. Sightability Analysis. Univariate analyses of the independent variables indicated that topography, cover type, study area, primary observer, group size, and percent vegetative cover significantly influenced sightability (P < 0.019; Table 1). regression analysis, however, Multivariate logistic indicated percent vegetative cover was the only important predictor of moose sightability (P < 0.001) from helicopters. None of the other variables strongly influenced sightability once percent vegetative cover was entered during stepwise Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �analysis (P > 0.067; Table 1), suggesting that topography, cover type, observer differences, study area differences, and group size were correlated with percent vegetative cover. The influence of topography, cover type, study area differences, and group size did not become important when each was forced into the model containing percent vegetative cover (P > 0.086). I was unable to examine the influence of observer differences on moose sightability, however, when included in the model containing percent vegetative cover. The maximum likelihood estimate could not be calculated because of the low sample sizes for some observers making it impossible to estimate Wald or likelihood ratio statistics. I was also unable to compute the exact scores statistic for observers. I therefore assumed that the stepwise significance of observers with only percent vegetative cover in the model indicated that observer differences did not have an important influence on moose sightability (P = 0.067; Table 1) . I also examined non-linear transformations of group size, percent vegetative cover, percent snow cover, and perpendicular distance to determine if these relationships with sightability were stronger than their linear forms but their contribution was less important in all cases. The significance of the natural log (P = 0.084) and square Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �39 root (P = 0.083) transformations of group size, however, were similar to the first-order term (P = 0.082). All 2-way interactions of independent variables were tested for significance. The interaction between group size and topography was significant (P = 0.038) when included in the model with percent vegetative cover. All other interactions were not significant, including the 3way interaction between percent vegetative cover, topography, and group size (P = 0.718). Sightability Model Selection. My objective during model selection was to identify the model that fit best while maintaining adequate predictability and reasonable precision. I felt it was necessary to group percent vegetative cover into broader categories to clarify the relationships between percent vegetative cover and the group size*topography interaction. Vegetative cover classes (VCC) of 17.5% provided the lowest SE relative to the estimated coefficient (CV = 0.182) of categories compared (from 10-25%) and should maintain estimator efficiency and, by grouping, reduce the potential bias from field measurement errors of percent vegetative cover. Percent vegetative cover is estimated in 5% categories during actual surveys. Previous analyses suggested that percent vegetative cover and the interaction of group size*topography Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �strongly influenced moose sightability. Examination of the model containing VCC, group size, topography, and group size*topography (model A) suggested that the main effects (group size and topography) may not be important predictors (P > 0.058; Table 2) once VCC and the group size*topography interaction are included. I therefore selected the simpler model containing only VCC (P < 0.001) and the group size*topography interaction (P = 0.023; model B ) . Variance was reduced and predictability was only 2.9% less with this model (Table 2). To interpret predicted sightability from model B, I plotted the probability of detecting moose groups for the range of variables encountered during the field phase of model development (5 vegetative cover classes, group sizes from 1-3, and flat or broken topography; Fig. 4). The predictions from model B indicated that sightability should increase with increasing group size in flat topography, however, the model also predicted that sightability should decrease as group size increases in broken topography (Fig. 4). counter-intuitive. The latter prediction is My sample size (n = 104) may not have been large enough to adequately investigate this interaction because the data were spread over 30 cells (5 vegetative cover classes, 3 group sizes, and 2 topography categories = 3 0 possible outcomes). Most cell frequencies Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �41 Table 2. Statistical significance, parameter estimates, and percent of observations (n = 104) correctly classified for 3 moose sightability models developed from data collected in western Wyoming during winter, 1993. Model Model parameters Significance® Parameter estimates Percent correctly classified A 84.6 Intercept Vegetative cover classb Group size Topography Interaction0 0.005 3.60 <0.001 0.058 0.077 0.021 -1.86 1.34 -1.75 1.69 Intercept Vegetative cover class Interaction <0.001 5.26 <0.001 0.023 -1.81 0.47 Intercept Vegetative cover class <0.001 5.04 <0.001 -1.77 B 81.7 C 82.7 aSignificance levels from Wald x2 test where significant results (P < 0.05) indicate the coefficient does not equal 0. bl = 0-17%, 2 = 18-35%, 3 = 36-53%, 4 = 54-71%, 5 = 72-89%, 6 = 90-100%. cGroup size*topography interaction. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �42 0.9 -© 0.9 0.6 0.6 0.7 0.7 S 0.6 O £ 0.6 0.6 § 0.4 0.4 0.3 0.3 0.5 0.2 0.2 0.1 COVER CLASS 1 (0- 0.1 COVER CLASS 2 (18-35%) T 3 1 0.9 1 COVER CLASS 3 (36-53%) 0.9 0.8 0.8 0.7 0.7 COVER CLASS 4 (54-71°/ 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 1 0.9 3 GROUP SIZE COVER CLASS 5 (72-1 i°/ 'O 0.8 0.7 0.6 = BROKEN TOPOGRAPHY Q- 0.5 = FLAT TOPOGRAPHY 0.4 0.3 0.2 0.1 GROUP SIZE Fig. 4. Estimated detection probabilities (model B) for 5 vegetative cover classes containing 1-3 moose/group in flat or broken topography. Comparisons based on 104 moose groups surveyed during helicopter sightability trials in western Wyoming during winter, 1993. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �43 were <5 and 9 were 0 (groups of 3 were rarely encountered; Table 3). Predictions from model B appeared unduly influenced by relatively few observations. The third model I evaluated (model C) excluded the apparently spurious interaction term (group size*topography) and retained only the independent variable VCC (Table 2). This model was the most precise (i.e., lowest number of parameters), provided adequate predictability (82.7%), and was well represented by the data (i.e., 104 observations distributed from 5-80% vegetation cover; Table 1). Also, the relatively small change in the parameter estimate for VCC among the 3 models indicated that exclusion of the main effects and the interaction did not greatly influence the relationship of vegetation cover with moose sightability (Table 2). Therefore, model C was selected as the final model. Predictive Sightability Model. Model C was developed to predict aerial moose sightability from the parameter estimates in Table 2, where the linear regression portion of the model is: u = 5.044 - 1.772(VCC). Vegetative cover class takes values 1-6 for each increase of 17.5% vegetative cover. The estimated SE for the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �44 Table 3. Frequency at which 104 groups containing radiocollared moose were seen and missed during helicopter sightability trials in western Wyoming by vegetative cover class, group size, and topography, winter 1993. Group size® Vegetative cover classb Flat topography 1 2 Broken topography 3 1 2 1 Seen Missed 8 0 (100) 7 0 (100) 1 0 (100) 8 0 (100) 2 0 (100) 2 2 (50) 8 0 (100) 1 0 (100) 6 2 (75) 6 2 (75) 1 5 (17) 3 0 (100) 0 0 3 6 (33) 1 3 (25) 2 3 (40) 1 1 (50) 0 0 1 10 (9) 0 5 (0) 0 2 (0) 0 1 (0) 0 0 0 1 (0) 0 0 2 Seen Missed 3 Seen Missed 4 Seen Missed 5 Seen Missed “Values are n (% seen). bl = 0-17%, 2 = 18-35%, 3 = 36-53%, 4 = 54-71%, 5 = 72-89% (radiocollared moose were not encountered in > 80% vegetative cover). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �45 intercept and vegetative cover class coefficients were 0.919 and 0.322, respectively. The negative coefficient for vegetative cover class indicates that sightability decreases with increasing vegetation cover (Fig. 5). These data were appropriate for the logistic regression model based on Brown's goodness-of-fit test (P = 0.360). The Hosmer-Lemeshow test, which is based on the distance between observed and fitted values, also suggested that these data fit the logistic regression model (P = 0.492). Model Precision. Moose population estimates and variances (90% Cl) for the dense canopy (<70% vegetative cover) and open canopy (<50% vegetative cover) scenarios were 423 + 106 and 342 ± 62, respectively, when search intensity was 91% in high density strata and 50% in low density strata. Sampling variance, sightability variance, and model variance for the dense canopy scenario comprised 0.59, 0.35, and 0.06 of the total variance, respectively. For the open canopy scenario, sampling variance, sightability variance, and model variance were 0.77, 0.22, and 0.01 of the total variance, respectively. Moose population size and variance estimates were 449 + 129 and 352 + 74 for the dense and open canopy scenarios, respectively, when sampling intensity was reduced to 82% and 40% in the high and low density strata. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �46 0.9 £ 0.8 55 0.7 I 0.6 DC a- 0.5 Q 0.4 mI— 0.3 LJJ Q 0.2 0.1 V E GE TA TIV E C O V E R C L A S S Fig. 5. Estimated moose sightability (model C) by vegetative cover class (1 = 0-17%, 2 = 18-35%, 3 = 36-53%, 4 = 54-71%, 5 = 72-89%, 6 = 90-100%), based on 104 moose groups observed during helicopter sightability trials in western Wyoming during winter, 1993. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �The relative contribution of sampling variance, sightability variance, and model error was 0.68, 0.27, 0.05 and 0.83, 0.16, 0.01 for the dense and open canopy scenarios, respectively. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �DISCUSSION Model Selection. I developed 3 predictive models of moose sightability from helicopter surveys using 2 observers and a pilot. All models appeared, based primarily on the effect of vegetation cover, to provide adequate predictability. Two models, however, included an interaction between group size and topography. Inclusion of this interaction was based on a relatively small proportion of the data set (Table 3). Moose observed in VCC 1 and 5 did not influence this interaction since all groups were either seen or missed, respectively, in these cover classes. Group size in broken topography did not influence the sightability of moose occurring in VCC 2, but slightly fewer groups of 2 moose were seen during surveys than single animals in VCC 3 and 4. In flat topography, group size may have increased sightability in VCC 2 and 3 but did not in VCC 4. The importance of the group size*topography interaction in predicting moose sightability is highly questionable since the increase in sightability with group size in flat topography is based on the 7 single moose missed in VCC 2 and 3, and sightability was proportionately higher in broken topography for single moose because of 2 single animals seen in VCC 3 and 4. Observations of the influence of 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �group size on sightability indicate that larger groups may be more easily seen (Cook and Jacobson 1979, Samuel and Pollock 1981, Gassaway et al. 1985, Samuel et al. 1987, Ackerman 1988), supporting the model's prediction in flat topography while contradicting its prediction in broken topography. I am aware of no biological reason that this general trend should be reversed in broken topography. This data set may simply be too small to address this relationship. Model A provided the best classification of the model observations, however, this model contained 2 non­ significant parameters resulting in unnecessary model variance. Model B contained the significant interaction term group size*topography, however, it was the least predictive of the 3 models and was not biologically plausible. Model C was the simplest and most biologically reasonable of the 3 models, and therefore should provide the most precise and intuitive estimates. If the interaction of group size*topography is later found to influence moose sightability, model C would be biased. It appears that this potential bias would be slight, however, because the parameter estimate for vegetative cover class did not change markedly after removal of the main effects (model A) and the interaction (model B ) . Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Factors Influencing Moose Sightability. Based on univariate comparisons, it appeared cover type, topography, group size, study areas, and observers would likely influence sightability of moose. These factors, however, did not contribute to discrimination in the logistic model likely because they were correlated with the dominant variable, percent vegetative cover. et al. Samuel (1987) noted that studies estimating sightability based on univariate analyses will tend to overestimate the number of factors significantly influencing sightability. Gassaway et al. (1985) reported that habitat type, group size, and activity influenced the sightability of moose in Alaska. They suggested that group size and activity might be correlated but did not address the relationship of habitat characteristics with other factors. Habitat type and group size were correlated with percent vegetation cover during my evaluations in Wyoming but moose activity did not appear important. (1986) Crete et al. suggested that results of fixed-wing aircraft surveys may be more sensitive to moose behavior than results from helicopter surveys. If this is the case, the relative importance of activity between Gassaway et al.'s (1985) study and mine may have been a product of aircraft type used. Gassaway et al. (1985) used fixed-wing air craft whereas I used helicopters for surveys. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �I assumed that the activity of moose missed during surveys corresponded with their observed activity after the survey (n = 8) or the pulse rate of the i r .recorded radio signals (n = 35; slow pulse rate = bedded and variable or fast pulse rate = active). Activity of the moose immediately after the survey in which they were not seen should not have changed in most cases because moose were typically located 15-20 minutes after they were missed. Although, pulse rates corresponded well with bedded and active moose observed, it was possible for an active moose to exhibit a constant slow pulse rate and be recorded as bedded when missed during surveys. potential bias should be conservative. This Misclassifying missed moose as bedded when they were active should increase the chance of finding an influence of activity on sightability, however, this influence was not present (Table 1). The topographic diversity of moose range in Wyoming is greater than that in much of the species range in North America. I attempted to control for this factor by flying contour intervals in broken topography to maintain a relatively constant distance from observers to the ground. Failure of topography to enter the model is not surprising since percent vegetative cover and topography are likely related in Wyoming. Dense conifer typically dominates Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �52 upland slopes. The Hams Fork Area, for example, was the least topographically diverse and had little conifer cover whereas the Greys River Area was very rugged and dominated by conifers. Previous studies evaluated the influence of observer differences during aerial surveys over a wide range of observer experience levels and found important differences (LeResche and Rausch 1974, Caughley et al. 1976). In contrast, Ackerman (1988) found that sightability did not differ when observers were experienced, and Samuel et al. (1987) reported that differences between experienced observers were correlated with vegetation cover and group size. Similarly, I found that observer differences (all experienced) did not appear to influence moose sightability after percent vegetation cover was entered in the model. Fifty-two percent of my observations were made while 2 of the 8 primary -observers were present, however, likely reducing the chance of finding an influence of observers on moose sightability (Table 1). My intent in the design of model development was to represent the variability among experienced observers expected during normal surveys rather than collecting statistically adeguate samples from a few observers to examine differences. The model error component of the variance estimates from this model should account for the variation Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �53 expected among experienced observers. Additional trials may be necessary, however, to adequately determine if differences among experienced observers significantly influence moose sightability. We saw 59% of the 104 groups containing radiocollared moose over a range of habitat types and vegetative cover densities. This detection rate is similar to rates reported by LeResche and Rausch (1974; 43-68%), Thompson (1979; 57%), and Rolley and Keith (1980; 64%) but lower than those reported by Crete et al. Peterson and Page (1993; 78%). (1986; 73%), and Although comparisons may be tenuous because of differences in aircraft type, number of observers, search intensity, and moose habitat use patterns among studies, the detection rate during my study does appear low considering the high level of search intensity (x = 4.7 min/km2, SD = 1.2 min/km2) and that 3 persons were present in the helicopter. I conducted surveys from February-April when moose are generally most difficult to detect due to shifts in habitat use patterns (Lynch 1975, Karns 1982, Crete et al. 1986, Gassaway et al. 1986). I observed declining moose numbers on riparian areas during early February, and 56% of radiocollared moose surveyed were in conifer habitat (Table 1). winter surveys were applied intentionally to best represent the range of variables expected during Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Late �54 application since sighting probabilities are estimated on a group by group basis. Also, more information is gained on moose sightability under difficult rather than simple conditions (e.g., moose in heavy conifer vs. open willow). My detection rate may be lower than would be experienced earlier in the winter when moose in western Wyoming tend to occupy more open habitats. Snow cover, although not included in this model, is generally believed to greatly influence moose sightability (LeResche and Rausch 1974, Novak 1981, Gassaway et al. 1986, Bisset and Rempel 1991). I observed a confounding influence between percent snow cover and percent vegetative cover in my data set as most moose encountered in areas of sparse snow cover were also in open habitat types where late winter snow melt had occurred. I believe absence of percent snow cover in the model, however, primarily reflects the limited range of snow cover measured during my study; 90% of moose observations were in areas >65% snow cover (Table 1). Samuel et al. (1987) developed their elk model on a limited range of snow cover and found percent snow cover not to be important in estimating aerial elk sightability. Later, however, when Leptich and Zager (1991) evaluated this elk sightability model during variable snow conditions they found percent snow cover an important predictor of elk sightability. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I �55 suspect that snow cover may impact moose sightability more than this model indicates. Until the influence of snow cover on moose sightability is understood, application of this model should be restricted to times when snow cover >65%. Survey design reduced my ability to investigate the relationship between distance and the probability of detecting moose. Transects were intentionally flown close to one another to minimize the potential influence of distance. Thus, while perpendicular distance did not appear to influence moose sightability during my study (Table 1), it should not be concluded that distance will not play a part in sightability of moose. While it is tempting to increase flight transect widths to reduce flight time, more work needs to be done to determine at what distance, if any, distance will begin to influence moose sightability. Distance measurements were obtained for 86 of the 104 sightability observations used in analysis. These missing data should not have inhibited my ability to detect an influence of distance on moose sightability because the proportion of moose seen for the full data set (59%) and the reduced data set (58%) were similar. Sixty-seven of the 86 distance measurements were obtained from differentially corrected 3-dimensional GPS positions, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �which should have provided reliable results since position accuracy is expected to be within 5 m (Trimble Navigation, Ltd., 1992:17-19). The remaining 19 distance measurements were from differentially corrected 2-dimensional GPS positions. If the elevations associated with 2- dimensional positions were incorrect, the corresponding UTM coordinates may have been inaccurate. Elevations were adjusted to minimize this potential bias. In some cases, however, I could not determine exact elevations from topographic maps because of variable topography. The true elevation may not have been important when both the transect path and moose location were 2 dimensional, however, since they were typically collected from similar altitudes. Although these positions may differ from their true coordinates, the relative distance between them should not have changed. Potential error was greatest when distance measurements were taken between 2dimensional and 3-dimensional positions. I attempted to reduce this potential error by combining distance measurements into 25-m categories. A high proportion of groups near the flight line were apparently missed during sightability surveys (Table 1). This suggests line transect surveys (Buckland et al. 1993) to estimate moose abundance in Wyoming would be biased because the primary assumption, that all animals on the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �line are seen, is likely violated. Line transect sampling may provide valid population estimates, however, if correction factors for animals missed on the line are developed. The logistic regression estimator may provide a means of correcting for animals missed on the line since correction factors are applied to each group. If confidence intervals are wide from surveys with narrow transect spacing, however, combining these methodologies and applying wider transect spacing may not be beneficial since estimating an additional correction factor (i.e., distance) would result in an additional source of variation and increase confidence interval widths. Although sex/age groupings did not influence aerial moose sightability, bull groups did selected higher vegetative cover densities than cow and cow/calf groups. If homogeneous sightability is assumed across sex/age groups, bull numbers will be underestimated during composition surveys under these conditions. This sightability model accounts for different habitat selection by sex/age groups and should provide reliable sex and age composition data for moose, however. Comparison to Other Sightability Model Research. The degree to which percent vegetation cover drives this sightability model is not surprising given the importance of habitat factors identified in other moose sightability Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �studies (LeResche and Rausch 1974, Thompson 1979, Gassaway et al. 1985, Gassaway et al. 1986, Becker and Reed 1990, Bisset and Rempel 1991). Previous moose studies that have addressed detection, however, have not quantified this relationship, making comparisons difficult. Sightability research addressing vegetative cover has been conducted for elk (Samuel et al. 1987, Otten et al. 1993) and mule deer (Ackerman 1988) . al. Samuel et al. (1987) and Otten et (1993) reported that percent vegetation cover had a strong influence on elk sightability during logistic regression model development in Idaho and Michigan, respectively. Samuel et al. (1987) also found a strong relationship with group size and sightability which is not surprising based on the propensity of elk to occur in groups. Ackerman (1988), however, reported that mule deer sightability was influenced by cover type, group size, and animal activity level, and not percent vegetative cover during logistic regression model development in Idaho. He also reported that the lack of influence of vegetative cover may have been due to the manner in which cover was measured (vertically rather than obliquely through the canopy). Because moose encountered during my surveys were consistently seen and missed at oblique angles, and this will likely be the case during surveys, percent vegetative cover should be measured obliquely. Ackerman Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �59 (1988) also reported that mule deer sightability was most affected by animal activity level and that this was largely due to the response of mule deer to the approaching helicopter. Moose rarely reacted to the helicopter during this study and usually only when they were circled during classification. Model Precision. The population estimate for moose sampled from the dense canopy scenario created from my data set was 2.3 times the number of moose encountered (N = 423, n = 181); 90% Cl represented 25% of the estimate. The corrected population estimate for moose sampled from the population in more open habitat was only 1.9 times the number seen (N = 342, n = 181); 90% Cl represented 18% of the estimate. Estimator precision was greatly reduced when moose occurred in dense cover types and were less visible. Population estimates were similar when sampling intensity was reduced, however, the relative variance increased to 29% and 21% of the estimate for moose in dense and open canopy habitats, respectively. This suggests that sampling intensity will have a greater influence on the variability of moose population estimates when surveys are conducted in areas with more dense canopy cover. The relative contribution of each of the 3 variance components to the overall variance of the population Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �60 estimate also changed considerably when moose were sampled in areas of different vegetative cover densities. Sampling error was the dominant contributor to the variance estimates regardless of average overstory canopy density. However, when maximum cover increased to 70%, sightability error increased from 22% to 35%, model error increased from 1% to 6%, and sampling error was reduced from 77% to 59% of the total variance. Trends in the contribution of the variance components were similar when sampling effort was reduced except that the relative influence of sampling variance was greater for both scenarios. While it is possible to reduce or even eliminate sampling error by increasing sampling effort, sightability error can only be reduced by sampling moose when they are predominantly using open habitat types and thus, correction factors are low. Survey conditions encountered during model development likely represent a worst case scenario since surveys were conducted during late winter when moose predominantly used closed canopy vegetation types and sighting conditions were difficult. Application of this model during early winter when moose were more often found in open habitats should result in more precise population estimates. If this model is applied to moose using areas with greater vegetative cover densities than we Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �61 encountered (>70%), estimates should be expected to be less precise and perhaps have limited usefulness for management. A carefully thought out sampling design, which includes timing surveys when moose are using open habitat, is important to achieve the most precise population estimates. Management Summary. After percent vegetative cover was entered in the model, there were no important effects from the 3 study areas though habitat types and topography varied substantially. Therefore, this model should be suitable throughout Wyoming moose range and possibly over much of the species' range. Sampling protocols established during model development should be rigorously followed during application. Primary observers should be experienced in aerial moose surveys, and survey time should be limited to 4 hours/day for each observer to maintain a high search intensity. Estimating percent vegetative cover will require some training; observers, however, tend to derive similar estimates after only 1 or 2 days of estimating cover. Surveys should be flown using the types of helicopters used to develop this model, preferably Hiller-Soloy or 12E helicopters since 94% of the observations were collected with them. Snow cover should be >65%, and transect widths should be maintained at 150-250 m. Following fresh tracks during surveys can Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �62 result in incomplete or excessive coverage of a search unit. Therefore, only 1 or 2 short passes should be made when tracks are seen. As moose shift from more dense to more open canopy habitats, confidence in estimating moose population parameters increases, resulting in narrower confidence intervals. Consequently, surveys should be done when moose are most likely to occupy open habitats (e.g., early winter) thus providing the most precise estimates. This model should be applied cautiously if characteristics of the area (e.g., vegetation cover >70%) or behaviors of moose (e.g., larger groups, flight response to helicopters) appear to differ substantially from the range of these variables under which the model was developed (Table 1). Sightability observations from radiocollared moose can be continually added to this model to account for these differences or other deficiencies should they exist. With the addition of observations to the model it should become more robust to a variety survey conditions. Applying this model to a moose population of known size could help identify potential short-comings. Moose population data obtained from application of this sightability model with the sampling protocols applied during model development should prove more reliable than trend data. Although trend data are not Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �63 accompanied by estimates of variance, variability associated with trend data is at least as great and likely greater than variance associated with sightability survey results obtained using this model. Obtaining moose population data with estimates of variance will allow stronger inferences on year to year population changes. Moose sightability model estimates should be more robust to variable survey conditions than trend counts and population composition data should be more reliable because differential habitat selection by sex and age groups is accounted for. Applying a random sampling strategy (e.g., stratified random sample of search units) will ultimately improve information on moose distribution as well. Sightability survey costs, however, will likely be greater than those of traditional trend surveys because areas must be sampled intensively. Any increase in cost, however, should be offset by the enhanced value of having more defendable moose management programs. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �64 LITERATURE CITED Ackerman, B. R. 1988. Visibility bias of mule deer aerial census procedures in southeast Idaho. Ph.D. Diss. Univ. Idaho, 105 pp. Bartman, R. M . , G. C. White, L. H. Carpenter, and R. A. Garrott. 1987. Aerial mark-recapture estimates of confined mule deer in pinyon-juniper woodland. J. Wildl. Manage. 51:41-46. Becker, E. F . , and D. J. Reed. moose population estimator. Bisset, A. R . , and R. S. Rempel. 1990. A modification of a Alces 26:73-79. 1991. Linear analysis of factors affecting the accuracy of moose aerial inventories. Alces 27:127-139. Buckland, S. T . , D. R. Anderson, K. P. Burnham, and J. L. Laake. 1993. Distance sampling: estimating abundance of biological populations. London. Chapman & Hall, 446pp. Burnham, K. P., and D. R. Anderson. 1984. distance data in transect counts* The need for J. Wildl. Manage. 48:1248-1254. Caughley, G. 1974. Bias in aerial survey. J. Wildl. Manage. 38:921-933. _____ . 1977. Sampling in aerial survey. J. Wildl. Manage. 41:605-615. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �65 _____ , R. Sinclair, and D. Scott-Kemmis. Experiments in aerial survey. 1976. J. Wildl. Manage. 40:290-300. Cook, R . , and J. 0. Jacobson. 1979. A design for estimating visibility bias in aerial surveys. Biometrics 35:735-742. Crete, M . , L. Rivest, H. Jolicoeur, J. Brassard, and F. Messier. 1986. Predicting and correcting helicopter counts of moose with observations made from fixedwing aircraft in southern Quebec. J. Applied Ecol. 23:751-761. Dixon, W. J . , M. B. Brown, L. Engleman, M. A. Hill, and R. I. manual. Jennrich. 1988. BMDP statistical software Univ. Press of Ca., Berkley, Ca. Gassaway, W. C., and S. D. Dubois. moose population parameters. (Suppl.) 1987. 1234pp. Estimating Swed. Wildl. Res. 1:603-617. _____ , _____ , and S. J. Harbo. 1985. Biases in aerial transect surveys for moose during May and June. J. Wildl. Manage. 49:777-784. _____ , _____ , D. J. Reed , and S. J. Harbo. 1986. Estimating moose population parameters from aerial surveys. Biol. Pap. No. 22, Univ. Alaska, Fairbanks. 108pp. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Hosmer, D. W . , and S. Lemeshow. regression. 1989. Applied logistic John Wiley & Sons, Inc., New York, N.Y. 307pp. Johnson, B. K., F. G. Lindzey, and R. J. Guenzel. 1991. Use of aerial line transect surveys to estimate pronghorn populations in Wyoming. Wildl. Soc. Bull. 19:315-321. Karns, P. D. 1982. Twenty-plus years of aerial moose census in Minnesota. Alces 18:186-196. Kleinbaum, D. G . , L. L. Kupper, and K. E. Keith. 1988. Applied regression analysis and other multivariable methods. PWS-Kent, Boston, Mass. Leptich, D. J, and P. Zager. 1991. 718 pp. Validation of the Idaho elk sightability model at the Starkey Experimental Forest, Oregon. A Proposal. Idaho Dep. of Fish and Game. Boise. LeResche, R. E . , and R. A. Rausch. 1974. precision of aerial moose censusing. Accuracy and J. Wildl. Manage. 38:175-182. Lockwood, R. 1991. Lincoln moose study. Wyoming Game and Fish Dept. 15pp. Lynch, G. M. Alberta. 1975. Best timing of moose surveys in Alces 11:141-153. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Mehta, C . , and N. Patel. 1993. LogXact-Turbo: logistic regression software featuring exact methods, user manual. Cytel Software Corp. Cambridge, MA 284pp. Neal, A. K . , G. C. White, R. B. Gill, D. F. Reed, and J. H. Olterman. 1993. Evaluation of mark-resight model assumptions for estimating mountain sheep numbers. Novak, M. J. Wildl. Manage. 57:436-450. 1981. The value of aerial inventories in managing moose populations. Alces 17:282-315. Otis, D. L . , K. P. Burnham, G. C. White, and D. R. Anderson. 1978. Statistical inference from capture data on closed animal populations. 62. Wildl. Monogr. 135pp. Otten, R. M . , J. B. Haufler, S. R. Winterstien, and L. C. Bender. Michigan. 1993. An aerial census procedure for elk in Wildl. Soc. Bull. 21:73-80. Peterson, R. O., and R. E. Page. 1993. Detection of moose in midwinter from fixed-wing aircraft over dense forest cover. Wildl. Soc. Bull. 21:80-86. Pollock, K. H . , and W. L. Kendall. 1987. Visibility bias in aerial surveys: a review of estimation procedures. J. Wildl. Manage. 51:502-510. Rolley, R. E . , and L. B. Keith. 1980. Moose population dynamics and winter habitat use at Rochester, Alberta, 1965-1979. Can. Field-Nat. 94:9-18. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Rudd, L. T., and L. L. Irwin. 1985. Wintering moose vs. oil/gas activity in western Wyoming. Alces 21:279- 298. Samuel, M. D . , and K. H. Pollock. 1981. Correction of visibility bias in aerial surveys where animals occur in groups. J. Wildl. Manage. 45:993-1000. _____ , E. 0. Garton, M. W. Schlegel, and R. G. Carson. 1987. Visibility bias during aerial surveys of elk in northcentral Idaho. J. Wildl. Manage. 51:622-630. _____ , R. K. Steinhorst, E. 0. Garton, and J. W. Unsworth. 1992. Estimation of wildlife population ratios incorporating survey design and visibility bias. J. Wildl. Manage. 56:718-725. SAS Institute Inc. 1988. SAS language guide for personal computers, release 6.03 ed. N.C. _____ . ed. SAS Inst. Inc., Cary, 558pp. 1990. SAS technical report P-200, release 6.04 SAS Inst. Inc., Cary, N.C. Steinhorst, R. K . , and M. D. Samuel. 235pp. 1989. Sightability adjustment methods for aerial surveys of wildlife populations. Thompson. I. D. Biometrics 45:415-425. 1979. A method of correcting population and sex and age estimates from aerial transects surveys for moose. Alces 15:148-168. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �Timmermann, H. R. review. 1974. Moose inventory methods, a Nat. Can. 101:615-629. Trimble Naviagation, Ltd. 1992. General reference guide for the GPS Pathfinder™ system. Sunnyvale, Cal. 148pp. Unsworth, J. W . , L. Kuck, and E. 0. Garton. 1990. Elk sightability model validation at the National Bison Range, Montana. Wildl. Soc. Bull. 18:113-115. _____ , F. A. Leban, G. A. Sargeant, E. 0. Garton, and J. R. Pope. 1991. Aerial survey 3.1: user's manual. Idaho Dept, of Fish and Game. 46 pp. White, G. C., R. M. Bartman, L. H. Carpenter, and R. A. Garrott. 1989. Evaluation of aerial line transects for estimating mule deer densities. J. Wildl. Manage. 53:625-635. Zar, J. H. 1984. Biostatistical analysis. Hall, Inc., Englewood Cliffs, N.J. Prentice- 718pp. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. �70 APPENDIX A. Aerial survey data form used during helicopter sightability trials of moose in western Wyoming during winter, 1993. Date Page of Lat. Long. Location:_ o ______ o Pilot_ _ _ _ _ _ 1 observer_ _ _ _ _ _ 2 observer_ _ _ _ _ _ (m ark observer that 1 st spots m oose) Search Time: start group # Stop_ _ _ freq / # of Bulls # o f Cows Unk. group % cover Perp. Tim e S een % Activity Light Terrain during/ collar adults adults Calves size cover yrlg. sex/age type Obs. y ig Dist. S n ow # after? 1 2 3 4 5 ACTIVITY: 1 = B EDD ED , 2 = S T A N D IN G , 3 = M 0 V IN G LIGHT: 0 = FLAT, 1 = BRIGHT TERRAIN: 1 = F L A T ( < 20° slope), 2 = M 0 D E R A T E (20° to 50° slope), 3 = S T E E P ( > 50°) C O V ER TYPE: 1 = W IL L O W RIPARIAN, 2= S A G E B R U S H SHRUBLANDS, 3 = O P E N FORB/G RASS M E D O W S, 4 = M T N . SHRUBLANDS, 5 = C O N IF E R , 6 = A S P E N , 7=A S P E N /C O N IF E R MIX, 8 = C O T T O N W O O D , 9 = B U R N E D FOREST, 10 = C L E A R CUT, 1 1 = O T H E R (specify). CO M M EN TS: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. � Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Dissertations/Theses Type The nature or genre of the resource Text Text A resource consisting primarily of words for reading. Examples include books, letters, dissertations, poems, newspapers, articles, archives of mailing lists. Note that facsimiles or images of texts are still of the genre Text. Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource A sightability model for moose developed from helicopter surveys in western Wyoming Rights Information about rights held in and over the resource <a href="http://rightsstatements.org/vocab/InC-NC/1.0/" target="_blank" rel="noreferrer noopener">In Copyright - Non-Commercial Use Permitted</a> Date Created Date of creation of the resource. 1994 Subject The topic of the resource Moose Mammal populations Wyoming Aeronautics in wildlife management Description An account of the resource Aerial surveys are the only practical way to estimate ungulate numbers in most of North America (LeResche and Rausch 1974, Timmerman 1974, Gassaway and Dubois 1987). These surveys, however, often provide biased estimates and only under specific conditions do they allow detection of even large population changes (Caughley 1974, Gassaway et al. 1985). Ideally, aerial survey estimators should be accurate, precise, cost effective (Gassaway et al. 1986), and repeatable to provide timely management decisions. Type The nature or genre of the resource Text Creator An entity primarily responsible for making the resource Anderson Jr, Charles R. Language A language of the resource English Format The file format, physical medium, or dimensions of the resource application/pdf Extent The size or duration of the resource. 78 pages Bibliographic Citation A bibliographic reference for the resource. Recommended practice is to include sufficient bibliographic detail to identify the resource as unambiguously as possible. Anderson, C. R. Jr. 1994. A sightability model for moose developed from helicopter surveys in western Wyoming. Thesis, University of Wyoming, Laramie, USA. Publisher An entity responsible for making the resource available University of Wyoming