Understanding Seasonal Variation in Detection of Martens Using Radio-Marked Individuals William J. Zielinski1, Katie M. Moriarty2, Thomas A. Kirk1 and Keith M. Slauson1 1 2 USDA Pacific Southwest Research Station, Arcata, California 95521 Oregon State University, Department of Fisheries and Wildlife, Corvallis, OR 97331 Final Report to the Lassen National Forest 7 September 2011 SUMMARY We sought to understand the effects of season of detection on the development of a habitat model for the American marten on the Lassen National Forest (Lassen NF). Monitoring the effects of land management practices requires a habitat model, and resulting map, that is an unbiased representation of important habitat areas. However, previous research on the Lassen NF revealed differences in the locations where martens are detected in summer and winter vs. primarily in winter, leading to marked differences in the habitat maps that are developed from these location data. The demographic structure (age, sex ratio) and residential status of martens differs by season, which may explain some of this difference, especially if locations where detections occur primarily in the winter include animals that are predominately young and dispersing. Habitat models developed using the “winter only” locations may not represent the most important habitat of breeding age animals, in particular females. We radio-marked 12 individual martens captured in areas where previous surveys have detected them primarily in the winter (Swain Mountain [Swain] and Humboldt Peak [Humboldt]) and monitored their fates. In doing so we tested the working hypothesis that areas where martens are detected primarily in the winter include locations without year-round occupancy and detections in these locations occur in the winter only because young, dispersing animals spend a short period of time assessing the general habitat quality and, finding it lacking, either move on, die, or persist for unknown periods of time but do not breed. These animals are much less likely to contribute to population persistence and, thus, the habitat choices they make during their movements may be unrepresentative of the resident breeding segment of the population. Ten of the 12 animals were male and 50% of the individuals that could be aged were < 1 year old. Monthly survival rates declined most during the period from July to September. There was no evidence, from either telemetry or seasonal camera surveys, for pronounced changes in the size of the area used by individual martens in winter versus summer. Camera surveys verified the same phenomenon reported in previous surveys: that occupancy rates – adjusted for detectability – were significantly higher in winter than in summer. As of July 2011 (14 - 18 months after their initial captures) only 3 of 12 animals were alive, but each was located within the boundaries of their original survey grid. The majority of the mortalities were of animals < 1 year of age. Martens in our study areas exhibited marked seasonal variation in occupancy and were primarily males and young animals that exhibited increased mortality during the summer. Importantly, our sample included only 1 breeding-age adult female. Collectively, these characteristics are indicative of relatively poor habitats, where breeding is rare, leading to the conclusion that they are likely “sink” habitats, where death and immigration rates exceed birth and emigration rates. We conclude that our working hypothesis was supported by our data on age and sex structure and mortality. Habitat identified on the basis of surveys conducted during the summer is the most likely to identify areas that are important to the marten population, not only because it is less extensive than habitats occupied during the non-breeding season, but also because it is known to support adults during the parturition and breeding season. Including in the development of habitat models survey locations where martens are primarily detected only in winter (the nonbreeding season) may confound the important relationship among the habitat predictors associated with year-round residence. This is because many of these locations will be sites where juveniles or males occur during the non-breeding season only. However, some of these detections will also be females and we don‟t know yet if their ranges increase in winter. Thus, new work should focus on the spatial dynamics of breeding-age males and females and the potential seasonal variation in home range size and habitat use. Ideally, our ultimate goal should be to distinguish two classes of suitable habitat: “suitable for reproduction” and “suitable for non-reproductive activities”. Future research will attempt to achieve this standard. INTRODUCTION The American marten (Martes americana), a member of the weasel family (Mustelidae), is associated with structurally complex, high elevation late-seral coniferous forests (Spencer et al. 1983, Buskirk and Powell 1994, Payer and Harrison 2003, Zielinski et al. 2005). Martens are considered sensitive to habitat fragmentation (e.g., Bissonette et al. 1997), but they can tolerate some forms of forest management and other disturbances. Lassen National Forest (NF) has a rich history of research and monitoring for sensitive species, including martens. Marten habitat use patterns were investigated at Swain Experimental Forest and vicinity (Ellis 1998), landscape habitat suitability was modeled throughout the Greater Southern Cascade area (Kirk 2007, Kirk and Zielinski 2009) and connectivity analyses using Least Cost Path modeling (LCP) have been undertaken (Kirk and Zielinski 2010). Marten distribution has also been described for the nearby Lassen Volcanic National Park (Perrine 2005). We have also conducted systematic and rigorous camera surveys that demonstrate significant seasonal effects on occupancy (Zielinski et al. 2009). 2 The Lassen NF is part of a greater land management area defined within the HergerFeinstein Quincy Library Group (HFQLG) Forest Recovery Act. This pilot project encompasses over 10,000 km2 within 3 national forests in northern California: the Lassen, Plumas, and Sierraville Ranger District of the Tahoe National Forest (Davis and King 2001). The HFQLG Forest Recovery Act supports collaborative projects which aim to balance timber management and local job stability with long-term conservation of natural resources, including wildlife sustainability and watershed restoration. Essentially, this plan outlined a 5-year (now 15-year) design to intensively monitor the effects of strategic timber extraction, largely focused on fuel reduction for fire barriers, ecosystem processes, and wildlife species (HFQLG-Revision 2008). Given the HFQLG‟s goals, and the previous work on martens in the area, Lassen NF is an ideal location to further assess the effects of management activities on martens. Monitoring the effect of HFQLG projects on martens requires a habitat model that faithfully represents important habitat areas. However, martens are detected at different locations during different times of the year, particularly when the results of summer surveys are compared with those conducted during winter (Kirk and Zielinski 2009, Zielinski et al. 2009). These seasons probably correspond to times of the year when population sizes differ (e.g. significantly larger in the fall/winter due to the presence of young-of-the-year), food is differentially available (e.g., more abundant in summer), and whether reproduction is occurring (summer) versus dispersal (winter). Seasonal differences in abundance or behavior have implications for the results of landscape habitat suitability modeling because most models for small carnivores are constructed using detection / non-detection data from track plate or camera surveys (Carroll et al. in press). However, empirical landscape habitat suitability models created using locations where martens are detected primarily in the winter identify very different habitat 3 areas than similar models created using survey data from summer or from surveys where martens were detected in summer and winter (Rustigian-Romsos and Spencer 2010). Typically, adult breeding martens have strong fidelity to their home range for their lifetime and do not engage in significant seasonal changes in home range (Phillips et al. 1998, Gosse et al. 2005, Hearn 2007). Thus, habitat associations created from summer survey data represent habitats chosen as home ranges by adults for the purposes of reproduction. Some of the animals detected in winter are undoubtedly dispersing juveniles, which are much less likely to be selective about the areas they traverse when searching for their ultimate home range. This may result in detections that occur in a wide variety of habitats, some of which happen to be traversed during the course of searching for suitable sites. Conversely, the greater variety of conditions where martens are detected in winter may instead occur because all martens are simply more detectable at baited stations during the winter when, compared to summer, food is more limiting and martens make extensive forays as they are attracted to carrion (including baits used at detection stations). We also don‟t know whether martens in the southern Cascades increase their home ranges in winter, which may account for some of the increase in areas where martens are detected in winter. Studies elsewhere on martens are inconsistent in their findings on seasonal change in the size of home ranges (e.g., Phillips et al. 1998, Fuller and Harrison 2005) but Ellis (1998), who conducted work in our study area, did not reveal substantial seasonal changes in home range size in either males or females. Male martens, however, make prebreeding and breeding-season forays to new areas (K. Slauson, pers. obs., K. Moriarty, pers. obs). The development of a credible habitat model is critical to achieving the goals monitoring questions that are mandated to be answered by HFQLG. One question, in particular, asks for a 4 consistent definition of suitable habitat for martens; a goal that was highlighted by the Pinchot Institute‟s recommendations (Anonymous 2008). Achieving this goal will not be possible until we can understand the modeling implications of using locations from different times of the year, when they represent different activities and perhaps different cohorts of individuals. Our previous work on the Lassen NF has demonstrated that martens have significantly lower occupancy rates during the summer (Zielinski et al. 2009), despite the fact that they are also statistically less detectable at camera stations during the summer (most likely due to the removal of bait by black bears [Ursus americana]). Our central working hypothesis, based on preliminary data, is that the areas where martens are detected primarily in the winter include relatively poor habitat that is only occupied in the winter because young, dispersing animals spend a short period of time assessing the habitat quality and, finding it lacking, move on or die. If this is true, including these observations in the development of an empirical habitat model may be misleading. Thus, understanding the reasons for seasonal differences in occupancy and – most importantly – determining how the season of data collection affects the habitat model that is created, is an important task. Intensive monitoring of the fate of individual martens, via radio-telemetry, is necessary to understand the cause of the seasonal change in occupancy pattern. To achieve this objective, we radio-marked martens in locations on the Lassen NF where seasonal variation in detection was greatest, the Humboldt Peak and Swain Mountain areas (Zielinski et al. 2009). These locations also are predicted to have relatively poor habitat (Kirk and Zielinski 2009, Rustigian-Romsos and Spencer 2010). We assume, for the purposes of this paper, that areas identified by these previous modeling efforts to have high probability of detection are areas with the best quality habitat and areas that have the lowest probability of detection area areas with the poorest quality 5 of habitat. By capturing and collaring animals at Humboldt Peak and Swain Mountain in the fall, and monitoring them through the following summer, we could determine whether they persisted in these regions throughout the winter and into the summer when they establish breeding home ranges, or whether they ultimately die there or seek more favorable habitat elsewhere. Our previous survey data were too coarse to answer these questions. We also deployed remotelytriggered digital cameras in winter and summer to determine how detectable martens were by season. The primary objective of this work is to understand the fate and characteristics of the American martens that appear to occur in specific regions on the Lassen NF primarily during the winter. This information will help us understand how annual variation in detection and occupancy can affect the development of habitat models, and other aspects of conservation planning for this species that are affected by changes during the annual life cycle of martens. METHODS Study Area Lassen NF is located in the mountains of northeastern California at the junction of the Southern Cascades and Sierra Nevada ranges. Our work focused on areas that were a subset of previous marten survey grids (Zielinski et al. 2009) at elevations ranging from 1500-2100 m (Fig. 1). Primary vegetation types include red fir (Abies magnifica), white fir (Abies concolor), lodgepole pine (Pinus contorta) and subalpine conifer forest types (Mayer and Laudenslayer 1988). Sierran mixed conifer, Douglas-fir (Pseudotsuga menziesii), montane hardwood, montane hardwood conifer, ponderosa pine (Pinus ponderosa), Jeffery pine (Pinus jeffreyi) and chaparral vegetation types are also present, but more widespread at lower elevations. Seasonal and perennial streams occur in both study areas, wet meadows are also common landscape features. The regional weather pattern is typical of California‟s Mediterranean climate with cool, 6 wet winters followed by summer growing seasons that are hot and dry. Precipitation ranges from 50 – 203 cm with most occurring as snow above 1800 m (Bailey et al. 1994). Marten Trapping and Processing We set traps within two of the three previously surveyed grids: Swain Mountain (Swain) and Humboldt Peak (Humboldt) (Fig. 2) (Zielinski et al. 2009). We did not trap, or do any new work, within the Mineral grid, located primarily within the Lassen Volcanic National Park (Fig. 1, 2). Initial trapping and radio-marking was focused within the perimeter of the two grids, each occupying approximately 15,000 ha. Each grid was comprised of 20 sites, about 3-km apart (Fig. 1, 2), which were previously used to conduct detection surveys via remote camera (Zielinski et al. 2009). We set traps at all previously surveyed stations within the Humboldt and Swain grids and at other locations within each grid where previous habitat modeling (Kirk and Zielinski 2009) predicted marten reproductive habitat. We used Tomahawk live-traps (model 108) that were modified to decrease potential injury and stress to the animals. A piece of masonite was fitted onto the bottom of each trap to discourage digging. Wooden „cubby boxes‟ were attached to the back of each live trap which provided trapped animals with a dark enclosed area for refuge (Wilbert 1992). It also contained pieces of fleece and conifer boughs for warmth. Martens were chemically anaesthetized using either ketamine-diazepam or ketamine-midazolam. The size of the animal was visually estimated and given an appropriate dose, based on previous marten research (Kreeger and Arnemo 2007). Processing included attachment of a VHF radio-collar, the collecting of morphometric data, and tooth removal. Each radio collar had a unique visual pattern consisting of three 1-cm reflective bands to aid in individual identification in remote-camera photographs (3M® 7 ScotchliteTM Very High Gain Reflective Sheeting 3000X). We collected Dacron swab samples at the nasal, ocular, and rectal regions for disease testing, we collected hair from the nape and tail for genetics or isotope analysis, we measured standard features (neck circumference, total length, tail length, fore and hind foot length and width), and we took ventral photographs. The vestigial first upper premolar was removed for cementum annuli ageing (Matson 1981, Bodkin et al. 1997). Our trapping and handling protocol was authorized by the Oregon State University Institutional Animal Care and Use Committee (ACUP permit 3944) and the California Department of Fish and Game (Scientific collecting permit 803099-01; Memorandum of Understanding). Locating Martens and Estimating Areas of Use We attempted to relocate each radio-marked individual once per week using radio-signal triangulation. We used Yagi three-element directional antennas and R-1000 telemetry receivers (Communications Specialists, Orange, CA) to obtain these locations. To maximize accuracy we attempted to triangulate from at least 3 locations, a minimum of 20° apart, in a period less than 20 min (Powell 2000, Millspaugh and Marzluff 2001). Approximately once per month we also attempted to locate each individual in a resting location. During these attempts the signal was approached on foot until, typically, it could be received without an antenna. This usually meant that we were within 20 m of the animal. Once in the rest site area, we attempted to narrow down the marten‟s location to the specific structure used (e.g. live tree, snag, log). All locations, resting sites, triangulations and capture locations were used to estimate 3 versions of the area used by each radio-marked individual: one that includes all locations, from the initial capture in the fall of 2009 to the conclusion of the study on 15 September 2010 (“total use area”), a second that reflects only those locations from the period beginning 2 weeks before 8 and extending to 2 weeks after the winter camera survey (“winter survey use area”), and a third that reflects only those locations from the period beginning 2 weeks before and extending to 2 weeks after the summer camera survey (“summer survey use area”). Each area was a 100% minimum convex polygon, created by joining the perimeter locations for each individual. When a perimeter location was derived from a triangulation, the outside margin of its error polygon (estimated using LOAS; Ecological Software Solutions) was used to denote the location. Note that we refer to these as “use areas” because we did not believe that we would be collecting sufficient information to denote an animal‟s home range. Estimating Survival Rates We estimated the monthly survival rate for martens that had been radio-collared using the Kaplan-Meier procedure, which allows for periodic entry of animals (“staggered entry”) into the analytical framework (Pollock et al. 1989). The survival function is the probability of an animal in a population surviving t units of time (in our case, t = months) from the beginning of the study. Each month, from September 2009 to September 2010 we accounted for the status of our sample animals by recording their entry into the marked population and their current status (alive, dead or censored). A “censored” individual is one whose status is unknown, which occurs most often when a radio-collared animal‟s collar has malfunctioned or an animal has emigrated from the study area. Camera Surveys Camera surveys were conducted to determine the likelihood of detecting radio-marked, and any other, martens within each grid. We conducted the first camera survey in the winter (12 January – 21 February 2010 in Swain and 1 March – 9 April 2010 in Humboldt) and the second camera survey in the summer (8 July – 5 August 2010 in Swain and 10 August – 9 September 9 2010 in Humboldt). Several cameras in each survey were run a few days beyond the end dates to compensate for malfunctioning cameras or those disturbed by black bears. We used standard camera methods (Kays and Slauson 2008) to replicate similar surveys that were conducted earlier, during summer 2007, winter 2007/08 and summer 2008 (Zielinski et al. 2009). This time, unlike the previous surveys, we were able to determine where martens occurred during the survey period and they were individually marked to be identifiable in photographs. We used both Cuddeback Excite (Non Typical Inc., Green Bay, WI) and ScoutGuard™ SG550 IR cameras (HCO Outdoor Products, Norcross, GA) and set them as close to each of the previously used survey points as possible. Two live trees spaced 2-4 m apart were chosen, with the camera on one tree facing north towards the second (bait) tree. We used as bait whole chickens (approximately 2.5 kg) that were staked to the tree at 1-2 m height. We also used a commercial scent lure (Gusto, Minnesota Trapline Products, Pennock, MN) which was placed on a nearby branch. Cameras were set on high sensitivity with a 1-min delay. We serviced cameras weekly to download images, check camera functions, and replace bait; 4 service occasions resulted in an approximately 4-week (28-day) sample period for each camera. If, however, there was evidence that a camera had not been functioning properly during the previous week (e.g., due to weather or black bear disturbance) the survey period for that camera was extended to attempt to achieve 28 effective survey days. To achieve an objective related to a separate project on the genetics of martens and the Sierra Nevada red fox (Vulpes vulpes necator), we placed a “hair snare” device (Kendall and McKelvey 2008, P. Figura, pers. comm.) at each station after it had received its first visit by either a marten or a red fox. The device used 5 gun-cleaning brushes as snares, each of which was attached about 5 cm apart along a 25cm strip of corrugated plastic that was tacked to the tree 10 below the bait. Estimating Probability of Detection and Occupancy Detection of a target species is never certain, but the probability of detection can be estimated by various methods, and can then be used to adjust occupancy and to interpret seasonal patterns. The probability of detecting a marten during a given sampling period varies with a number of factors and is represented by p, the probability for each sampling occasion (sometimes referred to as a“visit”), and P, the probability of detection over all sampling occasions in the sampling period (MacKenzie et al. 2006). We estimated probability of detection using the maximum likelihood method (MacKenzie et al. 2006). This approach uses the pattern of detections, represented by detection histories, to estimate the probability of detection assuming the target species is present. Maximum likelihood estimators are the estimated values for parameters given a set of empirical data. A detection history is such a data set and is the pattern of detections over the survey period. For example, a detection history of “100100” represents a survey period with 6 sample occasions where a marten was photographed during the 1st and the 4th occasion. We used program PRESENCE (version 2.4, Hines 2006) to estimate P using the multi-season analysis and the model parameterization that estimates an initial occupancy rate, extinction, colonization, and detection probability. “Extinction” (έ) and “colonization” (γ) are transition probabilities used in occupancy estimation (MacKenzie et al. 2006:206) that were adapted from the metapopulation literature (Levins 1969). Extinction is the probability that a sample site has changed from “detected” to “not detected” from one survey season to the next, and colonization is the opposite pattern. They are not estimable for the initial survey season because there is no data preceding this year. The camera data were aggregated by 24-hour 11 period (from 0000 to 2400 h) yielding a 28-day detection history for each camera station. Thus, the sampling occasion, for the purpose of estimating P was a 24-hour period. Previous analyses discovered that the presence of bait had a significant effect on probability of detection in our study area (Zielinski et al. 2009). As with previous surveys, the relatively long interval between checking and reprovisioning a station (1 week) made it likely that bait removed by a bear early in the interval would be absent during the remainder of the survey period. As such, we investigated the effect of the covariate “bait” (present or absent), as determined by photographs and field notes, on the probability of detection. Preliminary analysis included examining when detections occurred within the 28-day sampling period for each season, and how days added to the 28-day protocol may have affected detection probability. Based on this information, custom models were developed in PRESENCE to estimate how probability of detection changes over the course of the 28-day survey duration and by the presence/absence of bait. Patterns observed in the change in the proportion of detections over the survey duration were used to create several covariates that reflect variation in “time”, where estimates of p – the daily probabilities of detection – were modeled as being equal within each week (1 through 4), but different across weeks. Study area effects were modeled on both p and occupancy (ψ; see below) and used the variable “Swain” to determine if either probability differed significantly between the two study areas. Two “season” variables were created, representing winter and summer (season) or to differentiate winter (December – February), spring (March) and summer (June – August). Models were compared using Akaike‟s Information Criteria (AIC) (Akaike 1973) with lower AIC values indicating better fitting models. The AIC weight (w) was also calculated for each model, providing a measure of model fit relative to other candidate models. 12 Estimates of P generated in this fashion can also be used to adjust estimates of occupancy (ψ) (MacKenzie et al. 2006). Naïve estimates of ψ are simply the proportion of sample sites in each grid where a marten is detected (i.e., the observed proportion). If P is low, then the final – adjusted – estimate of ψ would be substantially higher than its naïve estimate. This occurs when a detection protocol is not very effective at detecting animals that are present. If, on the other hand, the estimate of P is high (i.e., most animals that are present are detected) then the adjusted estimate of ψ would be similar to the naïve estimate of ψ. Season-specific estimates of the probability of detection (P)were compared using paired t-tests and occupancy estimates (ψ ) were compared using McNemar‟s test to assess the difference between 2 proportions (McNemar 1947). RESULTS Martens and their Fates We trapped 49 sites (401 trap nights) at Humboldt and 58 sites (369 trap nights) at Swain (Fig. 2) during two trapping sessions: October – November 2009 and March – April 2010. We captured, processed and attached radio-collars to 12 martens, the majority of which were males (n = 10), 5 at Humboldt and 7 at Swain (Table 1). The majority of animals were < 2 years of age, with 50% of the animals for which age could be estimated, < 1 year of age (Table 1). By the time we had concluded monitoring the animals, on September 23, 2010 (5-10 months after they had been originally captured) 7 were known to be alive (58.3%), 3 were known to be dead (25%) and 2 were of unknown fate (Fig. 3). Monthly survival estimates were high from November through June (91.1%) but then decreased abruptly from July (70.9%) to September (49.6%) (Fig. 4). Two animals (M04, M11) were killed by a predator, most likely a raptor, and one died of unknown cause (M06). 13 Use Areas We collected an average (SD) of 35.6 (17.6) locations per individual during the entire sampling period, but 6.5 (3.2) and 9.4 (2.0) during the winter and the summer camera period, respectively. The majority (67%) of the total number of locations were triangulations, with accuracies that averaged 128 m (171.8 m). The balance of locations were from camera stations, incidental captures, aerial telemetry, or mortality locations. During the period that they were alive and being monitored, most of the radio-collared animals in the Humboldt area were located within the boundaries of the original camera station grid (Fig. 5). These individuals were not all available to be detected during both winter and summer (due to collar loss or death; see Fig. 3), but 4 of the 5 were on the grid and available during the winter camera survey. In contrast, a number of animals at the Swain area were found off the grid on multiple occasions, almost always to the west (Fig. 5). This behavior includes periods of time when camera stations were operating in summer (the red polygons; see especially M08 and M10) and when camera stations were operating in winter (the blue polygons, see especially M09) (Fig. 5). Total use areas ranged from 3.0 – 56.1 km2; males averaged 16.0 and females 0.9 km2. Adult male use areas averaged 9.3 and adult females 0.9 km2. Juvenile males had the 2 largest areas (M02, M09) but also two of the smallest areas (M07, M11; Table 1, Fig. 5). Size of area was not related to number of locations; one of the smallest total use areas had the greatest number of locations (i.e., F02) and one of the largest areas was a male with the fewest number of locations (i.e., M09) (Table 1). 14 The number of locations verified during summer and winter were relatively few, but casual inspection indicates that summer and winter use areas were not substantially different in size (Fig. 5). Camera Surveys Camera surveys in winter and summer produced results that are consistent with the seasonal pattern demonstrated for Humboldt and for Swain during previous years. A relatively high proportion of cameras had marten detections in the winter and a relatively low proportion had detections in the summer (Fig. 6). The seasonal cycle is dampened, somewhat, in 2009 and 2010, but the pattern is still present with the seasonal change in occupancy especially obvious at Humboldt (Fig. 6). At Humboldt in winter, martens were detected at 11 of 20 camera stations (naïve occupancy rate = 55.0%). Radio- marked martens (M01, M03, M04) were detected at 3 different stations and unmarked martens at 10 different stations (Fig. 5, 7). At the Swain grid in winter martens were detected at 4 of 20 camera stations (naïve occupancy rate = 20%); marked martens (M06, M07 and M09) at 2 stations (Fig. 5) and unmarked martens at 3 stations (Fig. 7). At Humboldt in summer, martens were detected at 2 camera stations (naïve occupancy rate = 10%). A radio-marked marten was detected at 1 station and an unmarked marten at 1 station (Fig. 5, 7). At the Swain grid in summer, martens were detected at only 1 camera station (naïve occupancy rate = 5%); radio-marked and an unmarked marten were both detected at this single station (Fig. 5, 7). None of the marked martens that were detected at a camera were detected at more than one camera, regardless of season. Thus, the camera detections provide no indication that 15 individuals used larger areas during winter than during summer, consistent with examination of the use areas for each season (Fig. 5). A surprising number of unmarked marten detections occurred (Fig. 7). Like the marked martens, there were more stations with unmarked marten detections during winter (left hemispheres; Fig. 7) than in summer (right hemispheres). These martens were either present on the grids and not captured during one of our two capture sessions, or they arrived at the survey grids after we had concluded our trapping efforts. Only some of them could be identified (via their snared hair) but they contributed to the detection history data for each survey, and thus contributed to estimates of probability of detection. Probability of Detection and Adjusted Estimates of Occupancy Review of the relationship between the proportions of detections per survey day across the entire survey duration (for the first summer and winter; 2007-2008), revealed that survey days that were added after the standard 28-day survey period – to compensate for inoperable survey days – varied substantially, compared to the first 28-day period (Fig. 8). Therefore, for the purposes of estimating probability of detection for each season, for all years, we truncated all datasets at 28 calendar days. For the multi-season occupancy analysis, a single top model clearly outperformed all other candidate models (Model 1, AIC weight = 0.76, Table 2). This model included custom parameters, including week-specific estimates for probability of detection (p), seasonal variation in p for the winter and summer seasons, and the effect of the presence of bait. The data suggested that probability of detection was different for three portions of the survey duration: week 1, weeks 2-3, and week 4, increasing during the survey duration (Table 3). Estimates of detection probability for each survey week were nearly double during the winter versus summer 16 (Table 3), however probabilities of detection were so high that, when compounded over the 28day survey period, winter and summer surveys both had overall estimates of P that were >0.99 (Table 3). These estimates apply to stations where the complete 28-day protocol was achieved, when survey days are lost due to camera inoperabilities or bait is not present, station-specific estimates of P were lower, but in all but the most extreme cases by less than 5%. The top model also included the bait status variable (Table 2), indicating that the presence or absence of bait helped account for additional detection heterogeneity. Bait was removed primarily during the summer, when the most frequent culprit, the black bear is active. In the summer 2010 surveys 28.5% (SE = 0.03) of the survey days had bait missing whereas in the winter of 2010, 22.7% (SE = 0.03) of the surveys days had bait missing. Most of the missing bait in the summer was due to visits by bears, whereas most of the missing bait in winter occurred after a marten had removed it (bears were in dens and unavailable in winter). The presence of bait resulted in a >170% increase in the likelihood (odds) of detecting a marten after accounting for the effects of survey week. This reflects the high incidence of detections that occur when bait was replenished. Adjusted Estimates of Occupancy On the basis of the best-fitting model, derived estimates of occupancy (ψ) varied significantly by both season and during the winter season by study area (Table 3). These estimates were significantly different (McNemar‟s χ2 = 30.9, df = 1, P < 0.0001) indicating that despite the greater detectability in winter, the adjusted estimates of occupancy still differed significantly between summer and winter. Martens were likely to be, on average, present at a total of only 5% (2 of 40 total stations) in summer, but 48% (19 of 40 total stations) in winter. 17 Furthermore, occupancies were significantly greater in Humboldt compared to Swain, but only during the winter (Table 3). The transition probabilities for colonization and extinction were best modeled using season and study area effects. As expected, significant differences were present between seasons in both probabilities (Table 3). Colonization transitions only occurred from summer to winter, while extinction transitions occurred in both seasonal transitions but were most pronounced from winter to summer (Table 3). This means that stations without detections in summer were much more likely to have detections in winter, than vice versa. DISCUSSION Martens occupied our survey grids at higher rates during the winter than the summer, a pattern verified over a 4-year period. They demonstrated 48% occupancy, on average, during the winter but only 5% of stations were estimated to be occupied during the summer. The higher occupancy in winter was statistically significant, despite the fact that martens are easier to detect during winter. Thus, the diminished probability of detection in summer was statistically overwhelmed by the fact that martens more commonly occur in the study areas during winter than summer. Sample sites that were without detections in summer were significantly more likely to have detections in winter (i.e. significant “colonization” of survey sites), but these same sites were also significantly less likely to stay occupied during the subsequent summer season (i.e. significant “extinction” of sites). Because our sampling methods were consistent across years and seasons (consistent survey locations and consistent detection devices) the difference in occupancy between winter and summer was a function of marten population dynamics or behavior, rather than variation in survey methods or effort. Thus, our most recent survey seasons 18 – winter 2010 and summer 2010 – confirmed and reinforced the seasonal pattern of occupancy that we had observed previously, with less comprehensive data. These general results are quite similar to the results of research on probabilities of detection and occupancy for the American mink (Neovison vison) (Reynolds et al. 2010) in which probabilities of detection were consistently lower in summer than in fall/winter, yet average occupancy estimates did not differ between seasons. Thus, although seasonal differences in detectability may be a common phenomenon in carnivores (see Hackett et al. 2007 and references therein), this means that there can, and typically are, more individuals available to be detected in winter than in summer. This is not too surprising given the general ecological principle that there are insufficient resources for all the young of the year that are produced by a species to survive the winter and reproduce. Annual cycles of abundance in temperate species, and the forces of natural selection, generate the phenomenon of overabundance in fall and density-dependent regulation of population size. We believe that these principles are ultimately responsible for the seasonal patterns in occupancy we observed. This is probably most evident at locations like Swain and Humboldt Peak which, over time, have become located at the margin of the marten‟s geographic range in the southern Cascades. Testing the Working Hypothesis Our radio-collared sample of animals provided some insights as to the proximate reasons for the seasonal pattern. The working hypothesis we sought to evaluate was whether the winter occupants were young-of-the-year animals that were temporary residents captured while they were in the process of finding adult home ranges (i.e., dispersers). Indeed, we verified that most of the animals captured during the fall and winter trap sessions were young animals. Of the individuals that could be aged via tooth cementum, 50% were < 1 year of age and only 1 of 13 19 individuals (7.7%) we captured was estimated to be > 2 years old. This appears to be a younger age distribution than in the Lake Tahoe Basin (where the habitat is perceived to be higherquality; W. Zielinski, pers. obs.). In the Basin there is a similar proportion of young animals but 14 of 71 (19.8%) of the individuals were > 2 years old (K. Slauson, unpubl. data). Contrary to our hypothesis, however, most radio-marked animals in this study resided on the study areas for longer periods of time than expected. From 5 to 10 months after initial capture most (58.3%) radio-marked individuals were still found within their respective study areas. Thus, we did not confirm the limited tenure in the study areas that we assumed would occur if the habitat was indeed so poor that individuals only needed to reside there long enough to assess, and then reject it as a home range site. The primary hypothesis also predicts an increase in mortality rates from the winter to the following summer. This is because if the habitat was unsuitable, and the animals did not leave, then we assumed that they were likely to die there. We believe that this prediction was realized in that survival rates decreased precipitously from July to September. It does not seem likely that if habitat conditions were favorable within the Swain or Humboldt study areas, martens would succumb at the rates we observed. Unfortunately, our surveillance of individual animals was not frequent enough to verify death (for 2 with unknown fate) or to verify cause of death for the 3 carcasses that were recovered. Nonetheless, we were surprised that if the habitat was significantly poorer in our study areas – compared to locations where seasonal patterns are less profound – that the martens would not have either left the study areas sooner, or that mortality rates didn‟t increase sooner. Unfortunately, we did not have a control area where survival rates at our study areas could be compared to a location where habitat conditions are predicted to be 20 more benevolent. It is possible that martens that do not find an acceptable home range reside in suboptimal habitat and wait until better sites become available. Originally, we also entertained the possibility that the seasonal difference in detections could be influenced by changes in the size of home ranges in winter compared to summer. We did not have sufficient data to address this possibility, but a cursory examination of the summer and winter use areas (Fig. 5) does not indicate marked differences. This agrees with others that did not report seasonal differences in either movement distances or home range sizes in martens (Phillips 1994, O‟Doherty et al. 1997, Gosse et al. 2005, Hearn 2006). Payer (1999) found smaller home ranges in winter, contrary to what we would expect if the higher winter detection rates were simply due to larger home ranges. In sum, the 2 study areas exhibited the marked seasonal variation in occupancy reported previously and they include male and juvenile-biased populations and increased summer mortality. We also verified the presence of only a single reproductive-age female. All these measures are indicative of potentially poor habitat, where martens are short lived (0-3 years) and successful breeding and parturition is rare, leading to the conclusion that these may be “sink” habitats, where death and immigration rates exceed birth and emigration rates (Pulliam 1988). An Update on Fate of Individuals: July 2011 Although we truncated the official collection of data at the end of September 2010, most of the surviving martens continued to be monitored for another purpose (K. Moriarty, unpubl. data). Thus, at the time of this writing (July 2011), we had additional information on the status of the subjects of our study. Of the 12 individuals that were originally captured (in the fall/winter of 2009-2010), 3 of them (25%) – all males – were still alive and located within the boundaries of the original survey grid. Two other males were alive, but also used areas outside 21 the original grid (i.e., both were Swain animals that were making significant use of the wilderness area to the west of this study area). Two of the animals of “unknown fate” (Fig. 3) were now presumed dead, given that substantial trapping effort in their former use areas did not result in their capture, and two additional individuals (F02, M07) were either dead or presumed dead. Thus, of the original 12 animals capture and marked, 7 (58.3%) were dead or presumed so as of mid-July 2011 (from 14 to 19 months after their original captures). This updated view of survival indicates that most of the 7 animals that died during the course of the study were young; 4 of the 7 (57.1%) were 0 or 1 year of age, the rest were 2 years of age. In summary, the original data – and the updated information – indicate that the working hypothesis was supported in terms of the expected age distribution and the mortality schedule. However the hypothesis was not necessarily supported in terms of the period of time we expected individuals to reside in the areas. This is because 3 of the original 12 animals (25%; all males) continued to live exclusively within the boundaries of the grid – for greater than one year – in an area that earlier work predicted to be of relatively low habitat value (Kirk and Zielinski 2009, Rustigian-Romsos and Spencer 2010) and which we expected few animals to use except during dispersal. However, we reiterate that none of the demographic information we collected suggests that the 2 study areas contribute to the populations in any significant way, particularly due to the juvenile biased age structure, high mortality rate for juveniles, and the fact that we captured only 1 breeding age female among the 12 individuals. Thus, the prediction of low habitat value for these study areas was supported by the demographic characteristics of the martens captured there. Estimating Probability of Detection 22 The weekly probability of detection (p) was significantly less in the summer than the winter, confirming our earlier work and the long-standing observation among marten trappers and researchers that martens are more difficult to capture or detect during the summer than winter (e.g., Campbell 1979, Soutiere 1979, Steventon and Major 1982, Bull et al. 1992, Wilbert 1992, Buskirk and Powell 1994, Fowler 1995, Ivan 2000). However, when the weekly probability was compounded over the entire survey period P exceeded 99% regardless of season. Thus, despite the effect that removal of bait by bears had on detection rates, our bait replacement rate and the survey duration were apparently sufficient to almost guarantee a certainty of detection if a marten was present. Implications for Habitat Modeling The temporal distribution of animal location data can dramatically affect the predictive habitat model that is created from these data (Schooley 1994, Arthur et al. 1996, Forsyth et al. 2005, Gayet et al. 2011). For example, Coe et al. (2011) evaluated the temporal performance of resource selection functions (habitat models) for elk (Cervus elaphus) for 9 monthly or seasonal periods over 3 years. They found that models were validated for some, but not all, of the months and seasons. Gayet et al. (2011) also discovered that seasonal differences in locations of mute swans (Cygnus olor) resulted in different predictors in the best seasonal habitat. In our study area, Rustigian-Romsos and Spencer (2010) predicted marten habitat using summer data, winter data and annual data and revealed significant differences in the distribution and abundance of habitat. Predicted habitat based on summer survey data was more limited in extent than habitat areas predicted when winter survey data were used. The challenge is in determining how to define marten habitat under these circumstances. 23 If, as we have suggested, the martens that were in detected in our study areas – which are locations that are primarily occupied during winter – are young animals searching for suitable habitat and are unlikely to maintain year-round home ranges there, then including detections of these animals in predictive habitat models will over represent suboptimal habitat in the resulting model. A model created with these data will include locations that are regularly assessed by young animals but rarely occupied for the purposes of breeding. Our data support our belief that this is the case and, therefore we agree with Rustigian-Romsos and Spencer (2010) that summer habitat – when females are occupying home ranges where they bear young and breed – is likely the most important to the marten population, not only because it is more restricted than habitats occupied during the non-breeding season, but also because it is known to support adults during the parturition and breeding season. Including survey locations where martens are primarily detected only in winter (the non-breeding season) in the development of habitat models may confound the important relationship among the habitat predictors associated with year-round residence. This is because many of these locations will be sites where juveniles or males occur during this season. However, some of these detections will also be females, thus future work should focus on the spatial dynamics of breeding adults and their seasonal variation in home range size and habitat so that we can understand how modeling habitat primarily for breeding adults will affect the results. When a habitat map is created from summer survey data, or from survey sites where martens are detected during both summer and winter, we can be assured that the animals that are detected are more likely to be contributing to future population growth. Thus, habitat models (and maps) created from these data identify important areas for marten management. The same cannot be said about models that are developed using data from locations where martens are 24 detected only during the winter, non-breeding season. Given that marten distributions have decreased in portions of the Cascades and Sierra Nevada mountains in California (Zielinski et al. 2005, Moriarty et al. in press), it is important that land management activities be especially careful not to adversely affect the high-value areas identified by models that were derived from data collected during the breeding season. However, as Rustigian-Romsos and Spencer (2010) also noted, the expanded area of occupancy during winter, which includes more lower-elevation habitats that are less likely to be used during the breeding and kit-rearing seasons, may also be important for sustaining the marten population by providing habitat for dispersing juveniles during winter as they search for suitable areas to settle. Alternatively, areas identified as supporting only winter occupancy may represent opportunities where habitat conditions can be restored or improved such that summer occupancy could be supported there. Our data lead us to conclude that, in response to the Pinchot Institute‟s request for “a suitable habitat definition for marten” (Anonymous 2008), the most defensible approach is to base the definition of suitable habitat (at the landscape scale) on a habitat model built using locations where martens are detected during summer or during both summer and winter. Thus, we support – as definition of suitable landscape-scale habitat – the results of modeling by Kirk and Zielinski (2009) and the summer model of Rustigian-Romsos and Spencer (2010). Locations where martens are detected primarily during winter can be included in a secondary model, which will identify a wider array of habitat types that include locations that support movement that is largely unrelated to breeding or home range establishment. Our study brings to light a fundamental shortcoming in the development of habitat models for martens, as well as for other species: the ambiguity of the definition of “suitable habitat”. Most often suitable habitat is simply defined as where an individual of a species 25 occurs – whether it be a dispersing juvenile, an adult male or a reproductive female. Obviously, the key to managing habitat for a species is to understand what habitat features are essential for maintenance of the population. Thus, managers interested in maintaining and increasing populations should consider suitability to vary as a function of the population segment of most importance, the reproductive female. This is an especially important consideration in a polygamous species such as the marten. Slauson (in prep) is exploring the implications of this variation in data sources for marten habitat modeling in the Lake Tahoe basin. This work suggests that knowledge of the spatial locations of breeding females (and males) would be the most important data source for the development of credible habitat models. Ideally, our ultimate goal should be to distinguish two classes of suitable habitat: “suitable for reproduction” and “suitable for non-reproductive activities”. Our study lacked a control area and suffered from relatively small sample sizes. These shortcomings will be avoided in new, ongoing work in the Lake Tahoe Basin which will attempt to describe source and sink habitats with better precision (Slauson, in prep.). Although we did not have the data in this study to assess the breeding status of the subjects, our data support the notion that martens that are detected in winter only, in the Swain and Humboldt Peak study areas, are less likely to be individuals that contribute to future population growth than martens that are detected during the summer. ACKNOWLEGEMENTS We thank Tom Frolli and Mark Williams, of the Lassen National Forest (LNF), for their support of this project. We appreciate the financial support of the Lassen National Forest, the Pacific Southwest Region and the Pacific Southwest Research Station of the USDA Forest Service, and the Herger Feinstein Quincy Library Group steering team and staff. Discussions with LNF biologists were instrumental in developing plans for field work and analysis. We 26 thank the following for their help in collecting and summarizing field data: Cassie Kinnard, Colleen Heard, Kaley Phillips, Mark Linnell, Mathew Delheimer, Katie Mansfield, and Patrick Lieske. LITERATURE CITED Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267-281 in Petran, B. N., and F. Csaki, editors. International Symposium on Information Theory. Second edition. Akademiai Kiadi, Budapest, Hungary. Anonymous, 2008. HFQLG independent science panel “red flag” issue monitoring report. Pinchot Institute for Conservation, Washington, D.C. Arthur, S. M., B. F. J. Manly, L. L. McDonald, and G. W. Garner. 1996. Assessing habitat selection when availability changes. Ecology 77:215-227. Bailey, R. G. 1994. Descriptions of the ecoregions of the United States. Second edition. Miscell. Publication 1391. US Forest Service, Washington, D.C. Bissonette, J. A., D. J. Harrison, C. D. Hargis, and T. G. Chapin. 1997. The influence of spatial scale and scale-sensitive properties on habitat selection by American marten. Pages 368-385 in J. A. Bissonette, editor. Wildlife and landscape ecology: effects of pattern and scale. Springer, New York. Bodkin, J. L., J. A. Ames, R. J. Jameson, A. M. Johnson, and G. E. Matson. 1997. Estimating age of sea otters with cementum layers in the first premolar. Journal of Wildlife Management 61:967973. Bull, E. L., R. S. Holthausen, and L. R. Bright. 1992. Comparison of 3 techniques to monitor marten. Wildlife Society Bulletin 20:406-410. Buskirk, S. W., and R. A. Powell. 1994. Habitat ecology of fishers and American martens. Pages 283-296 in Buskirk, S. W., A. S. Harestad, M. G. Raphael, and R. A. Powell, editors. Martens, sables and fishers: biology and conservation. Cornell University Press, Ithaca, New York. Campbell, T. M. 1979. Short-term effects of timber harvests on pine marten. Thesis, Colorado State University, Fort Collins. Carroll, C., W. D. Spencer, and J. C. Lewis. In press. Use of habitat and viability models in Martes conservation and restoration in Biology and Conservation of Martens, Sables, and Fishers: A New Synthesis. K.B. Aubry, W.J. Zielinski, M.G. Raphael, G. Proulx, and S.W. Buskirk, editors. Cornell University Press, Ithaca, New York, USA. 27 Coe, P. K., B. K. Johnson, M. J. Wisdom, J. G. Cook, M. Vavra, and R. M. Nielson. 2011. Validation of elk resource selection models with spatially independent data. Journal of Wildlife Management 75:159-170. Davis, C. and M. D. King. 2001. The Quincy Library Group and collaborative planning within U.S. national forests. Unpubl. report, Colorado State University, Fort Collins, Colorado. Ellis, L. 1998. Habitat use patterns of the American marten in the Southern Cascade Mountains of California, 1992-1994. Thesis, Humboldt State University, Arcata, California. Forsyth, D. M., W. A. Link, R. Webster, G. Nugent, and B. Warburton. 2005. Nonlinearity and seasonal bias in an index of brushtail possum abundance. Journal of Wildlife Management 69:976-984. Fowler, C. H. 1995. Techniques for detecting and monitoring martens and fishers in forest habitats of California. Thesis. Humboldt State University, Arcata, California. Fuller, A. K., and D. J. Harrison. 2005. Influence of partial timber harvesting on American martens in north-central Maine. Journal of Wildlife Management 69:710-722. Gayet, G., M. Guillemain, M. Benmergui, F. Mesleard, T. Boulinier, J-P. Bienvenu, H. Fritz and J. Broyer. 2011. Effects of seasonality, isolation and patch quality for habitat selection processes by mute swans Cygnus olor in a fishpond landscape. Oikos 120:801-812. Gosse, J. W., R. Cox, and S. W. Avery. 2005. Home-range characteristics and habitat use by American martens in eastern Newfoundland. Journal of Wildlife Management 86:1156-1163. Hackett, H. M., D. B. Lesmeister, J. Desanty-Combes, W. G. Montague, J. J. Millspaugh, and M. E. Gompper. 2007. Detection rates of eastern spotted skunks (Spilogale putorius) in Missouri and Arkansas using live capture and non-invasive techniques. American Midland Naturalist 158:123131. Hearn, B. J. 2007. Factors affecting habitat selection and population characteristics of American marten (Martes americana atrata) in Newfoundland. Dissertation, University of Maine, Orono. Hines, J. E. 2006. PRESENCE2. Software to estimate patch occupancy and related parameters. USGS-PWRC. Available from http://www.mbr-pwrc.usgs.gov/software/presence.html (Accessed 02 August 2011). Ivan, J. S. 2000. Effectiveness of covered track plates for detecting American marten. Thesis, Purdue University, West Lafayette, Indiana. Kays, R. W., and K. M. Slauson. 2008. Remote cameras. Pages 110-140 in Long, R. A., P. MacKay, W. J. Zielinski, and J. C. Ray, editors. Noninvasive survey methods for carnivores. Island Press, Washington D.C. 28 Kendall, K. C., and K. S. McKelvey. 2008. Hair collection. Pages 135-176 in Long, R. A., P. MacKay, W.J. Zielinski, and J. C. Ray, editors. Noninvasive survey methods for carnivores. Island Press, Washington D.C. Kirk, T. A. 2007. Landscape-scale habitat associations of the American marten (Martes americana) in the Greater Southern Cascades region of California. Thesis, Humboldt State University, Arcata, California. Kirk, T. A., and W. J. Zielinski. 2009. Developing and testing a landscape habitat suitability model for the American marten (Martes americana) in the Cascades mountains of California. Landscape Ecology 24:759-773. Kirk, T. A., and W. J. Zielinski. 2010. Modeling functional habitat connectivity of the American marten (Martes americana) in northeastern California using least-cost corridors. Report to the Lassen National Forest. Pacific Southwest Research Station, Arcata, California. Kreeger, T. J., and J. M. Arnemo. 2007. Handbook of wildlife chemical immobilization. Third edition. Self-published, Laramie, Wyoming. Levins, R. 1969. Some demographic and genetic consequences of environmental heterogeneity for biological control. Bulletin of the Entomological Society of America 15:233-240. MacKenzie, D. I., J. D. Nichols, J. A. Royle, K. P. Pollock, L. L. Bailey, and J. E. Hines. 2006. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Academic Press, San Diego, California. Matson, G. M. 1981. Workbook for cementum analysis. Matson's Lab, Milltown, Montana. Mayer, K. E., and W. F. Laudenslayer, Jr. 1988. A guide to wildlife habitats of California. California Department of Forestry and Fire Protection, Sacramento, California. McNemar, Q. 1947. Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika 12:153-157. Millspaugh, J. J., and J. M. Marzluff, editors. 2001. Radio tracking and animal populations. Academic Press, San Diego, California. Moriarty, K. M, W. J. Zielinski, and E. Forsman. In press. Decline in American marten occupancy rates at Sagehen Experimental Forest, California . Journal of Wildlife Management. O'Doherty, E. C., L. F. Ruggiero, and S. E. Henry. 1997. Home-range size and fidelity of American martens in the Rocky Mountains of southern Wyoming. Pages 123-134 in Proulx, G., H. Bryant, and P. M. Woodard, editors. Martes: taxonomy, ecology, techniques, and management. Provincial Museum of Alberta, Edmonton, Canada. 29 Payer, D. C. 1999. Influences of timber harvesting and trapping on scale-specific habitat selection and demographic performance of American marten. Dissertation, University of Maine, Orono, Maine. Payer, D. C., and D. J. Harrison. 2003. Influence of forest structure on habitat use by American marten in an industrial forest. Forest Ecology and Management 179:145-156. Perrine, J. D. 2005. Ecology of red fox (Vulpes vulpes) in the Lassen Peak region of California, U.S.A. PhD dissertation, University of California, Berkeley, California. Phillips, D. M. 1994. Social and spatial characteristics, and dispersal of marten in a forest preserve and industrial forest. Thesis, University of Maine, Orono, Maine. Phillips, D. M., D. J. Harrison, and D. C. Payer. 1998. Seasonal changes in home-range area and fidelity of martens. Journal of Mammalogy 79:180-190. Pollock, K., S. R. Winterstein, C. M. Bunck, and P. D. Curtis. 1989. Survival analysis in telemetry studies: the staggered entry design. Journal of Wildlife Management 531:7-15. Powell, R. A. 2000. Animal home ranges and territories and home range estimators. Pages 65-110 in L. Boitani and T. K. Fuller, editors. Research techniques in animal ecology: controversies and consequences. Columbia University Press, New York. Pulliam, R. 1988. Sources, sinks and population regulation. American Naturalist 132: 652-661. Reynolds, J. C., T. A. Porteus, S. M. Richardson, R. J. Leigh, and M. J. Short. 2010. Detectability of American mink using rafts to solicit field signs in a population control context. Journal of Wildlife Management 74:1601-1606. Rustigian-Romsos, H. L., and W. D. Spencer. 2010. Predicting Habitat Suitability for the American Marten on the Lassen National Forest. Unpublished report. USDA Forest Service, Lassen National Forest, Susanville, California. Schooley, R. L. 1994. Annual variation in habitat selection: patterns concealed by pooled data. Journal of Wildlife Management 58:367-374. Soutiere, E. C. 1979. Effects of timber harvesting on marten in Maine. Journal of Wildlife Management 43:850-860. Spencer, W. D., R. H. Barrett, and W. J. Zielinski. 1983. Marten habitat preferences in the northern Sierra Nevada. Journal of Wildlife Management 47:1182-1186. Slauson, K. M. In prep. Linking landscape pattern to population process in a carnivorous mammal. PhD proposal, University of Montana, Missoula, Montana. 30 Steventon, J. D., and J. T. Major. 1982. Marten use of habitat in a commercially clear-cut forest. Journal of Wildlife Management 46:175-182. Wilbert, C. J. 1992. Spatial scale and seasonality of habitat selection by martens in southeastern Wyoming. Thesis, University of Wyoming, Laramie. Zielinski, W. J., T. A. Kirk, and K. M. Slauson. 2009. The effect of season on detectability of martens in the Greater Southern Cascades region: Lassen National Forest, California. Unpublished report. USDA Forest Service, Pacific Southwest Research Station, Arcata, California. Zielinski, W. J., R. L. Truex, F. V. Schlexer, L. A. Campbell, and C. Carroll. 2005. Historical and contemporary distributions of carnivores in forests of the Sierra Nevada, California, USA. Journal of Biogeography 32:1385-1407. 31 Figure Legends Figure 1. Previously surveyed carnivore grids (Mineral, Humboldt Peak, and Swain Mountain) within the Lassen National Forest and Lassen Volcanic National Park, California. Remote camera sites are designated by circles. Figure 2. Camera, trap and initial capture locations, relative to the previously surveyed Humboldt and Swain camera survey grids during two trapping sessions (October- November 2009 and March-April 2010). A: grid locations used for current and previous camera surveys, B: locations where traps were set, which included previous cameras stations and some new locations within or just outside the grids, C: locations where individual martens were captured (closed circles) relative to all trap sites (open circles). Figure 3. Timeline of fates of individual radio-marked martens, from the date they were first captured (closed circles), to they were either confirmed to have died (closed squares), or when their fate became unknown. Time horizon represents one year, from 28 September 2009 – 30 September 2010. Solid lines represent time periods during which the individual was known to be within the survey grid (either Swain or Humboldt Peak) and dashed lines indicate when the animal was either known to be off the grid (large dashes: M06, M10, and M11) or suspected to be off the grid (M09). Figure 4. Kaplan Meier monthly survival rate estimates for all martens, beginning in November of 2009 and concluding in September of 2010. Dotted lines indicate standard errors (upper value not shown when > 1.0). 32 Figure 5. Minimum convex polygons and camera stations for Humboldt Peak (A - D) and Swain Mountain (E - H) study areas, Lassen National Forest. Camera stations are indicated by circles; empty half-circles represent no detections during either winter (left hemisphere) or summer (right hemisphere). Black polygons represent 100% minimum convex polygons for all known locations (total use area), blue dashed polygons are winter use areas (from 29 December - 7 March 2010 in Swain and from 15 February - 23 April 2010 in Humboldt: 2 weeks before until 2 weeks after the winter camera survey), and the red dashed polygons enclose the summer use areas (from 24 June – 20 August 2010 in Swain and from 27 July to 23 September 2010 in Humboldt: 2 weeks before until 2 weeks after the summer camera survey). Numbers near each polygon identify the individual marten. Figure 6. (A) Detections at camera stations on the Humboldt, Swain and Mineral grids over 3years, replicated during 3 summers (between June and September) and 2 winters (between November and January). Open circles are camera stations with no marten detection, closed circles indicate a detection. Mineral grid sampled only during summer of 2007. (B) Proportions of stations at each grid with detections for each of the 3 summer and 2 winter sample periods. Data correspond to the season directly above in panel A. Figure 7. Detection locations of unmarked martens at Humboldt (A) and Swain (B). Camera stations are indicated by circles; empty half-circles represent no detections during either winter (left hemisphere) or summer (right hemisphere). Surveys in Humboldt were conducted from 1 March – 15 April 2010 in winter and 10 August – 17 September 2010 in summer and surveys in 33 Swain were conducted from 11 January – 7 March 2010 in winter and 8 July – 13 August 2010 in summer. Figure 8. Plot of the proportion of stations where martens were detected using cameras (y axis) during each survey day (x axis) in the summer of 2007 and the winter of 2007-2008. 34 Figure 1. 35 36 Figure 2. 37 Figure 3. 38 1.0000 0.9000 Survival rate 0.8000 0.7000 0.6000 0.5000 0.4000 0.3000 0.2000 Figure 4. 39 Figure 5. Proportion of Satations 40 60 50 40 30 20 10 0 Mineral Figure 6. Swain Humoldt Peak 41 Figure 7. 42 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 Summer Winter 1 Figure 8. 4 7 10 13 16 19 22 25 28 31 34 37 40 43 Table 1. Individual characteristics of martens captured in the vicinity of either the Humboldt or Swain camera survey grids from October 2009 – March 2010. Study Site ID Initial Capture Date Estimated Age (yrs)1 Sex Initial Weight (g) No. Locations Estimated Use Area (km2) M01 Humboldt 7-Oct-09 2 Male 1050 61 15.60 M02 Humboldt 26-Oct-09 0 Male 840 45 42.06 M03 Humboldt 27-Oct-09 2 Male 970 44 5.99 M04 Humboldt 28-Oct-09 2 Male 950 34 5.28 M05 Humboldt 28-Nov-09 0 Male 815 03 --- M06 Swain 2-Nov-09 1 Male 1002 49 4.51 M07 Swain 4-Nov-09 0 Male 840 47 2.95 M08 Swain 8-Nov-09 1+ Male 1060 53 10.02 M09 Swain 11-Nov-09 0 Male 920 15 56.09 F01 Humboldt 19-Nov-09 Adult2 Female 700 6 0.63 M10 Swain 24-Feb-10 3 Male 1090 20 10.23 F02 Swain 8-Mar-10 1 Female 770 38 1.14 M11 Swain 8-Mar-10 0 Male 1095 via tooth cementum annuli; age = 0 refers to young of the year. 2 estimated; premolars missing. 3 small juvenile, not collared. 15 7.05 1 44 Table 2. Candidate models fit to the detection data for the purposes of estimating probability of detection (p), occupancy (ψ, psi), extinction (έ, eps), and colonization (γ, gamma). Covariates include “Season” (winter v summer), “Swain” (study area effect),” wk” (week within the survey duration; 1 - 4), “W” (the months December – February), “Spr” (the month of March), and “Bait” (the presence or absence of bait). Models compared using Akaike‟s Information Criteria for which a lower value suggests a better-fitting model. Models were generated from 5 seasons of occupancy data in 2 study areas, Swain and Humboldt Peak, combined. Akaike weights (w) represent the strength of the model and K is the number parameters in each model. Model # 1 2 3 4 5 6 7 Model psi,gamma(Season Swain),eps(Season Swain),p(wk_1_23_4 Bait Season_W_Spr) psi,gamma(Season), eps(Season),p(wk_1_23_4 Bait Season_W_Spr) psi,gamma(t),eps(t),p(wk_1_23_4 Bait Season_W_Spr) psi,gamma(Season),eps(Season),p(wk_1_23_4 Bait Season_W) psi,gamma(Season),eps(Season),p(wk_1_23_4 Bait) psi,gamma(),eps(),p(wk1_23_4) psi,gamma(),eps(),p() ΔAIC AIC weight Model Likelihood K 0.0 2.8 5.5 15.6 20.2 28.1 83.4 0.76 0.19 0.05 0.00 0.00 0.00 0.00 1.00 0.25 0.06 0.00 0.00 0.00 0.00 13 11 15 10 9 8 4 45 Table 3. Parameter estimates (SE) for individual variables in the top model for modeling marten detection probability for 5 seasons from 2007 to 2010 in the Humboldt and Swain study areas. Variable Summer 2007 Winter 2008 Summer 2008 Winter 2010 Summer 2010 0.03 (0.025) 0.03 (0.025) 0.60 (0.08)* 0.32 (0.08)** 0.06 (0.04) 0.07 (0.04) 0.60 (0.08)* 0.34 (0.08)** 0.06 (0.04) 0.07 (0.05) Colonization (γ) NA 0.60 (0.08) 0 (0) 0.60 (0.08) 0 (0) Extinction (έ) NA 0.47 (0.35) 0.90 (0.06) 0.47 (0.35) 0.90 (0.06) Week 1 Week 2-3 Week 4 0.087 (0.03) 0.142 (0.03) 0.288 (0.06) 0.159 (0.03) 0.247 (0.02) 0.444 (0.04) 0.087 (0.03) 0.142 (0.03) 0.288 (0.06) 0.159 (0.03) 0.247 (0.02) 0.444 (0.04) 0.087 (0.03) 0.142 (0.03) 0.288 (0.06) Detection Probability for entire 28-day Protocol 0.994 0.999 0.994 0.999 0.994 Occupancy (ψ) Humboldt Swain Detection Probability *P-value <0.05 for McNemar's chi-squared test for seasonal comparisons; indicating that winter ≠ summer. **P-value <0.05 for McNemar's chi-squared test for study area comparisons; indicating that Swain and Humboldt are only different during winter.