• Open Access

Detecting a population decline of woodland caribou (Rangifer tarandus caribou) from non-standardized monitoring data in Pukaskwa National Park, Ontario


  • Associate Editor: Glenn.


Observation bias from methodological inconsistencies plague many long-term ecological monitoring studies, leaving land managers to question the validity of apparent population trends over time. Furthermore, some species are cryptic and have low detectability, so assessments are naturally imprecise. We assessed the utility of aerial surveys for woodland caribou from 1972 to 2009 at a Canadian national park in detecting a reliable population trend for this threatened species. The surveys varied in flight patterns, total distance flown, observer experience, speed, altitude, timing, temperature, and snow depth. Of these, distance and the speed/altitude index influenced the population estimates, whereas no variables influenced the calf:female ratio or winter range size. Year was included in all plausible models for population estimates, and the majority of plausible models for calf:female ratio and winter range size. Population size, recruitment and winter range size all declined over time in the respective models with the lowest AICc. Switching methodologies mid-way through a long-term aerial survey monitoring program creates greater complexity for trend analysis over time; however, this study suggests that reliable conclusions can still be drawn from long-term monitoring data if confounding factors are accounted for in the analysis. © 2014 The Wildlife Society.

Temporal change in population size remains poorly understood because of the scarcity of long-term surveys and methodological issues often associated with those data sets (Magurran et al. 2010). Nevertheless, population trends are needed to assess the status and predict the trajectories of future population growth of species at risk of extinction (Gerber et al. 1999, Shea and Mangel 2001). Given that population sizes vary with shifting environmental conditions and infrequent extreme events, long-term data are needed to distinguish population trends from short-term variability (Peters 2010). Despite their value, wildlife population studies based on >10 years of data are rare (Pelton and van Manen 1996). Moreover, it is common for data collection methods and observers to change over the course of the surveys. Changes and improvements to methodology occur because of staff turnover; new knowledge or technology; and updated objectives, funding, capacity, and priorities (Peters 2010). These inconsistencies make it difficult to determine whether a trend detected by monitoring is occurring in nature or if it is a result of the varying methodology (Oakley et al. 2003).

A common method for estimating population size of large terrestrial mammals is aerial surveys (Caughley 1977, Gasaway et al. 1985), although alternative techniques such as forward-looking infrared surveys and genetic analysis have recently gained popularity (Thompson 2004, Carr et al. 2012). Despite widespread use, traditional aerial surveys often result in low rates of detectability, which may distort inference about patterns, such as temporal trends (Hochachka and Fielder 2008). Detectability in aerial surveys may be affected by factors such as vegetation cover (Thomas 1998), type of aircraft (Fleming and Tracey 2008), observer experience (Thomas 1998), time of day and year (VanderWal et al. 2011), altitude (Shupe and Beasom 1987, Graham and Bell 1989), speed (Hone 1986, Shupe and Beasom 1987), temperature (Bleich et al. 2001), and group size (Samuel and Pollock 1981). Methods used to estimate detectability include comparisons with ground surveys, infrared imagery, double-counting, mark–recapture, or line transects (reviewed in Pollock and Kendall 1987). These generally require inclusion into the survey design and must be conducted concurrently with aerial surveys, and are thus inappropriate for long-term data already collected. Virtually no studies have retrospectively examined the comparability of long-term aerial surveys with varying methodology (but see Thomas 1998).

Pukaskwa National Park (NP) protects a small and relatively isolated population of woodland caribou (Rangifer tarandus caribou), a species assessed as “Threatened” in Canada (COSEWIC 2002) and protected under the Species at Risk Act (Government of Canada 2002). Factors contributing to population declines for caribou include anthropogenic landscape disturbance (Vors et al. 2007; Environment Canada 2008, 2012), predation by wolves (Bergerud and Elliot 1986, Bergerud 2007), low recruitment (Bergerud and Elliot 1986, Cumming and Beange 1993, Pinard et al. 2012), and reduced range occupancy (Ontario Woodland Caribou Recovery Team 2008). Caribou have been a conservation priority and a focal species for monitoring since the park's inception. A 2010 workshop on Pukaskwa NP caribou conservation was attended by government biologists, land managers, and university researchers; therein, participants questioned the validity of the monitoring data given inconsistency in the survey methods, and questioned whether the park's caribou population was truly declining and at risk of extirpation, as suggested by Bergerud et al. (2007). The validity of year-to-year comparisons for aerial surveys in Pukaskwa NP had also been questioned by Thompson and Peterson (1988). Although some studies have inferred trends about the population from the survey data (Bergerud 1985, 1996; Neale 2000; Environment Canada 2012), and determined the Pukaskwa NP and/or the north Lake Superior population to be declining (Bergerud et al. 2007, Environment Canada 2008, Parks Canada 2008) or at risk of extirpation by 2018 (Bergerud et al. 2007), no studies have tested the variation in survey parameters across years to determine their influence on results. Failure to properly account for observation bias in population analysis may lead one to underestimate abundance and to obtain distorted inference about patterns such as temporal trends (Hochachka and Fielder 2008). Compounding this uncertainty, an aerial survey completed in 2009 used forward-looking infrared technology and detected 4 times more animals than the traditional aerial survey completed that same month. If forward-looking infrared results can be assumed to approach 100% detectability, and if the 1:4 ratio of animals detected in the aerial survey to total animals documented in 2009 is representative of ratios in other years, then traditional aerial surveys may have sampled only 25% of the park's caribou population. It was not surprising that aerial surveys did not achieve a complete count of caribou because observed counts usually do not equate to true abundance (Kéry et al. 2009) and detectability of caribou in the Pukaskwa NP area has not been reliably estimated. The degree to which a survey detects animals is a function of the availability of the animal for detection (pa), the proportion of the area that is sampled (parea), and the fraction of animals that are detected (pd; Pollock et al. 2004). If pa, pd, and parea differ systematically across surveys, a trend over time could be biased and produce misleading results.

The objective of this study is to determine whether Pukaskwa NP's caribou monitoring data, a data set spanning 37 years with inconsistent methodology, observer turnover, and imperfect detection, can produce a reliable trend. We investigate 1) whether the caribou population has declined over time even when temporal variation in survey parameters is controlled for statistically; 2) whether recruitment and winter range size within Pukaskwa NP have declined over time; and 3) how survey parameters affect population estimates, calf:female ratios, and winter range size. Calf:female ratios are positively correlated with both recruitment and population growth rate (Harris et al. 2008). Moreover, according to the “melting range” hypothesis, the range of declining population contracts toward high abundance areas (Lawton 1993, Brown 1995). Thus, declining recruitment and winter range size are 2 corroborating forms of evidence that would support a true population decline versus a methodological artifact.


This research was conducted in Pukaskwa National Park, a 1,878-km2 wilderness park on the northeast shore of Lake Superior in Ontario, Canada (Fig. Fig. 1). Two distinct physiographic regions, coastal and interior, existed within the Park. The coastal region was characterized by rugged topography with steep rocky outcrops and bedrock-controlled soils. The interior region was a relatively flatter terrain with coarse-loamy soils. Almost 94% of the park's land area included forest, interspersed with many lakes and rivers comprising 22 watersheds. The southern coastal area had a rugged topography with elevations ranging up to 650 m, whereas elevations in the northern coastal area and interior of the park were relatively lower, seldom exceeding 400 m. Small near-shore islands (<1 km from the mainland) were used regularly by caribou throughout the year (Bergerud 1985). The climate in Pukaskwa NP was maritime-like because it was moderated by its proximity to Lake Superior. Mean daily winter and summer temperatures range from −13°C to 15°C for the coastal region and −17°C to 16°C inland. Average annual cumulative snowfall was greatest in the center of the Park (500 cm) and least along the coast (300 cm; Findlay 1973). Approximately 50% of Lake Superior adjacent to the coast was covered with land-fast ice on an average year, but in severe winters the entire coast could be ice covered (Bergerud et al. 1983). The dominant tree species included white birch (Betula papyrifera), black spruce (Picea mariana), and jack pine (Pinus banksiana); and the distribution of age classes was skewed to older stands, reflecting a paucity of natural disturbances in the park's recent history.

Figure 1.

Aerial surveys for woodland caribou (Rangifer tarandus caribou) were conducted between 1974 and 2009 on islands and in the coastal strip, within 5 km of Lake Superior's shoreline of Pukaskwa National Park, Ontario, Canada.

Woodland caribou are medium-sized, sedentary, forest-dwelling ungulates (Thomas and Gray 2002) that tend to aggregate in small groups during the winter but remain relatively solitary in the summer (e.g., Darby and Pruitt 1984, Bergerud 1985, Pare and Huot 1985, Cumming and Beange 1987). Since the early 1900s, woodland caribou numbers have declined, and the distribution has receded northward by approximately 40–50% since the mid-1800s because of natural and anthropogenic factors such as habitat loss, fragmentation, human disturbance, and increased predation (Thomas and Gray 2002, Schaefer 2003, Festa-Bianchet et al. 2011). Despite this recession, a few small populations have persisted along the north shore of Lake Superior, in the southern part of their historical range. They consist of 2 large predator-free island populations and a small continuous but dispersed mainland population, which includes Pukaskwa NP (Environment Canada 2012). Animals have likely persisted along the coastal mainland because of the preponderance of near-shore islands and steep terrain, which may serve as refugia from predators (Bergerud 1985).


Survey Methods

Aerial surveys were conducted in Pukaskwa NP between 1972 and 2009, taking place annually from 1972 to 1983, and biennially from 1985 to 2009 between December and March. From 1972 to 1980, surveys were conducted as part of faunal investigations, which covered the entire park and whose purpose was to estimate the size of large mammal populations, including woodland caribou (B. C. Johnson, Pukaskwa National Park, unpublished report). Rotary-wing aircraft were used exclusively in all years except 1997, where a fixed-wing aircraft was used for part of the survey. The crew typically consisted of a pilot, navigator, and 2 backseat observers (one on either side of the aircraft). Surveys were conducted on sunny days with winds of <30 km/hour, and between 24 and 72 hours after a fresh snowfall. The area surveyed was the strip of Lake Superior coastline within the Pukaskwa NP boundary inland to 5 km, as well as the near-shore islands. It was assumed that surveying 5 km inland provided an adequate sample of the population, and was very close to the actual population, because caribou seem to be restricted in their winter movements to this area. For example, no caribou have been sighted beyond 5 km of the coast in the 11 park-wide moose (Alces alces) surveys conducted since 1976 (Pukaskwa National Park, internal reports). Flight paths were recorded on topographic maps (National Topographic System 1:50,000 or Provincial Series 1:126,740) prior to or during the flights. Flights followed a transect system in an approximate north–south direction following the coast for 1974, 1975, and 1977, and an east–west direction from 1987 to 2009. Transect spacing was 0.8 km in 1974, 1 km in 1975, random in 1977, 1 km from 1987 to 1995, and 0.5 km from 1997 to 2009. Estimated observer distance was 0.4 km in 1974, 0.5 km for 1993–1997, 0.25 km for 1999–2009, and not noted for other years. The flight pattern for 1976, and 1978–1985, was random. The location and number of caribou and fresh tracks sighted during a survey were recorded and summarized in a report. Whenever possible, animals were classified by age (calves or adults) and sex. In most cases, sex was confirmed by track size and urine spot location. The reports also contained estimates of Pukaskwa NP's caribou population, which were not based on techniques designed to compensate for missed animals (e.g., Burnham et al. 1980), but rather were calculated by counting the number of observed caribou, fresh tracks, and other evidence (e.g., scat) in areas with no observed caribou, and subtracting any suspected double-counts from this number. Snow depth was measured at several locations within 5 km of the coast during the survey.

Statistical Methods

We performed all statistical analyses (principal components analysis, linear regression, Pearson's product–moment correlations, backward stepwise [analysis of covariance] regression, calculation of Akaike's Information Criterion for small sample sizes, generalized linear models with Poisson and quasipoisson distributions, dispersion test, generalized additive model, and linear piecewise regression) in R (R Development Core Team 2010). We conducted all spatial analyses (digitizing flight lines and calculating total distance) in ArcGIS 10.1. In years where faunal investigations occurred, we only included caribou data collected in the coastal portion of the surveys, which were dedicated to searching for caribou. Though some surveys extended beyond Pukaskwa NP, we only used data collected within the park. We omitted 3 years (1972, 1973, and 1983) from the population trend analyses: 1972 because there was no coastal portion to the survey and the population estimate was derived from 4 survey types, including a canoe survey; 1973 and 1983 because of poor data quality (C. R. Mullen, Pukaskwa National Park, unpublished report; G. Keesey, Pukaskwa National Park, unpublished report). We excluded 5 years (1972, 1973, 1983, 2007, and 2009) from the recruitment analysis because individuals were sexed and aged unreliably or not at all.

We assumed that pa (availability of animals for detection; Pollock et al. 2004) was constant over time because terrain and vegetation did not change significantly within the park between 1972 and 2009. We used the total distance flown (km) as a measure of parea (the proportion of the area sampled; Pollock et al. 2004) instead of area surveyed (km2) because surveys conducted in 1976 and from 1978 to 1982 followed random flight lines, and some locations were surveyed multiple times. The use of transect length is also believed to increase precision of a trend estimate (Kéry et al. 2009). To obtain total distance flown, we digitized the flight lines from original survey maps (within 0.1 km for National Topographic System 1:50,000 and within 1 km for Provincial Series 1:126,740) and calculated the sum of transect lengths. For surveys with random flight patterns, we calculated total distance. To quantify pd (the fraction of animals that are detected, Pollock et al. 2004), we used observer experience (years), an index of flight speed (km/hr) and altitude (m), timing of the survey (Julian start date), flight pattern (north–south transects, east–west transects, or random flight lines), average snow depth (cm) and average daily high temperatures (°C). Observer experience was the sum of the number of years of aerial surveys for caribou at Pukaskwa NP for each observer divided by the total number of observers that year. We used the midpoint of flight speed and altitude because these were typically given as a range of values. Average flight speed and altitude were correlated (r2 = 0.51, P < 0.001), so we ran a principal components analysis to obtain a combined index (Kaiser and Norman 1991). To obtain snow depth, we averaged the snow depth measurements taken manually at random locations during each survey. When survey measurements were unavailable, we used the predicted value from a linear regression of observed depths with depths from the nearest Environment Canada weather station in Wawa, Ontario, approximately 80 km east of the southern tip of the park. We also obtained the average daily maximum temperature for the months over which each survey occurred from this weather station. Finally, to assess the degree of potential multi-collinearity between non-categorical survey parameters and year, we calculated the Pearson's product–moment correlations.

We used the population estimates obtained from survey reports as the dependent variable to assess the population trend between 1974 and 2009. Population estimates are based on all evidence observed during the surveys (animals and fresh tracks) and eliminate suspected duplicate sightings. To assess the effect of year and the survey parameters on population estimates, we ran a backward stepwise (analysis of covariance) regression. We selected the most parsimonious model using Akaike's Information Criterion for small sample sizes (AICc; Burnham and Anderson 2002), including only main effects to avoid over-parameterization (Crawley 2009). When residuals were not normal or were heteroscedastic, we used a generalized linear model with a Poisson distribution and log-link function. The Poisson model with population estimates was not over-dispersed (z = − 0.72, P = 0.76).

To determine how recruitment changed over time, we calculated the annual ratio of calves/100 females (calf:100 females, henceforth referred to as the calf:female ratio) and ran a backward stepwise (analysis of covariance) regression to determine whether any survey parameters affected the number of calves observed, despite evidence that the calf:female ratio is unaffected by variation in survey parameters (Shupe and Beasom 1987). We used the calf:female ratio because surveys in different years varied in their timing and this metric is independent to changes in the herds' adult sex ratio over the winter than calf:adult ratios (Gunn et al. 2005). Because of issues with non-normality, heteroscedasticity, and over-dispersion (z = 3.68, P < 0.001), we ran a generalized linear model with a quasipoisson distribution for the most parsimonious model. To determine whether caribou winter range size within the park was changing over time, we calculated the distance between the 2 farthest points of caribou evidence (animals, tracks, and scat) observed during the aerial caribou surveys for every year. The ground surveys from Otter Island were included for 1975 because it was not surveyed aerially that year. For 1975 and 1977, a portion of the location data was available as areas instead of points. Some individual animal sightings were combined with the wintering areas to run the analysis. For these years, centroids for the caribou wintering areas were created and the distance was measured between centroids. We ran a backward stepwise (analysis of covariance) regression using AICc as a selection criterion to determine whether survey parameters affected the range size. Given that the rate of decline of the winter range size in relation to time appeared to change partway through the data set, we ran a non-parametric generalized additive model with distance as the dependent variable and year as the smoothed independent variable. Then, using the “break” function in R, we ran a linear piecewise regression of distance by year.


Twenty-five aerial surveys were completed over 37 years (1972–2009). In addition to the difference in flight pattern, the average temperature, snow depth, observer experience, parea (distance flown), aircraft speed, altitude and timing of the survey varied among years, and all were poorly to moderately correlated with year (Table 1). In the principal components analysis of speed and altitude, the first principal component explained 85.7% of the variation among surveys and was negatively correlated with both speed (r2 = 0.86) and altitude (r2 = 0.86).

Table 1. Range, means, and standard deviations (SD) of survey parameters during woodland caribou (Rangifer tarandus caribou) aerial surveys conducted at Pukaskwa National Park, Ontario, Canada, from 1974 to 2009. Correlation coefficients (r), degrees of freedom (df), t- and P-values between year and survey parameters are also presented
  1. aSignificant result (P < 0.05).
Mean temp (°C)−17−6−10.82.9−0.3720−1.780.09
Snow depth (cm)269751.219.90.01190.0630.95
Speed/altitude index    −0.4319−2.080.052
Speed (km/hr)89130116.39.6    
Altitude (m)6820097.933.8    
Observer experience (yr)
Distance (km)185996477.9210.4−0.1319−0.570.57
Julian start date−29774321.1−0.0420−0.170.87

The most parsimonious model using population estimates included year, the speed/altitude index, and distance: caribou declined by 3.7%/year (estimate = − 0.037, SE = 0.007, t17 = − 5.71, P < 0.001; Fig. Fig. 2). Population estimates were higher when less distance was flown (estimate = −0.001, SE = 0.0003, t17 = −3.03, P = 0.002) and at higher speeds and altitudes (estimate = 0.15, SE = 0.05, t17 = 3.1, P = 0.002). All plausible models (ΔAICc < 7; Burnham and Anderson 2002) included year (Table 2). Temperature, crew experience, and snow depth were included in at least one of the other plausible models.

Figure 2.

Results of aerial surveys conducted for woodland caribou (Rangifer tarandus caribou) between 1974 and 2009 in Pukaskwa National Park, Ontario, Canada: (a) estimated number of caribou (n = 22 surveys), and (b) the number of calves/100 females in the population (n = 20 surveys).

Table 2. The model structure, number of parameters (k), deviances, Akaike's Information Criterion values corrected for small sample sizes (AICc), and ΔAICc values from the most parsimonious model in each category (ΔAICc = 0). Variables considered include year, a speed/altitude index (SpAlt), crew experience (CrewExp), distance flow, temperature (Temp), flight pattern, and the Julian date the survey began (JulianStart). All the plausible models (ΔAICc < 7) considered in backward stepwise regressions for the effects of survey parameters on population estimates, calf:female ratios, and range size of woodland caribou (Rangifer tarandus caribou) from aerial surveys in Pukaskwa National Park, Ontario, Canada, from 1974 to 2009 are shown. Models containing only the intercept are denoted by 1
Dependent variableModel structurekDevianceAICcΔAICc
Population estimatesYear + SpAlt + Distance313.599.20.0
Year + SpAlt + CrewExp + Distance410.7100.41.3
Year + SpAlt + Distance + Temp411.7101.42.2
Year + SpAlt + CrewExp + Distance + Temp58.1102.83.6
Year + CrewExp + Distance + Temp413.7103.44.3
Year + SpAlt + CrewExp318.5104.14.9
Year + SpAlt + CrewExp + Distance + Snow510.3105.05.9
Year + SpAlt + Distance + Temp + Snow511.3106.06.9
Calf:100 femalesYear + Distance21,577.3136.60.0
Year + Distance + Snow31,381.1138.51.9
Year + Snow21,792.9138.82.2
Year + Distance + Temp31,555.2140.53.9
Distance + Snow22,080.2141.34.7
Year + Distance + Temp + Snow41,284.5142.25.6
Year + Temp + Snow31,778.6142.86.2
Year + SpAlt + Distance + Snow41,367.3143.26.6
Range sizeYear12,911.7157.10.0
Year + CrewExp22,876.0160.23.0
Year + Pattern22,518.5161.44.3
Year + CrewExp + JulianStart32,875.4163.96.8

For the calf:female ratio, 73% of plausible models included year. The model with the lowest AICc included year and distance (Table 2). The calf:female ratio declined over time (estimate = −0.047, SE = 0.022, t = −2.17, P = 0.046; Fig. Fig. 2) but did not change with the distance surveyed (estimate = −0.001, SE = 0.0008, t = −1.18, P = 0.25). For the duration of the surveys, the mean number ± SE of calves was 10.7 ± 2.1, though no calves were observed from 2003 to 2009. Snow depth, temperature, and the speed/altitude index were also included in at least one other plausible model.

The size of the winter range of caribou within Pukaskwa NP shrank in the 1990s and 2000s, concentrating around Oiseau Bay and Otter Cove (Fig. Fig. 3). Of the plausible models, 80% included year as an explanatory variable (Table 2). The model with the lowest AICc only included year as an independent variable. Year explained 33.6% of the variation in range size given the generalized additive model parameters (r2 = 0.28, generalized cross validation score = 134.2, F = 4.07, estimated degrees of freedom = 2.01, estimated residual degrees of freedom = 2.52, P = 0.02). From the generalized additive model analysis, range size appeared to be stable initially and then to begin declining in the late 1980s. The break during this time period occurred at 1987 (r2 = 0.38, F[3,21] = 4.63, P = 0.02; Fig. Fig. 4) and the winter range size declined for 22 years thereafter at a rate of 1.51 km/year (F[1,10] = 6.013, P = 0.03). Crew experience, flight pattern, and Julian start date were included in at least one of the other plausible models.

Figure 3.

Woodland caribou (Rangifer tarandus caribou) observations (animals, tracks, and scat) within Pukaskwa National Park (Ontario, Canada) from winter aerial surveys for (a) 1972–1979, n = 8 surveys, (b) 1980–1989, n = 7, (c) 1990–1999, n = 5, (d) 2000–2009, n = 5.

Figure 4.

Piecewise regression, showing a break in 1987, of the furthest distance between signs of woodland caribou (Rangifer tarandus caribou) evidence (sightings, tracks, scat) recorded per year during aerial surveys (n = 25) conducted between 1972 and 2009 in Pukaskwa National Park, Ontario, Canada.


Despite inconsistencies in survey methodology over the span of this study, there is substantial evidence indicating there has been a temporal decline in the Pukaskwa NP caribou population. Population estimates have declined over time, recruitment has been consistently below the viability threshold and at zero for the past decade, and the winter range size has been reduced. Since 1974, the annual rate of decline, based on the population estimate, has been 3.7%/year.

Detectability in aerial surveys can range from 26.0% (Farnell and Gauthier 1988) to 80.0% (Thomas 1998) and may be affected by factors such as vegetation cover (Thomas 1998), type of aircraft (Fleming and Tracey 2008), observer experience (Thomas 1998), time of day and year (VanderWal et al. 2011), altitude (Shupe and Beasom 1987, Graham and Bell 1989), speed (Hone 1986, Shupe and Beasom 1987), temperature (Bleich et al. 2001), and group size (Samuel and Pollock 1981). In the most parsimonious model in this study, only the speed/altitude index and the distance flown affected the population estimates. Higher population estimates occurred in surveys conducted at higher average speeds and altitudes. This contrasts with findings that higher speeds and altitudes cause observers to miss animals (Caughley 1974, Shupe and Beasom 1987, Graham and Bell 1989). However, as altitude increases, the number of items obscuring target animals and the required eye movement both decrease, which could increase detectability (Caughley 1974). Alternatively, caribou may alter their behavior with varying flight altitude and speed as a response to noise disturbance (Reimers and Colman 2006), thereby altering their probability of detection. Population estimates also decreased with distance flown (parea), which might indicate a trade-off between effort and observer fatigue (Reilly and van Hensbergen 2002). If the minimum distance flown was sufficient to detect most visible animals, and observers missed relatively more animals later in long surveys because of fatigue, then counts may have been lower than expected in longer surveys. It is also possible that observers covered a greater distance in relatively low caribou years before the standardization of the protocol, hoping to find more animals. Based on these results, future research using aerial survey techniques should be cognizant that speed, altitude, and total distance flown may affect the ability of observers to detect animals; however, this does not imply that other survey parameters should not be considered in an analysis of aerial survey results. In fact, nearly every parameter considered was included in at least one of the plausible models for population estimates, calf:female ratios, and/or winter range size. This underlines the importance of recording all information during a survey that could influence the results, to change these parameters as little as possible, and to include them in analysis.

The calf:female ratio declined significantly over time, averaging 14 calves/100 females and has remained at zero in the past decade. Recruitment drives ungulate population growth (Gaillard et al. 1998, 2000; Harris et al. 2008), and population declines occur below 10–15% recruitment (Bergerud 1974, Bergerud and Elliot 1986, Thomas and Gray 2002). According to Environment Canada (2008), a ratio of 28.9 calves/100 females is required for the population to remain stable. The majority of surveys (18 of 20 that reported both no. of females and no. of calves) had recruitment below this threshold, and no surveys since 1981 have met this recruitment target. Recruitment represents a combination of fecundity and calf survival and may be influenced by weather (Hegel et al. 2010), snow depth (Bergerud and Page 1987, Hebblewhite 2005), wolf (Canis lupus) predation (Farnell and McDonald 1988, Whitten et al. 1992), black bear (Ursus americanus) predation (Pinard et al. 2012), and summer nutritional conditions (Brown et al. 2007). Observed age class ratios may be biased because of observer error (Bender et al. 2003) or behavioral differences between females with and without calves (Thompson 1979). However, the persistent low recruitment in this population, including a lack of recruitment in the past decade, is consistent with and likely contributed to the observed decline of the caribou population in Pukaskwa NP. In the most parsimonious model, no parameters other than year influenced calf:female ratios, suggesting that the detectability of calves in a group is independent of survey parameters.

In addition to declining population estimates and recruitment, Pukaskwa NP caribou became more restricted in their winter range size and occupancy over the course of this study. Although caribou in northwestern Ontario can be flexible in choosing winter areas year-to-year (Ferguson and Elkie 2004), caribou are also known to reduce individual movements and group together during winter (Stuart-Smith et al. 1997). As the caribou population declined, remaining animals appeared to group on or around Otter Island and One Lake Island, 2 of the larger near-shore islands along Pukaskwa NP's coast. These 2 areas were consistently occupied across surveys, and typically with multiple individuals (Bergerud 1985, Neale 2000). This finding corroborates the “melting range” hypothesis where, in declining populations, there is a contraction of range toward high-abundance areas (Lawton 1993, Brown 1995). Early studies demonstrate near-shore islands of Lake Superior to be important refugia for mainland caribou (Bergerud 1985, 1988), which may explain the persistence of caribou in these locations. In the most parsimonious model, no factor other than year affected range size, suggesting that the survey parameters did not affect the breadth of the area within which caribou were found in Pukaskwa NP.

When inconsistencies in methodology exist, they make it difficult to determine whether a detected temporal trend in population size is real or if it is an artifact of the varying methodology (Oakley et al. 2003, Lovett et al. 2007). However, this study suggests that temporal trends in aerial survey data with inconsistent survey methods may still be disentangled by statistically controlling for other factors influencing detectability. A long-term data set coupled with analyses of contributing factors provided evidence that the size of Pukaskwa NP's caribou herd has been declining over a ≥30-year period.


This study demonstrates that effects of methodological differences may be adequately accounted for when analyzing data sets to characterize temporal patterns. However, we caution researchers considering making changes to surveys that have been in place for many years because the process to parcel out effects is time-consuming, complex, and may not be as defensible if multiple dependent variables (cumulative evidence) are not considered. Should changes need to be made to methodologies, a detectability estimate for each method (to serve as a correction factor between survey methods) would facilitate analysis. Finally, we advise minimizing any changes to methodologies in long-term monitoring because clarifying the results can be complex and time-consuming, and can waste precious conservation resources available for a species in decline that has little time left to spare.


We would like to acknowledge Parks Canada for supporting this study and the many dedicated surveyors, researchers, and pilots over the years. S. Johns assisted with earlier drafts, and S. Hayes, K. Prior, E. Gonzales, and an anonymous reviewer provided very thorough and helpful comments.