2018 status Kangerlussuaq-Sisimiut caribou West Greenland. Technical Report No. 117

1 2018 Status Kangerlussuaq-Sisimiut caribou West Greenland Technical Report No. 117, 2021 Greenland Institute of Natural Resources 2 Title: 2018 status Kangerlussuaq-Sisimiut caribou West Greenland Authors: Christine Cuyler1, Tiago A. Marques2, Iúri J.F. Correia3, Beatriz C. Afonso3, Aslak Jensen4, Peter Hegelund1 and Jukka Wagnholt1 1 Greenland Institute of Natural Resources, P.O. Box 570, 3900–Nuuk, Greenland 2 CREEM, University of St Andrews, School of Mathematics and Statistics, Scotland 3 University of Lisbon, Faculty of Sciences, Portugal 4 Solviaq 15, lej. 203, 3900–Nuuk, Greenland Series: Technical Report No. 117, 2021 Date of publication: 11 May 2021 Publisher: Greenland Institute of Natural Resources Financial support: Greenland Institute of Natural Resources Cover photo: Aslak Jensen: Three polled caribou cows and a male calf, with top of one prong visible. ISBN: 978-87972977-0-4 ISSN: 1397-3657 EAN: 97887972977054 Cited as: Cuyler, C., Marques, T.A., Correia, I.J.F., Afonso, B.C., Jensen, A., Hegelund, P. & Wagnholt, J. 2021. 2018 status Kangerlussuaq-Sisimiut caribou, West Greenland. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 117. 79 pp. Contact address: The report is only available in electronic format. PDF-file copies can be downloaded at this homepage: http://www.natur.gl/publikationer/tekniske rapporter Greenland Institute of Natural Resources P.O. Box 570 DK-3900 Nuuk Greenland Phone: +299 36 12 00 Fax: +299 36 12 12 E-mail: [email protected] www.natur.gl 3 2018 status Kangerlussuaq-Sisimiut caribou, West Greenland By Christine Cuyler1, Tiago A. Marques2, Iúri J.F. Correia3, Beatriz C. Afonso3, Aslak Jensen4, Peter Hegelund1 and Jukka Wagnholt1 1Greenland Institute of Natural Resources, P.O. Box 570, 3900–Nuuk, Greenland 2CREEM, University of St Andrews, School of Mathematics and Statistics, Scotland 3University of Lisbon, Faculty of Sciences, Portugal 4Solviaq 15, lej. 203, 3900–Nuuk, Greenland Technical Report No. 117, 2021 Greenland Institute of Natural Resources 4 [Empty page] 5 Table of Contents Summary...................................................................................... 7 Resume (Dansk).......................................................................... 9 Eqikkaaneq (Kalaallisut) ....................................................... 12 Introduction ............................................................................ 127 Methods ..................................................................................... 20 Results ....................................................................................... 27 Discussion ................................................................................. 45 Acknowledgements .................................................................. 50 Literature cited ......................................................................... 50 Figures 1. Borders of the North region containing KS caribou… Page 17 2. Area covered by 2018 caribou survey of the North region (23,303 km2) … Page 23 3. The 19 line transects used in the 2018 caribou survey of the North region… Page 24 4. KS caribou survey 2018: exploratory analysis plots… Page 29 5. KS caribou survey 2018: exploratory analysis: group size distribution… Page 30 6. KS caribou survey 2018: exploratory analysis for caribou encounter rate Page 30 7. Observer effect: histograms: detected distances for the two observers Page 31 8. Histogram of two binning options for the caribou distance data… Page 31 9. Relationship between group size and observed distances… Page 33 10. The detected distances with the estimated detection function overlaid… Page 34 11. Estimated probabilities of detection for each observed group size… Page 36 12. Relative distribution of caribou numbers along the line transects… Page 37 13. Caribou density estimates with corresponding confidence intervals… Page 38 14. Plot sampling grid example of total area A divided into smaller plots… Page 54 15. Example of a patch of tundra with the transect in the middle… Page 57 16. Half-normal (top row) and hazard-rate (bottom row) detection functions… Page 59 17. Possible shapes for the detection function when cosine adjustments are…. Page 60 18. A good model for the detection function should have a shoulder… Page 63 19. Rugged terrain with good sunlit conditions and excellent almost complete… Page 69 20. Portions of line transects flown across high elevations, Sisimiut South… Page 70 21. Flat light combined with ground showing through thin layer of snow. Page 71 22. Vegetation poking through thin snow layer in flat light or with shadows. Page 72 23. Thin snow layer with ground showing through, or rocks and vegetation. Page 73 24. Thin snow layer with vegetation showing through combined with… Page 74 25. Dead ground to the left of a transect line Page 75 26. Fog with flat light and ground showing through thin snow layer. Page 75 27. Thin snow layer combined with grounds showing through and in flat light. Page 76 28. Variable snow cover in rugged terrain, with and without shadows… Page 77 6 Tables 1. Late winter population parameters, KS caribou, 1993-2010 Page 18 2. Summary of unprocessed results… Page 27 3. The 2018 caribou survey observations lacking recorded distances… Page 28 4. Summary of the coefficient characteristics of the GLM … Page 33 5. Model comparison across three Conventional Distance Sampling models… Page 35 6. Detection function parameters’ estimates. Page 35 7. Encounter rate estimates per sub-area (stratum) for caribou groups… Page 36 8. KS caribou abundance estimates and densities in the North region… Page 38 9. KS caribou movement, or lack thereof, in reaction to helicopter… Page 39 10. KS caribou details movement, or lack thereof, in reaction to helicopter… Page 40 11. Demographics for KS caribou, North region. March 2018. Page 41 12. Group size relative to composition from demographics, KS caribou… Page 43 13. Approximate elevations for caribou groups observed… Page 44 14. Commonly used key functions & series expansions for detection function Page 58 15. Population estimates & minimum counts caribou in Greenland, 1977-2018 Page 78 Appendices 1. Statistical methods behind Distance Sampling Page 54 2. Distance Sampling Assumptions – short summary Page 67 3. Recommendations for improving future surveys Page 68 4. Photographs of KS caribou survey conditions March 2018 Page 69 5. Past and recent Greenland caribou population estimates & minimum count Page 78 Raw data may be accessed by contacting the Greenland Institute of Natural Resources, Department of Mammals and Birds. 7 Summary West Greenland (south of 69°N) has six caribou (Rangifer tarandus) regions that contain several distinct populations. This report presents new information, from a survey carried out in 2018, about the KangerlussuaqSisimiut (KS) population, which inhabits the North region. The KS caribou were last surveyed in March 2010. Since then, there have been long autumn hunting seasons of unlimited harvest, as well as a winter season. A new estimate of abundance was overdue. Helicopter surveys in 2000, 2005 and 2010 used strip transect counts. In March 2018, helicopter was again used, and for the first time Distance Sampling methods and analyses were applied. Previous surveys have documented that Greenland caribou are extraordinarily camouflaged against typical environmental conditions in Greenland, and how this could reduce detection of caribou present within the surveyed area. While almost anyone can detect running animals, stationary animals can be difficult to detect. To investigate the proportion of non-moving caribou, the 2018 survey recorded caribou flight responses or lack thereof for every group observed. Flight movement was absent in almost 32% of all caribou groups observed during the survey. This underlines the importance of skilled observers, as well as flying low and slow to make detection of caribou easier. The 2018 survey’s Distance Sampling methods and analyses corrected for undetected caribou and provided a robust estimate for caribou abundance and density (below). It is reasonable to expect that any survey for caribou would have some proportion of non-moving caribou present in the surveyed area of the line transects. Additional results from two Greenland caribou surveys completed in 2019, will confirm whether the observed proportion in 2018, almost 1/3 non-moving caribou groups, is atypical or typical. If typical, this suggests that a survey dataset including few observations of stationary caribou groups would underestimate population size correspondingly. In early March 2018, observed KS caribou were at relatively low elevations, mean 361 m. A high proportion of polled KS cows was observed, 46%, which is similar to earlier reports for this population. Polled cows are not likely due to poor body condition, as is the common assumption for populations elsewhere. For KS, polled cows may be the result of a reduced need for the dominance conferred by antlers, given their small group sizes, and the xeric 8 climate reducing the need of competing for feeding craters dug down through deep snow to obtain forage. The March 2018 demographics were improved relative to observed in 2005 and 2010. Specifically, late winter calf (age ≤ 10-months) percentage was ca. 21.8%, and calf recruitment was ca. 42 calves per 100 cows. However, 26-32% of all calves were likely orphans without dams. This suggests that the true late winter value for calf percentage was closer to 17% and recruitment ca. 31 calves per 100 cows. This level of recruitment is higher than the observed in 2005 and 2010. The March 2018 sex ratio was ca. 51 bulls per 100 cows. The March 2018 demographics describe a caribou population that appears capable of withstanding current harvests, while the calf recruitment is not high enough to suggest the possibility of rapid population growth. Stochastic catastrophic events excepted, there appears to be a low risk for future population decline, while there is potential for slow growth. For March 2018, survey coverage was 10.6% of the study area, which is a substantial improvement from the 1% coverage of the 2000-2010 strip transect count surveys. The North region’s 2018 KS caribou population abundance was estimated at ca. 60,469 caribou (95% CI: 51,932–70,410; CV = 0.074; SE = 4,501), with a density of ca. 2.59 caribou/km2 (95% CI: 2.23–3.02). This Distance Sampling estimate was precise (CV = 7.4%). The population estimate is ca. 38.5% lower than the estimated number of KS caribou in 2010. Before concluding that a large decline has occurred, caution is needed because several mitigating factors must be recognized. The 2010 survey had low coverage and a high Coefficient of Variance (CV). Thus, it was likely not as accurate or precise as the 2018 survey. Also, better GIS mapping in 2018 resulted in a smaller total area, which means that the 2010 estimate was inflated. Furthermore, in 2018 survey methods changed to Distance Sampling. This by itself precludes trend projections based on just the current and the 2010 strip transect count surveys. To predict a somewhat reliable population trend, a time series of at least three estimates is needed and these must be obtained with comparable methods. Albeit the 2018 Distance Sampling estimate of ca. 60,469 caribou suggests decline in KS caribou abundance and some decline could be expected, given both the poor calf recruitment of the 2005-2010 period and over a decade of harvest management aimed at reducing KS caribou abundance. Regardless, this report’s good late winter calf recruitment for 2018 does not support future decline. Instead, it suggests possible stability or slow growth in future. It is also worth mentioning that an 9 alternate Model-based analysis of the 2018 dataset estimated a somewhat higher 73,895 caribou (95% CI: 65,983-82,757, CV = 0.037) (Correia 2020). Given there are two estimates begs the question, which is the most accurate and precise? This is currently being investigated, requires additional results from two other West Greenland caribou surveys completed in 2019, and conclusions regarding Distance Sampling and Model-based estimates will be published in a peer-reviewed journal. Whether 60,469 (95% CI: 51,932–70,410) or 73,895 (95% CI: 65,983-82,757), the 2018 KS caribou population size remains large relative to the area available, 23,303 km2. The KS caribou density from Distance Sampling was 2.6 caribou per km2. Given good calf recruitment, population decline is not expected in the immediate future. Like all estimates since 2000, the 2018 density exceeds the recommended management target of 1.2 caribou per km2. Exceeding the target density was assumed to raise risk of overgrazing and lead to abundance decline. In Alaska and Canada, when overgrazing played a major role, caribou declines took place over 15 to 20 years. Nevertheless, even after almost two decades of high densities exceeding the target have passed, there is no strong evidence of extensive overgrazing or decline in the KS caribou. Since in 2018 recruitment improved to at least 31 calves per 100 cows, despite an overall density of 2.6 caribou per km2, it appears that the North region can support a higher density than expected. Pending additional results from two other West Greenland caribou surveys completed in 2019, the target density for caribou management will receive re-evaluation regarding what level is compatible with demographics that facilitate sustainable populations and harvests in Greenland. Resume (Dansk) Vestgrønland (syd for 69°N) har seks regioner med rensdyr (Rangifer tarandus), der indeholder flere forskellige populationer. Denne rapport præsenterer nye oplysninger fra en undersøgelse, der blev udført i 2018, om Kangerlussuaq-Sisimiut (KS)-bestanden, i Nordregionen. KS-rensdyr blev sidst undersøgt i marts 2010. Siden da har der været lange efterårsjagtperioder med ubegrænset fangst samt en vintersæson. Et nyt skøn over bestandsstørrelse var på høje tid. Helikopterundersøgelser i 2000, 2005 og 2010 anvendte striptransekter til optælling. I marts 2018 blev der igen 16 periuseq sorleq tutsuiginaateqarnerunersoq. Apeqqut akissutissarsiniarlugu maannakkorpiaq tamannarpiaq ulapputigineqarpoq, tassa Kalaallit Nunaata kitaaani 2019-imi kisitsinerit marluk suliarineqarmata. Taakkunannga inernerusut ilisimatuussutsikkut allaasereriarlugit saqqummiunneqarumaarput, taakku aamma Distance Sampling naapertorlugu suliarineqarput. KS-imi tuttoqassuseq nunap tuttoqarfittut aqutsiveqarfiusup 23.303 km2 angissuseqarnera eqqarsaatigalugu tuttut amerlassusiat 60.469-galuarpata (95 % CI: 51.932 – 70.410; CV = 0,074; SE = 4.501) imaluunniit 73.895-uppata (95 % CI: 65.983 – 82.757; CV = 0,037) nuna uumaffigisaat amerlassusiannut sanilliullugu mikivallaarpoq. Distance Sampling atorlugu naatsorsuusiortitsinermi KS-imi tuttut 2018-imi eqimassusiat kvadratkilometer-imut 2,6-iuvoq. Siunissami qanittumi ikiliartulernissaat naatsorsuutigineqanngilaq piaqqiorluaqimmatami. Tuttut eqimassusiannik naatsorsuisarnerit 2000-miilli ingerlanneqartalernikuusut assigalugit 2018-imi tuttut eqimassusiannik kisitsit pissarsiarineqartoq aamma Aqutsinikkut anguniagaagaluartoq tuttut kvadratkilometer-imut 1,2-junissaannik aaliangiussaq qaangerneqartuaannavinneqarsimavoq. Killiliussap qaangerneqartuaannarnera peqqutaalluni nunap neriniarfiusartup naggorlutsinnissaa naggataagullu allaat tuttoqassutip appariartulerneranik kinguneqarsinnaaneranik pisoqarsinnaanera erseqqissaatigineqassaaq. Tuttoqarpallaalernera peqqutaalluni nunap neriniarfiusartup aserugaaneranik assersuutissat pipput Canadami Alaskamilu naggorlutsikkiartuaarnera ilutigalugu tuttut ukiut 15-it 20-llu akornanni ikiliartuaarsimammata. Taamaattorli KS-imi tuttoqatigiiaat ukiuni qulikkaani marlunni amerlavallaarsimagaluarlutik nunap naggorlutsinneranik uppernarsaatitaqartunik paasititsisoqarneq ajorpoq, taamaattumillu oqartoqartariaqarpoq tuttoqarfiup Avannaata tuttut taama amerlatigisut ”nerisaqartinnissaat – uumatinniarnissaat” naatsorsuutaanngikkaluartumik nappassinnaasimavaa. Kalaallit Nunaata kitaani tuttunik 2019-imi kisitsinerit marluk suliarineqartut naammassinissaasa tungaanut, Aqutsinikkut tuttut annerpaamik eqimassusissaannik anguniakkap, tassa tuttut kvadratkilometer-imut 1,2 – junissaannik aaliangiussap naliliiffigeqqinnissaa eqqarsaatigineqarsinnaavoq. Nunap saannaa isiginiarlugu killissarititaasup sumiinnissaa pitsaanerpaaq, piniagaanerallu eqqarsaatigalugit piujuartitsiniarnerpaaq ujartorneqartariaqarmat. 17 Introduction In Greenland, most caribou occur along the central to southwest coast, which is locally known as West Greenland, coastal area south of ca. 76° Lat. In 20002001, the Greenland Institute of Natural Resources (GINR) gave the names North, Central and South, to the regions in West Greenland that together contain several distinct caribou populations (Jepsen et al. 2002). This report focuses on the North region and Kangerlussuaq-Sisimiut (KS) caribou population (Fig. 1). The North region lies between Sukkertoppen (66° N) and Nordre Strømfjord (68° N) and corresponds with the Greenland government’s caribou management hunting area 2 and is within the Qeqqata municipality of West Greenland. Within the North region is the coastal city of Sisimiut, and on the eastern side of region near the inland Ice Cap is the country’s largest international civil airport, Kangerlussuaq, also previously known as Søndre Strømfjord airport. Like elsewhere in Greenland, the KS caribou are a financial resource for local hunters, both professional and recreational, as well as for the service industries associated with boating and outdoors. Monetary profits aside, caribou are the prized game meat served at all major family events and many official ones. Caribou are undeniably intrinsic to the hunting traditions and culture of all the communities in West Greenland. Figure 1. North region borders (caribou management hunting area 2) containing the KangerlussuaqSisimiut caribou population. Elevations over 200m are in light yellow, below 200m are green. 18 Since 1977, the KS population has been monitored for abundance with debatable results that were often invalidated by harvest data (Born et al 1998, Cuyler et al. 2005, Cuyler 2007). In the period 1993-1996, surveys employed fixed-wing aircraft, high altitude, high speed, long systematic transects, disregarded observer fatigue and were unable to maintain a constant altitude over Greenland’s mountains and rugged terrain. The surveys of 1993 and 1996 resulted in late winter pre-calving population estimates of ca. 3,788 and 7,727 respectively (Table 1, Ydemann & Pedersen 1999 unpublished report to GINR). Despite Distance Sampling analyses of the data set, these were underestimates (Cuyler et al. 2005; Cuyler 2007) caused by undetected caribou at all distances, specifically on the transect’s 0-line (centreline of strip flown). The latter violates the primary assumption of Distance Sampling, i.e., all animals/objects-of-interest on the transect’s 0-line are detected. To increase detection of caribou, beginning in 2000, aerial surveys employed helicopter flying slowly at low constant altitude and short length random transect lines with a narrow strip width while avoiding solar glare and considering observer fatigue. Consequently, more animals present on the transects flown were detected. The new logistics of the post-2000 survey optimized sample size, variance, detectability, observer concentration and with the result that estimates of population size far exceeded pre-2000 estimates. For the North region helicopter surveys were repeated in 2005 and 2010. Further background as well as detailed descriptions of design and methods for the aerial surveys from 2000 to 2010 follow Cuyler et al. (2002, 2003, 2005 & 2011). Since 2000, the KS caribou (Rangifer tarandus groenlandicus) population has been documented as the largest in Greenland (Cuyler et al. 2011). Table 1. Late winter population parameters of the Kangerlussuaq-Sisimiut caribou population of the North region, West Greenland, taken from aerial surveys of 1993 to 2010 (Cuyler et al. 2002, 2005, 2011; Ydemann & Pedersen 1999 unpublished). Parameter 1993 1996 2000 2005 2010 Population size estimate 3,788 7,727 51,600 90,464 98,300 90% Confidence Interval (CI) – lower - - 40,400 70,276 71,500 90% Confidence Interval (CI) – upper - - 62,800 113,613 132,400 Coefficient of Variance (CV) - - - - 0.19 Standard Error (SE) - - - - 18,500 Mean group size  SD 1.95  0.33 2.5  0.48 2.8 4.63  3.4 2.96  2.14 Max group size - - 17 17 17 Density per sq km 0.16 0.33 1.2 to 2.8 2 to 6 2 to 7 Calf percentage 1.3 % 17.2 % 26.6 % 11 % 15.2 % Recruitment (Calf /100 Cow) - - 68 16.2 27.7 Sex ratio (Bull /100 Cow) * - - 87 33 54 *Age classes; calves (age ≤ 10-months), adults (age > 1-year) 19 Survey methods altered somewhat from 2000 to 2005 (Cuyler et al. 2005). Further, the 2000 survey had significantly (p < 0.001) lower caribou detection on one side of every transect (Cuyler et al. 2002). The former makes it impossible to directly compare the 2000 and 2005 abundance and density values and the latter suggests the possibility that the 2000 survey underestimated both. The abundance and density of KS caribou appears to have remained unchanged in the 2005-2010 period (Table 1). Simultaneously, calf percentage and recruitment increased after an initial drop in the 20002005 period. Given the large number and relatively high density of KS caribou in the 20002010 period, density-dependent forage limitation was considered a risk, which could cause population size instability and possibly decline. Therefore, wildlife management aimed at reducing caribou abundance and density to a target stocking rate of 1.2 caribou per sq km (Cuyler et al. 2007). The target density was based on studies elsewhere that document associations between observed densities and changes in 1) caribou productivity, 2) dispersal, and 3) condition of the range, as described in Cuyler et al. (2007). Therefore, initially there were higher quotas followed by unlimited harvests. The autumn season, which was originally 1-month was lengthened several times over the years. A winter hunting season with quota was added and it became permissible to harvest all sexes and ages. Details are available in Cuyler et al. (2016). Despite these management measures, by 2010 there was no reduction in KS abundance or density. Today, unlimited autumn harvesting continues, and in 2019 the autumn hunting season became the longest ever (01 August – 31 December). Meanwhile, the winter harvest was recently discontinued, however, not prior to the March 2018 survey. Present survey The international network of caribou knowledge holders, CARMA (Circumpolar Rangifer Monitoring & Assessment network), advises monitoring caribou population abundance every three years. Given the last survey of the KS caribou population was in March 2010, and there have since been eight unlimited autumn harvests, had abundance, density, or demographics of the Kangerlussuaq-Sisimiut caribou population changed? In early March 2018, GINR again examined the Kangerlussuaq-Sisimiut caribou population in the North Region of West Greenland by aerial helicopter survey. The 2018 survey methods for collecting herd structure data remained unchanged from earlier surveys. The 2018 survey, however, replaced the 20 multiple short length random transect lines strip method used for KS caribou in 2000, 2005 and 2010, with systematic transect lines and Distance Sampling, in which distances from a line to animals detected are recorded and from those distances, abundance and density of animal populations are estimated (Buckland et al. 2001, Thomas et al. 2010). This report investigates the Distance Sampling data set collected during GINR’s 2018 caribou survey of the KS population in the North region. It then presents the 2018 pre-calving caribou abundance and density. Further, this report presents information on the movement or lack thereof of caribou detected. The herd structure data set is also investigated, and we report the pre-calving demographics for the KS caribou population. Note that an earlier analysis for 2018 caribou abundance and density was run on the same data (Marques 2018), however, the then known area (km2) was incorrect. Marques’ (2018) is an internal CREEM (Centre for Research into Ecological and Environmental Modelling (St. Andrews, Scotland)) report, which is available upon request. Methods Study area The North region is within the Qeqqata municipality. Although the Qeqqata municipality has a 2020 human population of ca. 9,400 not all live within the boundaries of the North region. The only large settlement within the region is the city of Sisimiut, with ca. 5,600 inhabitants, followed by the ca. 500 residing at the Kangerlussuaq international airport. Together, the hamlets of Itilleq and Sarfannguit contain a further 200-300 people. The North region is seasonally ice-free. Improvements in Geographic Information System (GIS) excluded lakes, rivers, sand, glaciers, and islands for a more accurate land area of 23,303 km2. Previous surveys reported a less precise land area of ca. 26,000 km2 (Cuyler et al. 2002, 2005, 2011). Located between 66-68° N Lat, the Arctic Circle passes through its middle. The northern border is provided by the Nordre Strømfjord, which has numerous turbulent maelstroms and thin, treacherous, incomplete winter ice. The southern border is framed by a combination of the Greenland Ice Cap, the Sukkertoppen Ice Cap, and the outer portion of the Kangerlussuaq fjord. The 21 latter is ice-free year-round and dominated by cliffs of ca. 1000 m. The western border is the permanently ice-free seacoast of the Davis Strait, and eastern border is the Greenland Ice Cap. The west coast topography is mountainous with peaks whose elevation can be 1000 to 1800 m and glaciers are common. Moving eastward the mountains gradually give way to rugged terrain generally ranging 10-900 m elevation. The wide and typically cliff sided Kangerlussuaq fjord penetrates the region stopping just short of the Greenland Ice Cap. It effectively separates the North region into two thirds above it and one third below it. In the southern third, the terrain immediately north of the Sukkertoppen Ice Cap is generally barren highlands >1000 m elevation. Moving north towards the Kangerlussuaq airport the terrain includes lowland valleys under 400 m elevation and highlands of generally under 1000 m elevation. Common to West Greenland, the North region exhibits a climate gradient on a west-east axis. The western seacoast is wet maritime; however, the climate becomes dry continental as one moves east towards the Greenland Ice Cap. Climate and weather in the west are influenced by the ice-free Davis Strait and the low-pressure oceanic storm systems that sweep in from the southwest. The climate in the inland of the North region is influenced by the Sukkertoppen Ice Cap at its southern boundary. Sukkertoppen’s elevation acts as a barrier to the oceanic storm systems above, creates a precipitation shadow on its northeastern side, and in combination with the dominating high pressure over the Greenland Ice Cap creates the inland’s xeric continental climate. Loess/sandstorms are common in the vicinity of the Greenland Ice Cap and are caused by katabatic winds (often gale force and dry) descending off the Ice Cap (Cuyler et al. 2005). The North region may be described as open or alpine tundra. At lower elevations, vegetation involves low arctic species of mainly dwarf shrub heath, which changes to predominantly steppe and grassland when moving east towards the Greenland Ice Cap (Tamstorf et al. 2005). Lichen heaths are rare, and further, higher elevations are often fell field, abrasion plateaus and bare ground (Tamstorf et al. 2005). Aside from the caribou, the only native wild mammals present in the North region are arctic hare (Lepus arcticus Rhoads) and arctic fox (Vulpes lagopus Linnaeus). Large mammalian predators are absent. In the early/mid 1960’s, 22 muskoxen (Ovibos moschatus Zimmermann), were translocated from NE Greenland to a location close to the Kangerlussuaq international airport. Muskoxen are now firmly established in the southern third of the North region. The borders of the North region are semi-permeable permitting limited animal movement between adjacent regions, i.e., Naternaq region to the north (above Nordre Strømfjord) and Central region to the south (below Sukkertoppen Ice Cap). Nevertheless, the borders are likely effective barriers preventing mass caribou movements (Linnell et al. 2000). Field methods Since 2000, early March has been the chosen period for surveys because caribou dispersion is high, group size is small with low variability and daily movement is at the annual minimum (Cuyler et al. 2007, 2011, 2016; Poole et al. 2013). The former two reduce variance among transects, diminish counting error, and maximize precision, while the latter lowers movement between or along transects. Meanwhile, snow cover is highly variable, and the terrain rugged (Appendix 4). Further, erratics (glacial boulder debris) are common and appear similar to caribou in size and colour. Singularly or in combination, these attributes of the habitat provide small groups of caribou with outstanding background camouflage and reduce detectability (Cuyler et al. 2005, 2011). Failure to detect caribou (often stationary, specifically even when on the 0-line) must be recognized as a source of negative bias (inaccuracy) for caribou surveys in Greenland. To mitigate the combinations of conditions that lower detectability of caribou, a helicopter is necessary to enable a constant altitude above ground level while flying low (40 m, ca. 120 feet) and slow (ca. 65 km/hour). The aerial survey of the KS herd occurred 01-15 March 2018 and a helicopter AS350 was the platform for observation. Participants included three observers, all with previous survey experience: GINR’s senior scientist Christine Cuyler, GINR’s project coordinator Peter Hegelund, and professional hunter Aslak Jensen (Greenland Association of Professional Hunters (KNAPK)) from Nuuk. Jensen and Hegelund were seated in the rear of the helicopter and observed animals for all distances from the side they were sitting, which alternated. Cuyler always sat in front, observed the 0-line, including distances to either side up to 100 m, and was the data recorder. Verbal contact among the observers permitted the digital audio recording of all observations. Two audio devices (SONY IC recorder, ICD-SX712) were used to record separately the observations specific to the left and right side of the line transect. Audio recording devices were on continual 23 recording for each line transect. At the end of each survey day, audio data was downloaded to computer for storage and back-up. Observations were later paired with Global Positioning System (GPS) coordinates of the helicopter at the time of observation. The audio recording included distance to, size, behaviour of each caribou group observed and name of the observer. Often flight and environmental conditions were recorded. Manual clickcounters, logging the number of caribou seen by each observer, provided lowtech back-up for the digital audio observations from each line segment. Figure 2. Area covered by the 2018 caribou survey of the North region (23,303 km2). Three different colours illustrate the three sub-areas, designated as Sisimiut (blue), Sisimiut South (orange) and Angujaartorfiup (purple). The term ’Byer bygder’ identifies the four human settlements. Survey design The 2018 survey differed in design from the 2000-2010 random transect line strip-counts. The surveyed North region area, 23,303 km2, was divided into three sub-areas, arbitrarily named Sisimiut (12,658 km2), Sisimiut-South (3,512 km2) and Angujaartorfiup (7,133 km2) (Fig. 2). The sampling design for the 2018 survey considered 19 systematic parallel line transects of variable length separated by 15 km and placed over the three sub-areas (Fig. 3). Those transects provide the maximum area coverage possible given the financial resources available. An initial line transect was computer generated at random, and the others followed 15 km apart. Aligning line transects perpendicular to known gradients within the surveyed area can maximize precision of the resulting estimate by lowering the encounter rate variance 24 (Buckland et al. 2001). Thus, the transect axis direction was chosen as perpendicular to previously known animal distribution gradients in March. Lines 1 to 13 followed a west-east axis, which also reflects the climate gradient from wet maritime to dry continental. Line transects 14 to 19 followed a north-south axis, reflecting animal, climate, and topological gradients between Sukkertoppen Ice Cap and the Kangerlussuaq International airport. Figure 3. The 19 line transects used in the 2018 caribou survey of the North region, illustrating end points, and numbering of line transects and employing the same three colours as applied to the three sub-areas in above figure 2: Sisimiut (blue), Sisimiut South (orange) and Angujaartorfiup (purple). Elevations between 0 and 200 m are pale green, while any above 200 m are pale yellow. Distance collected was the perpendicular distance from the helicopter’s flown 0-line to a caribou group (object-of-interest). A caribou group was a relatively tight aggregation of animals. Since flight response to the approach of the helicopter was common, distance collected was the distance to the center of the caribou group from the 0-line before any movement by the caribou occurred. Exact distance measurements were not possible primarily because impractical, e.g., maintaining flight speed, number of caribou groups encountered within a short time, the range finder too time consuming to use with errors occurring, plus time constraints on audio recordings, which could create confusion as to which distance applied to which group observation. Additionally, helicopter time was limited by financial constraints, which prevented stopping and flying out to individual groups before returning and continuing along the 0-line. Instead, distance measurement used the following 25 distance bins: 0, 50, 100, 200, 300, 400, 500, 750 and 1500 meters perpendicular to the line transect. These values correspond to the upper limit for a specific bin that the caribou observation was included in. For analysis, these were recoded to the mid distance for a specific bin. Note, binning accuracy relies heavily on observer ability to correctly estimate distance to the observed animals. Thus, before starting the survey the helicopter hovered at the 40m altitude used during line transects, while each observer used a Leica laser range finder 1600 to gauge distances. Then they marked their window with masking tape delineating the approximate distances for each bin. When possible while flying line transects, the laser range finders were used to double-check reported bin distances to detected caribou. Distance sampling The caribou group was the selected sample unit for the Distance Sampling analysis of the 2018 survey. Neither the individual caribou within a group, nor individual line transects were considered as the sample unit. The recorded distances to the caribou groups observed were used to estimate a detection function, then estimate the detection probability and finally to estimate the density of the caribou within the surveyed area (Buckland et al. 2001). The detection function, 𝑔(𝑦), describes the probability of detecting an object of interest (caribou group) given that it is at a distance 𝑦, from the centreline (0-line), thus being a non-increasing function of 𝑦 (Buckland et al. 2015). For line transects, 𝑦 is the perpendicular distance from the 0-line to the detected object. Within Distance Sampling methods, the probability of detection is explained recurring to these observed distances (Buckland et al. 2001). Prior to Distance Sampling analysis, the raw data was first processed for inconsistencies. Then extensive exploratory data analysis was completed, including evaluation of observed distances, before proceeding to determining the detection function through model fitting and selection, as per recommendations by several authors (Buckland et al. 2001; Marques et al. 2011; Thomas et al. 2010). To determine the detection function, several models were considered, since this is the standard approach introduced by Buckland (1992) and popularized by being available in the software Distance (Thomas et al. 2010). The model presenting the lowest AIC value was chosen. The subsequent analysis was based on Marques (2018). Details regarding Distance Sampling theory, methods and analysis are available in Buckland et al. (2001, 32 Distance Sampling analysis Before conducting any modelling, an analysis of the observed distances was made to evaluate whether any major assumption violation occurred or other data-related issue, as stated in previous sections. These analyses are from Correia (2020). The histogram of observed distances with no defined truncation distance is similar to typical Distance Sampling data, perhaps showing some over-dispersion, with not-equally-spaced bins (Fig. 8). Given the histogram of binned distances, a strip half-width of 𝑤 = 0.75 km was selected (i.e., all observations at distances beyond 750 meters were discarded). This truncation reduced the sample size from 2079 to 1640 caribou groups for the Distance Sampling analysis. Data truncation is a common procedure because otherwise extra adjustment terms may be needed to fit the long tail of the detection function. Further, little information is lost by truncation, since data observations located more than 0.75 km from each side of the line make a minimal contribution to the abundance estimate. The alternative binning Option 2, less bins, reduces the influence of potential measurement errors in the observed distances. This alternative binning option includes bin cut points of 0, 0.10, 0.30, 0.50 and 0.75 km (Fig. 8). With the original binning option, there seem to be less than expected observations on the 0.10-0.20 km and 0.30-0.40 km intervals, when compared to the 0.20-0.30 km and the 0.40-0.50 km bins. This might be evidence of heaping. This phenomenon occurs when observers tend to record some preferred values over others (Buckland et al. 2001). Here, the heaping would have occurred for distances 0.25 km and 0.50 km, which are round distances that are easily chosen in the absence of a rigorous distance measuring method. Both binning options were considered in model fitting, albeit only Option 2 minimizes the effect of measurement error induced by heaping. Since binning Option 1 was not suitable for grouping, only the analyses whose fitted models consider the second binning option are illustrated below. The advantage for choosing the second binning option is that it results in more reliable detection functions. However, owing to fewer degrees of freedom, the small number of bins affects the 𝜒2 Goodness-of-Fit tests following model fitting. A scatter plot, with a GLM fitted between two variables, observed distance as explanatory variable, and group size as response variable, suggested a faint tendency for larger groups being associated with greater distances (Fig. 9). 33 Note that the maximum group size is no longer 20 as this data set has been truncated, considering a strip width of 𝑤 = 0.75 km, therefore, the most distant observations, which corresponded to large group sizes, were excluded. This regression analysis suggests that distance is a statistically significant variable explaining group size (Table 4). Group size also seemed marginally related with the spatial coordinates (Correia 2020). Table 4. Summary of the coefficient characteristics of the GLM between observed distance and the group size while considering a Poisson distribution. Parameter Estimate Standard Error z-value p-value Intercept 0.747 0.029 26.20 0.00000 Distance 0.311 0.080 3.88 0.00011 Note: AIC = 5613.7, Null Deviance = 1361.4, Residual Deviance = 1346.4. Figure 9. Relationship between group size and observed distances and respective regression fit using GLM. Detection function models fitted with the first binning option did show poor fitting, including the best fit within this group, since these presented several adjustment terms, due to the heaping phenomena. Below, the detection functions are fitted to the data considering the second binning option. For these models, every combination of key function and adjustment terms was tested. The only additional covariates assessed were observer and group size, considering 𝑤 = 0.75 km. A summary of the information from each 34 model fitted to the data (Table 5) provides a simple overview of several models, and includes the respective key functions, adjustment terms, model formula, 𝜒2 Goodness-of-Fit test p-value, estimates of the detection probability, respective standard error (se (𝑃 a)), and Δ𝐴𝐼𝐶 comparison between each model and the model with the lowest AIC. The best model fitted to the data possesses the lowest change in AIC value (Δ𝐴𝐼𝐶 = 0). For the 2018 caribou survey data, this model has the hazard rate function as a key function, no adjustment terms added and only group size as covariate (AIC = 4414.52). The hazard rate key function was selected because it was the most flexible key. The second-best model includes the half-normal key with group size as a covariate (AIC = 4422.17, i.e., Δ𝐴𝐼𝐶 = 7.66). This strongly suggests that group size is a relevant covariate in detectability. The best fitted detection function parameters’ estimates indicate a slight positive relationship between group size and detectability, superimposed with the observed distances’ histogram (Table 6, Fig. 10). The estimated averaged probability of detection for the North region was 𝑃 a = 0.541 (se = 0.025, Table 5). Remaining detection functions and summary table are found in Correia (2020). It is an averaged estimate since group size is included in the model. Consequently, each group size has its separate detection function, corresponding to different estimates for the probability of detection (Fig. 11). Figure 10. The detected distances with the estimated detection function overlaid, considering the binning option that reduces the effect of heaping. 35 Table 5. Model comparison across the three Conventional Distance Sampling models and models considering group size and observer as covariates. Key function Formula 𝒙𝟐 p-value 𝑷 a se (𝑷 a) ∆AIC Hazard-rate Group size NA 0.541 0.025 0.000 Half-normal Group size 0.000 0.603 0.013 7.658 Half-normal with cosine adjustment terms of order 2,3 1 NA 0.512 0.026 8.582 Uniform with cosine adjustment terms of order 1,2,3 NA NA 0.513 0.025 8.582 Hazard-rate with cosine adjustment term of order 2 1 NA 0.519 0.025 8.647 Hazard-rate with simple polynomial adjustment term of order 2 1 NA 0.533 0.032 10.123 Hazard-rate Observer NA 0.544 0.025 10.672 Hazard-rate with Hermite polynomial adjustment term of order 4 1 NA 0.535 0.032 10.679 Uniform with simple polynomial adjustment terms of order 2,4,6 NA NA 0.577 0.029 15.371 Half-normal Observer 0.000 0.605 0.013 15.635 Half-normal 1 0.001 0.606 0.013 19.097 Uniform with Hermite polynomial adjustment term of order 4 NA 0.000 0.643 0.010 30.245 Note: Under Formula, explanatory variables: Group size = group size as variable, 1 = for Uniform key, NA = no explanatory variables, Observer = observer as variable. Under Chi-square p-value, NA = not enough degrees of freedom for the Goodness-of-Fit (GOF) test, thus the ‘NA’ values. (Degrees of freedom are calculated considering the model parameters and these vary considering which key function is used and how many/which explanatory variables are considered.). For each key function, all three series expansions (Cosine, Simple Polynomial, Hermite (Appendix 1, Table 14) were applied. Table 6. Detection function parameters’ estimates. Estimate Standard Error Intercept +1.681 0.154 Group size 0.152 0.048 Note: Estimates are on log scale. 36 Figure 11. Estimated probabilities of detection for each observed group size obtained with the fitted model. A group size of 2 caribou presents an estimated probability of detection of 0.533, while a group size of 10 has an estimate of 0.909 (Fig. 11). With increasing group size, the probability of detection also increases. This is consistent with the author’s intuition as larger groups are easier to detect than smaller ones. The estimates for encounter rates suggest the Sisimiut sub-area has the most caribou, since its estimate is larger than the other sub-areas (Table 7). Visualization of the detected caribou distribution shows this was almost continuous along the line transects of the Sisimiut sub-area, with a few ‘hot’ spots (Fig. 12). Meanwhile, caribou were seldom observed at elevations over 1000 m. This involved much of the Sisimiut South sub-area, south end Angujaartorfiup sub-area, and a couple of line segments in Sisimiut sub-area: line transect 6 for duration of the glaciated Qáqapalât (1200-1600 m); line transect 5 over the Akuliaruserssuaq Peninsula and associated Tugtoqarajôg (900-1500 m). Concerning the design-based estimates for caribou abundance and density, Sisimiut is also the subarea presenting more caribou (Table 8, Fig. 13). Table 7. Encounter rate estimates per sub-area (stratum) for caribou groups considering three strata, five bins, and a detection function fitted with group size as covariate. Sub-area Encounter rate Standard Error (se) Coefficient of Variance (cv) Sisimiut 1.389 0.084 0.060 Sisimiut South 0.403 0.110 0.273 Angujaartorfiup 0.559 0.085 0.152 TOTAL 0.997 0.120 0.120 37 Figure 12. Relative distribution of caribou numbers along the line transects. Smudge shading indicates fewest caribou. This darkens until becomes black, turns into purple-violet, shifting to pink and ending with yellow, which is the most caribou. Underlying map: elevations between 0 and 200 m are pale green, above 200 m are pale yellow. The 𝜒2 Goodness-of-Fit test could not be performed to the selected model because there were not enough degrees of freedom (Appendix 1: Equation (19), 𝑢 − 𝑞 − 1 = 4 − 3 − 1 = 0 degrees of freedom, observed and expected values in Correia (2020)). Additionally, the Kolmogorov-Smirnov and Cramér-von Mises tests (Appendix 1) could not be applied since the distances were represented as a discrete variable. In March 2018, the North region had an estimated population size of approximately 60,469 caribou (95% CI: 51,932 – 70,410), with a CV of 7.4% (Table 8). The latter is an exceptionally low value, which indicates relatively accurate caribou abundance estimates for 2018. The design-based density estimate for the whole survey region was 2.59 caribou per km2, with 95% CI: 2.23 – 3.02 (Table 8, Fig. 13). Further details in Correia (2020). 38 Table 8. Kangerlussuaq-Sisimiut caribou abundance estimates and densities in the North region, March 2018, considering three sub-areas (strata), five bins and a Hazard rate detection function with group size as a covariate. Sub-area Population Estimate SE CV 95% Confidence Interval Density (caribou / km2) Lower Upper Sisimiut 46,724 3,745 0.080 39,392 55,422 3.7 Sisimiut South 3,931 1,134 0.289 1,820 8,492 1.2 Angujaartorfiup 9,814 1,502 0.153 6,758 14,252 1.4 TOTAL 60,469 4,501 0.074 51,932 70,410 2.6 Note: SE = Standard Error, CV = Coefficient of Variance. Figure 13. Caribou density estimates with corresponding confidence intervals for the three sub-areas, Sisimiut, Sisimiut South and Angujaartorfiup, and finally for the total North region. Caribou detectability As noted under all helicopter surveys since 2000, detecting caribou was again difficult owing to background conditions camouflaging the caribou from view. These included incomplete or patchy snow cover, substrate (including grass, low vegetation, ground) poking or showing through thin snow layer, rocky terrain, fog, and light/shadow conditions typical to latitudes around the Arctic Circle in early March. Detecting caribou was further compromised by the west-east orientation of most lines, which ensured that on the south-facing side of the helicopter in the absence of cloud cover, the sun was in observer eyes and reflecting off the snow surface causing solar glare. Despite observers using polarized sunglasses, this intense sunlight in the eyes may have reduced detectability of caribou. The flight altitude of 40 m reduced the amount of dead ground (land blocked from view by terrain features, e.g., Appendix 4, Fig. 25), resulted in more 39 ground to search and scan over for animals. When combined with a long line length, the subjective result was a feeling of there not being enough time to scan all terrain properly, and sometimes diminished observer concentration could occur. Both could lower caribou detection given the high camouflage conditions. Finally, the helicopter windows sometimes frosted, that frost and the time for physical removal (scrapping off using credit card) could have decreased caribou detection. Additionally, low caribou group size and specifically lack of movement by the caribou made sighting them difficult. Table 9. Kangerlussuaq-Sisimiut caribou movement, or lack thereof, in reaction to helicopter flying line transect survey of North region, March 2018. The dataset for observations of caribou group size which included behaviour was n= 1880, while the dataset which included distance of the caribou group from the line transect was n = 1870. Kangerlussuaq-Sisimiut caribou Caribou Groups Exhibiting Movement Lacking Movement p – value Number of groups 1288 592 % observations 68.5% 31.5% Mean group size 2.54 2.14 < 0.0001 Confidence Level (95%) 0.106110162 0.112602502 Standard Error 0.054087919 0.057333656 Median 2 2 Mode 2 1 Standard deviation 1.941145995 1.394988056 Sample Variance 3.768047773 1.945991677 Maximum 20 10 Minimum 1 1 Distance from 0-line1 1283 587 Mean distance 539.36 m 663.80 m < 0.0001 Confidence Level (95%) 28.5451518127234 39.8522696732454 Standard Error 14.5503705 20.29116828 Median 300 500 Mode 1500 1500 Standard deviation 521.1795663 491.6161067 Sample Variance 271628.1403 241686.3964 Maximum 1,500 m 1,500 m Minimum 50 m 50 m 1 0-line is the centre of the line transect flown by helicopter. Caribou behaviour: flight reaction or lack thereof The caribou survey of 2018 was the first to use digital audio recorders to collect the observation data. The digital recorders permitted including in the dataset what, if any, was the behavioural reaction of the caribou group to the helicopter flying a line transect 40 past or over them. Behaviour could then be put in relation to group size and distance from the transect line (Table 9). There was a significant difference between the size of caribou groups that exhibited movement and those that did not, mean 2.5 and 2.1, respectively (t Stat = 5.104806; twotailed testing P < 0.0001, t = 1.961497638, df = 1548). Non-moving caribou groups averaged ca. 125 m further away from the line transect flown by the helicopter than those caribou groups showing movement (Table 9). There was a significant difference between the mean distance for groups with movement, 539.36 m, relative to the groups lacking movement, 663.80 m (t Stat = -4.983728647; two-tailed testing P < 0.0001, t = 1.961944491, df = 1199). Table 10. Kangerlussuaq-Sisimiut caribou details for movement, or lack thereof, in reaction to helicopter flying line transect survey of North region, March 2018. Dataset of observations that included caribou group size, behaviour, and distance from line transect. Dataset was n= 1,880 groups, which contained n = 4,545 individual caribou. Kangerlussuaq-Sisimiut caribou Category Groups (n = 1,880) % Individuals (n = 4,545) % Exhibiting Movement Running away 1000 53.19 2548 56.06 Running away high speed 111 5.90 345 7.59 Walking 92 4.89 188 4.14 Approach* 20 1.06 63 1.39 Confused, circling tightly 20 1.06 39 0.86 Running parallel to line transect 18 0.96 38 0.84 Running, later standing looking 17 0.96 29 0.64 Trotting away 7 0.37 14 0.31 Mixed: movement + lack of 3 0.16 13 0.29 TOTAL 1,288 68.51 3,277 72.10 Lacking Movement Standing still 459 24.41 992 21.83 Standing, later walk approach* 66 3.51 145 3.19 Lying down 32 1.70 51 1.12 Lying down, later stood up 17 0.90 39 0.86 Some lying, others standing still 10 0.53 28 0.62 Lying down, later walk movement 8 0.43 13 0.29 TOTAL 592 31.49 1,268 27.90 *Approach movement was towards the helicopter position. Caribou groups reacting to the helicopter fly-by with movement made up 68.5% of all observations. Conversely, 31.5% of all caribou groups exhibited little or no movement. The results were similar when considering the absolute number of caribou involved (Table 10). 41 Of 4,545 individual caribou, 72.1% exhibited movement and lack of movement 27.9%. Almost a third of all caribou observed lacked movement. Among the 1146 ‘running’ groups (Table 10), 939 of those groups exhibited unabated flight, i.e., they never stopped while within view of the helicopter. Group composition (sex, age) was determined for 316 of those groups and 86% were composed of cows with calves. Only 7% were bull groups (juveniles, adults), with the remaining 7% being groups of cows only, calves only, cows and juvenile bulls or adults of unknown sex. Considering only the 233 caribou groups whose original position was on or within 50 m of the 0-line, 140 of those groups (60%) never stopped running away. Six times such unabated flight began while a group was far distant on a section of the 0-line yet to be flown, e.g., 0.5 to 1.5 km ahead of the helicopter. Meanwhile, 15 groups lacked movement, although five of those groups did move once the helicopter was directly over top of them. In ten of those groups the caribou were standing and in five they were lying down. Table 11. Demographics for Kangerlussuaq-Sisimiut caribou, North region, March 2018. Parameter Kangerlussuaq-Sisimiut caribou Number of groups observed 894 Mean group size 2.45 Confidence Interval (95%) 0.1148 Standard Error 0.0585 Standard Deviation 1.75 Sample Variance 3.0605 Median group size 2 Mode group size 1 Maximum group size 12 Minimum group size 1 Original data Removed 124 orphan calves Total individuals sexed & aged (n) 2188 100 % 2064 100 % Cow (age > 1 year) 1136 51.92 % 1136 55.04 % Calves from previous spring 476 21.76 % 352 17.05 % (231 females) 10.56 % - - (228 males) 10.42 % - - Bull (age > 1 year) 576 26.33 % 576 27.91 % (231 adults, age > 3) 10.56 % (231 adults, age > 3) 11.19 % (202 juveniles, age 2½) 9.23 % (202 juveniles, age 2½) 9.79 % (143 juveniles, age 1½) 6.54 % (143 juveniles, age 1½) 6.93 % Recruitment (calves / 100 cows) 41.90 30.99 Sex ratio (Bull >1 year / Cow) 0.51 0.51 48 egged twin foetuses have occasionally been observed (Cuyler & Østergaard 2005). Nevertheless, twinning is considered uncommon in caribou (Skoog 1968, Bergerud 1969, Dauphiné 1976). Either twinning is more common than expected in the KS caribou population or the extra calves also had lost their dam. Considering the 124 orphans, their survival into March of their first winter is remarkable but not unexpected since large predators are absent in West Greenland. It indicates that food availability, quantity and quality were not a problem for KS caribou in the months leading up to March 2018. Unfortunately, the high number of orphan calves (i.e., missing cows) indicates that the calf recruitment of 42 calves per 100 cows is artificially high. One option would be to remove the orphan calves from the dataset, since these may not survive, assuming these have a higher mortality rate than calves with dams (Bergerud et al. 2008). If just the known 124 orphan calves are removed from the dataset, then the revised demographics results are cows 55%, bulls 28% and calves 17%, with a late winter calf recruitment of 31 calves per 100 cows and a bull to cow ratio of about 0.51. Even the revised calf values are better than the low 2005 and 2010 values (Tables 1, 11). The March 2018 demographics, revised or not, describe a caribou population that appears capable of withstanding current harvests, while the calf recruitment is not high enough to suggest the possibility of rapid population growth. Conversely, the 2018 calf recruitment indicates a low risk for future population decline, while there is potential for possible stability or slow growth (Bergerud et al. 2008). Still, stochastic catastrophic events could bring abrupt changes in abundance (CAFF 2021). 2018 caribou population size & density At 23,303 km2, the North region is the largest of all the caribou regions in West Greenland. Due to the high cost of helicopter time in Greenland, the 2000, 2005 and 2010 strip count surveys covered only about 1% of the North region area. In 2018 and with increased funding, Distance Sampling methods with systematic transects were adopted. This increased area coverage to 10.6% (given truncation limiting strip width to 750 m either side), which contributed to improved estimate accuracy. The 2018 KS population estimate was 60,469 caribou (95% CI: 51,932–70,410; SE = 4,501; CV = 0.074), with density of ca. 2.59 caribou/km2 (95% CI: 2.23–3.02). The Distance Sampling estimate was exceptionally precise (CV = 7.4%), specifically in regards past less precise estimates. On the surface the 2018 population estimate is ca. 38.5% lower than the estimated number of KS caribou in 2010, i.e., ca. 60,500 versus 98,300 (Tables 1, 8). Before concluding that a large decline in abundance has occurred, caution is needed because several mitigating 49 factors must be recognized. The 2010 survey had low area coverage (1%) and a high Coefficient of Variance (CV), (19%). Thus, the 2010 estimate was likely not as accurate and was certainly less precise than the 2018 survey. Also, unlike in 2018, the 2010 analyses used a less refined area, i.e., included lakes, rivers, and islands. The larger area entailed would have inflated the 2010 estimate. These may account for the lack of overlap in the confidence intervals for the 2010 and 2018 estimates. Expanding to include the 2005 survey estimate and the overlap of the confidence intervals suggests a lack of significant difference between the 2005-2010 and 2018 population sizes. Population trend can be predicted if the same methods are repeated over a time series of surveys. However, the 2018 survey adopted Distance Sampling methods and analyses to maximize estimate accuracy and precision. This change of methods precludes trend projections based on just the current and the 2010 strip transect count surveys. To predict a somewhat reliable population trend, a time series of at least three estimates is needed and these must be obtained while repeating the same methods. Albeit the 2018 Distance Sampling estimate of ca. 60,469 caribou suggests decline in KS caribou abundance and some decline could be expected given both the poor calf recruitment of the 2005-2010 period and almost two decades of harvest management aimed at reducing KS caribou abundance. Regardless, the 2018 late winter calf recruitment does not support future population decline. The 2018 Distance Sampling design-based estimate, 60,500 caribou, was based on the selected 19 line transects, which may over-represent some features within the North region, while under-representing others. In an alternate approach, Correia (2020) applied Generalized Additive Model (GAM) and Density Surface Model (DSM) to the same 2018 survey data and thus considered the entire North region. Correia’s GAM/DSM analyses resulted in a 2018 Model-based population size estimate of 73,895 (95% CI: 65,983-82,757) KS caribou, which had the exceptional CV of 0.037 (3.7%), lower than that for Distance Sampling. Given there are two estimates begs the question, which is most accurate and precise? Addressing that issue is beyond the scope of this technical report. Instead, it is currently being investigated, requires additional results from two other West Greenland caribou surveys completed in 2019, and conclusions regarding Distance Sampling and Model-based estimates will be published in a peer-reviewed journal. Given the Distance Sampling estimate of 60,500 caribou and the Model-based estimate of 73,895 presented by Correia (2020), we can be certain that despite 18 years of harvest management to the contrary, the KS caribou population size remains large in relation to 50 the area available, 23,303 km2. The overall 2018 estimate for KS caribou density was 2.6 caribou per km2. Given good calf recruitment, population decline is not expected in the immediate future. As with the 2000, 2005 and 2010 surveys, the 2018 KS caribou density exceeded the recommended management target density of 1.2 caribou per km2, above which there is assumed an increased risk of overgrazing leading to caribou decline (Kingsley & Cuyler 2002, Cuyler et al. 2007). In North America, when overgrazing played a major role, caribou declines took place over 15 to 20 years (Schaeffer et al. 2016, Soulliere & Hammel 2015). Nevertheless, even after almost two decades have passed with high densities exceeding the target, there is no strong evidence of extensive overgrazing or decline in the KS caribou. Since in 2018 recruitment improved to at least 31 calves per 100 cows, despite an overall density of 2.6 caribou per km2, it appears that range conditions in the North region support a higher density than expected. Pending additional results from two other West Greenland caribou surveys completed in 2019, the target density for caribou management will receive re-evaluation regarding what level is compatible with demographics that facilitate sustainable populations and harvests. Acknowledgements This project was financed primarily by the Greenland Government and otherwise by the Greenland Institute for Natural Resources, Nuuk Greenland. Grateful thanks go to Air Greenland Charter and their helicopter pilot Kåre Berli for his safe flying. Thanks also to the Greenland Association of Professional Hunters (KNAPK) for providing an experienced observer, excellent at spotting caribou despite poor detection conditions. We also thank Josephine Nymand for review of the manuscript, and Emma Kristensen for review of the summary. The summary was translated into Danish by Anna Haxen and to Greenlandic by Emma Kristensen. Literature cited Bergerud A.T. 1967. Management of Labrador caribou. J. Wildl. Manage. 31:621-642. Bergerud A.T. 1969. The population dynamics of Newfoundland caribou. Ph.D. Thesis. University of British Columbia. 140 pp. Bergerud A.T. 1971. The population dynamics of Newfoundland caribou. Wildl. Monogr. 25: 55 pp. Bergerud A.T. 1980. A review of the population dynamics of caribou and wild reindeer in North America. Proc. 2nd Reindeer/Caribou Symp., 556-581. Trondheim: Direktoratet for Vilt og Ferskvannsfisk. 51 Bergerud A.T., Luttich S.N. & Camps L. 2008. The return of caribou to Ungava.. McGill-Queen’s University Press, Montreal & Kingston, London, Ithaca. 586 pp. Born E.W. (ed.), Heide-Jørgensen M.-P., Merkel F., Cuyler C., Neve P.B. & Rosing-Asvid A. 1998. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 16. 70 pp. Brewer M.J., Butler A. & Cooksley S.L. 2016. The relative perperformance AIC, AICc and BIC in the presence of unobserved heterogeneity. Methods in Ecology and Evolution 7(6): 679–692. Buckland S.T. 1992. Fitting density functions with polynomials. Applied Statistics 41: 63-76. Buckland S.T., Anderson D.R., Burnham K.P. & Laake J.L. 1993. Distance Sampling: Estimating Abundance of Biological Populations. Springer. Buckland S.T., Anderson D.R., Burnham K.P., Laake J.L., Borchers D.L. & Thomas L. 2001. Introduction to Distance Sampling. Oxford: Oxford University Press. Buckland S.T., Anderson D.R., Burnham K.P., Laake J.L., Borchers D.L. & Thomas L. 2004. Advanced Distance Sampling: Estimating abundance of biological populations. Oxford University Press. Buckland S.T., Rexstad E.A., Marques T.A. & Oedekoven C.S. 2015. Distance Sampling: Methods and Applications. Springer. CAFF. 2021. State of the Arctic Terrestrial Biodiversity: Key Findings and Advice for Monitoring. Conservation of Arctic Flora and Fauna International Secretariat, Akureyri, Iceland. ISBN: 978-9935-43190-5. www.arcticbiodiverstiy.is/terrestrial CARMA (Circum Arctic Rangifer Monitoring & Assessment network). www.carma.caff.is/ Correia I.J.F. 2020. Estimating caribou abundance in West Greenland using distance sampling methods. MSc. Thesis. University of Lisbon, Portugal. 63 pp. Couturier S., Dale A., Wood B. & Snook J. 2018. Results of a Spring 2017 aerial survey of the Torngat Mountains Caribou Herd. Technical report, Torngat Wildlife, Plants and Fisheries Secretariat. Cuyler C. 2007. West Greenland caribou explosion: What happened? What about the future? Proceedings of the 11th North American Caribou Workshop, Jasper, Alberta, Canada, 23-27 April 2006. Rangifer, Special Issue No. 17: 219-226. Cuyler C., Nagy J. & Zinglersen K. 2017. Seasonal movement and activity of Akia-Maniitsoq caribou cows in West Greenland as determined by satellite. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 99. 94 pp. Cuyler C., Nymand J., Jensen A. & Mølgaard H.S. 2016. 2012 status of two West Greenland caribou populations, 1) Ameralik, 2) Qeqertarsuatsiaat. Greenland Institute of Natural Resources Technical Report No. 98, 179 pp. Cuyler L.C., Rosing M., Egede J., Heinrich R. & Mølgaard H. 2005. Status of two West Greenland caribou populations; 1) Akia-Maniitsoq, 2) Kangerlussuaq-Sisimiut. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 61. Part I-II, 64+44 pp. Cuyler C., Rosing M., Heinrich R., Egede J. & Mathæussen L. 2007. Status of two West Greenland caribou populations 2006, 1) Ameralik, 2) Qeqertarsuatsiaat. Greenland Institute of Natural Resources. Technical report No. 67. 143 pp. (Part I: 1-74; Part II: 75-143). 52 Cuyler C., Rosing M., Linnell J.D.C., Loison A., Ingerslev T. & Landa A. 2002. Status of the KangerlussuaqSisimiut caribou population (Rangifer tarandus groenlandicus) in 2000, West Greenland. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 42. 52 pp. Cuyler C., Rosing M., Linnell J.D.C., Lund P.M., Jordhøy P., Loison A. & Landa A. 2003. Status of 3 West Greenland caribou populations; 1) Akia-Maniitsoq, 2) Ameralik & 3) Qeqertarsuatsiaat. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 46. 74 pp. Cuyler C., Rosing M., Mølgaard H., Heinrich R. & Raundrup K. 2011. Status of two west Greenland caribou populations 2010; 1) Kangerlussuaq-Sisimiut & 2) Akia-Maniitsoq. Greenland Institute of Natural Resources. Technical Report No. 78. 158 pp. (Part I: 1-86; Part II: 87-158). Cuyler L.C. & Østergaard J. 2005. Fertility in two West Greenland caribou populations 1996/97: Potential for rapid growth. Wildlife Biology. 11(3): 221-227. Dauphiné T.C. 1976. Biology of the Kaminuriak population of barren-ground caribou. Pt. 4, Growth, reproduction, and energy reserves. Can. Wildl. Report Series 38. Gaare E. & Skogland T. 1980. Lichen-reindeer interaction in a simple case model. Proc. 2nd International Reindeer/Caribou symposium 47-56. Trondheim: Direktoratet for vilt of ferskvannsfisk. Gibbons J.D. & Chakraborti S. 2011. Nonparametric Statistical Inferencing. Chapman & Hall. Gates C.C., Adamczewski J. & Mulders R. 1986. Population dynamics, winter ecology and social organization of Coats Island caribou. Arctic. 39(3): 216-222. Heard D.C. & Ouellet J.P. 1994. Dynamics of an introduced caribou population. Arctic. 47(1): 88-95. Jepsen B.I., Siegismund H.R. & Fredholm M. 2002. Population genetics of the native caribou (Rangifer tarandus groenlandicus) and the semi-domestic reindeer (Rangifer tarandus tarandus) in Southwestern Greenland: Evidence of introgression. Conservation Genetics. 3: 401-409. Kelsall L.B. 1968. The Caribou. Ottawa: Queen’s Printer. Kingsley M.C.S. & Cuyler C. 2002. Caribou harvest 2002: advisory document. Pinngortitaleriffik – Greenland Institute of Natural Resources, Nuuk. 12 pp. Landa A., Jeremiassen S.R. & Andersen R. 2000. Rensdyr og moskusokser i Inglefield Land, Nordvestgrønland. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 31. 21 pp. Linnell J.D.C., Cuyler C., Loison A., Lund P.M., Motzfeldt K.G., Ingerslev T. & Landa A. 2000. The scientific basic for managing the sustainable harvest of caribou and muskoxen in Greenland for the 21st century: an evaluation and agenda. Technical Report 34, Greenland Institute of Natural Resources, Pinngortitaleriffik. Loison A., Cuyler C., Linnell J.D.C. & Landa A. 2000. The caribou harvest in West Greenland, 1995-1998. Pinngortitaleriffik – Greenland Institute of Natural Resources. Technical Report No. 28. 33 pp. Miller D.L., Rexstad E., Thomas L., Marshall L. and Laake J. L. 2016. Distance Sampling in R. Journal of Statistical Software 89(1): 1–28. Marques T.A. 2009. Distance Sampling: estimating animal density. Significance 6(3): 136–137. 53 Marques T.A. 2018. Estimating caribou abundance for GINR s 2018 West Greenland caribou survey. Techincal Report 3, Centre for Research into Ecological and Environmental Modelling. Report produced for GINR under a research contract between CREEM and GINR. 33 pp. Marques T.A., Buckland S.T., Borchers D.L., Rexstad E. & Thomas L. 2011. Distance Sampling. International Encyclopedia of Statistical Science, 1: 398–400. Marques T.A., Thomas L., Fancy S.G. & Buckland S.T. 2007. Improving estimates of bird density using multiple covariate distance sampling. The Auk 124(4): 1229–1243. Poole K.G., Cuyler C. & Nymand J. 2013. Evaluation of caribou Rangifer tarandus groenlandicus survey methodology in West Greenland. Wildlife Biology 19: 225–239. Reimers E. 1983. Growth rates and body size differences in Rangifer, a study of causes and effect. Rangifer 3: 3-15. Reimers E. 1993. Antlerless females among reindeer and caribou. Can. J. Zool. 71: 319-325. Reimers E., Lund S. & Ergon T. 2011. Vigilance and fright behaviour in the insular Svalbard reindeer (Rangifer tarandus platyrhynchus). Canadian Journal of Zoology 89(4): 753-764. Schaeffer J.A, Mahoney S.P, Weir J.N., Luther J.G. & Soulliere C. 2016. Decades of habitat use reveal food limitation of Newfoundland caribou. Journal of Mammalogy 97(2): 386-393. Skoog R.O. 1968. Ecology of the caribou (Rangifer tarandus granti) in Alaska. Ph.D. Thesis. University of California at Berkeley, 699 pp. Soulliere C. & Hammel C. 2015. A report on the Newfoundland caribou: a summary and interpretation of the state of knowledge of the island of Newfoundland’s caribou population and key considerations for sustainable management. Newfoundland Labrador, Department of Environment and Conservation, 779 pp. Tamstorf M. P., Aastrup P. & Cuyler C. 2005. Modelling critical caribou summer ranges in West Greenland. Polar Biology. 28: 714–724. Thing H., Olesen C.R. & Aastrup P. 1986. Antler possession by west Greenland female caribou in relation to population characteristics. Rangifer, Special Issue 1: 297-304. Thomas L., Buckland S.T., Burnham K.P., Anderson D.R., Laake, J.L., Borchers D.L. & Strindberg S. 2002. Distance Sampling. Encyclopedia of Environmetrics 1: 544–552. Thomas L., Buckland S.T., Rexstad E.A, Laake J.L., Strindberg S., Hedley S.L., Bishop J.R.B., Marques T.A. & Burnham K.P. 2010. Distance software: design and analysis of distance sampling surveys for estimating population size. J. Appl. Ecol. 47(1): 5-14. Wolfe S.A., Griffith B. & Wolfe C.A.G. 2000. Response of reindeer and caribou to human activities. Polar research 19(1): 63-73. Ydemann D. & Pedersen C.B. 1999. Rensdyr i Vestgrønland 1993-1996. Unpublished report to Pinngortitaleriffik – Greenland Institute for Natural Resources, Nuuk, Greenland (in Danish). 68 pp. 54 Appendix 1 Statistical methods behind Distance Sampling This appendix presents the basic building blocks and reasoning behind Distance Sampling (DS) design-based methods, followed by some details. This summary of statistical methods is from Correia (2020). Fundamental concepts Before entering into the detailed theory behind the Distance Sampling methodology, we present a simpler design, which is quadrat or plot sampling (Buckland et al. 2001; Marques, 2009). In plot sampling, a region of interest with total area 𝐴, is divided into small plots of area 𝑎𝑝𝑙𝑜𝑡 (Fig. 14). Some of these small plots are randomly chosen for sampling and the total number of individuals within these, 𝑛𝑝𝑙𝑜𝑡, is recorded. Figure 14. Plot sampling grid example of total area 𝐴 divided into smaller plots of area 𝑎 𝑝𝑙𝑜𝑡 . The density within each plot, 𝐷𝑝𝑙𝑜𝑡, is the number of individuals per unit area for the respective plot so, by definition, it is given by Equation (1) where 𝑎 is the total area sampled within 𝐴. (i.e., 𝑎 = 4 ⋅ 𝑎𝑝𝑙𝑜𝑡 = 4𝑘𝑚2 for Fig. 14) Since a random design was used, the density is a representative estimate, by design, for the total area 𝐴. Hence, an estimate for the abundance, 𝑁 , can be obtained by simply multiplying 𝐷 plot by the total area 𝐴, Equation (2) 55 𝑃 The DS methodology is an extension of quadrat-based sampling methods. The detail that creates the bridge from one methodology to the other is the fact that the method described above assumes that every individual of interest is detected (Miller et al. 2016). Frequently, this assumption cannot be met, specifically if among the individuals of interest there are animals impossible to observe owing to low sightability. Several factors cause low sightability, including topographical barriers, weather conditions, ground surface conditions and many others related to observer training and survey design. The proportion of individuals that were not detected can be estimated using the detection function fitted to the observed distances (Thomas et al. 2002). Once this proportion is estimated, it can be considered to obtain more accurate estimates and then, an extrapolation for a wider region can be done similarly as shown in Equation (2). In Distance Sampling, this proportion of detected objects in the area 𝑎 is defined as the probability of detection, 𝑃𝑎. Therefore, a density estimate can be obtained as per Equation (1) by adjusting 𝑛𝑝𝑙𝑜𝑡 by 𝑃𝑎, i.e., by correcting the detections for those that were missed. Since the latter cannot be known, in general, an estimate must be also obtained, thus Equation (3) where 𝑃 a is an estimate of 𝑃𝑎 obtained from the distance data, and 𝑎 is the area of the sampled region. Usually 𝑎 = 2𝑤𝐿, with 𝑤 as the truncation distance, for both sides of the centreline, and the total transect length 𝐿 = ∑𝑙𝑗 𝑘 𝑗=1 , where 𝑙 is the length of transect 𝑗. Abundance can be determined using a reasoning analogous to that above (Equation 2). The truncation distance is defined as the distance beyond which distances are not recorded. This can be defined in the field or at the analysis stage. The coefficient of variation of 𝐷 , c𝑣(𝐷 ), is related with two random components referred above, encounter rate (𝑛𝑝𝑙𝑜𝑡/𝐿), and 𝑃 a, plus a third one that is the estimate of the expected size of detected clusters (𝐸  (𝑠)). Assuming independence between these, the former is given by Equation (4) An approximation of the standard error of 𝐷 , 𝑠𝑒(𝐷 ), is defined as Equation (5) 56 Once these are obtained, an approximate 100(1 − 𝛼)% confidence interval (CI) can be determined by Equation (6) Where is the quantile of the N(0,1) distribution 1.96 for a 95% confidence interval). However, the distribution of the 𝐷 is positively skewed, thus an interval assuming that 𝐷  is log-normally distributed has better coverage. According with Buckland et al. (2015), a 100(1-alpha)% confidence interval can be given by Equation (7) where Equation (8) and Equation (9) For further details see Buckland et al. (2001) and Buckland et al. (2015). Probability of detection Given the above, the probability of detecting an object, giving that it is within the area covered by the transects, 𝑃 a, needs to be estimated. For this project, the object of interest consists in caribou groups. To illustrate the importance of this probability, consider that an observer walks across a large patch of tundra and detects 8 caribou (Fig. 15). While discussing with the local biologist, and considering the biologist’s experience, he/she will state that, on average, only one third of all caribou present are detected (i.e., 𝑃 a = 1/3) meaning that probably there were around 24 caribou within that patch of tundra and 16 have been missed. That is where Distance Sampling is useful, since it allows a rigorous framework for the estimation of Pa and then an estimate of abundance can be obtained as shown in Equation (3). 57 Figure 15. Example of a patch of tundra with the transect in the middle. Blue dots represent eight observed caribou, while orange dots represent the 16 undetected ones. The lines perpendicular to the transect represent the recorded distances. Distance Sampling methods The detection function, 𝑔(𝑦), describes the probability of detecting an object of interest given that it is at a distance 𝑦, from the centreline (also known as 0-line), thus being a nonincreasing function of 𝑦 (Buckland et al. 2015). For line transects, 𝑦 is the perpendicular distance from the 0-line to the detected object. Within Distance Sampling methods, the probability of detection is explained recurring to these observed distances (Buckland et al. 2001). Sometimes covariates may be added to explain their relationship with the detection probability. In this situation, we are within the Multiple Covariate Distance Sampling (MCDS) framework (Buckland et al. 2001). Conventional Distance Sampling Conventional Distance Sampling (CDS) occurs when no additional covariates are added to the model. Once the detection function is estimated, 𝑃 a can be obtained via the following equation Equation (10) where 𝜋(𝑦)= 1 𝜔 and, therefore, used to estimate density using Equation (3). For 𝑔(𝑦) it is also specified a flexible semi-parametric model, composed by a key function and some 64 von Mises test and the 𝜒2 Goodness-of-Fit test (GOF test). The likelihood ratio test can also be used but, since it is only applicable for nested models, AIC is the recommended method (Marques et al. 2007). A proper model should be simple with an adequate fit without overfitting the data. Akaike Information Criterion The relative fit of alternative models may be evaluated recurring to AIC, or AICc, in case of small samples, providing a small sample bias correction (Buckland et al. 2001). These criteria can be determined as follows Equation (15) Equation (16) where ℒ is the likelihood function, 𝑞 is the number of estimated parameters in the model, and 𝑛 is the sample size. This measure provides a trade-off between bias and variance. AIC includes two terms, one related with the fitted model, and the other working as a penalty considering the excess of parameters in the model (Brewer et al. 2016). Kolmogorov-Smirnov test The Kolmogorov–Smirnov test is one of the tests that can be applied to the detection function to assess model fit (Buckland et al. 2004). This test is only applicable for continuous data, being preferable to the 𝜒2 GOF test for MCDS methods. Considering the cumulative distribution function (c.d.f.) 𝐹 (𝑥) = 𝑃 (𝑋 ≤ 𝑥) and the empirical c.d.f. (e.d.f.) 𝑆(𝑥), the null hypothesis to be tested is 𝐻0 ∶ 𝐹 (𝑥) = 𝐹0(𝑥), ∀𝑥. The alternative hypothesis states that both functions differ for at least some value of 𝑥. In practice, 𝐹 (𝑥) is replaced by its estimate, and 𝐻0 states that the assumed model is the true model for the data (Buckland et al. 2004). The largest absolute difference between 𝐹 (𝑥) and 𝑆(𝑥), denoted 𝐷𝑛, is the test statistic (Gibbons and Chakraborti 2011). The corresponding 𝑝-value can be approximated by Equation (17) 65 Cramér-von Mises test Similar to the Kolmogorov-Smirnov test, the Cramér-von Mises test shares the same null hypothesis and basis on differences between c.d.f. and e.d.f. However, instead of considering only the largest difference between the two functions, this test is based on their entire range (Buckland et al. 2004). The test statistic can be given by Equation (18) Chi-square Goodness-of-Fit test The 𝜒2 Goodness-of-Fit test (Buckland et al. 2001, 2015) compares the observed frequencies, ni, with the expected frequencies under the model E(ni) and it is given by Equation (19) under the null hypothesis (H0) of good model fitting, i.e., the difference between the observed (ni) and expected (E(ni)) counts is close to zero. In Equation (19), n is the total number of observations, u is the number of groups (or bins) within the distance data, and q is the number of model parameters estimated. Reject H0 if 𝑋𝑜𝑏𝑠 2 > 𝑋1−𝛼;(𝑢−𝑞−1) 2, with the latter representing the 1–𝛼 quantile from a 𝜒2 distribution with u-q-1 degrees of freedom. As the number of parameters of the fitted model increases, the bias decreases, but the sampling variance increases (Buckland et al. 2001). While the Goodness-of-Fit test results should be considered in the analysis of distance data, they will be of limited value in selecting a model since these tests are sensitive to heaping. Therefore, care is needed in choosing suitable distance intervals. If data are collected with no fixed 𝜔, it is possible that a few extreme outliers will be recorded. These values are not useful, and the data should therefore be truncated. This can be checked using the distances’ histogram, and whether there is evidence of heaping or not (Buckland et al. 2001; Couturier et al. 2018). 66 Goodness-of-Fit tests allow formal testing of whether a detection function model provides an adequate fit to the data. Since the GOF test cannot be used on continuous data, unless grouped, it is of limited use for testing MCDS models (Buckland et al. 2015), being useful for testing models using CDS methods. However, if distances are not grouped, they must first be categorized into groups to allow the test to be conducted. Thus, there is a subjective aspect to the test, and different analysts, using different group cut points, may reach different conclusions about the model adequacy. In contrast, the Kolmogorov– Smirnov and Cramér–von Mises tests can only be applied to continuous data (Buckland et al. 2015). 67 Appendix 2 Distance Sampling Assumptions – short summary Line transect Distance Sampling assumptions and design are described in Buckland et al. (1993) and a summary of the assumptions for survey of large herbivores in Greenland provided below are from Cuyler et al. (2016). 1. All caribou on the 0-line are detected. This is critical and must be true. 2. Caribou are randomly distributed. (Lacking this will not bias abundance estimates if the transect lines are randomly placed, which they were.) 3. Detection of caribou is independent. (Although detection was dependent in our survey, the lines had random start-end points, so this assumption is not violated). 4. No caribou movement prior to detection. The method is a ‘snapshot’ method. In practice this assumption is not violated if the observer moves faster than the animal, e.g., if movement of caribou to the next transect line to be surveyed is rendered impossible, which it was. 5. Distance measurements are exact. Provided distance measurements are approximately unbiased, bias in line transect estimates tends to be small in the presence of measurement errors. In our survey we binned the observations into distance intervals which decreases measurement error. 6. Clusters (caribou groups) close to the 0-line are accurately sized. 7. Other assumptions include those for other survey types, e.g., that each population is closed, being confined within a clearly defined area. 68 Appendix 3 Recommendations for improving future surveys. Aerial survey methods & design The 10.6% survey coverage in 2018 promotes accuracy of abundance estimates and should be continued in future to facilitate evaluating population trends. The flight altitude could be reduced to 30-35 m. The flight altitude of 40m while observers scanned the landscape out to 1000-1500m from the 0-line was mentally exhausting. This was because the amount of terrain to be scanned was too great for even the relatively slow speed flown (60-70 km/hour), given the high degree of background camouflage that hides caribou. Although an altitude of 30-35m likely will cause amount of ‘dead’ ground to increase, the Distance Sampling will mitigate for any caribou missed, provided that dead ground is not on the line (which there is no reason to expect it might be). Meanwhile, observer ability to judge correct distance bin may improve. It is the author’s experience that without practice people commonly misjudge distance. Looking down from above can exacerbate this tendency. A lower angle to the terrain could provide a more (normal) horizontal line-of-sight to the animals and may increase binning accuracy. The timing for aerial surveys could remain early March because it coincides with annual minimum caribou movement (avoids double counting), and enough day length for flying the pilot maximum of 7-hours per day. Experience from eight surveys since 2000 has illustrated that snow cover and depth is variable regardless of the winter period chosen. Demographics When flying line transects, distance and other factors often make identification of calves impossible, resulting in an underestimate of calf number. Herd structure data must continue to be collected in efforts separate from flying the line transects for Distance Sampling. Logistics Check whether other helicopter options are available. To date, the smallest helicopter available is the AS350 from Air Greenland (Charter). The AS350 permits limited vision for rear observers, owing to the small window size containing several bar/struts, and which under cold ambient temperatures always fog with ice-frost. These factors reduce visibility of terrain. 69 Appendix 4 Photographs of KS caribou survey conditions March 2018 Figure 19. Rugged terrain with good sunlit conditions (above), and excellent, almost complete snow cover with shadows (below). Photos C. Cuyler. 70 Caribou survey conditions March 2018 Figure 20. Portions of line transects flown across high elevations, Sisimiut South sub-area. Photos C. Cuyler. 71 Caribou survey conditions March 2018 Figure 21. Flat light combined with ground showing through thin layer of snow. Photos C. Cuyler. 72 Caribou survey conditions March 2018 Figure 22. Vegetation poking through thin snow layer in flat light (above) or with shadows (below). Photos C. Cuyler. 73 Caribou survey conditions March 2018 Figure 23. Thin snow layer with ground showing through (above), or rocks and vegetation (below). Photos C. Cuyler.