Satellite imagery can be used to detect variation in abundance of Weddell seals (Leptonychotes weddellii) in Erebus Bay, Antarctica
- 487 Downloads
The Weddell seal population in Erebus Bay, Antarctica, represents one of the best-studied marine mammal populations in the world, providing an ideal test for the efficacy of satellite imagery to inform about seal abundance and population trends. Using high-resolution (0.6 m) satellite imagery, we compared counts from imagery to ground counts of adult Weddell seals and determined temporal trends in Erebus Bay during November 2004–2006 and 2009, and December 2007. Seals were counted from QuickBird-2 and WorldView-1 images, and these counts were compared with ground counts at overlapping locations within Erebus Bay during the same time. Counts were compared across years and within individual haul-out locations. We counted a total of 1,000 adult Weddell seals from five images across all years (for a total of 21 satellite-to-ground count comparisons), approximately 72% of the total counted on the ground at overlapping locations. We accurately detected an increase in abundance during 2004–2009. There was a strong, positive correlation (r = 0.98, df = 3, P < 0.003) between ground counts and counts derived from the imagery. The correlation between counts at individual haul-out locations was also strong (r = 0.80, df = 19, P < 0.001). Detection rates ranged from 30 to 88%. Overall, our results showed the utility of high-resolution imagery to provide an accurate way to detect the presence and variation in abundance of Weddell seals. Our methods may be applied to other species in polar regions, such as walruses or polar bears, particularly in areas where little is known about population status.
KeywordsWeddell seal Leptonychotes weddellii Remote sensing Satellite imagery Antarctica Erebus Bay
Weddell seals (Leptonychotes weddellii) are found along the coast of Antarctica, where fast ice is present for a significant portion of the year. Females are philopatric (Stirling 1969a; Stirling 1974; Croxall and Hiby 1983; Cameron and Siniff 2004; Cameron et al. 2007; Hadley et al. 2007) and form traditional haul outs for pupping where persistent tide cracks reliably offer access from the ocean to the ice surface (Tedman and Bryden 1979; Siniff et al. 2008). While one of the best-studied marine mammal populations in the world exists in Erebus Bay (Hastings and Testa 1998; Cameron and Siniff 2004), less is known about populations of Weddell seals elsewhere around the Antarctic continent. The paucity of data regarding population status of seals is largely due to the logistical difficulties of accessing potential seal habitat in areas of Antarctica that are not in close proximity to research stations. Knowledge of seal distributions and numbers in other areas would be valuable for a variety of reasons. First, when attempting to understand the dynamics of local populations, it would be useful to know the status of nearby populations that may be involved in temporary immigration/emigration exchanges (Cameron and Siniff 2004; Rotella et al. 2009). Further, information on population status from numerous locations around the continent will be important for monitoring the species’ overall status and its responses to environmental change (Siniff et al. 2008). Thus, developing methods for attaining knowledge of Weddell seal status throughout Antarctica is important to understand the ecology of sea ice seals and to assess the potential impacts of climate change or other anthropogenic influences on upper top-trophic predators.
Knowledge of Weddell seal abundance at diverse locations will also be useful for more general monitoring of the Antarctic marine system. For example, Weddell seals are an important predator of Antarctic toothfish (Dissostichus mawsoni; Ponganis and Stockard 2007), which have become the target of a fishery in the Ross Sea (Ainley and Siniff 2009; Blight et al. 2010). Given the difficulties of using standard methods to monitor toothfish in Antarctic pack-ice conditions, and in keeping with CCAMLR’s precautionary and ecosystem management principles (Constable et al. 2000; Croxall and Nicol 2004), it would be desirable to monitor seal numbers along with fish harvest to evaluate fishery impacts on this important top-level predator and its food web.
Traditional monitoring methods for pinnipeds include using aircraft or aerial photography to census populations in the Arctic (McLaren 1966; Burns and Harbo 1972; Lavigne et al. 1982; Gilbert 1989; Johnston et al. 2000; Bester et al. 2002; NMML 2007), and intensive on-the-ground counts on shore-fast ice (Siniff et al. 1977) or ship-based line-transect surveys through pack ice (Gelatt and Siniff 1999; Ackley et al. 2002; Southwell et al. 2004; Flores et al. 2008; Bengston et al. 2011) in the Antarctic. Weddell seals haul out on the surface of shore-fast ice to give birth, suckle young, rest and molt; so employing imaging techniques to obtain population trends along the Victoria Land coast of the Ross Sea is feasible. Aerial surveys have previously been used to count seals in the Ross Sea (Smith 1965; Stirling 1969b). However, technology has advanced such that the use of satellite imagery may now be used to conduct significant research in the Antarctic (Barber-Meyer et al. 2007; Fretwell and Trathan 2009; Fretwell et al. unpublished data), while minimizing the substantial efforts and impact to the Antarctic environment of ground and aerial survey methods (Eberhardt et al. 1979; Green et al. 1995; Southwell 2005a).
Studying Weddell seals in the Ross Sea using satellite imagery has already been suggested in the literature (Barber-Meyer et al. 2007), and distribution and abundance around the continent has only recently been addressed by the international research program, Antarctic Pack Ice Seals (Bester et al. 2002; Bester and Stewart 2006; Bengston et al. 2011); United States’ efforts took place within the Ross Sea during 1999–2000 (Ackley et al. 2002; Bester and Stewart 2006; Bengston et al. 2011). Thus, we wished to determine whether satellite imagery could be used to reliably identify adult Weddell seals hauled out on the ice and whether counts determined from imagery could provide accurate information about known abundances. Here, we present findings that suggest that high-resolution (0.6 m) satellite imagery can, indeed, be used to identify occurrence and to detect changes in abundance of a Weddell seal population. The cost of high-resolution imagery has made this kind of research difficult in the past, but satellite imagery is becoming more available and costs are declining. Our methodology will likely facilitate similar research in the future and may allow an efficient, cost-effective way to study polar pinnipeds in areas where little is known about distribution and abundance.
Materials and methods
We gathered high-resolution satellite images of Erebus Bay for dates during November 2004–2006 and 2009, and December 2007. We used WorldView-1 (panchromatic, 0.6 m resolution) and QuickBird-2 (2.4 m multispectral and 0.6 m panchromatic) images, which were identified through the vendor’s search tools (http://www.digitalglobe.com). Local times of image capture ranged from 1000 to 1300 hours, and these images were the only suitable, cloud-free images of the area as of December 2009. Each image was then analyzed for its utility. To do so, a remote sensing analyst determined, without knowledge of ground count results, which images and haul-out locations within each image to use based on image quality. Criteria for including a haul-out location in our analysis were as follows: (1) sufficient quality of the image (i.e., low banding, cloud-free, and few shadows); (2) the entirety of a haul-out location was captured within the image; and (3) the dates of ground counts and satellite images were within 7 days of each other. If any haul-out location on any image did not meet all three criteria, it was omitted from analysis.
One observer counted seals from the imagery at each suitable haul-out location on each image. The number of haul-out locations compared per date differed based on image quality and the spatial coverage of each image. For example, Big Razorback Island (Fig. 2) was captured on the image dated November 12, 2006, so we counted the number of seals present and compared those results to ground counts made at that location and on that date. Ground counts at locations that were not represented on a given image were not used for comparison for the image date. We only made comparisons of seal counts at haul-out locations where the image and ground counts were spatially and temporally comparable (within 7 days).
Ground counts were conducted in Erebus Bay, Antarctica, from 1000 to 1800 hours at 3- to 6-day intervals from early November until mid-December each year. These counts recorded all individual adult seals, individual pups, and adult–pup pairs. Because Weddell seals that are hauled out are highly detectable (Rotella et al. 2009), we are confident that these ground counts missed very few seals that were present on the ice surface at the time of the count. Thus, our measure of abundance from ground counts was the actual count on the surface on a given day at a given haul-out site. Repeated counts on marked animals can be used with mark-recapture methods (Williams et al. 2002) to estimate actual abundance (Rotella et al. 2009) but that was not our focus in this effort. Here, we were interested in knowing whether counts of seals on the ice made on the ground would strongly correspond with counts of seals via satellite imagery.
For each location and date for which we had associated ground and satellite counts, we calculated the number of seals counted by each method and the proportion of ground-counted adult seals that were detected by satellite across all haul-out locations that could be used across a given year. We excluded pups from all comparisons because we assumed that pups, which even at weaning are generally less than half the size of adults, would be less likely to be detected in 0.6 m resolution imagery. Pearson’s correlation coefficient was calculated for annual counts from the two approaches and for counts from the two approaches at individual haul-out locations. We also determined change in abundance through years at three haul-out locations where ≥3 satellite-to-ground count comparisons were possible and calculated Pearson’s correlation coefficient for each location through time.
Summary of the comparison of counts of adult Weddell seals derived from high-resolution satellite imagery and ground counts in Erebus Bay, Antarctica
Ground count date(s)
Count on images
Ground count of adults
22 Nov. 2009
17, 21, 25 Nov. 2009
TH, TP, NB, TR, HC
10 Dec. 2007
4 Dec. 2007
TI, BR, II, TR
12 Nov. 2006
12 Nov. 2006
TI, TP, BR, HC, TR
9 Nov. 2005
9, 13 Nov. 2005
TH, NB, HC
19 Nov. 2004
18 Nov. 2004
TH, SB, HC, TR
Across the 5 years, 1,000 seals were detected on the five annual images, which represented 71.7% of the 1,394 seals known to be present from ground counts (Table 1). However, satellite counts did not detect a constant proportion of the seals detected in ground counts. In 2004, when the fewest seals were recorded on ground counts, counts from imagery detected only 30% of the seals known to be present. In contrast, during the two most recent years, in which ≥385 seals were detected on ground counts, image-based counts detected ≥82% of seals known to be present (Table 1).
Our work provides an important step forward in polar ecology by demonstrating that remote sensing data can be used effectively to identify presence and determine abundance of the Weddell seal population within Erebus Bay, Antarctica. This study combined a few important factors that contributed to our strong results. First, the Weddell seal population of the Erebus Bay area was an ideal test population because of its accessibility and proximity to McMurdo Station and because of how much is known about the current and historical population abundance of adult seals there (Smith 1965; Stirling 1969a; Siniff et al. 1977; Testa and Siniff 1987; Hastings and Testa 1998; Cameron and Siniff 2004; Hadley et al. 2007). Long-term population datasets are rare for large, long-lived animals (Fossey and Harcourt 1977; Croxall and Kirkwood 1979; Garrott and Taylor 1990; Micol and Jouventin 2000), and the population of seals in Erebus Bay provided a unique opportunity for the count comparisons needed to establish strong correlations between ground and satellite counts.
Secondly, because it is nearly impossible to count all animals present in a population (Williams et al. 2002), it is important to conduct surveys during a peak in the population to minimize the number of missing individuals (Eberhardt et al. 1979; Green et al. 1995; Boyd et al. 2010). Ideally, such counts of seals should occur after birthing when adult females are more likely to be on the surface and available for detection (Eberhardt et al. 1979). Thus, the time of year we compared our images to ground counts was important because the population peaks in October–December (Stirling 1969a; Tedman and Bryden 1979; Siniff 1981), when pregnant Weddell seals haul out in groups in Erebus Bay and remain on the ice for several weeks after pups are born (Lindsey 1937; Stirling 1969a). This is also the season when annual ground counts are conducted (Rotella et al. 2009; D.B Siniff, J.J. Rotella, R.A. Garrott, personal communication). Our comparisons were made during an ideal time, when the population was at its peak and when a relatively large proportion of individuals were visible on the ice. It is, however, worth noting that the counts here do not represent the actual abundance of seals present in the study area because on any given survey a large proportion of adult seals, especially males (Stirling 1969a; Gelatt et al. 2000; Gelatt 2001), may be in the water and undetectable with ground or satellite counts. Regardless, mark-recapture estimates of population abundance from repeated surveys within a year do indicate that counts from any single survey are positively correlated with abundance estimates that do take failed detections into account (J.J. Rotella, R. A. Garrott, and D. B. Siniff unpublished data).
Marine mammal populations are generally difficult to census, and direct observations in particular are often hindered by inaccessibility and difficult logistics (Eberhardt et al. 1979; Green et al. 1995; Gelatt and Siniff 1999; Boyd et al. 2010; Bengston et al. 2011). Our approach minimizes several disadvantages of traditional census methods. First and quite importantly, high-resolution satellite imagery eliminates any effects of observer presence on the individuals in the population that would normally occur by walking near, sailing by, or flying over the area (Buckland et al. 2001; Southwell 2005a). Although such effects are not problematic for Weddell seals, human presence can cause major disturbance in breeding colonies of some pinniped species. Use of satellite imagery provides a passive way of viewing a truly undisturbed population. Second, the imagery we used has a wide swath, which provided us with an image of ≥400-km2 for each comparison date. The possible area covered by several satellite images per day (>5,000 km2) would potentially allow the observation of a much larger region than would be logistically possible to cover on foot, by plane, or by ship in 1 day. Third, satellite-based survey methods would allow us to readily obtain replicate counts and would provide the potential to develop rigorous sampling schemes across large areas. For example, coastal areas already known to be breeding sites for Weddell seals could be repeatedly surveyed during the peak of pupping in November to evaluate changes through space and time. Analyses of data across multiple years could then be used to learn about how similar or dis-similar population changes might be across large areas. The technique described here could also be used to identify other large aggregations of seals, if images for coastal areas were repeatedly surveyed when many seals are hauled out (e.g., November pupping season or January when adult seals stay near breeding colonies during the annual molt [Burns and Kooyman 2001]). Surveys in pack-ice areas that have been shown to support large numbers of other seals (Bengston et al. 2011) can be also be used to monitor other species and non-breeding aggregations of Weddell seals. As the methods here did not evaluate the performance of the survey technique in pack ice, it would be ideal to initially pair satellite surveys with information from traditional methods to facilitate comparisons. If correlations between counts from traditional and satellite methods are strong, it could be possible in the future to obtain more information remotely.
Another disadvantage of conducting such rigorous sampling by ship or aircraft-based platforms is that such methods are typically limited by high cost and logistical difficulties, and some areas of Antarctica are impossible to reach. Our methods that used satellite imagery, which was provided through federal licensing agreements, GIS software, and 1–2 observers would cost a fraction of what would be spent gathering the same data on the ground, especially when scaled to cover large or inaccessible areas. Finally, optical satellite imagery can be combined with remotely sensed sea ice data to correlate patterns of abundance and distribution of seals, providing a broader-scale understanding of the effects of sea ice on seal ecology.
We recognize that counts from satellite imagery were consistently lower than the ground counts of adult Weddell seals. However, we accurately captured changes in abundance across years at three haul-out locations (Fig. 5). But the disparity between the ground counts and the satellite counts may be due to the time of day images were acquired. Because the QuickBird-2 and WorldView-1 satellites have polar orbits, the time of day when our images were on-nadir (i.e., directly vertical over the study area) was always between 1000 and 1300 hours. It is well-known that Weddell seals have a diurnal haul-out pattern (Smith 1965; Stirling 1969a, b; Siniff et al. 1971; Lake et al. 1997), with the largest proportion on the ice between 1200 and 1900 hours (Siniff et al. 1971; Lake et al. 1997). Satellite images were collected just before the most inactive portion of the day, and the seals counted were only a proportion of what would likely be hauled out later during the day. It seems likely that part of the discrepancy between counts was due to the time satellite images were collected during the day.
Further, we noticed that detection rates of seals from the imagery were not consistent, ranging from a low of 30% in 2004 to a high of 88% in 2009. Annual variation in total ground counts and in ice surface conditions was high, which provided a useful dataset for evaluating the utility of counts from satellite images. During the study period, ice conditions and seal numbers were strongly influenced by a massive iceberg that blocked the usual advection of sea ice from the area during 2004–2006 (Arrigo et al. 2002; Siniff et al. 2008). During the first several years of the study, the ice was unusually thick with larger and more extensive pressure ridges than are typically experienced in the area. These large pressure ridges and jumbled ice within Erebus Bay made the detection of seals on the ice more difficult, leading to lower detection rates from the satellite imagery. After the iceberg broke up in 2006, pressure ridges gradually diminished in size and extent and the smoothness of the ice surface increased. Our data succeeded in capturing seal response to these conditions (Fig. 5).
However, the consistent under-identification of seals from satellite imagery suggests that for future work, applying some kind of correction factor may be warranted (Eberhardt et al. 1979; Erickson et al. 1989; Bengston and Stewart 1992; Lake et al. 1997; Southwell 2005b; Boyd et al. 2010; Bengston et al. 2011), should the objective be to determine exact population densities of Weddell seals. We did not include a correction factor for this study because this was simply beyond the scope of our initial investigation. Our goal here was to demonstrate the utility of high-resolution satellite imagery for identifying and enumerating seals on the sea ice surface, which could potentially be used for providing an index of abundance, and further as a trend indicator for seal populations in the Antarctic.
Remote sensing of Weddell seals in Erebus Bay indicated a strong, positive correlation (r = 0.98) between counts of adult Weddell seals from satellite imagery and actual ground counts collected during the same time. Given our indications here that satellite counts can provide information about relative abundance and, more importantly, changes in relative abundance, we feel confident that our technique can be applied to search for seals in larger areas where abundance is unknown and where general population trends have never been observed. For example, across broad areas one could examine which populations grow, shrink, or remain unchanged as sea ice conditions change, as fish harvesting practices vary, or as other environmental conditions change through time. One could also investigate possible spatial structuring of population units by monitoring populations separated by varying degrees of distance and occupying locations with different environmental attributes.
It would be useful to conduct further work in Erebus Bay using repeated counts of colonies, assessment of diurnal haul-out patterns, and comparisons of rigorous mark-recapture estimates of absolute abundance (Rotella et al. 2009) with estimates obtained from satellite imagery to determine how well counts from imagery represent absolute abundance. Regardless of the results of such work, our current results indicate that much can be learned about relative abundance. Given that, it is clear that the method presented here can readily identify which sites along the Antarctic coast are or are not occupied by seals under various environmental conditions. Although presence/absence data do not contain as much information on population state as what is provided by data on absolute abundance, recent work has demonstrated the great utility of having occupancy data, especially at broad spatial scales and over a broad range of conditions (MacKenzie et al. 2005; Fretwell and Trathan 2009). Moreover, Weddell seals are disappearing fast enough in some areas of the Antarctic Peninsula (Siniff et al. 2008) that quantitative, relative abundance may provide a useful indication of ecosystem change.
This work was funded by National Science Foundation grants OPP-0753663, OPP-0944411 and OPP-0635739. Satellite imagery data were copyright of DigitalGlobe, Inc. and provided through the Commercial Imagery Program by the National Geospatial-Intelligence Agency. We thank the United States Air Force, Raytheon Polar Services Company, and Petroleum Helicopters International for assistance with field logistics. B. Herried provided consultation for figures and S. Barber-Meyer, C. Southwell, and M. Cameron provided valuable insight into a previous draft of the manuscript.
- Ackley SF et al (2002) A top-down, multi-disciplinary study of the structure and function of the pack ice ecosystem in the eastern Ross Sea, Antarctica. Polar Rec 39:219–231Google Scholar
- Arrigo KR, van Dijken GL, Ainley DG, Fahnestock MA, Markus T (2002) Ecological impact of a large Antarctic iceberg. Geophys Res Lett doi:10.1029/2001GL014160
- Bengston JL, Stewart BS (1992) Diving and haul-out behavior of crabeater seals in the Weddell Sea, Antarctica, during March 1986. Polar Biol 12:635–644Google Scholar
- Bengston JL et al. (2011) Distribution, density, and abundance of pack-ice seals in the Amundsen and Ross Seas, Antarctica. Deep Sea Res. doi:10.1016/j.dsr2.2010.10.037
- Bester MN, Stewart BS (2006) The international antarctic pack ice seals (APIS) program multi-disciplinary research into the ecology and behavior of Antarctic Pack Ice Seals Summary Update. SCAR Expert Group on Seals, 25 ppGoogle Scholar
- Bester MN, Ferguson JWH, Jonker FC (2002) Population densities of pack ice seals in the Lazarev Sea, Antarctica. Antarc Sci 14:123–127Google Scholar
- Blight LK et al (2010) Fishing for data in the Ross Sea. Sci Lett 330:1316Google Scholar
- Boyd IL, Bowen WD, Iverson SJ (2010) Marine mammal ecology and conservation: a handbook of techniques. Oxford, New YorkGoogle Scholar
- Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers DL, Thomas L (2001) Introduction to distance sampling: estimating abundance of biological populations. Oxford University Press, OxfordGoogle Scholar
- Burns JJ, Harbo SJ (1972) An aerial census of ringed seals, northern coast of Alaska. Arctic 25:279–290Google Scholar
- Croxall JP, Hiby L (1983) Fecundity, survival and site fidelity in Weddell seals (Leptonychotes weddellii) population in McMurdo Sound, Antarctica. Can J Zool 82:601–615Google Scholar
- Croxall JP, Kirkwood Ed (1979) The distribution of penguins on the Antarctic Peninsula and island of the Scotia Sea. Brit Antarc Surv 10:1–186Google Scholar
- Eberhardt LL, Chapman DG, Gilbert JR (1979) A review of marine mammal census methods. Wildl Monogr 63:3–46Google Scholar
- Environmental Systems Research Institute (2009) ArcGIS 9.3. Redlands, CaliforniaGoogle Scholar
- Fossey D, Harcourt AH (1977) Feeding ecology of free-ranging mountain gorilla. In: Brock TH (ed) Primate ecology: studies of feeding behavior in lemurs, monkeys, and apes. Academic, New York, pp 449–479Google Scholar
- Gelatt TS (2001) Male reproductive success, relatedness, and the mating system of Weddell seals in McMurdo Sound, Antarctica. Dissertation, University of MinnesotaGoogle Scholar
- Gelatt TS, Siniff DB (1999) Line transect survey of crabeater seals in the Amundsen-Bellingshausen Seas, 1994. Wildl Soc Bull 27:330–336Google Scholar
- Gelatt T, Davis CS, Cameron M, Siniff D, Strobeck C (2000) The old and the new: integrating population ecology and population genetics of Weddell seals. In: Davison W et al (eds) Antarctic ecosystems: models for a wider ecological understanding. Caxton, Christchurch, pp 63–70Google Scholar
- Lavigne DM, Innes S, Kalpakis K, Ronald K (1982) An aerial census of Western Atlantic harp seals (Pagophilus groenlandicus) using ultraviolet photography. In: Mammals in the Seas, FAO Fisheries Series (5) IV, pp 295–302Google Scholar
- MacKenzie DI, Nicols JD, Royle JA, Pollock KH, Bailey LL, Hines JE (2005) Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Elsevier, San DiegoGoogle Scholar
- NMML (National Marine Mammal Laboratory) (2007) Polar ecosystems research: Harbor seal census in Southeast Alaska. Quarterly Report, Alaska Fisheries Science Center, Juneau. http://www.afsc.noaa.gov/Quarterly/jas2007/divrptsNMML4.htm. Accessed 15 Feb 2011
- Ponganis PJ, Stockard TK (2007) The Antarctic toothfish: how common a prey for Weddell seals? Antarc Sci 19:441–442Google Scholar
- Siniff DB (1981) Seal population dynamics and ecology. J R Soc NZ 11:317–327Google Scholar
- Siniff DB, Testa JR, Kuechle VB (1971) Some observations on the activity patterns of Weddell seals as recorded by telemetry. Antarct Res Ser 18:173–180Google Scholar
- Siniff DB, Garrott RA, Rotella JJ, Frase WR, Ainley DG (2008) Projecting the effects of environmental change on Antarctic seals. Antarc Sci 20:425–435Google Scholar
- Williams BK, Nichols JD, Conroy MJ (2002) Analysis and management of animal populations. Academic, San DiegoGoogle Scholar
- Wilson EA (1907) Mammalia (whales and seals). In: National Antarctic expedition 1901–1904. Nat Hist Vol 2, Zoology Pt 2, Aves LondonGoogle Scholar