Accuracy of the method

A primary concern when monitoring emperor penguins using VHR satellite imagery is the accuracy of the classification method. Uncertainty comes from two sources: (1) the errors associated with classifying the area of penguins as a metric of population; and (2) the potential variation in the population in late spring when the imagery is acquired45,46 The levels of variability from each of these two sources are potentially important45. With respect to the first issue, LaRue et al.30 and this study use images linked to Bayesian statistics to produce an overall population trend. Bayesian statistics are based on the concept that probability is a measure of belief in an event and gives confidence values based upon not just the overall probability, but the amount of data used to calculate those statistics (La Rue et al. analysed 460 images30, here we use 241). The population trend for both the LaRue et al.30 and our analysis is based upon methods that have been developed and refined by different groups over many years. The common practices that have evolved are considered robust and best practice for the data available. With respect to the second issue, it is important to recognise that the timing of image acquisition in late spring, dictates that only a proportion of the adult population is in attendance at each colony, with an unknown proportion at sea foraging, or having already failed in their breeding attempt. As such, it is uncertain whether the area classified as penguins in the images reflects a true population estimate. However, recent work at Atka Bay suggests that adult attendance remains stable throughout spring, extending into late November or early December46, so the estimate of population in spring is plausibly based upon a constant demographic component. Further, the image acquisition date is used as a parameter in the model to reflect this.

Previous work24,30 has converted the index of area representing penguins into a number of penguin pairs, using aerial survey observation to ground truth satellite imagery; effectively equating one square metre to a single penguin based on a linear regression of the relationship between satellite derived area and actual counts from aerial survey24, or via a parameter in the Bayesian model30. These population estimates then reflect spring adult attendance at the colony, rather than a full breeding population estimate. In our analysis we are concerned with relative change, rather than population size per sea. Thus, we remove this potential source of error by reporting only the area of penguins as an index of abundance.

Interpretation of results

Our results build on those of LaRue et al.30 and show an evident decline in population numbers in our sector of interest; this study suggests a decline in emperor penguin populations with a best fit of −22% over the 15 year period, equivalent to an absolute decline of 1.6 % per year (N15 = (100 − 22) × N0 = λ15 × N0, which gives λ = 0.984, meaning an annual decline of ca. 1.6%) or a log-linear annual rate of change of −2.25% per year. We report this decline with a very high probability (0.91 for a 30% decline over three generations (a total of 48 years as each generation is 16 years on average; cf 16.4 years reported by Bird et al.47). Our results indicate a steep decline in population until 2016, followed by an undulating plateau in population size after that point. Although the 15-period of our study is relatively short for a long-lived species, it does approximately represent one breeding lifetime of the emperor penguin, which on average lives 20 years and first breeds at around 5 years of age. As indicated above, successful annual breeding for this species is more critical than most other birds and skipping breeding in poorer years is not a long-term option. Therefore, the relatively short time series shown here is important and highly relevant to our study that it might be for many other avian species.

Our analyses reflect a quarter of the Antarctic coastline and, as such, may not be representative of the whole continent. Environmental conditions and patterns of sea-ice extent are regional (Fig. 4). Our sector of interest has been subject to a greater decline in sea-ice extent, especially in the Bellingshausen Sea region.

Fig. 4 Change in sea ice duration (in days per year) around Antarctica (Sea Ice Index dataset: https://nsidc.org/data/g02135/versions/3 accessed 01/08/2024. Full size image

Interestingly, although the sea-ice duration in the Bellingshausen Sea sector has reduced significantly (Fig. 5A), the decrease in our emperor penguin index for this region has not tracked sea-ice decline. This is in marked contrast to the Weddell Sea (Fig. 5B), where sea-ice duration has been variable31, but the decline in our emperor penguin index has been significant. In the Bellingshausen Sea region, sea-ice loss in 2022 was extensive and three of the four colonies analysed in the region in this study had no adults present33. It is plausible that breeding was initiated at these colonies, but breeding failure subsequently occurred. When the sea ice returned in 2023, numbers were lower than average.

Fig. 5: Antartic sea ice decline. A Bellingshausen Sea regional index of abundance (based on x colonies) showing the extreme event in 2022. B Weddell Sea regional index of abundance (based on y colonies). The loss of the Halley Bay colony in 2016 is evident, followed by a slight upturn as adults relocated to the Dawson-Lambton colony. Full size image

In the Weddell Sea, the collapse of the Halley Bay colony in 201634 is a major factor. This colony was originally one of the largest in Antarctica but suffered breeding failures in each year subsequently. After the sea ice break-up in 2016, most of the adult population emigrated to the Dawson Lambton colony, 85 km to the south, where sea ice conditions remained stable. Taken together, the Halley Bay and Dawson Lambton index of abundance showed that the adult population had recovered by 2018 (see Supplementary Fig. 2). The loss of production between 2016 and 2018 is likely to have been significant and may account for a degree of inter-annual variation in the Weddell Sea combined index. Importantly, however, the combined index for the two colonies shows a decline prior to 2016, with an abundance index of ~20,000 in 2010 and ~12,000 in 2015 (Supplementary Figs. 1 and 2). The underlying cause of this decline before 2016 remains unclear.

Comparison with the published results from LaRue et al.30 (see Supplementary Fig. 3), on a regional and overall level, shows good correspondence between 2009 and 2015, with each year’s combined overall value within 7% of each other. However, in the years 2016–2018 our estimates are lower than the LaRue estimates of the same colonies (by 18.6%, 20.4% and 17.8% respectively, see Supplementary Data 6). In 2016 and 2017, this can be partially explained by a slight difference in methodology, specifically of treatment of the zero count at Halley Bay due to early sea ice loss in these years. In our analysis, colonies that have no penguins in the imagery are counted as zero, reflecting the abundance at the time, but in the LaRue et al.30 analysis they are classed as null on the assumption that a breeding population was at the site in the winter before the image was taken in the spring. This can be seen as a difference between our aim to estimate relative springtime abundance, rather than provide an absolute population estimate at each site in each year. However, this reasoning does not fully explain all of the discrepancy in 2016, 2017 and 2018. In these years, lower estimates at three sites: Atka Bay, Gould Bay and Smyley Island are responsible for 88% difference, with the largest difference at Smyley Island (44% of the total difference). Most colonies, however, still show similar estimates in these years. Why these three colonies exhibited very different results is difficult to assess, but it does highlight the high potential variance of estimating emperor population at individual colonies using this methodology.

Why have emperor penguin populations declined?

Attributing a reason why emperor penguin numbers have declined in this sector is complex and, as yet, poorly understood. Previous work has shown that population numbers are affected by fast ice extent, with low fast ice leading to early ice break-out and poor breeding success and excessive fast ice extent leading to higher rates of adult mortality5,48,49. Reduced sea ice extent and early ice-breakout since 2016 have been noted in several areas, but the majority of the population decline shown in this paper has been in the period between 2009 and 2015, before the well documented years of extreme low sea ice that have been observed recently7,8. It must follow, therefore that other environmental factors are the main drivers of population decrease in this sector.

One colony, located within this sector but not included in this study, is the site on the Dion Islands that is likely to be functionally extinct. This colony decreased in population almost continually from the late 1970s until 2009, when no breeding birds were recorded37. Since its discovery in the 1940’s the small colony was located on an island, so breeding success due to lack of sea ice as a breeding platform could not have been a major factor, and other drivers must have reduced the population. The location, in Marguerite Bay on the West Antarctic Peninsula has witnessed rapid warming over the last 50 years50,51 and the Trathan et al. 37 study lists some of the implications of this warming as potential drivers of change. They include reduced prey availability as pack-ice conditions change and oceans warm; increased competition from other predators as they move into the area to exploit the marine resources; lack of fresh snow at the site that emperors use to re-hydrate while breeding; and increased disturbance and predation as other predators move into the area with the warmer conditions and more open ocean environment. It is likely that these factors may now be affecting several of the other emperor colonies in this sector.

The other ice-obligate penguin species, the Adélie (Pygoscelis adeliae), has been shown to have adapted its foraging strategy when sea ice cover is low. In times of no or little sea ice, adults must dive deeper to catch prey, which takes more effort and lowers breeding success52,53. Conversely, and much like earlier studies on emperor penguins, excessive sea ice cover can also be detrimental as adults need to walk further across the ice to access foraging grounds. Recent work has also shown that juvenile emperor penguins exhibit very high interannual variability in body mass, which has important impacts on juvenile survival, and which is tightly associated with sea ice extent during the chick-rearing period12.

There are also other potential drivers linked to rainfall and storminess. Studies on Adélie penguins have shown that increased storms and extreme rainfall can seriously impact breeding success54,55. Extreme storms and precipitation events result in flooding of nests, which drenches chick,s necessitating the use of more energy to regulate temperature. Although nest flooding is not a problem for emperors, personal observations (PTF) support this hypothesis that chick drenching can contribute to mortality as an unusually early extreme rainfall event in mid-October was observed causing the death of several hundred otherwise healthy chicks at the Snow Hill Island colony in 2010. The Antarctic Peninsula has already seen increased storminess, snowfall and rainfall56 and it is likely that, for more northerly or warmer colonies, increased and earlier rainfall events around the coast could be a driver of population change. Other potential causes of chick wetting linked to warming temperatures, such extensive melt pools on the surface of the sea ice in warm years, could also have detrimental effects on breeding success. Analogous to the situation for Adélie chicks, these pools can lead to chick wetting and increased energy expenditure, but like all of these environmental factors, these potential drivers have not yet been quantified for emperor penguins.

A final consideration is huddle size57. As colony size decreases, the huddling ability of emperors in winter or during spring storms becomes less effective. As extreme events are predicted to increase and colonies decrease in size, a smaller huddle size may lead to failed incubation and poor body condition in over-wintering adults.

If we are to understand the drivers of emperor penguin populations and more effectively project their future status it is essential that studies on various potential drivers linked to warming temperatures are implemented. However, what is clear is that emperor penguin populations are already in decline.