Recent changes in the diet and survival of Atlantic puffin chicks in the face of climate change and commercial fishing in midcoast Maine, USA

We present data regarding puffin productivity, chick diet, chick condition, and postfledging survival at three Maine colonies (Fig. 1). One of these, Matinicus Rock, is a relic population, surviving since the feather-hunting days of the late 19th century when most Maine puffin colonies were extirpated due to hunting (Drury 1973). The other two sites, Eastern Egg Rock and Seal Island National Wildlife Refuge, are restored colonies (Jones and Kress 2012). All three colonies are managed by the National Audubon Society, and puffins at all three colonies nest under large granite boulders.

Eastern Egg Rock (43°52′N, 69°22′W) is a 2.9 ha treeless island located in the mouth of Muscongus Bay off midcoast Maine, USA, approximately 9 km from the nearest mainland. The island consists of a central meadow surrounded by a perimeter of granite boulders. Its highest ground is only 7 m above the sea level. An estimated 150 pairs of Atlantic puffins nested here in 2015.

Matinicus Rock (43°47′N, 68°51′W) is a 7 ha treeless island with a 3 ha central meadow overlying a shallow peat soil with outcropping granitic ridges. It is located approximately 32 km from the mainland in the outer Penobscot Bay. By 2015, over 400 puffin pairs were estimated to be nesting on the island.

Seal Island National Wildlife Refuge (43°53′N, 68°44′W) is a 40 ha treeless island located about 14 km northeast of Matinicus Rock, 32 km from the mainland in the outer Penobscot Bay. The colony had about 500 puffin pairs in 2015.

We examined the prey items delivered to puffin chicks at three midcoast Maine puffin colonies (Eastern Egg Rock, Matinicus Rock, and Seal Island National Wildlife Refuge) over the years 2005–2014 to determine if the diets were similar at the three islands. Observers trained in fish identification recorded observations with binoculars and/or cameras from 1 × 1 m observation blinds. The blinds were located at a distance from where 10 or more active puffin burrows were visible within 5–20 m of the blind. Observers conducted 3-h observation stints at different times of the day that spanned the entire chick-rearing season from mid-June through early August, for a total of about 120 h/island/year. For each observed prey delivery, the species and estimated size of each prey item were recorded. In our analysis, we included only identified prey items; an average of 13% (range 0–30%) of prey items delivered each year were unidentified due to insufficient views of the feeding.

The efficacy of training observers to accurately identify the foods in puffin bill loads was independently demonstrated by Bowser et al. (2013). Additionally, to verify the composition of prey loads, we collected specimens dropped near nests and used photographs to confirm identification of prey. Photographs were taken with a Canon 7D SLR with a 300 mm telephoto lens equipped with a 1.4 extender.

Voucher specimens collected from chicks and near nests were identified by staff at the Sandy Hook Marine Lab, Highlands, NJ, the Maine Department of Marine Resources in Boothbay Harbor, ME, or the National Marine Fisheries Service, Northeast Fisheries Center. On laboratory examination of specimens obtained, pelagic-juvenile Urophysis proved to be white hake (Urophycis tenuis). Redfish were not identified to species in the bill loads delivered by parent puffins to their chicks; so we refer to this group as Sebastes spp. However, all collected specimens were identified as Acadian redfish (Sebastes fasciatus).

To determine the “prey signature” for each island, we added the number of specimens of each prey species delivered by all puffins and divided this by the total number of identified prey items (of all prey species) to arrive at a percent by number for each of the 10 years. We then computed the median percentage for each species observed. We used the median rather than the mean to reduce sensitivity to unusually high or low percentage values. We defined the “signature” of a puffin colony by ranking the three prey species with the highest median percentages. To see if the signature changed over the 10 year period, we compared the three most frequently delivered prey during two 5-year observation periods, 2005–2009 (early) and 2010–2014 (late).

We measured hatching success as the percent of eggs that hatched and productivity as the percent of nests that produced a fledgling. We examined whether hatching success or productivity had changed significantly over the years 2004–2014 for Matinicus Rock, and 2006–2014 for Seal Island. Nesting burrows are largely inaccessible at Eastern Egg Rock. Hence, this island was excluded from hatching and productivity studies. We began measuring productivity parameters at Matinicus Rock in 2004, conducting a direct observation of approximately 50–75 puffin nests each year, and followed the same protocol at Seal Island beginning in 2006 when the colony had grown large enough to sample a similar number of nests.

Burrows were checked for eggs, starting in mid- to late-May. Once an egg was found, burrows were not checked again until hatching was expected, and then they were checked every 4 days until hatch was detected. After hatching, burrows were checked periodically (every 1–2 weeks) until it was determined that the chick fledged or the nest failed. Due to annual variation in burrow occupancy and breeding activity, the number of active monitored burrows each year ranged from 37 to 78 per island. The final fate of a small number of nests that hatched late in the season each year could not be determined, and these were excluded from productivity calculations.

We analyzed data of all years at both islands and then compared an early period (2004–2009 for Matinicus Rock and 2006–2009 for Seal Island) with a later period, 2010–2014 at both islands. We chose 2010 as a break point using a statistical measurement called the total sum of squared errors (SSEs). We computed this based on two separate linear regression models, one for 2004/2006 through the “splitting year” (SY) and then the next year through 2014. We selected the splitting year that produced the lowest total SSE value as the best model fit. In this manner, the dataset was split into two sets, the first from 2004 through SY, and then from SY + 1 through 2014 for all years 2005 through 2014. Finally, with two islands, each of which featured an analysis of hatching and productivity success, there were four different splitting years to consider, and it was desirable to choose the same splitting year per analysis. The year 2010 was the best for productivity for both islands. For hatching success, 2010 was the second best for Matinicus Rock behind 2009, whereas 2010 was third best for Seal Island behind 2008 and 2009. Clearly the optimal overall splitting year is between 2008 and 2010; therefore, 2010 was selected because it produced the lowest total SSE twice, whereas the other two years only produced the lowest total SSE once each.

To test for changes in hatching success and productivity, we fit least squares linear regression models to both the early and later periods. We also used least squares linear regressions for the full study period. However, in the case of Seal Island productivity, we used median regression (quantile regression) due to outliers in the data. The explanatory variable (i.e., independent) was always the year, and the response variable was either the hatching success or productivity. We express the proportion between 0 and 1 because of the puffin’s one egg clutch.

We compared condition of chicks at Matinicus Rock and return rate (percent of cohort resighted) as a function of year for the period 1993–2009. Chick condition is reflected in the weight/wing chord ratio of puffin chicks measured during the linear growth stage (age 10–30 days, although most chicks were measured between days 21 and 30). We selected the linear growth stage chicks because puffin chicks typically lose weight during the final week before fledging, a fact that would affect the accurate comparison of weight/wing chord ratio between chicks. A total of 888 puffin chicks were included in this sample.

Chicks were removed from their burrows for banding with a USGS issued band in combination with a field readable band (USGS banding permit No. 20308). We measured unflattened wing chord with a 30 cm stopped wing ruler. Weight (mass) was measured by placing each chick in a cloth bag suspended from a 300 or 500 g capacity Pesola scale. The data that we used for chick condition is a weight/wing ratio, measured only during the linear growth phase. These data were obtained only at the time of banding, and thus, there was just one data point for each bird, rather than a series of data points over time. Hence, providing variance around this data point for each bird was not possible.

Most puffin chicks typically spend their first 2 or 3 years at sea before returning to land (Kress and Nettleship 1988), but some are not sighted on land until they are 5 years old. For Matinicus Rock puffins, we determined that at least 95% of surviving puffin chicks will have been resighted on land by 5 years after fledging. Because resighting effort was very thorough at all five puffin breeding colonies in the Gulf of Maine (the three study islands as well as the nearby puffin colonies at Machias Seal Island, New Brunswick, and Petit Manan National Wildlife Refuge, Maine), we believe that we resighted most of the returning puffins. We use the return rate of banded birds as a metric that approximates survival, an approach that is common in avian studies (Hamel et al. 2004; Oro et al. 2011). To allow for maturation at sea by some individuals, we did not include chicks weighed and measured during 2010–2014 period in this analysis of returning birds. The number of chicks in each banded cohort ranged from 37 to 105.

To determine if there was a difference in postfledging survival in relation to the chick condition, we used a T-test comparing the weight/wing cord ratio of 612 chicks that were later resighted at any breeding colony in the Gulf of Maine and 276 chicks that were not resighted.

To determine if there may have been a change in the trajectory of condition and/or survival over this 19-year period, we examined the data for a change point using the regression statistic R-squared (R ²). This measurement indicates how well the model fits the data, and values closer to 1 indicate better fit. The R-squared value was computed based on two separate linear regression models, one for 1993 through the “splitting year” (SY) and then the next year through 2009. We selected the splitting year that produced the highest R-squared value as the best model fit. In this manner, the dataset was split into two sets, the first from 1993 through SY, and then from SY + 1 through 2009, for all years 1994 through 2009. Because there are many data points and considerable dispersion, the R-squared value is naturally low (usually between 0.10 and 0.15).

Although the three puffin colonies investigated in this study are relatively close together, the constellation of prey delivered (i.e., prey signature) varied between islands, though several key species dominated the catch at all islands (Fig. 2).

We have analyzed changes (Fig. 3) and included statistics for those that made significant changes. Eastern Egg Rock puffin chick diet has remained relatively constant over the 10-year period (2005–2014), with white hake (Urophysis tenuis) comprising about 75% of the identified food items delivered to chicks throughout the study. Likewise, Atlantic herring (Clupea harengus) was a consistent second most important food throughout the 10-year study period. American butterfish (Peprilus triacanthus) doubled in frequency within chick diet from the first half of the study to the second (Z-statistic = −4.6610, p-value = 3.15 × 10⁻⁶) but still comprised only about 5% of food items in the later period. Redfish (Sebastes spp.), previously unknown at Eastern Egg Rock since 2005, comprised nearly 10% of food items in 2013 and 2014.

Matinicus Rock puffins fed their chicks predominantly on a diet of white hake throughout the study, especially in the first half of the study; herring and sand lance (Ammodytes dubius) are represented in the puffin chick diet by occasional “boom” years (e.g., Atlantic herring comprised 72% of food items in 2009 and sand lance comprised 42% of food in 2008). In the recent period, herring has declined from the third to fourth most frequently delivered food (Z-statistic = 2.6976, p-value = 0.0070); haddock (Melanogrammus aeglefinus), not even observed in the first half of the study, has increased in abundance to the second most frequently delivered food; butterfish (though still far below other foods) has increased to the third most frequently delivered food (Z-statistic = −2.3298, p-value = 0.0198). Also, redfish (Sebastes spp.) was not detected in the first half of the study, but it began showing up in the puffin diet in 2011 when it comprised about 24% of food items delivered to puffin chicks. Since 2011, it has remained a consistent addition to the diet.

Seal Island puffins delivered more Atlantic herring to their chicks in the first half of the study than other foods, but herring was replaced as the most frequently delivered food by white hake in the second half of the study (Z-statistic = −10.1377, p-value = 3.84 × 10⁻²⁴). As at Matinicus Rock, sand lance was important in certain years, but as it is highly variable, comprising 44% of food items in 2007 and 20% in 2014, it falls to less than 1% in 4 of the 10 years. Butterfish, not in the top four foods for 2005–2009, has increased to the third most common prey in the 2010–2014 period (Z-statistic = −13.8411, p-value = 1.50 × 10⁻⁴³). Also, as at Matinicus Rock, haddock first appeared at Seal Island in 2010 when it comprised up to 42% of food items and has remained a regular, though smaller part of the diet since 2010 (Fig. 3).

From 1993 to 2009, the weight/wing chord ratio of puffin chicks significantly declined by an average of 0.0305 units per year (T-statistic = −15.4411, p-value = 1.38 × 10⁻⁵⁰). The weight/wing chord ratio significantly declined by an average of 0.0663 units/year from 1993 to 2002 (T-statistic = −9.1908, p-value = 7.58 × 10⁻¹⁹) and by 0.0936 units/year from 2003 to 2009 (T-statistic = −10.2291, p-value = 1.24 × 10⁻²²) (Fig. 6).

The percent return of Matinicus Rock puffin cohorts (with individuals resighted at age 2 years or older) varies between years from about 40% to 90% with the lowest percent returns in the most recent years. The percentage of each cohort resighted has significantly decreased by 2.525 percentage points per year for 1993–2009 cohorts (T-statistic = −5.3013, p-value = 0.0001) (Fig. 7).

The mean weight/wing chord ratio is significantly different between chicks that survived and chicks that were not resighted (p < 0.001). On average, survivors had a 6.4% higher weight/wing chord ratio than nonsurvivors (T-statistic = 5.6182, p-value = 2.82 × 10⁻⁸) (Fig. 8).