Regional impact of large-scale climate oscillations on ice out variability in New Brunswick and Maine

General trends in lake ice phenology

While ice out dates can fluctuate by as much as 20 + days year over year, long-term averages in lake ice phenology are generally more consistent, only varying up to 10 days decade over decade (Fig. 2). On West Grand Lake, ice persisted latest in the spring on average from ∼1880–1890 before trending to earlier ice out dates until ∼1900, shifting later in the spring again until ∼1910. In contrast, average ice out dates for Oromocto Lake were relatively consistent from 1876–1910. From the mid to late 1910s onward, ice out on both West Grand Lake and Oromocto Lake shifted earlier in the year until the 1930s. Average ice out dates shifted later again through the next decade until the early 1940s on each of West Grand Lake, Oromocto Lake, and Skiff Lake before becoming earlier around 1950. These three lakes each exhibit a later ice out peak, spanning the entire record in the 1960s and 1970s before generally declining towards 2000 (Harvey Lake also exhibits this decline). Since the early 2000s, average ice out dates have generally shifted later in the year on all lakes except West Grand Lake. West Grand Lake ice out indicates the most substantial change between the late 19th century and the early 21st century, with the average ice out date lowering from approximately Julian Day 124 to Julian Day 115. Oromocto Lake does not show this same change, with average late 19th century ice out occurring prior to Julian Day 120 and early 21st century ice out occurring around Julian Day 115.

Lake ice out coherence and correlation

Oromocto Lake exhibited very strong coherence with each of the three other lakes in this study, particularly in the latter 20th century. The coherence with each lake spans interannual to interdecadal bands at the 95% confidence level (Figs. 3D, 3E and 3F), though notable non-significant coherence bands include ∼10 years (all lakes), ∼21 years (Skiff Lake), and ∼42 years (West Grand Lake). West Grand Lake has the most comparable record length to Oromocto. Prior to ∼1940, these records have less coherence and only a moderate correlation (r² = 0.355, p = 0.004, df = 62), but are strongly coherent and correlated afterwards (r² = 0.921, p < 0.001, df = 78; Figs. 3F and 4). The Skiff Lake record is similar in that correlation was initially lower (r² = 0.401, p = 0.038, df = 26) with coherence primarily at interdecadal scales but increased after 1960 (Fig. 3E), whereafter the records are coherent and strongly correlated (r² = 0.895, p < 0.001, df = 49). Harvey Lake has the shortest available ice out record and shows strong and significant coherence (Fig. 3D) and correlation with Oromocto Lake over the available dates (r² = 0.966, p < 0.001, df = 45, Fig. 4). The strong overall correlation and coherence of all three lakes with Oromocto Lake ice out justifies the use of the Oromocto Lake record as a representative sample for the region, particularly for the mid-20th century to present-day interval.

Periodicity in lake ice out

Figure 5 shows a periodogram and a continuous wavelet transform for the ice out record from Oromocto Lake. The time series is primarily characterized by interannual oscillations with periods ranging from 2.1 to 8.3 years. These oscillations are most significantly (>95% confidence) concentrated between approximately 1900–1940 and 2000–2015, but they also occur intermittently with lower significance throughout the ice out record, as recognized in both the MTM periodogram and the CWT. Some interdecadal oscillations are also present at a lower significance; ∼9–18 year oscillations persist from approximately 1940–1990 (significant at the 95% confidence level from approximately the mid-1940s to ∼1970), as well as a ∼23–34 year oscillation from 1920–2000.

The XWTs between the Oromocto Lake ice out time series and the records for seven known regional climate oscillations are provided in Fig. 6, while the XWTs for Harvey, Skiff, and West Grand lakes where observed large-scale climate oscillations can be viewed in Fig. S1. A comparison of the XWT results with the CWT and periodogram helps clarify which climatic oscillations best explain observed oscillations in the ice out record.

At the interdecadal to multidecadal scales, the PDO, SSC, and AMO are of particular interest. The PDO shows strong relationships with cycles within the 15–18-year band (most significantly from ∼1940–1960) as well as a 23–34-year oscillation from 1920–1965. The SSC is linked to ∼7–13-year oscillations in the ice out dates, most significantly from 1900–1980. Less significant relationships exist during times of weak quasi-decadal oscillations in ice out (e.g., 1995–2015). While the relationship for AMO is less significant (<95% confidence), there nevertheless exists an area of high incidence between the AMO and the ice out time series from approximately 1920 and 1980, with oscillations of 23–30 years. Furthermore, the beginnings of a 45+ year oscillation related to the AMO is just visible at the bottom of the AMO—ice out scalogram, indicating that the longer band of the AMO may have a relationship with ice out dates.

The most prevalent oscillations at the interannual scale are primarily associated with ENSO, NAO, AO, and QBO (Figs. 6F–6I). Both ENSO and NAO appear to have the strongest relationship with the interannual oscillations occurring between 1900–1920 (∼5–8-year oscillations). The AO exhibits a weaker relationship at this time-frequency location, though this may be influenced by the AO’s shorter temporal record. Each of these three climatic oscillations also exhibit significant relationships with the ice out record from approximately 2000–2015, in the 3–6-year return time range. The QBO primarily exhibits significant cyclic relationships with the ice out record at a period of 2.1–3.1 years (e.g., ∼1966–1968, 1980–1990, Fig. 6I). A 12–18-year common oscillation is also present between ice out and the QBO from approximately 1953–1975. While they are usually associated with interdecadal to multidecadal oscillations, the PDO and AMO also show some instances of significant (>95% confidence) relationships, primarily in the 3–8-year return time bands during the time intervals spanning 1900–1940 and 2000–2015.

Regional coherence and long-term patterns in ice out

The strong coherence and correlation exhibited between lake ice out records between closely spaced Harvey, Oromocto, Skiff and West Grand lakes (Figs. 3 and 4), and the similarity in long-term observed trends (Fig. 2) are indicative of regionally consistent patterns in ice out and ice out forcings. While localized factors such as the surrounding landscape, lake basin and shoreline characteristics, and wind exposure may influence some minor variation between ice out dates of these four lakes, the main drivers of ice out appear active at a regional scale. The long-term patterns in lake ice out in these lakes suggests only limited changes in year over year (20 + days) ice out variation since the late 19th century. While several shifts towards earlier or later average ice out have been recognized, a substantial change in average ice out since the late 19th was not strongly apparent (late 19th century average ice out occurred around Julian Day 120 (125) on Oromocto (West Grand) Lake whereas early 21st century ice out occurred around Julian Day 115 (115)). Rather, interdecadal shifts in ice out dates dominate the long-term variation, culminating in a shift toward later average ice out beginning in the early 2000s and continuing to present.

The general long-term pattern in lake ice out closely resembles the findings of Hodgkins, James & Huntington (2002) for ice out on northern New England lakes(which are latitudinally similar to the four lakes examined here) between the late 19th century to the year 2000. Hodgkins, James & Huntington (2002) note that lakes in the northern and mountainous regions of New England did not exhibit as strong of shifts toward earlier ice out dates in this time frame as lakes of southern New England did. The authors suggest that greater snow cover in the northern region decreases the sensitivity of lake ice cover to late winter regional air temperature, which is a primary driver of ice out. For the four lakes examined here, a similar moderation of ice out day may be occurring due to snow cover, subsequently decreasing the correlation to air temperature.

Prior to the mid-20th century, coherence and correlation between time series were lower though, for which a possible, but not confirmed explanation, may be related to a change in reporting methodology. For example, at Oromocto Lake, reporting of ice out dates have for decades been documented by citizen scientist Clayton Piercy who took over the duty of collating and recording ice out dates as a teenager in the 1940s. He used a single set of criteria (ice out declared when ice has disappeared from all coves) from the mid 20th century until his recent passing. The reporting of ice out on Oromocto Lake is now carried out collaboratively by many members of the Oromocto Lake Association using the same criteria as used by Clayton Piercy, but now aided by real time simultaneous social media monitoring of the entire lake. Prior to the mid 20th century, reports were made by many people at different locations on the lake, where based on examination of the data, ice out may have been declared when ice disappeared from the main body of the lake, and not slightly later when it disappeared from coves (Article S1). This unconfirmed hypothesis may explain the reduced coherence between the ice out record for Oromocto Lake and that from West Grand Lake through the earlier part of the ice out record.

Interannual variations

Ice out in the study area is primarily dominated by interannual oscillations in the ∼2–8 year range (Figs. 3 and 5). A correlation with these periodicities is most apparent in the scalograms with ENSO and NAO (Fig. 6). These oscillations have also previously been related to variations in temperature and precipitation in the region (Shabbar, Bonsal & Khandekar, 1997; Shabbar, 2006; Toniazzo & Scaife, 2006; Bonsal & Shabbar, 2011; Zhao et al., 2013; Patterson & Swindles, 2015; Nalley et al., 2019). The effects of ENSO and NAO on ice out are similar to one another, for which there are two main explanations. The simplest explanation is that the dominant frequencies of the two oscillations are similar. Both oscillations have a particularly strong 4–8 year band (Rasmusson & Carpenter, 1982; Kestin et al., 1998; Setoh et al., 1999; Pozo-Vázquez et al., 2001; Barbosa, Silva & Fernandes, 2006; Olsen, Anderson & Knudsen, 2012), which is abundant in the ice out time series. The alternate explanation is that there is a nonstationary relationship between the ENSO and NAO, in which the two climatic oscillations interact and covary depending on their phases (Zanchettin et al., 2008; Seager et al., 2010; Zhang et al., 2019). These interactions may amplify late winter/early spring temperature and precipitation anomalies depending on the state of the oscillations’ nonstationary relationship, each of which are correlated to ice out dates (Hodgkins, James & Huntington, 2002; Hewitt et al., 2018).

The AO exhibited a relationship to ice out similar to that of the ENSO and NAO, though it was generally weaker (∼8 year common oscillations, ∼1910; 3–8 year common oscillations, ∼2000–2015). The NAO is a geographically localized subset of the AO with both being products of oscillations in atmospheric pressure which, in turn, impact atmospheric circulation and the movement of airmasses (Thompson & Wallace, 1998; Ambaum, Hoskins & Stephenson, 2001; Rogers & McHugh, 2002; Thompson & Lorenz, 2004; Cohen & Barlow, 2005). While the AO has been found to influence ice out and other climate variables elsewhere in eastern North America (Bonsal et al., 2006; Wang et al., 2010; Mishra et al., 2011; Wang et al., 2012; Imrit & Sharma, 2021), its weaker influence on Oromocto Lake ice out may be a result of the AO’s broader geographic expanse, in comparison to the NAO.

The QBO exhibited a relationship to observed ∼2–3 year oscillations in ice out, which suggests that it plays a role in driving interannual ice out variability, albeit less importantly than for the ENSO and NAO though. The QBO has previously been attributed as the driver of ∼2-year variations in ice out along the east coast of North America (Sharma & Magnuson, 2014; Patterson & Swindles, 2015; Imrit & Sharma, 2021), and has been related to other climatic variations, such as those in temperature or precipitation (Lau & Sheu, 1988; Baldwin et al., 2001; Chattopadhyay & Bhatla, 2002; Becker et al., 2008; Cherenkova, Bardin & Zolotokrylin, 2015). The QBO is linked to the strengthening/weakening and the overall stability of the Arctic polar vortex. The strength and stability of the polar vortex impacts the movements of large-scale airmasses, thus impacting mid-latitude temperatures, strom tracks, and precipitation (Holton & Tan, 1980; Walter & Graf, 2005; Anstey & Shepherd, 2014; Overland & Wang, 2019).

Interdecadal variations

Interdecadal oscillations in ice out in these lakes were primarily correlated to the SSC and the PDO. While the quasi-decadal signal in the ice out time series was generally weak, it has a strong relationship to the SSC. This result suggests that while quasi-decadal oscillations are not always prominent in the ice out record, in the instances when they are present, they are primarily driven by the SSC. A strong relationship between the SSC and quasi-decadal oscillations has been previously observed for both ice out records (Sharma et al., 2013; Sharma & Magnuson, 2014; Fu & Yao, 2015; Hewitt et al., 2018) and temperature and precipitation records (Vines, 1984; Currie & O’Brien, 1988; Currie, 1993; Van Loon & Shea, 1999; Ogurtsov et al., 2008; Lockwood, 2012; Laurenz, Lüdecke & Lüning, 2019; Walsh & Patterson, 2022; Walsh & Patterson, 2022) due to its relation to total solar irradiance, which subsequently influences atmospheric, oceanic, and terrestrial heating (Yang et al., 2010; Kopp, 2014).

The PDO appears to be the primary driver correlated to 15–18 year and 23–34 year oscillations in ice out record, which is largely consistent with the prominent 15–25 year band of the PDO (Minobe, 1997; Minobe, 1999; Mantua & Hare, 2002; D’Orgeville & Peltier, 2007; Wu et al., 2011). Similar to what was observed with the quasi-decadal oscillations, these oscillations are not very prominent in the ice out record, but they have a strong relationship with the PDO when they are present. The correlation of the PDO to ice out is strongest prior to the 1977 PDO regime shift. Other PDO regime shifts (e.g., 1890, 1925, and 1947) are somewhat recognizable in the PDO–ice out relationships (Fig. 6), though not amongst the statistically significant correlations. PDO regime shifts, particularly the 1976–1977 “Great Regime Shift” from PDO+ to PDO-, have been recognized as contributors to abrupt changes in climate and ecological systems worldwide (Hare & Mantua, 2000; Scheffer et al., 2001; Litzow, 2006; Le Goff et al., 2007; Wang et al., 2014; Jacques-Coper & Garreaud, 2015), and have been noted for influencing ice phenology, primarily via atmospheric temperature changes (Magnuson, Benson & Kratz, 2004; Bonsal et al., 2006; Ghanbari et al., 2009; Mishra et al., 2011; Wang et al., 2018).

The AMO also exhibited a correlation to the 23–34 year band in ice out, but not at a 95% confidence level. The observed AMO–ice out relationship is similar to that of the PDO, though weaker overall, exhibiting a statistically insignificant correlation to the 23–34 year band in ice out. As the AMO influences Atlantic climate events and patterns and is characterized by sea surface temperature variation, a relationship to east coast ice out variation is not unreasonable (McCabe, Palecki & Betancourt, 2004; Alexander, Kilbourne & Nye, 2014; Wang et al., 2018; Abiy, Melesse & Abtew, 2019). This weaker but persistent AMO–ice out relationship could also be related to the PDO, as it mirrors the PDO’s relationship to 23–34 year oscillations in ice out. There exists a moderate correlation between the indices for the AMO and PDO (McCabe, Palecki & Betancourt, 2004; D’Orgeville & Peltier, 2007; Wu et al., 2011; Ogunjo & Fuwape, 2020; Zhang et al., 2020); thus, the similarity in their influence on ice out variation may stem from mutual interactions.

Furthermore, the beginnings of a 45+ year AMO relationship to Oromocto Lake ice out were exhibited in Fig. 6. This provides a preliminary indication that the primary multidecadal mode of the AMO likely influences ice out in the New Brunswick and Maine region; however, further validating this relationship with the methods used here is not possible due to the limited time frame of ice out data available. A longer ice out record is needed to robustly assess the common multidecadal signals of the AMO (50–90 years) and regional ice out.

It should be noted that the interdecadal oscillations are most strongly recognized in the ice out record and most strongly correlated to other oscillations within the cone of influence on each CWT or XWT. While this may be an accurate representation of the time series, it is also possible that the concentration of these interdecadal oscillations within the cone of influence could be attributed to edge effects, where the frequency information of the input signal(s) may be distorted outside of the cone of influence (Grinsted, Moore & Jevrejeva, 2004). The CWT and XWT techniques are limited in this sense. The timing of low frequency oscillations cannot be conclusively determined without longer data records.

Lastly, a brief instance of an interdecadal-scale relationship was depicted between ice out and the QBO between approximately 1953–1980. This result likely stems from a modulatory effect between the SSC and the QBO, reported in several studies (Quiroz, 1981; Hamilton, 2002; Fischer & Tung, 2008), which has not persisted since the early 1980s.