Periodicity in Extreme Weather in the ‘Maritime Region’ of Eastern North America

Spectral and wavelet analysis were used to identify trends and cycles in extreme temperature and precipitation events based on historical data (~100-150 years) from six climate stations within the “Maritime Region” of eastern North America. Many statistically signicant climate cycles were identied using both spectral and Morlet wavelet analyses at each of these locations for both extreme high and low temperature and precipitation (rain, snow) data, with periodicities typically ranging from ~ 2–30 years. To assess potential drivers of these cyclical extreme weather events, the records of these events were compared, using cross wavelet analysis, to the climate indices of several teleconnections, including the 11-year Schwabe solar cycle, Atlantic Multidecadal Oscillation, North Atlantic Oscillation, Arctic Oscillation, El Niño Southern Oscillation and the Quasi–Biennial Oscillation. It was found that the 11-year solar cycle had the strongest inuence over extreme temperature and precipitation in this region, whereas the remaining oscillations, with the exception the Quasi–Biennial Oscillation, exhibited complex interactions with one another, characterized a variety of both positive and negative modulating effects. The Quasi–Biennial Oscillation was found to drive high–frequency oscillations in extreme weather, particularly extreme precipitation. Overall, the ndings of this study indicate that extreme weather events in this region have not substantially increased or decreased in number over time, but have been predominantly inuenced by several cyclic climate phenomena.


Introduction
Historically, extreme weather events have posed undeniable risk and consequence to both humans and the environment (Parmesan et al. 2000;Greenough et al. 2001;Znachor et al. 2008;Ummenhofer and Meehl 2017;van de Pol et al. 2017;Campbell et al. 2018; Smith and Sheridan 2019) and have been considered among the highest global risk factors in terms of both likelihood and impact (World Economic Forum 2019). Unlike seasonal weather changes, extreme weather events occur within considerably shorter time frames, rendering them nearly impossible to adapt to which, in turn, can be detrimental to human and ecological health (e.g. heat waves associated with increased emergency medical needs (Dolney and Sheridan 2006); storm-induced disturbances distressing phytoplankton communities (Stockwell et al. 2020)).
Changing patterns in the frequency and intensity of extreme weather events driven by anthropogenically driven climate change potentially represents yet another factor to be considered when developing future risk assessment scenarios. Indeed many studies have already reported evidence of increased frequency and intensity in extreme events (e.g. Rahmstorf and Coumou 2011;Gao et al. 2012;Wang et al. 2015), and based on these results computer modeling predicts a concomitant increase in medical and environmental impacts (Wu et al. 2014;Oliver et al. 2019). In contrast, other researchers suggest that natural cyclic climate oscillations within the relatively short historical climate record dataset is a primary in uence on the frequency of extreme events (e.g. Knight et al. 2006;En eld and Cid-Serrano 2010).
With models forecasting an increase in the frequency of extreme weather throughout North America and other mid-latitude regions, (Cai et al. 2014;Cohen et al. 2014;Wang et al. 2015) analysis of regional data subsets of extreme weather events provide an opportunity to speci cally assess more localized trends and cycles, and to ascertain whether they are widespread throughout a given region. Furthermore, such analyses provide an opportunity to directly compare instances of extreme weather to various documented cyclic climate phenomena to determine whether cyclical phenomena constitute localized or regional drivers of extreme weather events. Developing such a regional-scale understanding of the contributors to extreme weather is important to society as localized instances of extreme weather can result in ood damage, dangerous road conditions, increased risk of temperature-related health concerns (e.g. heat stroke, frost bite), and may ultimately be related to a myriad of health and safety hazards, as well as a broader ecological impact (Bouwer 2019).
Understanding whether there are trends and cycles in the occurrence of extreme weather events can also provide important information on changes in climate dynamics, an important metric when measuring overall climate change (Linnenluecke et al. 2012;Moazami et al. 2019). Analysis of extreme weather phenomena has also received increased scrutiny within the climate change research community in recent years as new information on the broad impacts of extreme weather has emerged (Ummenhofer and Meehl 2017;Bouwer 2019) This study examined trends and patterns in daily extreme high and low temperature and precipitation (rain, snow) in eastern North America, speci cally a regional subset informally dubbed here as the Maritime Region (MR; see section 4.3 below) and comprising historical weather records from 24 stations in New Brunswick, Nova Scotia, Prince Edward Island, eastern Maine and the Gaspé region of Quebec.
Historical records selected for detailed analysis in the MR comprised stations with near-continuous instrumental data records spanning the past 100-150 years. Spectral, Morlet wavelet and cross wavelet time series analysis techniques were then employed to detect trends and patterns within individual records, and to determine whether they correlate with trends and cycles in the records of known drivers of climate in the region (Bonsal and Shabbar 2011;Patterson and Swindles 2015).

Previous Work
Most previous studies that have examined extreme weather phenomena within or near the MR study area have not been carried out to understand the nature of extreme weather itself, but instead have principally either assessed the impact of these events on the environment factors (Payette et al. 1985;Scott et al. 2003;Boland et al. 2004;Andalo et al. 2005;Nye et al. 2009;Ugarte et al. 2010;Lemieux and Scott 2011), or have focused on the in uence of extreme weather on human health and safety (Andrey et al. 2003;Dolney and Sheridan 2006;Ziska et al. 2008;Balbus and Malina 2009;Cheng et al. 2012;Wu et al. 2014).
Amongst the few studies that have focused on the incidence of extreme weather phenomena Douglas and Fairbank (2010) for example, examined trends in extreme precipitation in New England, nding that the trend in extreme precipitation in New England was stationary (near-zero linear slope) from 1893-2005, although some select stations did shown a positive linear trend. In contrast, Frei et al. (2015) identi ed a pattern of decadal uctuations in the frequency of extreme precipitation events in the northeastern United States, and an increase in the occurrence of extreme precipitation events during the warm season (June-October) through the twentieth century. Other studies have reported on general trends in extreme weather across other parts of eastern North America outside the MR, presenting evidence of general increases in frequency or intensity of both extreme temperature and extreme precipitation (Bonsal et al. 2001;Zhang et al. 2001;Vincent et al. 2018;Tan and Gan 2017).
Cyclical Drivers Of Regional Climate Change Drivers of climate oscillations known to impact weather in eastern North America include the Schwabe sunspot cycle (SSC, ~11 years), Atlantic Multidecadal Oscillation (AMO, 50-90 years), El Niño Southern Oscillation (ENSO, 2-10 years), and Quasi-Biennial Oscillation (QBO, 2.1-2.5 years). The North Atlantic Oscillation (NAO) and Arctic Oscillation (AO) are also known to in uence eastern North American weather patterns, though these oscillations do not have well de ned periodicities.

Sunspot Cycle (SSC)
The SSC, which is characterized by the quasi-periodic rise and fall in the number of observed sunspots, cycles approximately every 11 years (9-14 year variability in duration; Lassen and Friis-Christensen 1995;Hathaway 2010Hathaway , 2015Jørgensen et al. 2019). Variations in the number of sunspot numbers are related to changes in total solar irradiance and solar variations have been shown to affect temperatures on both long and short timescales. Waple et al. (2002) through proxy record analysis demonstrated both longand short-term correlations between changes in solar irradiance and global temperature from 1650 -1850 A.D. A suite of mechanisms control the in uence of solar irradiance on the Earth's atmosphere, which can be simpli ed to two main mechanisms: dynamic changes in the troposphere caused by stratospheric ozone absorption of UV light, and tropospheric heating due to increased solar irradiance (Gray et al. 2010;Rind et al. 2008;Lockwood 2012).
Earlier studies of North American climate patterns have noted the presence of ~11-year cycles in temperature and precipitation that have generally been attributed to the in uence of the SSC. For example, Vines (1984) noted the presence of 9.5 and 11 year cycles in rainfall between 1880 -1960 in the Canadian Maritimes. Similarly, Prokoph et al. (2012) recognized an ~11 year cycle in stream ow records from the St. John River in New Brunswick and the Northeast Margaree River in Nova Scotia. Currie (1988) noted 10 -11 year cycles in air temperature in New England, as well as throughout the continental USA. Currie and O'Brian (1993) observed 10 -11 year cycles in New England precipitation records. In fact, the SSC plays a major role in climate variability across the globe. For example, Indian rainfall totals have indicated moderate correlation with the SSC (Ananthakrishnan and Parthasarathy 1984), and both ~11 and ~22 year cycles (22 years being the double sunspot cycle) have been recognized in Indian, African, and Australian precipitation patterns (e.g. Vines 1977Vines , 1980Vines , 1986Mazzarella and Palumbo 1992;Currie and Vines 1996;Thresher 2002).
Others have linked the SSC to short term variations in near-surface temperatures in various locations, though many of these studies have focused on latitudinal zones spread across often vast geographical distances Shea 1999 2000;van Loon et al. 2004;Maliniemi et al. 2014). However, the relationship of the SSC to regional variations in temperature have also been noted in several regional studies, including Romania (Sfîcă et al. 2018), China (Du et al. 2017), and North America (Currie 1993;Mendoza et al. 2001;Mendoza et al. 2006;Nalley et al. 2013).

Atlantic Multidecadal Oscillation (AMO)
The AMO is an oceanic alternation between warm and cool sea surface temperatures throughout the Atlantic Ocean, characterized by a quasi-periodic ~64-year cycle (~50-90 year variation) with about 30-35 years per phase (Knudsen et al. 2011). There is evidence that the AMO may be derived from an oscillatory component in the strength of North Atlantic thermohaline circulation (Dima and Lohmann 2007), which is characterized by a coherent pattern of warm and cold sea surface phases in the North Atlantic (Schlesinger and Ramankutty 1994). Negative (cold) AMO phases persisted from approximately 1900-1925 and 1965-1995 and positive (warm) phases characterized approximately 1925-1965-present (En eld et al. 2001; En eld and Cid-Serrano 2010; Knudsen et al. 2011).
AMO positive phases have been associated with several changes in climate, such as decreased precipitation, increased temperature, and droughts in North America (En eld et al. 2001;Hodson 2005, 2007;Nigam et al. 2011). However, in reference to extreme temperatures and precipitation, the AMO has a more prevalent impact on North America by modulating the effects of other climate phenomena (Mo et al. 2009;Zhang et al. 2012;Kang et al. 2014;Park and Li 2019;Zhang et al. 2019).
When isolated, signi cant in uence of the AMO by itself on North America often exhibits a coastal-inland gradation (Knight et al. 2006;Patterson and Swindles, 2015). In an analysis of lake ice out records from Maine and New Brunswick, Patterson and Swindles (2015) recognized signals of not only the 32-year AMO phase record, but also ~16-year cycles that they attributed to being secondary subharmonics of the primary AMO multidecadal signal.

North Atlantic Oscillation (NAO) and Arctic Oscillation (AO)
The NAO is a weather phenomenon caused by uctuations in atmospheric pressure at sea-level between two pressure centers in the North Atlantic; the Azores High and the Icelandic Low. Variation in this differential pressure in uences the strength and directionality of North Atlantic westerly winds, which in turn in uences temperature and precipitation patterns in the Atlantic Northern Hemisphere (Hurrell 1995;Olsen et al. 2012;Patterson and Swindles 2015). The NAO and to a lesser extent AMO have been shown to have the most signi cant overall impact on winter climate in North America (Burakowski et al. 2010;Hubeny et al. 2006) The related AO is a measure of broader change in sea-level pressure of Arctic air masses in the Northern Hemisphere, of which the NAO is a localized constituent of (Deser 2000;Ambaum et al. 2001;Olsen et al. 2012;Young et al. 2012;Patterson and Swindles 2015). Both the AO and NAO have similar known in uences on eastern North American weather during their positive (negative) phases, resulting in typically colder (warmer) temperatures, particularly during winter (Patterson and Swindles 2015;Yang et al. 2020). While both phenomena have relatively poorly de ned frequencies, they typically range from subdecadal to interdecadal, and are occasionally characterized by 20+ year multidecadal oscillations (D'Arrigo et al. 2003;Olsen et al. 2012). Steinschneider and Brown (2011), Coleman and Budikova (2013), and Armstrong et al. (2014) have observed linkages between the NAO and hydroclimate variations in the eastern North America. Li et al. (2017) correlated intervals of persistent cold spells in eastern North America to the in uence of the AO. Moreover, the AO is known to be associated with sea ice formation/melting in the Canadian Arctic (Rigor et al. 2002).

El Niño-Southern Oscillation (ENSO)
The ENSO originates in the equatorial Paci c where strengthening or weakening of equatorial trade winds in turn results in a strengthening or weakening of the upwelling of cold deep-ocean waters off the west coast of South America. This phenomenon has a far-reaching impact, in uencing much of the world's climate with both its warm (El Niño) and cool (La Niña) phases (Ropelewski and Halpert 1986;Rodbell et al. 1999;Moy et al. 2002). In eastern North America, the in uence of the ENSO is the result of the interaction of various atmospheric teleconnections (Bjerknes 1969;Diaz et al. 2001). Typically, only strong ENSO phases are re ected in weather patterns in eastern North America, leading to warmer, drier conditions during El Niño and cooler, wetter (snowier) conditions during La Niña, particularly in winter (Ropelewski and Halpert 1986;Harrison and Larkin 1998;Wolter et al.1999;Patterson and Swindles 2015;Yang et al. 2020).

Quasi-Biennial Oscillation (QBO)
The nal major climate/weather driver considered here is the QBO, which is a measure of the variation between easterly and westerly winds in the equatorial stratosphere. The QBO record is typically characterized by 2.1-2.2 and 2.3-2.4 year cycles, which have been shown to in uence various aspects of near-surface climate via polar, subtropical, and tropical teleconnections (Gray et al. 2018). In eastern North America, the westerly phase of the QBO often leads to cooler surface temperatures, particularly in winter (Baldwin et al. 2001;Chattopadhyay and Bhatla 2002;Patterson and Swindles 2015). The QBO has also been shown to impact ice break-up in North American lakes (Sharma and Magnuson, 2014;Patterson and Swindles, 2015). The timing of lake ice break-up each year is primarily a function of late winter air temperatures, with other factors such as distance from coast, elevation, wind direction and wind intensity also being contributing factors (Patterson and Swindles, 2015).
Interactions between the cycles described above and their resultant in uence on climate are welldocumented. For example, AMO is known to weaken the effects of ENSO, NAO, and AO during its positive phase and amplify them during its negative phase (Zhang et al. 2012;Kang et al. 2014;Park and Li 2019;Zhang et al. 2019;Chen et al. 2020). Other climate oscillations thought to in uence weather only weakly in uence climate in the region, or occur with frequencies of hundreds or thousands of years, well beyond the instrumental records examined in this study (Mann et al. 1995;Domínguez-Villar et al. 2017).

De ning extreme weather
Various methods have been used to record and analyze extreme events (e.g. event counts or incremental classi cation; statistical, time series, and geospatial analysis; modelling), each tailored to differing approaches to what de nes an extreme weather event (e.g. severe or unseasonal weather; weather at the limits of the historical distribution; e.g. Mondal and Mujumdar, 2015;Panda et al., 2016). Phenomena associated with the broad term "extreme weather event" can include, but are not limited to, rain or snowstorms, heat waves, cold snaps, droughts, oods, or hurricanes (Meehl et al. 2000;Stephenson et al. 2008;Huber and Gulledge 2011). Thus, a more speci c de nition of extreme weather in the context of this study is required. A common method for de ning an extreme weather event is a statistical approach using the 95th percentile of historical climate data collected from a speci c weather station or groups of stations (Rahmstorf and Coumou 2011;Gao et al. 2012). This concept can be simpli ed mathematically using the assumption that given aspects of weather (e.g. temperature, precipitation) follow a Gaussian distribution, with 95% of all weather events falling within two standard deviations of the mean (µ± 2σ, where µ is the mean and σ is the standard deviation; Frei et al. 2015;Wang et al. 2015). Beyond these two standard deviations, weather events are considered extreme. This method uses univariate data (e.g. daily temperature records) and is only suitable for isolated extremes in that data (e.g. one day of abnormally high temperatures). Here, this method is employed for single day events and is not suitable if one wishes to recognize longer and/or less isolated events, such as hurricanes, the effects of which can last several days or weeks. While multi-day events can be recognized and analyzed using this approach, corrections must be applied to ensure multi-day events are mutually independent (Frei et al. 2015).

Data and analyses
Daily-resolution weather (temperature and precipitation) data was obtained from the historical data archives of Environment and Climate Change Canada and the National Oceanic and Atmospheric Association (NOAA) of the United States for 90 weather stations with long high-quality records from eastern North America ( Fig. 1; stations listed in Appendix 1). The raw data spans various date ranges, typically beginning in the late 19 th to early 20 th centuries and continuing up to the 2010s, with the latest year analyzed being 2017. These 90 stations, which were distributed through the provinces of Ontario, Quebec, New Brunswick, Nova Scotia, and Prince Edward Island, and states of New York, Vermont, New Hampshire, Maine, Massachusetts, Connecticut, and Rhode Island, were subdivided using hierarchical agglomerative clustering (Sup. Fig. 1). Cluster analysis took into account the seasonal extreme weather event totals of each weather variable (maximum temperature, minimum temperature, rain, snow) for each given location, as well as latitude and longitude, using Ward's minimum variance linkage (Ward 1963).
This approach resulted in recognition of three distinct climatic regions (informally designated as the Continental Region, Central Region, Maritime Region (MR), Sup. Fig. 1), which re ects the diversity of subclimates found within this relatively large area of eastern North America.
This study focuses on the MR, which itself comprises distinct climatic subzones that are heavily in uenced by proximity to the moderating in uence of the Atlantic Ocean. Six stations within the MR were selected for detailed analysis based on the availability of long-running, high-quality climate records with no, or negligible missing data. They also provide a wide geographic distribution throughout the MR. MATLAB (MathWorks 2019) was used to separate each record by meteorological season (December-February, March-May, June-August, September-November). The means and standard deviations for temperature (high, low) and precipitation (snow, rain) were calculated for these four seasons for the entirety of the available weather records. For precipitation, means and standard deviations were only calculated for non-zero values, as days with zero precipitation values would unfavourably skew the data. Temperature and precipitation values that fell outside of two standard deviations from the mean (the 95th percentile) were considered extreme. The number of extreme events for each season and year were tallied for further analysis.
The variability of extreme weather events through time was analyzed by calculating linear best ts as well as by red noise spectral and wavelet time series analyses. Spectral and wavelet time series analysis provide complementary results. Spectral analysis provides a very good estimate of signal periodicity but presumes stationarity, while non-stationary wavelet analysis provides an indication of the distribution and interrelationship of signals with varying frequencies through time (Prokoph and Patterson 2004).
Red noise spectral analysis, a function based on Thomson's multitaper method (Dorothée 2020), was used to transform time-domain data into its frequency domain counterpart. As opposed to a basic Fourier transform, which provides a single spectral power estimate, this method utilizes several estimates of a power spectrum and averages them, eliminating potential biases that can arise when using a simple Fourier transform (Thomson 1982). Red noise and con dence levels were estimated using the Monte Carlo and chi-square methods (Schulz and Mudelsee 2002). Typically, a 95% con dence interval is used to validate statistically viable periodic trends, though slightly lower con dence intervals remain statistically signi cant. Con dence levels of 90% and 99% are also often used in statistical analysis.
Here, any periodic signals above the 90% con dence interval were considered signi cant, as this level still maintains data robustness (Zar 1998). Continuous wavelet transforms (CWTs) were applied to the extreme weather records, presenting the spectral information with the additional element of time, which provides information on the continuous or discontinuous nature of the periodic signals.
Cross wavelet transforms (XWT) were then carried out between the various extreme weather records and the six climate oscillations discussed in Section 3 (SSC (SILSO, 2021), AMO (Trenberth et al. 2021), AO (NCAR 2021), NAO (Hurrell and NCAR 2020), ENSO (Trenberth and NCAR 2020) and QBO (NCAR 2013)) using the MATLAB functions by Grinsted et al. (2004), yielding comparisons between the extreme weather records and climate oscillations. While a CWT is useful to analyze a single time series for oscillatory components, an XWT can be constructed to reveal the common spectral power between two time series (i.e. exposing the oscillatory components that both time series share at a given point in time; Grinsted et al. 2004). This transform is particularly useful for assessing the relationship between two cyclic time series, providing a visual representation of statistically signi cant correlated oscillations between two data sets. The XWT scalograms indicate the 95% con dence interval of statistically signi cant common oscillations (shown as areas of high spectral density outlined with a solid black line). The XWT analyses were carried out to assess the relationships that the various aspects of extreme weather have to various climate oscillations.

Linear regressions
The seasonal linear regressions for each intra-station extreme weather variable are typically characterized by near-zero to very shallow positive or negative slopes depending on the season and weather variable (-1.78-0.79 events/100 years; Sup. Fig. 1-4, Sup. Table 1). The shallowness of the slopes and variability in sign across the region indicates that there has been no consistent, long-term, regional trend in the periodicity of extreme weather phenomena in the timeframe spanned by the MR instrumental record. However, there do appear to be several trends in the periodicity of extreme weather events that are unique to individual or small groupings of climate stations.
In general, there has been an increase in regional seasonal variability of extreme rainfall events with notably consistent seasonal increases in extreme rainfall events in Houlton, ME and Moncton, NB. There has also been a consistent increase in extreme snowfall events in Moncton. The remaining stations display near-zero trends in extreme snowfall.
Four of these larger changes were correlated to summer extreme weather records, and three of these occurred at Nepisiguit Falls, which is characterized by the strongest weather linear trends of any of the studied stations. Nepisiguit Falls is experiencing declines in both extreme maximum and minimum temperature events for a given season, indicative of a narrowing of local climate variability (Sup. Fig. 1-4, Sup. Table 1).
For extreme maximum temperature events the St. Margaret's Bay, NS, station has been characterized by consistently positive slopes, whereas the extreme minimum temperatures recorded for the same station display consistently negative slopes. This step-wise increase in extreme warm temperatures and decrease in cold temperatures is consistent with a progressively warming local climate at St. Margaret's Bay.

Spectral analysis and wavelet transforms
Spectral analysis and CWTs display many statistically signi cant cycles at the 90% con dence interval and higher throughout the MR (Figs. 2-9). Across the six stations, there was variation in the observed cycles within each of the seasonal weather records, although regional trends and spatial gradients were observed between stations. Extreme maximum and minimum temperature data exhibited 2.8-3.2 year non-stationary cycles spanning all the seasons at nearly all stations (Figs. 2-3 and 6-7). This cyclicity is particularly strong at all stations for 1997-2000 autumn extreme maximum temperature (Fig. 6). Similarly, a 2.8 year cycle occurred during summer in the mid-1940s at all stations except for St. Margaret's Bay and Annapolis Royal, NS. Other cycles in extreme maximum temperature that can be seen across several, or all stations, are 5.1-6.6 years (spring, 1970s), 9-11 years (spring, 1940-1960), and 3.2-4 years (winter, 1940s). Cycles observed for extreme minimum temperature across nearly all stations include: ∼3.7 years (autumn, 1980s); 3.2-4.2 years (autumn, 1930s); 2.8-3.3 years (summer, ∼2000); 5.1-5.7 years (spring, ∼1940); and 3.9-5.3 years (winter, 1930s and 1960s).
Most stations are also characterized by statistically signi cant non-stationary extreme rain and snow precipitation cycles of 2-3 years across all seasons (Figs. 4-5 and 8-9). However, these cycles are not quite as well correlated across the entire region as was the case for extreme temperature.
There were non-stationary 3.4-3.7 year extreme rainfall cycles (spring, 1990s) observed at all stations, although this cycle was only statistically signi cant at some stations. There was also a 9.5-11 year nonstationary cycle (winter, 1970s-1990s) and ∼3 years (spring, 1910s) observed at Houlton and Moncton. The records at these two stations also included non-stationary 3.1-3.2 year cycles in extreme snowfall (spring, ∼2000). There was also a 2-3 year snowfall cycle (autumn, 1950s) present only at St. Margaret's Bay and Annapolis Royal.
Regional variation in MR precipitation cycles re ects the in uence of the Atlantic Ocean. The weather at stations in more coastal environments (St. Margaret's Bay and Annapolis Royal, NS) typically receive more precipitation as rainfall throughout the year, as well as being more likely to be impacted by the remnants of tropical storms than the inland stations. These two coastal stations have similar extreme precipitation records, quite distinct from the records from the other stations. More inland, continentalweather-in uenced Houlton and Moncton are also more similar to one another for the same reason.

Cross wavelet transforms
The The strongest observed relationship between the extreme temperature and precipitation data and cyclic climate drivers was an ~ 11 year cycle that correlates with the SSC (Sup. Figs. 5-9). This strong relationship with the SSC included all stations and all extreme weather types analyzed, and although nonstationary, in most cases extended throughout the entire instrumental record. Examples of nonstationarity included a discontinuity between the SSC and maximum temperature in summer (∼1960-1980) Moncton, Houlton, and St. Margaret's Bay, as well as during autumn (∼1930-1950 and ∼1970-1985) in St. Margaret's Bay and Annapolis Royal. Similar discontinuities were recognized in the SSCextreme minimum temperature relationship. The Causapscal, QC, and Nepisiguit Falls stations also display non-stationarity in the relationships between extreme temperatures and the SSC, though to a lesser degree. Extreme precipitation shows the highest degree of continuity in common spectral power with the SSC across the region.
The XWTs for AMO and each of the extreme weather variables (Sup. Figs. 10-13) were characterized by many small areas of weak common spectral power, although with extreme temperature and precipitation there was for many stations an area of high common power with a cycle of ∼16-30 years, typically centered upon ∼1950. This relationship was most apparent with winter and summer maximum temperature, summer minimum temperature and winter snowfall and was variable by location with rainfall, appearing most strongly in the coastal stations, St. Margaret's Bay and Annapolis Royal. Several occurrences of common ~4 and ~8 year cycles were also observed.
The XWTs for both the NAO (Sup. Figs. 14-17) and AO versus each extreme weather variable  indicate that these drivers have had a varying impact on extreme high and low temperatures across the region, with many intermittent occurrences of common cycles ranging from 2 -16 years. The climatic in uence of NAO across the region varies considerably between stations, with relatively few, and typically weak regional patterns. In contrast the XWTs between extreme precipitation and AO indicates that an abrupt climatic shift occurring around ~1965 for all stations, although prior to this part of the regional record the common spectral power between extreme precipitation and AO is sparse and weak. During the post-1965 interval, this relationship became stronger, with higher common spectral power between the signals. This shift is not as clear when AO is compared to extreme temperatures, though a clear regional pattern exists. While NAO does not show the time-varying pattern that the AO does, many of the strongest common cycles between extreme weather and NAO are similar to those shown with the AO.
The ENSO-extreme weather XWT results  are broadly similar to those patterns observed for the NAO and AO XWTs. In particular, the XWTs for ENSO and temperature extremes are strongly correlated between stations, with extreme precipitation showing an abrupt change in common spectral power beginning in ~1965, as occurred in the AO results. As observed with AO, areas of high common spectral density increased post-1965, although this trend is less pronounced. The abrupt shift correlated to ENSO is also apparent across the region with the winter and spring rainfall and spring snowfall extremes. The high and low frequency signals within the NAO, AO, and ENSO XWTs show distinct similarities to the resulting AMO XWTs, suggesting a relationship between the effects of these four phenomena and their in uence on extreme weather.
Each of the ENSO, NAO, and AO versus extreme weather XWTs are also characterized by a relatively strong 16-30+ year common cycle with various extreme weather records (e.g. winter maximum temperature -AO, Annapolis Royal (Sup. Fig. 18); summer minimum temperature -NAO, Moncton (Sup. Fig. 15)), reminiscent of the relationships found between extreme temperature and the AMO. This is most often found during winter, but these relationships are present in other seasons as well.
The QBO versus extreme weather XWT results  indicate that this driver has had a pulsed non-stationary in uence on extreme weather throughout the MR with the most signi cant in uence being on extreme precipitation, and less so on extreme temperature. The QBO XWT data is strongly correlated across the region with the notable exception of the Annapolis Royal station. With extreme maximum temperature, QBO has had a stronger in uence during winter, and has been weakest during summer. The relationship between QBO and extreme minimum temperature, is fairly uniform regardless of the season. The most common XWT periodicity observed with temperature has primarily been in the 2-3 year range, although particularly with extreme maximum temperature there is a semiconsistent 10-20 year cyclicity that spans from the beginning of the QBO record (1953) to 1980. This 10-20 cycle is even more strongly expressed with extreme precipitation, with the 2-3 year cycles described above still present.

Maritime Region (MR) extreme weather trends
Weather patterns in the MR of eastern North America are signi cantly in uenced by its proximity to the Atlantic Ocean, which moderates the climate there, particularly in coastal areas. However, areas only a few kilometres inland are signi cantly in uenced by continental climate conditions (Trewartha and Horn 1980;Patterson and Swindles 2015). For example, in summer the Atlantic coast is often shrouded in fog with temperatures of < 20°C, while it may be sunny and >30°C only a few km inland. Then in winter the coast may be snow free and rainy while areas only a short distance from the coast may be covered in a thick blanket of snow. Thus, considering the spatial distribution of weather stations used in this study, a gradient in climatic trends is fully expected. The gradient in climatic conditions was most signi cant between the two most coastal weather stations, St. Margaret's Bay and Annapolis Royal, and the remaining four stations, which are characterized by more continental climates (e.g. Fig. 7, Sup. Fig. 10).
Of the inland stations the climatic patterns characterizing Moncton and Houlton were most similar to each other, most likely related to the close latitudinal positioning of the stations at 46.09°N and 46.13°N, respectively. These results suggest that even at the small regional scale studied, latitudinal climate gradients impact the periodicity of extreme weather occurrences.

Linear trends in extreme weather events
Nearly all of the extreme weather records for the MR are, based on regression analysis results, characterized by fairly shallow slopes, with only relatively small changes, both positive and negative, in the periodicity of extreme weather through the past ∼100 years, with little spatial coherence. There were nine weather records in the region though that exceeded an increase/decrease threshold of 2 events/100 years. These records were primarily concentrated in the minimum temperature extremes, particularly for Nepisiguit Falls and Annapolis Royal (Sup. Table 1). With the vast majority of linear regressions in the extreme precipitation and temperature records for the MR being negligible through the span of the instrumental data record examined, the confounding in uence of natural climatic oscillations further complicate detection of any trajectory in these records. These results, at least for the MR, contrast with some recently published studies, which report that extreme weather events are becoming more frequent as global temperatures continue to rise (Perkins-Kirkpatrick et al. 2017;Francis et al. 2018;Keellings et al. 2018;Cowan et al. 2020; Perkins-Kirkpatrick and Lewis 2020).
6.3 In uence of Cyclic Climate teleconnections 6.3.1 Schwabe solar cycle (SSC) The prominent 9-12 year cycles that recurs throughout the extreme weather data for the MR, and the strong common periodicity in this range between extreme weather and the SSC observed in the XWT analysis results, indicates that this phenomenon is the most signi cant in uence on each aspect of climate analyzed in the region (Sup. Fig. 5-9).
The observed linkage between MR extreme weather records to the SSC is similar to the ndings of Laurenz et al. (2019), where the SSC was found to correlate with European rainfall totals. However, in that study it was observed that there was a moderate month to month correlation between variation in the strength of solar in uence on extreme precipitation, which in contrast to the results for the MR, were signi cantly reduced at the seasonal level. The wider geographic coverage, accompanied by a wider regional climatic variability, used by Laurenz et al. (2019) may explain the greater month to month variation between the two studies.
Additionally, Prokoph et al. (2012) attributed an ~11-year cycle in MR steam ow records to the SSC, although they also noted that this signal was comparably weaker than what was observed in rivers elsewhere in Canada. Moreover, the SSC is known to in uence other natural cyclic climate phenomena (e.g. QBO and ENSO (Quiroz 1981;Hamilton 2002;Kuroda 2007;Fischer and Tung 2008;Calvo and Marsh 2011;Zhou et al. 2013)), which could possibly explain the observed non-stationary relationship between the SSC and extreme weather in the MR, particularly the relationship with extreme minimum temperature.

Atlantic Multidecadal Oscillation
The XWTs comparing the AMO index and MR extreme weather data revealed a common ~16-30 year cycle recognizable in varying degrees in both temperature and precipitation data, which was widespread across the region between 1930 and 1970. This period range corresponds to a suspected subharmonic in uence of the AMO on this region; 16-24 year subharmonics of the AMO have been reported by Ruiz- Barradas et al. (2013) and have been found to in uence climatic trends in the MR, particularly those related to temperature uctuations, by Patterson and Swindles (2015). Furthermore, this 1930Furthermore, this -1970 interval, corresponding to a positive phase of the AMO, was most common in extreme temperatures and snowfall data from Nova Scotia stations, although extreme rainfall data also was characterized by this pattern, albeit with a weaker spectral power. While this effect was observed most strongly in the Nova Scotia stations, it appeared more weakly in the remaining MR stations, indicating a coastal-inland effect of the AMO in this region. This gradational effect can be seen in both extreme temperature/AMO XWTs (Sup. Figs. 10,11), in which the coastal Nova Scotia stations experience the strongest relationships, particularly in winter months.

NAO, AO, and ENSO
Several parallels were observed in the relationships between NAO, AO, and ENSO XWT results. Each of these oscillations exhibited common cycles of 2-16 years that appeared at similar times between the three oscillations. Furthermore, both ENSO and AO exhibited a shift from low common spectral power to high common spectral power at ~1965. These patterns suggest the NAO, AO, and ENSO share an interactive in uence on extreme weather in the MR.
From the XWT results, both ENSO and AO have a relatively uniform in uence on extreme weather across the MR, with similar signals observed across all stations for a given season and weather type. In particular, ENSO has the most signi cant in uence over extreme temperatures in the MR, particularly in winter and spring (Sup. Fig. 22-23). ENSO and AO collectively have a similar but more moderate in uence on extreme precipitation (Sup. Fig. 24-25, Sup. Fig. 20-21), with the NAO having a comparatively weaker in uence on this parameter. In contrast to ENSO and AO, the in uence of the NAO often has a higher spatial variation with respect to extreme weather throughout the MR (Sup. Fig. 14-17).
There is a notable common 16-30 year cycle observed in the XWT extreme temperature records for the NAO, AO, and ENSO, particularly during winter. These lower frequency cycles parallel similar XWT results observed with the AMO data. In addition, ~4 and ~8 year XWT links between the AMO and extreme weather records also parallel many of the signi cant relationships observed with the XWT derived correlations between extreme weather and ENSO, NAO, and AO. These resemblances suggest a strong interaction and signi cant modulation of ENSO, NAO, and AO by the AMO in the MR (e.g. Fig. 10 displays stark similarities between the maximum temperature XWT analyses for AMO, NAO, AO, and ENSO at Annapolis Royal, NS).
The interrelationship between AMO, NAO, AO, and ENSO is particularly strong in coastal areas, particularly in the respective extreme temperature XWTs. For example, the extreme maximum temperature record from Annapolis Royal record is characterized by a close similarity observed within each season between each of these four climate oscillations (Fig. 10). This close relationship between these oscillations is also observed at many inland stations, albeit generally less strongly (e.g. Moncton extreme rainfall, Fig.   11). This same pattern is also weakly observable in the extreme precipitation XWTs, although it is much more variable.
In the MR, this in uence was further supported by observed patterns within the AO-extreme precipitation XWTs. For example, the common periodicity between AO and extreme precipitation became ampli ed after 1965 following the establishment of an AMO-phase. A similar post-1965 ampli cation was observed in the ENSO XWT record, although these effects were more varied by season and weather type. The signi cant in uence of the AMO in modulating the effects of other cyclic climate phenomena (such as the NAO, AO, and ENSO) has also been previously documented elsewhere in North America, where it has been observed that this phenomenon suppresses the in uence of their effects during an AMO+ phase and amplifying them during an AMO-phase (Mo et al. 2009;Zhang et al. 2012;Kang et al. 2014;Park and Li 2019;Zhang et al. 2019).

The Quasi-Biennial Oscillation
In the MR the in uence of the QBO appeared strongest within the extreme precipitation record (both rain and snow) and was weakest within the extreme minimum temperature record. This result is partially contradictory to the spectral analysis results, where 2-3 year cycles appeared in nearly all extreme temperature records and, less commonly in the extreme precipitation records. A possible explanation for the discrepancy is that there is probable overlap between lower frequency QBO and higher frequency ENSO signals. This strong observed correlation between QBO and uctuations in extreme rainfall and snowfall patterns is also supported by research from elsewhere in North America and around the globe (Lau and Sheu, 1988;Brázdil Zolotokrylin 1995;Inoue and Yamakawa 2010;Seo et al. 2013). Similarly, several extreme rainfall records have been found to be characterized by quasi-biennial patterns, although in many cases they have not necessarily being linked to the QBO (Nastos and Zerefos 2007;Becker et al. 2008;Han et al. 2020).
The lower-frequency cycles observed in many of the QBO XWTs from 1953-1980 (e.g. Fig. 10, QBO row) are likely a result of modulation of what has been dubbed the quasi-decadal oscillation, which has been attributed to in uence of the SSC (e.g. Quiroz,1981;Hamilton, 2002;Fischer and Tung, 2008). It is hypothesized that changes in TSI through the 11-year SSC in uences movement of upper stratospheric air masses, where there are also strong SSC-linked changes in solar-ozone radiative forcing.
Overprinting on the QBO signal results in a quasi-decadal trend within in the QBO index. In the MR this modulation is seen most strongly in the pre-early 1980s QBO.

Conclusions
Through implementation of spectral, Morlet wavelet and cross wavelet analysis we have demonstrated that the occurrence of extreme weather (precipitation (snow, rain) and temperature (high, low)) in the instrumental record (~100-150 years) from six climate stations in the MR of eastern North America is largely cyclical, showing only localized evidence of either negative or positive trends. A regional gradation in extreme weather occurrence is recognizable when comparing coastal and inland/continental areas. It was found that the observed cyclic components of seasonal extreme weather are principally driven by several natural climate oscillations, the strongest of which is the ~11-year SSC. The SSC cycle exhibited both continuous and discontinuous impact through time in all seasons on the occurrence of extreme maximum and minimum temperature, rainfall, and snowfall.
Other cycles found to in uence the occurrence of extreme weather were the AMO, NAO, AO, ENSO, and QBO. The in uence of NAO, AO, and ENSO on extreme weather were modulated by negative and positive phases of the AMO, with distinct similarities between the four oscillations that would otherwise be unexpected. In particular the negative phase of the AMO appears to amplify the impacts of the AO and ENSO. Moreover, comparison of the AMO, NAO, AO, and ENSO records with extreme weather records in the MR indicates that there seems to be a complex interplay between them that causes some uniformity in their in uence, particularly on extreme temperature events through the 20 th and early 21 st centuries for a given station.
The QBO had a frequent but discontinuous impact on extreme weather in the MR through time, having a stronger relationship with extreme precipitation than extreme temperature. Furthermore, there was a modulating effect of the SSC on the QBO spanning from the 1950s to the 1980s, which was re ected in the XWT analysis for most stations in the MR. Red noise spectral analysis for extreme maximum temperature records at the six stations used in this study, with AR1 and various con dence levels indicated. All peaks above 90% con dence are labelled in years Figure 3 Red noise spectral analysis for extreme minimum temperature records at the six stations used in this study, with AR1 and various con dence levels indicated. All peaks above 90% con dence are labelled in Red noise spectral analysis for extreme rainfall records at the six stations used in this study, with AR1 and various con dence levels indicated. All peaks above 90% con dence are labelled in years Page 31/36 Red noise spectral analysis for extreme snowfall at the six stations used in this study, with AR1 and various con dence levels indicated. All peaks above 90% con dence are labelled in years Page 32/36 Figure 6 Continuous wavelet transforms (CWTs) for extreme maximum temperature records at the six stations used in this study Areas of high spectral density that are greater than the 90% con dence level are labelled with the cycle length in years Page 33/36

Figure 7
Continuous wavelet transforms (CWTs) for extreme minimum temperature records at the six stations used in this study. Areas of high spectral density that are greater than the 90% con dence level are labelled with the cycle length in years Page 34/36

Figure 8
Continuous wavelet transforms (CWTs) for extreme rainfall records at the six stations used in this study.
Areas of high spectral density that are greater than the 90% con dence level are labelled with the cycle length in years Continuous wavelet transforms (CWTs) for extreme snowfall records at the six stations used in this study. Areas of high spectral density that are greater than the 90% con dence level are labelled with the cycle length in years