Ocean–Atmosphere Variability in the Northwest Atlantic Ocean during Active Marine Heatwave Years

: The Northwest (NW) Atlantic has experienced extreme ecological impacts from Marine Heatwaves (MHWs) within the past decade. This paper focuses on four MHW active years (2012, 2016, 2017, and 2020) and the relationship between Sea Surface Temperature anomalies (SSTA), Sea Surface Salinity anomalies (SSSA), North Atlantic Oscillation (NAO), Geopotential Height anomalies (ZA), and anomalous Jet Stream positions (JSPA). Multichannel singular spectrum analysis (MSSA) reveals the strongest temporal covariances between SSSA and SSTA, and JSPA and SSTA for all years, particularly for 2020 (SSSA–SSTA: 50%, JSPA–SSTA: 51%) indicating that this active MHW year was more atmospherically driven, followed by 2012, which had the second highest temporal covariances (SSSA–SSTA: 47%, JSPA–SSTA: 50%) between these parameters. Spatial correlations for SSSA and SSTA between NAO during MHW active years disrupt the long–term (2010–2020) positive relationship in the NW Atlantic. SSSA and JSPA, and SSSA and SSTA were strongly correlated across the NW Atlantic; 2012 SSSA–JSPA correlations were strong and positive between 56–62 ◦ W, and 2016, 2017, and 2020 SSSA–JSPA correlations were mostly strong and negative, with strong positive correlations present near the coastline (70–66 ◦ W) or off the NW Atlantic shelf (52–48 ◦ W). SSSA–SSTA showed the opposite correlations of similar spatial distributions of SSSA–JSPA for all MHW active years. This indicates strong relationships between JSPA, SSSA, and SSTA during MHWs. Understanding the temporal and spatial interplay between these parameters will aid in better monitoring and prediction of MHWs.


Introduction
Marine Heatwaves (MHWs) are extreme warming events that have been increasing in duration, frequency, and intensity, posing adverse effects to local marine ecosystems [1,2]. They are defined by an event of Sea Surface Temperature anomalies (SSTA) above the 90th percentile of climatological seasonal SST variations for at least 5 continuous days [3]. MHWs are globally occurring events that have a variety of drivers [4][5][6] with notable events in the west coast of Australia (2010/2011) [7], Tasman Sea (2015/2016) [8], the Northeast Pacific (2015, 2016) [9], the Northwest Pacific (2021) [10], the Red Sea (2010,2017,2018,2019) [11], and the Mediterranean Sea (2020) [12,13]. The Northwest (NW) Atlantic experiences frequent MHWs, and these events have been caused by oscillations of the Jet Stream (JS) and maintained by its pertinent northward shift [4,14]. MHWs are difficult to predict due to their variety of spatio-temporal scales and the nonlinear physical properties associated with MHW formation [15]. Throughout the past decade, there have been four anomalously MHW active years (2012, 2016, 2017, and 2020) in the NW Atlantic. The 2012 MHW formation was induced by an anomalously northward JS through the autumn and winter of 2011-2012, with strong JS variability into the spring and summer,

Materials and Methods
To identify and quantify the location and timing of MHWs, the daily Optimum Interpolation Sea Surface Temperature (OISST) was used for our SST analysis. SST analyses used in this study were provided by NOAA's National Center for Environmental Information at a global resolution of 0.25 • . SSTAs are calculated with daily OISST minus a 30-year climatological mean at the same 0.25 • spatial resolution. Hobday et al. (2016) [3] recommends that SST data used in MHW studies have 30-year climatological baselines. OISST has been reliably used to identify and study MHWs [1, 3,4,6,13,16,24,27,28].
Debiased SMOS SSS Level 3 version 7 maps generated by LOCEAN/ACRI-ST Ocean Salinity Center of Expertise for CATDS provided the salinity data from January 2010 to November 2021 to calculate the SSSAs for this study. The V7 maps are provided every 4 days, and the debiased SSS are temporally averaged using a slipping Gaussian kernel with a full width at half maximum of 9 days for the 9-day product. Maps are at a spatial resolution of 25 km with an applied mean over neighbor pixels at less than 30 km for the purpose of producing debiased SSS maps. To account for the differences in temporal and spatial resolution between the debiased SSS and OISST, the SMOS SSS maps were re-gridded to OISST resolution.
All atmospheric parameters used in this work (zonal and meridional winds at 300 hPa, geopotential height at 300 hPa, and monthly averaged surface thermal radiation) are courtesy of the European Centre of Medium-Range Weather Forecasts (ECMWF). We use their newest product (ERA5), ECMWF's fifth generation reanalysis, which improves upon the previous ERA-Interim dataset with better temporal and spatial resolution, more available parameters, and a variety of new instruments and reprocessed datasets. Like the OISST, we use 0.25 • daily files and the 300 hPa pressure level to observe the upper atmosphere including the JS. We are confident in this dataset's capability for the study of MHWs, as shown in Perez et al. (2021) [16] and Schlegel et al. (2021) [24].
The National Weather Service's Climate Prediction Center NAO Index was used in this work to determine the state of the NAO. This daily index is computed by applying Rotated Principal Component Analysis to 500 mb heights north of 20 • N with respect to the 1950-2000 climatological mean, and anomalies are with respect to this mean.
General Bathymetric Chart of the Oceans provided the global terrain model, GEBCO_2021 Grid, used to interpret spatial correlation maps and dynamics within the NW Atlantic [29]. GEBCO_2021 provides elevation data, in meters, on a 15 arc-second interval grid. A Type Identifier (TID) grid accompanies the GEBCO Grid, where corresponding grid cells indicate the source data for that cell with a variety of direct and indirect measurements.
An MHW is a prolonged discrete anomalously warm water event. Hobday et al. (2016) [3] were the first to define an MHW as an event with temperatures that exceed the 90th percentile of an established climatology (recommending a climatology based on at least 30 years) for five or more days. This definition has been adopted by the literature [2,4,16,17,28,30] on MHWs. Using literature that cited Hobday et al. (2016) [3] to define multiple or long-lasting MHW events in the NW Atlantic, we identified four MHW active years, 2012, 2016, and 2017. The 2012 MHW is considered the most extreme year, with SSTAs in the GOM with over 100 days above the 90th percentile; the 2016 MHWs lasted over 4 months over 2 events with conditions consistently above the 90th percentile from January to mid-April and September to late November [2,16,24]. In 2017, a similar pattern of split MHWs occurred with February through April having strong positive SSTAs [17]. MHWs from 2020 are not well studied in the NW Atlantic, but as they have record ocean temperatures and ocean heat content for the NW Atlantic in a region with frequent MHWs, it is a year important to include in this study [4][5][6]18,19]. The region selected for the NW Atlantic (40-48 • N, 48-70 • W) focusing on the Center Shelf, Center Slope, North Shelf, and North Slope are as outlined in Perez et al. (2021) [16].
To determine the location of the Northern Hemisphere Jet Stream (JS) to the GOM, we use the same methodology as Belmecheri et al. (2017) [31]. The JS in the northern hemisphere is often fragmented over land but becomes a single stream over the North Atlantic, therefore making a single-stream index more accurate [31][32][33]. The Jet Stream Index (JSI) identifies the latitude of the strongest 300 hPa wind speed at each available longitude [34]. To remove the seasonality of the JS, the JSI was calculated daily for each longitude from 1981-2020 and averaged to calculate the daily climatological mean JS position. The Jet Stream Position Anomaly (JSPA) is with respect to this position. The seasonality of daily geopotential height at 300 hPa, monthly surface thermal radiation (STR), and SST was removed when the respected climatological mean from 1981-2020 was subtracted from the data, providing the geopotential height anomaly (ZA), STR anomaly (STRA), and SSTA. SSS anomalies (SSSAs) were calculated from the available SMOS data, 2010-2021.
Multichannel singular spectrum analysis (MSSA), which is an ideal method for analyzing temporal and spatial correlations between different time series related to climate variability or ocean-atmosphere dynamics [35,36], was applied to the time series of each parameter averaged in the NW Atlantic focused box (40-48 • N, 48-70 • W). The bounds of this box were determined to include the Center and North Shelf and Slope around and outside biodiverse areas, such as the GOM and the Gulf of St. Lawrence (GSL) [16]. More detailed steps of MSSA are included in Grusszczynka et al. (2019) [36], Groth and Ghil (2011) [37], and Zotov and Scheplova (2016) [38]. Two time series are combined in a trajectory matrix, where the corresponding eigenvectors and eigenvalues are produced by applying Singular Value Decomposition. With the known original structure of the matrix, the Principal Components (PCs) are reconstructed, where each PC mode decreases in amplitude, representing the time series variability [36]. With seasonality removed from each time series by using the calculated anomalies, the variance of the PCs will be reduced, but the first PC having a high variance indicates a strong correlation in the variability of each time series. Seasonally averaging SSTAs and the position of the JS, during the formation and decline of MHWs, removes the noticeable features of these events as these parameters have a high temporal variance. To account for this, six dates for each year were chosen to show days prior to onset, early onset, close to peak, and declines in the MHW active years while also being seasonally comparable to other years. The dates were chosen in January, March, May, July, September, and October.           Figure 5 shows persistent and strong negative SSSA anomalies throughout the NW Atlantic in 2012. This is likely from the shift in ocean circulation from changes in NAO and associated wind stress curl patterns which caused a freshening event to occur from       Figure 9 shows the spatial correlation coefficients between daily SSTA and NAO (Figure 9a,c,e,g,i), and SSSA and NAO (Figure 9b,d,f,h,j) for 2012 (Figure 9a   In contrast to 2012, 2016 SSSAs were more positive throughout the year ( Figure 6). Within the confines of the NW Atlantic box used in this study, positive SSSAs were more prevalent in the southeast edge of the box, away from the coastline and freshwater inflow. Figure 6a shows January to have a widespread positive SSSA with an average of 0.17 psu and a maximum of 1.99 psu. Figure 6b  It was found that 2020 had more persistent and stronger SSSAs in the NW Atlantic than all other MHW years (Figure 8). Positive SSSAs were also persistent outside of the GSL and to the northeast of the boxed region ( Figure 8). From the beginning of the year to late spring, a strong negative SSSA moved westward into the GOM and increased in size, with the center of this water body staying between −1.40 to −1.80 psu (Figure 8a-c).  Figure 9 shows the spatial correlation coefficients between daily SSTA and NAO (Figure 9a,c,e,g,i), and SSSA and NAO (Figure 9b,d,f,h,j) for 2012 (Figure 9a Figure 10a shows the 2012 time series where SSSA was very low for most of the year with peaks into positive anomalies occurring with and slightly prior to SSTA peaks in April and July to November. NAO also followed this pattern with positive peaks occurring with SSSA peaks (Figure 10a). In early 2012, while JSPA does not become positive, NAO peaks were preceded by peaks in JSPA, but this pattern stopped occurring after May, as the JS remained south of the climatological mean in the summer and autumn as seen in Figure 1d-f (Figure 10a). Figure 10b shows the 2016 time series.

Spatial
SSSA was more variable and had mostly positive anomalies for 2016. Mid-January to mid-March there was a small MHW where SSTA, SSSA, and NAO had a high plateau before dropping; in early January, JSPA was positive before this event occurred (Figure 10b). Other peaks of high SSTA and SSSA occurred in July to August and November to December, and NAO had a peak prior to these events (Figure 10b). These peaks were connected by constant SSTAs above 1 • C; this was where JSPA reached its maximum decline for 2016 (Figure 10b). Figure 10c shows the 2017 time series. The beginning of the year had high SSSAs which SSTAs were decreasing, becoming negative, and NAO slowly decreasing with oscillations through spring (Figure 10c). During this time, JSPA was oscillating just below zero (Figure 10c). Later in the year, July to August and November to December, peaks of NAO were followed by higher SSTAs and SSSA peaks (Figure 10c). Figure 10d shows the 2020 time series. JSPA from January through April experienced a positive peak monthly before rapidly moving south of the climatological mean for the rest of the year (Figure 10d). NAO appeared to follow the reverse of JSPA, with negative peaks while JSPA was positive (Figure 10d). Once JSPA steeply dropped in mid-May, SSTAs increased through the end of the year, while reaching a brief peak in August (Figure 10d). SSSA began close to zero and slowly increased from July to November with elevated SSTAs (Figure 10d). To focus on how SSTA and atmospheric parameters vary temporally, Figure 11 (Figure 11c), and July 2020 (Figure 11d). ZA oscillated around zero without a noticeable increase or decrease, but did experience larger positive peaks during elevated SSTAs, indicated by the light gray boxes (Figure 11).
Time lag composites show the spatial cross-correlations from 48-70 • W within the NW Atlantic. Figure 12    Time lag composites show the spatial cross-correlations from 48-70°W within the NW Atlantic. Figure 12 shows the time lag composites between atmospheric parameters and SSTA. Cross-correlations between JSPA and SSTA (Figure 12a   Time lag composites show the spatial cross-correlations from 48-70°W within the NW Atlantic. Figure 12 shows the time lag composites between atmospheric parameters and SSTA. Cross-correlations between JSPA and SSTA (Figure 12a d,m,q), thin contours were added. Figure 13 shows time composites between SSSA and the other parameters. From JSPA and SSTA having such a strong negative correlation, the strong correlations SSSA  Using the variance of PCs calculated in MSSA, the temporal relationship between the parameters, JSPA, SSSA, SSTA, and NAO, can be compared across years ( Figure 14). The first PC should have the most variance, as it was reconstructed to model the most Using the variance of PCs calculated in MSSA, the temporal relationship between the parameters, JSPA, SSSA, SSTA, and NAO, can be compared across years ( Figure 14). The first PC should have the most variance, as it was reconstructed to model the most oscillations between the parameters. The higher the variance of this first PC, the stronger the temporal covariance pattern will be [36]. Figure 14 shows pairs of parameters JSPA-SSSA (Figure 14a-d (Figure 14k), and 24% for 2020 (Figure 14l). For every parameter pair, 2016 had the least temporal covariance indicating that this MHW active year had other drivers that were likely oceanic-related, creating these MHWs. For SSSA and SSTA paired with NAO, 2012 had the highest covariance compared to the other years, indicating that NAO variability played a role in the intense MHW formation that occurred.

Discussion
The NW Atlantic has experienced four MHW active years within the past decade. Individual studies have identified 2012, 2016, and 2017 to be characterized by longer duration MHW events with severe quantifiable impacts on the ecosystem and fisheries. Global temperature and ocean heat content records identified 2020 as one of the warmest years in the NW Atlantic, experiencing conditions experienced in the other MHW active years. To compare the air-sea interactions of these four years, we used oceanic and atmospheric parameters (SSTA, SSSA, JSPA, ZA) and the NAO index. Using spatial maps of SSTA and SSSA, we were able to observe the formation of long duration MHWs in the NW Atlantic for the identified MHW active years with reference to the position of JS and its climatological position (Figures 1-8). Time series also presented observations of MHW parameters with NAO ( Figure 10) and atmospheric parameters with SSTA ( Figure 11) throughout the active MHW years. MHW events in 2012 began in the GOM from April to May (averaged maximum = 1.75 • C) and formed a large event spread into the NW Atlantic from July to November (averaged maximum = 2.50 • C) (Figures 1 and 10a). This year also experienced a prolonged northward position in the first half of the year which was not present during the other MHW active years [14]. Salinity for the 2012 MHW active year was negative for most of the NW Atlantic due to circulation shifts from NAO [39], but experienced SSSAs up to 1.68 psu (Figure 5d). The events in 2016 occurred from January to March (averaged maximum = 1.0 • C), July to August (averaged maximum = 1.90 • C), and November to December (averaged maximum = 2.0 • C) (Figures 2 and 10b). SSSA had an average maximum of 0.25 psu for each MHW event in 2016 (Figure 10b). The events in 2017 occurred from July to August (averaged maximum = 1.0 • C) and November to December (averaged maximum = 1.0 • C) (Figures 3 and 10c). The formation of MHW in 2017 in the Middle Atlantic Bight was attributed to a Warm Core Ring [17], which heavily influenced SSSAs in this region in the beginning half of the year but did not exceed 0.25 psu during the identified MHW months (Figure 10c). The events in 2020 occurred mid-June to October (averaged maximum = 1.95 • C) (Figures 4 and 10d). SSSAs in 2020 stayed negative near the coastline while positive SSSAs increased in the east side of the NW Atlantic with an average maximum of 0.20 psu, and peak maximums reaching 1.96 psu (Figures 8e and 10d).
To understand the dynamical relationship between these parameters during MHW years, we used spatial correlations, lag-lead correlations, and MSSA PCs' variance for combined time series covariance. Figure 9 shows how during MHWs, NAO variability had an influence over SSTA and SSSA. In a long time series (2010-2020) (Figure 9i,j) these correlations had weak correlation coefficients but experienced a slight positive relationship in the NW Atlantic. During the MW active years, there was an increase in correlations and a disruption of their distribution. While major increases did not occur within the focused region, NAO increased positively with SSTA in the North Atlantic and west along the coast for SSSA. Figures 12 and 13 show lead-lag days across the NW Atlantic for each pairing of parameters. Figure 14 shows PCs from MSSA and each combined time series covariances. As JS position is a previously known driver of MHWs in the NW Atlantic, strong relationships between JSPA and SSTA were expected to be seen (Figure 12a-d) [14]. The correlations between JSPA and SSTAs being strong negatives in Figure 12 were likely from the preconditioning that the JS position caused for the MHW, which occurred outside the laglead time relevant to the year, and the climatological position of the JS being very variable and seasonal, where the daily position anomaly became negative frequently [14,15]. Salinity and ocean temperatures experienced a similar relationship when MHWs form as shown in SSSA and SSTA (Figure 13a-h). Due to JSPA and SSTA's strong negative relationship, the relationship between SSSA-JSPA and SSSA-SSTA is reversed (Figure 13a-h). Covariances of these 3 parameters were the 3 highest overall values, JSPA-SSTA, SSSA-SSTA, and JSPA-SSSA ( Figure 14).
The 2012 MHW active year had a unique relationship with NAO, likely related to the shift occurring from 2012-2016 [39], as Figure 12m,q show all positive or negative correlations between JSPA and SSTA, respectively. This is further shown in the PCs' covariance, where 2012 had the highest value compared to the other MHW years for SSSA-NAO and SSTA-NAO (Figure 14i,q). This shows the 2012 MHW was influenced by NAO variability along with JS variability as found in Chen et al. (2014) [14].
In 2020, the highest covariance of both JSPA-SSTA and SSSA-SSTA (51%, 50%) was observed, followed by 2012 (50%, 47%). The 2012 event was considered to be a MHW that was predominantly forced by atmospheric processes [14], indicating that the 2020 MHW is likely predominantly driven atmospherically. While 2020 was not as strongly influenced by NAO as 2012 was, the spatial correlations (Figure 9g,h) show a similar distribution where positive correlations increased along the shelf break.
In 2016 and 2017, more advectively formed MHWs [16,17] were observed where atmospheric forcing possibly aided in continuing their duration. While outside of the focus area, there were wider-spread positive correlations for both SSTA-NAO and SSSA-NAO in 2016 and 2017 (Figure 9c-f). Within the NW Atlantic, there were positive correlations along the shelf break, and between the shelf break and the coast (Figure 9c-f). This relationship was likely due to NAO influencing Warm Core Rings and MHW events occurring as these eddies crossed the shelf break. Figure 12b,c differed from Figure 12a,d by having strong positive correlations east of 50 • W. This shift in correlation was likely from 2016 and 2017 being more advective, and the MHWs' influence on SSTA occurring closer to the coast. The highest value of covariance between JSPA and SSSA was in 2017 (39%), indicating the advective MHW with its changes in salinity was influenced and intensified in the NW Atlantic by the JS. Finally, 2016 had the lowest covariance values for each paired parameter indicating that this MHW had predominantly oceanic forcings.

Conclusions
As MHWs continue to increase in frequency and intensity, understanding patterns of their formation is key to future prediction and monitoring of these events. Using atmospheric and oceanic parameters, the spatial and lead-lag correlations, and temporal covariance, we observed the variability occurring in the MWH active years 2012, 2016, 2017, and 2020 in the NW Atlantic. Spatial maps displayed the changing and propagation of the MHW active years in SSTA and SSSA, as the JS shifted position throughout the year. Each year experienced unique formations, but as positive SSTAs strengthened and became widespread, SSSAs became more extreme within the NW Atlantic. SSTA and NAO, and SSSA and NAO spatial correlations were not widespread with strong correlations located in small regions along the NW Atlantic shelf or slope. These small regions shown in the spatial correlations are essential locations where NAO affects MHW temperatures and salinity.
Due to JS position being a major driver in the NW Atlantic, and SSTA and SSSA related to characteristics of MHWs, lead-lag correlations were very strong between JSPA-SSTA and SSSA-SSTA. JSPA-SSTA and SSSA-SSTA also had the strongest temporal covariance as seen in the variance of PCs from MSSA. The highest covariance for JSPA-SSTA and SSSA-SSTA was observed in 2020 indicating that this MHW active year had the most forcings being atmospheric-related, much like 2012 which was known as a very atmospherically driven MHW [14]; 2012 had the second highest covariances for JSPA-SSTA and SSSA-SSTA, and the highest covariances for SSTA-NAO and SSSA-NAO, indicating 2012 was heavily influenced by the NAO index in addition to the JS. Finally, 2016 had the lowest covariance for each correlation pair, indicating that this MHW active year was not primarily atmospherically driven and instead was maintained by oceanic processes [16].
These parameters having strong correlations is not a new discovery; as temperature has been used to determine MHW events, JS are known drivers of MHWs in the NW Atlantic, and salinity is affected by other drivers of these MHWs [4,14,17,24]. The relationship of these parameters can be used to develop better monitoring of MHWs, as it is critical to understand the indicators and development of MHW activity. Future work quantifying the relationship of SSS and SST variability with atmospheric parameter variability, such as JS position, in the NW Atlantic is necessary to understand the formation of MHWs. More research on air-sea responses to other MHW drivers, such as ocean currents, is also needed in creating models designed to monitor and track MHWs. As accurate models can be expanded and applied to MHWs occurring globally, more research is important for being able to deploy relief measures for key ecosystems and fisheries.