1. Introduction
Wind-generated surface ocean waves are an illustration of the atmospheric forcing on the ocean, where steady surface winds drive wave growth and subsequent swell propagation across the ocean (e.g., [
1]). The North Atlantic wave climate has the largest seasonal variation among ocean sub-basins, which results from a combination of highly variable seasonal wind forcing and Atlantic basin geometry that prevents the penetration of year-round Southern Ocean swells [
2,
3,
4]. The largest (smallest) wave heights and periods are observed in winter (summer) due to increased (decreased) average westerly wind intensity [
2]. Climatological wind intensity is modulated by the meridional pressure gradient between the subtropical anticyclone and the subpolar low, the pressure gradient itself being driven by the temperature gradient between the tropics and high latitudes, which peaks in boreal winter.
While the atmospheric processes that drive ocean waves are mostly restricted to specific latitudinal bands within the North Atlantic sub-basin (e.g., mid-latitude storm tracks, easterly trade winds, and tropical cyclones), waves can propagate over long distances and in various directions in the form of energetic, long-period swell (e.g., [
5]). In particular, boreal wintertime mid-latitude storm activity to the east of the North American continent occasionally triggers large waves that can travel as far east as the European and West African coasts, and as far south as the Eastern Caribbean islands and South American shorelines [
6,
7], where they represent a hazard to coastal populations and can cause significant damage to coastal ecosystems and infrastructure [
8,
9]. This hazard may be further amplified in the future as small island states, such as the Caribbean islands, are especially vulnerable to climate change-driven marine hazards such as sea level rise, changes in tropical cyclone activity, and increasing North Atlantic wave heights [
10,
11,
12]. Indeed, the climate and topography of the Caribbean islands have favoured the development of social and economic activities that strongly rely on tourism along the exposed coastal area. Despite notable improvements in the prediction, management, and mitigation policies of both present and future marine hazards, most of the literature focuses on extreme waves caused by tropical cyclones (e.g., [
11,
13,
14,
15]). Few published works are dedicated to understanding extreme wave climatology of extratropical origin in the Caribbean region (e.g., [
16]).
A few case studies of swell propagation from distant sources have been conducted for the Caribbean [
9], most of them focusing on the March 2008 extreme swell event [
6,
7,
8]. This event resulted from a deep low-pressure system located offshore of the North American east coast (
Figure 1). Jury (2018) suggests similar atmospheric conditions drive other major such north swell events in the Caribbean islands, consistent with dominant wintertime northerly swell direction in the tropics and despite the year-long presence of easterly trade winds [
2]. A more systematic attribution of the characteristics of the atmospheric drivers of such events is, however, lacking.
The present study, therefore, focuses on these wintertime north swell events, propagating southward and reaching the Lesser Antilles, to further understand the atmospheric conditions that generate them, based on a global coupled atmospheric/wave reanalysis. The main objectives consist of (i) identifying historical north swell events that reached the eastern Caribbean based on swell height, period, and direction; (ii) determining the mid-latitude synoptic atmospheric patterns that generate these events; and (iii) assessing why some synoptic events generate waves that affect the eastern Caribbean, while other synoptic events mostly affect other parts of the sub-basin (e.g., eastern boundaries). Investigating these objectives is important, not only for the short-term anticipation of coastal hazards and the improvement of operational early warning systems, but also for the assessment of future hazards in the region in relation to projected changes in weather and climate patterns [
15,
17].
Section 2 of the paper identifies the study region and describes the data used to address the above objectives, followed by a detailed description of the methods used to identify and characterize swell events and the associated distant storms. The results are given in
Section 3 and discussed in
Section 4.
2. Materials and Methods
2.1. Study Area
The map boundary (5° N–70° N, 0° E–100° W) (
Figure 1) shows the full study area for which atmosphere and swell composite maps are created (see
Section 2.5). The large internal rectangle (25° N–50° N, 35° W–85° W) is the area in which synoptic atmospheric systems are examined (
Section 2.6), and the small internal rectangle (11.7° N–19.7° N, 59° W–64° W) is the area in which the north swell is examined for the eastern Caribbean (
Section 2.4).
2.2. ERA5 Reanalysis
Swell and atmospheric variables derived from hourly ERA5 reanalysis data [
18,
19] are obtained from the Copernicus Climate Change Service (C3S) climate data store (
https://cds.climate.copernicus.eu/cdsapp#!/home (accessed on 19 January 2021)) for “significant height of total swell”, “mean period of total swell”, “mean direction of total swell”, “mean sea level pressure” and, “10 m u and v components of wind” from the 40 November to April (NDJFMA) periods within the November 1979 to April 2019 record. From this point forward, variables are referred to by the following short names: swell height, swell period, swell direction, sea level pressure, surface wind speed, and surface wind direction. Data resolution is 0.5° for the swell variables within the eastern Caribbean study area and 1° for swell and atmospheric variables within the full study area (
Figure 1). Both 0.5° and 1° data are obtained directly from the C3S climate data store.
Daily averages are computed for all variables. For swell direction, hourly values are decomposed into zonal and meridional (x and y) components before averaging. Unity magnitude is used as swell direction is derived from wave energy spectra and therefore lacks an explicit associated magnitude, unlike surface wind vectors. For consistency, surface wind u and v components are normalized by wind speed. Finally, the resulting average x/y and u/v components are used to calculate average daily swell and wind directions.
2.3. NOAA Buoys
Two wave buoys located near the eastern Caribbean study area (
Figure 1) are chosen to validate the use of daily ERA5 gridded reanalysis swell data. NOAA buoys 41040 and 41044 are both spectral wave buoys and partition wave energy into swell and wind sea categories. Daily averages are calculated from sub-daily data. Buoy observations begin in 2005 and 2009, resulting in periods of record of 14 and 10 NDJFMA seasons, for buoys 41040 and 41044 respectively. Additionally, buoys 41040 and 41044 are missing 515 of 2537 days (20%) and 141 of 1812 days (8%), respectively. Along with the scarce buoy spatial distribution and comparatively shorter records, these limits suggest the use of ERA 5 reanalysis data is preferred over buoy only analysis.
Swell heights measured from the NOAA buoys are compared with those derived from the ERA5 data at the grid point nearest the buoy locations. The distances between the buoys and ERA5 grid points are 63 and 60 km for buoys 41040 and 41044 respectively. Statistical comparisons are made between datasets to assess the efficacy of the gridded product to represent in situ observations.
2.4. Characterizing and Identifying Swells
The 90th percentile of swell height is computed for each grid cell in the eastern Caribbean study area and compared to a 2.2 m threshold value. Grid cells where the 90th percentile is less than 2.2 m are masked from further analysis (
Figure 2) due to the clear reduction in swell height in the Caribbean Sea from the islands. Mask location calculated from the 90th percentile of swell height versus the mean swell height does not show any signs of meaningful differences in spatial pattern.
Daily swell data for the eastern Caribbean are summarized by calculating spatial averages and then monthly averages from the daily spatial averages. Such summary data are also grouped by swell direction in 22.5° bins.
An empirical orthogonal function (EOF)-clustering procedure is used to identify the most recurrent patterns in the swell data, as well as north swell days and events (defined below) [
20,
21,
22]. Masked grids for the eastern Caribbean swell study area of daily swell height, period, and the x and y components of direction serve as initial inputs to the EOF-clustering procedure. Prior to calculating the EOF, daily input values are standardized by subtracting the long-term gridded mean over the available 40-year period and dividing by the gridded standard deviation. Inputs are weighted by the cosine of the grid cell latitude to account for the decreasing grid cell size moving poleward [
21]. Following Dawson [
23], an EOF analysis is computed to reduce the redundancy in the input data and provide orthogonal inputs for the cluster analysis. Five EOF structures (spatial patterns) and their corresponding principal components (PC’s; time series), which collectively explain 97.4% of the total variance in the input data and individually each explain over 1% of the variance, are retained for continued analysis (
Figure S1a).
Spatial patterns of daily swell height, period, and direction are grouped using a k-means clustering analysis calculated from the five PC’s [
24,
25,
26]. The k-means method iteratively repartitions days into clusters to minimize the sum of variances within the clusters. The analysis is executed ten times with a predetermined number of clusters ranging from 1–10. Each of the ten individual analyses is initialized 50 times with different randomly generated cluster centers. The maximum number of iterations to minimize the sum of variances for each initialization is set to 500. This maximum is never reached as the algorithm converged on a solution well prior to this maximum. The initialization with the smallest sum of squared error (SSE) when comparing the Euclidian distance between cluster centers and cluster members is chosen. The three clusters used for this analysis are identified by the curve “elbow”, which highlights the optimum cluster number where a decrease in SSE tends to plateau as the number of clusters increases (
Figure S1b). Two and four cluster results are also examined and found to be biased: two clusters did not account for one of the three major swell patterns and four clusters unnecessarily subdivided one of them.
Cluster composite maps of swell height, period, and direction are created by averaging data from all the member days for each cluster. Monthly counts of cluster occurrence are calculated from the time series of daily cluster membership. Daily spatial averages of swell height, period, and direction are calculated for the cluster representing north swells. North swell days, defined as days belonging to the north swell cluster, are aggregated into north swell events by grouping consecutive north swell days. To separate back-to-back north swell events without non-north swell days in between, the second event is determined to begin when the spatially averaged swell height increased after falling at the end of the first event. Finally, the parameters of a generalized extreme value (GEV) distribution [
15,
27,
28] are fit to the time series of the annual maximum spatially averaged swell height to assign a probability of occurrence to each north swell day. North swell events are assigned the probability associated with the lowest probability (i.e., highest height) day of the event.
2.5. Atmosphere and Swell Composites and Anomalies
Atmosphere and swell composites are created for the entire study area to examine the broader geographic conditions and forcing mechanisms associated with each cluster. Composites of sea level pressure, wind speed and direction, and swell height, period and direction are calculated for the full study area by averaging the gridded data for the days in each of the three clusters. Additionally, days within the north swell cluster are also composited in three classes according to their swell height probability (Low (Lo): <0.25, Medium (Md): 0.25–0.75, and High (Hi): >0.75). Composite anomalies are mapped based on differences between the previously described composites and the gridded long-term NDJFMA mean of each variable. Composites and anomalies are mapped for the four days prior to the day belonging to a specific cluster or a specific probability category of the north swell cluster to show the evolution of the atmospheric patterns leading to the swell generation. Swell days prior to November 4 are not included in this analysis because the preceding days are not available in the considered dataset.
2.6. Identifying Storms and Their Relationship to the North Swell
A methodology is developed to identify transient centers of low surface pressure offshore of the Atlantic coast of North America, which are hypothesized to generate most north swells that propagate into the eastern Caribbean. Numerous examples exist for identifying storm tracks in reanalysis data [
29,
30,
31,
32,
33,
34,
35,
36]. The present study most closely follows Hirsch et al. [
30]. For each day, the minimum sea level pressure grid cell location is identified as well as the 40 cells that form the edge of an 11° × 11° square centered on the minimum surface pressure location. The minimum surface pressure center is determined to be a closed low if at least 80% of the edge cells of the 11° × 11° square are over 4 hPa higher than the minimum pressure. A closed low is a standard criterion in automated storm identification [
30]. Numerous combinations of geographical domains to examine storms, distances away from the minimum pressure, and pressure differences are all examined and lead to similar results. The storm identification criteria are confirmed by visual inspection of daily sea level pressure animations.
Consecutive days with closed low pressure systems are identified as belonging to the same storm event if the low-pressure system on each consecutive day remains east of the previous day’s location. Different thresholds of eastward displacement on consecutive days are tested and do not have meaningful impacts on the results. One-day storms are removed from the list as the overwhelming majority are assumed to be too small and short-lived to be consequential for swell generation. The following parameters are recorded for each storm system: start and end dates, duration, daily minimum pressure, and daily latitude and longitude. A probability of occurrence of each storm’s lifetime minimum surface pressure value was assigned from a GEV distribution based on the inverted annual minimum pressure time-series. The annual minimum time series is inverted by subtracting from a constant value of 2000 because GEV distributions are typically based on maxima in the data [
28].
Each swell event is matched with a storm event if the swell event start date was greater or equal to the storm event start date and less than or equal to 2 days after the storm event end date. The two-day lag is based on the approximate time for swells to reach the eastern Caribbean based on the composite analysis described previously. Swells with a probability below 0.75 (59 highest-height swell events) and their corresponding storms are manually inspected to ensure that swells are properly matched to storms. Four one-day storms that produced swells are added back to the dataset. Composites and anomalies of ten medium probability swell events and one low probability swell event, all with no associated storm system, are mapped following the previously discussed technique (See
Table S1 for the list of these events).
Storm tracks are mapped based on the following categories: storms that produced north swells (in the eastern Caribbean) with swell height probabilities <0.25, 0.25–0.75, and >0.75, and storms that did not produce north swells (as defined by membership in the north swell cluster) with pressure probabilities <0.25, 0.25–0.75, and >0.75. The average latitude and longitude are determined for the track start, end, overall track average, and location of minimum pressure for each of these six storm categories.
4. Discussion
This study makes use of ERA5 reanalysis data to examine north swell events in the eastern Caribbean. Compared with buoy observations, ERA5 underestimates swell height; an observation consistent with previous studies [
15,
37,
38]. ERA5′s higher spatial and temporal coverage, however, make it an acceptable dataset for this study. Underestimates of swell height may have implications when assessing the magnitude of coastal hazards.
Jury [
7] notes that near Puerto Rico, northerly (315°–45°) swells occur 29% of the time on an annual basis with a mean height and period of 2 m and 8 s respectively. For the present study in the eastern Caribbean, the north swell cluster occurs 16% of the time during the NDJFMA period with monthly average swell height and period ranges of 2.2–2.4 m and 10.4–10.5 s respectively. Additionally, a more southeasterly and zonal storm track is associated with larger north swell heights in the eastern Caribbean.
The highest-height north swell event for both the present study and Jury [
7] is associated with a storm on 16–20 March 2008. Jury [
7] shows a blocking high sea level pressure system near Greenland slows the storm movement and allows more persistent northerly winds to generate the swell that ultimately impacts Puerto Rico. Jury [
7] also notes that a similar blocking high exists for the next five highest events in the study. A similar sea level pressure pattern is seen in the low probability north swell composite anomalies (
Figure 11, row 1), where the negative pressure anomalies around 35° N associated with storms immediately transition to positive pressure anomalies to the north in a location similar to Jury [
7]. Pressure anomalies with a similar pattern but smaller absolute values exist for both the moderate and high probability north swell events (
Figure 11, rows 3 and 5). The blocking high prolongs the duration of northerly winds generated from a region favorable for north swell propagation toward Puerto Rico and the eastern Caribbean. After reaching Puerto Rico, north swells take approximately 1 day longer to propagate to the eastern Caribbean.
In the North Atlantic, the annual cycle and low-frequency modes of both large-scale and synoptic atmospheric patterns induce strong seasonality and interannual variability of wave climate. On interannual time scales, Woolf et al. [
3] have shown the strong relation between ocean waves, particularly in the northeastern part of the sub-basin, and the North Atlantic Oscillation (NAO), which is the dominant mode of atmospheric variability over the sub-basin and is associated with the pressure difference between the Azores high and the Icelandic low [
39] (
Figure 9a). Morales-Marquez et al. [
40] further demonstrate that the NAO is a major driver of the interannual variability of wintertime extreme wave heights in this region. Future work in the eastern Caribbean should examine the relationship between north swells and the NAO. The phase and intensity of the NAO could possibly influence the storm speed and track location and therefore the generation and propagation of north swells toward the eastern Caribbean.
Additionally, to our knowledge, this is first study to identify high sea-level pressure systems in the western North Atlantic as generating moderate probability north swells in the eastern Caribbean. As such, the relationship between these high-pressure systems and their covariability with NAO phase and strength should also be examined.
The algorithm to identify storms in this study is limited. It is possible to have multiple low-pressure systems within the study area at the same time, but the algorithm only identifies the single low associated with the minimum pressure, missing any other local minima. This is due to the computational demands required to locate all closed lows for each day. Generally, such multi-low situations persist one day before one of the lows propagates out of the study region, at which point the second low is identified. Furthermore, the minimum pressure system likely has stronger winds and is therefore more likely associated with swell generation, although storm area and track speed may be of importance as well. Finally, there are never two major low-pressure systems in the study area at the same time. Rather, whenever two systems are observed, the larger one is typically in the process of moving out of the eastern or northern edge of the study area while the smaller one is forming along the western or southern edge.
Many storms in the North Atlantic do not generate north swells in the eastern Caribbean. Conceivably, some fraction of these storms generates significant swell events in locations other than the eastern Caribbean. Boucharel et al. [
5] examine this idea in the Pacific basin by assessing local versus remote atmospheric drivers of swell. A similar methodology could be applied to the Atlantic basin. At the Atlantic basin or North Atlantic sub-basin scale, it would also be possible to replicate the EOF-clustering procedure presented here. This procedure, however, would illuminate basin-scale swell patterns and not highlight patterns that are unique to a specific region such as was done here for the eastern Caribbean. As such, future research should consider a methodology to identify unique swell climate regions for more in-depth analysis or perhaps identify all storms in the sub-basin and develop composites and anomalies of swell conditions based on different storm categories.
Finally, long-term historic trends and future projections of wave-climates in the eastern Caribbean should be investigated. Future warming projections generate decreases in significant wave heights over the North Atlantic sub-basin (e.g., [
41]). Belmadani et al. [
15] confirm this result but also show marked increases in extreme wave heights associated with tropical storms. Future work should make use of climate projections to drive wave models for the eastern Caribbean and other local regions.