The upper ocean thermal structure is largely influenced by natural internal variability, and it modulates global surface temperature as well as regional sea level anomalies over decadal time scales. The internal variability of some important oceanographic processes, such as the Pacific decadal oscillation, can cause differences of 10 to 20 cm relative to the global mean in many coastal locations. These differences are sufficient to alter decisions between soft and/or large protective measures on every coast. Yet, current indicators of sea level changes are based on either surface variables or subsurface ocean measurements through the entire water column. There is also a lack of focus on short-term predictions of sea level change and concrete applications of these predictions at regional scale, which clouds local decision-making. The results herein confirm that the information contained within the top 300 m of the ocean is essential to capture decadal, regional relative sea level variability, whereas depths well below the thermocline do not appear to be closely tied to these large oscillations. Hence, the authors propose a regionally scoped indicator based on upper-ocean temperature as a first step toward identifying trend changes in short-term sea level rise over large coastal regions of the United States. The proposed indicator is a promising new tool that could help close the gaps mentioned above and improve the utility of existing local sea level records.
Characterization of natural internal variability and its influence on future regional sea level change is crucial to help adaption planners assess key coastal regions vulnerable to regional relative sea level rise. Decadal internal variability is often large enough to overshadow anthropogenic trends over periods of several decades (IPCC 2013; Lyu et al. 2014). The local rate of sea level change that is associated with internal cycles such as the Pacific decadal oscillation (PDO) can be as much as 4 times the rate of global mean sea level over a couple of decades (Hamlington et al. 2014). In particular, in the most vulnerable regions of the United States, such as coastal communities along the Gulf of Mexico and the Atlantic seaboard, water resources managers/planers will encounter new risks (US CLIVAR 2013; Melillo et al. 2014). These sectors will need a more accurate evaluation of the chances of local flood damage from the rising seas in coastal areas than existing practices can provide. Furthermore, there is also an important disconnect between the long-term focus of most sea level rise research—projections for 2100 and beyond—and the shorter-term needs. Most infrastructure planning efforts operate on a time scale of years and decades, not centuries. This is enough to change how planners take protective measures over the next 20 years. For example, soft protective measures (mangrove planting, shoreline restoration) and large infrastructure investments like sea walls have very different capital and maintenance costs (Donner and Webber 2014) and the best solution for a given region will depend on the amount of rise that is likely over the next 10 to 20 years.
Some attempts have been made to establish the relationship between multidecadal changes in extreme sea levels and large-scale climate variability associated with both traditional indices and tailored indices (e.g., Wahl and Chambers 2016). These indices, however, are based on surface variables (sea surface temperature, sea level pressure, winds). Note that surface information alone is not sufficient to characterize internal climate variability (Roemmich et al. 2015). Instead, upper-ocean (above 300 m) temperature provides a more slowly varying and robust signal than sea surface temperature or sea level pressure, which respond more rapidly to atmospheric forcing (National Academies of Sciences, Engineering, and Medicine 2016). We should also point out that although the quality of ocean and atmospheric data has gradually improved with time, the reliance on atmospheric products is particularly problematic over the Pacific Ocean before the 1980s (Merrifield et al. 2012). On top of that, there are other problems associated with these climate indices. For example, climate indices fail to explain the mechanistic understanding of what drives local sea level patterns. How to combine the influences of different climate modes in one location is also still a topic of debate (Meehl and Hu 2006; National Academies of Sciences, Engineering, and Medicine 2016).
Other studies have also explored the ability to use sea surface height (SSH) data to estimate the effect of internal variability on sea level along coastlines over shorter time scales using empirical mode decomposition (EMD) and empirical orthogonal functions (EOFs) or cyclostationary EOFs (CSEOFs) (Cummins et al. 2005; Li et al. 2012; Hamlington et al. 2013, 2014, 2015). We note that although these statistical methods can help to infer the sea level contribution of the dominant internal variability, this choice is limited by the integrative nature of the SSH variable and the time span of the record. SSH provides integrated information from the surface to the bottom of the ocean, and the SSH record does not go back nearly as far as the temperature record. We believe that a better assessment of this variability can be achieved by using subsurface temperature. Subsurface temperature provides a direct way to isolate the most active layers regarding decadal internal variability (against the long-term trend), which in certain instances allows early detection of large-scale sea level changes, as shown in this paper. While the information confined within the top 300 m of the ocean is a good proxy for decadal climate variability on basinwide and global scales (Roemmich et al. 2015; Nieves et al. 2015; Yan et al. 2016), deep ocean heat uptake (below 700 m) seems to be influenced by other mechanisms and disconnected from the regional pattern of internal variability (e.g., Liu et al. 2016). SSH includes other effects that correspond poorly to decadal climate variations (e.g., Church and White 2011). In shallow coastal regions, there is also the issue of wind-driven ocean bottom changes that can “disturb” the spectrum of the sea level on time scales of a few years to decades (Dangendorf et al. 2015). A temperature-based index is smoother than sea level, and we should also acknowledge its longer memory behavior (Dangendorf et al. 2014). The long record of subsurface temperature data enables us to learn about climate variability in the past as we develop proxy-derived records for decadal projection trends of sea level (outside the range of climate models).
Although we do not directly address wind-driven coastal sea level variability (Hong et al. 2000; Bromirski et al. 2011; Thompson et al. 2014), upper-ocean temperature is coupled to wind speed (shear) (England et al. 2014) and, as a result, it reflects the history of wind forcing on a short-term time scale. Interbasin connections through atmospheric bridges have also not been discussed here. However, regardless of whether changes in decadal variability in the Pacific can be triggered by variables in the Atlantic Ocean or the other way around (McGregor et al. 2014; Li et al. 2016; Yan et al. 2016), these processes leave some fingerprints in the upper-ocean layer of each respective ocean basin (see also Miller and Schneider 2000).
As an alternative to near-surface or full depth-integrated variables, we suggest using regional upper-ocean temperature estimates to identify potential changes in evolving regional sea level fluctuations within a short-term planning horizon. Although we are not yet providing formal projections or predictions, our analysis of the history of upper-ocean temperature will provide planners and decision-makers with insight into how much of the observed rate of regional relative sea level change is caused by internal variability (as opposed to anthropogenic change) over periods of a decade or two.
2. Data and methods
Observational yearly subsurface temperature anomalies from the World Ocean Database (WOA; https://www.nodc.noaa.gov/OC5/3M_HEAT_CONTENT/) were used in this study. These temperature estimates were averaged over the top 300 m for the regions under study (described below). The adequacy of WOA data to study climate variability was discussed in Nieves et al. (2015). Yearly estimates from AVISO gridded multimission monthly mean sea level anomalies (delayed-time “all sat merged” product; http://www.aviso.altimetry.fr/en/data.html) and from CSEOF monthly reconstructed sea level records (http://podaac.jpl.nasa.gov/dataset/RECON_SEA_LEVEL_OST_L4_V1) have also been used in this study.
The regions referred to herein as West, East, and Gulf are defined by the following coordinates, respectively: (19–80°N, 100°E–110°W), (19–80°N, 90–9°W) and (19–35°N, 105–9°W). The selection of the regions is based on proximity to the respective coast. The North Pacific basin has a strong impact on the western side of the United States and the Atlantic basin has a strong impact on the East and Gulf Coasts (e.g., DiLorenzo et al. 2008; Hamlington et al. 2013, 2015; Goddard et al. 2015; Wahl and Chambers 2016). We found that to study the Gulf Coast, a relatively larger region had to be taken into account due to the influence of the Rossby waves that propagate into the western boundary, redistributing heat across the Atlantic Ocean (e.g., Hong et al. 2000). Tropical and midlatitude decadal variability involves coupled Rossby waves (see also Goodman and Marshall 1999; Capotondi and Alexander 2001).
3. Results and discussion
We aim to constrain the relationship between internal climate variability and regional sea level patterns using upper-ocean temperature estimates. Even though formal separation of the human impact from internal sea level explicitly is still challenging (e.g., Marcos and Amores 2014), we can reduce the influence of the human imprint by removing the long-term trend. Note that the long-term warming contribution to regional sea level is small compared to internal variability on a short enough time scale (Perrette et al. 2012; Slangen et al. 2014). Ultimately, targeting ocean layers independently allow us to identify what part of the ocean holds clues to internal climate variability and its contribution to short-term sea level changes. Our method is especially useful in regions in which sea level rise is expected to be at least in part anthropogenic (see, e.g., Lyu et al. 2014).
We first computed observational temperature trends as a function of depth over three main regions along the U.S. coastline that show different oceanic regimes (see Fig. 1). These regions are referred to as West, East, and Gulf herein (see section 2). We can see that the largest temperature changes occur in the top 300 m of the ocean for the selected periods (1955–76, 1977–2002, 2003–12), regardless of whether the long-term trend is removed. Periods have been chosen based on the phase changes (enhanced or reduced warming) associated with decadal internal variability, which not only dominate on the global climate (e.g., Fyfe et al. 2016) but are also similar at regional scale (as will be seen later) even if they are driven by different physical mechanisms. Whichever the case, we expect regional relative sea level to be largely driven by upper-ocean variability, and therefore we estimated regionally averaged temperature over the top 300 m for the above regions. For this purpose, subsurface temperature was low-pass filtered—using standard moving average filtering (Kendall et al. 1983; Oppenheim et al. 2009) for different window lengths (of 3, 5, 10, and 15 years)—after the long-term trend (i.e., for the entire period) was removed. Results are rather insensitive to the effects of the particular filter used. In addition, different depth layers were also considered (e.g., top 500 and 700 m), yielding similar results (not shown). This further supports our assertion that the most active layers of the ocean are within the top 300 m and those contain the most relevant information needed to capture decadal internal variability, as shown in a previous study (Nieves et al. 2015).
Next, the three regional, upper-ocean estimates (referred hereinafter as the “New Indicator” or NI) were evaluated against observed sea level signatures at specific coastal locations within each region using both altimetric sea level anomalies and reconstructed sea level records (also detrended and filtered) to assess the relationship at multiple time scales (3-, 5-, 10-, and 15-yr filter) for the entire period of sea level record that is available (since 1992 and 1955, respectively, until 2012). Given that the analysis revealed regionally consistent patterns (shown later), for illustrative purposes, only three locations are given here for each region: South Beach, San Francisco, and San Diego (West); Portland (in Maine), New York, and Wilmington (East); and Turkey Point, Grand Isle, and Galveston (Gulf). The statistical significance of the linear correlations was evaluated following Ebisuzaki (1997), a nonparametric method that is particularly useful when serial correlations are a concern, as is the case of smoothed or filtered data (e.g., Marcos et al. 2015). It was found that the correlation becomes significant when the time lag is taken into account, mainly in the Atlantic coast (with the new indicator leading sea level). The time difference for which the two series have the highest correlation is of several years on average for both the East and Gulf regions (see Fig. 2). The time scale of the Rossby wave propagation most likely accounts for this lag between the basinwide average of upper temperature and western boundary sea level variability on the North Atlantic (see section 2). This suggests that the new indicator is seemingly leading sea level changes in the Atlantic region, especially during the negative phase (e.g., from the late 1970s to the early 2000s; see Fig. 3). In the West region, our indicator and the sea level signal are usually synchronous (time lag is 0 or 1 on average). The different phases of short-term climate oscillations—characteristic of tropical Pacific origin (Hamlington et al. 2013)—are also noticeable in this region (Fig. 3). Regardless of the physical processes underlying the observed decadal oscillations in each region, interestingly they all show similar cycles that occur around the same time (see the right two panels in Fig. 3). The selection of the above-mentioned periods to estimate trends is hence justified. Similar to the characterization of enhanced/reduced rate of global surface temperature warming (Fyfe et al. 2016), the choice of trend length is also critical to assess changes in the rate of short-term regional sea level rise, which in this case should be based on decadal fluctuations in the upper ocean.
The same conclusions can be reached using altimetric sea level or long-term reconstructions. However, because of the short length of the altimetry record, the level of statistical significance of correlations was smaller (Fig. 4). Yet, the analysis with reconstructions (for time lags ranging from 0 to 5 years) confirms that our indicator and sea level are in good agreement (with 90% confidence level) on decadal time scales for any of the above-selected locations (see Fig. 2). Overall, correlation coefficients are most often higher than 0.80, 0.80, and 0.60 for time lags of 0 or 1 (West), from 1 to 3 (East), and from 3 to 5 years (Gulf), respectively. The same applies to all coastal locations investigated along the United States (Fig. 5). This agreement reinforces the idea that a physical process tied to climate variability is thus linking both upper-ocean temperature and sea level in these regions. Over periods of 3 to 5 years, although the visual correspondence is quite remarkable, it is difficult to establish a robust relationship, especially in the Gulf region, where other local effects may also play an important factor (e.g., Sweet and Park 2014). In those cases, where correlations are significant, an ordinary linear regression model was developed from upper-ocean estimates to obtain the explained variance. Our model can explain about 69%, 81%, and 43% of the variability in regional sea level on decadal time scales for the West (t lag = 0), East (t lag = 1), and Gulf (t lag = 4) regions, respectively.
It is also clear from Fig. 6 that, rather than closely obtaining the exact amount by which sea level rise is in excess or deficit relative to the global mean, the main advantage of the proposed indicator resides in its capability to indicate trend changes. We can make a first guess as to whether observed regional sea level rise is part of a long-term anthropogenic trend or whether it can be expected to accelerate or decelerate under certain (warming/cooling) conditions due to internal climate variations—even if these internal cycles give way to the long-term anthropogenic trend after many decades.
Because natural internal variability involves movement of heat over broad regions around the upper layers of the ocean (National Academies of Sciences, Engineering, and Medicine 2016; Yan et al. 2016), our focus has been on producing regionally scoped estimates rather than averaged over small location-based boxes. The selection of large broad regions also helps to reduce the noise and emphasize statistically significant large-scale climate variability. On the other hand, the coastal ocean response is similar whether we use upper-ocean depth-integrated temperature (i.e., upper-ocean heat content) or upper-ocean averaged temperature estimates. They both reflect the thermosteric nature of regional relative sea level changes over interannual to decadal periods. Nevertheless, the mass component of sea level variability (loss of land-based ice) and other ocean and terrestrial effects can simply be included from other sources.
Thus, subsurface temperature provides the opportunity to gain more information about changes responding to internal variability than solely surface variables. It also allows us to exclude those specific layers that are irrelevant to those changes in particular areas. In doing so, not only we can generally mimic the distinctive pattern of large-scale oscillations in sea level but also, more importantly, detect preceding changes a few years ahead during the negative phase (all while remaining dynamically consistent with well-documented physical mechanisms). Furthermore, in general the predictive skill of temperature will be higher compared to coastal sea level itself, especially (but not only) in shallow coastal seas, where the barotropic motions can dominate sea level variability over periods up to a decade (Dangendorf et al. 2015). Based upon the climate cycles from short-term oscillations, our method provides the means to identify key coastal regions vulnerable to internally induced regional relative sea level changes in the United States within the 10- to 20-yr horizon, which, up until now, existing indices have been unable to do.
By analyzing long records of large-scale, decadal upper-ocean temperature variability in the Pacific and Atlantic Oceans, we can uncover potential coastal changes in regional relative sea level. Our approach seems to work best in the Atlantic coast, showing sea level changes ahead of time. Nevertheless, our indicator can also be useful for the West coast during its negative phase most of all. Sea level change along the Gulf of Mexico, however, remains a challenge as other local effects appear to dominate in that region, mainly over periods smaller than 10 years. On decadal time scales, our findings demonstrate that upper-ocean temperature estimates provide a useful tool for understanding regional relative sea level changes due to internal climate variability. The simplicity of the proposed indicator and its ability to explain observed signals makes it a useful support system for decision-makers. A good knowledge of multidecadal records of regionally scoped sea level “proxies” is pivotal to evaluate potential future sea level risk and to target adaptation and mitigation options.
The new regional sea level indicator could be further improved by accounting for changes in water mass in the oceans using data from GRACE and its follow-on mission. Nevertheless, this is not necessary to detect the large fluctuations by internal variability discussed here. These decadal-scale climate oscillations are also superimposed on the influence of tidal effects, topographic vulnerability, or storm surges. This variety of additive factors can be included later using information from tide gauges and other national coastal products (e.g., Sweet and Zervas 2011; Sweet et al. 2013; Sweet and Park 2014).
The research described in this paper was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. This work was partly supported by NASA NNH12ZDA001N-MEaSUREs and Spanish Ministry of Economy CLIMPACT CGL2014-54246-C2-1-R grants. We thank B. Hamlington for generously providing the up-to-date sea level reconstructions used in this study.