The thermosteric contribution of the North Atlantic Oscillation (NAO) to the North Sea sea-level for the winter period is investigated. Satellite sea surface temperature as well as in situ measurements are used to define the sensitivity of winter water temperature to the NAO as well as to determine the trends in temperature. The sea surface temperature sensitivity to the NAO is about 0.85 °C per unit NAO, which results in thermosteric sea-level changes of about 1–2 cm per unit NAO. The sensitivity of sea surface temperatures to the NAO is strongly time-dependent. Model data from a two-dimensional hydrodynamic tide+surge model are used in combination with the estimated thermosteric anomalies to explain the observed sea-level changes and, in particular, the sensitivity of the datasets to the NAO variability. The agreement between the model and the observed data is improved by the inclusion of the thermosteric effect.
Sea-level rise is a major threat to the coastal environment and is expected to accelerate with global warming (Church et al. 2001). The longest time-series from tide-gauges extend more than 150 years back. These indicate a value for the global sea-level rise in the range of 1–2 mm yr−1 (Church et al. 2001). Deriving global estimates from single point tide-gauge measurements has many difficulties. Tide-gauges measure sea-level changes relative to land, thus their records are contaminated by local land movements, which in turn may be either local or global in character. In addition, atmospheric and oceanographic forcing affects the tide-gauge measurements at decadal, inter-decadal and possibly centennial scales and thus it contaminates the long-term sea-level trends estimates (e.g. Tsimplis et al. 1994). Thus, it is arguable that the correct way of obtaining reliable global sea-level rise estimates is by removing the effects of the local atmospheric and steric forcing together with any known land movements. In this paper, we will attempt to resolve the contribution of the various forcing parameters to sea-level at one tide-gauge located at Den Helder at the southern boundary of the North Sea and strongly affected by variations in the North Atlantic Oscillation (NAO) (Wakelin et al. 2003).
The NAO is one major atmospheric, basin scale pattern, which affects sea-level around Europe (Tsimplis & Josey 2001; Wakelin et al. 2003; Woolf et al. 2003; Yan et al. 2004). The NAO dominates atmospheric variability over the North Atlantic and Europe (Hurrell 1995), especially in winter. The NAO is essentially a measure of the atmospheric pressure difference between the Icelandic Low and the Azores High. The effects of the NAO on various meteorological parameters have been well established (Hurrell 1995; Osborn et al. 1999), with high NAO winters being wetter, warmer and windier over northern Europe and the North Atlantic. In addition to a large inter-annual component in the index, a multi-decadal positive trend occurred in the NAO index from the 1960s onwards, raising the possibility that climate change might be strongly linked with changes in the NAO, a possibility that remains uncertain (Osborn et al. 1999). Recent studies on the impact of the NAO on sea-levels around Europe indicate that a linear relationship between the winter sea-level anomalies and the NAO Index can be used to explain most of the variability in the North Sea, the Mediterranean and the eastern parts of the North Atlantic (Woolf et al. 2003). The mechanisms contributing to this correlation have partly been resolved. Wakelin et al. (2003) have shown that the wind is the major forcing factor affected by the NAO in the shallow North Sea, while atmospheric pressure changes are the major contributor over the Mediterranean Sea (Tsimplis & Josey 2001) and the Atlantic coasts of Europe (Wakelin et al. 2003; Tsimplis et al. 2005). Wakelin et al. (2003) compared the output of a two-dimensional, depth averaged tide+surge model to the NAO and estimated the change of sea-level for 1 unit change of the NAO (which was termed the sensitivity to the NAO). They repeated the same estimate for nearby tide-gauges and found that the sensitivities from tide-gauges were, in general, higher than those estimates from the model grid-points close to the tide-gauges. A possible explanation for this discrepancy could be an additional thermal contribution of the NAO to sea-level or some deficiency of the t+s model. A thermal contribution by the NAO would be consistent with the general description of the NAO by Hurrell (1995) and the findings of Tsimplis & Rixen (2002) for the upper waters of the northern parts of the Mediterranean Sea.
This paper describes the influence of the NAO on the sea surface temperatures (SST) over the North Atlantic and, in particular, the North Sea. We then proceed to estimate the thermosteric contribution to sea-level changes for the winter season and investigate whether the inclusion of thermosteric sea-level effects improves the agreement between the observed sea-level data and a two-dimensional hydrodynamic model in respect of the sea-level sensitivity to the NAO. Sea-level trends decontaminated by the atmospheric and steric forcing are also derived.
Section 2 describes the data and the methodology used. Then, in §3, the SST records for the North Atlantic are described and their North Sea values compared with in situ measurements at Marsdiep. The thermosteric contributions to sea-level variability are then combined with the two-dimensional model data and compared with the observed sea-level values at Den Helder. Finally, in §4, our findings are summarized.
2. Data and methodology
(a) Sea surface temperature data
The sea surface temperature data used in this study are derived from the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) sensor. The AVHRR sensor is a filter-wheel spectrometer with three thermal and two visible channels (Pickard & Emery 1990). The AVHRR channels are then processed to form multi-channel sea surface temperature (MCSST) data. This processing is carried out by the University of Miami producing weekly 8-day averages, with each bin overlapping the previous by 1 day. McClain et al. (1985) estimated the SST accuracy to be approximately 0.3 °C. The University of Miami produces day-time (ascending) and night-time (descending) SST data having a spatial resolution of 18 km. The MCSST data span a number of satellites (NOAA-7, -9, -11 and -14). To reduce the uncertainty of the data because of diurnal warming, we only used the night-time data. The data coverage starts at 1982, but the values for 1982–1983 were not included due to aerosol contamination after the eruption of the El Chichon in 1982. In 2000, the NOAA-14 satellite drifted from its original equatorial crossing, therefore to avoid any ambiguities MCSST data beyond December 1999 were not included. The MCSST data were then averaged into monthly values in a 0.5°×0.5° grid.
To verify the AVHRR (MCSST) data, eight locations within the North Atlantic and Mediterranean regions were compared with data from the Along Track Scanning Radiometers (ATSR) 1 and 2. ATSR 1 data cover the period August 1991 to April 1996 while ATSR 2 data cover the period July 1995 to December 1999. The correlation coefficients of the ATSR 1 and 2 mean monthly values with those derived from the AVHRR were, in all eight control points, better than 0.92 (significant at the 99% level). Two of the control points are further used for comparison with in situ temperature values (figure 1).
(b) In situ temperature
Near surface water temperatures were kindly provided by van Aken (2003). These cover 1861–2004. Daily temperature measurements were made for 1861–1962 at the Marsdiep tidal inlet near Den Helder by taking the temperature of bucket samples. After 1947, temperature measurements at the coast of the Island Texel were made with the same method. Based on the overlapping period van Aken (2003) has homogenized the time-series. Salinity measurements are also available. The mean monthly AVHRR and the in situ Marsdiep temperatures are consistent. The correlation between the in situ data and point A (figure 1) in the southern North Sea was 0.62, while for the northern North Sea point B the correlation was 0.42. There were 181 and 187 observations at points A and B giving 158 and 164 degrees of freedom (we are losing 12 degrees of freedom for removing the seasonal cycle and one for the mean). Thus, the correlation coefficients are significant at the 99% level. Notably, both the seasonal cycle and the departures from it are larger at the southern shallower part of the North Sea (figure 2). The largest differences between the in situ and the AVHRR time-series occur when the maxima and the minima of the seasonal cycle occur. The availability of the Marsdiep data permits the comparison between sea-level and temperature for the whole period of the model data (1955–2003), while the shorter AVHRR time-series which begin at 1983 permit an assessment of the spatial variation of temperature.
(c) Tide-gauge data
(d) Two-dimensional model sea-level data
Model sea-level data from a two-dimensional barotropic model known as CSX (Flather et al. 1998) for the period 1955–2003 were used (Wakelin et al. 2003). The model covers the northwest European continental shelf from 45°40′ N to 62°20′ N and 15° W to 12°30′ E with a resolution of about 35 km (1/3° latitude by 1/2° longitude). The model is forced by six-hourly wind and atmospheric pressure data and by tides (Flather 1981). Meteorological forcing was provided by a meteorological set developed at the Norwegian Meteorological Institute (DNMI) (Reistad & Iden 1995). At the open boundary, a radiation condition is used with input of eight tidal harmonics (Q1, O1, P1, K1, M2, S2, K2, N2) and a meteorological component estimated from surface atmospheric pressure using the inverse barometer correction (Flather et al. 1998). The surge component was computed by subtracting the output from two runs, one with the tide and the meteorological forcing yielding total sea-level values and one with tides only. The difference includes the interaction of the tides with the surge (Flather et al. 1998). Mean monthly values were formed from the hourly model values.
The monthly values of SST, in situ temperature, observed and model sea-level were separated into a seasonal signal (calculated by averaging over the monthly values for each of the 12 calendar months), and a residual anomaly. The residual anomalies represent inter-annual variance about the mean seasonal cycle. The entire monthly dataset has been studied, but as the NAO influence is more pronounced during the winter, which is also the period where storms cause most damage, we concentrate on the period December to March. We mainly discuss anomalies from the mean seasonal cycle unless stated otherwise.
Steric anomalies were estimated on the basis of the full equation of state. The air temperature variation in Marsdiep is highly correlated with the air temperature in Den Helder (van Aken 2003), where a tide-gauge has been operating. The thermal response of the sea in the Marsdiep region to air temperature changes is about two weeks (van Aken 2003). Thus, for mean monthly values we consider the Marsdiep water temperature as representing the sea temperature variation in Den Helder. We do not use the salinity measurements at Marsdiep because these are affected by the reducing outflow of a nearby river (van Aken 2003). Instead, we assume the salinity at Den Helder to be constant at the average winter value (28.27 psu). Knowledge of the temperature variation near the surface and the value of salinity permits the estimation of steric sea-level variability for the shallower parts of the North Sea during winter when the water column is well mixed. Thus, the temperature-related steric variability was calculated for a nominal depth of 18 m which corresponds to the grid depth of the two-dimensional model.
(a) Linear trends and the NAO influence on SST
The results of the linear regression with a dependent parameter of winter SST anomaly, and independent parameters winter NAO index and time, are shown in figure 3. The time period was split into two 8-year periods. The period 1992–1999 shows stronger linear trends in SST (figure 3c) than the period 1983–1991 (figure 3a) and dominates the overall average (figure 3e). The spatial characteristics of the trends are different between 1983–1991 and 1992–1999 with a cooling area in the east Atlantic during the former period. The trends in figure 3c for the Mediterranean Sea are in agreement with those produced by Cazenave et al. (2001) for the period 1992–1999. The North Sea, where we are focusing, has positive SST trends during the whole period of observation with weaker trends in the first part of the measurements. The Marsdiep dataset reveals a small positive trend in the winter temperatures from 1956 to 2003 of 0.001 °C yr−1. For the 1983–1991 period, the trend was in winter 0.12 °C yr−1 while for 1992–1999 the temperature trend was 0.159 °C yr−1. Figure 4 shows the temperature variation in winter together with the trends stated above. It is evident that the large trends observed in the 1983–1991 and the 1992–1999 periods are the product of inter-annual variability and not representative of the long-term trends. Thus, the AVHRR changes in the SST trend (figures 2 and 3) must similarly be seen as a result of inter-decadal variability. The winter Marsdiep temperatures appear to be above the mean most of the time after the mid-1980s (figure 4). However, this appears to be linked with the NAO variability (figure 4), which is highly correlated (the correlation coefficient is 0.75 for 1956–2003), with the sea temperature at Marsdiep. The sensitivity found at Marsdiep (i.e. the regression coefficient for the NAO index) is 0.9 °C per unit NAO. The spatial distribution of the sensitivity from the AVHRR data can be seen in figure 3b,d,f. The North Sea area appears consistent with the Marsdiep results (figure 2).
(b) The contribution of the temperature forcing to sea-level variability in Den Helder
The estimated thermosteric values, together with the observed (from the tide-gauge) and the modelled (from the two-dimensional model) time-series, are shown in figure 5. The two-dimensional model, which includes the direct meteorological forcing of wind and atmospheric pressure variability as well as some of the nonlinear interaction between the tidal signal and the storm surges, resembles the observed data very well. The steric sea-level also covaries for certain periods of time (e.g. 1972-onwards), but it is also very different during other periods (1968–1975). The cause of the discrepancies is uncertain. However, as we have found good correlation between the NAO and the temperature time-series, we consider that the differences are not due to error in the measurement of sea temperature. In table 1, the results of linear regression between the various sea-level components and the NAO are shown. There is a significant difference between the sensitivity of the observed sea-level (5.5 cm per unit) and that of the modelled sea-level (4.4 cm per unit). This was noted first by Wakelin et al. (2003). Thus, there is a difference of about 1.1 cm per unit NAO between observed and modelled sea-level. The steric sea-level sensitivity to the NAO (table 1) can give a value of 1.1 cm per unit if the depth is 13.75 m, thus resolving the discrepancy. The water depth at the grid-point of the two-dimensional model is 18 m giving a higher sensitivity of 5.8 cm per unit NAO. The small difference may have to do with the fact that the average grid-point depth differs from the depths near the coast. If one considers the possibility of the discrepancy between the two-dimensional model and the observed values being due to the model resolution, one can reproduce the observed sensitivity by multiplying the two-dimensional model data by a 1.31. Whether there could be a contributing two-dimensional model resolution problem will be considered elsewhere. The inclusion of the total steric effects (i.e. the salinity contribution) did not affect the NAO-related sensitivity value significantly (5.6 mm per unit NAO), but it increased the discrepancy between the observed sea-level anomalies and those included in the steric+two-dimensional model values (not shown).
(c) Decontaminated sea-level trends
It is worth noting that sea-level trends calculated from the observed data give values of 3.3 mm yr−1. When the steric effect and the meteorological forcing are removed from the observed data either by assuming the depth to be known at 18 m or by linear regression with the two-dimensional model and the steric levels as independent parameters, the estimated trend reduces to values in the range of 1.7–2.0 mm yr−1. This is consistent with the observations of Tsimplis et al. (1994) and Wakelin et al. (2003). Thus, estimates of long-term sea-level rise based on tide-gauge time-series that are 47 years long include trends of 1.3–1.6 mm yr−1 due to atmospheric and steric forcing changes. For Den Helder, the steric effect has a trend of 0.163 mm yr−1, but the two-dimensional model does not have a trend even if these data affect the trend of the observed sea-levels.
The winter sea surface temperature over the whole of the Atlantic is affected by the NAO. In the North Sea, in particular, a very good correlation between the measurements at Marsdiep and the NAO index has been established. Nevertheless, the estimated steric variability is not always coherent with the observed sea-level variability. The two-dimensional model as well as the observed and the steric sea-levels are highly correlated with the NAO index for the period 1955–2003. The thermosteric sea-level contribution is enough to explain the discrepancy in the NAO sensitivity between modelled and observed data observed by Wakelin et al. (2003). However, scaling of the sea-level anomalies can have the same effect and it slightly improves the description of the observed sea-level. Thus, further work is needed to confirm whether higher resolution is needed to describe the sea-level more accurately near the coast. Notably, the relationship between the winter NAO and the winter Marsdiep temperature is not steady (figure 6). During the periods 1890–1910 and 1930–1940, the correlation is not significant at the 95% level. Similarly, it is reduced in the 1970s, where we have observed some discrepancy between the steric sea-levels and the observed and modelled ones (figure 5). Tsimplis et al. (2005) have suggested that for periods of time the southern North Sea is dominated by the central European weather rather than the North Atlantic. Plag & Tsimplis (1999) have observed coherent changes in the annual cycle consistent with atmospheric pressure changes, indicating the dominance of the continental climate for periods of time. The results in figure 6 suggest also a similar effect. Thus, we suggest that there are periods of time where the NAO is high, in which the storm surges as well as the steric cycle is driven by the NAO. There are also periods of time where the NAO is strongly negative and both the sea-level and the temperature variability are not driven by the NAO and the westerlies. Nevertheless, we also found a period in the early 1970s where the temperature correlation with the NAO is significant, but the sea-level becomes uncorrelated with the steric variability. One reviewer pointed out that the environment near Marsdiep changed in the 1930s as a result of the construction of the closure of the Zuiderzee which then became the IJsselmeer. It is possible that sea-level variability and temperatures at Marsdiep have been affected by this civil engineering construction activity, thus casting some doubt on the above discussion on the time variability of the NAO influence. Finally, we note that during the analysis the seasonal cycle of water temperature was found to precede the seasonal cycle of sea-level by about a month. This may allow for improved forecasts of sea-level.
We are grateful to Hendrik van Aken for providing the mean monthly values for temperature at Marsdiep and to four reviewers for their helpful reviews.
One contribution of 20 to a Theme Issue ‘Sea level science’.
- © 2006 The Royal Society