Climatic variability of the sub-surface sea temperatures in the Aegean-Black Sea system and relation to meteorological forcing

Abstract

Non-smoothed yearly temperature records with minimal statistical uncertainties are constructed for winter and summer of the period 1950–2000 in two areas in the Aegean Sea, for the sub-surface layer of 80–120 m, and two areas in the Black Sea, for the sub-surface layer of sigma-theta isopycnals between 14.5 and 15.4. The specific areas are selected mostly because of the dense hydrographic-data coverage they have during the period 1950–2000. Two trend regimes appear in both Seas: a period of decreasing sea temperatures from the early/mid 1960s to the early/mid 1990s and an apparent warming afterwards. Trends in sea temperatures correlate with trends in the North Atlantic Oscillation (NAO) and partly the East Atlantic West Russian (EAWR) indexes, but the signs of NAO and/or EAWR cannot sufficiently justify the winter-to-winter temperature changes in the entire study area. In examining the wind flows in the sea-level-pressure maps for characteristic winters in which local peaks in the sea-temperature records occur, we identify particular sea-level-pressure structures that are not accounted for by the typical North-Atlantic or East Atlantic-West Russia positive or negative dipoles. In addition, there are winters when the Siberian High induces local maxima in sea-temperatures in the study area. A spectral-coherence analysis of the unfiltered winter sea-temperature and the corresponding teleconnection NAO/EAWR records, shows that common spectral and coherence peaks exist at ~5–6, ~9–10 and ~15–17 years.

Appendices

Appendix 1: Statistical uncertainties of the mean temperatures

Each estimate of a mean temperature value (t_mean) is based on a given sample of N data points obtained within a particular area and during a given time period, i.e., year days 10–120 for winter and year days 200–280 for summer. The question to be answered is how much a t_mean value differs from the true mean T_mean that would have been obtained if the entire area had been covered with data continuously for the specific period. Apparently the statistical error in t_mean depends (a) on the actual spread of temperature around the T_mean, i.e., the variance, throughout the particular area and the considered time period and (b) on the available N data points that are used in the averaging. For each area considered and for both winter and summer, apart from area B2 in the north-east Black Sea, there were winter and summer periods of specific years in which the entire area was continuously sampled so as to assume that t_mean is nearly equal to T_mean. In these periods N ranged from 47 to 84. For those cases in which T_mean is practically known, we considered hypothetical samplings with smaller sub-samples N_sub in which N_sub = 2, N_sub = 3, N_sub = 4, … etc. We computed the standard deviation from the T_mean of all sampling combinations with given N_sub. A plot of the standard deviations as functions of N_sub for all cases with the hypothetical samplings is in Fig. 11. The largest standard deviations in the hypothetical samplings occur in area A1, because this area has the largest north–south extent in an existing north–south mean temperature gradient. For an approximate 35–40 independent temperature samples, the estimated mean temperatures are very nearly equal to the true mean temperatures.

Fig. 11

The standard deviations from the true mean temperatures for the different subareas in winter and summer when estimated temperatures are based on N_sub data points. All possible combinations with given N_sub and the resulting mean temperature estimates are considered in hypothetical samplings drawn from a large volume (N_∞) of available data sufficient to determine the true mean temperature when all of them are used

In constructing the temperature records, in each area and for a specific season, summer or winter of a given year, N_year data points were used to estimate the t_mean. An estimate of the error in t_mean can be approximated by the standard deviation obtained in the corresponding hypothetical sampling, as described above for the specific area and season, in which N_sub = N_year. The errors in area B2 are computed based on the hypothetical-sampling standard deviations of area B1. All winter and summer temperature records along with the resulting curves when we add and subtract one error bar, i.e., one standard deviation, to each mean temperature estimate are shown in Fig. 3. The errors, for a given area and season depend on the number of the data samples used for the mean temperature estimate. Periods from the mid/late 1960s to the mid/late 1980s in the Aegean are characterized by higher data coverage and the uncertainty ranges decrease. The same happens for periods from the mid 1970s to the late 1980s/early 1990s in the Black Sea. The smallest uncertainties are in the southwest Aegean in winter because it is characterized by low standard deviation values (Fig. 11) and dense data coverage for most of the 1950–2000 period.

Appendix 2: A note on the reliability of the NCEP monthly winds in our study area

During the review process, a query was raised with respect to the coarse NCEP grid and the possibility that this may not adequately represent the actual wind disturbances caused by orographic and other land features nearby the four sea areas that we consider. The question particularly concerned area B2 which is to the west of the nearby Caucasus mountains. For this reason we have compared recent monthly-mean NCEP wind data with corresponding wind data from a meteorological station nearby Athens/Greece and from a high-resolution operational meteorological model, the SKIRON model (Spyrou et al. 2010), for the B2 area in the northeast Black Sea. The first comparison is for the period from January 1998 to May 1999 and is exhibited in Figure 2 of the article by Kontoyiannis (2010). The NCEP monthly wind data can successfully describe the seasonal wind changes in the wider sea area south of Athens. The second comparison, for area B2 and during the period from January 2003 to December 2007, is shown in Fig. 12. There are several periods, each lasting a few months, during which the wind is characterized by a persistence is some general directions. For example, during most of 2003, the winds are to the south-southeast, whereas in the first half or 2004 they switch into north-northeast directions. Both datasets, NCEP and SKIRON, successfully describe these particular wind regimes during the entire 5-year period.

Fig. 12

Lower panel monthly-mean wind vectors at area B2, as spatial averages of NCEP monthly-mean winds at grid points ‘6’ and ‘7’ shown in Fig. 2. Upper panel monthly-mean wind vectors at 44°N 37°E from the high-resolution, operational, atmospheric model ‘SKIRON’ (Spyrou et al. 2010)