North Atlantic extratropical and subpolar gyre variability during the last 120 years: a gridded dataset of surface temperature, salinity, and density. Part 1: dataset validation and RMS variability

Abstract

We present a binned annual product (BINS) of sea surface temperature (SST), sea surface salinity (SSS), and sea surface density (SSD) observations for 1896–2015 of the subpolar North Atlantic between 40° N and 70° N, mostly excluding the shelf areas. The product of bin averages over spatial scales on the order of 200 to 500 km, reproducing most of the interannual variability in different time series covering at least the last three decades or of the along-track ship monitoring. Comparisons with other SSS and SST gridded products available since 1950 suggest that BINS captures the large decadal to multidecadal variability. Comparison with the HadSST3 SST product since 1896 also indicates that the decadal and multidecadal variability is usually well-reproduced, with small differences in long-term trends or in areas with marginal data coverage in either of the two products. Outside of the Labrador Sea and Greenland margins, interannual variability is rather similar in different seasons. Variability at periods longer than 15 years is a large part of the total interannual variability, both for SST and SSS, except possibly in the south-western part of the domain. Variability in SST and SSS increases towards the west, with the contribution of salinity variability to density dominating that of temperature in the western Atlantic, except close to the Gulf Stream and North Atlantic Current in the southwest area. Weaker variability and larger relative temperature contributions to density changes are found in the eastern part of the gyre and south of Iceland.

Appendices

Appendix 1. Comparison with Friedman et al. (2017)

The BINS boxes mainly use the same underlying SSS data as the large boxes north of 40° N in Friedman et al. (2017). The datasets cover a similar area (Fig. 10a); though as mentioned previously, the BINS boxes more carefully avoid shelf regions (except for southwest Greenland and the southern part of the Grand Banks). Most of the source data used in the two analyses are the same. Additionally, BINS also incorporates two small datasets from the 1900s and 1910s, plus a few recent transects (and 2014–2015). There are more gaps in the 34 BINS time series than in Friedman et al. (2017): in particular in 1918–1921 and during and just after WWII; these gaps are linearly interpolated in the smaller boxes. Also, when sampling is poor, but varying geographically within the larger boxes, it is possible that some spatial variability is aliased in the temporal variability in the larger boxes.

Fig. 10

a Grid boxes from Friedman et al. (2017) (blue) and BINS (black). Thick lines indicate the NATL region. c Comparison of NATL index from Friedman et al. (2017) and BINS grid boxes. Anomalies are from 1896 to 2012; shading indicates ± 2 error terms

Figure 10b compares the NATL index from Friedman et al. (2017), area-averaged SSS from 45° to 62° N, with SSS averaged over a similar area in BINS. (NATL from Friedman et al. (2017) is only plotted through 2012, as 2013 was subject to endpoint smoothing). The two products are very highly correlated (r = 0.94, 1896–2012) and compatible considering the differences in area and error estimates. Greater differences are found for smaller regions in the first half of the record, particularly during the gap years mentioned above (not shown).

Appendix 2. Comparison with Reverdin et al. (2018)

As a check on the BINS time series, we compare the boxes with time series constructed with mostly similar data, monthly binned along two ship routes since mid-1993, intersecting near 59.5° N/32° W: AX02 between Iceland and southern Newfoundland and AX01 between the North Sea and southern Greenland, mostly along 59.5° N (Reverdin et al. 2018), shown in Fig. 11a. Time series along AX02 start in July 1993 with few gaps, whereas for AX01 some large data gaps were filled until late 1997. These time series, referred to as B-AX01 and B-AX02, provide increased spatial resolution at seasonal timescales and portray very coherent variability where they intersect.

Fig. 11

a AX01 (red) and AX02 (blue), from (Reverdin et al. 2018). b Comparison of B-AX01 with BINS, 1993–2015. The monthly time series of B-AX01 have been yearly-averaged (Dec–Nov), low-pass (1-2-1) filtered over successive years, and then averaged over the bins in BINS: correlation coefficient (top) for T (blue) and S (red) and RMS (middle for S and bottom for T) (B-AX01 in black, BINS in blue (T) and red (S))

We illustrate the comparison of interannual variability of T and S between B-AX01 and BINS for overlapping boxes in the common period 1993–2015 (Fig. 11b). For S, the corresponding filtered RMS variability is larger in B-AX01 than in BINS by up to 20%, but with very high correlation coefficients (all larger than 0.95). The smaller RMS amplitudes of salinity in BINS probably result from the larger box sizes and the resulting spatial averaging. RMS variability is more similar for T, but with a slightly smaller correlation in the Iceland Basin (0.80), where gaps in 1993–1997 were the most common in B-AX01. Altogether, the comparisons for the two ship tracks suggest that the method used for BINS in the box averaging to produce interannual variability yields correct results when data coverage is sufficient.

Appendix 3. Locations of BINS and HadSST3 grid boxes

Fig. 12

Locations of BINS and HadSST3 grid box regions used in Sect. 3.3. Thick blue lines show the 5° × 5° HadSST3 grid boxes; shading shows the corresponding BINS grid boxes: a central Labrador/West Greenland, b southern Nordic, c central SPG, d north-east SPG, e south-west Labrador, f intergyre, g Grand Banks/Labrador Current, h east Atlantic