APPENDIX 1

NOTES ON PROCESSING AND QUALITY CONTROL

OF THE SEA SURFACE TEMPERATURE DATA

FROM THE METEOROLOGICAL OFFICE MAIN

MARINE DATA BANK (MOMMDB)

1. REMOVAL OF DUPLICATES (Figure 3 (in main text), Box 1).

Observations originally entered the MOMMDB on punched cards or on data sets keyed in 'card image' format. Any observation in a card batch or data set was removed from the batch or data set if it was found to have all 80 columns identical with any other observation in that card batch or data set: if there were 75 to 79 identical columns, the observation was printed out for visual inspection and possible removal. When card batches or data sets were being merged into the MOMMDB, observations identical to ones already in MOMMDB were not added. The subsequent averaging of data for 1 deg. areas for individual pentads (Section 3(c) of the main text) will have reduced the effects of undetected duplicates.

2. CREATION OF BACKGROUND FIELDS (Figures 4a, b, c, d (in main text))

(a) Formation of 'first guess' 1 deg. area background field averages

Accepted SAT or NMAT observations (Box 1 of Figure 4a) f or the whole period of the record (1854-1981) were averaged, irrespective of the number of observations in each contributing year, into a first-guess climatology for each 1 deg. area and pentad (1-5 January etc.) (Box 2 of Figure 4a). Away from the main shipping lanes, these averages were frequently based on small samples of data, and in many 1 deg. areas an appreciable number of pentads had no data.

When filling the gaps, the first step (Boxes 3 to 5 of Figure 4a, expanded in Figure 4b) was to combine (without weighting) averages from consecutive pentads in each 1 deg. area to produce up to 24 calendar 'half-monthly' first guess values, using the mainly 3-pentad groupings in Table A1.1. A single 4-pentad half month is required because there are 73 pentads in a year: the period 2nd to 21st March was chosen for this because at this time of the year SST usually changes only slowly with time, at least in the northern hemisphere extratropics. A half-monthly value was permitted to be formed from as little as a single pentad .

1 deg. areas that contained all possible calendar 'half-monthly' values were defined as 'qualifying areas' (Box 2 of Figure 4b). If there were fewer than four such 1 deg. areas in any 10 deg. area (Box 4 of Figure 4b), no background field was created for the 10 deg. area, except for a few 10 deg. areas with more than about 3000 observations but with data mostly missing at particular times of the year, e.g. some sub-arctic regions with winter ice (Boxes 6 and 7 of Figure 4b). In these cases, a similar selection of 'qualifying areas' was carried out using calendar 'monthly' averages from pentads grouped as in Table A1.2 (Boxes 8 to 11 of Figure 4b).

(Table A1.2) Definition of calendar 'months' used in processing SST and NMAT data from the MOMMDB

Month Pentads Calendar Equivalent

January 1-6 1-30 January February 7-12 31 January - 1 March (in all years) March 13-18 2-31 March April 19-24 1-30 April May 25-30 1-30 May June 31-36 31 May - 29 June July 37-42 30 June - 29 July August 43-49 30 July - 2 September September 50-55 3 September - 2 October October 56-61 3 October - 1 November November 62-67 2 November - 1 December December 68-73 2-31 December

(b) Smoothing of first guess 1 deg. area background field averages to produce initial pentad 10 deg. area background field averages

For each qualifying 1 deg. area the 24 (or 12) first guess background field averages were harmonically analysed (Box 6 of Figure 4a, expanded in Figure 4c). The results were used to synthesize 1 deg. area calendar pentad values, which varied smoothly in time. These values were averaged to make a first estimate of the 10 deg. area background average for the given pentad.

(c) Production of pentad 1 deg. area background field averages and derived 5 deg. area monthly averages

The above calendar pentad 10 deg. area averages were combined with available incomplete 1 deg. area calendar pentad background field averages from non- qualifying 1 deg. areas within the same 10 deg. area, to complete initial calendar pentad background field averages for these 1 deg. areas (Box 7 of Figure 4a, expanded in Figure 4d). This synthesis was, however, not carried out for 1 deg. areas having fewer than four pentad averages (Figure 4d). The 1 deg. area calendar pentad background averages for the 10 deg. area were then smoothed with the aid of timewise harmonic analysis and synthesis. The results were re-analysed into sets of harmonic coefficients (Boxes 8 to 10 of Figure 4a). However, some of the harmonic coefficients were still found to fluctuate spatially by amounts exceeding the likely real variations. Accordingly, each set of harmonic coefficients of a given order was smoothed using a polynomial filter applied firstly to each of the ten 1 deg. longitude bands and then to each of the ten 1 deg. latitude bands (Box 11 of Figure 4a); the order of the polynomial was automatically reduced if there were fewer than ten coefficients per row or column. This could result from the presence of land or ice, or from 1 deg. areas being absent because they had fewer than four original pentad values. The order of the polynomial was five if there were ten or nine coefficients in the row or column, four if there were only eight or seven coefficients, while a third order polynomial was used if fewer than seven coefficients were available. In this way all 1 deg. areas in processible 10 deg. areas were provided with harmonic coefficients, from which a final set of background 1 deg. area calendar pentad averages was synthesized (Box 12 of Figure 4a). Background field values were derived for 5 deg. areas and calendar months (as defined in Table A1.2) by averaging the appropriate 1 deg. area calendar pentad values.

(d) Discussion of background fields

The calendar pentad 1 deg. area means were now smooth within any 10 deg. area, but some discontinuities across 10 deg. area boundaries required smoothing. This was not done for the background field but the necessary smoothing was incorporated at a later stage for SST when the MOMMDB-MIT climatology was blended with the Alexander and Mobley climatology (see Section 4 of this Appendix).

Near ice-edges, the annual cycle of the background field may have been distorted by the limitation of SST and NMAT data to warmer than average (i.e. ice-free) years, with consequent biases of the order of 0.5 deg. C to 1 deg. C in the background field. Although systematic biases of this size should have had much smaller biasing effects on the accuracy of later stages of the procedure, similar problems are likely in the data used for the final SST and NMAT climatologies.

3. QUALITY CONTROL FLAGS (Figure 3 (in main text), Boxes 6 to 16)

(a) Range test (Figure 3 (in main text), Box 11)

In the range test, the standard deviation of all the winsorised monthly mean anomalies available for a given 5 deg. area and calendar month was calculated once and for all for the whole period 1854-1981. If the 5 deg. area monthly mean anomaly for a given year lay outside the limit ±3 standard deviations, the corresponding SST or NMAT value was flagged when archived

(b) Variance test (Figure 3, Box 13)

In the variance test, the pooled, weighted variance ₂ of winsorised pentad 1 deg. area values about their 5 deg. area mean was calculated for the given calendar month using all years (1854-1981) where at least 16 such values were available. The weights were set proportional to the number of 1 deg. area pentad values available in each year. Also calculated was the variance ₁ of the winsorised set of 1 deg. area values for the month and year under examination. When ₁ exceeded the empirically- chosen quantity 0.08n(₂ + 1), where n is the number of 1 deg. area pentad values available in the year under test, that monthly 5 deg. area value was flagged. This empirical expression was chosen, after large number of experiments, to flag approximately 0.5% of those monthly 5 deg. area values that were derived from 12 to 15 deg. area values, rising to 25% of those monthly 5 deg. area values that were derived from only four 1 deg. area values.

4. BLENDING OF MOMMDB/MIT AND ALEXANDER AND MOBLEY SST CLIMATOLOGIES

(a) Rationale

The climatology of Alexander and Mobley (1976) was used as a method of filling the empty areas in the SST climatology mainly because it was available in computer-compatible form. However, it is based on data for a variety of periods earlier than 1951-80 (mainly 1920-60), so the temperatures contained in it are probably negatively biased with respect to 1951-80 because of large-scale slow climatic changes (Folland, Parker and Kates 1984) Uncompensated changes of instrumentation may have caused an additional overall negative bias, though these instrumental biases appear to be small in the Southern Ocean (Section 3(h) of main text). Reynolds (1983) also suggested that the Alexander and Mobley climatology has a winter bias to lower temperatures that is greater than other climatologies. We therefore merged the Alexander and Mobley climatology into the MOMMDB-MIT SST climatology in such a way that the resulting values were as consistent as possible with the MOMMDB-MIT climatology.

(b) Procedure

An MOMMDB-MIT 1 deg. area monthly SST climatology was produced by averaging the constituent pentad values (see Box 8 of Figure 5 and Table A1.2). A few modifications were made to the resulting climatology e.g. in the Sea of Japan where the MOMMDB-MIT climatology had failed to represent adequately the large west-east gradients, especially in winter. Revised values in and around the Sea of Japan were derived using the 1951-80 climatology of the Japan Meteorological Agency (1983) as a strong guide to the 'true' climatology.

The following procedure achieved the goal of providing the empty areas with an SST climatology that retained the patterns of geographical variation contained in the Alexander and Mobley climatology but whose absolute values were anchored as far as possible to those of the generally warmer and more reliable MOMMDB-MIT climatology.

Empty areas of the MOMMDB-MIT climatology were filled in by solving the following Poisson equation in spherical coordinates:

² = ² A

where A is the Alexander and Mobley climatology (in a modified form that resulted from timewise harmonic analysis and synthesis to ensure smooth annual cycles) and is the resulting blended climatology (Reynolds and Gemmill 1985). Thus, essentially, the spatial rates of change of SST gradient in these areas of the MOMMDB-MIT climatology were set equal to those in the modified Alexander and Mobley climatology. All SST values were constrained to remain above -1.8 deg. C.

The solution of equation A1 is subject to internal and external boundary conditions imposed by the MOMMDB-MIT SST values and by Alexander and Mobley's SSTs along their ice limits. These ice limits are the only places where Alexander and Mobley data are used in a direct way to produce the blended field. However, their use is important as they will strongly influence the derived blended values over much of the Southern Ocean because of the great lack of MOMMDB-MIT data there. At lower latitudes MOMMDB-MIT data become more plentiful and their influence on the blended solution predominates.

The blended 1 deg. area climatology was smoothed with a 1:2:1 filter, first east-west and then north-south. This ensured that any residual 'jumps' across 10 deg. area boundaries in the MOMMDB-MIT climatology (for instance see Section 2(d) of this Appendix) were well-smoothed without affecting SST gradients on >200 km scales.

5. DETERMINATION OF LAND/SEA/ICE CLASSIFICATION FOR 5 deg. AREA CLIMATOLOGY

A 5 deg. area is defined to be land if and only if all 25 constituent 1 deg. areas are land. The 5 deg. area is ice if the number of ice 1 deg. areas exceeds the number of sea 1 deg. areas. Otherwise the 5 deg. area is sea.

APPENDIX 2

First list

Second list

Third list

Fourth list

APPENDIX 3