THE DEVELOPMENT OF MONTHLY TEMPERATURE SERIES FOR SCOTLAND AND NORTHERN IRELAND

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 24: (2004) Published online in Wiley InterScience ( DOI: /joc.1017 THE DEVELOPMENT OF MONTHLY TEMPERATURE SERIES FOR SCOTLAND AND NORTHERN IRELAND P. D. JONES* and D. LISTER Climatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UK Received 11 June 2003 Revised 19 December 2003 Accepted 8 January 2004 ABSTRACT We develop monthly temperature series (mean, maximum and minimum) for the Scottish mainland (SMT), the northern and northwestern Scottish Isles (SIT, Orkneys, Shetlands and Outer Hebrides) and for Northern Ireland (NIT). The series were developed from eight long station records, five over SMT, two for SIT and one for NIT. The series are reliable back to 18 for all three regions, but extensions to 1800 and 1827 for SMT and SIT are less so, likely due to direct solar insolation effects in the summer months. All three regions show long-term statistically significant seasonal warming over both and , except for the winter season. For the calendar year, the temperature change explained by a linear trend over is 0.9 C, 0.4 C and 0.77 C for SMT, SIT and NIT respectively. Of the 51 (12 months + 4 seasons + annual = 17 3 regions) warmest values, 23 have occurred after Of the coldest extremes, 30 occurred between 181 and 1899 and the last cold extreme months were in Copyright 2004 Royal Meteorological Society. KEY WORDS: temperature; Scotland; Northern Ireland; North Atlantic oscillation 1. INTRODUCTION The central England temperature (CET) series (Manley, 1974) is the longest monthly series in the world, reaching back to 159, and is widely used as a measure of thermal conditions over the British Isles. It has been updated, extended to a daily series from 1772 and adjusted for a small urbanization influence by Parker et al. (1992). CET is areally representative of the triangle of Manchester, Oxford and Cambridge; but, despite this, seasonal correlations with the most extreme western and northern observation sites in the British Isles (Valentia and Lerwick) are highly statistically significant (0.75 and 0.5 respectively on an annual basis, over the period). As expected, correlations with CET are stronger in the zonal, relative to meridional, directions (Jones and Hulme, 1997). The purpose of this paper is to develop additional national monthly temperature series (mean T,maximum T x, minimum T n and diurnal temperature range (DTR)) for Scotland and Northern Ireland. The study extends initial work by Purves et al. (2000). The emphasis of this study is the T records, as greater potential homogeneity issues affect DTR and hence the T x and T n series. The impetus for the series partially relates to devolution within the government of the United Kingdom (UK), and the need for more locally representative series (see Kerr et al. (1999)). The greater spatial variability of precipitation has already led to the development of regional and national series (see Alexander and Jones (2001) and references cited therein). The paper is organized as follows: Section 2 documents the candidate data sources; Section 3 documents the homogeneity (both over the long term and with respect to outlier assessment) and reasons for the use of the selected sites; Section 4 details the development of the national series and compares the series with * Correspondence to: P. D. Jones, Climatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UK; p.jones@uea.ac.uk Copyright 2004 Royal Meteorological Society

2 570 P. D. JONES AND D. LISTER neighbouring sites (CET, Valentia in the Republic of Ireland, Tørshavn on the Faroe Islands and Bergen in western Norway); Section 5 considers relationships with the North Atlantic oscillation (NAO); and Section concludes and discusses suggestions for further work. 2. DATA SOURCES Several long T records are included in datasets available within the Climatic Research Unit (CRU) and the Global Historic Climate Network (GHCN; Peterson and Vose, 1997). The original source of this information is the World Weather Records (WWR) volumes, data internationally exchanged over the World Meteorological Organization (WMO) networks (both the climate monitoring (CLIMAT) network and through the WMO publication Monthly Climatic Data for the World). Some Scottish and Irish stations are included in the North Atlantic Climatological Dataset (NACD; Frich et al., 199), this source providing homogenized monthly series for the period. Data for the last few years of a number of records were obtained from the UK Met Office, through the British Atmospheric Data Centre (BADC). Nordic stations (including Tørshavn and Bergen) have been updated to 1999 by Tuomenvirta et al. (2001). In the earlier work of Purves et al. (2000), original archives were digitized for monthly averages of T, T x and T n, and homogeneous series developed for a number of locations. These are referred to here as the SNIFFER series. In our work, however, we have used the original series digitized by Purves et al. (2000), who also combined a number of shorter length records, using monthly regression relationships from overlapping periods, to produce longer and complete composite series for the Edinburgh and Glasgow areas. Extensive digitization of all daily data from the original ledgers has recently been undertaken for Armagh Observatory (see Butler and Johnston (199; Butler et al., 2003)) in Northern Ireland. Table I lists the candidate stations, together with their various principal sources. 3. SELECTION AND HOMOGENEITY OF THE SITES TO BE USED 3.1. Rationale for selection Five basic principles determined the site selection: 1. Long continuous series. 2. Homogeneous over their entire length. 3. Well-spaced locations over Scotland and Northern Ireland. 4. Easily updateable in near-real time and likely to continue for the foreseeable future. 5. Located at relatively rural locations. Although the first three principles are imperative in any study of this kind and the homogeneity assessment took the most time, the other two were the most important in determining the final choice. The ease of updating was paramount: all the selected sites currently provide monthly National Climate Messages that are available to the Met Office within a few days of the end of each month. Values for the national series will then be available for press releases when near-extreme monthly, seasonal and annual temperatures are being recorded. Eight key sites (Lerwick, Stornoway, Wick, Braemar, Leuchars, Auchincruive, Eskdalemuir and Armagh) were selected (see Table II and Figure 1). Armagh is the only long-term record in Northern Ireland, but it is the most widely studied of the series, with all the daily data available in digital form. Digitizing all the original daily observations has been shown to be the best approach for unravelling all issues related to homogeneity for a number of European locations (Camuffo and Jones, 2002). For Scotland, the seven sites are widely dispersed, and all but two would be classed as being both coastal and relatively low elevation. The exceptions are Eskdalemuir (242 m), representative of the southern Scottish uplands and Braemar (339 m) of the Scottish Highlands. No other long upland sites have a century of records, although observations were taken on Ben Nevis for the period (Roy, 1983), and in time

3 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES 571 Table I. List of the candidate stations, their sources and lengths of record. Abbreviations given in the text in Section 2 and additional details of the comment section discussed in Appendix A Location Altitude (m) Source Period Comments a Aberdeen 59 CRU Dyce Airport Armagh 2 Met Office Armagh 2 Armagh Observatory Homogenized series from the observatory Armagh 2 SNIFFER Auchincruive 48 BADC Missing data filled by Scottish Agricultural College Bergen 3 CRU To be used later to check regional series Braemar 339 CRU/BADC Identical to NACD file Braemar 339 NACD Braemar 339 SNIFFER Balmoral data CET 50 Met Office Parker et al. (1992) updated Dumfries 49 CRU Identical to NACD file Dumfries 49 NACD Dumfries 49 SNIFFER Edinburgh ROE 134 CRU Royal Observatory Edinburgh Airport 35 CRU Turnhouse Airport Edinburgh ROE 134 GHCN Edinburgh 2 NACD This shows differences with CRU ROE Edinburgh 2 SNIFFER Based on LNS, ROE and RBG (see Appendix A) Eskdalemuir 239 CRU/BADC Glasgow Unknown CRU Glasgow 20 SNIFFER See Paisley (below) Kirkwall Variable CRU Identical to NACD for the overlap period Kirkwall 2 NACD Kirkwall 20 SNIFFER Many missing values Leith Navigation School Unknown SNIFFER Lerwick 82 CRU/BADC Identical to NACD for the overlap period Lerwick 82 NACD Lerwick coast guard 50 SNIFFER Lerwick observatory 82 SNIFFER Leuchars 10 BADC Paisley 32 SNIFFER based on Glasgow University Stornoway 15 CRU/BADC Identical to NACD for the overlap period Stornoway 15 NACD Stornoway 11 SNIFFER Tørshavn 39 CRU To be used later to check regional series Wick 3 CRU/BADC Identical to NACD for the overlap period Wick 3 NACD Wick 3 SNIFFER Valentia 14 CRU To be used later to check regional series a See Appendix A for details. automatic weather stations may provide long series. More upland sites recorded temperatures in the late- 19th century (see the annual volumes of the Journal of the Scottish Meteorological Society for the 1870s to about 1920). Eskdalemuir, Stornoway and Lerwick have been designated by the Met Office to be part of the Global Climatological Observing System (GCOS) surface network (GSN) for the UK. The origins of GSN are discussed in Peterson et al. (1997).

4 572 P. D. JONES AND D. LISTER Table II. The eight selected series Site Period (T ) Period (DTR) Comments a Lerwick NACD/CRU/BADC extended back before 1890 with CRU Kirkwall Stornoway NACD/CRU/BADC extended back using SNIFFER series Wick NACD/CRU/BADC extended back using SNIFFER series Braemar NACD/CRU/BADC extended back using SNIFFER series Leuchars Leuchars series extended back using the Edinburgh record (which uses the GHCN and SNIFFER series) Auchincruive Auchincruive extended back using the Paisley SNIFFER series Eskdalemuir Eskdalemuir extended back using the various Dumfries series Armagh This series provided by Armagh Observatory a See Appendix A for details Homogeneity of the selected series Although a number of the series have been extensively assessed for homogeneity by a variety of methods (for a discussion of methods see Peterson et al. (1998)) in earlier studies, this section discusses the additional homogeneity procedures undertaken during this study. Both the NACD source and Purves et al. (2000) have used the objective homogeneity approaches introduced by Alexandersson (198; Moberg and Alexandersson, 1997). Here, we consider each site record by a subjective assessment of annual station difference plots (between neighbours) of T and DTR, combined with knowledge of each station s history from metadata. Our approach was designed to locate all the potentially important discontinuities in the station series and to derive appropriate adjustment factors. We also strove to determine the reason for the jumps, but not to correct series where jumps were only sustained for 3 to 4 years. The results of this extensive exercise for the Scottish and Northern Irish sites are given in detail in Appendix A for each of the eight key sites. This gives a brief history of the sources available, details of the neighbouring sites used for difference plots, and the periods and values of adjustments calculated. The procedure was not straightforward, and for much of the period before about 1920 requires some trial and error, as many of the sites and their neighbours need adjustment during similar periods. As the more complex stations were in the central lowlands of Scotland, we began the homogeneity exercise at Lerwick/Kirkwall and Wick and progressed south; at the same time, we worked north from Armagh and Eskdalemuir. For the period from about 1870 onwards, T was calculated from the average of T x and T n. It is somewhat easier to assess homogeneity by considering T and DTR (T x T n ). Prior to about 1870, only T is available, generally calculated from observations at a few fixed hours during the day. Possible procedures for assessing the homogeneity of these types of record are addressed in Camuffo and Jones (2002). In Figure 2 we give two examples of the station difference plots, one for T and one for DTR. All possible difference plots for adjacent pairs were examined for jumps and/or trends. Homogeneity adjustments to parts of the series were calculated for periods before and after the presence of clear jumps (which could generally be associated with known causes) on a monthly basis. As the break points were determined on an annual basis, we also inspected difference plots of the extreme seasons (winter December February (DJF) and summer June August (JJA)) to ensure that there were no cancelling jumps that would not be found in the annual plots. The DTR difference plot is erratic and indicative of several jumps, but none can be simply related to changes at either site (see also the discussion with Figure 4). Appendix A gives details of all the adjustments

5 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES 573 Figure 1. Locations of the eight selected series and additional sites named in the text in Table A.I for T and in Table A.II for DTR. After completion of the exercise for each station, T x and T n series were produced by addition/subtraction of 0.5 DTR from the monthly T series. This ensures that, for the whole period of T x and T n measurements, T is the average of the two observations. Figure 3 shows decadally smoothed plots of the eight series for annual average T and for the winter and summer seasons. For annual temperatures, Armagh is the warmest location, with Braemar the coolest. In summer, the coolest location is Lerwick, with Braemar and Eskdalemuir the coldest in winter. Figure 4 shows similar time series for DTR. Changes in DTR are markedly less coherent than for T, reflecting the differences between coastal and inland sites. DTR station difference plots (e.g. the lower panel of Figure 2) are also often erratic and suggestive of additional homogeneity problems. Station history information, however, generally does not always indicate site, instrument or observational practice changes. Further analysis of the inspection records, held by the Met Office, might indicate possible reasons relating to repair or repainting of screens, thermometer calibrations and/or changes in the immediate vicinity of the site. Changes such as these are likely to have differential effects in the T x and T n series, so will be accentuated in DTR and reduced in T. It is also likely that some of these changes took place but were not recorded due to the changes being considered unimportant. With this in mind, we have much greater confidence in the long-term homogeneity

6 574 P. D. JONES AND D. LISTER (a) Difference ( C*10) (b) Difference ( C*10) (c) Difference ( C*10) Figure 2. Example homogeneity plots: (a) the difference between Armagh and Stornoway annual temperatures ( ) before adjustment; (b) after adjustment; (c) the difference between Stornoway and Auchincruive annual DTR of the T records than the T x and T n (and hence DTR) series. The conclusion with respect to DTR is in accord with Wijngaard et al. (2003). They found that only % of the 195 daily temperature series in the European Climate Assessment dataset were homogeneous for DTR for the period. All sites show a slight reduction in annual average DTR, but none of the trends is statistically significant. DTR reductions have been found in many regions of the world since the 1950s (Karl et al., 1993; Easterling et al., 1997) and, when spatially averaged for continental scales, are highly statistically significant. No significant reductions over the British Isles, however, have been found (Horton, 1995) Assessment of outliers During the homogeneity exercise, several odd monthly mean values for T, T x, T n and DTR were noticed. To determine whether the value was correct, we calculated anomaly values (based on monthly data for the period) for each series and on the basis of neighbour comparisons assessed all monthly values in excess of an absolute difference of 2 C. We first determined from the neighbour differences which site was likely giving the incorrect value; then we examined the various sources for this month. Values were then subject to alteration to that given in one of the other sources and were corrected (generally changing the sign of the erroneous value) or were accepted. Most of the outliers relating to Braemar in the winter half of the

7 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES 575 Figure 3. Decadally smoothed mean temperatures for the eight selected series. Values are plotted for the calendar year and for the winter (DJF) and summer (JJA) seasons. Smoothing is achieved using a 10 year Gaussian filter year were accepted, these clearly relating to the presence of extensive snow cover during these months at this site. Such cold anomalies were not present at the coastal sites of Wick and Leuchars. This exercise only corrected seven observations, at Wick, Stornoway and Dumfries (see Appendix A for details). 4. NATIONAL SERIES DEVELOPMENT Table III gives correlations between the eight sites based on all months for the period (1572 values). Prior to calculation, the annual cycle was removed from each station by subtracting the monthly average for the period. As expected, correlations are highest between adjacent locations. Correlations are generally lowest with the most northerly location at Lerwick. Here, correlations are highest with Wick and Stornoway (about 0.1 higher than the other sites). Similar correlation matrices for T x and T n (lower part of Table III) give slightly lower values (by on average), more markedly so for the T n series for Braemar. Of the eight sites, Auchincruive, Braemar, Leuchars and Eskdalemuir form a distinctive group. Average site correlations (each with the other seven sites) are lower for the other four locations.

8 57 P. D. JONES AND D. LISTER Figure 4. Decadally smoothed DTR series for the eight selected sites. Values are plotted for the calendar year and for the winter (DJF) and summer (JJA) seasons. Smoothing is achieved using a 10 year Gaussian filter 4.1. Calculation of the national series Based on this correlation matrix (Table III) it was decided to develop two Scottish series (as also undertaken by Purves et al. (2000)), one for the five mainland locations (SMT) and one for the northern islands (SIT), using the two sites at Stornoway and Lerwick. The Northern Ireland temperature series (NIT) will be based solely on the Armagh record. An argument can be made for Wick being part of the SIT pair, but we kept it in the SMT group as it is located on the Scottish mainland. Averages for SMT, SIT and NIT were simply the unweighted averages of monthly temperatures, expressed as anomalies from the base period. For SMT, areal weighting was not considered necessary, as the sites were relatively well spaced. For the earliest years, when fewer than the complete set of stations were available, the averages could be adjusted (variance reduction) so that they conform statistically to the full (for SMT, five stations) site average (Jones et al., 1997a). This step was not undertaken, however, as the reductions were less than 0.1 C.

9 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES 577 Table III. Correlations between the eight key sites based on all months for the 1872 to 2002 period. Each month was expressed as an anomaly from the period before calculation of the correlation coefficients. T correlations are in the top matrix, with T x (above diagonal) and T n (below diagonal) in the lower matrix Auchincruive Braemar Eskdalemuir Leuchars Wick Stornoway Lerwick Armagh Auchincruive Braemar Eskdalemuir Leuchars Wick Stornoway Lerwick Auchincruive Braemar Eskdalemuir Leuchars Wick Stornoway Lerwick Armagh Absolute temperatures for the three national series The simplest way of producing absolute temperatures would be to add back the base-period means by averaging the five sites for SMT and two for SIT. As this may be potentially biased for SMT, owing to there being only one truly high-elevation site at Braemar, we have calculated true areal means for the three regions. We did this by averaging all 5 km grid-box values developed for in the recent UKCIP02 (UK Climate Impacts Programme) report (Hulme et al., 2002: appendix 7). This uses data from many more shorter record stations, which cover the period, and additional factors such as latitude, longitude, elevation and distance from the coast. Regression relationships were then derived in the UKCIP02 report to estimate temperature for each 5 km grid box. The average of these, for the three regions, will provide better estimates of true surface mean temperature for each of the three regions. From the UKCIP02 appendix, for SMT we used all 5 km grid values for the Scottish mainland and all islands (except for the Outer Hebrides, Orkneys and Shetlands) large enough to have estimates. SIT was the average for the outer islands, and NIT used all grid-box values in Northern Ireland. Table IV gives true monthly values for each of the three regions for T, T x and T n and compares these with the average of sites that make up the regional averages. For SMT and NIT, the UKCIP02 values are lower. For SMT, differences are in the range 0.14 to 0.91 C, with a more marked annual cycle for T x with larger differences in summer. The differences are larger still for NIT (0.13 to 1.48 C), again with larger values in summer, particularly for T x. For SIT, differences are negative in the range 0.0 to 0.51 C, with a marked annual cycle for T x compared with T n. The SIT differences are negative, principally because the area of the Outer Hebrides is larger than the combined area of the Orkneys and the Shetlands. The 5 km grid-box averages were then added to the anomaly series to produce absolute temperature series for each of the three regions. The differences given in Table IV are simply offsets and do not affect the time series properties (e.g. correlations, trends) of the series. They only influence the reference-period means. Figures 5, and 7 show seasonal and annual temperatures (in absolute degrees Celsius) for the three national series (SMT, SIT and NIT respectively), all plotted using the same vertical scales. Extremes in each record and trends will be discussed in later sections. For all three national series, the variability from year to year and decade to decade is slightly greater in winter than the other seasons and is markedly lower in the annual series. For SMT and SIT, when the early parts of the record are based on only one location before 18 in both series, there are suggestions of potential inhomogeneities in the early part of the Leuchars

10 578 P. D. JONES AND D. LISTER Table IV. Absolute temperatures for derived from UKCIP02 for SMT, SIT and NIT compared with the respective simple averages of the data for the selected sites. is the station average minus UKCIP02 T ( C) T x ( C) T n ( C) UKCIP Site ave. a UKCIP Site ave. a UKCIP Site ave. a SMT Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year SIT Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year NIT Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year a Site averages are for five, two and one value for SMT, SIT and NIT respectively.

11 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES Winter Spring Summer 10 Autumn 8 10 Annual Figure 5. Seasonal and annual temperature time series for SMT. Smoothing is by means of a 10 year Gaussian filter. The horizontal line is plotted at an arbitrary whole degrees Celsius value to assist viewing. The line at 18 is plotted to indicate greater uncertainty before this date and Lerwick records, particularly during the summers. In Figures 5 and, the pre-18 part of the series is differentiated from the latter period by a vertical line. This greater uncertainty is a common feature of a number of long northern European records, and the influence of direct sunlight on the thermometers in summers has been suggested as a cause (Parker, 1994). An extensive study of two Swedish records that extend back to the 18th century (Uppsala and Stockholm) has provided evidence for this and the suggestion is that for these two sites the summer temperatures are biased warm by C before the 180s (Moberg et al., 2003). Stevenson (see discussion in Mawley (1884)) developed his now almost ubiquitous louvred screen in the late 180s early 1870s. Prior to this time, thermometers were sited on north-facing walls of buildings; so, the further north the location, there is an increased chance of direct exposure of the sun on the thermometer bulb, particularly on summer mornings. Because the early series have these potential biases, extremes and trends are only considered from 181 onwards. In Appendix A we state that both the Leuchars (Edinburgh) and Lerwick (Kirkwall) records

12 580 P. D. JONES AND D. LISTER Winter 0 10 Spring Summer 10 Autumn 8 10 Annual Figure. Seasonal and annual temperature time series for SIT. Smoothing is by means of a 10 year Gaussian filter. The horizontal line is plotted at an arbitrary whole degrees Celsius value to assist viewing. The line at 18 is plotted to indicate greater uncertainty before this date would benefit from more detailed homogeneity assessment, particularly by digitizing the entire daily series (cloudiness estimates, as well as temperatures) enabling the summer issue to be dealt with in detail, as in the Uppsala/Stockholm case Monthly, seasonal and annual extremes in the three records Table V lists the warmest and coldest values based on the period. Of the 51 ((12 months + 4 seasons + annual) 3) warmest values, 23 have occurred since The most recent cold extreme was September 1952 for NIT. Of the coldest extremes, 30 occurred in the period compared with only six for the warmest extremes.

13 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES Winter 10 Spring Summer 10 8 Autumn 10 Annual Figure 7. Seasonal and annual temperature time series for NIT. Smoothing is by means of a 10 year Gaussian filter. The horizontal line is plotted at an arbitrary whole degrees Celsius value to assist viewing 4.4. Trends in the regional series and comparisons with neighbouring sites Table VI gives the temperature change ( C) explained by a linear trend fit over the and periods for the three series developed here, the four neighbouring locations (CET, Tørshavn, Bergen and Valentia) and the Northern Hemisphere (NH) average. The NH average series is based on land and marine data (HadCRUT2v; Jones and Moberg, 2003). The chosen periods were those used in the Third Assessment Report of the Intergovernmental Panel on Climate Change (Houghton et al., 2001). All seasons, except winter, show significant warming. SMT trends are very similar in both periods to CET, although SMT warms more than CET in spring (MAM) and CET more than SMT in autumn (SON). SIT warms by slightly lower values than SMT, but more than Tørshavn on the Faroe Islands. Trends for NIT are similar to SIT and to Valentia in southwest Ireland. Over the 20th century, SMT warms by a similar amount compared with the NH average, with SIT and NIT warming slightly less. A sea-surface temperature (SST) series for 53 5 N by 1 E 10 W was derived from the dataset described by Rayner et al. (2003). This series extends from 1870 to 2002 and over the period warms by 0.37 C (significant at the 95% level).

14 582 P. D. JONES AND D. LISTER Table V. Monthly, seasonal and annual temperature extremes for three regional temperature series based on the period (DJF dated by the year of the January) Period a SMT SIT NIT Year Extreme ( C) Year Extreme ( C) Year Extreme ( C) Warmest Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec DJF MAM JJA SON Year Coldest Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec DJF MAM JJA SON Year a MAM: March May; SON: September November. Figure 8 compares the three regional series with neighbours (SMT with CET, SIT with Tørshavn and Bergen, and NIT with Valentia). SMT and CET are in excellent agreement throughout the 200 years, with the exception of the 1840s and 1850s. The coolest decade since 181 in all three series is the 1880s, which is in agreement with CET and also the NH average series. For SIT and NIT the comparison series (Tørshavn and Valentia respectively) appear to have homogeneity problems (before 1880 and 1890 respectively). The Valentia site was moved from Valentia Island to the mainland in the late 19th century. Tørshavn has warmed less in recent decades compared with SIT and Bergen, as has Valentia compared with NIT. The SST series for the waters around Scotland is shown in the panel with the SIT series.

15 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES 583 Table VI. Seasonal and annual temperature change explained by a linear trend ( C) in the three regional, four neighbouring sites and NH temperatures over and (bold indicates significance at the 5% level) SMT SIT NIT CET Bergen Tørshavn Valentia NH a DJF MAM JJA SON Annual DJF MAM JJA SON Annual a HadCRUT2v from Jones and Moberg (2003). 5. RELATIONSHIPS BETWEEN THE TEMPERATURE SERIES AND NAO The NAO is known to exert a strong influence on European temperatures (Hurrell, 1995; Jones et al., 1997b). Table VII gives correlations between seasonal values of the NAO (based on normalized pressure series for Ponta Delgada (Azores) and Reykjavik (Iceland)) and the seven local temperature series discussed in Section 4.4 for the period. Correlations are strongest in the winter season and just reach significance in the spring season. Exceptionally cold winters are all related to average NAO values < 2 (weaker westerlies and even easterlies during part of the season) and exceptionally mild winters are linked to NAO values >2 (stronger westerlies). Figure 9 shows comparisons of the three regional series and the NAO for the winter season. The NAO clearly exerts a strong influence, but this varies with time and is noticeably weaker between 1900 and 1940 (also see Jones et al. (2003)). Significant economic benefits would be made with successful NAO forecasts, but these are not that likely to be good enough in the near future. The latest views on the potential for NAO predictability are given by Rodwell (2003). Osborn and Jones (2000) identified the amount of temperature change that could be explained by atmospheric circulation variability, as measured by wind directions, strength and vorticity using daily CET data. Though much of the seasonal time-scale variability can be explained, the long-term trends are barely altered, but they become much more statistically significant because of the reduced year-to-year variability. We expect similar results for the three series here, but we would need daily data to repeat the exercise, and this is only available at present for Armagh. In a related study of the long De Bilt record in the Netherlands, van Oldenborgh and van Ulden (2003) have found that temperatures have increased over the 20th century for all wind directions except northeast. So, both studies indicate that the century time-scale warming is not a result of more days with winds from warmer directions.. CONCLUSIONS AND SUGGESTIONS FOR FURTHER WORK Based on eight long temperature records, we have developed regional monthly temperature series (for T, T x and T n ) for SMT, SIT, NIT back to the mid-19th century. The most time-consuming aspects of the work were

16 584 P. D. JONES AND D. LISTER SMT and CET SIT, Bergen, Thorshavn and SST NIT and Valentia Figure 8. Decadally smoothed annual temperatures comparing the three regional series (SMT, SIT and NIT, bolder lines) with CET, Valentia, Tørshavn (Faroes) and Bergen (thinner lines). On the top panel, the SMT scale is to the left and the CET scale is to the right. On the bottom panel the NIT scale is to the left and the Valentia scale is to the right. Vertical scales on the middle panel are the same for all the station series, where Bergen is the upper of the thinner lines. The SST series (from Rayner et al. (2003) and representative ofthearea53 5 N by1 E 10 W) is the thicker of the lines running from 1870 to 2002 in the middle panel with the scale ( C with respect to ) on the right Table VII. Seasonal correlations between the NAO and the three national and the four neighbouring series over the period (bold indicates correlations significant at the 5% level). NAO is defined as the normalized pressure difference between Ponta Delgada (Azores) and Reykjavik (Iceland) DJF MAM JJA SON SMT SIT NIT CET Bergen Tørshavn Valentia

17 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES SMT and NAO SIT and NAO NIT and NAO Figure 9. Winter (DJF) SMT, SIT and NIT temperatures (thinner lines) compared with the winter NAO series (thicker lines). The NAO series used here is based on the difference of the normalized pressure series at Ponta Delgada (Azores) and Reykjavik (Iceland). Each series has been smoothed with a 5 year Gaussian filter. The temperatures scales are to the left and the NAO scale is to the right the construction of the eight homogeneous station series. Selection of the eight records was based on length of record and on the ease of updating the records in real time for climate monitoring purposes. Homogeneity was assessed using annual, winter and summer station difference time series between all adjacent pairs of stations. Adjustment to parts of all records was made on a monthly basis, using the plots and information about station relocations. Homogeneity proved more difficult with the pre-1920 records and, because of a lack of neighbouring records, was problematic before about 180. We recommend that more extensive homogeneity assessments of the Leuchars (Edinburgh) and Lerwick (Kirkwall) records be undertaken using daily data (which will have to be digitized for the period before the 1950s) along the lines of that undertaken for Armagh and for a number of European locations. Subsequent homogeneity of the records would be aided if the daily records of pressure, cloudiness and precipitation were digitized at the same time. This could lead eventually to a Scottish daily temperature series. The three series SMT, SIT and NIT show long-term annual warming of 0.9 C, 0.4 C and 0.77 C respectively over the period. Warming is generally slightly greater in the autumn and is negligible and statistically insignificant during the winter. The three serieswere comparedwith long neighbouring records (SMT with CET, SIT with Tørshavn (Faroes) and Bergen, and NIT with Valentia, Ireland). SMT and CET have similar trends. For SIT, Bergen warms by a greater amount and Tørshavn less. For NIT, Valentia warms a little less. Both the Valentia and Tørshavn records appear to have homogeneity problems in the mid-to-late 19th century. Twenty-three of the warmest months, seasons and years (out of a total of 51 series: 17 time series and three locations) in the period 181 to 2002 have occurred since The last coldest month was September 1952 for NIT, with 30 of the 51 coldest extremes occurring in the period. The NAO is the principal circulation influence on the three series. Correlations are of the order for the winter and are also weakly significant in spring.

18 58 P. D. JONES AND D. LISTER ACKNOWLEDGEMENTS We are grateful for the support of SNIFFER (contract number CC(01) 230/8039, Project Manager, Marc Becker) and the help of the earlier Purves et al. (2000) report. John Butler of the Armagh Astronomical Observatory and Billy Harrison of the Scottish Agricultural College at Auchincruive gave their data freely and helped with a number of follow-up questions. Simon Tett, Briony Horton and David Parker (of the Hadley Centre, Met Office) assisted in the determination of the key sites and commented upon the draft manuscript. The comments of one of the reviewers also helped. APPENDIX A DETAILS OF THE HOMOGENEITY ASSESSMENT OF THE EIGHT SELECTED SITES A.1. Armagh (180) The original daily records for the Armagh Observatory have been digitized and homogeneous monthly (and daily) records for T, T x and T n developed by Butler and Johnston (199; Butler et al., 2003). The current record extends back to 1844, and ongoing work at the observatory is expected to extend this back for T to the late-18th century (Butler, personal communication). Coughlin and Butler (1998) have also assessed monthly data for the last 50 years for urban influences, but found these to be negligible. All our homogeneity checks revealed no problems with the T and DTR records. All daily data will also soon be available, complementing the precipitation data already on the Armagh Observatory Website ( A.2. Eskdalemuir/Dumfries (1871) Observations began at Eskdalemuir in about 1908, but we have only considered those digitally available from Homogeneity assessment using the records at Armagh, Dumfries, Paisley and Edinburgh revealed no problems with post-1930 T and DTR records. To extend the record we have added the adjusted Dumfries record for the period. The CRU and NACD records for Dumfries were identical for the overlap period of The SNIFFER series (Purves et al., 2000) extends from 1871 until Based on comparisons with Armagh and Paisley, adjustments for two periods ( and ) were calculated. For the first period, the adjustments were the same as the difference between the SNIFFER and NACD series for at least 20 years after The reason for the second correction for is unclear. The adjustments are given in Table A.I for T and in Table A.II for DTR, the latter being necessary for the Dumfries DTR measurements prior to The T values in March in 1892 and 1893 were both adjusted based on comparisons with Armagh and Paisley. A.3. Auchincruive/Paisley (188) The Paisley record is based on Glasgow University records for and Paisley from 1884 and was developed in the earlier SNIFFER work (Purves et al., 2000). As the Paisley site does not report in real time, the Auchincruive record will be used for the period after Auchincruive is a more coastal location and the adjustment factors for T and DTR for Paisley ( ), derived by comparison of the two records over the period, reflect this. The SNIFFER record for Paisley was additionally adjusted for the two periods (188 7 and ) based on comparisons with Armagh, Dumfries, Edinburgh and CET. The adjustments are given in Table A.I for T and in Table A.II for DTR, the latter being necessary for the Paisley DTR measurements prior to The use of Auchincruive data for the recent 40 years eliminates the possibility of any small urban influence in the Paisley data for this period. A.4. Leuchars/Edinburgh (1800) Edinburgh records were available from CRU, GHCN/WWR, NACD and SNIFFER. The longest continuous records, which extend back to 174, were developed by Mossman (189, 1897, 1902). This source is used for the records given in WWR and GHCN, where the site is stated to be the Royal Observatory. The earlier

19 SCOTLAND AND NORTHERN IRELAND MONTHLY TEMPERATURE SERIES 587 Table A.I. Monthly adjustments added to the source records to achieve homogeneity for T for the periods stated Station Period Adjustment ( C 10) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Dumfries Dumfries Dumfries a Paisley Paisley Paisley a Edinburgh Edinburgh a Braemar Braemar Wick Stornoway Kirkwall Kirkwall Kirkwall Kirkwall a a This record forms part of one of the eight long composite records. The modern part of the Dumfries record is Eskdalemuir, Paisley is Auchincruive, Edinburgh is Leuchars and Kirkwall is Lerwick. Table A.II. Monthly adjustments added to the source records to achieve homogeneity for DTR for the periods stated Station Period Adjustment ( C 10) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Dumfries a Paisley a Edinburgh a a This record forms part of one of the eight long composite records. The modern part of the Dumfries record is Eskdalemuir, Paisley is Auchincruive, Edinburgh is Leuchars and Kirkwall is Lerwick. SNIFFER project digitized separate records for the Royal Botanical Gardens (RBG), the Royal Observatory (ROE) and Leith Navigation School (LNS), developing a composite record based on LNS for , ROE for and RBG from Based on comparisons of data from the various sources and neighbouring records from Paisley, Armagh and Wick, we decided to use the composite SNIFFER record for , extended to 1800 by the GHCN/WWR/Mossman source. Homogeneity issues remain for the pre-180 period, based on comparisons with CET. None of the Edinburgh sites reports in real time, so it was necessary to use the nearby record from Leuchars, across the Firth of Forth. An adjustment was necessary for the Edinburgh data (based on Armagh and CET) and for the whole of the Edinburgh record up to 195 to make it compatible with the Leuchars record from 1957 onwards. The adjustments are given in Table A.I for T and in Table A.II for DTR, the latter being necessary for the SNIFFER Edinburgh DTR measurements prior to 195. As for Auchincruive, the use of Leuchars data from 1957 eliminates possible urban effects at the Edinburgh sites during the last 45 years. The Edinburgh record clearly warrants more extensive study, along the lines of that undertaken for Armagh and for a number of European locations in Camuffo and Jones (2002). Digitizing all the original daily data

20 588 P. D. JONES AND D. LISTER (not just for temperature, but for precipitation, cloudiness and pressure) would probably indicate some further adjustments, particularly for years before 180. Edinburgh, along with London, Armagh, Durham and Oxford, represents a location where long continuous daily records have been taken within the UK. At the present time we consider the homogeneity checks for Edinburgh we have performed as being adequate, whereas we would class the work performed at Armagh as excellent. A.5. Braemar (18) This is the only long record high-elevation site in the Scottish Highlands. The CRU and NACD records for the site are identical and cover the period. The SNIFFER records are longer ( ) and differ from the other sources for the period before Comparisons with the neighbouring sites of Wick, Aberdeen, Paisley and Edinburgh revealed two problem periods: in all sources, and before 190 in the SNIFFER record. The original records for the period indicate that the measurements were taken at Balmoral, with the original instruments from Braemar. It appears that they were returned to Braemar in late 1911, but not to the original site pre-190. Adjustments were made for two periods ( and ) and are given in Table A.I for T. No additional adjustments were made for possible DTR differences between Braemar and Balmoral for the period. The early part of the Braemar record is discussed in more detail by Mossman (1892). A.. Wick (1872) The CRU and NACD series are identical over the period. The SNIFFER series, which covers the period , is considerably longer, but indicates significant differences with the other sources over the period. For 1911 onwards, the SNIFFER source agrees with CRU/NACD. Comparisons with the neighbouring sites at Stornoway and Kirkwall indicate that the SNIFFER series is more likely to be the correct one. Adjustments were still necessary for the period even with the better SNIFFER series. The adjustments are given in Table A.I for T.TheT value for February 1934 was corrected based on comparisons with Stornoway and Kirkwall. A.7. Stornoway (18) The CRU and NACD records were identical over the period. The SNIFFER series was generally the same except for the period, where differences appeared to be somewhat random in nature. We extended the CRU/NACD series by using the SNIFFER series for and we assessed the homogeneity based on Wick, Armagh and Kirkwall. Adjustments were necessary for the period. The adjustments are given in Table A.I for T.ValuesofT were corrected for Januaries in 1880 and 1883, February 1890 and September 1903, based an anomaly comparisons with Wick and Kirkwall. A.8. Lerwick/Kirkwall (1827) Both the Lerwick and Kirkwall records for the period were identical in their respective sources from NACD and CRU. Comparisons between the two sites indicated problems with the Lerwick record before As the aim was to use Lerwick for 20th century and Kirkwall for the 19th century, it was decided not to try and correct the Lerwick T record for the years before 1910 but to use the Kirkwall record up to More details are given about the 19th century Kirkwall record by Spence (1907). The homogeneity of the Kirkwall record was assessed using comparisons with Stornoway, Tørshavn and CET, the latter being necessary as the Edinburgh record required adjustments for similar time periods. Three periods required adjustment ( , and ). A second adjustment was necessary to add the Kirkwall to the Lerwick record for The adjustments are given in Table A.I for T. For DTR, we use the complete SNIFFER Lerwick data, which extends back to 1871 and was updated to 2002 using BADC data. Again, digitizing the original daily records from Kirkwall and Lerwick would be an extremely useful exercise.