Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://star.arm.ac.uk/~ambn/376.ps Äàòà èçìåíåíèÿ: Thu Dec 13 20:03:34 2001 Äàòà èíäåêñèðîâàíèÿ: Mon Oct 1 20:17:19 2012 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: http www.astronomy.ru forum index.php topic 4644.0.html

Comparison of Sunshine Records
and Synoptic Cloud Observations:
A case study for Ireland
E. Palle 1;2 & C.J. Butler 1
1 Armagh Observatory, College Hill, BT61 9DG. Armagh, N. Ireland. Fax:+44
02837 527174; e-mail: epb@star.arm.ac.uk, cjb@star.arm.ac.uk
2 Big Bear Solar Observatory, 40386 North Shore Lane, Big Bear City, CA 92314-
9672, U.S.A. Fax:+1 909-866-4240; e-mail: epb@bbso.njit.edu
1 Abstract
We compare the trends and inter-annual variability of sunshine duration and synoptic
cloud data for three sites in Ireland. Our results suggest that, although observers are
usually consistent in their synoptic cloud estimates, they can show a signicant personal
bias which makes it diôcult, if not impossible, to extract meaningful information on
long term trends in synoptic cloud cover. A study of the frequency of totally clear skies,
however, suggests that this parameter may be less aected by bias and could indicate
meaningful trends in total cloud factor. It is concluded that sunshine measurements
are likely to provide the most reliable indicators of long term trends in cloudiness.
2 Introduction
Comparisons of synoptic cloud records and sunshine duration have been made in several
areas of the world. Angell (1990) and Angell et al. (1984) compared cloud and sunshine
records for the United States for the period 1950-1988, and found a decrease in sunshine
hours accompanied by an increase in cloud cover. A strong negative correlation between
1

the two datasets was reported, even though the magnitude of the increase in cloud
cover was greater than the decrease in sunshine records. This result was attributed to
an increase in thin cirrus clouds which were not thick enough to interrupt the sunshine
record but nonetheless were included in the cloud report. Changnon (1981) found
virtually the same results for the midwest United States during the period 1901-1977.
On the other hand, Raju and Kumar (1982) argued that the fraction of possible sunshine
will closely approximate the fraction of possible clear weather when averaged over long
periods in any given region, and they demonstrated how sunshine-derived cloud cover
in India can be considered a better measurement of actual cloud cover than visual
estimates of cloudiness.
In previous studies, it has been established that the annual total of bright sunshine
hours has been decreasing over Ireland as a whole. Stanhill (1998) analysed several
sunshine series for Ireland for the period 1954-1995, and noted that a statistically
signicant decrease had occurred at some of the stations. However, when he compared
the decreasing sunshine trend with synoptic cloud observations at Valentia during the
same period, he found no sign of increasing cloudiness. He concluded that the observed
decrease in sunshine hours was to be attributed to other mechanisms such as increasing
aerosol or pollution or changes in cloud types. Palle and Butler (2001), extended this
work over the period 1881-2000, reaching similar and statistically stronger results for
the sunshine trends. Synoptic cloud data for Armagh, over the reduced period 1951-
1998 available at that time, showed an increasing trend similar in magnitude to the
decreasing trend in sunshine. Moreover, they found sunshine records and temperature
to vary in the same direction, and attributed the sunshine decrease to a cloudiness
increase, possibly as a result of increasing evaporation rates in the North Atlantic
ocean.
It is important to establish the cause of the sunshine decrease in Ireland, since it
may have major implications for local and regional climate. As Ireland lies near the
Atlantic Ocean and westerly winds prevail for most of the year, trends in Ireland can be
relevant for larger areas like the N. Atlantic ocean and oceanic mid-latitude regions in
2

general (Palle and Butler, 2001). A cloudiness increase during the last century would
most probably introduce some additional feedback in the Earth's radiative budget and
climate, the strength and sign of which will depend on the altitude and latitudinal
distribution of the cloud change.
The purpose of this study is to establish whether or not the decrease in sunshine hours
can be attributed to an increase in cloudiness. As a test case, we have compared the
synoptic cloud data and the sunshine data for three stations in Ireland. In section (3),
the synoptic cloud and sunshine data are described, and in section (4) their seasonal
variability compared. In section 5, we have analysed the long-term trends in the two
datasets and, in section 6, assessed the similarities and dierences between the two.
In section 7, the clear sky frequency at Armagh is shown to be a better indicator of
cloudiness trend than the synoptic cloud cover. Our conclusions are presented in section
8.
3 Data
Sunshine records at four Irish stations have been obtained since 1881 by means of a
standard Campbell-Stokes recorder. A Campbell-Stokes recorder consists of a glass ball
that concentrates the sunlight into a thin beam which is focussed onto a card. If there
are no clouds in the line of sight between the recorder and the Sun, the concentrated
light burns the card and leaves a trace for the duration of the clear condition. Monthly
and annual total sunshine hours were available to us for four stations, namely: Armagh,
Valentia, Birr and Dublin (see Figure 1). Daily records are also available for Armagh
for the period 1951-1998 and are currently being extended back to 1881. The sunshine
factor is dened as the observed sunshine duration divided by the period during which
the Sun lies above 3 o altitude - that is the maximum possible sunshine hours for that
particular date. A more detailed description of the Irish sunshine records can be found
at Palle and Butler (2001).
3

Synoptic observations of the total cloud cover are taken in many meteorological stations
worldwide. The procedure is simple; at specied times of the day, the observer estimates
the number of oktas (1/8 th of the sky) covered by clouds. A totally overcast sky is
registered as 8 and a clear sky as 0 (Observers Handbook, 1982). Estimation of the
cloud coverage in oktas has been in use since 1949; measurements prior to this were
made in tenths of the sky (1/10). Similarly as in the case of oktas a totally overcast sky
is registered as 10 and a clear sky as 0. At some stations, the coverage by dierent cloud
types is also recorded. It is evident that, to a greater or lesser extent, such estimates
are observer dependent and therefore subject to personal bias.
The synoptic cloud data for Armagh Observatory goes back to 1884 and thanks to
a grant from the Heritage Lottery Fund (Grant number: RF-98-01507) this data has
recently been entered in digital form, together with many of the historical meteorological
records from the Observatory archives. The Armagh cloud data consists of four series of
total synoptic cloud measurements. Observations were made at 9:00 am GMT for the
period 1884-2000 and also at 9:00 pm GMT for the period 1884-1964. In both series,
clouds were measured in tenths for the period 1884-1948, and subsequently, in oktas.
Synoptic total cloud cover data for Valentia (1940-1999) and Birr (1956-1999) was
kindly provided by Met Eireaan (Dublin). Data are given in monthly means and mea-
surements have been made at 9:00 am GMT. Unfortunately observations of cloud type
were not available to us for any of the three sites.
4 Seasonal variability
For the Armagh station, daily measurements of both total synoptic cloud cover and
sunshine hours are available in digital format. This has allowed a straight forward
comparison to be made. In Figure 2 (a), we plot the seasonal variability of synoptic
cloud cover and the cloud factors (F c
) derived from the sunshine records for the period
1951-1988; where F c
= 1 F s
. The sunshine factor F s
is the total number of sunshine
4

hours divided by the total number of available sunshine hours. It is evident that there
is a qualitative similarity in the annual variability of synoptic cloud cover at 9:00 am
and the cloud factor derived from sunshine records, with a peak in cloudiness around
mid-summer and another in mid-winter.
However, the overall level in summer relative to winter is dierent for the synoptic cloud
measurements in the upper panels of Figure 2 to cloud factor derived from sunshine
in the lower panels. One possible explanation could be that the sunshine records were
more strongly aected by trees on the horizon in winter than in summer. Another
explanation would relate to the time of measurement: whereas sunshine records cover
the whole day light period, synoptic cloud observations are only made once a day (9:00
am). A seasonal eect resulting from under-recording of winter sunshine due to haze
and wetter and cooler conditions may also contribute by reducing the ability of the Sun
to burn a trace in the (cold and damp) card.
For Birr and Valentia the same procedure was applied to the data, but employing
monthly means (Figures 2 (b) and (c) respectively), with similar results as in the case
of Armagh. Because of the scarcity of trees on the western seaboard (where Valentia
is situated), changes in the local horizon are unlikely to be responsible for seasonal
dierencies between cloud factor derived from sunshine and synoptic cloud cover.
5 Long-term trends in Ireland
5.1 Sunshine Records
The most prominent feature of the data, for all four sites, is a gradual decline in the
total annual sunshine hours over much, if not all, of the 117 year period during which
records have been obtained. In Figure 3, we plot the total number of hours of bright
sunshine for the four Irish sites for the period 1881-1998. The eect is particularly
conspicuous at the most westerly site of Valentia Island/Cahirciveen, on the County
5

Kerry coast, where the number of sunshine hours has dropped by  20% since the end
of the last century. If we plot the seasonal averages, the gradual decrease is seen in all
stations in all seasons (not shown).
In Table 1, the slope of the decreasing trend and its statistical signicance is shown
for the annual and seasonal means. It is clear that the decreasing trend is present and
highly signicant for all of the stations except for Armagh, where the decrease is not
so statistically signicant due to the anomalously low sunshine at the beginning of the
records, possibly due to a change in the condition of the local horizon (e.g. trees). We
also observe that the trends are less signicant in winter and strongest in spring and
summer.
The four meteorological stations lie between approximately 52 o :0 and 54 o :3 North, and
6 o :3 and 10 o :3 West. Thus, although there is likely to be a strong degree of correlation
between the sunshine records of the four sites, regional dierences will occur. The
correlation coeôcients between the annual total sunshine hours at the four locations
are listed in Table 2. In Table 2, we also show the correlation with the sunshine data
for Kew Observatory near London, England over the years in common (1881-1964). A
high degree of correlation between the data for the dierent sites is evident though, as
expected, this becomes lower as the distance between the sites increases.
5.2 Synoptic Cloud Observations
All four cloud series for Armagh are shown in Figure 4 (a), with the two series measured
in tenths converted to oktas. There is strikingly good agreement between the am and
pm measurement trends, and no discontinuities at the changeover points(1948-1949)
from tenths to oktas. Clouds in Armagh seem to indicate a decreasing trend during the
rst 50 years of the record and an increasing trend since the 1930's, except for the last
decade or so.
The cloud data available to us for Valentia and Birr cover a much shorter time interval
6

and are also shown in Figure 4 (b). In Table 2, the correlation coeôcients between the
three am cloud series for the period in common, 1956-1999, are given. The agreement
between the Valentia and Birr series is quite strong, but unlike the Armagh data, neither
show a long-term trend in cloud cover.
As might have been expected from eye estimates, dierent observers do not always
record the same degree of cloud cover. In Figure 4 (c) we have again plotted the cloud
series for Armagh (9:00am only), for the period during which meta-data relating to the
observers identity is available. Since 1901 the names of the meteorological observer and
deputy observers have been retrieved from the Meteorological Oôce inspection reports
and the Observatory archives. It is often the case that when an observer leaves or
retires he is substituted by one or a succession of temporary observers. In Figure 4 (c),
we have also plotted the periods during which observers are known to have taken the
measurements for the whole of the calendar year. Unshaded periods are times when it
is uncertain if the preceding or leading observer was taking the measurements or, more
often, a period when several temporary observers were responsible.
It looks likely, from the gure, that the mean recorded cloud cover strongly depends
on the observer, and therefore we cannot rely on long-term trends indicated by synoptic
cloud cover. Thus the disagreement in the long-term trends between synoptic cloudiness
and sunshine can be explained, at least for the case of Armagh, through systematic
errors introduced by the observers. Particularly clear is the jump in the Armagh cloud
series in 1993, after a new observer started in December 1992. In Figure 5 (left top
panel), the drop in sunshine and synoptic cloud towards the end of the series seem to
agree, suggesting that the cloudiness drop may be real, however this is just an eect of
the data normalisation. The drop in the synoptic cloud series occurs in 1993 and the
cloudiness level remains lower than average for the whole duration of that observer's
employment until 1998. At the same time, in 1993, there is a dip in sunshine which
disagrees with the decrease in cloudiness (since we would expect them to vary in opposite
directions). Two years later, in 1995, there is a peak in sunshine with a corresponding
dip in cloudiness (respective to the mean level of that observer), as one would expect.
7

This is explained by the fact that introducing a new observer changes the mean level of
observed cloudiness, thus destroying the possibility of long-term trend studies, however
during the interval in which the same observer is taking the measurements, the year-to-
year variability matches well with the sunshine records. In other words, the observers
are consistent in their own measurements but dier systematically from each other.
The cloud series for the Valentia and Birr stations do not show such observer-dependent
discontinuities; or at least they are not evident at rst glance. However we do not have
access to the relevant data to check this in detail.
6 Comparison of Sunshine and Synoptic Cloud Trends.
Initially, only the monthly mean synoptic cloud cover for the period 1951-1988 was
available in digital form for Armagh. For cloud cover we nd a 15% increase over this
period as compared to a 13.5% decrease in sunshine. These trends were so similar that,
we concluded initially that, contrary to Stanhill's (1998) results, an increase in cloud
cover was the most likely cause of the drop in sunshine over Ireland since the late 19th
century (Palle and Butler 2001).
In Figure 5, synoptic cloud cover and sunshine records at the three Irish stations are
compared. Both series have been normalised to ease the comparison, and the sunshine
series has been inverted. In the left hand panels we can see how except for Armagh
the agreement between the cloud and sunshine series in their long-term trends is poor.
However, the inter-annual variability is well matched for all stations. In the right hand
panels, the sunshine records have been detrended with a chi-square linear t. Once
the linear decreasing trend in the sunshine is removed, the correspondence between the
records for Birr and Valentia improves (see Table 3). The agreement is particularly good
for the Valentia station, where the observations are taken by professional meteorological
observers.
We nd then, in general, that even though the inter-annual cloud and sunshine vari-
8

ability is strongly related, the trends in the two data series sometimes dier. A decrease
in the sunshine records not accompanied by an increase in cloud cover is diôcult to
explain. One possible reason could be an increase in atmospheric pollution or aerosol
concentration leading to changes in the transparency of the atmosphere or changes in
the radiative properties of clouds (e.g thickening of cirrus clouds). It is also possible
that observer changes can play a fundamental role and in uence long-term cloud cover
trends at a particular station, as we have seen in Armagh. This problem could be
overcome through averaging the data for a large number of stations.
In Figure 6, the high-frequency variability of the Armagh sunshine and synoptic cloud
series is shown. To derive this quantity we have smoothed the raw data for both series
with a ve year running mean, and then subtracted the running mean from the raw
data. In this way, the long-term variability in both series has been eliminated. The
resulting cloud and sunshine series correlate with each other, for the period 1884-1998,
with a Pearson's correlation coeôcient of -0.56 (P << 10 4 ). Such a good agreement
is striking and much more convincing than the removal of a simple linear trend as in
Figure 5. For the restricted period 1949-1998 used in Figure 5 (to compare with Valentia
and Birr), the correlation coeôcient is -0.49, slightly lower than for the whole series but
still reaching less than 99.9% probability of occurrence by chance. For Birr and Valentia
this procedure leads to correlation coeôcients of -0.6 and -0.79, respectively, both of
them signicant at higher condence levels than Armagh during the same period. This
method gives a very similar correlation coeôcient to the detrended sunshine for Valentia
(Figure 5), but results in a substantially improved correlation for Birr. Thus, the
mismatch between sunshine and synoptic cloud observations involves more than a simple
linear trend, which would tend to suggest that the long-term trends in the two series
are not always physically related.
Another possible factor aecting cloud trends is the resolution of the measurements.
Synoptic cloud cover is measured in integral numbers of oktas, i.e. a resolution of
just 12.5%. Although it is diôcult to translate percentage changes in cloud cover into
percentage changes of sunshine or vice-versa, the decreasing trends detected in the
9

sunshine records are around 14% for Valentia and around 11% for Armagh and Birr
during the last ve decades, very close indeed to the synoptic cloud resolution.
7 Clear Sky Frequency at Armagh
We have seen in the previous sections how synoptic cloud observations are subjectively
biased by the observer making it diôcult to obtain reliable information on long-term
trends. This systematic error cannot be corrected easily (e.g, by calibrating the means
for dierent observers) without destroying any long-term trends. Here however, we
suggest another method to obtain reliable information on long-term trends in cloudiness
- namely to use the clear-sky frequency.
One thing that all observers should agree on, independently of their subjective bias,
is a sky completely free of clouds. The Observer's Handbook (1982) is very clear in
this respect, synoptic cloud is recorded as 0 (clear sky) if there is absolutely no cloud
visible. If a small cloud is present, even if it extends much less than one eighth of the
sky, synoptic cloud has to be recorded as 1. With this in mind, we analysed the clear
sky frequency at Armagh.
It must be noted that, whereas sunshine and cloudiness trends would be expected to
vary oppositely, the clear sky frequency (absence of clouds) is expected to vary in the
same way as sunshine. In Figure 7, the number of days per year in which the sky
appeared clear at 9:00 am (a) and 9:00 pm (b) are plotted. We can see here how the
number of clear mornings (9:00 am) has been monotonically decreasing during the last
century, after an initial sharp increase between 1880 and 1900. Although the possibility
of a systematic observer bias in the data cannot be excluded, the variability of this
series is not the same as the total synoptic cloud and there are no apparent jumps in
the series at the times when the observers changed. The am and pm series agree in their
year to year variability, although the decreasing trend is only evident in the am series.
Henderson-Sellers (1992) reported that synoptic cloud estimations at 8:00 or 9:00 am
10

local time give the best estimate of the daily average.
In Figure 7 (c) we have plotted together the sunshine records (scaled but not inverted)
and the clear sky frequency. Although these two measurements are not equivalent,
the correlation between the two series is about 0.4, signicant at least at the 99.9%
level. Overall this picture suggests that cloudiness has in fact increased during the past
century over Armagh, in agreement with the sunshine records. Though, it is diôcult
to evaluate the implications of the change in clear sky frequency and its signicance for
climate, the downward trend is striking.
As well as the frequency of clear skies, our rst thought was also to use the frequency
of a completely overcast sky. However, analysis of the frequency of an overcast sky
reveals the same discontinuities in the series as seen in the total synoptic cloud cover,
coinciding with changes in the observer (see Figure 4 (c)). A possible explanation is
that the Observer's Handbook (1982) is not so clear about this point. If there is a
very small patch of the sky (far less than one okta) the observation can be qualied
as 7 or 8 depending on the observer, although strictly it should be recorded as 7. The
diôculty in recording an overcast sky resides in determining whether there is a clear
patch in a sky full of clouds. The Observer's Handbook (1982) states that \if even
the smallest amount of blue sky is visible then you must report a 7.", but also states
that \Sky during the day has a blue coloration. However, during conditions of haze
or mist, the sky towards the horizon may become quite discoloured. Observers should
beware of mistaking such phenomena with cloud". Thus, the decision as to whether the
observer sees a small amount of `blue sky' or not can be subjective. Angell et al. (1984)
also signaled a tendency for ground-based observers to overestimate cloudiness because
of their inability to detect clear patches at a distance when clouds have a signicant
vertical extent. It is also worth noticing how a clear sky is a relatively \rare" event at
Armagh, going from around 30 days/year at the beginning of the century to around 10
days/year in the last couple of decades, and it is also easy to evaluate. On the other
hand, an overcast sky is a quite common observation, around 150-200 days/year (almost
2/3 of the days in some years), which would lead to a much stronger eect (through
11

accumulation) of the bias introduced by observers.
A nal comment is that the decreasing trends in the sunshine records agree with the
global trend in increasing cloudiness detected in studies all around the world (Norris,
1999; Sun and Groisman, 2000; Henderson-Sellers, 1992). Since those major studies
involve thousands of stations the systematic errors introduced by the observers should
be much reduced or eliminated.
8 Conclusions
A gradual decrease in the number of sunshine hours over Ireland is observed since
records began in 1881. To assess whether or not this is due to increased cloudiness we
have analysed the synoptic cloud and sunshine data from three stations.
The synoptic cloud records have a strong inter-annual correspondence with sunshine
records, but they fail to show any consistent trend during the last 50 years. However
we have demonstrated how trends in synoptic cloud cover have little weight due to
their observer dependence. We conclude therefore that long-term trends in synoptic
cloud records are not reliable for a single station, particularly for stations with non-
professional observers. The data from Valentia could be an exception to this rule,
however, the synoptic cloud cover at 9:00 am at Valentia has no signicant trend over the
past 50 years. The use of sunshine records in preference to synoptic cloud observations is
recommended in view of the agreement between sites and because they are instrumental.
Trends derived from synoptic cloud cover observations from a small number of stations
should be conrmed if possible with the use of sunshine recorders.
Finally, we propose that the clear sky frequency is a good indicator of cloud trends
and is less susceptible to observer bias than synoptic cloud cover. Clear sky frequency
at Armagh, the only station for which daily data was available, show a very signicant,
robust, decreasing trend during the last century. These results seem to support the
suggestion that increasing cloudiness is responsible for the observed decrease in sunshine
12

hours over Ireland.
9 References
Angell, J.K., 1990. Variations in United States cloudiness and sunshine duration be-
tween 1950 and the drought year of 1988. Journal of Climate 3, 296-308.
Angell, J.K., Korshover, J., Cotton, G.F., 1984. Variation in United States cloudiness
and sunshine, 1950-82. Journal of Climate and Applied Meteorology 23, 752-761.
Changnon, S.A., 1981. Midwestern cloud, sunshine and temperature trends since 1901:
possible evidence of Jet Contrail eects. Journal of Applied Meteorology 20, 496-508.
Henderson-Sellers, A., 1992. Continental cloudiness changes this century. Geo-Journal.
27.3, 255-262.
Norris, J.R., 1999. On trends and possible artifacts in global ocean cloud cover between
1952 and 1995. Journal of Climate. 12, 1864-1870.
Observers Handbook, 4th Edition, 1982. UK Meteorological Oôce (Eds). Her Majesty's
Stationery Oôce, London, p 153.
Palle, E., Butler, C.J., 2001. Sunshine records from Ireland, cloud factors and possible
links to solar activity and cosmic rays. International Journal of Climatology 21, 709-
729.
Raju, A.S.N., Kumar, K.K., 1982. Comparison of point cloudiness and sunshine de-
rived cloud cover in India. Pure and Applied Geophysics (PAGEOPH) 120, 495-502.
Stanhill, G., 1998. Long-term trends in, and spatial variation of, solar irradiance in
Ireland. International Journal of Climatology 18, 1015-1030.
Sun, B., Groisman, P.Y., 2000. Cloudiness variations over the Former Soviet Union.
International Journal of Climatology 20, 1097-1111.
13

Table 1: Seasonal and annual values for the trend in sunshine at the four Irish stations.
The rate of the trend is given, as well as the determination coeôcient for the trend
t and its statistical signicance (ns: not signicant). Also we give the seasonal and
annual trends for the synoptic clear sky frequency at Armagh (both 9:00am and 9:00pm
series).
Sunshine Period Decr. [hours/year] r 2 P value
Armagh Annual -0.78 0.057 98%
Winter -0.03 0.002 ns
Spring -0.33 0.045 95%
Summer -0.28 0.021 ns
Autumn -0.14 0.021 ns
Valentia Annual -2.50 0.371 99.9%
Winter -0.25 0.072 99%
Spring -0.81 0.220 99.9%
Summer -0.82 0.151 99.9%
Autumn -0.63 0.227 99.9%
Dublin Annual -2.45 0.373 99.9%
Winter -0.20 0.058 99%
Spring -0.95 0.282 99.9%
Summer -0.90 0.182 99.9%
Autumn -0.39 0.118 99.9%
Birr Annual -1.74 0.219 99.9%
Winter -0.17 0.046 99%
Spring -0.54 0.116 99.9%
Summer -0.60 0.088 99.9%
Autumn -0.43 0.132 99.9%
Clear Sky Freq. Period Decr. [days/decade] r 2 P value
9:00 am Annual -2.55 0.486 99.9%
Winter -0.77 0.395 99.9%
Spring -0.61 0.225 99.9%
Summer -0.31 0.099 99.9%
Autumn -0.85 0.430 99.9%
9:00 pm Annual 1.45 0.047 ns
Winter 1.02 0.184 99.9%
Spring 0.56 0.042 ns
Summer -0.72 0.103 99.5%
Autumn 0.59 0.056 95%
14

Table 2: The correlation coeôcient between the mean annual sunshine hours over the
four Irish sites and Kew (England). All coeôcients are signicant at the 99.9% level or
higher (less than 0.001 probability of occurrence by chance). The distance (D) between
the sites in kilometers is also indicated. Correlations Coeôcients between the 9:00 am
cloud data series for three Irish sites for the common period 1956-1999 (44 years) are
also indicated. The correlation between Armagh site and either of the other two sites is
not signicant. Only the correlation between Valentia and Birr (r=0.39) is signicant
at the 0.01 level.
Site Armagh Valentia Birr Dublin
Correl D Correl D Correl D Correl D
Armagh Sunshine { { 0.55 367 0.74 175 0.69 116
Cloud { 0.09 0.14 {
Valentia Sunshine 0.55 367 { { 0.71 212 0.76 317
Cloud 0.09 { 0.39 {
Birr Sunshine 0.74 175 0.71 212 { { 0.76 113
Cloud 0.14 0.39 { {
Dublin Sunshine 0.69 116 0.76 317 0.76 113 { {
Kew Sunshine 0.55 612 0.34 750 0.56 600 0.52 510
15

Table 3: Correlation coeôcients and signicance level between sunshine records and
synoptic total cloud cover at three Irish stations. Two case have been considered, using
raw data with no corrections and removing a chi-square linear t trend to the sunshine
records. Also indicated is the period for which both datasets are available at each site.
Site Raw Data Detrended Sunshine High-Frequency
Years r P r P r P
Armagh 1949-1998 -0.55 <<99.9% -0.49 <99.9% -0.56 <<99.9%
Valentia 1940-1998 -0.64 <<99.9% -0.81 <<99.9% -0.8 <<99.9%
Birr 1956-1998 -0.14 ns -0.42 99.5% -0.6 <<99.9%
16

Figure 1: Outline map of Ireland. The geographical locations of the four Irish stations
(A:Armagh; D:Dublin; B:Birr; V:Valentia) are shown.
17

(a) Armagh
(b) Birr
(c) Valentia
Figure 2: The average seasonal cloud variability at (a) Armagh for the period 1951-
1988, (b) Birr and (c) Valentia. Top panels: seasonal variability from synoptic cloud
observations at 9:00 am GMT. Lower panels: seasonal variability of the cloud factor
(1-Sunshine factor) from the sunshine records. Note that daily data have been used for
Armagh (a) and monthly data for Birr (b) and Valentia (c).
18

Figure 3: Total annual sunshine hours for four Irish sites 1881-1998.
19

(a) (b)
(c)
Figure 4: (a) Synoptic cloud observations at Armagh Observatory. The thick solid line
represents the observations taken at 9:00am GMT and the thin broken line observations
taken at 9:00pm GMT. Data for the period 1884-1948 were measured in tenths of the
sky and have been scaled by the factor 0.8. The vertical line at 1949 marks the change
from tenths to oktas. (b) Synoptic total cloud observations at Valentia (solid line)
and Birr (dotted line) stations. Observations are taken at 9am GMT.(c) Total synoptic
cloud observations at 9:00am GMT for Armagh (solid line, as in panel (a)) and overcast
sky frequency at 9:00am at Armagh (broken line). Shaded areas are periods for which
a single observer is believed to have made the observations. Note how discontinuities
in both series tend to coincide with the start or end of a shaded period.
20

Figure 5: Total synoptic cloud cover (solid line) and sunshine (dotted line) observations
at Armagh (top), Valentia (middle) and Birr (bottom) sites. Both series have been
normalised to allow a more straight forward comparison and the sunshine series have
been inverted. In the left panels both datasets are as measured. In the right panels the
sunshine data is detrended.
21

Figure 6: High frequency variability at Armagh for the sunshine records (dotted line)
and synoptic cloud cover (solid line). Both sets of data are normalised and sunshine
is inverted. The raw data for both series have been smoothed with a 5-year running
mean and the smooth values subtracted from the raw data.
22

(a) am (b) pm
(c) am-sunshine (d) all
Figure 7: Top panels: Clear sky frequency at Armagh for the 9:00am series (a) and
9:00pm series (b). In panel (c) the 9:00am clear sky frequency (solid line) is plotted
together with the sunshine records at Armagh (dotted line) arbitrarily scaled for com-
parison. The correlation between the two series is of 0.4 (P << 0:001). In panel (d) the
clear sky frequency at 9:00am (solid line), clear sky frequency at 9:00pm (dashed line)
and sunshine records (dotted line) have been smoothed with a 3-year running means to
allow an easier comparison of the long-term variability of the three series.
23