[Previous Article] [Next Article]
Linear Regression Forecast of
Central England Temperature for July-August 1997
contributed by Andrew Colman and Michael Davey
Hadley Centre for Climate Prediction and Research
UK Meteorological Office, Bracknell, United Kingdom
The UK has had some notable exceptionally warm summers in recent years, prompting the
question: Are seasonal summer temperatures in this region predictable? At UKMO, empirical
techniques using global scale patterns of historical sea surface temperature anomalies (SSTA)
have been used successfully to make experimental predictions of seasonal rainfall in NE Brazil,
tropical northwestern Africa and eastern Africa (Ward et al. 1993; see also recent March and June
issues of this Long-Lead Bulletin). The problem of predicting UK climate is more difficult than
that of predicting rainfall in these tropical regions. However, there is some evidence that UK
climate may be predictable from SST (e.g. Ratcliffe and Murray 1970, Palmer and Sun 1985).
This year an experimental prediction of July-August average Central England Temperature (CET)
has been made at UKMO using SSTA data. CET is a homogenized temperature series from 1659
to the present, first put together by Manley (1974) and updated by Parker et al. (1992). From
1877 on, CET is based on daily maximum and minimum temperatures at selected stations in the
region. July and August are usually the warmest months in the UK. Weather conditions also tend
to be less variable in July and August than in other months, and persistence in CET between these
months is high. Correlating July and August monthly average CET for the period 1896-1995
results in a temporal anomaly correlation of 0.53, compared with 0.17 for June vs. July, and 0.35
for August vs. September.
THE SST PREDICTOR
By correlating July-August CET with monthly and bi-monthly 10x10 latitude-longitude area
averages of SST, significant correlations were found between January-February SST anomalies in
the extratropical north Atlantic and July-August CET. An eigenvector analysis of North Atlantic
winter (January-February) SSTA was also carried out. The pattern (Fig. 1) which explained the
most variability in the SSTA data over the period 1901-1990 is similar to the correlation pattern
between SST and CET. The time series of this eigenvector, which indicates the strength and sign
of the eigenvector pattern, was calculated for each winter (January-February average SSTA), and
used to develop a simple linear regression prediction scheme for CET.
PREDICTION SKILL ESTIMATE
The temporal anomaly correlation between the predictor time series and July-August CET over
the period 1871-1995 is 0.40, which is significant (different from zero) at the 99.9% level. Over
the shorter period 1971-1995 the correlation is 0.51 (99% significant).
Inflated regression predictions have been made for 1956-1996, using a jack-knife technique. The
term "jack-knife" refers to forecasts being made for each year using a regression equation
obtained from a dataset that excludes the year being forecast. The subsequent two years are also
excluded, as these may be related to the forecast year through persistence. The term "inflation"
indicates that a scaling factor is applied to match the variance of the predictions to that of the
observed predictand. This rescaling does not affect the anomaly correlation, which is 0.56 for
1956-1996. This level of skill is promising, considering the 6-month lead time, and compares well
with other long lead extratropical statistical prediction skills (e.g. Barnston 1994).
Fig. 2 shows these jack-knife forecasts plotted against observed CET. The forecasts are plotted as
a temperature range, with the range chosen such that, in the period 1956-1996 shown, 50% of the
observations fall within the indicated range.
In a simple "climatological" forecast strategy, the predicted range for each year is centered on
some average temperature. When this strategy is applied for 1956-1996, using the median
temperature for those years, the range required to cover 50% of the observations is very close to
that indicated in Fig. 2 for the jack-knife regression forecasts. By this "50% range" measure, the
jack-knife regression method and the climatological method have the same skill. However, by the
anomaly correlation measure of skill the jack-knife regression method is superior.
FORECAST FOR 1997
An experimental forecast has been made using the regression scheme described above. The
statistical model prediction is that there is a 50% probability that the July-August Central
England Temperature will be within the range 14.4 to 16.1C in 1997. By contrast, a
"climatological" prediction suggests a 50% probability that CET will be within the range 15.1 to
16.8C. Thus, the forecast values are slightly lower than the climatological values. However,
the difference is small compared with expected variations.
NOTE: This forecast for 1997 is experimental, with substantial uncertainty in accordance with the
weak nature of the statistical link between SST and CET. The method should be regarded as
develop-mental, and practical use is not recommended.
Barnston, A.G., 1994: Linear statistical short-term climate predictive skill in the northern
hemisphere. J. Climate, 7, 1513-1564.
Manley, G., 1974: Central England temperatures, monthly means 1659 to 1973. Q. J. Roy. Met.
Soc., 79, 242-261.
Palmer, T. and Z. Sun, 1985: A modeling and observational study of the relationship between sea
surface temperature in the north west Atlantic and atmospheric general circulation. Q. J. Roy.
Met. Soc., 111, 947-975.
Parker, D.E., T.P. Legg and C.K. Folland, 1992: A new daily Central England Temperature
series, 1772-1991. Int. J. Climatol., 12, 317-342.
Ratcliffe, R.A.S. and R. Murray, 1970: New lag associations between north Atlantic sea
temperatures and European pressure, applied to long-range weather forecasting. Q. J .Roy. Met.
Soc., 96, 226-246.
Ward, M.N., C.K. Folland, K. Maskell, A.W. Colman, D.P. Rowell and K.P. Lane, 1993:
Experimental seasonal forecasting of tropical rainfall at the UK Meteorological Office.
In:Prediction of interannual climate variations, NATO ASI Vol. 16, pp 197-216.
Fig. 1: The predictor SST leading eigenvector pattern, based on 1901-1990 January-February sea
surface temperature anomalies in the North Atlantic. Observed SST anomalies are projected onto
this pattern to obtain a predictor value for the corresponding year. Positive values of this pattern
are associated with warmer Central England temperature (CET).
Fig. 2: July-August Central England Temperature, as observed (diamonds) and as forecast
(vertical bars) using linear regression and January-February North Atlantic sea surface
temperature anomalies, for 1956-96. The indicated forecast range is chosen such that 50% of the
observations fall within that range.