18
the 1920's and '30's, cooling in the 50's and '60's and warming from 197690 is nonetheless quite compelling, even in this unfiltered version of the
record. For example it is evident, even to the naked eye, that a large fraction of the monthly mean temperatures observed during the decade of the
1980's were above the normals for their respective calendar months based
on averages over the previous three decades. If the data were smoothed,
say with a one year running mean filter, the interdecadal to century trend
would be no less apparent, but visual information on the 'signal to noise
ratio', as manifested in the amplitude of the long term changes relative to
the width of the envelope within which the individual data points tend to
be clustered, would be substantially degraded. Formal statistical measures
of the signal to noise ratio could, of course, be calculated, but it is not clear
that they would be any more discerning or informative than the subjective
impression that emerges from the image processing that goes on inside the
human brain when it is provided with the raw data in the analogue format
of Fig. 2 of chapter 2.
The warm season (May-October) data points, indicated by the darker
dots in the top panel of Fig. 16 of chapter 2 , exhibit substantially less
month-to-month variability than their cold season counterparts, rendering
the interdecadal to century scale "signal" more prominent. A distinction is
also apparent in the corresponding 5-year running mean time series shown
in the top panel of Fig. 17 of chapter 2. The cold season curve exhibits
a richer spectrum of variability in the range of periods from ten years,
which corresponds to the high-frequency cutoff of the filter, up to several
decades, whereas the variability in the warm season time series is more
concentrated at the interdecadal-to-century time scale. In view of the
much larger month-to-month variability in the cold season time series, it
is conceivable that much of the excess variability at periods ranging from
10 to 40 years could merely be sampling fluctuations, as discussed in the
previous section. If one only had access to the smoothed data in Fig. 17 of
chapter 2, one might be inclined to accept the differences between the warm
and cold season time series at face value, without questioning whether they
have any physical significance. It will be shown in the next chapter that
most of the excess variability of hemispheric mean temperature during the
cold season is associated with a particular spatial pattern of hemispheric
circulation anomalies, whose month-to-month variations are almost, but
not quite random.
There is no guarantee that the noteworthy interdecadal-to-century scale
the 1920's and '30's, cooling in the 50's and '60's and warming from 197690 is nonetheless quite compelling, even in this unfiltered version of the
record. For example it is evident, even to the naked eye, that a large fraction of the monthly mean temperatures observed during the decade of the
1980's were above the normals for their respective calendar months based
on averages over the previous three decades. If the data were smoothed,
say with a one year running mean filter, the interdecadal to century trend
would be no less apparent, but visual information on the 'signal to noise
ratio', as manifested in the amplitude of the long term changes relative to
the width of the envelope within which the individual data points tend to
be clustered, would be substantially degraded. Formal statistical measures
of the signal to noise ratio could, of course, be calculated, but it is not clear
that they would be any more discerning or informative than the subjective
impression that emerges from the image processing that goes on inside the
human brain when it is provided with the raw data in the analogue format
of Fig. 2 of chapter 2.
The warm season (May-October) data points, indicated by the darker
dots in the top panel of Fig. 16 of chapter 2 , exhibit substantially less
month-to-month variability than their cold season counterparts, rendering
the interdecadal to century scale "signal" more prominent. A distinction is
also apparent in the corresponding 5-year running mean time series shown
in the top panel of Fig. 17 of chapter 2. The cold season curve exhibits
a richer spectrum of variability in the range of periods from ten years,
which corresponds to the high-frequency cutoff of the filter, up to several
decades, whereas the variability in the warm season time series is more
concentrated at the interdecadal-to-century time scale. In view of the
much larger month-to-month variability in the cold season time series, it
is conceivable that much of the excess variability at periods ranging from
10 to 40 years could merely be sampling fluctuations, as discussed in the
previous section. If one only had access to the smoothed data in Fig. 17 of
chapter 2, one might be inclined to accept the differences between the warm
and cold season time series at face value, without questioning whether they
have any physical significance. It will be shown in the next chapter that
most of the excess variability of hemispheric mean temperature during the
cold season is associated with a particular spatial pattern of hemispheric
circulation anomalies, whose month-to-month variations are almost, but
not quite random.
There is no guarantee that the noteworthy interdecadal-to-century scale