Section 4.5: Concl usions
75
since 1930 in the Southern Hemisphere, while spatially some regions have
warmed by nearly 2°C while a few areas indicate cooling (Jones and Briffa,
1992). Analyses with maximum and minimum temperatures over 1952-89 indicate that three quarters of the rise has occurred during nighttime compared
to daytime (Karl et al. 1993).
The interrelationships of the climate system mean that changes in surface
variables will be accompanied by changes in the atmospheric circulation.
Additional checks on the homogeneity of any of the surface data sets and
associated time series can be made by finding synchronous and connected
changes in the other two.
A spatial pattern correlation statistic, R(t), has been used, in order to
attempt to detect the GCM-based climate change signal due to increasing
greenhouse gas concentrations in the observed temperature data. The statistic is related to the C(t) statistic defined earlier by Barnett and Schlesinger
(1987). Uncentered statistics like C(t) are inappropriate for detection because
they do not separate the effect of the changing me an from the true pattern
similarity. C(t) is, in effect, equivalent to the spatial-mean time series (i.e.,
global-mean temperature for the data considered here).
The detection exercise using R(t) to measure the pattern similarity between
the five model signals and the observed temperatures during the present
century failed. There was no evidence of an increasing expression of the
signal in the observed data. Possible methods of optimizing detection by
concentrating on regions of high signal-to-noise ratios are discussed by Sant"er
et al. (1993) but appear to have limited value due to a number of statistical
problems.
Of the set of possible spatial similarity fingerprint detection methods, it
is the most obvious and basic of these (pattern correlations using the raw
data) that appears to be the most useful. Alternatives, using non-centered
moments and/or normalizing to highlight certain (high SNR) regions, have
drawbacks that make them oflesser use. Results obtained here with the direct
method (i.e., using R(t)) are negative, indicating that the spatial details of
the greenhouse signal are still within the noise, or that the signal itself is
poorly defined.
75
since 1930 in the Southern Hemisphere, while spatially some regions have
warmed by nearly 2°C while a few areas indicate cooling (Jones and Briffa,
1992). Analyses with maximum and minimum temperatures over 1952-89 indicate that three quarters of the rise has occurred during nighttime compared
to daytime (Karl et al. 1993).
The interrelationships of the climate system mean that changes in surface
variables will be accompanied by changes in the atmospheric circulation.
Additional checks on the homogeneity of any of the surface data sets and
associated time series can be made by finding synchronous and connected
changes in the other two.
A spatial pattern correlation statistic, R(t), has been used, in order to
attempt to detect the GCM-based climate change signal due to increasing
greenhouse gas concentrations in the observed temperature data. The statistic is related to the C(t) statistic defined earlier by Barnett and Schlesinger
(1987). Uncentered statistics like C(t) are inappropriate for detection because
they do not separate the effect of the changing me an from the true pattern
similarity. C(t) is, in effect, equivalent to the spatial-mean time series (i.e.,
global-mean temperature for the data considered here).
The detection exercise using R(t) to measure the pattern similarity between
the five model signals and the observed temperatures during the present
century failed. There was no evidence of an increasing expression of the
signal in the observed data. Possible methods of optimizing detection by
concentrating on regions of high signal-to-noise ratios are discussed by Sant"er
et al. (1993) but appear to have limited value due to a number of statistical
problems.
Of the set of possible spatial similarity fingerprint detection methods, it
is the most obvious and basic of these (pattern correlations using the raw
data) that appears to be the most useful. Alternatives, using non-centered
moments and/or normalizing to highlight certain (high SNR) regions, have
drawbacks that make them oflesser use. Results obtained here with the direct
method (i.e., using R(t)) are negative, indicating that the spatial details of
the greenhouse signal are still within the noise, or that the signal itself is
poorly defined.