Section 4.4: The Greenhouse Detection Problem
67
The basic detection strategy is to compare the observed time series of this
vector either with an estimate of the equilibrium greenhouse-gas signal (as
may be inferred by differencing the results of equilibrium model experiments
for 1xC02 and 2xC02), or with a time-evolving signal (inferred from a
model simulation with time-dependent greenhouse-gas forcing).
4.4.1
Definition of Detection Vector and Data Used
Before performing any analysis, some consideration should be given to the
choice of the detection vector. There are at least four criteria which can be
used to evaluate the usefulness of a given vector for detection studies.
• First, the individual components should have high signal-to-noise ratios
(SNR). Results from climate model experiments indicate that temperature and atmospheric moisture content are relatively good choices in this
regard, while preeipitation and sea level pressure have low SNR (Barnett
and Schlesinger, 1987; Santer et al., 1991).
• Second, the signal vector should not be model-speeific. If different models gave different signal vectors, then this would lower confidence in the
signal from any one model.
• Third, the signal vector should be easily distinguished from (i.e., near
orthogonal to) both the signals due to other forcing factors and the noise
of natural internal variability on the 10-100 year time scale relevant to
the enhanced greenhouse effect (Barnett and Schlesinger, 1987; Wigley
and Barnett, 1990). A poor choice in this regard might be the tropospherejlower stratosphere temperature change contrast (but see also
Karoly et al., 1994), where the expected greenhouse signal is apparently
similar to observed natural variability (Liu and Schuurmanns, 1990).
The spatial pattern of near-surface temperature changes, which is used
here, is also less than ideal, since the feedback processes that influence
the spatial character of the enhanced greenhouse signal must also operate with other extern al foreing mechanisms, such as changes in solar
irradiance (Wigley and Jones, 1981).
• Fourth, suitable observation al data must exist. Because it is the 10100 year time scale that is of concern, long data re cords are needed.
Only surface-based data have records exceeding ",40 years in length,
and of these, only temperature data satisfy the high signal-to-noise ratio
condition.
Part of the reason why detection is so difficult is because the intersection
of the sets satisfying the above criteria is effectively empty. The observed
data we use satisfy these criteria better than any other data set, viz. 50 X 50
gridded monthly-mean, land-based surface air temperatures and sea surface
67
The basic detection strategy is to compare the observed time series of this
vector either with an estimate of the equilibrium greenhouse-gas signal (as
may be inferred by differencing the results of equilibrium model experiments
for 1xC02 and 2xC02), or with a time-evolving signal (inferred from a
model simulation with time-dependent greenhouse-gas forcing).
4.4.1
Definition of Detection Vector and Data Used
Before performing any analysis, some consideration should be given to the
choice of the detection vector. There are at least four criteria which can be
used to evaluate the usefulness of a given vector for detection studies.
• First, the individual components should have high signal-to-noise ratios
(SNR). Results from climate model experiments indicate that temperature and atmospheric moisture content are relatively good choices in this
regard, while preeipitation and sea level pressure have low SNR (Barnett
and Schlesinger, 1987; Santer et al., 1991).
• Second, the signal vector should not be model-speeific. If different models gave different signal vectors, then this would lower confidence in the
signal from any one model.
• Third, the signal vector should be easily distinguished from (i.e., near
orthogonal to) both the signals due to other forcing factors and the noise
of natural internal variability on the 10-100 year time scale relevant to
the enhanced greenhouse effect (Barnett and Schlesinger, 1987; Wigley
and Barnett, 1990). A poor choice in this regard might be the tropospherejlower stratosphere temperature change contrast (but see also
Karoly et al., 1994), where the expected greenhouse signal is apparently
similar to observed natural variability (Liu and Schuurmanns, 1990).
The spatial pattern of near-surface temperature changes, which is used
here, is also less than ideal, since the feedback processes that influence
the spatial character of the enhanced greenhouse signal must also operate with other extern al foreing mechanisms, such as changes in solar
irradiance (Wigley and Jones, 1981).
• Fourth, suitable observation al data must exist. Because it is the 10100 year time scale that is of concern, long data re cords are needed.
Only surface-based data have records exceeding ",40 years in length,
and of these, only temperature data satisfy the high signal-to-noise ratio
condition.
Part of the reason why detection is so difficult is because the intersection
of the sets satisfying the above criteria is effectively empty. The observed
data we use satisfy these criteria better than any other data set, viz. 50 X 50
gridded monthly-mean, land-based surface air temperatures and sea surface