A key issue is when, if ever, an indisputable signal (``fingerprint'') of greenhouse warming will be visible among the noise of natural climate variability. To assess this, direct measures of climate variability are needed. The observed records of global mean surface temperature and precipitation, considered reliable back to the late 1850s [ Wigley, 1989], are characterized by oscillations on a variety of timescales. Diaz and Bradley [in press] caution that the spatial scale and region selected for time series analysis influences results. An analysis of the temperature record by Dettinger and Ghil [1992] revealed oscillations of 1.7 years to 25 years in length over the continental U.S., with cycle length dependent on region and spatial empirical orthogonal function (method of calculating the time series).
With respect to climate change detection, Karl et al.
[1991] argued that while temperature has increased and
precipitation has decreased over the central U.S., in agreement
with greenhouse projections, natural variability is so great that
it will be 20 years or more before a signal will be apparent.
Ghil and Vautard [1991] used singular spectrum analysis to
determine if a significant warming trend could be distinguished
from the natural variability in 135 years of observed data.
Their analysis showed that both interannual and decadal
oscillations were large enough to conceal a greenhouse signal for
one to two decades. Stouffer et al. [1994] used a coupled
ocean-atmosphere GCM to determine if a greenhouse signal should
be detectable in the observed record. Their results showed that
in a 1,000-year time series of global temperature, no temperature
change as large as the observed 0.5
C per century was
sustained for more than a few decades.