Developing a Historical Precipitation Record - Satellite-based Applications on Climate Change - page 98

satellite period, all of the reconstructions give similar multi-decadal variations and
indicate an increase in oceanic precipitation with time. The AR4 ensemble indicates
an increase with time over the twentieth century, which reflects the theoretical
response of precipitation to increased greenhouse gas and aerosol forcing. With a
warming troposphere, theory states that there will be more evaporation and higher
rainfall (e.g., Allan and Soden 2008 ). For the first half of the twentieth century, the
REOF(blend) shows an opposite tendency, suggesting that the gauge network may
not be adequate to specify the oceanic multi-decadal signal for much of the
twentieth century.
The RCCA indicates a general warming tendency throughout the period, but it is
not as steady as the AR4 ensemble model tendency. One major difference is the
influence of interannual variations in the RCCA. Interannual variations occur in the
individual AR4 models, but they are not phase locked and so are averaged out of
the ensemble. Another difference is the positive shift in the RCCA in the mid 1970s.
That shift is associated with a climate shift reflected in the Pacific SSTs (Trenberth
1990 ; Zhang et al. 1997 ). Evaluation of individual RCCA modes indicates that
the 1970s shift is mostly due to an ENSO-like mode which responds to the ENSO-
like shift in SSTs (Smith et al. 2009b ). The AR4 models do not consistently resolve
the 1970s climate shift, although models are capable of demonstrating such shifts.
For example, a model was used to show that the 1970s shift is likely caused by a
combination of external forcing and internal Pacific multi-decadal variability,
which influenced the timing of the shift (Meehl et al. 2009 ).
7.5 Merged Reconstruction
Our analyses showed that the REOF(blend) is capable of describing seasonal to
interannual variations, but its representation of oceanic multi-decadal signals
appears to be less reliable. The RCCA yields accurate multi-decadal signals
where it can be validated with independent data, but since it is an analysis of annual
averages, its representation of interannual and shorter-period variations is damped.
The REOF(blend) and RCCA standard deviations have consistent differences over
the analysis period, with the REOF(blend) systematically higher (Fig. 7.3 ). There is
a slight negative trend in both standard deviations, but that is much less than the
interannual changes over the period. In addition, although the RCCA multi-decadal
representation of land variations is strong, the availability of local gauges makes the
REOF(blend) representation of land variations superior. In order to take advantage
of the best qualities of each analysis, we merged the REOF(blend) and the RCCA.
Details of the merging are given elsewhere (Smith et al. 2010 ) and outlined here.
To merge the analyses, both are first filtered using the same weighted 7-year
filter to separate a low- from a high-frequency component of each. In ocean areas,
the low-frequency component of the REOF(blend) was replaced with the low-
frequency component of the RCCA. In 5 areas that are partly ocean, the adjustment
is proportional to the fraction of ocean area. In areas that are all land, no adjustment