Developing a Historical Precipitation Record - Satellite-based Applications on Climate Change - page 95

for possible use as base data, including GPCP and CMAP (CPC Merged Analysis of
Precipitation, Xie and Arkin 1997 ). We found that both gave consistent
representations of interannual variations. However, the CMAP had a decreasing
multi-decadal tendency, opposite the tendency of GPCP and other satellite
estimates such as Wentz et al. ( 2007 ). That opposite CMAP tendency appears to
be caused by how satellite data are adjusted to reduce their satellite-to-satellite
biases. In CMAP large-scale tropical adjustments are performed against a few
island stations. Those few stations reasonably represent the interannual variations
that CMAP was developed to evaluate, but they may not be representative of multi-
decadal variations. By comparison, GPCP was developed for climate studies on a
range of time scales, including decadal. Therefore, the GPCP satellite-to-satellite
bias adjustments are made relative to a satellite standard and do not depend on
limited sampling.
Several gauge-based analyses were also evaluated for reconstructions. Gauge
analyses were available on different spatial grids, most either 2.5 or 5 spatially.
For these analyses, the 5 grid was chosen as the reconstruction grid so that all of
the gauge analyses may be used. Grid analyses also cover different periods, but the
longest analyses began in 1900. Historical gauge analyses evaluated include the
analyses from the Global Historical Climatology Network (GHCN, Vose et al.
1998 ) 1900-2008, the Global Precipitation Climatology Centre (GPCC, Rudolf
2005 ; Schneider et al. 2008 ) 1901-2007, and the Climate Research Unit (CRU,
Hulme et al. 1998 ) 1900-1998.
In an earlier study (Smith et al. 2010 ), GPCP base data were used to compute
EOF modes, which were then used with each of these gauge analyses to perform a
reconstruction. Over land those reconstructions are similar for most of the overlap
analysis period. Before about 1930 the EOF-based reconstruction using GHCN had
weaker variations. That is because the GHCN does not fill regions with missing
data, while the others each use some limited interpolation to fill missing stations.
To ensure that noise is filtered out of our EOF-based reconstruction, the relative
sampling of each mode is checked using the data in each historical month. Modes
that are not adequately sampled are not used in the reconstruction. Omitting modes
damps the reconstructed signal. Since the GHCN analysis tends to be sparser than
the others before about 1930, its signal is damped for that period.
These EOF-based reconstructions were found to consistently represent interan-
nual oceanic variations such as those associated with ENSO and the North Atlantic
Oscillation (NAO). In addition, the improved reconstructions retain more Northern
Hemisphere oceanic variance than earlier reconstructions. However, the oceanic
multi-decadal components of the EOF-based reconstructions are less consistent.
This suggests that the weaker multi-decadal signal requires more sampling, while
the stronger interannual climate modes can be resolved by available gauges. To
better resolve the multi-decadal signal, we developed an indirect method of
reconstructing precipitation, which is described below.
The EOF-filtered GPCP, called REOF(GPCP), has variations similar to the
gauge-based reconstructions in the overlap times (see Fig. 7.3 ). Without filtering,
the GPCP variance increases in time, because the newer satellites resolve more