Geoscience Reference
In-Depth Information
Fig. 16.6
Mean squared skill score (
MSSS
) differences for decadal temperature hindcasts from a
12 member multi-model ensemble from CMIP5, for the initialized hindcasts (“forecasts”) minus
the uninitialized hindcasts (“reference”) as predictions of the observed climate. The forecast target
is year 2-9 following the initialization every 5 years from 1961 to 2006 (i.e. 10 hindcasts for each
model). Contour line indicates statistical signifi cance that the MSSS is positive at the 95 % confi -
dence level, with areas in yellow and orange indicating enhanced predictive skill from initialization
(Adapted from Meehl et al.
2013
)
results (see Fig.
9.6
,
Doblas-Reyes et al.
2013
; Meehl et al.
2013
) confi rm earlier
fi ndings that initialization improves skill mainly in the North Atlantic Ocean, with
some modest improvement in the tropical Pacifi c. Furthermore, a physical basis for
improved skill in the North Atlantic is potentially provided by skilful predictions of
an estimate of the Atlantic overturning circulation obtained from multi-model syn-
theses of available observations (Pohlmann et al.
2013
).
Improved skill in the Atlantic Ocean is encouraging since it might be expected to
provide improved predictions of Atlantic hurricanes and climate impacts over land
in many regions, including Africa, America and Europe (Sect. 16.2.1, Fig.
16.3
)
.
Indeed, retrospective forecasts of multi-year Atlantic hurricane frequency have high
correlations with observations (Smith et al.
2010
; Vecchi et al.
2013
). Initialization
improves the skill, consistent with improved surface temperature predictions in the
North Atlantic and tropical Pacifi c. However, much of the skill is attributable to
external forcing especially from anthropogenic aerosols (Dunstone et al.
2013
;
Villarini and Vecchi
2012
), which have also been suggested to be responsible for
much of the observed changes in AMV (Booth et al.
2012
). However, anthropo-
genic aerosol effects are uncertain (Zhang et al.
2013
), and their role in decadal
climate variability is a key area for future research.
There is currently little evidence that initialization improves predictions over
land in general (see Fig.
16.6
). This is surprising given the improvements in the
Atlantic and Pacifi c, which in observations appear to be related to impacts over
land, possibly indicating an inadequate atmospheric response to ocean anomalies in
models. However, initialization does improve predictions of climate impacts over
Search WWH ::
Custom Search