Transcript Case Studies in Decadal Climate Predictability Leon Hermanson and Rowan Sutton, Department of Meteorology, University of Reading, UK 1.

Case Studies in Decadal Climate Predictability
Leon Hermanson and Rowan Sutton, Department of Meteorology, University of Reading, UK
1. Introduction
Is climate predictable beyond seasonal (El Niño)
timescales? To what extent does knowledge of initial
conditions constrain long-term climate forecasts?
Previous studies show potential predictability on annual
time scales is greatest for ocean variables and is weak
for climate variables. This study investigates whether
there is predictability beyond the limit of seasonal time
scales for climate variables. Recognising that
predictability is likely to be dependent on initial
conditions we consider model-based case studies
rather than average predictability. Two case studies
examined here show potential predictability in tropical
precipitation and West European surface temperatures
two years ahead, but differ substantially in the details.
Initial phase: Pre-industrial
to modern day
1860
Generation of initial
conditions
1950
Two initial conditions
chosen for case study
~1980
Figure 1 Schematic of experiment set-up. An ensemble
is started in 1950 from a single run (black). From this
ensemble case studies are chosen. Each case study consists
of two ensembles (red and purple) of ten members each.
All integrations are forced by observed changes in radiation.
Figure 2 Difference in the first annual mean ensemble mean 500m ocean
heat content for each of the four case studies. Units are 10-19J.
2. Generation of Initial Conditions
Figure 1 shows the methodology. An ensemble of 20th century simulations with
HadCM3 are used to generate initial conditions for the case studies. Members of
this ensemble which show large, persistent, regional differences in ocean heat
content are used to initialise the case study ensembles. Figure 2 gives an
indication of the differences in the initial ocean heat content for the four case
studies shown here. Note that large anomalies exist in all ocean basins.
3. Predictability Plumes
Figure 3 Predictability plumes from the four case studies (one per row). The first column
is global 500m ocean heat content, the second an SST index of the Interdecadal Pacific
Oscillation and the third the Atlantic meridional overturning circulation. The dark line
shows the ensemble mean and the shading one standard error for that mean. The
predictability as determined by a t-test is printed on each plot.
Figure 3 shows examples from all four case studies of plumes for three different
ocean variables: global mean ocean heat content (OHC), a Pacific sea surface
temperature (SST) index and the Atlantic meridional overturning circulation (MOC)
at 30°N. As expected from previous studies the OHC (first column) is predictable
between 2—7 years ahead. This is despite the ocean heat content being strongly
influenced by external forcing. The second column shows a sea surface
temperature (SST) index of the Interdecadal Pacific Oscillation (IPO). The IPO is a
basin-wide multi-decadal pattern in the Pacific that can modulate El Niño Southern
Oscillation (ENSO) teleconnections. Predictability of the index is generally low, but
interestingly in plot (e) it looks like predictability returns in year 8. Further work is
needed to determine whether cases of “returning predictability” such as this one,
are really predictable. The third column shows the Atlantic MOC, which is linked to
ocean meridional heat transport and important for western European climate. It is
potentially predictable between 3─6 years ahead here.
4. Predictability of Climate in the Second Year
Figure 4 shows differences between annual mean ensemble mean maps of
surface temperature and precipitation in the second year for experiments 2
and 4. It is clear that in some regions predictability exists for both these
variables beyond ENSO time scales. Comparing the two experiments it is
clear that the initial conditions strongly influence the predictability. The other
case studies also show evidence of climate predictability in the second year,
but not always in the same areas. Figure 5 shows predictability plumes for
some seasonal and regional mean quantities for all four experiments. It shows
that two experiments (1&4) show 3 year predictability for western European
temperatures. Two experiments (1&2) also have two year predictability in
tropical South Atlantic precipitation. The mechanisms which give rise to these
predictabilities are currently being investigated, but the most common
mechanism is persistence of ocean heat content anomalies.
Figure 5 Same as figure 3, but for various seasonal mean quantities (season indicated on plot).
5. Conclusions
Figure 4 Differences in second annual mean ensemble mean 1.5m surface
temperature (°C) and precipitation (mm/day) for experiments 2 and 4. Line
contours show significance levels (dashed for negative anomalies).
Perfect model experiments have been carried out with HadCM3 on a case-by-case
basis to investigate decadal predictability of transient climate.
• Predictability in the ocean varies from 0—2 years for SST to 2—7 years for OHC
and 3—6 years for the MOC, this is broadly in line with previous studies.
• Though predictability of atmospheric climate variables is usually less than for ocean
variables, for some initial conditions there can be predictability up to 3 years for
regional variables.
• The length of predictability and the region and season in which it occurs is very
state dependent and differed in all four experiments.
• The mechanism that gives rise to predictability is commonly persistence of OHC
anomalies.
• Cases of returning predictability will have a more complex mechanism.
GCEP