213 / 218
[
] 211
over a season – that has potential predictability.
Though the weather on any given day is entirely
uncertain at long lead times, the persistent influence
of the slowly evolving surface conditions may change
the odds for a particular type of weather occurring
on that day. In rough analogy to the process of throw-
ing dice, the subtle but systematic influence of the
boundary forcing can be likened to throwing dice
that are “loaded”. On any given throw, we cannot
foretell the outcome, yet after many throws the biased
dice will favour a particular outcome over others. This
is the sort of limited predictability that characterizes
seasonal prediction
3.5 Currently, seasonal predictions are made using both
statistical schemes and dynamical models. The statis-
tical approach seeks to find recurring patterns in
climate associated with a predictor field such as sea-
surface temperature. Such models have
demonstrated skill in forecasting El Niño and some
of its global climate impacts. The basic tools for
dynamical prediction are coupled models – models
that include both the atmosphere and the other
media of importance, particularly the oceans. Such
models are initialized using available observations
and integrated forward in time to produce a seasonal
prediction. The issue of uncertainty is handled using
an ensemble approach, where the climate model is
run many times with slightly different initial condi-
tions (within the range of observation errors or
sampling errors). From this, a distribution of results
is obtained, whereupon statistics of the climate can
be estimated. Recently, encouraging results have
been obtained from ensemble outputs of more than
one model being combined
3.6 There are several limitations attending current predic-
tions. Most coupled models (and to a lesser extent
uncoupled models) exhibit some serious systematic
errors that inevitably reduce forecast skill. Data avail-
ability is a limitation for both statistical models and
for dynamical models. In the latter case, very limited
information is available for much of the global oceans
and for the land surface conditions. Also, current
initialisation methods do not account properly for
systematic model errors, further limiting forecast
performance. A final set of limitations arises for prac-
tical reasons. Due to resource requirements, most
seasonal predictions cannot be done at resolutions
comparable to weather prediction
Furthermore, rather small ensemble sizes (of the
order of 10) are used for some models, certainly less
than is optimal for generating robust probabilistic
forecasts. Current research is addressing the poten
tial for regional “downscaling” of climate forecasts
by various means and the possibilities for more
detailed probabilistic climate information from
expanded ensembles of one or more models
3.7 Possible use of seasonal forecasts is currently being
explored in various contexts. In each case, effective
use will require careful attention to the issue of uncer-
tainty inherent in seasonal forecasts. Future
advancements can be expected to improve the esti-
mates of uncertainty associated with forecasts, thus
allowing better use of forecast products.
4. Projection of future climate
4.1 As explained above, based on the current observed
state of the atmosphere, weather prediction can
provide detailed location and time-specific weather
information on timescales of the order of two weeks.
Some predictability of temperature and precipitation
anomalies has been shown to exist at longer lead
times out to a few seasons. This comes about because
of interactions between the atmosphere, the oceans,
and the land surface, which become important at
seasonal timescales. At longer timescales, the current
observed state of the atmosphere and even those
large-scale anomalies which provide predictive skill at
seasonal to interannual timescales are no longer able
to do so due to the fundamental chaotic nature of
the Earth-atmosphere system. However, long-term
changes in the Earthatmosphere system at climate
timescales (decades to centuries) are dependent on
factors which change the balance of incoming and
outgoing energy in the Earth atmosphere system.
These factors can be natural (e.g. changes in solar
output or volcanoes) or human induced (e.g.
increased greenhouse gases). Because simulations of
possible future climate states are dependent on
prescribed scenarios of these factors they are more
accurately referred to as “projections” not “predic-
tions” or “forecasts”
4.2 In order to perform climate projections, physically-
based climate models are required in order to repre-
sent the delicate feedbacks which are crucial on
climate timescales. Physical processes and feedbacks
that are not important at NWP or even at the
timescales of seasonal prediction become crucial
when attempting to simulate climate over long
periods, e.g. cloud-radiation interaction and feed-
back, water vapour feedback (and correctly modelling
long-term trends in water vapour), ocean dynamics
and processes (in particular an accurate representa-
tion of the thermohaline circulation). The treatments
of these key features are adequate to reproduce many
aspects of climate realistically though there remain
many uncertainties associated with clouds and
aerosols and their radiative effects, and many ocean
processes. Nevertheless, there is reasonable confi-
dence that state-of-the-art climate models do provide
useful projections of future climate change. This
confidence is based on the demonstrated perfor-
mance of models on a range of space timescales