14-1190b-innovation-managing-risk-evidence

122
THE SCIENTIST’S PERSPECTIVE
Julia Slingo (Met Office Chief Scientist)
Protecting life, livelihoods and property is a
fundamental responsibility of national governments.
Weather-related hazards, including coastal and inland
flooding, are high priorities in the UK government’s National
Risk Register of Civil Emergencies 1 . The atmospheric, ocean
and land-surface observations and forecasts provided by the
Met Office play a central role in helping to mitigate these
risks.
Our world is increasingly and intricately interdependent,
relying on global telecommunications, efficient transport
systems, and the resilient and reliable provision of food,
energy and water. All of these systems are vulnerable to
adverse weather and climate. The additional pressure of
climate change creates a new set of circumstances and poses
new challenges about how secure we will be in the future.
More than ever, the weather and climate have considerable
direct and indirect impacts on us — our livelihoods,
property, well-being and prosperity — and increasingly we
rely on weather forecasts and climate predictions to plan
our lives.
Uncertainty is an inherent property of the fluid motions
of the atmosphere and oceans, which determine the
weather and climate at the regional and local level. This
was recognized in 1963 by Ed Lorenz in his seminal paper
Deterministic Nonperiodic Flow 2 , in which he introduces the
concept of the atmosphere as a chaotic system subject to
small perturbations that grow through non-linear processes
to influence the larger scale: as Lorenz said, “the flap of
a seagull’s wings may forever change the course of the
weather”.
It is important to understand that a chaotic system is not
the same as a random system. Chaos is manifested through
the physical processes that allow energy to cascade from
one scale to another and influence the final state of the
system. The evolution of a chaotic system depends on the
current state of the system, whereas a random system has
no knowledge of the current state and assumes that each
subsequent state is independent.
The concept of the weather and climate as chaotic
systems has had a profound impact on the way in which
forecasting has evolved over recent decades. No longer do
we produce a single, deterministic forecast, but instead we
perform an ensemble of forecasts that seek to capture the
plausible range of future states of the weather and climate
that might arise naturally from ‘the flap of the seagull’s
wings’. This enables the forecaster to assess the probability
of certain outcomes and to couch the forecast in terms of
likelihoods of hazardous weather. In some circumstances the
weather is highly predictable, and in other circumstances
there is a wide degree of spread and hence a high level of
uncertainty.
There are two major sources of uncertainty in the
prediction process. The first involves the certainty with
which we know the current state of the atmosphere
(and ocean), known as initial condition uncertainty. Despite
remarkable progress in Earth observation this uncertainty
will always be present, because instruments have
inaccuracies and we cannot monitor every part of the
system at every scale. Tiny perturbations are therefore
introduced in the initial state of the forecast, which then
grow and cause the forecasts to diverge.
The second source of uncertainty comes from the model
itself — model uncertainty — and recognizes that there are
unresolved ‘sub-grid scale’ processes that will affect the
evolution of the system. These include turbulent processes
in the atmospheric boundary layer, the initiation and
evolution of cumulus convection, the formation of clouds
and the production of precipitation. The sub-grid scale
variability in these processes is represented by random
perturbations at the resolved scale, and increasingly draws
on information from detailed observations and fine-scale
models that seek to characterize the spatial and temporal
characteristics of this sub-grid scale variability.
Uncertainty in climate prediction follows the same
principles that are used in weather forecasting, incorporating
both initial condition and model uncertainty. In climate,
however, initial condition uncertainty is only important
out to a few years ahead; beyond that, model uncertainty
dominates. Indeed, both model uncertainty and the
uncertainty in future emission scenarios dominate the range
of possible climate change outcomes towards the end of
the century. In this situation, model uncertainty goes beyond
the unresolved, sub-gridscale processes, and includes the
uncertainty in key physical parameters in the climate system,
such as the response of the carbon cycle to a warming
world and how readily cloud droplets are converted to rain
drops through cloud microphysics.
The reliability of ensemble forecasting depends on
whether the forecast probabilities match the observed
frequencies of predicted outcomes. In weather forecasting,
a reliable ensemble is one in which the ensemble spread is
representative of the uncertainty in the mean. In the context
of climate prediction, a reliable ensemble tends to be one in
which the ensemble forecasts have the same climatological
variance as the truth. This means that probabilistic
forecasting systems require substantial re-forecasting of past
cases to characterize the reliability of the system. In climate
change, of course, the past is not an analogue for the future,
and therefore gauging reliability depends much more on
scientific assessment.
One recent development in weather forecasting and
climate prediction is the translation of these probabilities in
terms of risk. The UK’s National Severe Weather Warning
Service warns the public and emergency responders of
severe or hazardous weather (rain, snow, wind, fog, or heat)
that has the potential to cause danger to life or widespread
disruption. Since 2011, the severity of the warning (yellow,
amber or red) depends on a combination of both the
likelihood of the event happening, and the impact that
the conditions may have at a specific location (flooding,