|
|
FORECASTING CHANGES IN WEATHER REGIMES
What is a weather regime?
The notion of weather regimes originates from empirical observations
of midlatitude weather throughout this century. It has been observed
that, on time scales of a few days up to a few weeks,
'quasi-stationary anomalies' repeatedly occur at the same
locations. For example, weather in the UK is strongly influenced by
the midlatitude westerly jet, which tends to steer extratropical
cyclones across the north Atlantic and into northern Europe. However,
this upper level 'zonal flow' is sometimes blocked by large-scale high
pressure systems over the UK as shown in the figure below. This figure
shows an example of a dipole-like 'blocked regime', with anomalous
easterly winds in the upper level flow along the north coast of
France.
The easterly flow associated with blocking can lead to significant
weather in the UK such as droughts in summer, or extended cold periods
in winter. It is therefore very important to be able to forecast such
events well. Unfortunately, current forecast models have problems
predicting changes in weather regime, such as block onset, more than a
few of days in advance.
The aim of my research is to investigate the nature of regime changes,
such as block onset, using the Ensemble Prediction System
at the European Centre for Medium-range Weather Forecasting (ECMWF).
The Ensemble Prediction System
Numerical forecasts are 'deterministic', that is to say that, given an
initial state of the atmosphere, a forecast model can be run forward
in time to produce one forecast of the state of the atmosphere at some
future time. However, after a few days, errors in the specification of
the initial atmospheric state may start to dominate the forecast such
that it is no longer useful. The Ensemble Prediction System (EPS) at
ECMWF is intended to extend the range of usefulness of numerical
forecasts by exploring the concept of 'probabilistic forecasting'. The
main idea behind the EPS is to take into consideration the
uncertainties in the description of the initial atmospheric state,
exploring what role they may play in the development of the forecast.
To simulate the effects of possible analysis errors, the basic
analysis is slightly modified, or 'perturbed', in sensitive areas
identified by the system depending on the current flow pattern. The
perturbations are then combined to form 25 different perturbation
patterns. By just reversing the signs, another 25 mirrored patterns
are created, thereby creating 50 slightly different initial analyses
and 50 possible forecasts of the future state of the atmosphere.
The EPS forecasts can be used as a measure of the predictability of
the atmosphere or as an indicator of possible alternative atmospheric
developments. They may also be used to produce local probabilistic
forecasts of weather parameters.
PV/PT diagnostics
Before considering changes in weather regime, it is first necessary to
be able to define each regime in an objective way. In order to do
this, I am using 'potential vorticity' (PV) and 'potential
temperature' (PT) as diagnostics. Potential vorticity is a useful
variable for two reasons. Firstly, PV may be used as a Lagrangian
tracer as it is 'quasi-conserved' on isentropic surfaces (surface of
constant PT) for periods of up to a week or so. Similarly, PT is
quasi-conserved on PV surfaces. Secondly, one can deduce
diagnostically the complete flow structure from the spatial
distribution of PV. This is known as the 'invertibility
principle'.
As an example of PV/PT diagnostics, the figure below shows the
analysed field of PT on PV=2 corresponding to the 250mb height field
in the previous figure. The blocking high over the UK is associated
with a positive potential temperature anomaly. In fact, the 250mb
height field appears very much as a highly smoothed version of the PT
on PV=2 map. This is a common property of such maps, and is
particularly notable away from the earth's surface, as in this case.
Using PV/PT diagnostics, I have defined a 'blocking
index' B, which is described in the next section.
A PV/PT blocking index
The general westerly flow at upper levels in the extratropics is
associated with a negative latitudinal gradient in PT on a PV
surface. During a blocking episode, this gradient tends to be
reversed, with high PT to the north and low PT to the south in the
region of the block. Therefore, we may define our blocking index B at
a reference longitude as the difference in mean potential temperature
between high and low latitudes relative to that longitude:
B = mean PT in northern area - mean PT in southern area
where the northern and southern areas refer to the relative locations
of the averaging areas.
The figure below shows a plot of B against longitude calculated from
the potential temperature field shown above. This figure indicates
that B>0 from 10 degrees W to about 35 degrees E, so this region may
be instantaneously defined as blocked. Blocking 'episodes' can be
defined by B>0 over a range of longitudes for a certain number of
days. In the case of the block shown here, B>0 from September 20 to 24
over at least 30 degrees of longitude (or thereabouts).
Future work
The work presented here pertains to a case study of a blocking episode
which started on 20 September 1998. Just before the onset of this
block, the ECMWF EPS had significant forecasting problems such that
the 5- and 6-day forecasts of September 18 showed a huge drop in
skill. This is depicted in below in terms of the 'anomaly correlation
coefficient' of the 500mb geopotential height over Europe for the
forecast of September 13. Some of the 4-day EPS forecasts showed
considerably more skill, although September 18 still presented a
'predictability barrier'. By examining individual EPS member forecasts
I hope to gain some insight into what caused this predictability
barrier and perhaps gain some information on the nature of
perturbations that are most likely to cause a transition to a blocked
regime.
Having analysed the data from this case study, I intend to look at
further case studies, including cases where the EPS has performed
particularly well. After sufficient case studies, I hope to be able to
refine my blocking index to be able to define blocking generally in
terms of PT on the PV=2 surface. In the longer term, I also intend to
consider regimes other than blocking.
Return to DYNAMICAL PROCESSES home page.
This page is maintained by J.L.Pelly@reading.ac.uk
|
|
