department of meteorology – university of reading ...ben/blocking2016/talks/rodwell.pdf6-8 april...
TRANSCRIPT
Diagnostics of the prediction and maintenance
of Euro-Atlantic blocking
Mark Rodwell, Laura Ferranti,
Linus Magnusson
European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, RG2 9AX, UK © ECMWF March 3, 2016 [email protected]
Workshop on Atmospheric Blocking
6-8 April 2016, University of Reading
1
Difficulties in predicting blocking
European Centre for Medium-Range Weather Forecasts [email protected] 2
Animation of ‘bust’ forecast
Animation of forecast started at 0 UTC on 10 April 2011
European Centre for Medium-Range Weather Forecasts [email protected] 3
Potential Vorticity on 320K
Animation of ‘bust’ forecast
Animation of forecast started at 0 UTC on 10 April 2011
Block forms in observations, but not in forecast
It is difficult, by day 6, to disentangle model error from the natural growth
of initial condition uncertainty (chaos)
European Centre for Medium-Range Weather Forecasts [email protected] 4
Potential Vorticity on 320K
Animation of ensemble forecast: Initial peturbations
Potential Vorticity on 320K
Animation of ensemble forecast
Ensemble forecasting (flow evolution to day-6)
Potential Vorticity on 320K
Occasional ‘busts’ in forecast performance
Spatial Anomaly Correlation Coefficient for 500 hPa geopotential height in [12.5oW –42.5oE, 35oN–75oN]. Date is forecast start
Rodwell et al, 2013, BAMS
European Centre for Medium-Range Weather Forecasts [email protected] 7
European Z500 skill at day 6
Bust around 10 April 2011
• Initial condition error?
• Model error?
• Reduced predictability?
Verifying conditions composited over many bust forecasts
Composite of 584 busts in ERA Interim forecast prior to 24 June 2010
Rodwell et al, 2013, BAMS
European Centre for Medium-Range Weather Forecasts [email protected] 8
Rex-type block
Unit = m Bold colours = statistical significance at 5% level
500 hPa geopotential height (Z500) anomaly
Regimes
Regimes based on clustering of daily anomalies for 29 cold seasons (1980-2008)
Ferranti et al. 2015, QJRMS
European Centre for Medium-Range Weather Forecasts [email protected] 9
m2s2
500 hPa geopotential
Blocking: More poorly predicted and not persistent enough
From Laura Ferranti
European Centre for Medium-Range Weather Forecasts [email protected] 10
Day 0 Day 1 Day 5 Day 7 Day 10
100/100 70/70 44/52 36/47 29/41
NAO+ European Blocking
Skill in predicting regime projection
Blocking persistence: ECMWF model/analysis
1 day worse than for
NAO+ at CRPS=0.5
Ensemble evolution in phase space
Initial date: 22 September 2015 0UTC
In presentation to ECMWF Scientific Advisor Committee, 2015
European Centre for Medium-Range Weather Forecasts [email protected] 11
Analysis
HRES
ENS member
• Nice way to summarise ENS in
two dimensions
• Transition to blocking well-
predicted 4 days ahead
• Blocking projection perpetuates to
day 10, but spread increases
Euro
pean B
lock →
- ← NAO → +
“Ferranti - Magnusson diagram”
(similar to phase-space diagram of MJO)
Blocking onset and forecast ‘reliability’
12European Centre for Medium-Range Weather Forecasts Mark J Rodwell
Composite initial conditions of bust forecasts
Rodwell et al, 2013, BAMS
European Centre for Medium-Range Weather Forecasts [email protected] 13
There is an initial flow regime: “Rockies
trough” with high CAPE ahead
Conducive to the formation of
mesoscale convective events (MCS)
Remarkable that we can identify any
significant initial conditions 6 days
ahead of the busts – this must be due to
the large composite (584 events) used
Other bust causes not so geographically
fixed and are not highlighted by this
composite-mean
‘CAPE’ = Convective Available Potential
Energy
Bold = 5% significance
Spread-error for Trough/CAPE composite (MCS)
• Following conditions conducive to MCS development, enhanced errors and spread propagate east towards Europe → ‘Busts’
• Note: -ve residuals occur in non-trough/CAPE situation too.
• +ve residual at D+5 is not significant (Chaos? → use bigger sample or shorter leadtime? But analysis uncertainty at D+1?)
Rodwell et al., 2015, Report to SAC
European Centre for Medium-Range Weather Forecasts [email protected] 14
95% confident
Not significant
54 cases 200 hPa geopotential
D+1
Error2 Ensemble Variance Residual
D+3
D+5
Error2 = EnsVar + Residual
Reliability [Residual]=0
EDA reliability budget: Trough/CAPE composite
• Residual highlights MCS, and suggests lack of background variance. (Obs uncertainty changes 2nd-order)
• MCS uncertainty (existence, intensity, location) not well reflected in Jetstream uncertainty (with downstream consequences)
• Budget useful to diagnose biases, modelling of observation error and representation of model uncertainty (including
stochastically-formulated process parametrizations)
European Centre for Medium-Range Weather Forecasts [email protected] 15
Depar2 = Bias2 + EnsVar + ObsUnc2 + Residual
Reliability [Residual]=0
Relative to aircraft observations of zonal wind 200hPa (±15)54 cases Rodwell et al., 2015, Report to SAC
EDA = “Ensemble of Data Assimilations”
MCS – Jetstream interaction (composite)
• Increments emphasize model systematic error: MCS does not interact enough with Jetstream
• Also need to strengthen stochastic physics to increase background variance?
European Centre for Medium-Range Weather Forecasts [email protected] 16
PhysicsPhysics + analysis increment
u=25ms-1
3Kd-1
Jetstream
MCS
Met3D: Marc Rautenhaus
MCS – Jetstream interaction (composite)
• Increments emphasize model systematic error: MCS does not interact enough with Jetstream
• Also need to strengthen stochastic physics to increase background variance?
European Centre for Medium-Range Weather Forecasts [email protected] 17
PhysicsPhysics + analysis increment
u=25ms-1
3Kd-1
Jetstream MCS
Met3D: Marc Rautenhaus
Maintenance of blocking
18European Centre for Medium-Range Weather Forecasts Mark J Rodwell
Initial process tendencies and analysis increment (DJF 2016)
Dynamical and physical process tendencies integrated over the 12h background forecast of the data assimilation (EDA cntl)
European Centre for Medium-Range Weather Forecasts [email protected] 19
Analysis increments
suggest model warms
over land a little too
much
(note different contour
interval to tendencies)
T500
Initial process tendencies and analysis increment (Blocked)
Composite over three blocked periods: Dec 4-9, Dec 26-28, Jan 26-28 (24 analysis cycles)
European Centre for Medium-Range Weather Forecasts [email protected] 20
Cloud forcing (net latent
heating associated with
microphysics) highlights
Warm Conveyor Belt (WCB)
Note negative dynamical
tendency in this region
T500
Initial process tendencies and analysis increment (Blocked-DJF)
Composite over three blocked periods: Dec 4-9, Dec 26-28, Jan 26-28 (24 analysis cycles) minus DJF 2015/16
European Centre for Medium-Range Weather Forecasts [email protected] 21
Convective forcing
unchanged relative to full
season mean
Increments suggest WCB
cloud forcing too weak?
(can probably discount effect
on increments of
compositing on observed
blocking)
T500
Barotropic vorticity equation in upper troposphere (analysis)
Rossby Wave Source and advection by rotational flow smoothed, averaged over composite and integrated 100 – 300 hPa
European Centre for Medium-Range Weather Forecasts [email protected] 22
Divergence associated with
WCB
Down-stream advection (+ve)
and β-effect (-ve)
Deficiencies in perpetuating blocking associated
with slight weakness of WCB heating?
Rossby wave source: -𝛻. 𝑣𝜒𝜁
Vorticity advection by rotational flow: -𝑣𝜓. 𝛻𝜁
DJF 2016 Blocked
DJF 2016 Blocked Blocked-DJF
Blocked-DJF
Onset of blocking
• Difficult to predict beyond a few days
… associated with ‘busts’
• Diabatic processes important
… produce large-amplitude waves that break to form a block
… associated with instabilities that decrease predictability
• We may under-represent uncertainty
… due to systematic errors or deficiencies in stochastic physics
Maintenance of blocking
• Warm conveyor belts important for vorticity forcing that stabilises block
• We may have too weak cloud forcing in the WCBs
Summary
23European Centre for Medium-Range Weather Forecasts Mark J Rodwell
Extra slides
European Centre for Medium-Range Weather Forecasts [email protected] 24
Skill in predicting from regimes
Skill in predicting from a given regime. October – March, 2007 – 2012. 95% confidence based on bootstrapping
Ferranti et al. 2015, QJRMS
European Centre for Medium-Range Weather Forecasts [email protected] 25
ACC
European (and Atlantic) blocking is more
poorly predicted than NAO
Reliability in ensemble forecasting
(Cross-terms on squaring have zero expectation. EnsVar is scaled variance to account for finite ensemble-size)
European Centre for Medium-Range Weather Forecasts [email protected] 26
Error2 = EnsVar + Residual
The importance of
reliability is the motivation
for using ‘proper’ scores
(such as the Brier Score or
CRPS).
Reliability (at all leadtimes)
should reduce ‘jumpiness’
of ensemble forecasts
Adapted from Rodwell et al., 2015, QJRMS
Reliability in ensemble data assimilation
European Centre for Medium-Range Weather Forecasts [email protected] 27
Adapted from Rodwell et al., 2015, QJRMS
Depar2 = Bias2 + EnsVar + ObsUnc2 + Residual