© University of Reading 2007 www.reading.ac.uk

THORPEX PDP/ WGNE workshop, Zurich 2010

Using observational experiments to isolate dynamical processes and deficiencies in their representation by models

John Methven, Department of Meteorology

What is the scope of this talk?

Not – How best to use obs to reduce forecast error?

(not about targeting to reduce IC uncertainty)

Not – What new instrumentation is required?

Stems from – diagnostics for the extratropics that inform

us about atm evolution (history and anticipated future)

Motivated by – what can high resolution obs tell us that

we could not learn from a virtual experiment?

Dependent on quality of model and its structure (not just parametric model error)

Motivation

Analysing the atmosphere is complicated by its non-local nature

1. Long-range transport of nearly conserved properties by winds.

“Origin” is remote from observer (e.g., origin of water in precipitation)

2. Balance between variables, mediated by fast wave propagation

action-at-a-distance and concept of “induced flow”.

3. Wave-mean flow interaction

Waves effects felt in dissipation region, remote from source.

Diagnosed effects depend of partition of background flow from full state.

Consequences

Diagnostics are complicated by the non-local nature of the

atmosphere

1. Observed variability at a point dominated by changing air mass and not processes occurring within air masses.

2. Some variables, such as geopotential height, are inappropriate for identifying model error

3. Localised forecast error must be traced back in time and method depends on variable.

1. Transport diagnostics

Winds are 3-D and time-varying, but dominated by larger scales

chaotic advection where trajectories of air masses are sensitive

to initial conditions and separate exponentially on average.

Example:

Trajectories calculated forwards and backwards in time (for 5 days) from an aircraft flight track.

Note sampled air originates from a wide area and also spreads into a wide area.

Numerical integration of

Where winds from ECMWF analyses.

( ( ), )ii

Dxu x t t

Dt

Quasi-Lagrangian experiments

Do not tag air masses throughout history (e.g., with dye or balloon)

BUT need to forecast air-mass trajectories in order to direct aircraft to intercept the same air several times over a long range.

Can observe chemical and thermodynamic tracers with aircraft to establish air-mass identities and matches.

Motivation for the ITCT-Lagrangian 2K4 Experiment which took place within the framework of the ICARTT campaign in summer 2004:

Sample polluted air masses leaving the continental BL.

Follow across Atlantic since no emissions after USA.

Deduce chemical transformation en route.

Scale of problem requires several coordinated research aircraft.

JGR, ICARTT special issue

Fehsenfeld et al [2006] – campaign overview

Methven et al [2006] – Lagrangian experiment

e.g., New York to Spain – 4 flights linked by trajectories

1 2

3

Trajectories from BAe146 flight track (blue) back and forwards for 3.5 days.

Best matches with trajectories from other flight tracks.

2 3 1

4

4

+3

Did the aircraft sample the same

air mass many times?

Two independent matching methods:

1. Trajectory models driven by met. analyses 2. Hydrocarbon fingerprints (bottled air samples)

Search for coincident matches: two samples with matching HC fingerprint are also linked by matching trajectories. Quality of matches assessed using independent observations of thermodynamic tracers

Lagrangian match quality evaluated with temperature and humidity observations

Coincident HC/FLEXPART matches

Hydrocarbon matches

Pairing random time points

Trajectory matches

Coincident matches strongly peaked (almost adiabatic)

New York - Ireland. Mixing and cooling in North Atlantic MBL.

Latent heat release (ascent).

Trajectory-only matches good in theta. Analysis close to obs.

Coincident HC/traj matches

Matching HCs alone does not make a Lagrangian match

D/Dt De/Dt

Using tracer conservation properties

Reverse domain filling trajectories

Finescale tracer structure can be diagnosed by calculating many back trajs from a dense 3-D grid and colouring each grid-point by value of tracer at origin.

e.g., Sutton et al [1994], JAS

Back trajectories from points A, S and E on section XY

Sampled using MRF aircraft in 2000. Methven et al [2003], JGR

Meteosat WV chan dry intrusion

X Y

S

Is finescale structure

accurately represented?

Allowing for non-conservation

• RDF trajectory reconstructions assume tracer conservation.

• Strong gradients arise as air masses of different origins are

brought together by strain flow (tracers are long-lived relative to

Lagrangian decorrelation timescale ~1 day in mid-lats).

• As trajectory length in time increases, RDF structure becomes

increasingly finescale until unrealistic.

• Relevant trajectory length is determined by non-conservative

processes following air-masses.

Next level of sophistication is to use simple models along trajs.

Application for water vapour.

(e.g., Sherwood [1996], J.Clim.; Pierrehumbert and Roca [1998], GRL)

Advection-condensation model (1)

qmin = min[ qsat ] (occurs at t=-τ min)

qtraj = min[ q(-T), qmin ]

Back trajectories from

coordinates of radiosonde.

T, q interpolated from analyses to

trajectory coords.

RH traj.

RH analysis

100 hPa

RH traj.

RH analysis

100 hPa

black areas

grey areas

Cau, Methven and Hoskins (2005), JGR, 110, d06110

Sonde observations of RH in TOGA COARE expt

Dry events simulated using trajectories

Trajectory statistics

• Trajectory ensembles from a small volume rapidly diverge

and can cross.

• Even trajectories from a point will form a tangle if collated

over time.

• Hard to visualise typical trajectory behaviour.

• One approach is to identify special events along trajectories

(e.g., time of last condensation) and create number density

distributions characterising location of events.

discarding information about complex path between the

event and trajectory release point.

Origins for isolated dry regions

Dry regions in subtropics are isolated by boxes.

Contours show corresponding density of origin.

Many contributors to each dry region.

Destination of air saturating in isolated regions

Condensation events within isolated boxes.

Contours show arrival locations of these dry air masses.

Each condensation region contributes to one/two dry regions.

2. PV diagnostics

Ertel PV is conserved by the unapproximated dynamical equations

following adiabatic, frictionless flow.

Step 1: Calculate Ertel PV as diagnostic from model variables

Step 2: Assume material conservation and use it to infer finescale

structure using RDF trajectories

Step 3: Integrate tracers which accumulate non-conservative effects

on PV

Step 4: Define a background state and anomalies from it.

Step 5: Make a balance approximation and invert the PV anomaly

distribution or portions of it to obtain “induced flow anomalies”

PV structure assuming conservation

RDF trajectories

Air mass: history of ascent

Change in pressure along trajectory before arrival over UK Green/blue=ascent 1.75 days travel

0.75 days travel

w = warm conveyor belt of cyclone over UK

w

w

Air mass: integrated heating

w

w

0.75 days travel

w = warm conveyor belt, latent heating rate ~ 10 K day -1

PV non-conservation

Change in potential temperature along trajectory before arrival over UK Yellow/red=increase 1.75 days travel

Sensitivity of model PV to

representation of diabatic processes

Experiments running the global and 12km LAM versions of the Met Office Unified Model

(Jeffrey Chagnon, NCAS-weather)

Same case as RDF example. Shown 4 days prior to forecast bust over Europe identified by Thomas Jung.

Same cross-section taken.

Global vs mesoscale model PV

Tracking non-conservative changes in PV

• New set of diagnostics has been developed by Bob Plant

(Reading) based on a Lagrangian decomposition of the PV field.

• Full PV conservation equation can be written:

p

p

DqS

Dt where Sp denotes the Lagrangian tendency

resulting from one physical process in model

• Tracers qp are initialised as zero but experience only one of the Sp

terms as well as being advected by the semi-Lagrangian scheme.

each tracer shows accumulated contribution of one process to PV.

passive p

p

q q q error

Diagnosing processes affecting PV

●

●

●

●

Accumulated PV tracers for the effects of convection and large-scale rain parameterisations in LAM model.

3.Cyclone tracks in ensemble forecasts

Lizzie Froude, ESSC, Reading. Weather and Forecasting (2010)

Using Kevin Hodges’ cyclone tracking algorithm on ensemble forecasts from different centres archived on TIGGE database @ ECMWF.

Example of the control forecasts for one cyclone collated from 9 operational centres.

Bias in intensity and speed

Forecasts from most centres underestimate cyclone intensity (except ECMWF, CMA, CMC).

Cyclones propagate too slowly in forecasts (feature of all centres).

Plans for future observational expts T-NAWDEX THORPEX- North Atlantic Waveguide and Downstream Impact Experiment Has been proposed by the THORPEX working group Predictability and

Dynamical Processes for the European THORPEX Science Plan.

Its overarching scientific goal is to investigate in detail the

physical processes that are primarily responsible for degradation in 1-7 day

forecast skill in global prediction systems and of their representation in NWP

models.

An international field experiment is proposed for autumn 2012 observing

diabatic processes within Atlantic weather systems.

DIAMET DIAbatic influences on Mesoscale structures in ExTratropical storms Consortium led by Geraint Vaughan (NCAS-weather director) with Methven,

Parker and Renfrew as other lead PIs + Met Office partners. Response to NERC Natural Hazards theme action call (shorter lead times and

smaller scales than T-NAWDEX plan). DIAMET overarching theme is the role of diabatic processes in generating

mesoscale PV and moisture anomalies in cyclones, and the consequences of those anomalies for weather forecasts.

Three-pronged approach: a) Determining influence of diabatic processes on mesoscale structure (PV

tracers partitioned by process) b) Improving parameterisation of convection (in cyclone environment), air-

sea fluxes and microphysics. c) Using feature-tracking within the MOGREPS ensemble to quantify the

predictability of mesoscale features and the dependence of the skill of weather forecasts (precip and winds) on mesoscale features.

T-NAWDEX pilot Brief project with flight-time only funded (Lead PI: Ian Renfrew). 3 flights in November 2009 trying out flight plans to identify influence of diabatic processes on mesoscale structure within growing cyclones. 3/11/09: intense cold front case 13/11/09: attempted two flights ahead and behind frontal cyclone, crossing WCB both times to infer Δ. First flight crossed intense warm front and WCB dropping sondes. Forced to abandon second flight due to malfunction. 24/11/09: Box within marine BL either side of cold front plus transect across WCB behind.

Cold-front case (3/11/09)

Strong rearward sloping cold front with tropopause fold over-running high -e air. Location of front well forecast and -e contrast predicted to within 1K. But, precip forecast was poor - model went for strong precip along front over ocean, but very active convection broke out across England where fold over-ran (this was not forecast).

Friday 13th: Frontal cyclone

ozone

ozone

theta

u

v

Zooming in on shear lines

Sudden change in ice crystal shape crossing shear lines

No coincident temperature gradient

Origin unknown – slantwise circulation in near-neutral environment?

ozone

ozone

theta

u

v

u

theta

UM LAM (12km) simulation

38 Reading, Jan 2010

Hindcast appeared to show patterns of vertical motion and other variables along front.

Jeffrey Chagnon, NCAS-weather

39 Reading, Jan 2010

PV

Heating

w

Analysis of FASTEX IOP16 cyclone

Nigel Roberts, Met Office (JCMM)

IOP16 - Model simulated run 1

Sid Clough et al

UM simulation at 1.5, 4, 12km

42 Reading, Jan 2010

Precipitation

In this case roll structure in model is lost as horiz resolution increased.

Origin of observed wind shears still unknown.

WCB case: 24th Nov 2009

Radar (21:00 UTC) T+18 operational forecast Met Office UKV model (1.5km)

WCB case: 24th Nov 2009 • PV sources/sink

accumulated from 00

utc on 22 November

2009, UMet12 km run

•The strip of high PV

along the cold front is

diabatically

generated.

• Main contributions

are from boundary-

layer heating, cloud

microphysics, and

convection schemes.

• The high PV strip is

involved in the

generation of a large-

amplitude gravity

wave packet.

Strip of high PV above

surface front

Upper-level

trough

Summing up: Nature of the atmosphere

2. Balance and PV

• PV anomalies induce flow at-a-distance.

• Inversion to obtain vertical motion or overturning circulations is also non-local.

1. Transport

• Thermodynamic and chemical tracers carried vast distances

Complex source-receptor relationships.

3. Waves and eddies

• Waves forced in one location can propagate across globe and break/dissipate far away wave-transport.

Summing up: Suitable diagnostics

2. Balance and PV

• PV error diagnostics

• Piecewise PV inversion and

PV modification of forecasts

1. Air-mass modification

• Diabatic tendencies partitioned by process (as YOTC)

• Trajectories and Lagrangian experiments

• Tracer fields (accumulating non-conservative effects of processes)

3. Coherent eddies and waves

• Cyclone tracking and composite analysis.

•Teleconnections, Rossby wave “source” and wave activity flux.

DIAMET DIAbatic influences on Mesoscale structures in ExTratropical storms Consortium led by Geraint Vaughan (NCAS-weather director) with Methven,

Parker and Renfrew as other lead PIs + Met Office partners. Response to NERC Natural Hazards theme action call. DIAMET overarching theme is the role of diabatic processes in generating

mesoscale PV and moisture anomalies in cyclones, and the consequences of those anomalies for weather forecasts.

Three-pronged approach: a) Determining influence of diabatic processes on mesoscale structure (PV

tracers partitioned by process) b) Improving parameterisation of convection (in cyclone environment), air-

sea fluxes and microphysics. c) Using feature-tracking within the MOGREPS ensemble to quantify the

predictability of mesoscale features and the dependence of the skill of weather forecasts (precip and winds) on mesoscale features.

DIAMET WP A: Mesoscale structure and diabatic effects Two flight campaigns with FAAM aircraft spanning: Autumn 2011: 3 month period of opportunity with 2 week IOP. Late summer 2012: 2 month period link with T-NAWDEX/HYMEX? New streams of Doppler radar data from Met Office Radar group and

data assimilation for convection-resolving model (JCMM) Nested Met Office operational forecasts at varying resolution (currently 40km global, 12km LAM, 1.5km UK) plus hindcasts at varying resolution using tracers to partition effects of

model processes on heating and PV.

DIAMET WP B: Parameterisation of key processes

1. Detailed examination of convection parameterisation: • Can existing parameterisation be adapted to treat elevated

convection? • Choice of closure timescale for embedded convection • Decomposition of bulk mass flux detrainment

2. Quantify contribution of surface and boundary layer fluxes to

mesoscale PV anomalies and storm evolution 3. Measure microphysical properties and use them to derive latent

heating estimates and improve parameterisations.

DIAMET WP C: Predictability and DA for high resolution forecasts

1. Quantify forecast statistics for objectively identified (Hewson/Titley)

mesoscale features in the MOGREPS ensemble (Swinbank). Measure distances between forecasts in terms of feature tracks. Characterise dependence of precip forecast skill stratified by

mesoscale feature type. 2. Use short-range convection-resolving ensemble, perturbing

parameterisations (with and without stochastic terms) to disentangle model error from IC error.

3. Use ensemble forecasts to assess the nature of balance between

variables at high resolution and influence on forecast error stats.