Dust storm forecasting at the UK Met Office

© Crown copyright Met Office

Dust storm forecasting at the UK Met OfficeMalcolm E. Brooks1*, Kerry Day1, Bruce Ingleby2, Yaswant Pradhan1, David Walters1,1 Met Office, Exeter, UK

2 ECMWF, Reading, UK

Castellanetta Marina, Italy, June 2014


Global Model Forecasts

• N512 (~25km) resolution, 70 levels

• 4D VAR ensemble-hybrid data assimilation of wind, temperature, humidity etc.

• 4D Var assimilation of MODIS dust obs. over Land

• Soil Moisture assimilation uses ASCAT/Synop obs

• Dust advected with 2 bins

• Forecasts daily at 00Z and 12Z, runs for 144 hours

• N768 (~17km) resolution upgrade due July 2014

• N1024 (~10km) due in 2016(ish!)


Downscaling the Global capability to high resolution

• Global resolution converging on SAM• Implies SAM retirement• Does a higher resolution model work for dust?

Global (25km) SAM (12km) Afghan (4km)



• Global model drives an Afghan 4km ‘dynamical downscaler’• Initialised from global, for every forecast,

with global data at boundaries• No independent assimilation of obs



• Meant to ‘add detail’ to global forecasts. Does it?• It appears to do that – needs more detailed verification.



Consistent performance across global, SAM and 4km resolutions.


Using dust observations to initialise forecasts

• 12km LAMs will be retired in late 2014.

• Focus on improving the global model,• either in the forecast model or assimilation of dust observations


Assimilation of Dust Observations

Merged MODIS “DEEP BLUE” and standard AOD product:

• Near global coverage over a day (before filtering).

• Uses obs only over land.• Standard MODIS filtered by type.• All DEEPBLUE obs included.

• Results from ocean assimilation to come later.


Assimilation of Dust Observations:mean behaviour

• Test results of re-running global forecasting system for December 2011, into January 2012.

• Assimilating MODIS obs mostly adds dust• Esp. over Asia• Dust redistributed in Sahara

• Improves skill vs AERONET.

• Went operational in April 2013.


Upcoming developments - global resolution and dynamics

• Global model upgrade due in July 2014 includes:• New dynamical core:

• Improved solver, slight change of grid.

• Less diffusive, more energetic, as forecast evolves.• More expensive, but more scalable on many cores.

• Resolution upgrade from N512 (~25km) to N768 (~17km).• Physics upgrades to improve ‘weather’: surface T, cloud etc.• The most significant NWP upgrade at the Met Office in at least a

decade.• No direct impact on dust forecast,

• but does the dust forecast maintain skill?• Part of the ‘Global Atmosphere’ (GA) model development process.

• Current model is GA3.1, upgrading to GA6.1.



• Comparison of dust AOD from current operational and resolution/dynamics/physics upgrade (untuned).

• No major differences stand out.• Time mean AODs also similar.

Current (GA3.1) model GA6.1 model


• Long range forecast drift away from the DA analysis (forecast bias):• Slightly reduced in the GA6.1 model.

• Forecast model is slightly more consistent with the DA, and hence obs.

Current (GA3.1) model GA6.1 model



Skill scores vs AERONET:• Equitable Threat Score

• 0 – no skill• 1 – perfect model.

• L1.5 data, 1hr window

• Forecast shows skill, and GA6.1 neutral to slightly positive.

• Poor skill at low AOD events – non dust aerosol?

• Moderate dust events improved in GA6.1

• ETS always poor for rare events.


Upcoming developments - Dust interacting with radiation

• Proposed model upgrade for late 2014• Interactive dust used in radiation (instead of climatology).

• No change at analysis time.• Bias at T+120 broadly similar, with a dipole pattern over N. Africa.• With a slight reduction in biases over N. Africa bias pattern.• Small (positive) impact on dust and general forecast evolution.

• A reasonable dust climatology does most of the work.• Our dust has reflective optical properties (SSA = ~0.95 to ~0.97).


Upcoming developments - Include MODIS obs. over ocean

• Proposed model upgrade for late 2014

• Bellouin, N., Boucher, O., Haywood, J., and Reddy, M. S. (2005) Global estimate of aerosol direct radiative forcing from satellite measurements Nature, 2005, 438, 1138-1141.

• Jones, T. A., and Christopher, S. A. (2011) A reanalysis of MODIS fine mode fraction over ocean using OMI and daily GOCART simulations, Atmos. Chem. Phys., 11, 5805-5817, doi:10.5194/acp-11-5805-2011

• Includes MODIS observations over ocean, in specified regions:

• Non dust aerosol filtered using additional MODIS retrievals, using criteria:

• Fine Mode Fraction ≤ 0.4• Angstrom Exponent ≤ 0.5• Effective Radius > 1.0 μm• Mass Concentration ≥ 1.2×10−4 kg m-2

• AOD > 0.1 (still under review)



• Proposed model upgrade for late 2014• An example set of MODIS obs, after filtering, for a typical DA cycle.



• Comparison of dust AOD the upcoming GA6.1 model, and including MODIS over ocean (plus interactive dust).

• Dust is being added

in Saharan/other outflow.

GA6.1 model + MODIS ocean


• By improving the analysis, the forecast drift from analysis changes:• Highlights future model developments to improve (long range transport):

• Change fallspeeds? size distribution? Retune emissions? Soil properties in W. Africa?

GA6.1 model + MODIS ocean



MODIS Ocean ETS:• Improves all AODs at T+0• Skill score increase

persists to T+24• and to T+120,• throughout the forecast


Upcoming developments - Dust included in global ensemble

• The GA6.1 operational suite is now running, in parallel, for final testing of performance, timeliness and robustness.

• Dust included in the global ensemble when this parallel suite was set up.

• Global 12 member ensemble, twice daily.

• N400 (~30km) resolution, forecasts to T+144.

T+0, main run T+120, main run


T+0 T+120

• Individual N400 ensemble members, broadly comparable to the N768 main run.

Mai

n ru

nE

nsem

ble


• This is very much initial work – where do we go from here?• N400 dust configuration may need tuning.• Ensemble DA now an option for dust.

• Ensemble mean also broadly comparable to N768 main run.

• Ensemble Std. Dev. is interesting.

Mai

n ru

nE

ns. m

ean


Summary:

• Dust forecasting at the Met Office started with Local Area Models for defence applications.

• We have since moved to a global model, • which successfully drives dust in high resolution dynamical

downscalers.

• Global Model forecasts benefit from MODIS dust observations over land (merging standard and DEEPBLUE).

• Met Office to upgrade global resolution, dynamical core and model physics. Dust forecast performance is neutral to slightly improved.

• Using forecast dust interactively in model radiation gives a very small benefit (relative to a dust climatology).

• Assimilating filtered MODIS AOD over ocean gives a larger improvement in dust forecast skill.

• Dust now included in our Global Ensemble forecasts, but we are not sure what to do with it yet.


Questions?

• Dust AOD for 21Z, Thursday 6th June 2014, from 12Z Friday 30th May.


Questions?

• Dust AOD for 12Z, Thursday 6th June 2014, from 12Z Friday 30th May (GA6.1, parallel suite)


Questions?

• Ensemble mean dust AOD for 12Z, Thursday 6th June 2014, from 12Z Friday 30th May


Questions and answers


Dust Model Details

• NWP only: Verical flux partitioned to bins with prescribed emission size distribution


Met Office/Imperial College London AOD retrieval using SEVIRI (MSG):

• Uses differences in IR channels

• and a radiative transfer model, with 16 days of NWP model data to find a dust-free comparison.

• Produces hourly observations.

Assimilation of Dust Observations


Assimilation of Dust Observations: Assessment

• Comparing against AERONET observations.

• An ETS of 1 is a perfect forecast

• 0 has no skill – this is a hard score as the obs a daylight only, so have a shorter time window than precip verification.

• Dust assimilation gives a large increase in skill

• Including SEVIRI is not quite right yet.


Upcoming developments - Dust interacting with radiation

• Proposed model upgrade for late 2014• Interactive dust used in radiation (instead of climatology).• NWP index: an internal metric of large scale global forecast

performance• Interactive dust has a small positive impact.


Dust emission control from soil properties

• Libyan coast dust: regular occurrence in the model, during the 2011 Air campaign• Intense scrutiny of forecasts and SEVIRI pink imagery during this period.• Benghazi was a source, but the coastal dust did not happen!


• Current combination of constraints and data do not give ‘optimal’ results…

• Need to look to other soil datasets


)6),2.0max(6.13(10 cFHG

Total Vertical flux: Gillette (1979)



Reprocess current HWSD data?

Even old datasets like Zobler 1degree look useful…

Recent datasets like GMINER30 have a lot more detail. Are they useful?

Geomorphology from Digital Elevation Models.



A preferential source map:

• Ginoux (widely used)

• Marticorena ’97:

• or Bullard ’11:

Both global coverage and high resolution is required.


Dry river beds, lakebeds, Wadis

• Our emission scheme lacks alluvial sediement…

e.g. Sistan Basin – often dry

Dust storm forecasting at the UK Met Office

Documents