DA needs for nowcasting
19 people
- specific development/improvement
Luc – observation correlations
Sue studies have shown very short range
Bias – may be quality control issue
What do we know about R, dependent on resolution – need greater than 2-3 times beam width. At 1-2 km scale , beam width – what does cell mean, reflectivity. Strong gradients – how much does it matter?
Beam attenuation issues for reflectivity
C,X band severe problem – OU build into operator
Iszka X.BAND kdp
Observation based correction or use forward operator? 1+3-D beam along beam?
Need to do it quickly so may need short-cuts for obs processing ahead of time
Fall speed not a problem at low beam angles. But are seeing a bias
If observation operator very nonlinear – big problem if bad first guess,
Can we make the problem more linear? MF – seek around area so get more sensitivity/better first guess. Cycling should improve first guess
Cycling with ongoing MCS may be harder – fronts more linear – multiple scale interactions – with diabatic DFI initialization. What are the drivers? Cold pools/outflow – what scales?
RR – cold pools come from precip. May be too hard to get cooling initialized – what obs tell us about that.
Better use of radial velocity – good background term
Thinning, superobbing, observation errors
Radar refractivity – too close to radar –
Need to understand the problem and what will help with the problem
Cold pool very sensitive to microphysics timescale mismatch, CASA network low level wind network very helpful to correct gust front location, 5min mesonet data helps
Clear air Doppler can help with convergence lines – Taiwan threshold hides clear air winds – sea clutter contamination, mountain contamination
MF get more positive impact by removing clear air echoes –but don’t give up
WRF has problem dealing with clear air – VDRAS OK so may be technique
Ground clutter – lower velocity to convergence to need very good quality control – no motivation of no client – need to talk to radar data providers
C-band insects are below the noise level. Need to improve the signal processing
Wrong environment for DA nowcasting – how to modify the forcing? Need to deal with synoptic scale
Need a 4D DA system to get large scale as well as convective/storm scale
Need to improve the low level analysis – CAPE, shear
Need to look at whole 3D-structure and use satellite data
Can 4D-DA modify the environment – needs continuous assimilation
Need big profiler networks – deep winds
Problem with driving model – need larger domain
Problem with phase errors – how to correct, radar data cannot correct phase error.
Need inversions all weather – downward pointing lidar – give size of inversion
Thermodynamic profiles
Modelling of dynamical balances – can we simplify them? Covariances derived from ensembles
Weak constraint eg mass-continuity
Balance idea comes from large scale – minimising time tendancy – now need right time tendancy – may be strong pressure tendancy. mimic the synoptic scale still. 2,5km model run to developed convection balance diabatic forcing and vertical velocity – otherwise no good relationships
Freezing mechanism – need to get correct phase to prevent latent heat release, need model close to atmospheric balance – DFI does some of this
Model errors – need to improve models
3 suggestions for international collaboration
Hymex
Isztar – concern about how quickly loses the impact of the data – model lives in its own world
need good flow dependent background error covariances to get good JB eg ensemble based methods or enkf
Jenny thinks BG less important for radar da assimilation
How long will radar data last – predictability limit – hurricanes lasts longer
Try to reach predictability limit by improving model and initial conditions
BG 100km scale winds impt, 1000km height field
Radar impact to 12hours at 20-80km scale with 3km model
We should worry about process verification not user based verification
International collaboration – common test cases? TOMAX field experiment? Japan – need observation database and cleaned data – problem with forward models and observation processing
Single ob experiments, same environment, need the ensemble as well
Some basic states
Eg squall line/MCS similar situation to understand process eg low level cold pool
Identical twin experiments OSSE of squall line
WGNE – large eddies
Issues
Observation errors and correlations
Radar observation operators – eg beam width
MCS assimilation – cold pools
Thinning/superobbong
Background errors
Wrong environment
Look at full 3d structure
Balance
Model errors
Intercomparison – single ob, OSSE of squall line etc