Preprocessing of radar data

Lesley De Cruz*, Annelies Duerinckx

Royal Meteorological Institute of Belgium, Ringlaan 3, 1180 Brussels *E-mail: Lesley.DeCruz@Meteo.be

Radar data assimilation

Assimilation of GNSS and radar data in ALARO cy38t1 at RMIB

• Preprocessing of radar data is done with the package rmiradlib, a local fork of wradlib(maintained by E. Goudenhoofdt, Department Observations, RMIB).

Dynamical clutter identification algorithms: - precipitation structure - correlation with satellite images (clouds)Static (monthly updated) clutter identification: - static clutter map - beam blockage map

Preprocessing of GNSS data

• The Royal Observatory of Belgium (ROB) [4] provides hourly updated Zenith Tropospheric Delay (ZTD) estimations, within the framework of E-GVAP.• Multi-GNSS ZTDs are estimated if available: improves reliability and stability

• E-GVAP ascii files are converted straightforwardly into OBSOUL format, and assimilated as SYNOP data• Ongoing work: whitelisting, bias correction, height-difference correction

Conclusions and outlook

GNSS DA set-up went more smoothly: • Canonical format (E-GVAP) which can easily be converted into OBSOUL.• ZTDs are treated as SYNOP data.• First experiments already show an improvement for short-range forecasts.Next steps:• Variational bias correction• Spatial thinning, as the Belgian GNSS receiver network is quite dense.• Validation of precipitation structure; test longer periods, more test cases.

Radar DA is not straightforward to set up:• "Canonical" ODIM format: but still needs a lot of custom work to use in DA.• Key information is missing from ODIM (e.g. radar sensitivity, MDS): format is targeted at nowcasting rather than NWP ⇒ Opportunity for next version of ODIM?• Some technical issues remain, work on this is still ongoing• Changes to the code are required for radar DA to work locally: e.g. blacklisting, rain thresholds, etc.

Introduction

Goal: improve initialization of moisture variables in our LAM in order to obtain better precipitation and cloud forecasts, especially for severe weather eventsCurrent operational set-up: ALARO-0, cy38t1, 4km horizontal resolution, 46 model levels, 180s timestep ⇒ Data assimilation of moisture-related observations with a high temporal and spatial resolution:

Reflectivities and radial velocities from precipitation radars & Zenith Tropospheric Delay (ZTD) estimations from GNSS receivers

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

● ●

●

●●

●

●

●

●●

●

●

●

●

●

●

●

● ●

●

●

●●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●●

●

●

ZTD data assimilation

First results

Case study: Pentecost storm (June 7-8-9 2014): over €500M in damage claims• Preliminary results for assimilation of SYNOP data + non-bias corrected ZTDs: improved RMSE and bias of the 2m relative humidity for short (<9h) forecast range (gnssDA_rt uses the "realtime", and gnssDA the "hourly" stations shown above) compared to SYNOP data only.

• Impact of GNSS DA on pseudosoundings & total precipitable water (TPW):

Forecast time since 0000 UTC

Rel

ativ

e H

umid

ity (%

)

0 3 6 9 12

1015

20 synopDAgnssDAgnssDA_rt

2m Relative Humidity RMSE (01−09 June 2014) run 0

Forecast time since 0000 UTC

Rel

ativ

e H

umid

ity (%

)

0 3 6 9 12

−10

−50

510 synopDA

gnssDAgnssDA_rt

2m Relative Humidity BIAS (01−09 June 2014) run 0

• Conversion of ODIM to MF-BUFR with extended ConRad [1] tool.

• Results in MF-BUFR format: (a) "raw" dBz values; (b) categorized in rainy (red) and non-rainy (blue) data; (c) categorized, with dynamic quality flags (teal), (d) categorized, with dynamic and static quality flags (teal); note e.g. the beam blockage in the NNW direction.

Strategy: 1D + 3D-Var [2]: • First, use a Bayesian 1D approach to determine a humidity profile for each retained reflectivity observation: this is constructed as a a linear combination of nearby vertical columns, weighted by the departures between observed and simulated reflectivities (observation operator: reflectivity simulator GPROF [3]).• Secondly, assimilate the constructed humidity profiles as if they were observations, using 3D-Var.

Care should be taken how to treat non-rainy pixels and pixels at the MDS threshold level!

Results: work in progress

MD

S (

dBZ

)

r

noise

signal rain

no rain

!!!

• 3D-Var, using the ZTD observation operator [5, 6]: ∫(index of refractivity).dz over the model column above the GNSS receiver ⇒ function of pressure, temperature and water vapor pressure• Constant error (will be removed through bias correction): part of the atmosphere above the highest model level.

[1] Salomonsen, M., Grønsleth, M. S. et al., tech. report, 2012.[2] Wattrelot, E., et al., Mon. Wea. Rev., 142(5), 1852-1873, 2014[3] Caumont, O., et al., J. Atmos. Oceanic Technol., 23(8), 1049-1067, 2006.

[4] Bruyninx, C., et al., EUREF Permanent Network, http://www.epncb.oma.be.[5] Poli, P., et al., J. Geophys. Res., 112(D6), 1984–2012, 2007.[6] Boniface, K., et al., Annales Geophysicae, Vol. 27, 27–35, 2009.

RMI

−30

−20

−10

0

10

20

30

40

50

201285321

32

32

28

28

24

24

20

20

16

16

12

12

8

81000

700

500

400

300

200

150

100

P (h

Pa)

gnssDAsynopDA

−3

−2.5

−2.0

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5

2.0

2.5

3

TPW(gnssDA)-TWP(synopDA)

−0.005

−0.0045

−0.0040

−0.0035

−0.0030

−0.0025

−0.0020

−0.0015

−0.0010

−0.0005

0.0000

0.0005

0.001

S046HUMI.SPECIFI2012/3/5 z0:0 Uninitialized

⇒spuriousdrying?

●

●●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●●

●

●

●

"hourly" stations "realtime" stations

Acknowledgements. Thanks to Eric Pottiaux (ROB) for the ZTD data, Edouard Goudenhoofdt (RMIB) for the radar data and clutter identification,Máté Mile (OMSZ), Eric Wattrelot (MF), Loïk Berre (MF), Xin Yan (ZAMG), Alena Trojáková (CHMI), Alex Deckmyn (RMIB), Daan Degrauwe (RMIB), and many others for help with the data assimilation setup.