mark naylor data assimilation, nwp

25
NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 1 Mark Naylor Data Assimilation, NWP NAE 4D-Var – Testing and Issues EWGLAM/SRNWP meeting Zurich 9 th -12 th October 2006

Upload: aimee

Post on 11-Jan-2016

36 views

Category:

Documents


1 download

DESCRIPTION

NAE 4D-Var – Testing and Issues EWGLAM/SRNWP meeting Zurich 9 th -12 th October 2006. Mark Naylor Data Assimilation, NWP. NAE 4DVar. The Met Office’s 4DVar NAE went operational on the 14 th March 2006 after much testing and tuning. We will present results from:- - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 1

Mark NaylorData Assimilation, NWP

NAE 4D-Var – Testing and IssuesEWGLAM/SRNWP meeting Zurich 9th-12th October 2006

Page 2: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 2

NAE 4DVar

The Met Office’s 4DVar NAE went operational on the 14th March 2006 after much testing and tuning. We will present results from:-

Two season’s trials (Spring and Summer 2005)Pre-operational real-time trials – 5 weeks in Dec 2005

Problems, particularly with screen temperature (T2m) scores.

Future development plans

Page 3: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 3

Spring 2005 Trial

6th – 20th March

Mixed conditions:

anticyclonic with widespread frost

lows bringing gale force winds

very mild south-westerlies

Page 4: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 4

Spring 2005 Trial

6th – 20th March

The NWP UK index is used to asses skill in the NAE and consists of verification against screen temperature, visibility, wind, cloud

amount and precipitation amount.

NWP UK Index (NAE area): +1.5%

NWP UK Index (UK Mes area): +2.6%

NWP UK Index (UK stations): +1.9%

Page 5: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 5

Spring 2005 Trial - Pressure

Page 6: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 6

Spring 2005 Trial – Screen Temperature

Page 7: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 7

Spring 2005 Trial

Summary

• good positive impacts

• wind and pressure particularly good

• hint of improved balance at T+0

• detriment in screen temperature fit up to T+12

Page 8: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 8

Summer 2005 Trial

18th June – 2nd July

Week 1 anticyclonic with thunderstorms

Week 2 anticyclonic and more mobile weather with rain over UK

Page 9: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 9

Summer 2005 Trial

18th June – 2nd July

NWP UK Index (NAE area): +2.0%

NWP UK Index (UK Mes area): +3.7%

NWP UK Index (UK stations): +5.4%

Page 10: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 10

Summer 2005 Trial - Pressure

Page 11: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 11

Summer 2005 Trial – Screen Temperature

Page 12: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 12

Summer 2005 Trial summer rainfall

t+9 3DVAR

t+9 4DVAR radar

Page 13: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 13

Summer Trial 2005

Summary

• again good scores, especially ppn

• screen level temperature ok after T+0

• better balance in analysis

Page 14: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 14

Real-time trial setup

4DVAR real-time trial (Dec 2005) final settings

12km UM /36km PF Visibility Assimilation on ‘operational’ 3-hourly 3D-Var control 6 weeks real-time from Dec 2005 4 forecasts per day

4DVar gave consistently positive results, similar to the seasonal trials

Page 15: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 15

Equitable Threat Score precipitation ETS

3D-Var

4D-Var

Page 16: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 16

Screen Temperature Scores

3D-Var

4D-Var

Page 17: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 17

Screen Temperature investigation

An investigation into the poor initial screen temperature scores, involving experimental reruns in March 2006, was undertaken.

The 1st week of March 2006 included some especially poor screen temperature forecasts.

Page 18: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 18

Single reruns

18Z on 2nd March was particularly bad so was rerun and compared with experiments including:-

PF Persistence

No Screen Temperature obs

3DVar

No T2m obs after T+0 (i.e. obs only from T-3 to T+0)

Halving period of all obs (i.e. obs only from T-90m to T+90m)

Page 19: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 19

Screen Temperature VER results

The UM ran to T+6 and RMS’s produced:-

Operational NAE (3DVar 3-hourly) is still better than 4DVar Control at T+0 RMS fit.

PF Persistence has a lower RMS fit than Control 4DVar at T+0 !

Using T obs only upto T+0 increased RMS.

3DVar (with 6 hours obs) is similar to 3DVar

Using 3-hours of all obs halves the gap

Page 20: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 20

Issues with Screen Temperature experiments

PF persistence has a lower RMS error to obs at T+0 than 4DVAR.

Removing the 2nd half of the time-window for screen temperature obs increases the RMS error.

Halving the number of all the obs (to between T-90m to T+90m) decreases the T2m RMS but has a major detriment on scores for other variables (especially pressure).

Is the PF model dealing correctly with Surface Temperature?

Stats reflect only one case, but 1-week reruns indicate similar results.

Page 21: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 21

Linearisation tests

In PF-linearisation tests low-level theta in persistence also beats control for the first 2 hours (8 timesteps):-

4DVar Persistence

After the first 2 hours the PF model has a higher correlation. Why isn’t this feeding into the analysis?

Page 22: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 22

Damping Coefficient

Theta damping coefficient from analysis increment.

The Theta Damping coefficient is the ratio the average PF model increment size to the average UM increment size

In the lower levels, the Theta PF increment is up to 20% larger than the UM increment

Can we improve the physics to remedy this feature?

Page 23: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 23

Future Improvements

New PF physics package is expected later this year. Includes better boundary layer mixing and PF

convection.

We wait to see how this will influence the screen temperature near T+0.

Page 24: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 24

Recent Improvement

The Screen temperature skill in Summer was improved through soil moisture modifications

We now get soil moisture (outside the UK) from the Global Model nudging scheme replacing climatology

Page 25: Mark Naylor Data Assimilation, NWP

NAE 4DVAR Oct 2006 © Crown copyright 2006 Page 25

Questions?

Any questions?........