Download presentation
Presentation is loading. Please wait.
Published bySophie Owen Modified over 9 years ago
1
Page 1 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Andrew Lorenc Head of Data Assimilation & Ensembles Numerical Weather Prediction Met Office, UK Data Assimilation – WGNE etc. WOAP August 2006
2
Page 2 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Data Assimilation – Summary (2005) Growing field: 0 increase in Met Office R&D effort (1999-2005) 7%/year researchers (WMO DA Symposia, 1999-2005) 100%/year computer power (Met Office, 2000-5) 115%/year operational data volume (Met Office, 2000-6) NWP 4D-Var is most popular (for those who can afford it) Fitting model to observations DA for ObsSystem cal-val well established DA for model development increasing Assimilation products Good fields are sufficient for some users More information in ob-model assimilation diagnostics Problems & Issues
3
Page 3 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Problems and Issues Management: Data volume & diversity System complexity Resources Collaboration between operations & research Scientific: Error modelling Efficient use of all obs, allowing for all errors Representing uncertainty Nonlinear models, non-Gaussian errors
4
Page 4 Andrew Lorenc WOAP 2006 © Crown copyright 2006 WGNE: extracts from TOR development of atmospheric models for weather prediction and climate studies atmospheric physics processes, boundary layer processes and land surface processes in models variability and predictability data assimilation for numerical weather and climate predictions, and estimation of derived climatological quantities exchange of information through publications, workshops and meetings
5
Page 5 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Steady improvement in forecasts
6
Page 6 Andrew Lorenc WOAP 2006 © Crown copyright 2006
7
Page 7 Andrew Lorenc WOAP 2006 © Crown copyright 2006
8
Page 8 Andrew Lorenc WOAP 2006 © Crown copyright 2006
9
Page 9 Andrew Lorenc WOAP 2006 © Crown copyright 2006 S.Hem. Z500 T+24 rms v analyses 4D-Var 3DVar+ATOVS ATOVS Model+Cov Radiances+Cov NOAA16+AMSU-B FGAT+Cov 2nd ATOVS New stats 12hr 4D-Var Higher res.
10
Page 10 Andrew Lorenc WOAP 2006 © Crown copyright 2006 N.Hem. Z500 T+24 rms v analyses 4D-Var 3DVar+ATOVS ATOVS Model+Cov Radiances+Cov FGAT+Cov NOAA16+AMSU-B 12hr 4D-Var New stats 2nd ATOVS Higher res.
11
Page 11 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Relative scores 2003-5 + dates of 4D-Var implementation 4D-Var implementation
12
Page 12 Andrew Lorenc WOAP 2006 © Crown copyright 2006 THORPEX – DA OS WG (Mar’06) ATReC2003: value of targeted obs is ~twice normal, but overall impact is marginal & does not justify cost of deploying targeted obs on demand. It remains important to make significant progress on the assimilation of satellite data. Model error needs to be taken into account, but it is not obvious how. Links with multi- model ensemble research in TIGGE should help.
13
Page 13 Andrew Lorenc WOAP 2006 © Crown copyright 2006 ECMWF/GEO Workshop on Atmospheric Reanalysis (June’06) reported by Adrian “how to determine and convey to users information on uncertainty and problems is paramount” “many users want measures of expected accuracy or uncertainty”
14
Page 14 Andrew Lorenc WOAP 2006 © Crown copyright 2006 DA can estimate errors that are being modelled (1) variances OI gave analysis error variance for resolved random errors only VAR can approximate this (via Hessian) Deterministic ensemble methods (EnSRF, ETKF...) use same eqns as OI Stochastic ensemble methods (EnKF) rely on modelling of error distn – perturbed obs All these methods underestimate total error – ad hoc “inflation” to fit (o-b) 2 statistics is needed
15
Page 15 Andrew Lorenc WOAP 2006 © Crown copyright 2006 DA can estimate errors that are being modelled (2) biases Observation & model bias correction methods are being developed – could in principle estimate errors in determined bias Above methods are often described as dealing with model error. In fact they are assuming that a different model (stochastic, with a few unknown parameters) is perfect. Few methods consider “unknown unknowns”:- multi-model ensembles, “shadowing”. Obtaining reliable total error estimates from a single DA system will be difficult, requiring modelling of all significant error sources in DA, model & obs.
16
Page 16 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Recommendations from WOAP DA Report: Collate a list of groups with capability and interest to develop DA methods for fields of interest to WCRP but not currently part of established systems Encourage them to make their results system (near- real-time analyses, seasonal climatologies, or extended re-analyses) available to the established centres, as part of a loosely coupled system. Encourage the established centres to support these new developments: make available necessary output, validate and test, support bids WOAP: fostering the development of data assimilation techniques for components of the earth system which are not part of operational systems.
17
Page 17 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Recommendations from WOAP DA Report: Using DA in model development. Comparing analyses with research obs globally and mesoscale. Climate models validated in assimilation mode. Persuading operational centres to develop and maintain their DA systems in a way that they can be used for climate research such as re-analyses. (USA) Promoting coupled land-atmosphere assimilation. Focus attention on atmospheric model developments needed to help coupled modelling. How to improve models to better fit fluxes deduced from coupled ocean models? WGNE: fostering the use of data assimilation to benefit climate research.
18
Page 18 Andrew Lorenc WOAP 2006 © Crown copyright 2006 Recommendations from WOAP DA Report: GSOP should concentrate initially on all aspects of ocean re-analysis but should, in parallel, begin to approach the coupled problem involving ocean, atmosphere and sea ice. GSOP. Operational centres are focussing only on analyses for Seasonal-Interannual forecasting. Not yet a comparable sustained reanalysis activity addressing Dec-Cen and ACC prediction problems, (only in research). Nor adequate support of the general community.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.