240 likes | 400 Views
Mesoscale NWP developments. Jeanette Onvlee WGNE meeting 2014, Melbourne 11 March, 2014. Outline. Data assimilation Use of fine scale observations Towards more flow-dependent algorithms The forecast model Grey zone issues Towards sub-km scales Surface characterization
E N D
Mesoscale NWP developments Jeanette Onvlee WGNE meeting 2014, Melbourne 11 March, 2014
Outline • Data assimilation • Use of fine scale observations • Towards more flow-dependent algorithms • The forecast model • Grey zone issues • Towards sub-km scales • Surface characterization • Stable boundary layer • Probabilistic forecasting • Typhoon forecasting WGNE 2014, 20140311
Towards optimal use of hi-res observations Obs of interest: radar, GPS, high-resolution sounders, aircraft obs, surface (satellite) data, boundary layer remote sensing. Search for useful new observations from non-meteorological networks Impact assessment: OSE’s, increasingly also by routine monitoring of tools familiar to global modelling community: DFS, FSO, … Emphasis shifting towards: how to get highest/longest impact, with e.g.: • QC and bias correction • Cloud assimilation • Rapid update/rapid refresh modes and spinup studies • Scales to apply; blending with large scales from the nesting model? • Impact of model bias • Introduction of more advanced (flow-dependent) DA methods WGNE 2014, 20140311
Mesoscale data assimilation algorithms • Towards more sophisticated mesoscale data assimilation algorithms: • 4D-VAR (incl methods how to make faster/more scalable; e.g. Gaussian quadrature) • Many flavours of ensemble data assimilation, for use in both DA and LAM EPS: combinations of EnKF/(L)ETKF with 3-/4D ensVar • New methods to address non-additive (displacement) errors from gaming industry: field alignment / image warping techniques WGNE 2014, 20140311
Grey zone experiment:Cold air outbreak, 31st January 2010 WGNE 2014, 20140311
Cold air outbreak; WGNE grey-zone test; 2-km schemes’ with convection ALARO NCAR MODIS observation Unified Model (tamed conv.) NCEP AROME WGNE 2014, 20140311
Cold air outbreak; WGNE grey-zone test; 2-km schemes’ without convection ALARO NCAR MODIS observation Unified Model NCEP AROME (no sh. cv.) WGNE 2014, 20140311
Drivers towards hectometric resolution models: 1. complex terrain FROST-14: (sub-)km scale models for Sochi area
Tornadoes Occurred on 6th May 2012 12:53 12:51 12:49 12:47 12:45 12:43 12:41 12:39 12:37 12:35 Drivers towards hectometric scale models: 2. small-scale severe weather phenomena 住家の屋根はぎとられる 平沢 住家倒壊 北条 被害分布の幅:約500m 山木 被害分布の長さ:約17km Rainfall region that caused Tornado テクノパーク大穂 吉沼 つくば市 常総市 新石下 大沢新田 http://www.jma.go.jp/jma/press/1205/11c/120511tsukuba_tornado.pdf 非住家の屋根が東北東に約50m飛散 http://mainichi.jp
Positions of vortices more than 0.1(1/s) Observed positions 10 members (83%)duration: 6-36min ・Positions anddurations differed among the ensemble members. Tornados occurred in three areas, which were the same as the observations though they were shifted northward by 10 km. Courtesy: K. Saito
Driver towards hectometric solutions nr 3: Needs from aviation - AROME-airport system AROME hourly assimilation at 2.5km using AROME-France forecast as first guess (on red domain) : two additional wind profilers AROME forecast at 500m performed on green domain to produce boundary conditions to a Wake-Vortexprediction model RMSE of wind force for: AROME-France (green) AROME-airport 2.5 km (red) AROME-airport 500m (blue) X-axis is the forecast range in hours. Areas covered by the models, AROME-airport 2.5km domain (red) AROME-airport 500m domain (green) Zoom of an AROME-airport 500m forecast, vectors show wind direction and force, shaded areas is orography.
High resolution simulations for the COPE experiment, summer 2013 Model setup – UM vn8.2 PS32 UKV – 1.5km grid length, 70 levels, 2D subgrid turbulence scheme, BL mixing in vertical. 500m model – 500x400 km 200m model – 300x200 km 100m model – 150x100 km High res models: 140 vertical levels, 3D subgrid turbulence scheme, RHcrit is 0.97 (0.91) in 1st few layers decreasing smoothly to 0.9 (0.8) at ~3.5km. Also: research on scale-sensitive shallow convection formulation Set of nested models. Courtesy: M. Bush
3rd August – convergence line • 03rd August case had a nice line of showers down the centre of the peninsula which the 500m model captured quite nicely. • The higher resolution models have lots of little showers • Cells appear to get smaller as grid length is reduced Similar features are more generally observed also in other models. Also: more problems with numerical stability, need for significantly higher vertical resolution at these scales Courtesy: M. Bush
Higher resolution, scientific open questions • “Some open questions for hectometric resolutions • Do we need 3D features ? (physics, turbulence) : Probably not at 500m, our models have other uncertainties that must overwrite the refinement gain of 3D parameterizations. • Which shallow convection scheme do we need ? • Coupling frequency, coupling zone size. • Physiographic fields resolution. • Volume of the data to treat. • For finer resolutions (LES mode) • Issues like 3D parameterizations that might have been avoided for 500m resolutions should become problematic. • Which turbulence parameters, turbulence mixing length.” • Need for experimentation to clarify how best to handle shallow convection/turbulence grey zonesin parametrizations! Pier Siebesma: Extend existing grey zone project to higher resolutions? Courtesy: L. Auger WGNE 2014, 20140311
Are we trying to run before we can walk? • We are rushing towards hectometric scale resolutions, while we have only just gotten over the grey zone for deep convection • At present resolutions the convection-permitting models still suffer from several (severe) systematic errors, which will not magically disappear on finer scales; deal with them first?!? • We also have only started to e.g. get aerosols in and disentangle the complex interplay between radiation, clouds, microphysics and aerosols; do this first?!? • But under pressure of the driving forces, we enhance complexity by adding two new grey zone problems of shallow convection representation and (3D?) turbulence • Surely we cannot ignore the issue of very limited predictability at these small scales? Accompany this VHR experimentation with VHR ensemble information, if so, how? WGNE 2014, 20140311
Need for 3D-radiation? Tilted column modelling Courtesy: K. Nielsen WGNE 2014, 20140311
Physiography datasets • Detailed and high-quality datasets essential for surface characterization in VHR modelling. However, for many well-known (global) datasets • quality is generally NOT uniform over the dataset’s geographical area • some gross errors exist which can have significant local meteorological impact (and you do not want to attribute this erroneously to forecast model error!) • => Look at them critically for your own domain! Sand Clay
Stable boundary layer: • Preparations for GABLS-4 experiment • Testing new parametrizations 1D test with EFB closure: Comparison for GABLS1 (e.g. Holtslag et al, 2003): Energy- & flux-budget (EFB) turbulence closure modeling Energy- & flux-budget (EFB) turbulence closure modeling Proposed by Zilitinkevich et al. (2013): Proposed by Zilitinkevich et al. (2013): • Budget equations for basic second moments: two energies, turbulent kinetic energy (TKE) and turbulent potential energy (TPE), and vertical turbulent fluxes of momentum and potential temperature • New prognostic equation for the turbulent dissipation scale • Budget equations for basic second moments: two energies, turbulent kinetic energy (TKE) and turbulent potential energy (TPE), and vertical turbulent fluxes of momentum and potential temperature • New prognostic equation for the turbulent dissipation scale Comparison with GABLS1 (e.g. Holtslag et al, 2003): Comparison with GABLS1 (e.g. Holtslag et al, 2003): Energy- & flux-budget (EFB) turbulence closure modeling Energy- & flux-budget (EFB) turbulence closure modeling EFB closure mean profiles of wind speed and potential temperature for ninth hour compared with LES and pre-GABLS results from other models EFB closure mean profiles of wind speed and potential temperature for ninth hour compared with LES and pre-GABLS results from other models • No tuning of empirical constants • Very little sensitivity to spatial resolution • Works well with only one prognostic equation (TKE) • No tuning of empirical constants • Very little sensitivity to spatial resolution • Works well with only one prognostic equation (TKE) Time series for friction velocity and Obukhov length Time series for friction velocity and Obukhov length
Probabilistic forecasting on convection-permitting scales • # of convection-permitting LAM EPS growing • Increasing link with ensemble DA • Experimentation with different physics and surface perturbation types • Stochastic physics / SPPT • Multi-physics schemes • Multi-model schemes • Different surface quantities • Added value wrt deterministic model • Still quite underdispersive: are we capturing the mechanisms of uncertainty correctly? • Calibration important but often not in place • Much effort sofare spent in “getting EPS up and running”, is it time to start paying more attention to intrinsic (limitations of) predictability?? WGNE 2014, 20140311
TIGGE-LAM: GEOWOW archive at ECMWF Data provider progress status information http://tigge.ecmwf.int/tigge/d/inspect/tigge/ tigge/monitoring/lam/prod Courtesy: T. Pacagnella WGNE 2014, 20140311
WMO-Typhoon Landfall FDP (TLFDP) Products from limited area models (LAMs) are widely used by operational forecast agencies in the western North Pacific region, but their application is generally limited to the host agency. For track forecasts, TLFDP verification has shown that it is difficult for LAM’s to add value to the global models. But for intensity / precipitation / wind they might. Yu H., S.T. Chan, B. Brown, and Coauthors, 2012, TCRR
Specific objectives of TM-MWFR proposal: set up guidelines and establish a data center for international exchange of TC forecast products from NWP models and EPSs, both global and limited-area (preferably including convection-permitting); evaluate the participating models' performance on the prediction of TC intensity and precipitation up to 48 hours; understand the physical mechanisms for the performance of different LAM models/EPSs; understand the performance and impact of different data assimilation systems; understand the impact of different observing systems on TC forecast quality (e.g. TC bogus, targetted observations); propose skillful multi-model consensus forecast scheme for TC intensity and precipitation up to 48 hours, both deterministic and probabilistic, which should be region-dependent; find out the minimum forecast range that the model output becomes useful; enhance the application of advanced LAMs and LAM-EPSs in operational forecast agencies globally. WGNE 2014, 20140311