Assimilation of Radar Reflectivity in high resolution NWP Lee Hawkness-Smith MetOffice@Reading RMetS High Resolution Data Assimilation 19 April 2013 © Crown copyright Met Office Data Assimilation Group, MetOffice@Reading • Collaboration with Universityy of Reading g • Research theme: Use of novel observations for nowcasting and NWP prediction of convective or urban-scale weather • See our posters! © Crown copyright Met Office Overview • Weather radar reflectivity observations • What are they and what do they tell us? • How many of them are there? • What quality issues are there? • Why assimilate radar reflectivity observations? • Assimilation methods • The Met Office approach © Crown copyright Met Office Radar reflectivity y observations What are they and what do they tell us? © Crown copyright Met Office How many y of them are there? 18 operational radars in the British Isles U tto 5 scans every 5 minutes Up i t outt tto 250 kkm Millions of observations in just minutes! © Crown copyright Met Office What q quality y issues are there? wet radome -NOISEP rti l beamf ll ng © Crown copyright Met Office Why assimilate them? Nowcasting: forecast hazardous weather and precipitation quantitatively and promptly ~ to T+6 within 15 minutes of data time © Crown copyright Met Office Nowcasting techniques Extrapolation Quick and simple technique What about orographic enhancement, mesoscale dynamics, dynamics etc. etc ? NWP More physically realistic modelling of the evolution of weather events Requires equ es high g spat spatial a a and d te temporal po a resolution eso ut o modelling and data assimilation, and therefore rapid collection, processing and dissemination of g data volumes – expensive! p large Merged • Current compromise solution © Crown copyright Met Office Data assimilation techniques q used for radar reflectivity • Latent heat nudging Rescale model latent heat profiles by the ratio of observation / model p precipitation p ratios • 1D-Var+3D/4D-Var Use a 1D variational method to generate temperature and humidityy increments for assimilation in 3D or 4D-Var • Incremental 4D-Var Minimise the differences between the model and observations g a simplified, p , linear model byy iterating • Full-fields 4D-Var Iterate a full non-linear forecast model in the minimisation – very expensive! p • Ensemble methods Use ensemble members to represent background error covariances © Crown copyright Met Office Latent heat nudging • P Proven to t be b effective ff ti att improving i i precipitation i it ti iin fi firstt ffew hours of forecast • Latent heat increments may not be dynamically consistent with analysis • Relationship between surface precipitation and latent heat release in model column may break down at high resolution • Does not use full 3D information from volume scans, only derived surface rainrates • Deriving surface rainrates leads to further errors • Must maintain another system © Crown copyright Met Office 1D-Var+3D/4D-Var 1D Var 3D/4D Var (indirect) • Avoid handling the non-linearities of radar reflectivity observations in the full 3D/4D-VAR • Double use of background information may reduce impact of observations and reinforce incorrect features in the model • Information loss as observations treated in a column © Crown copyright Met Office 4D-Var 4D Var (direct) • No a priori diagnostic adjustment of moisture or heating rate • Directly assimilate all observations together in a consistent framework • (Aim for) consistency in microphysics • It has worked well for satellite radiances • Requires a simple (for convergence) yet physically reasonable adjoint model d l – challenging h ll i ffor precipitation i it ti processes • How do we assimilate observations when the background has zero rain? © Crown copyright Met Office Ensemble methods • For pure ensemble methods, no linearisation or adjoint required • More suited for massively parallel computing • Limited ensemble size due to computational p constraints means localization required © Crown copyright Met Office The Met Office Approach • Two methods developed: • 1D-VAR+3D/4D-VAR indirect assimilation (Nicolas Gaussiat) • Direct 4D-VAR assimilation (Lee Hawkness-Smith, Cristina Charlton Perez) Charlton-Perez) • Make of use of mature Met Office VAR system © Crown copyright Met Office Direct assimilation of radar reflectivity in Met Office 4D-Var • Use Nowcasting Demonstration Project (NDP) system • Southern UK domain • 1.5 km fixed resolution UM, 3 km PF model grid • Hourly cycling © Crown copyright Met Office NWP p production process p (from a radar data assimilation scientist point of view) Other obs RadarNet Observation Processing System High quality radar data Processed obs, model equivalent at obs locations Previous Unified Model run Model dump MetDB Variational assimilation (VAR) Increments Unified Model Forecast run From Radar to customer ASAP © Crown copyright Met Office My focus Post processing Customers OPS: Processing g and QC Q with respect to model Nicolas Gaussiat • Filter obs using quality flags: clutter, speckle, partial beam blockage • Forward model from UM rainrate + increment: • Corrections for beam bending and earth curvature • Gaseous attenuation is simulated • No beam integration, attenuation from rain or forward modelling of ice yet • Calculate observation – background (O-B) values and use in quality control • Superobbing of O-B and thinning (to a scale the model can represent) • Output statistics of OPS runs to improve error specification and monitor radar performance © Crown copyright Met Office VAR: forward model and adjoint j links increments to total water to Zobs b - Zmodel d l • PF model: Added rainrate on each model level as a field Autoconversion-like term from ((diagnostic) g ) cloud water,, rain falls out in single g timestep (no evaporation) • Radar Reflectivity y operator: p Same operator as used in OPS Z [dBZ] = 10 log10(ZR + ZI) Currently only assimilate rain obs ZR = 180R7/4.67 Rain component: UM + PF increment Linearisation issues – may be better to assimilate something that scales with qr... Minimisation assumes Gaussian error statistics(!) → transform variables Accumulation may help Currently just use obs where model simulated reflectivity > radar noise level © Crown copyright Met Office Research areas for 4D-VAR direct assimilation • High Hi h quality lit observations b ti required i d – collaboration ll b ti b between t U University i it off Reading, Met Office Observations R&D and Met Office Data Assimilation • Control variables – hydrometeor control variables required or let hydrometeors spin up? • Need a good covariance model – challenging at convective scale • Optimum length of time window - balance between allowing evolution of background covariances and timescale of validity of linearity assumptions • How to handle large innovations, including zero rain • Representation of ice microphysics © Crown copyright Met Office Case study: y 09Z 5 April p 2011 Radar obs Background derived model Latent 4D-Var heat Reflectivity nudging Assimilation © Crown copyright Met Office
© Copyright 2026 Paperzz