Assimilation of cloudy infrared satellite observations: The Met Office perspective Ed Pavelin, Met Office International Symposium on Data Assimilation 2014, Munich
Contents This presentation covers the following areas Cloudy IR radiances: Introduction Current approaches to cloudy IR assimilation Use of cloudy IR radiances at the Met Office Improvements to the Met Office scheme Towards the assimilation of cloud information Hope to stimulate discussion!
Cloudy IR radiances: Radiative effect Clear Cloud signal ~ >10K Temperature O-B ~ 0.1K Cloudy 14.8 m 14 m 9.6 m 7.2 m 4.5 m 4.1 m LW T Sounding BT ~ >10K (Cloudy Clear) 14.8 m 14 m 9.6 m 7.2 m 4.5 m 4.1 m Need: Accurate forward model Knowledge of background error covariances for cloud Very challenging to assimilate cloudy radiances without degrading T/q analysis!
Height Height Height Cloudy IR radiance assimilation 1. Clear scenes only 2. Clear channels only 3. Grey cloud methods Cloud Cloud A. Improved forward models Jacobian B. Variational all-sky schemes Jacobian C. Ensemble DA methods
1D-Var Grey Cloud Analysis (Met Office scheme) Aim: To extract T and q information in the presence of cloud Analyse two cloud variables: n eff, p ctp Simple cloud model: Single-level grey cloud L n neff Lc (1 neff ) L0 eff n Overcast component Clear-sky component Retrieve n eff, p ctp in 1D-Var to fit observed radiances (No direct assimilation of cloud information)
Problems with grey cloud model (results from 1D-Var simulation) Severely degraded analysis below cloud Errors in cloud analysis aliasing into T and q
Limitations in cloud model In many cases, 1D-Var cloud model is unrealistic Not (generally) single-level grey cloud Cloud is generally multi-level, 3D Leads to large biases below cloud top Effect of inhomogeneous cloud Solution: Remove channels most likely to be poorly modelled (from Prates et al., 2014) Simple automatic channel selection: Reject all channels peaking below retrieved cloud top 10% of weighting function area allowed below cloud top Channel selection carried out for each sounding
Met Office cloudy IR scheme CTP CF CTP, CF Retrieve cloud parameters in 1D-Var Using RTTOV: Single-level grey cloud Choose channels with <10% sensitivity below cloud top 4D-Var Pass cloudy radiances, retrieved CTP and CF to 4D-Var (Bias corrections calculated using clear sky scenes only)
Example cloudy weighting functions ( B i / T j ) Mid-level cloud Use 26 of 94 channels Retrieved CTP Low cloud Use 67 of 94 channels
Simulated 1D-Var analysis errors: Mid-level cloud cases Retrieval Retrieval Background Background CTP CTP From: Pavelin, English and Eyre, 2008, Q. J. Roy. Met. Soc.
Coverage: Clear IASI No. of channels passing 1D-Var QC
Coverage: Cloudy IASI No. of channels passing 1D-Var QC
Retrieved effective cloud fraction
Forecast impact (Cloudy AIRS+IASI vs Clear) Bad Bad Good Good Impact ~ double clear-sky IASI
Grey cloud scheme: Next steps System is very conservative Fewer low-peaking channels used than in previous system Over-detection of low cloud in clear sky Use more low-peaking channels Variable observation errors More advanced cloud analysis Improved forward model introduce extra layers Towards the assimilation of cloud information
Estimating grey cloud radiative transfer errors Generate simulated brightness temperatures Use 13495 diverse profiles from ECMWF/NWPSAF profile dataset (Chevallier, 2001) Simulate BTs using RTTOV v10 Include full liquid water and ice water profiles Multiple scattering parametrisation (Chou et al., 1999; Matricardi, 2003 & 2005) Estimate single-level grey cloud parameters using minimum residual retrieval (Eyre & Menzel, 1989) Compare equivalent grey-cloud BTs with original Chou scattering BTs Approximate grey cloud RT errors
CLEAR SKY CLEAR SKY CLEAR SKY CLEAR SKY CLEAR SKY CLEAR SKY Grey cloud BT errors: Dependence on CTP and CF Full channel selection Above the red line = extra obs Use with reduced weight
Bias look-up table
Std. dev. look-up table
Corrected BT histograms Ch. 262 710.25 cm -1 (Mid-trop) Ch. 428 751.75 cm -1 (Lower trop) Ch. 1027 901.5 cm -1 (Surface)
Improved cloudy IR processing scheme simulation Bias-corrected BTs Minimum residual + 1D-Var Simulate using profile dataset + RTTOV 10 Use NWP SAF standalone 1D-Var CTP, CF CTP, CF Look-up tables Cloud bias correction Observation error model Corrected BTs (all channels) Predicted residual errors Simulate using 1D-Var 4D-Var
Preliminary 1D-Var analysis errors +++ - - - Background Channel Selection Error Model q CF<0.5, Low cloud CF<0.5, High cloud CF>0.5, Low cloud CF>0.5, High cloud T
Towards the assimilation of IR cloud information Advanced cloudy IR radiative transfer schemes: e.g. Chou scaling (RTTOV-CLD) Forward modelling of full cloud ice/water profiles Variational analysis? Adjust cloud through TL and adjoint of NWP model physics Ensemble DA? Fit observations using a linear combination of ensemble forecasts --- Global and convective-scale ---
Cloud in convective-scale DA Convective-scale model (e.g. UK 1.5km model) Radiances (e.g. SEVIRI or MTG-IRS) Simulated image: Tom Blackmore SEVIRI 10.8 m Can improved cloud analysis improve the forecast?
Convective-scale DA Representativeness / scale matching? Initial analysis vs lateral boundary conditions (global)? Future application: NWP-based Nowcasting SEVIRI 10.8 m
Variational vs. Ensemble DA Methods Variational Methods (1D/3D/4D-Var) 4D-Var: Requires TL and Adjoint of forecast model Forecast error covariance model required Relatively complex to implement Covariances not flow dependent (unless hybrid) Balance imposed through choice of control variables Ensemble Methods (EnKF, ETKF, 4DEnVar ) No TL/Adjoint needed Forecast covariances obtained from ensemble spread Relatively easy to implement Localisation issues Balance imposed by ensemble forecast model
Cloud assimilation: Plans for 2014 Investigate viability of ensemble-based cloud analysis methods e.g. 4D-En-Var Analysis = linear combination of ensemble members Focus on UK regional ensemble (MOGREPS-UK) Localisation of analysis increments? Can we generate an improved cloud analysis? How do the cloud increments affect the forecast?
Summary (1): Met Office work on cloudy IR radiances Current operational scheme (AIRS, IASI, CrIS): 1D-Var CTP, CF analysis (grey cloud) Dynamic channel selection Next step (2014-15) Bias correction and variable obs error to account for grey cloud RT model errors Towards assimilating cloud information Experiments with ensemble DA methods Focus on convective scale model (UK)
Summary (2): Questions to answer Variational vs. ensemble methods? Benefits of cloud assimilation: Global vs regional? Regional scale: Importance of cloud analysis vs LBCs? Ensemble DA: Localisation methods? How to assimilate cloud information without damaging critical T/q information? Representativeness / scale matching Plenty to discuss!
Questions and answers