Assimilating cloud information from satellite cloud products with an Ensemble Kalman Filter at the convective scale Annika Schomburg, Christoph Schraff This work was funded by the EUMETSAT fellowship programme. International Symposium on Data Assimilation, Munich 24-28 February 2014
The COSMO model COSMO-DE : Limited-area short-range convectionpermitting numerical model weather prediction model x 2.8 km / 50 vertical layers Explicit deep convection Implementation of the Ensemble Kalman Filter (LETKF, Hunt et al., 2007): COSMO priority project KENDA (Km-scale ensemble-based data assimilation) KENDA workshop Friday 2
Local Ensemble Transform Kalman Filter OBS-FG Obs R Observation errors Analysis perturbations: linear combination of background perturbations OBS-FG OBS-FG LETKF OBS-FG Background error covariance P b ( k 1) k 1 b( i) b b( i) i 1 ( x x )( x x b ) T Local: the linear combination is fitted in a local region - observation have a spatially limited influence region Transform: most computations are carried out in ensemble space computationally efficient 3
Local Ensemble Transform Kalman Filter OBS-FG Obs R Observation errors Analysis perturbations: linear combination of background perturbations OBS-FG OBS-FG LETKF OBS-FG Background error correlations P b ( k 1) k 1 b( i) b b( i) i 1 ( x x )( x x b ) T Additional: one deterministic run: x a det x b det Kalman gain matrix from LETKF o b K( y H( xdet))
Observation systems Geostationary satellite data: Meteosat-SEVIRI (Δx ~ 5km over central Europe, Δt=15 min) Source: EUMETSAT NWCSAF Satellite product: cloud top height Cloud top height Cloud top height Relative humidity at cloud top height 1 2 3 4 5 6 7 8 9 10 11 12 13 Height [km] Cloud cover 5
Determine the model equivalent cloud top CTH obs Z [km] model profile Cloud top k 1 k 2 k 3 Avoid strong penalizing of members which are dry at CTH obs but have a cloud or even only high humidity close to CTH obs search in a vertical range h max around CTH obs for a best fitting model level k, i.e. with minimum distance d: d min k ( RH k RH obs ) 2 1 h max ( h k CTH obs ) 2 k 4 k 5 relative humidity = 1 height of model level k (make sure to choose the top of the detected cloud) RH [%] use y=cth obs H(x)=h k and y=rh obs =1 H(x)=RH k (relative humidity over water/ice depending on temperature) as 2 separate variables assimilated by LETKF 6
Example: 17 Nov 2011, 6:00 UTC Observations and model equivalents Cloud top height Observation Model RH model level k 7
Determine model equivalent: cloudfree pixels Z [km] 12 What information can we assimilate for pixels which are observed to be cloudfree? 9 no high cloud assimilate cloud fraction CLC = 0 separately for high, medium, low clouds 6 no mid-level cloud model equivalent: maximum CLC within vertical range 3 no low cloud CLC 8
Example: 17 Nov 2011, 6:00 UTC COSMO cloud cover where observations cloudfree Low clouds (oktas) Mid-level clouds (oktas) High clouds (oktas) 9
1-hourly cycling over 21 hours with 40 members 13 Nov., 21UTC 14 Nov. 2011, 18UTC Wintertime low stratus Thinning: 14 km Comparison cycling experiment: only conventional vs conventional + cloud data 10
Comparison only conventional versus conventional + cloud obs" Time series of first guess errors, averaged over cloudy obs locations assimilation of conventional obs only assimilation of conventional + cloud obs RH (relative humidity) at observed cloud top RMSE Bias (OBS-FG) Cloud assimilation reduces RH (1-hour forecast) errors 11
Comparison of cycled experiments Total cloud cover of first guess fields after 20 hours of cycling satellite obs conventional only conventional + cloud Satellite cloud top height 12 Nov 2011 17:00 UTC 12
Cycled assimilation of dense observations Time series of first guess errors, averaged over cloud-free obs locations (errors are due to false alarm clouds) mean square error of cloud fraction [octa] low clouds High clouds False alarm clouds reduced through cloud data assimilation 13
Comparison only conventional versus conventional + cloud obs" false alarm cloud cover (after 20 hrs cycling) low clouds mid-level clouds high clouds conventional obs only [octa] conventional + cloud 14
Comparison forecast experiment: only conventional vs conventional + cloud data 24 deterministic forecast based on analysis of two experiments (after 12 hours of cycling) 14 Nov., 9UTC 15 Nov. 2011, 9UTC Wintertime low stratus 15
Comparison of free forecast: time series of errors RH (relative humidity) at observed cloud top averaged over all cloudy observations Mean squared error averaged over all cloud-free observations RMSE Bias (Obs-Model) Low clouds Mid-level clouds High clouds Conventional + cloud data Only conventional data The forecast of cloud characteristics can be improved through the assimilation of the cloud information 16
Verification against independent measurements Errors for SEVIRI infrared brightness temperatures (model values computed with RTTOV) Conventional + cloud data Only conventional data RMSE Bias (Obs-Model) RMSE is smaller for first 16 hours of forecast for cloud experiment, bias varies 17
Verification against independent measurements: SEVIRI brightness temperature errors Only CONV experiment CONV+CLOUD experiment 14 Nov 2011, 18 UTC Cloud top height Also the high clouds are simulated better in the cloud experiment 18
Conclusions and next steps Use of (SEVIRI-based) cloud observations in LETKF: Tends to introduce humidity / cloud where it should and to reduce false-alarm clouds Improvement on cloud characteristics in free forecast for a stable wintertime high-pressure systems Usefulness for frontal systems or convective situations still needs to be proven If convective clouds are captured better while developing, convective precipitation may be improved Application in project EWeLiNE: Improving the forecast for renewable energy sector (clouds particularly important for photovoltaic power production) Thank you for your attention! 19
20
Single observation experiment Analysis for 17 November 2011, 6:00 UTC (no cycling) Each column is affected by only one satellite observation Objective: Understand in detail what the filter does with such special observation types Does it work at all? Analyse effect on atmospheric profiles Sensitivity to settings Very detailed evaluation 21
Single-observation experiments: missed cloud event 1 analysis step, 17 Nov. 2011, 6 UTC (wintertime low stratus) vertical profiles relative humidity cloud cover cloud water cloud ice observed cloud top 3 lines in one colour indicate ensemble mean and mean +/- spread 22
Single-observation experiments: missed cloud event Cross section of analysis increments for ensemble mean specific water content [g/kg] relative humidity [%] observed cloud top observation location 23
Deterministic run First guess Analysis Relative humidity Cloud cover Cloud water Cloud ice Observed cloud top Humidity at cloud layer is increased in deterministic run 24
Missed cloud case: Effect on temperature profile temperature profile [K] (mean +/- spread) first guess analysis observed cloud top LETKF introduces inversion due to RH T cross correlations in first guess ensemble perturbations 25
Single-observation experiments: False alarm cloud assimilated quantity: cloud fraction (= 0) vertical profiles relative humidity cloud cover cloud water cloud ice observed cloud top 3 lines on one colour indicate ensemble mean and mean +/- spread 26
Increment cross section ensemble mean Observation location Observed cloud top 27
Single-observation experiments: False alarm cloud Observation cloudfree assimilated quantity: cloud fraction (= 0) Observation minus model histogram over ensemble members low cloud cover fraction [octa] FG ANA mid-level cloud cover fraction [octa] FG ANA LETKF decreases erroneous cloud cover despite very non-gaussian distributions 28
Cycling experiment: only cloud data 1-hourly cycling over 21 hours with 40 members 13 Nov., 21UTC 14 Nov. 2011, 18UTC wintertime low stratus Sensitivity experiment: Thinning 8 km 14 km 20 km 29
Sensitivity experiment: Data density Comparing experiments with different data density: 8 km 14 km 20km RH (relative humidity) at observed cloud top Low cloud cover RMSE Spread For cloudy pixels best results are obtained for a 14 km thinning distance, for cloud-free observations no clear conclusion The spread for the 8km thinning experiment is lower than the other two, the difference in spread between the 14 and 20km experiments is smaller. Ensemble is underdispersive, but there is no sign of a further reduction of the spread during 30 the cycling
Sensitivity experiment: Data density Comparing experiments with different data density: 8 km 14 km 20km RMSE Low clouds Mid-level clouds High clouds Solid: 8km Dashed: 14km Dotted: 20km Bias (OBS-FG) RMSE and bias averaged over all cloudy observations Mean squared error for low/medium/high cloud cover averaged over all observed cloud free pixels For cloudy pixels best results are obtained for a 14 km thinning distance, for cloud-free observations no clear conclusion 31
Comparison of free forecast: Example fields Run forecast after 12h cycling satellite obs conventional only conventional + cloud Observed cloud top height Total cloud cover after 12h forecast (15. Nov 2011, 6:00 UTC) 32
Comparison of free forecast: Example fields after 12h forecast Satellite observation conventional only conventional + cloud cloud top Total cloud cover 14 Nov 2011, 18 UTC In the cloud data experiment, less of the stratus clouds are resolved too early 33