AMV impact studies at the Met Office James Cotton, Mary Forsythe, Francis Warrick, Richard Marriot 12th International Winds Workshop, Copenhagen, 16-20 June 2014 © Crown copyright Met Office

Overview Forecast Sensitivity to Observations MSG low level winds Filling the gap IODC investigation © Crown copyright Met Office

Forecast Sensitivity to Observations (FSO)

Impact on 24-hr forecast error - FSO Global FSO impacts now being calculated on routine basis in NRT Example from May 2014 Moist total energy norm Surface to 150 hPa 8.6% 5.5% Impact of observations on forecast error can be measured by the adjoint-based Forecast error Sensitivity to Observations (FSO) method. This works by calculating a single, scalar measure of the reduction in forecast error due to the assimilation of observations We are now able to calculate these impacts on a routine basis in NRT. Example shown from May 2014 and shows the impact on a 24-hour total (moist) energy norm, from surface to 150 hPa. Geo AMVs make a significant contribution at 8.6%, Scatterometers 5.5 % and polar winds around 0.5% AVHRR > MODIS © Crown copyright Met Office

Impact on 24-hr forecast error - FSO Global FSO impacts now being calculated on routine basis in NRT Example from May 2014 Moist total energy norm Surface to 150 hPa 9.1% If we now combine the AMV impacts then we see that, for May 2014, the AMV impact actually overtakes the Sondes. For years the AMVs have always been "the icing on the cake" in terms of impact but now they are more substantive which is good. This is a pretty impressive result for the AMVs, so its worth revisiting how we have got to this point.. © Crown copyright Met Office

Impact on 24-hr forecast error - FSO PS30-like Jan-Mar 2012 PS32-like Apr-July 2013 6.7% 4.3% +Metop-B The figure on the left shows the impacts from around the time of the last Winds Workshop in 2012. The Geo AMVs sit in 8th place on 4.3%, behind the scatterometers. By mid 2013, the Geo AMVs have jumped up to 5th (6.7%) ahead of Scat, AIRS and SYNOP AVHRR slightly ahead of MODIS winds due to addition of Metop-B (note: have also since lost Terra WV channel winds) Contributions to the total observation impact on a moist 24-hour forecast-error energy-norm (courtesy of Richard Marriot) © Crown copyright Met Office

Number of ‘soundings’ PS30-like Jan-Mar 2012 PS32-like Apr-July 2013 The reason for the large increase in AMV impact is largely due to the increase in the number of AMVs assimilated. GEO AMVs now equal with Scatwind in terms of number of ‘soundings’ i.e. wind vectors (u/v pair) NB. Scatwind will have ~2x number of 'observations' due to the ambiguous u/v wind components i.e. num wind vectors (u/v pair) for AMV and Scatterometers © Crown copyright Met Office

Implementation of temporal thinning Previously used one wind in each spatial box in the 6 hour window From PS31 (January 2013) assimilate in 2-hourly time slots 2-3x number of winds Time relative to Analysis (hrs) -3 -2 -1 1 2 3 2hr Thinning Boxes Meteosat-7 Meteosat-9 Implementation of 2-hourly thinning leads to 2-3x increase in number of winds MTSAT-1R/2 GOES-15/13 Polar winds

Change in MSG impact AMV impact by satellite PS30-like Jan-Mar 2012 PS32-like Apr-July 2013 Amongst the large improvement in AMV impact from 2012 to 2013, we also observed a large decrease in the impact from MSG. In mid 2013, MSG actually contributing less than Meteosat-7. This occurred because we had taken the decision to blacklist the low level IR/VIS winds following the implementation of the CCC scheme at EUMETSAT.. Observed significant drop in relative impact of MSG following blacklist of low level IR and visible channel winds Contribution similar to Meteosat-7

EUMETSAT CCC winds

EUMETSAT CCC winds Main changes Cross Correlation Contribution (CCC) method (Borde and Oyama, 2008) Maintain closer link between the pixels used in the height assignment with those that dominate in the tracking Makes direct use of pixel-based cloud top pressures from CLA product rather than generating AMV CTPs Pre-operational monitoring showed Significant improvements e.g. high level in the jet regions. Increase in RMSVD of ~0.6 m/s (20%) for IR and VIS winds at low levels CCC data became operational on 5 September 2012 Decided to blacklist the low level MSG data (16 Oct 2012) EUMETSAT ‘fix’ for low level data implemented 16 Apr 2013. Impact? © Crown copyright Met Office

Parallel monitoring - June 2012 VIS 0.8, Oper (pre-CCC), QI2>80 VIS 0.8, CCC, QI2>80 CCC higher Positive speed bias 0-30°S Negative bias mid latitudes/ Europe Assigned minus best-fit pressure Negative, ‘high’ height bias Shift to higher levels not in agreement with model This shows some comparisons for the parallel period of data we had in June 2012. The operational data at that time (left) and CCC data (right) With CCC scheme, wind distribution is generally shifted upwards. This led to a positive speed bias 0-30S over Atlantic. A larger negative speed bias over Europe If we compare to model best-fit pressure we see the CCC data has negative ‘high’ height bias, confirming that the shift to higher levels is not in agreement with the model. © Crown copyright Met Office

Impact experiment at ECMWF Courtesy Kirsti Salonen (ECMWF) Significant difference in mean wind analysis at low levels (850 hPa) Negative impact on forecasts -> blacklist the low level MSG winds (Sept 2012) At the time we weren’t able to run an assimilation experiment ourselves, so we relied to some extent on results from ECMWF. These are some results from Kirsti Salonen. Assimilating the CCC winds leads to a significant difference in the mean wind analysis at low levels Strengthening the mean wind field in the same region where we saw an increase in the positive bias This leads to a negative impact on forecasts (red shading on plot on right) This was why we decided to blacklist the data CCC AMVs tends to strengthen the mean wind field Same region as the increase in positive speed bias for CCC winds © Crown copyright Met Office

Update for low level HA Height difference mainly occurs in regions with a low level temperature inversion (Borde et al, 2013). Due to difference in the way the inversion correction is applied in CLA algorithm compared to the pre-CCC AMV algorithm Pre-CCC If final EBBT height below 650 hPa and above an inversion then corrected to base of inversion layer CCC If EBBT CTH below 650 hPa + temperature inversion is found then height is corrected to 1/3 way above base of inversion layer CTH can be corrected upwards as well as downwards Derivation updated on 16 April 2013 so more consistent with old inversion scheme © Crown copyright Met Office

CGMS Time series Met-9 -> Met-9 CCC -> Met-10 CCC Tropics VIS 0.8 IR 10.8 Best way to analyse impact of the low level update is from long time series. This shows the time series of Met-9 data (black), Met-9 after the CCC change (blue) and Met-10 CCC (purple) Solid line is RMSVD, dashed is speed bias The orange line shows when the low level update was applied This is for the tropics Clear increase in RMSVD and bias when CCC introduced (black/blue overlap) Following the update the stats look much better for the IR (particularly last 6 months or so) VIS still positive bias, RMS not quite back to pre-CCC levels Met-9 -> Met-9 CCC -> Met-10 CCC QI1 > 80 (IR) or QI1 > 65 (vis) © Crown copyright Met Office

CGMS Time series Met-9 -> Met-9 CCC -> Met-10 CCC Southern hemisphere extra-tropics VIS 0.8 IR 10.8 In the SH Bias looks good RMSVD still not back to pre-CCC level (particularly for VIS) Met-9 -> Met-9 CCC -> Met-10 CCC QI1 > 80 (IR) or QI1 > 65 (vis) © Crown copyright Met Office

Assimilation Experiment 45-day experiment to assess impact of reintroducing low level MSG winds to operations 10 July 2013 to 25 Aug 2013 Control: No Met-10 AMVs below 700 hPa Trial: Assimilate Met-10 IR, VIS, HRVIS below 700 hPa PS32 configuration, N320 L70, 4D-Var (N108-N216) We conducted a 45-day assimilation experiment to see the impact of reinstating the low level winds The period covered 10 July to 25 Aug 2013 We ran a control experiment, as per operations, with the Met-10 data below 700 hPa removed And then a trial experiment with the data reinstated © Crown copyright Met Office

Number of AMVs assimilated Control Trial In all these plots, blue is the control and red is the trial experiment 20% increase in AMV observations assimilated (from avg. 24,800 to 30,000) © Crown copyright Met Office

AMV zonal wind component O-B Mean u-wind O-B RMS u-wind O-B Control Trial RMS O-B ~ 0.2 m/s (6%) lower when the low level MSG winds are included. Mean O-B also shows a small improvement Similar results for O-A © Crown copyright Met Office

Scatterometer zonal wind O-B Mean u-wind O-B RMS u-wind O-B Control Trial RMS O-B for scatterometer winds also slightly lower in the trial (0.5% reduction for U and V) © Crown copyright Met Office

Impact on wind analysis 850 hPa wind field, averaged over trial period Diff: Green-blue indicates wind speeds are slower in the trial, yellow-red where they are faster. This shows the impact on the mean wind analysis at 850 hPa Trade winds in tropical South Atlantic increase by up to 0.7 m/s Slower in North Atlantic by up to 0.9 m/s © Crown copyright Met Office

Impact on 850 hPa wind forecasts Zonal wind: T+48 RMS error blue: positive impact yellow-red: negative impact This shows the impact on low level wind forecasts Areas shaded blue show positive impact, yellow/red shading shows negative impact Each experiment verified against own analysis Small region of negative impact ~ 20°S North Atlantic shows a widespread region of positive impact © Crown copyright Met Office

Change in forecast RMS (NWP Index) Standard (-0.11) Alternative (-0.13) Mixed impacts for PMSL in NH Tropical winds rather neutral Improvements for most forecast ranges in SH AMDAR sonde AMDAR sonde AMV sonde These plots show the change in forecast RMS (verified against obs) for the parameters that make up our Global NWP Index The different colours represent different forecast ranges Here we show 2 plots, “standard” and “alternative” AMDAR sonde Positive impact Positive impact © Crown copyright Met Office

Change in MSG impact (II) AMV impact by satellite PS32-like Apr-July 2013 May 2014 Meteosat-10 impact restored following reintroduction of low level winds in January 2014

Filling the gap

LeoGeo mixed satellite winds LeoGeo provides less complete coverage than seen in previous analyses Loss of GOES-12 (South America) and removal of Met-7, FY-2E (?) Assimilation trials ongoing with BUFR data

EUMETSAT Metop winds EUMETSAT data reaches as far as 50° N/S. CIMSS data is polewards of ~65° CIMSS Metop-A/B EUMETSAT vs NoPolar CIMSS vs NoPolar -0.06 -0.15 EUM Metop-A/B Positive impact Positive impact

Dual Metop-A/B winds Collocation with Meteosat-10 Good vector agreement More differences in height (HA limitations for AVHRR)

Indian Ocean intercomparison

Indian Ocean inter-comparison Report available from NWP SAF investigation web page http://nwpsaf.eu/monitoring/amv/investigations.html

Summary Latest FSO statistics show AMVs are having a substantial positive impact on Met Office global NWP Low level MSG winds show some improvement following update in April 2013. O-B’s perhaps not back to pre-CCC levels. Reinstating the data gives some improvement in forecast RMS scores AMV O-B and O-A fit improved and small benefit for scatterometer O-B Reinstated to operations 21 January 2014 Trials with LeoGeo and EUMETSAT Metop winds ongoing IODC inter-comparison report published on NWP SAF website Would be good to understand why stats not quite back to pre-CCC levels

Thank You Questions? 32

Met Office NWP model suites Global and MOGREPS-G 25-km in mid latitudes (17-km from July 2014) 70 levels (80-km model top) Hybrid 4D-Var (60-km) Analysis times: 0,6,12,18 Z T+67 forecast twice/day T+168 (7 day) forecast twice/day 12-member EPS - 32 km 4x/day T+168 Euro4 4.4 km, 70 levels (40-km model top) Global downscaler T+60 forecast twice/day T+120 (5 day) forecast twice/day UKV and MOGREPS-UK 1.5-km, 70 levels (40-km model top) 3D-Var (3 hourly) T+36 hr forecast 8/day 12-member EPS - 2.2 km 4x/day 36h Global model 25-km but will be increasing resolution to 17-km from July 2014 European model at 4-km, downscaled from the global High resolution UK regional model 1.5-km Short range forecasts over the UK at convection resolving resolution MOGREPS = Met Office Global Regional Ensemble Prediction System

Current operational use of AMVs Using data from 5 geostationary and 7 polar platforms Using data from standard set of 5 Geo platforms For polar winds, using MODIS winds from NESDIS and AVHRR winds from CIMSS (including Metop-A and Metop-B). NOAA-16 mission ended 6 June 2014. EUMETSAT Met-10, Met-7 JMA MTSAT-2 NESDIS GOES-13/15, Aqua/Terra MODIS CIMSS NOAA-15/16/18/19 AVHRR, Metop-A/B AVHRR © Crown copyright Met Office

EBBT CTH above inversion CCC Pre-CCC Courtesy Regis Borde (EUMETSAT) © Crown copyright Met Office

Impact of low level update Met-10 visible 0.8, QI2 >80 CCC: month prior updated-CCC: month after Updated AMVs assigned lower in atmosphere Reduced positive speed bias in tropical Atlantic. Negative speed bias in mid-latitudes also improved. We can then try and judge the impact by comparing the data before (left) and after the update (right). As expected the updated AMVs are generally assigned lower Impact on speed bias looks rather good Reduced the positive bias in the tropical Atlantic Negative speed boas in mid latitudes also improved © Crown copyright Met Office

CGMS Time series Met-9 -> Met-9 CCC -> Met-10 CCC Northern hemisphere extra-tropics VIS 0.8 IR 10.8 Northern Hemisphere: Negative bias improved RMSDV not quite back to pre-CCC (e.g. IR) Met-9 -> Met-9 CCC -> Met-10 CCC QI1 > 80 (IR) or QI1 > 65 (vis) © Crown copyright Met Office

Wind verification in VER Global NWP index verification 250 hPa wind against AMDARS 850 hPa wind against AMVs (tropics only) Other heights in extended index verified against sondes AMV observations used to verify the forecasts are those that were assimilated in operations at the time. The low level MSG winds are blacklisted so have fewer observations to verify against in the region where we are adding AMVs back in. Still some coverage in this region from GOES-13 and Met-7. Alternatively, the VER system also allows the option to verify all wind scores against Sondes © Crown copyright Met Office

EBBT CTH below inversion CCC Pre-CCC Red line shows final pressure. With pre-CCC (right), when we have a CTH below an inversion there is no adjustment With CCC the pressure is shifted up to 1/3 way up the inversion layer Courtesy Regis Borde (EUMETSAT) © Crown copyright Met Office

Scatwind FSO Impact by satellite January 2014 May 2014

Operational changes 2013 Emergency: GOES-13/14 Seasonal: MTSAT Derivation updates: [GOES hourly product implementation delayed] New satellites/transitions: Meteosat-10, Metop-B, MTSAT-1R/2 Assimilation changes: temporal thinning, QC, blacklisting © Crown copyright Met Office