The impact of dropsonde and extra radiosonde observations during the field campaigns NAWDEX and SHOUT in autumn 2016

Transcription

1 The impact o dropsonde and extra radiosonde observations during the ield campaigns NAWDEX and SHOUT in autumn 2016 Matthias Schindler, Martin Weissmann Hans-Ertel Centre or Weather Research (HErZ), Munich Andreas Schäler Deutsches Zentrum ür Lut- und Raumahrt (DLR), Oberpaenhoen George Craig Meteorologisches Institut LMU, Munich Thanks to Gabor Radnoti, Cristina Lupu, Massimo Bonavita, - <1>

2 Motivation NAWDEX campaign Diabatic processes disturbing waveguide Disturbances can grow, resulting in HIW aecting Europe Systematic errors in waveguide representation translate to errors in c Figure 1: Idealized weather situation during NAWDEX (rom Schäler et al., 2018). Gather comprehensive dataset No objective targeting, but observations in synoptically sensitive regions - <2>

3 Motivation Unique Dataset Dropsonde observations rom our research aircrats (N- and W-Atlantic) Additional radiosondes over CA and EUR Observations rom cloud radars, radiometers, wind lidars (not assimilated) Figure 2: Dropsonde and radiosonde observations. Favorable Conditions 13 intensive observation periods (IOPs) Events with low orecast skill and busts Increased requency o relevant weather: Cyclones, WCBs, ETs, NAWDEX AUGUST SEPTEMBER OCTOBER NOVEMBER Figure 3: 5 day ACC or 500 hpa geopotential height, NA region (rom Schäler et al., 2018). - <3>

4 Motivation SHOUT/ RECONNAISSANCE NAWDEX Unique Dataset Dropsonde observations rom our research aircrats (N- and W-Atlantic) Additional radiosondes over CA and EUR Observations rom cloud radars, radiometers, wind lidars (not assimilated) Figure 2: Dropsonde and radiosonde observations. Favorable Conditions 13 intensive observation periods (IOPs) Events with low orecast skill and busts Increased requency o relevant weather: Cyclones, WCBs, ETs, NAWDEX AUGUST SEPTEMBER OCTOBER NOVEMBER Figure 3: 5 day ACC or 500 hpa geopotential height, NA region (rom Schäler et al., 2018). - <3>

5 Experiments Approach Cycled data denial exps with global model o ECMWF (current version, including EDA) Cycled over whole campaign period CNTRL: Assimilating oper obs + extra obs DNL: Assimilating oper obs Evaluation o data assimilation diagnostics (e.g. FSOI, First guess departure, SG, ) Figure 4: Dropsonde and radiosonde observations, denial: green box. Goal Downstream impact o collected observations Model errors and their source as related to speciic weather eatures - <4>

6 Mean absolute orecast error Mean abs c error Dierence in mean abs c error Figure 5: Mean abs. orecast error (let) and dierence, 500 hpa geopotential, denial region Small improvement up to 96 hrs (denial region) Consistent or dierent levels and parameter - <5>

7 Timeseries o RMSE Matthew(?)/ Sanchez Karl NA development Initial Date Figure 6: Timeseries o RMSE (500 hpa Geopotential), denial area, 48 hrs. Dierence between CNTRL and DNL relatively small However: several cases with signiicant dierences (e.g. Karl) - <6>

8 IOP KARL Figure 7: Timeseries o total FSOI and impact-coded dropsonde location. Signiicant number o beneicial SH dropsondes NAWDEX light on with beneicial impact Date - <7>

9 IOP KARL Figure 7: Timeseries o total FSOI and impact-coded dropsonde location. Signiicant number o beneicial SH dropsondes NAWDEX light on with beneicial impact Date - <7>

10 IOP KARL Dierence in analysis spread Dierence in abs c error Figure 8: Dierence in analysis spread and in absolute orecast error 12UTC, 24 hrs, Reduced analysis spread or control, presumably related to cycled SH drops Smaller orecast error over widespread NA area - <8>

11 Mean impact per observation Waves to Weather DFG Collaborative Research Center 165 q dropsondes radiosondes u + v T Figure 9: FSOI per observation or dropsondes (let) and radiosondes (right), whole campaign period. Beneicial impact o dropsonde observations in denial region Large beneicial impact on wind or SH/RE dropsondes Beneicial impact o extra radiosonde observations (Canadian, all) - <9>

12 Mean impact per observation Waves to Weather DFG Collaborative Research Center 165 q dropsondes radiosondes u + v T Figure 9: FSOI per observation or dropsondes (let) and radiosondes (right), whole campaign period. Beneicial impact o dropsonde observations in denial region Large beneicial impact on wind or SH/RE dropsondes Beneicial impact o extra radiosonde observations (Canadian, all) - <9>

13 Mean impact per observation Waves to Weather DFG Collaborative Research Center 165 q dropsondes radiosondes u + v T Figure 9: FSOI per observation or dropsondes (let) and radiosondes (right), whole campaign period. Beneicial impact o dropsonde observations in denial region Large beneicial impact on wind or SH/RE dropsondes Beneicial impact o extra radiosonde observations (Canadian, all) - <9>

14 Mean impact per observation Waves to Weather DFG Collaborative Research Center 165 q dropsondes radiosondes u + v T Figure 9: FSOI per observation or dropsondes (let) and radiosondes (right), whole campaign period. Beneicial impact o dropsonde observations in denial region Large beneicial impact on wind or SH/RE dropsondes Beneicial impact o extra radiosonde observations (Canadian, all) - <9>

15 Modiied Dropsonde QC (Bonavita et al. 2017) q operational control u + v T Figure 10: FSOI or dropsondes: operations (let), control exp (right). ECMWF s modiied dropsonde quality control QC changes with signiicant inluence on wind However: horizontal drit not yet taken into account! - <10>

16 Relative total impact Waves to Weather DFG Collaborative Research Center 165 Relative total impact Number o observations Impact per observations Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Figure 11: Relative total impact (let), number o observations (middle) and impact per observation (right). Largest impact by SH/RE dropsondes and CA radiosondes Largest impact per observation: SH/RE dropsondes > NAW dropsondes > CA radiosondes Small impact by European extra radiosondes - <11>

17 Relative total impact Waves to Weather DFG Collaborative Research Center 165 Relative total impact Number o observations Impact per observations Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Figure 11: Relative total impact (let), number o observations (middle) and impact per observation (right). Largest impact by SH/RE dropsondes and CA radiosondes Largest impact per observation: SH/RE dropsondes > NAW dropsondes > CA radiosondes Small impact by European extra radiosondes - <11>

18 Relative total impact Waves to Weather DFG Collaborative Research Center 165 Relative total impact Number o observations Impact per observations Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Figure 11: Relative total impact (let), number o observations (middle) and impact per observation (right). Largest impact by SH/RE dropsondes and CA radiosondes Largest impact per observation: SH/RE dropsondes > NAW dropsondes > CA radiosondes Small impact by European extra radiosondes - <11>

19 Relative total impact Waves to Weather DFG Collaborative Research Center 165 Relative total impact Number o observations Impact per observations Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Naw_D_ml SH_RE_D CA_extra Other_extra Figure 11: Relative total impact (let), number o observations (middle) and impact per observation (right). Largest impact by SH/RE dropsondes and CA radiosondes Largest impact per observation: SH/RE dropsondes > NAW dropsondes > CA radiosondes Small impact by European extra radiosondes - <11>

20 some work in progress The concept o targeted observations: Predict location o high-impact observations, gain a lot rom ew adaptive observations Limited success in pratice during several expensive aircrat ield campaigns What determines observation impact? What is needed or predicting observation impact? For this we investigate: Relative importance o the two components (SG and g error) determining impact Role o ignoring covariances, just targeting at the location o maximum sensitivity/g error - <12>

21 some work in progress Sensitivity Gradient Pearson: First guess error Pearson: impact = innovation x K T x SG Pearson: Samplesize > 400 beneicial dropsondes Figure 12: FSOI binned or pointwise vertically integrated SG, g error and approximated FSOI. - <13>

22 Summary and Outlook Summary Slightly lower mean error o control Large dierences or Karl (and two urther cases) Overall beneicial impact o dropsondes and radiosondes Largest impact by SHOUT dropsondes (TC and ET) Followed by midlatitude NAWDEX dropsondes and CA radiosondes Small impact o European extra radiosondes Indication or weak correlation between FSOI and pointwise SG Better (but imperect) correlation or pointwise approximated FSOI Next steps FSOI comparison with NRL and Environment Canada Single cycle experiments (detailed case studies) Link impact to weather eatures - <14>

23 Reerences [1] Quantiying Observation Impact or a Limited Area Atmospheric Forecast Model Amerault et al., Data Assimilation or Atmospheric, Oceanic and Hydrologic Applications Vol. II, 2013 [2] On the initialization o Tropical Cyclones Bonavita et al. et al., ECMWF Technical Memorandum 810, 2017 [3] Monitoring the observation impact on the short-range orecast Cardinali, Quarterly Journal o the Royal Meteorological Society, 2009 [4] Estimation o observation impact using the NRL atmospheric var. data assimilation adjoint system Langland and Daley, Tellus, 2004 [5] The North Atlantic Waveguide and Downstream Impact Experiment Schäler et al., Bulletin o the American Meteorological Society, <15>

24 Forecast Error Waves to Weather DFG Collaborative Research Center 165 FSOI Basics Concept y e g e e g diagnostic tool to monitor obs impact inormation on error o new obs type, relative importance o obs etc. approx e x b x a x g x x t Theory (adjoint-based) Quadratic energy weighted orecast error J J g 1 e 2 1 e 2 g ( x ( x g x ), C( x t x ), C( x t g x ) t x ) t e g e e g -6h 00UTC t=24 hrs Figure 9: FSOI schematic (as in Langland and Daley, 2004). < 0 : obs improve c > 0 : obs degrade c Using adjoint o NWP and DA scheme to approximate e approx J J T g J y Hxb, K y Hxb xa x, b y e g - <Backup 1>

25 Forecast Error Waves to Weather DFG Collaborative Research Center 165 FSOI Basics Concept y e g e e g diagnostic tool to monitor obs impact inormation on error o new obs type, relative importance o obs etc. approx e x b x a x g x x t Theory (adjoint-based) Quadratic energy weighted orecast error J J g 1 e 2 1 e 2 g ( x ( x g x ), C( x t x ), C( x t g x ) t x ) t e g e e g -6h 00UTC t=24 hrs Figure 9: FSOI schematic (as in Langland and Daley, 2004). < 0 : obs improve c > 0 : obs degrade c Using adjoint o NWP and DA scheme to approximate e approx J J T g J y Hxb, K y Hxb xa x, b y e g impact = innovation x K T x sensitivity gradient - <Backup 1>