Optimisation of Aeolus Sampling

Transcription

1 Optimisation of Aeolus Sampling 1 Gert-Jan Marseille 1, Jos de Kloe 1 Harald Schyberg 2 Linda Megner 3 Heiner Körnch 3 1 KNMI, NL 2 Met.no 3 MISU/SMHI, SE

2 Vertical and Horizontal Aeolus Measurement Positioning (VHAMP) Maximum exploitation of wind observations in NWP Establish optimal ADM-Aeolus observation size and quality to maximize mission impact Simulate such ADM-Aeolus observations and investigate impact using Hi-Res radiosondes, CALIPSO, LES & NWP inputs, (KNMI Aeolus data base) Simple theoretical data assimlation tool Ensemble Data Assimilation System Review of Mission Requirements Document in light of ADM- Aeolus operation concept changes Vertical (range gate positioning) Horizontal (heterogeneous aggregation in 2D plane) Calibration, QC, accuracy, precision, biases, error correlation

3 Some Sampling Scenarios... 1-to-1 Mie/Rayleigh overlap Mie in PBL/troposphere Mie on Tropical cirrus

4 Aeolus vertical 24 bins of 250m to 2km depth May be changed 8x per orbit Recommendations for Aeolus integration and bin positioning? Impact assessment in optically heterogenous atmosphere, i.e., with clouds Input for the Mission Requirements Document (MRD) Input for the L2B processing and NWP data assimilation strategy PBL cross calibration Ground calibration

5 Aeolus in the atmosphere 0 Rayleigh HLOS One simulated orbit LIPAS 18 Mie HLOS S E N E S E N E Rayleigh clear, Mie cloudy => Complementarity Rayleigh and Mie Clear area >> Cloudy area => Rayleigh is critical for ADM-Aeolus Many variable/mixed scenes => R and M signal aggregation and QC?

6 Hi-res radiosonde work Radiosondes provide high-resolution vertical variability Houchi et al. (2010) studied wind and shear Extended now with cloud vertical variability (and aerosol) Radiosondes also provide T and p Zhang et al. (2010) applied to De Bilt for 2007

7 Hi-res radiosonde shear Collocation data base Winds agree very well Shear in ECMWF model 2-3 times lower Tropical tropopause strongly variable Effect of shear on Aeolus? RAOB RAOB Houchi et al ECMWF ECMWF

8 Cloud layer statistics 1/3 of the cloud layers are thinner than 400m Such layers cause non-uniform backscatter and extinction Mean height of backscatter particles will be uncertain Wind and wind shear will be biased

9 Centre-of-Gravity (COG) COG w(z) is the signal strength inside the Aeolus bin as a function of altitude z Simulation Analytical calculation (no T,p dependence) Using LIPAS and (T,p) from radiosondes

10 Particle free bin analytical Rayleigh height assignment error is height dependent Typical atmosphere Stratosphere, 2 km Rayleigh bin, wind-shear 0.01 s -1 H=40 m ~ 0.4 ms -1 bias Extreme: 0.05 s -1 shear and ~ 2.0 ms -1 bias Biases exceed mission requirement in more extreme scenes (tropopause jet stream, PBL) if height assignment error is not corrected height assignment error as function of Rayleigh channel bin size 2 km 1.5 km 1 km

11 (T,p) from radiosonde database analytical radiosonde (T,P) mean stddev 1 year, station De Bilt => (T,p) => m (z) => w(z) => COG Height assignment errors are slightly larger than from analytical expressions Not very sensitive to T,p errors and predictable Use AUXMET to correct for Rayleigh channel height assignment errors in L2B optical properties code

12 RMSE wind error (systematic) Mie Rayleigh z cloud layer bin Rayleigh HLOS insensitive to z c can be obtained from optics Mie HLOS sensitive to unknown z Mie wind performance is severely degraded in clouds

13 ECMWF B error mid-latitudes Horizontal analysis Single obs. Experiment Over the English channel 500 hpa analysis increment Courtesy: Andras Horanyi (ECMWF) Background error length scale ~ 400 km Aeolus burst-mode observation separated by 200 km (< B length scale) Not fully independent information, some redundancy Aeolus continuous mode observation separated by 86 km

14 Background error length scales B-matrix formulation for operational global and mesoscale models Daley (1991) definition of B length scales: ECMWF model Global (ECMWF; EnDA) and mesoscale (HARMONIE; NCEP method) model Observation-model intercomparison (o-b) statistics: COV(o-b) = HBH T + R; i.e., the sum of (i) Background error and (ii) Instrument and wind representativeness error How to separate B and R? Used Desroziers et al. (2005) Application to ASCAT, aircraft observations, ECMWF and HIRLAM 150 km ASCAT + HIRLAM

15 Effective resolution UTLS 500 km AMDAR/ACARS/AIREP (ODB)/Mode-S Mode-S ECMWF ECMWF model starts to loose variance > 500 km scales Model does not show a k -5/3 spectrum, i.e., turbulence spectrum

16 Representativeness error From ODB ECMWF T m/s on ASCAT 12.5 km wind Upper troposphere: 2.1 m/s on aircraft components Along-track accumulation reduces the representativeness error Accumulation length of observations such that the resulting spectrum matches the model spectrum: (1) Upper troposphere: aircraft accumulation along km track (2) Ocean surface: ASCAT accumulation along km track Log variance vs wave number Aeolus representativeness error negligible for ~ 100 km along-track accumulation

17 Aeolus simulation LIPAS (Lidar Performance Analysis System) Heritage since 1999 and has evolved with mission updates Input: KNMI atmospheric database of CALIPSO backscatter / extinction and ECMWF/UKMO dynamics Marseille et al., 2011 UKMO 30 km CALIPSO

18 LIPAS HLOS wind statistics LIPAS QC: SNR too low BM 110 mj, 50 km CM 110 mj, 85 km CM 80 mj, 85 km CM 80 mj, 250 km mission requirement mj reduced Rayleigh coverage

19 Added value NWP by Aeolus Theoretical tool Based on theoretical equations data assimilation No competitive observations Limited to analysis quality, no forecast projection EDA Ensemble Data Assimilation Experiments in operational ECMWF system Aeolus in competition with other observing systems Ensemble spread is a measure of impact Compare forecast spread for different sets of observations Less spread means better forecast Does Aeolus reduce forecast spread? x a = x b +W(y-Hx b ) A= (I-WH)B(I-WH) T + WOW T Impact: tr(a)/tr(b)

20 1D theoretical tool Usual meteorological analysis equations Fully solved 1D = horizontal (in VHAMP) Horizontal characteristics from ECMWF and HARMONIE model and (LIPAS) Aeolus observations Introduction of bias, correlation, averaging, thinning and misspecification Norwegian Meteorological Institute met.no

21 Correlated representativeness error 60N background statistics, continous 80 mj, 500hPa (LIPAS Rayleigh channel mean obs error), 2/3 B bandwidth Representativness error variances based on assuming global model effective resolution 7* x (112 km; Skamarock) Triangular O correlation structure with half basis of 112 km Much lower analysis quality Optimal accumulation length is now about the effective model resolution 22 Norwegian Meteorological Institute met.no

22 Conclusions theoretical tool Latitudinal dependence Impact increases substantially from Burst Mode (BM) Cont. Mode (CM) Impact reduces substantially at 80 mj (CM 80 mj ~ BM 110 mj) Aeolus impact appears maximum around 250 hpa Aeolus impact is maximum in the Tropics Impact is maximized for ~85 km accumulation length

23 Conclusions theoretical tool Observation error correlation up to is not detrimental Correlation 0.1 corresponds to an increase of random error of 0.2 m/s Correlation 0.38 corresponds to an increase of random error of 0.7 m/s Biases > 0.5 m/s are detrimental Negative impact for biases exceeding 1 m/s Impact of mis-specified B-matrix is substantial ESA VHAMP, TN8

24 Conclusions EDA experiments EDA experiments conducted : No sondes; to assess radiosonde impact as reference for Aeolus BM, 110 mj, 50 km accumulation CM, 110 mj, 85 km accumulation CM, 80 mj, 85 accumulation CM 80 mj, 250 km accumulation CM 80 mj, 85 km accumulation; 1-year mismatch; to test impact of erroneous observations Aeolus impact comparable to radiosonde - above 24 km Aeolus quality reduces - Below 10 km, Aeolus impact larger Impact all Aeolus scenarios very similar (large-scale impact) 100 hpa 500 hpa

25 Conclusions EDA experiments Maximum Aeolus impact in the tropics and in the UTLS In agreement with theoretical tool

26 Conclusions EDA experiments 85 km Impact at 500 hpa Impact reduces when going from 110 mj 80 mj, but not dramatic 85 km However, assumed Perfect calibration No instrument biases No laser jitter

27 Conclusions Issues of instrument wind calibration, zonal wind variability climate, atmospheric heterogeneity, expected beneficial impact, and data assimilation method are all at interplay The vertical bin sizes should be at least 1 km for the Rayleigh channel in the lower troposphere increasing to 2 km in the stratosphere to obtain accuracy over a ~100km horizontal context It is advantageous to change the Mie vertical sampling along track, i.e., positioning top Mie bins around 11 km over the Poles to up to 18 km in the tropics to better sample tropical cirrus and obtain maximum NWP benefit Along track accumulation in the range km for global NWP, but continuous mode allows context-sensitive aggregation, esp. for Mie Wind biases should be below 0.5 m/s for a successful mission Observation error correlations should stay below 0.15 Zero wind calibration on ground targets is probably favorable with the Mie channel (unfavourable for Rayleigh), but calibration procedures need further evaluation (cause detrimental vertically correlated error) Assimilation of wind shear causes the loss of information on observed deep vertical structures and is not a good alternative for lack of absolute calibration Recommend advanced NWP monitoring to obtain instrument biases and consistency with the Global Observing System

28

29 Conclusion Vertical Sampling Rayleigh bin size is driven by the mission HLOS wind quality requirements (1-2 km) Increasing the Mie channel sampling in heterogeneous atmospheric regions reduces height assignment errors of both Mie and Rayleigh winds near the tropopause and jet stream and provides NWP impact (EnDA, theoretical tool) Zero wind calibration on ground targets is probably favorable with the Mie channel (unfavourable for Rayleigh) Issues of instrument wind calibration, zonal wind variability climate, atmospheric heterogeneity, expected beneficial impact, and data assimilation method are all at interplay An advanced ADM-Aeolus vertical sampling scenario takes account of climate regions and ground calibration opportunities and has been proposed in the VAMP project Assimilation of wind shear causes the loss of information on observed deep vertical structures and is not a good alternative for lack of absolute calibration

30 VHAMP method Establish the ability to exploit wind observations in weather models Establish optimal ADM-Aeolus observation size and quality to maximize mission impact Simulate such ADM-Aeolus observations and investigate impact using Simple theoretical data assimlation tool Ensemble Data Assimilation System Review of MRD in light of ADM-Aeolus operation concept changes