1. Aeolus in heterogeneous atmospheric conditions
Gert-Jan Marseille

2. Polar Stratospheric Clouds
Quite a lot over Antarctic in August and Arctic in January
30 km
15 km
PSC not always well sampled with the Mie channel

3. true cloud/aerosol (CALIPSO)
Aeolus simulations
true HLOS wind (UKMO)
true cloud/aerosol (CALIPSO)
30 km
Aeolus Rayleigh channel
1 orbit, simulations with LIPAS
30 km
Aeolus Mie channel
18 km
Rayleigh clear Mie, cloud/aerosol => complementary
Clear area dominates => Rayleigh channel is most important for Aeolus
No winds below (optically) dense clouds

4. Aeolus Rayleigh wind error can be large depending on
Clear atmosphere
true atmosphere
Mie signal
Rayleigh signal
30 m/s
20 m/s
1 km
mean wind in bin: 25 m/s
measured: nothing
measured: < 25 m/s
Aeolus Rayleigh wind error can be large depending on
binsize
bin altitude
wind-shear inside the bin
COG

5. Rayleigh wind error in clear atmosphere
Rayleigh channel height assignment error is height dependent
Typical atmosphere example
Stratosphere, 2 km Rayleigh bin, wind-shear 0.01 s-1
H=40 m ~ 0.4 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

6. Cloud layer inside Aeolus bin
true atmosphere
Mie signal
Rayleigh signal
30 m/s
20 m/s
cross-talk
1 km
cloud
transmission
mean wind in bin: 25 m/s
measured: 25 m/s
measured: > 25 m/s
mean wind in bin: 25 m/s
measured: 20 m/s
measured: < 25 m/s
Aeolus wind error can be large depending on (i) bin size, (ii) cloud/aerosol layer location inside the Aeolus bin, (iii) layer size, (iv) layer transmission and (v) wind-shear over the bin

7. Mie Rayleigh z  c  RMSE wind error
Sun et al., 2014
m/s
m/s
Mie Rayleigh
z 
c 
Rayleigh HLOS insensitive to z, but sensitive to
particle layer transmission c
c can be obtained from Rayleigh channel signal
Rayleigh winds are under control
Mie HLOS however sensitive to z
cloud layer z 
Bin height

8. model vs. real atmosphere
Models are very smooth relative to the real atmosphere as measured by radiosonde
ECMWF underestimates real atmospheric wind shear by a factor of 3
ECMWF
radiosonde
Houchi et al., 2010

9. Radiosonde database
Radiosondes provide wind, temperature, humidity and pressure at high (10-m) resolution
cloud layers detected from humidity along the radiosonde path (Zhang et al., 2010). Applied to De Bilt radiosonde
One year (2007) database of high resolution (u, v, T, P) and cloud for Aeolus testing and L2Bp algorithm development
Retrieval of aerosol and cloud properties from radiosonde data remains a challenge
Focus on cloud layers,
assuming simplified back-
scatter and extinction

10. Cloud layer statistics from radiosondes
Aeolus height bins are typically 1 km
But, 1/3 of cloud layers are thinner than 400m
Such layers cause non-uniform Mie backscatter and extinction
Mean backscatter height is uncertain
Wind and wind shear will be biased (mean shear = 4 m/s per km height)
Advanced retrieval methods will be needed
Sun et al., 2014
Mie wind error

11. Optical Properties Code (OPC)
Data assimilation requires estimates of the errors of Mie and Rayleigh channel winds
Pretty well known in clear air (Rayleigh winds)
Less well known in heterogeneous atmosphere
This requires estimates of particle layer size, location and transmission
Optical Properties Code as part of the L2Bp
Main purpose: feature detection as input for classification before accumulation from measurement to observation level
Spin-off: estimation of particle backscatter and extinction and (sometimes) the location of the layer inside the Aeolus bin
Needs Rayleigh channel signal only!
Still under development

12. LITE scene – tropical cirrus over Indonesia??
Measured signal is at very low resolution and noisy
Thin layers hard to detect, even by eye
quite a challenge!

13. OPC – feature detection
Input:
AUX_MET: ECMWF forecast of P,T,u,v as a function of z
AUX_CAL: Rayleigh channel calibration info
Estimate the Rayleigh channel signal in clean (no aerosol/cloud) air
Compare with measured signal
Detect features

14. Feature detection
true
OPC
OPC

15. Score sheet
good detection
false alarm
missed detection
clear air

16. OPC layer location estimation
Black dots denote OPC layer top/bottom
OPC well fits the layer location at a resolution higher than Aeolus bins
Outliers for individual measurements

17. Aeolus main contribution is Rayleigh winds in clear atmosphere
Conclusions
Aeolus main contribution is Rayleigh winds in clear atmosphere
Error characteristics are well-known
Large signal from cloud and aerosols => good Mie SNR
But location of layer inside bin not known => large Mie wind errors, which cannot easily be estimated and/or corrected
Mie winds may not be very useful for NWP
Rayleigh winds in cloudy/aerosol conditions more reliable
But depending on layer size and transmission
Optical properties code (OPC) provides estimates
Feature detection (as part of OPC) is critical to decide between clear/cloud (aerosol) bins

18. backup

19. Feature detection of true input
scattering ratio > 1.15

20. This is the challenge
scattering ratio > 1.15
Noisy

21. (T,P) from radiosonde database
analytical
radiosonde (T,P)
mean and stddev
1 year radiosonde data over De Bilt => (T,P) => m(z) => w(z) => COG
Molecular (attenuated) backscatter profile smaller than from analytical expression
Height assignment errors are slightly larger than from analytical expressions
Not very sensitive to T errors
Use AUXMET to correct for Rayleigh channel height assignment errors

22. Results largely confirmed for LIPAS applied to the radiosonde database
Bin with cloud layer
Mie Rayleigh
Cloud layer with thickness z and one-way transmission c; linear inside cloud
cloud layer z 
bias
bin
z =>
H can not be corrected: cloud location and thickness are unknown
H for Rayleigh ch. relative insensitive for z, more to c but that can be obtained from optical prop. code
We can estimate Rayleigh wind bias, not for Mie; Rayleigh wind bias smaller, in particular for c > 0.8
std.dev
RMSE
c =>
Results largely confirmed for LIPAS applied to the radiosonde database

23. Radiosonde database – Mie wind error
309 radiosonde launches in De Bilt
About 25% of cloud layers is assigned ice cloud and generally thin:
25% < 300 m
60% < 1 km
Mie wind error reduces for smaller bin sizes
For 1 km bins: std.dev = ms-1
Slightly below the analytical calculations (1.66 ms-1)
2000 m
1000 m
500 m
250 m

24. Radiosonde database – Rayleigh wind error
309 radiosonde launches in De Bilt
Rayleigh wind error reduces for smaller bin sizes
For 1 km bins: std.dev = ms-1 in free troposphere
Compatible with analytical calculation (0.4 ms-1)
But classification may further reduce Rayleigh ch. wind error
Particle-free