1. Blind tests of radar/lidar retrievals in ice clouds: Assessment in terms of radiative fluxes
Robin Hogan, Malcolm Brooks, Anthony Illingworth
University of Reading, UK
David Donovan Claire Tinel
KNMI, The Netherlands CETP, France

2. Ice cloud retrievals: the problem
Radar only
Not well related to radiative properties
Lidar only
Difficult to correct for attenuation

3. Combined radar and lidar
Current plan for CloudSat and Calipso…
Combine Calipso Level 2 extinction profile with CloudSat reflectivity product to get IWC and effective radius (size is related to ratio of Z and extinction coefficient)
But the large errors in the extinction will feed through to very inaccurate retrievals, particularly at cloud base
Alternative: use radar to constrain lidar inversion
Donovan et al. (2000): find lidar ratio that minimizes the variation in implied effective radius in furthest few gates
Tinel et al. (2000): find lidar ratio that minimizes the implied variation in number concentration in the profile
CloudSat must ingest Calipso Level 1 backscatter product

4. Blind Tests
At last CloudSat science meeting a “Blind Test” of these algorithms was reported:
I used aircraft spectra to simulate radar/lidar profiles
Dave Donovan and Claire Tinel ran their algorithms to derive profiles of extinction, IWC and effective radius
What we found:
Both algorithms retrieved extinction very accurately, even for 1-way optical depths of up to 7 (lidar reduced by 10-6)
Stable to uncertain lidar extinction-to-backscatter ratio
However, re and IWC retrievals sensitive to mass-size relation

5. Blind Test 1 No instrument noise No multiple scattering
(From aggregation study)
No instrument noise
No multiple scattering
No molecular scattering
High lidar sensitivity
Two versions of each profile provided, with variable or constant extinction/backscatter ratio “k”, which was not known by the algorithms

6. Blind Test 1: Results 1 Constant k: Variable k:
Both Donovan and Tinel (after modification) algorithms produce highly accurate extinction
Variable k:
Error in extinction varies with k, but not unstable

7. Blind Test 1: Results 2 Effective radius: Ice water content:
Good, but difficult if re > 80 microns because of radar Mie scattering
Sensitive to particle habit
Ice water content:
Extinction ~ IWC/re
Hence if extinction is correct then the % error in effective radius is equal to the % error in IWC

8. In this talk…
Ultimate test: are these retrievals accurate enough to constrain the radiation budget?
Perform radiation calculations on true & measured profiles
Assess errors in terms of fluxes and heating rates
What are the most sensitive of the retrieved parameters?
But in the Blind Test, the “measured” profiles were noise-free and almost infinitely sensitive!
Second Blind Test simulates instrumental limitations that will be faced by EarthCARE for:
10-km dwell (1.4 seconds), 400-km altitude, 355-nm lidar
Conclusions similar for CloudSat/Calipso
Radar/lidar combination better than radar alone?
Also test Z-IWC, Z-, Z-re relationships (from EUCREX)

11. Case from First Blind Test
Excellent extinction (both Donovan and Tinel)
Good re if same mass-size relationship used (otherwise 40% too low)
Extinction coefficient
Effective radius
Francis et al. relationship
Radar only retrieval
Mitchell relationship

12. What about radiative fluxes?
Edwards-Slingo 1-D plane-parallel calculations
Excellent longwave, good shortwave, slight effect of habit and extinction-backscatter ratio, better than radar alone
Effective radius not very important?
Longwave up
Clear sky profile
Error W m-2 depending on habit and k
Cloudy profile
Shortwave up

13. Heating rates
Radar/lidar: very accurate
Radar alone: almost as good

14. Case from First Blind Test
Excellent extinction (both Donovan and Tinel)
re depends on mass-size relationship, but can still be too low
Extinction poor from radar only
Extinction coefficient
Effective radius
Poor radar-only retrieval, particularly at cloud top
Mitchell relationship
Francis et al. relationship
Effective radius underestimate

15. Radiation (Edwards-Slingo code)
Excellent longwave, good shortwave
Slight effect of habit and k; better than radar alone
Effective radius not very important?
Clear sky profile
Error Wm-2 depending on habit and k
Cloudy profile
Longwave up
Shortwave up

16. Error due to higher Z here
Heating rates
Radar/lidar: reasonably accurate
Radar alone: OK but some biases
Error due to higher Z here

17. Second Blind Test: more realistic
Ice size distributions from EUCREX aircraft data
Correction for 2D-C undercounting of small ice crystals
Radar
Simulate 100-m oversampling of 400-m Gaussian pulse
Noise added based on signal-to-noise and number of pulses
Lidar
Add molecular scattering appropriate for 355 nm
Instrument noise: photon counting - Poisson statistics
True lidar sensitivity: incomplete penetration of cloud
Multiple scattering: Eloranta model with 20-m footprint
Extinction-to-backscatter unknown but constant with height
Night-time operation, negligible dark current

18. Accounting for small ice crystals
2D-C probe undercounts small crystals
Assume gamma distribution for crystals < 100 m diameter
Mode at 6 m
Same conc. at 100 m
Conc. 2.5 times higher at 25 m

19. Instrument noise Radar Five new profiles from EUCREX dataset
This is what they would look like without instrument noise or multiple scattering
Note strong lidar attenuation
Lidar

20. Instrument noise Radar Five new profiles…
With instrument noise & multiple scattering
Radar virtually unchanged except finite sensitivity
Lidar noise noticeable
Lidar multiple scattering increases return
Note: “radar-only” relationships derived using same EUCREX dataset so not independent!
Lidar

21. Donovan retrieval: with multiple scattering

22. Tinel retrieval: no multiple scattering

23. Good case: lidar sees full profile
Extinction and effective radius reasonable when use same habit and include multiple scattering
Extinction coefficient
Effective radius
Donovan: includes multiple scattering
Difference between Mitchell
and Francis et al. mass-size
Tinel: no multiple scattering

24. Good case: radiation calculations
OLR and albedo good for both radar/lidar and radar-only (but radar-only not independent)
Longwave up
Mass-size relationship:
Error~10 Wm-2
Underestimate radiative effect if multiple scattering neglected
Shortwave up

25. Typical case: radar/lidar retrievals
No retrieval in lower part of cloud
Important to include multiple-scattering in retrieval
Extinction coefficient
Effective radius
Donovan: good retrieval at cloud top
Difference between Mitchell
and Francis et al. mass-size relation
Tinel: no multiple scattering
Wild retrieval where lidar runs out of signal

26. Typical case: radiative fluxes
At top-of-atmosphere, lower part of cloud important for shortwave but not for longwave
Underestimate radiative effect if multiple scattering neglected
Albedo too low (80 W m-2): lower part of cloud is important but mass-size less so (10 W m-2)
OLR excellent:lower part not important
Longwave up
Shortwave up

27. Typical profile: Heating rates
Heating profile would be reasonable if full profile was retrieved
What do we do when the lidar runs out of signal?
Erroneous 80 K/day heating
No cloud observed so no heating by cloud here

28. Simple solution: blend profiles
Where lidar runs out of steam, scale radar-only retrieval for seamless join
Much better result than pure radar/lidar or radar only
Lidar becomes unreliable here
Radar/lidar only
Radar only
Scale radar-only retrieval to match
Blended
Extinction coefficient
Shortwave up

29. Possible solution: blend profiles
Where lidar runs out of steam, scale radar-only retrieval for seamless join
Better result than pure radar/lidar or radar only
Lidar becomes unreliable here
Blended
Radar only
Radar/lidar only
Scale radar-only retrieval to match here
Extinction coefficient
Shortwave up

30. Sensitivity of radiation to retrievals
Longwave: Easy! (for plane-parallel clouds…)
Sensitive to extinction coefficient
Insensitive to effective radius, habit or extinction/backscatter
OLR insensitive to lower half of cloud undetected by lidar
“Blending” usually gets in-cloud fluxes to better than 5 W m-2
Shortwave: More difficult
Most sensitive to extinction coefficient
Need full cloud profile: blending enables TOA shortwave to be retrieved to ~10 W m-2, in-cloud fluxes less accurate
Some sensitivity to habit and therefore effective radius
Slight sensitivity to extinction/backscatter ratio

31. Conclusions Extinction much the most important parameter:
Good news: this can be retrieved accurately independent of assumption of crystal type
So lidar can provide the extra accuracy at cloud top necessary if retrievals are to be consistent with measured TOA fluxes
But need to include multiple scattering in retrieval
Must avoid erroneous spikes where lidar loses signal!
What is the best way to blend profiles?
Scale radar-only retrieval? Or switch straight to radar-only?
Need to analyze aircraft spiral descents through ice cloud
How could radiances from passive instruments be used to refine the retrievals?
E.g. do SW radiances provide multiple-scattering information?

32. Scaling the radar-only retrieval
Where radar/lidar retrieval fails, can we scale the radar-only retrieval to get a seamless join?
Dubious: the profiles are not real but simulated!
Good fit
Partial fit

33. An invitation!
Try your algorithm on profiles of radar reflectivity and attenuated lidar backscatter from the first blind test (variable lidar ratio, no instrument noise):
If it passes the test, try profiles from the second blind test (multiple scattering, instrument noise):