Analysis of D Band Cloud Flag Jane Hurley Anu Dudhia Graham Ewen University of Oxford.

Analysis of D Band Cloud Flag

Jane Hurley

Anu Dudhia

Graham Ewen

University of Oxford

Background• Presence of clouds in the field of view (FOV) of remote

sensing instruments influence observations

• Retrievals can normally deal with small amounts of cloud in the FOV by fitting a continuum term in parallel with that of the retrieved species

Important to recognize and reliably be able to identify presence of clouds

• A couple of common and simple techniques for cloud detection …

Simple Radiance Thresholding (basic)Color Index Thresholding (improvement, as

reduces influence of variations in p and T)

Colour Indices (CI) in a Nutshell• CIs work on the principle of ratios of mean radiances

between two spectral microwindows which respond differently to cloud.

• CI = Lav MW1 / Lav MW2

Large CI (ie. CI > 4) → cloud freeSmall CI (ie. CI ~ 1) → thick cloud Range of CIs represents range of optical

thicknesses of clouds present in FOV

• Presence of cloud determined by setting a threshold on the CI such that

for all CI > CIthresh → cloud free FOV

for all CI < CIthresh → cloudy FOV

• Cautions: Definition of CI breaks down above ~ 30 km with

decreased SNRAbove threshold (“cloud free”), cloud can still occur

if the cloud is optically thin or partially filling the FOV …

• MIPAS has CIs defined for 3/5 bands:

Colour Index MW1 (cm-1) MW2 (cm-1) Threshold

CI-A

CI-B

CI-D

788.0 - 796.0

1246.3 -1249.1

1929.0 -1935.0

832.0 - 834.0

1232.3 -1234.4

1973.0 -1983.0

1.8

1.2

1.8

Issues for Discussion

• Anomalous situations where the D band flags cloud but the A band does not – is it possible that the D band is better suited to detect certain cloud types?

• A correlation between these anomalous events (where the D band flags cloud while the A band doesn’t) and extremely low H2O volume mixing ratio (vmr)

• Day/night difference in behaviour of the D band

Correlation of Low H2O VMR and D/A Cloud Flagging Anomaly

• Previous work (Remedios) implied a correlation between anomaly and low H2O vmr values (10-10ppmv).

• These low H2O vmrs are markers for when retrieved H2O vmr is negative → a bad H2O vmr retrieval

• Using arbitrary day (15 August 2003), in keeping with previous work, found that in the range of definition of CI:

if D band flags and A band doesn’t, ~ 8% of the occurrences have low H2O vmr (are bad retrievals)

if have low H2O vmr, ~ 20% of the occurrences have the D band flagged but the A band unflagged.

A WEAK CORRELATION AT BEST. Not trustworthy enough to risk throwing away quite a bit of useful data …

If were to use D band as a filter for poor H2O retrieval, could potentially lose 92% of such points which could be good data

# flagging anomaly / # total points = 353 / 12041 ~ 3%# low H2O vmr / # total points = 173 / 12041 ~ 1.5%

Day/Night Difference in D Band

• Remedios found that anomaly of D band flagging and A band not flagging occurred for winter daytime points (60-75 S day, 60 – 75 S night, 75 – 90 S day);

• Could not replicate this result – but definitely found concentration of such anomalous winter daytime points in winter south pole;

• Difference in day/night D band spectra discussed later …

D Band Sensitive Cloud Detector?

• Mostly winter daytime points (but quite a few mid-lat nighttime points as well …)

• Polar points attributable to PSCs (A band close enough to threshold)

Standard Spectra

• A Band Clear: level zero-magnitude baseline; day/night spectra have same relative shape.

• A Band Cloudy: slanted baseline heightened in magnitude above zero; day/night spectra have same relative shape.

• D Band Clear: level zero-magnitude baseline; day/night spectra have same relative shape.

• D Band Cloudy: slanted baseline heightened in magnitude above zero; daytime spectra exhibit large non-LTE feature at ~ 2350 cm-1.

Anomaly Spectrum• Low altitude, mid-latitude daytime example of where D

band flags but A band doesn’t. CI-D = 0.94, CI-A = 1.88

• Both spectra exhibit cloudy features – there is cloud, but no surprise since A band quite close to threshold.

• Pick situations where flagging anomaly occurs but can’t be attributed to cloud – ie. mid-lats, 20 – 30km altitude.

• For a period of 15 July – 15 August 2003, such anomalous points (D flagged/ A unflagged) looked clear in the A band and had strange negative feature in D band at ~ 2350 cm-1. ANOMALY NOT CAUSED BY CLOUD PRESENCE!

• Thick cloud spectra should limit to Planck blackbody function at brightness temperature

Planck function at appropriate TB should flag as cloud

• Using ECMWF temperature profiles:

• A band reliably flags thick cloud but D band stops flagging cloud at higher altitudes/ lower brightness temperatures

15 km 18 km 22 km 25 km 28 km

CI-D 1.29 1.32 1.28 1.26 1.25

CI-A 1.15 1.17 1.14 1.13 1.12

• D band moves away from flagging Planck function at colder temperatures.• Suggested threshold of 1.2 would fail for most atmospheric temperatures.

RFM Atmospheric Models• Reference Forward

Model (RFM) used to simulate clear and thick cloudy atmospheric conditions

• A more realistic model than a blackbody, taking atmospheric absorptions/emissions etc into account

• Using a ‘step function’ cloud …

• Sample spectra at a tangent height of ~ 15 km for a cloudy atmosphere

• RFM shows range over which can confidently use CI-A and CI-D

• Clearly D band flag acts unreliably at most altitudes

General Behaviour of CI-D & CI-A

• A band behaves as expected, detecting cloud at low altitudes (5-20km) and not at higher altitudes (20-30km);

• D band less convincing, detecting cloud even where there should be none from a statistical standpoint;

A band seems a more reliable cloud flag than the D band!!

Conclusions• Not a high enough correlation between low H2O vmrs

(poor retrievals) and flagging anomaly to warrant using anomaly as a detection mechanism for poor data;

• Day/night difference between D band spectra due to a large non-LTE feature appearing in the daytime at ~ 2350 cm-1;

• D band more sensitive to clouds? Maybe, but probably not – and certainly not reliably so!D band flags cloud where there simply isn’t anyD band doesn’t consistently flag a blackbodyStrange features in anomalous spectra (non-LTE and

negative)

Recommendations: Stick with tried-and-true A band. D band simply not reliable.