Project Title: Detection and Correction of Aerosol Contamination in Infrared Satellite Sea Surface Temperature Retrievals
Principal Investigators: James Cummings, Doug Westphal Naval Research Laboratory, Monterey,
CA
Co-Investigators: Jeff Hawkins, Doug May, Andy Harris
Budget: $110 FY03 $115 FY04 $150 FY05
Talk Outline: Project Objectives and Tasks
Progress to Date
Future Plans
JCSDA Workshop onSatellite Data Assimilation
Project Project Objectives
• Detection of aerosol contamination in infrared satellite sea surface temperature (SST) retrievals using Navy Aerosol Analysis Prediction System (NAAPS) aerosol distributions.
• Correction of satellite SSTs for aerosol contamination using NAAPS aerosol products.
JCSDA Workshop onSatellite Data Assimilation
Project TasksProject Tasks
• Collocate NAAPS optical depth forecast fields valid for the time SST retrievals are generated (Doug May, NAVOCEANO).
• Estimate SST retrieval reliability relationship to AOD content (Doug May, NAVOCEANO - Jim Cummings, NRL)
• Develop SST quality control schemes to recognize aerosol contamination (Jim Cummings, NRL).
• Correct satellite SSTs for aerosol contamination (Andy Harris, NESDIS).
• Validate NAAPS aerosol products using using independent data - improve NAAPS model (Jeff Hawkins, Doug Westphal, NRL).
JCSDA Workshop onSatellite Data Assimilation
SST Retrievals and NAAPS Collocations at NAVOCEANO
• On going since February 2004– NAAPS AOD forecast fields obtained via ftp from NRL 4 times daily– Append AOD value closest in time and location to each MCSST retrieval
• total AOD used (sum of dust, smoke, sulfate components)• globally for N-16 and N-17 (26 Jan 2004)
– Global SST observation data file with NAAPS AOD values provided daily at 1000 UT to US GODAE server in Monterey
• New capabilities added May 2004– NAAPS AOD components plus total AOD collocated with MCSST– Cloud cleared radiances for AVHRR channels 3,4,5 saved with AOD values
JCSDA Workshop onSatellite Data Assimilation
QC of SST Retrievals with NAAPS Collocations at NRL
• Develop discriminant analysis functions to distinguish aerosol contaminated vs. uncontaminated SST retrievals
• SST retrievals from verified Saharan dust events are used as training data sets
• Discriminant functions computed using NAAPS AOD components (dust, sulfate, smoke), AVHRR channels 3,4,5 brightness temperatures, and SST innovation from 6 hourly global 9 km SST analysis
• Provides probabilistic framework for QC – outcome is probability SST retrieval is contaminated– allows simple query capability when gathering data for assimilation
JCSDA 2nd Workshop onSatellite Data Assimilation
Case 20050212Case 20040725
QC Discriminant Analysis Training Data Sets
• Jun 2-6, 2004
• Jul 15-17 and 20-25, 2004
• Sep 12-15, 2004
• Oct 10-13, 2004
• Nov 2-4 and 6-8, 2004
• Dec 13-15 and 28-29, 2004
• Jan 5-8, 2005
• Feb 10-13, 2005
12:48 GMT 14:18 GMT 15:55 GMT
Case 20040725: Visible & AOD
s u n
g l
I n t
s u n g l i n t
Resultant Composite AOD Image12:48 GMT 14:18 GMT 15:55 GMT
Composite
NPS AOD Image Reduction: Matching NAAPS DomainFinal: 20 X 20 pixelsOriginal: 2250x1200 pixels Intermediate: 60 x 60 pixels
NAAPS Dust AOD valid: 2004072512
0.1 0.4 1.6 6.4
MODIS (GSFC) AOD Image Reduction: Matching NAAPS DomainFinal: 20 X 20 pixelsOriginal: 2250x1200 pixels Intermediate: 60 x 60 pixels
NAAPS Dust AOD valid: 2004072512
0.1 0.4 1.6 6.4
202 observations 231 observations
NAAPS vs NPS AOD (left) &
NAAPS vs MODIS (GSFC) AOD (right)
AOD .25 .30 .40 .50R2: .68 .53 38 .29
AOD .50
R2: .37
202 observations
Scatter Plot: NPS vs NAAPS AOD for Case 20040725
NAAPS vs NPS AOD
Suggested Improvements
• Cloud Filtering
• Conversion of Image data to NAAPS grid
• Include AERONET measurements
RT Modeling of aerosol effectsConsider Merchant et al. notation…
SST = aTk k is aerosol ‘mode vector’
a is vector of retrieval coefficients
So, need to ascertain weights of mode vector for 3.7, 11 and 12 µm channels, i.e.
12
11
7.3
T
T
T
k
Effect on brightness temperatures
Dependency on total transmittance
Primary cause of scatter is attenuation of near-surface aerosol effect by intervening atmosphere
Can be mitigated by linear fit to total clear-sky transmittance
Different aerosol types have significantly different coefficients
Role of air-sea temperature differenceResidual error in fit depends on air-sea temperature difference
Magnitude and range of ASTD-dependence is a function of total clear-sky transmittance
Could parameterize ∂T/∂Χ as a function of both and ASTD…
dzTT
TTBTTBTBT
zzA
ASTi
iAS
i
S
...where
Suggested form of k-estimation
ASAS
ASAS
ASAS
TTcTTcccTTbTTbbbTTaTTaaa
12321210
11321110
7.3327.310
k
k-coefficients will be different for different aerosol types
Conclusions from RTM studiesMerchant et al. approach requires modification in the tropospheric case because aerosols are not at the top of the atmosphere
Similar reason for greater success of Nalli & Stowe methodology in stratospheric aerosol case
Addition of total atmospheric transmittance (from NWP or e.g. SSM/I water vapor) should assist in correcting for much of the scatter
Air-sea temperature difference (NWP) useful addition
Still need discrimination of aerosol type (e.g. via NAAPS)
Conclusions from RTM – part 2NAAPS data can permit full RT treatment of problem, but this is costly → reduced predictor approach proposed here
More work required in order to develop and validate this approach
May be desirable to adopt an interim empirical approach using satellite-derived AODs (analyses) and ancillary clear-sky transmittance, air-sea temperature differences (NCEP fields?) Beware of cross-talk between AOD & WV, ASTD
Stratospheric aerosols have much greater impact for given AOD – suggest using alternative sources (e.g. HIRS retrieval, or another analysis/product)