1. NMSC Daytime Cloud Optical and Microphysical Properties (DCOMP) 이은희

2. NMSC DCOMP  COD (Cloud Optical Depth)  CPS (Cloud Particle Size)  LWP (Liquid Water Path)  IWP (Ice Water Path)  Observing System Overview  Algorithm Description  Test Datasets and Outputs  Practical Considerations  Assumption and Limitation

3. NMSC OBSERVING SYSTEM OVERVIEW

4. NMSC Products Generated  COD : vertical optical thickness between the top and bottom of a cloud column : independent of wavelength in the visible range : no unit  CPS : the cloud droplet distribution : cloud effective radius : unit = micrometer (μm)  LWP/IWP : the total mass of water in a cloud column. : unit = g/m 2

5. NMSC Products Generated  Daytime : the solar zenith angle is less than or equal to 65 degree : To fill a temporal gap between DCOMP and NCOMP degraded products for solar zenith angles between 65 and 82 degrees  Nighttime : solar zenith angles greater than or equal to 90 degrees : algorithm use channels 7,14,15  Cloud Mask : clear, probably clear, probably cloudy and cloudy

6. NMSC DCOMP requirements specifications

7. NMSC Instrument Characteristics  GOES-R ABI : spatial resolution - IR ~ 2km - VIS ~ 0.5km : Important new feature - 2.26μm channel for better particle size retrieval  Proxy data : SEVIRI(MSG) : ABI 2(0.62 μm) and 6(2.26 μm) channels

8. NMSC ALGORITHM DESCRIPTION

9. NMSC Algorithm Overview  COD & CPS characterize the impact of clouds on the energy and radiative budget of the Earth.  So, they are used to parameterize clouds in global climate models, are critical to improving climate models  Knowing CPS and COD also enables retrievals of the amount of water within the cloud  This value into LWP and IWP to correspond with the dominant water phase in the cloud

10. NMSC Algorithm Overview  DCOMP is based on earlier methods from VIS and NIR [Nakajima and King, 1990,1992] - COD : use absorption-free wavelengths VIS, is determined by the amount of light scattering by cloud droplets - CPS: use an absorption solar channel (NIR), is mirrored in absorption amount of clouds - Radiance reflectivity, To avoid an overestimation of shortwave channels, where the solar irradiance is bigger than for longer wavelengths

11. NMSC Algorithm Overview  RTM is used to solve the forward problem. : t, q profile satellite sensor signals (simulating the transfer of solar radiation)  Optical properties from satellite radiances is the inverse problem. 1D-var optimal estimation approach.  Solving the RT equation for a single-layered, homogeneously distributed cloud above a Lambertian surface.  The current COD/CPS algorithm is implemented in the NOAA/NESDIS AIT processing framework.

12. NMSC Processing Outline  Required GOES-ABI products : cloud mask, cloud top press., cloud phase  Also needs 2 kinds of LUTs :The cloud LUTs - reflection, transmittance, cloud albedo, cloud spherical albedo tables :The ancillary data LUTs : coeff. to estimate transmission in cloud-free layers for ozone and water vapor.

13. NMSC Flow-chart  The actual DCOMP retrievals 1. First segments: testing the channel settings and loading all LUTs and the coeff. for WV correction in memory. 2. Each segment: validity test and the aliasing of the framework parameters to local pointers ( validity test: reject pixels a) outside valid sensor and sun angle range b) cloud-free, c) no valid cloud press. or cloud phase) 3. Correct atmosphere for upper layer by estimating the real top of cloud reflectance, by adjusting the TOA measurement, by estimating a virtual surface albedo 4. Observation vector y ( input of inversion of the optimal estimation technique with a modified surface albedo) => COD/CPS 5. LWP, IWP will be calculated from COD/CPS pair

14. NMSC Flow-chart

15. Basic Considerations  Cloud radiation in the shortwave range of IR spectrum are a function of cloud optical depth and cloud droplet distribution n(r)  [Hansen and Travis, 1974]  By using R instead of L – avoid an overstimation of shortwave channels  Cloud optical depth can be expressed

16. NMSC Basic Considerations  [Stephens, 1978], [Bennartz, 2007]  [Heymsfield,2003]

17. NMSC Radiative Transfer Calculations

18. NMSC Atmospheric correction  Rayleigh scattering  Aerosol scattering effects  Water Vapor Absorption and ozone

19. NMSC Optimal Estimation Inversion Technique

20. NMSC Flow-chart  The retrieval loop - iterative 1D-var optimal estimation(OE) technique 1. Definition of a priori values of the state vector and observation & atmospheric state covariance matrices. 2. The cost will be calculated for each iteration step. 3. Find the minima value on a cost surface function 4. If the cost falls below a pre-defined threshold, the solution is found and the retrieval loop will end. Otherwise, if a maximal number of iterations is exceeded, no solution could be found (not converged)

21. NMSC

22. Algorithm Input  Primary sensor data

23. NMSC  Ancillary Data : land mask, Surface clear sky reflectance (Albedo) snow mask, NWP data, LUT

24. NMSC  LUT

25. NMSC Algorithm Input  Derived data : cloud mask – determine which pixels are cloudy or cloud free : cloud top pressure – determine the amount of absorber mass by WV above the cloud for atmospheric correction : cloud phase – determine which LUT, ice or water are used for forward model calculations. : snow mask – flagging each pixel as snow or clear ( 11μm BT > 277K, then turn off)

26. NMSC Algorithm product Output  Data product - 2 float-typed datasets: cloud optical depth, cloud effective radius - format: HDF-4  Output - Liquid water path or Ice water path for each pixel  Quality flag - 0: Valid, good quality converged retrieval - 1: Not valid, quality may be degraded due to snow or sea ice sfc. - 2: Not valid, degraded quality due to twilight conditions ( 65° < solar zenith angle < 82° ) - 3: Invalid due to cloud-free condition - 4: Invalid pixel due to being outside of observation range - 5: Invalid pixel due to missing input data - 6: Invalid pixe, DCOMP attempted but failed retrieval

27. NMSC Output  Processing information flag

28. NMSC Metadata  Day/Night flag  Mean, Min, Max and standard deviation of cloud optical depth  Mean, Min, Max and standard deviation of cloud particle size  Number of QA flag values  For each QA flag value, the following information is required: - Number of retrievals with the QA flag value - Definition of QA flag  Total number of detected cloud pixels  Terminator mark or determination

29. NMSC TEST DATA SETS AND OUTPUT

30. NMSC Test Data sets  Simulated/Proxy Input Datasets: : SEVIRI observations (provided by SSEC Data center) : SEVERI SRF function(from EUMESAT) : RTM, LUT design, inversion technique are identical.

31. NMSC Output from Simulated Datasets

32. NMSC Quality flag - 0: Valid, good quality converged retrieval - 1: Not valid, quality may be degraded due to snow or sea ice sfc. - 2: Not valid, degraded quality due to twilight conditions ( 65° < solar zenith angle < 82° ) - 3: Invalid due to cloud-free condition - 4: Invalid pixel due to being outside of observation range - 5: Invalid pixel due to missing input data - 6: Invalid pixel, DCOMP attempted but failed retrieval

33. NMSC Precisions and Accuracy Estimates  Cloud optical parameters are difficult to validate. Because cloud optical thickness is a radiative property, it is not possile to validate from in-situ measurements.  Validating DCOMP products 1. Direct comparison with MODIS : precision & accuracy estimate 2. For liquid water cloud, the use of passive microwave retrievals from AMSR-E and SSM/I 3. A-TRAIN measurements- identify aerosol layer 4. Inter-comparisons with SEVIRI product, additional quality check - consistency check for new algorithm.

34. NMSC Inter-comparison with Products of Other Research Group  GOES ABI algorithm was compared with its SEVIRI counterpart at the EUMESAT workshop in Ascona, Switzerland in February 2009. Strict pixel-base 1:1 comparison for all cloud products  CMS,OCA-EUMESAT, DLR-German Aerospace Center, UKM-UK Met Office, GSF-Godard official products

35. NMSC

36. Caused by underestimating of a-priori error.

37. NMSC Error budget of DCOMP

38. NMSC Practical considerations  Numerical computation consideration - run the algorithm on a 200 scan-line basis to avoid memory issue. - run for full SEVIRI scene in under five minutes  Programming and Procedural consideration - Fortran 90, 1 fortran module+subroutine, include file - pixel by pixel algorithm - Global values were avoided as much as possible. - Use pointer variables for ouput.  Quality Assessment and diagnostics - Missing/No data - Cloud-free - Cloudy, but no convergence - High value of cost function

39. NMSC Exception Handling  If the MODIS surface albedo is missing, use a default value(for land surface) of 0.15  If NWP data are missing, we will use a default water vapor profile.  Algorithm validation – Mainly MODIS comparison

40. NMSC Assumptions and limitations  Assumptions for performance 1. NWP data/ current 6 hourly GFS forecast are available 2. surface albedo values form MODIS are available 3. All of the static ancillary data are available at the pixel level 4. Channel 2 is available 5. Channel 6 is available  No product improvements planned at the moment

41. NMSC THE END THANK YOU

42. NMSC Find a minima of cost function