1. ADB Assimilation Activities
Porting of All-sky code into GSI (with Yuanfu Xie)
Modularization of code with OO design
Feed various atmospheric, aerosol & land surface inputs
Use this for validation along with assimilation?
HRRR Alaska Retro run
Develop work flow / XML script
Takes a delayed real-time run
Runs without radar (control) and with radar (test)
Good domain to test all-sky synthetic imagery
Use this domain to derive visibility from camera images

2. Alaska no-radar run (Nov 10, 2016)

3. Alaska with-radar run (Nov 10, 2016)

4. Alaska run (Jan 26, 2017)
real-time
with radar no
radar

5. Alaska no-radar run (Jan 26, 2017)

6. Alaska with-radar run (Jan 26, 2017)

7. Alaska Cloud Verification (3000’ ceiling)

8. Alaska Cloud Verification (any cloud)

9. Alaska Cloud Verification (3000’ ceiling)

10. Alaska METAR Temp Verification

11. Alaska METAR Dewpoint Verification

12. Alaska HRRR radar assimilation?

13. Alaska Radar Run Comments
Real-time run may not be assimilating radar data
ERROR messages in radar stdout files
DPQC radar files being used, archive just has QC files
Only difference is removing ‘obs/reflect’ radar directory
Namelists are being kept the same
NCL graphics are missing the map overlay
GRIB output with Theia UPP version has incorrect projection in header
Only a subset of the domain gets plotted

14. Alaska RUA / HRRR cloud comparison

15. Alaska RUA / HRRR Sim-IR comparison

16. Alaska RUA / HRRR ceiling comparison

17. Alaska RUA / HRRR cloud-top comparison

18. Recent Feedback from Alaska
Unforecast Icing conditions are being reported
Wind direction driving mountain waves
Sufficient VV for -30C supercooled cloud liquid
Identify case date to look at RAP and possibly rerun HRRR

19. ADB Assimilation Activities
Simulated Visible Satellite
Code is being modularized for porting to UPP
Fast running approach
Useful for qualitative or semi-quantitative evaluation
All-Sky software improvements
Nighttime – including city lights and airglow (with CIRA)
Viewing from above the atmosphere (DSCOVR comparisons), starting to read in FIM 3-D cloud fields
High aerosol optical depth (AOD) cases
Consider Monte-Carlo techniques (e.g. PSD package)?
Data Display
Potential use of “on-the-fly” web viewing interface?

20. EMB Assimilation Activities
Variational Cloud Analysis work?
Consider concepts for blending satellite + METARs + First guess
Vertical partitioning of hydrometeors
Addition of radar data
What satellite data can be used?
Cloud-top pressure (existing)
CLAVR-X (also includes CWP, optical thickness)
Visible satellite (albedo)
GOES-R cloudy radiances (with CRTM), for 3D and 4D analysis
Following slides summarize earlier var cloud work in FAB

21. Variational Cloud Analysis
Modular variational cloud analysis currently under development
Based on existing LAPS and GSI cloud analyses
Simultaneous solution allowing merging of all data sources
Hydrometeor control variables (qc, qi, qr, qs, qg)
Use satellite radiance (e.g. CRTM) or retrievals (e.g. DCOMP)
Appropriate forward models and physically based constraints
Include all-sky cameras for validation and as input data
Establish testbed evaluating OO variational cloud analysis using constraints - augmenting ensemble methods
Assess physical and statistical constraints
Compare various approaches for GSI inclusion

22. Datasets of Global Interest
Geosynchronous
IR / Visible
GOES / GOES-R / Meteosat / Himawari
Polar Orbiting
JPSS
VIIRS (note nighttime / lunar capability)
GPM Constellation
Active Radar + Microwave
Microwave is available on more satellites and can be calibrated using the colocated radar on the GPM core satellite
Lightning
Forward model (e.g. for upward graupel flux)
Consider lightning is more predominant over land

23. All-sky Camera Assimilation
More cameras, via NOAA’s observing systems?
Add to ASOS?
FAA camera networks (e.g. Alaska)
Airborne cameras?
CSTAR / AWIPS
Data assimilation with variational cloud and GSI analysis
Efforts underway to use GSI cloud/hydrometeor analysis (used in HRRR/RAP) with all-sky forward model for nowcasting.
Use derived METAR, cloud mask, image correlation, or spectral radiances
Check applicabilty of available RTMs

24. All-sky Camera Assimilation
Camera image on the left. Cloud mask on the right is being interfaced with the GSI for assimilation.

25. All-sky Camera Assimilation
Clear sky simulated image panorama generated by GSI. A cloud mask associated with this forward model will be compared with camera cloud masks to clear out clouds. Clouds can be added as part of a variational analysis. These images (together with spectral radiance values) can also be used to evaluate analysis/forecast cloud 3D placement, microphysics, and radiation.

26. Visibility Visibility Measurement
Contrast between land and sky (on the horizon)
Contrast of features on the land scape (e.g. a runway)
Extinction coefficient (α) (related to visibility, AOD and aerosol scale height, as well as other hydrometeors)
Visibility = -ln(.05) / α
Measured with a transmissometer
Contrast is sensitive to both extinction “out-scatter” and
“in-scatter” or airlight due to aerosols, gas, hydrometeors
Simulated imagery provides 2-D radiance field over a 360 degree spherical FOV to help interpret these phenomena in camera images.

27. Visibility Visibility Measurement (cont)
Note that AOD can be inferred variationally from clear sky radiance and 3-D extinction coefficient can in principle be constrained from calibrated camera images of the entire sky & landscape. Simpler retrievals can also be considered.
Link between extinction coefficient and AOD (assuming exponential decay of aerosols with height)
α = AOD / scale height

28. Distance to Terrain (km)
Visibility
Distance to Terrain (km)
Possibly help augment LL edge detection algorithms.

29. Ceiling is low obscuring the terrain with simulated clouds
Visibility
Clear Sky
Low Clouds
Ceiling is low obscuring the terrain with simulated clouds

30. Full Disk Earth View Earth global view SIMULATED (FIM 00h fcst)
OBSERVED (DSCOVR)
Earth global view
Compare with
DSCOVR or Himawari

31. Variational Cloud Strategies
Choose Object / Fwd Model for Satellite / Radar Obs
Constraints
Physical
Statistical OR OR OR
Water Vapor RH & Q
Saturation Q
+ other state vars
These five constants represents the coefficient of hydrometer concentration, with the unit of 1/kg/m.
Cloud optical depth is equal to the integration of hydrometers from surface to model top vertically.
Cloud Tau, LWP, Aerosol Retrievals
(e.g. CLAVR-x)
IR B-temp
VIS Albedo
Goddard Satellite
Simulator
CRTM

32. Variational Cloud Observation Operator
Courtesy Juxiang Peng
Cloud optical depth (satellite retrieval) =
hydrometeor water content analyzed as control variables
Units: m
Hydrometeor Water Path
These five constants represents the coefficient of hydrometer concentration, with the unit of 1/kg/m.
Cloud optical depth is equal to the integration of hydrometers from surface to model top vertically.
constants account for
scattering cross-sections
effective particle radius

33. Variational impact on Cloud Ice Content at 225mb Sept. 13, 2013 1500UTC
Here are the cloud ice at 225hPa.The left is analyze from using cloud optical depth ,and the right is analyze from using traditional LAPS
The left Cloud ice seems add some detail characteristic when compare with LAPS
vLAPS + Satellite Cloud Optical Depth
vLAPS + non-var Cloud

34. Observation Operator with Temperature Constraints Added on Hydrometer Type
Cloud phase
In order to use cloud phase as constrain condition in minimization, add new coefficient in front of each hydrometer. The new coefficients control the hydrometer as switch. It’s value is based on Background temperature
Here the cloud ice and cloud water reference temperature temporarily set to be zero.
Courtesy Juxiang Peng

35. Cloud Ice Content along 40N Sept 13, 2013 1500 UTC
The Y axis is pressure of STMAS,and X axis is longitude .The left is from COD scheme and the right is from LAPS scheme.
Here is obviously diffenence between the two scheme.On the left hand ,the hight of cloud ice is higher.
Tradition LAPS analysis cloud height from 1200 m to 20000m. STMAS analysis cloud height from sfc to model top.
The large height of cloud ice corresponding to convective area.
kg/meter**3
kg/meter**3
vLAPS + Satellite Cloud Optical Depth
vLAPS + non-var Cloud
Courtesy Juxiang Peng

36. Variational Cloud Analysis Experiments
CRTM
Forward model
K-matrix model
Minimization
Surface Info
Control variables
Simulated radiances
Gradients
Atmospheric state along trace
Surface characteristics in FOV
Sensor and source locations
Special setup for each obs profile
Load coefficients
Atmosphere structure
Surface structure
Geometry structure
Options structure BK
Obs
Converge? Y
Update
Solution N
Satellite radiances as input
Design and coding underway
Interface
DA system
Now, we are building the interface between CRTM and our DA system.
The interface mainly needs to complete two pieces of work: one is loading the CRTM coefficient data files. The other is delaring and filling four structures.
Atmospheric structure is filled with the atmospheric state along the scanning line. The atmospheric state includes the distribution of hydrometeors and aerosols.
Surface structure is filled with the surface characteristics in each FOV.
Geometry structure is filled with the locations of sensors and sources.The source is typically the sun or moon.
Options structure allows users to choose the certain kind of scattering RTM, to enable and disable the antenna correction and so on.
After all the above setup, CRTM is called. Simulated radiance and gradients are produced by its forward model and k-matrix model respectively.
Then, simulated radiance and gradients are used for the minimization of the cost function. If convergence is reached, we get the solution; else the control variables are updated and the iterative loop continues.