Post on 16-Jan-2016
A NON-RAINING 1DVAR RETRIEVAL FOR GMI
DAVID DUNCAN
JCSDA COLLOQUIUM
7/30/15
The GPM satellite is in a non-sun-synchronous LEO at 65° inclination
Together, GMI and the Dual-frequency Precipitation Radar provide an active-passive combination designed for measuring light to heavy precipitation, rain and snow
GMI is nearly identical to TMI (17 functional years—don’t change it!) but with additional high frequency channels
GMI is the calibration standard for the GPM constellation
GPM MICROWAVE IMAGER (GMI)
NON-RAINING PARAMETERS AND GMI
Why develop a non-raining retrieval for GMI? Isn’t GPM tailor-made to sense rain and snow?
Other imager algorithms are sensitive to assumption of water vapor distribution—so I try to solve for it
Using GMI as an ‘ideal’ sensor to develop code and methods that can be applied to other sensors (AMSR2, SSMIS, etc.)
GMI is one of the best absolute-calibrated sensors in orbit (according to X-Cal), thus a good test bed for a new approach
‘Non-raining parameters’ means total precipitable water (TPW), wind speed, and cloud liquid water path (LWP) over ocean—also called ‘Ocean Suite’
From Imaoka et al. (2010)
Wind effect on Tb (RSS emissivity model)
NON-RAINING PARAMETERS AND GMI
OPTIMAL ESTIMATION / 1DVAR
Through iteration, the cost function is minimized to find an optimal solution to the inversion, given the measurement vector (Tbs) and a priori knowledge of the environment and the measurement:
Φ = (x-xa)TSa-1(x-xa) + [y-f(x,b)]TSy
-1[y-f(x,b)].
Iterate to find state vector that minimizes the difference between observed Tbs and simulated Tbs
One huge advantage to 1DVAR is output error diagnostics (posteriori errors) that come out of the formalism:
Sx = (KTSy-1K + Sa
-1)-1
To execute a 1DVAR retrieval, you need:
1. Prior knowledge of the environment
2. A good forward model—a method of modeling the atmosphere and surface to simulate what the satellite sees
Radiative transfer model Surface emissivity model Assumptions about the atmospheric profiles of water vapor,
cloud water, etc.
3. Knowledge of channel errors and their covariances
1. PRIOR KNOWLEDGE OF THE ENVIRONMENT
From analysis of ECMWF Interim Reanalysis 6-hourly data, LUTs consist of means/variances/covariances of 10m winds EOFs of water vapor, broken up by SST.
Mean10m Wind
Std DevWind
2. FORWARD MODEL
NOAA’s Community Radiative Transfer Model (CRTM v2.1.3)
Emission/absorption only—not a bad assumption in microwave unless rain or lots of ice present
User has option of FASTEM5 (or FASTEM4) or RSS ocean emissivity model
16 vertical layers defined by pressure
Liquid water cloud set at 850-750mb
Ice cloud may be added as well, and scattering turned on, but no skill in retrieving IWP currently
Reynolds OI SST used as base temperature, though retrieval shows some skill at retrieving SST if it’s allowed to vary
SST used as index for climatological mean water vapor profile from ERA-Int
10m wind, CLWP, SST and 3 EOFs of water vapor are retrieved parameters
3. CHANNEL ERRORS
Determining forward model error is an omnipresent issue for satellite retrievals
Most retrievals make up numbers, or at best assume a diagonal Sy matrix in which there are no channel error covariances
Sa determination is easier, since that can be taken from a model
Sy is necessarily different for every sensor, every forward model used!
How to get a ‘real’ Sy matrix?
The approach:
Run both the simplified forward model of the retrieval AND the fullest forward model possible, then analyze the difference:
(TbS, Simplified – TbO) - (TbS, Full – TbO) = TbS, Full – TbS, Simplified
3. CHANNEL ERRORS
Full
Simplified
3. CHANNEL ERRORS
A good retrieval needs a good Jacobian (δTb/δx), which stems from Sy
The approach takes into account all simplifying assumptions: no scattering, no ice, fewer levels, etc.
Attempted to screen out rain, sea ice, RFI-contaminated pixels
Even ECFull has trouble, especially at middle frequencies, though other channels in this analysis largely matched Xcal results
What about channel biases?
Success of retrieval depends heavily upon how Sy is formed!
RESULTSGLOBAL IMAGES—NON-RAINING PARAMETERS
LWP [mm]
TPW [mm] Wind [m/s]