1. Defining the Products: ‘GSICS Correction’
Tim Hewison
Marianne König

2. Scope What we need to decide: Draft Action GDWG03_17:
GRWG to define the content of their NetCDF files with support from the GDWG. End of March 2009.
Includes input and output datasets and intermediate files

3. Generic Data Flow GRWG to define content NetCDF files
Includes input and output datasets and intermediate files
Related to Definition of the ‘Product’
The ‘GSICS Correction’
i.e. Last Step in Data Flow
Other Products? (plots, tables, etc)
What needs to be published?
Correction algorithm
Input parameters needed (‘Statistical Data’)
Zoom on Last Step
Ensure we are clear
Can we make in generic?

4. Generic Data Flow Applying the Correction Statistical Data User Input
From time-series of inter-calibrations:
Satellite/Instrument/Ch
Time Series of Regression Coefficients
Version numbers
Users take the GSICS Correction and apply it to input data:
Generic Data Flow
User Input
Satellite/Instrument/Ch
Date/Time
Geometry (Lat/Lon/IncAng/AzAng/PolAng/SunAng, …)
Radiances/Counts
Correction Algorithm
e.g. Look-Up Table, FORTRAN subroutine, New calibration coefficients, …
Is this generic?
What is missing?
How can Correction be
Implemented?
Distributed?
Corrected Radiances
With Uncertainties

5. Defining ‘Statistical Data’
If we are happy with this generic form of correction Need to Define ‘Statistical Data’:
Contents
Satellite/Instrument/Channel
Time Series of Regression Coefficients with Uncertainties
Version numbers
Anything else?
Format
NetCDF
To be defined by GDWG

6. How Could Correction be applied?
In general, the correction is a function of multiple variables
e.g. Date, Time, Radiance, Incidence Angle, etc…
Need to establish sensitivity of bias to each of these
Need to develop algorithms to apply GSICS Corrections
Although these algorithms do not all need to be the same, we can consider generic correction methods:
Divide dataset into homogeneous chunks and correct each independently
E.g. Monthly mean biases, near-nadir, over sea, in tropics, etc…
Combine all inter-calibration results and develop multivariate correction
E.g. Multiple Linear Regression, Kalman Filter, etc.
Some combination of these
E.g. If sensitivities & non-linearity constant with time

7. Generic Trend Correction Algorithm
New Result Add to Time Series
Trend Significant
Trend Insignificant
Bias Significant
Time-dependent correction
Constant correction
Bias Insignificant
Time-dependent correction?
No correction
Inconsistent
Consistent
Test Consistency with Previous Results
Calc Trend in Results
Sound Alarm Reset Trend Stats
Calc Timescale over which trend produces significant changes
Insignificant
Test significance of Bias
Smooth results over this timescale
Could apply same method to correct other dependences in the calibration
e.g. Time of Day, Latitude, Scan angle,…
No Correction
Significant
Test significance of Trend
Time-dep. Correction
Constant Correction
Significant
Insignificant

8. General Structure of ‘Statistical Data’
What do we need in ‘Statistical Data’ to apply a correction?
In general, the correction may be a function with multiple inputs:
Satellite, Instrument, Channel
Date, Time
Geometry (Lat ,Lon ,IncAng, AzAng, PolAng, SunAng, …)
Radiances/Counts
Need to establish which of these input parameters are inter-dependent
Can we assume geometric corrections are constant with time, radiance?
Can we assume non-linearities are constant with time, geometry?
Difficult to automate this decision process
May be different for application to Near Real-time and Archive data

9. Proposed Content of ‘Statistical Data’
What do we need in ‘Statistical Data’ to apply a correction?
Time Series of Regression Coefficients (with uncertainties)
(Can build Look Up Table from this??)
Statistics of these time series:
Date/time of last trend reset
Alert flag for results not consistent with trend
Time constant over which inter-calibrations should be averaged?
Trend statistics – slope, offset (with uncertainties)
Sensitivity of bias to radiance (with uncertainties)
Sensitivities of bias to geometric factors (with uncertainties)
Lat, Lon, IncAng, AzAng, PolAng, SunAng, …
Assume these sensitivities to be constant within time constant (??)
What does correction mean in terms of Tb bias for Reference Scenes? (±)

10. Draft NetCDF format for Inter-Calibration Results
Coalesce time series of instantaneous results into ‘Statistical Data’ file and include:
Trends & Statistics of time series
Sensitivity of bias to other factors
Calculated over longer periods

11. Discussion Points Is this concept generic?
Could it be applied to your inter-calibration to produce a GSICS Correction?
What ‘Statistical Data’ is needed to apply GSICS Corrections?
Time Series of Regression Coefficients
Trends and Statistics of these time series
Sensitivity of bias to radiance, geometry, etc.
Can we define sensitivity of results in a generic way?
To factors such as time, geometry and radiance
These variables can be inputs into the Correction functions
How can Correction be
Implemented? E.g. Start simple and aim to evolve?
Distributed?