Detailed conversations about the Multi-sensor Matchup System developments.
1. Satellite image collection strategy where imagery does not wholly overlap.
Image collection is based on a fixed number of pixels for different sorts of sensor. For the most part, these extract sizes are designed to ensure coverage of the 101 x 101 pixel image from ATSR (main reference sensor), which is matched at its centre to an in situ observation.
Ideally, the overlaps would look like the above. But where a further sensor cannot fully cover the 101 x 101, there is an option on whether to extract the full number of pixels ...
... as above, or else, fill the uncaptured areas with fill values ...
... and we agreed the latter. This means that the centre pixel of any extract is always the one nearest in location to the reference point, which is much easier of users of datasets to deal with.
2. Strategy for numerical weather prediction fields.
Some MMS content will be created by applying simulations to NWP fields. The treatment of NWP needs to be as close as possible to the processor treatment in these cases.
We agreed that NWP would be interrogated on demand within the MMS, not pre-extracted and stored.
The extracted data necessary to support creation of additional content (e.g., simulated BTs, cloud masks) for an ATSR and GAC multisensor match [MM] will include NWP extracted on demand over a space-time domain permitting the full interpolation process over all the imagery in the MM. When MMDs are extracted later for algorithm development or RRDP purposes, etc, the NWP extraction can be simpler for ease of use (e.g., just one NWP profile, interpolated to the reference point).
3. Flags, instrument information and location
Original flags will be kept in the MMS. A simple good/bad flag will be made for retained pixels in image extracts that will be based on rules to be defined: OE to advise RQ for ATSRs; MF to advise for AVHRRs.
A common land-sea mask will be used to create a consistent flag for all sensors within the MMS, based on an existing Brockmann tool. For this, RQ will need a definition of a pixel size to assume (as a function of satellite zenith angle in some cases); OE and MF to advise.
In the case of ATSR1 there is a need to know the detector temperature. The ARC system uses 5 minute files (interpolated in time). OE will write a short description of the basis of these, and give information to GC for DARD about them. The note should also describe the models used for NeDTs.
Time information will be kept per row of image extracts.
4. Processing for additional fields for ATSR
The additional fields required for the RRDP are RTTOV BTs and their tangent linear with respect to SST, and the Bayesian cloud mask.
OE will provide an executable that will take in extracted netCDF auxiliary information and the orbit/image locator information and will output the above in netCDF. The executable will be a configuration of the ARC processor, in effect.
5. Processing for additional fields for AVHRR
The additional fields are calibrated BTs and flags (using Jon Mittaz methods) and the CLAVR-X post-processed cloud mask. It has turned out that the operational CLAVR-X mask available with some recent AVHRR GAC is of limited consistency with the post-processed version, and therefore it is a new requirement to process the whole GAC archive to get a CLAVR-X mask. This will be a delay for MMD extraction, most likely. However, the mask will be static (need be done only once).
So the agreed approach is:
MF to run all GAC data to generate netCDF files sufficient to geolocate by scan line and element for all AVHRR orbit files. In addition, he will output and keep the whole orbit CLAVR-X mask and basic optimal estimation SST results in netCDF as a static output for reference later (for efficiency reasons).
RQ will define to MF the list of extracts (as orbit, line, element).
MF will provide an executable that for a given orbit/line/element will output:
* calibrated BTs etc for extracts (this requires smoothing of calibration information beyond the extract boundaries)
* RTTOV BTs and tangent linear wrt SST for all extracts
* CLAVR-X mask for extract
* additional flags from JM reader method