GSICS Microwave SNO / Vicarious Calibration Focus Group meeting 2025-11-13

GSICS Microwave SNO/Vicarious Calibration Focus Group meeting

Agenda

1. Noemie Lalau (Magellium) – Contribution of microwave radiometer calibration to altimeter uncertainty budgets
2. Jorge Gill (Deimos-Indra) – Spectral Band Adjustment Factors to account for SRF differences
3. Joe Munchak (Tomorrow.io) – Inter-calibration of Tomorrow.io Microwave Sounder constellation

Attendees

Tim Hewison (Chair)

TBD

Summary

Noemie Lalau (Magellium) – Contribution of microwave radiometer calibration to altimeter uncertainty budgets

Noemie showed that the Wet Tropospheric Correction is a dominant source of uncertainty on sea level trends from altimeters, and that this is directly related to the calibration of the microwave radiometers that provide that information operationally. Uncertainties in the global trends can be reduced by using WTC from a water vapour CDR (HOAPS). The group plan to use WTC from CDRs to asses MWR WTC in Sentinel-6 Next Generation cal/val plan.

Altimetry MWRs are getting ever more complex – on-board calibration, more channels, higher frequencies and scanning in future – and their calibration is critical for their primary application. S6NG requires a MWR with calibration stability of ~0.015K/yr, which is difficult to verify by L1 vicarious calibration. So much of these comparisons are conducted at L2 (sea level height) or L3 (its trend).

Karsten Fennig (DWD) added that the Climate Monitoring SAF also do comparisons in L1-space as part of their development of the HOAPS CDR used here, and cooperate with other teams involved in L2 and L3 comparisons. CNES are also running a study to compare L1 observations.

The question was raised of whether GSICS should focus more on the inter-calibration of altimeter microwave radiometers – and even of whether they could be considered as reference instruments for some spectral bands.

 

Jorge Gill (Deimos-Indra) – Spectral Band Adjustment Factors to account for SRF differences

Jorge introduced the Spectral Band Adjustment Factors being developed to account for SRF differences as part of a EUMETSAT study to develop algorithms to inter-calibrate channels of the proposed Sterna constellation with other MWRs. This is based on a piecewise linear function (not a LUT), based on CDF matching of simulated observations. Other methods were attempted, but with more limitations. For example, the weighted linear regression used in the SBAFs for the generation of FCDRs of geostationary infrared radiances, and neural networks, which worked well on training and validation data, but failed when applied to real data. Jorge found that extremes need to be excluded when applying the SBAFs due to instability – although no physical cause has been established.

It was highlighted that the training dataset could introduce biases – for example due to unrepresentative diurnal sampling – and that different SBAF are needed for different scene types.

There is an outstanding question of how to handle uncertainty contribution from SBAF in inter-calibration algorithm error budget. This will be followed-up in the study and reported back to GSICS.

Action:

A.GMW.20251113.1: Tim Hewison (EUMETSAT) to consider sharing the SBAF algorithm, scripts and datasets with GSICS.

 

Joe Munchak (Tomorrow.io) – Inter-calibration of Tomorrow.io Microwave Sounder constellation

Joe introduced the Tomorrow.io Microwave Sounder, based an evolution of TROPICS. Its L1B calibration is based on views of deep space, noise diodes and an internal blackbody (based on TKRAM), with corrections are applied based on comparisons with ERA5 and double-differences, based on MiRS retrievals from ATMS. There are known biases in modelled data at 89GHz in the TMS polarisation due to SURFEM.

Tomorrow.io also generate L1C products, which include inter-calibration to compensate for radiometric and spectral differences between the satellites in the constellation. This is based on a QRNN, using simulation data from 40-day OSSEs (a Jan-Mar period, so including snow-covered land). This training data included simulated noise with realistic inter-channel correlations, but no 1/f noise – which could be added in a future version. A similar algorithm is used for limb adjustment, which is applied in a further product, L1C-TCR, which also includes remapping the observations to a common grid for ease of visualisation and L2 product generation. The QRNN also provides error estimates – which has nice gaussian distributions.

Tim noted the FIDUCEO terminology, the process used to generate the TMS L1C products is homogenisation, whereas other GSICS corrections are based on harmonisation, which requires the SRFs of each instrument be applied when using the data – e.g. when assimilating into NWP, but does not introduce additional uncertainties.

An alternative approach to providing inter-calibration information was discussed, whereby correction functions are distributed instead of L1C data. However, as the TMS L1C corrections are derived by QRNN, this may be more difficult to distribute – although they are static for each version of the data.

In response to a question about how much the calibration varies in each orbit, Joe explained that temperature variations cause changes in gain, and also noise diode brightness temperature, which are pre-calculated. He added that analysis suggests the ICT thermistors  were stable and accurate, except in some 45° orbits due to solar intrusion.

Outcome

TBD