Test cross-sensor transfer across divergent scanning geometries

Investigate whether feature transfer and joint training remain effective when adapting the artifact-detection framework from conical-scanning instruments such as SSMI and SSMIS to sensors with divergent scanning geometries, such as cross-track scanners.

Background

The paper evaluates Warm-Start and Joint training strategies using SSMI and SSMIS data. Because both instruments use conical scanning and therefore have broadly compatible field-of-view characteristics, the reported experiments do not establish whether cross-sensor transfer generalizes to substantially different instrument geometries.

This unresolved issue is particularly relevant to CubeSat radiometers, which commonly use cross-track scanning and can exhibit field-of-view elongation at high scan angles. Determining the efficacy of transfer or joint training across conical and cross-track sensors would clarify whether these strategies can support deployment on heterogeneous satellite platforms.

References

The efficacy of feature transfer or joint training across divergent scanning geometries, such as adapting from a conical to a cross-track scanner, remains untested.

— A Sensor-Adaptive Incremental Learning Framework for Artifact Detection in Satellite Precipitation Data  (2609.01514 - Monsalve et al., 1 Sep 2026) in Section 5.2, Methodological Limitations

The training geography is the main limitation of the student. The teacher exists only in the GOES-East/West overlap, so the student samples the Americas and eastern Pacific and its transfer to GK2A or SEVIRI is untested.

— Distilling deep optical flow stereo methods to retrieve dense three-dimensional wind fields  (2609.03100 - Vandal et al., 2 Sep 2026) in Discussion and Conclusion