Assess perceptual quality of satellite precipitation nowcasts

Identify and validate quantitative perceptual quality assessment methodologies tailored to satellite-derived precipitation nowcast images (such as GOES-16 RRQPE outputs from models including TUPANN and GAN-TUPANN) to rigorously evaluate visual realism and align it with forecast utility.

Background

Beyond conventional skill metrics like CSI and HSS, the authors note that GAN-based enhancements improve visual realism yet do not consistently translate into better quantitative performance, underscoring the need for appropriate perceptual assessment tools.

The Limitations and future work section explicitly identifies the assessment of perceptual quality as an open challenge, reflecting a gap in domain-specific metrics for evaluating the visual fidelity of satellite precipitation nowcasts.

References

Finally, GAN-based enhancements improve visual realism but degrade or inconsistently affect skill metrics; stabilizing adversarial training and assessing perceptual quality remain open challenges.

— Precipitation nowcasting of satellite data using physically-aligned neural networks (2511.05471 - Catão et al., 7 Nov 2025) in Section 6, Limitations and future work