Multilayer cloud detection is vital for numerical weather prediction. This study embeds channel selections derived from threshold-based algorithms into a 1D-CNN, utilizing machine learning (ML) to learn latent physical relationships for simplifying physical retrievals for operational deployment.
The results show that the 1D-CNN achieves a multilayer-cloud probability of detection ($\text{POD}_{\text{mul}}$) of 0.620 and a false alarm rate ($\text{FAR}_{\text{mul}}$) of 0.240, outperforming the conventional threshold algorithm ($\text{POD}_{\text{mul}} = 0.558$, $\text{FAR}_{\text{mul}} = 0.369$). These results demonstrate that prior physical knowledge derived from radiative transfer theory can serve as an effective feature-engineering prior.
Further experiments reveal that ML-revealed physical mechanisms can also enhance traditional algorithms. Replacing AGRI channel 12 (C12, centered at $10.8~\mu\text{m}$) with channel 13 (C13, centered at $12.0~\mu\text{m}$) increased $\text{POD}_{\text{mul}}$ from 0.558 to 0.609 without materially affecting $\text{FAR}_{\text{mul}}$.
However, for AHI, substituting the $11.2~\mu\text{m}$ channel with the $12.3~\mu\text{m}$ channel yielded negligible improvement. In addition to spectral response function (SRF) mismatches, a primary contributing factor is the channels' on-orbit radiometric stability.
Hence, physics-informed machine-learning methods appear promising for advancing remote-sensing AI, while sensor-specific characteristics must be considered during operational transfer.
Blogger's Review: This paper illustrates how physical knowledge can effectively enhance the performance of machine learning models, particularly in complex meteorological data processing. By integrating physical principles with modern deep learning methods, the study provides new insights for multilayer cloud detection, which is worth emulating and promoting in other fields.