POSTER SESSION 1
15:30–17:30, Monday, September 14
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
ABSTRACT 1119 | POSTER 133
RECONSTRUCTING OCI-CONSISTENT HYPERSPECTRAL Rrs FROM AQUA MODIS USING A UNIFIED SPECTRAL-ATTENTION DEEP LEARNING
NASA’s Plankton, Aerosol, Cloud, ocean Ecosystem Ocean Color Instrument (PACE OCI) provides unprecedented hyperspectral observations of ocean color, yet its relatively short observational record limits its use in long-term applications. To bridge the gap between the hyperspectral capability of OCI and the multi-decadal continuity of Aqua MODIS, this study developed two deep-learning models within a unified spectral-attention-based framework to reconstruct OCI-consistent spectra from MODIS multispectral remote sensing reflectance (Rrs). SA-CGAN integrates spectral attention, adversarial learning, and spectral response function constraints to reconstruct visible hyperspectral Rrs over 400–700 nm, whereas SA-UVNet extends the same framework to the ultraviolet domain, enabling reconstruction of Rrs from 346 to 400 nm. Using more than one year of daily 4-km collocated OCI–MODIS observations since March 2024, both models demonstrated strong global generalization and consistently outperformed existing baselines. In the visible range, reconstruction errors in blue open-ocean waters were generally below 5%, whereas uncertainties increased in coastal waters and in the red bands. In the ultraviolet range, SA-UVNet reduced MAE, RMSE, and RRMSE by approximately 10% and improved R² by up to 3%. The results further indicate that short visible wavelengths, particularly 412–488 nm, provide the primary constraints for reconstructing OCI-consistent spectra from MODIS and represent the main source of uncertainty propagation. The reconstructed spectra support downstream applications, including retrieval of seawater constituents, spectral slope analysis, and diagnosis of climate-related changes in ocean material distributions, offering a practical pathway toward long-term, spectrally complete ocean-color remote sensing.
Xiaolong Li, Institute of Oceanology, CAS, [email protected]
Yaopu Zhu, Institute of Oceanology, Chinese Academy of Sciences, [email protected], https://orcid.org/0009-0008-1469-3422
Xiaofeng Li, Institute of Oceanology, Chinese Academy of Sciences, [email protected]
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
