AUTHOR=Tan Jing , Frouin Robert , HΓ€entjens Nils , Barnard Andrew , Boss Emmanuel , Chamberlain Paul , Mazloff Matt , Orrico Cristina TITLE=Reconstructing hyper-spectral downwelling irradiance from multi-spectral measurements JOURNAL=Frontiers in Remote Sensing VOLUME=Volume 5 - 2024 YEAR=2024 URL=https://www.frontiersin.org/journals/remote-sensing/articles/10.3389/frsen.2024.1335627 DOI=10.3389/frsen.2024.1335627 ISSN=2673-6187 ABSTRACT= In this study, the downwelling planar irradiance reaching the surface 𝑬 𝒔 acquired by the current HyperNav systems in only four 10 nm wide spectral bands centered on 412, 489, 555, and 705 nm in clear sky conditions are used to reconstruct via a multi-linear regression model the hyper-spectral 𝑬 𝒔 signal at 0.5 nm resolution from 315 to 900 nm, the Ocean Color Instrument (OCI) spectral range, allowing an estimate of 𝑬 𝒔 at the HyperNav, OCI, and other sensors' resolutions. Based on simulations for Sun zenith angles from 0 to 75 o and a wide range of (i.e., expected) atmospheric, surface, and water conditions, the 𝑬 𝒔 spectrum is reconstructed with a bias of less than 0.4% in magnitude and an RMS error (RMSE) ranging from 0 to 2.5%, depending on wavelength. The largest errors occur in spectral regions with strong gaseous absorption. In the presence of typical noise on 𝑬 𝒔 measurements and uncertainties on the ancillary variables, the bias and RMSE become -2.5% and 7.0%, respectively. Using a General Additive Model with coefficients depending on Sun zenith angle and aerosol optical thickness at 550 nm improves statistical performance in the absence of noise, especially in the ultraviolet, but provides similar performance on noisy data, indicating more sensitivity to noise. Adding spectral bands in the ultraviolet, e.g., centered on 325, 340, and 380 nm, yields marginally more accurate results in the ultraviolet, due to uncertainties in the gaseous transmittance. Comparisons between the measured and reconstructed 𝑬 𝒔 spectra acquired by the MOBY spectroradiometer show agreement within predicted uncertainties, i.e., biases less than 2% in magnitude and RMS differences less than 5%. Reconstruction can also be achieved accurately with other sets of spectral bands and extended to cloudy conditions since cloud optical properties, like aerosol properties, tend to vary regularly with wavelength. These results indicate that it is sufficient, for many scientific applications involving hyper-spectral 𝑬 𝒔 , to measure 𝑬 𝒔 in a few coarse spectral bands in the ultraviolet to near infrared and reconstruct the hyperspectral signal using the proposed multivariate linear modeling.