Modeling underwater photosynthetically available radiation profiles from biogeochemical Argo floats using...

Biogeochemical (BGC) Argo floats currently measure downwelling planar irradiance (E_d) at three spectral wavelengths (380, 443, and 490 nm) and photosynthetically available radiation (PAR) in the 400-700 nm range. In next-generation floats, replacing the PAR sensor with a 555 nm band is under consideration to enhance spectral resolution while still enabling accurate PAR reconstruction from the existing measurements. This study proposes a General Additive Model (GAM) to estimate PAR at any given depth (z) from just below the surface (z = 0 m) to 200 m using E_d at 380, 443, 490, and 555 nm. The model coefficients are functions of z, and the possibility of introducing chlorophyll concentration ([Chl]) as an extra parameter is also evaluated. Theoretical simulations conducted under diverse environmental conditions show that PAR(z) can be estimated with high accuracy. When using depth as the sole explanatory variable, the estimate bias ranges from 0 to -0.59 µE/m²/s and root mean square deviations (RMSD) between 0.01 and 8.23 µE/m²/s. Percent bias is near-zero across all depths with slightly elevated values near the surface and around 200 m, i.e., -0.2% and 0.5%, respectively. Relative RMSD is about 1-2% at the surface depths and gradually increases to about 8% at 200 m. Including [Chl] as an extra explanatory variable did not significantly improve model performance, probably attributed to the uncertainties in [Chl] measurements. Validation against various in-situ E_d profiles confirms the model's robustness, i.e., with an overall model bias of -0.8% and an RMSD of 4.8% across 120,115 in-situ cases, accurately capturing near-surface variability and maintaining consistent performance, i.e., less than 10% relative error for PAR ranging from 10³ to 10^-2 µE/m²/s. Theoretical uncertainty of the PAR estimates was also quantified as a function of depth and estimated PAR, providing an uncertainty value for each estimate and showing good agreement with actual uncertainties. The proposed model benefits the BGC Argo program by expanding the E_d dataset within the photosynthetically active range and offering accurate PAR estimates across diverse environmental conditions.