A Model for Predicting the Amount of Photosynthetically Available Radiation from BGC-ARGO Float Observations in the Water Column

Abstract

Modern oceanography uses, amongst other platforms, automated diving devices, which are drifting with the ocean current whilst continuously collecting vertical profiles of environmental parameters. One of the important parameters is photosynthetically available radiation (PAR). It was studied in this work whether the PAR values can be reconstructed by combinations of measurements from the remaining onboard sensors with specific wavelength. If a reconstruction of PAR is possible, this would allow allocating the sensor with a further specific wavelength instead of PAR. Having available more spectral information could for example enable natural science researchers to better distinguish phytoplankton or UV radiation. Therefore, data from three different expeditions from different regions of the world were used to model PAR using multiple linear regression and regression trees (RT). multiple linear regression achieved an R2 value of 0.970 for the combined dataset and RT achieved an R2 value of 0.960. Hence, the models are accurate enough to predict the PAR parameter without the need for a dedicated PAR sensor. Thus the PAR sensor could be allocated instead with a further specific wavelength.

Vertically Resolved Global Ocean Light Models Using Machine Learning

“…Pioneering work by [26] showed that global 3D reconstruction of the b bp could be performed thanks to the development of the first SOCA model. More recently, subsets of irradiance data (ED380, ED412, and ED490) acquired by BGC-Argo floats have been used to predict PAR either through statistical approaches [59] or the use of neural networks [60].…”