Accurate estimation of photosynthetic available radiation from multispectral downwelling irradiance profiles

Abstract

Photosynthetic available radiation (PAR) is the light usable by photosynthetic organisms. Photosynthetic available radiation measurements at depth are required to quantify the light availability for primary production. Direct PAR measurements may be measured with full-spectrum quantum sensors for the range 400 to 700 nm. When spectrally resolved light is measured, as for the downwelling irradiance spectrum $E_d$, PAR may be computed by numerically integrating $E_d$ within those limits. As radiation varies across a spectral continuum, $E_d$ needs to be resolved at a sufficiently large number of bands, to provide an unbiased PAR estimate. When $E_d$ is available at a small number of spectral bands, as for multispectral $E_d$ sensors, it is still possible to numerically integrate $E_d$, but the estimation will contain errors. Here, we propose a method that delivers unbiased PAR estimates, based on two-layer neural networks, formulable in a small number of matrix equations, and thus exportable to any software platform. The method was calibrated with a dataset of hyperspectral $E_d$ acquired by new types of BioGeoChemical (BGC)-Argo floats deployed in a variety of open ocean locations, representative of a wide range of bio-optical properties. This procedure was repeated for several band configurations, including those existing on multispectral radiometers presently the standard for the BGC-Argo fleet. Validation results against independent data were highly satisfactory, displaying minimal uncertainties across a wide PAR range, with the performance varying as a function of each sensor configuration, overall supporting the operational implementation in the Argo program. Model codes are findable at https://github.com/jaipipor/PAR_BGC_Argo.