Comparing in situ and satellite-derived primary production estimates in the Canary Current upwelling region

Abstract

Satellite-based Net Primary Production (NPP) estimates are arguably the best way to improve our understanding of large-scale ocean productivity and to validate Earth System Models. Despite significant progress over recent decades, satellite-derived NPP estimates still suffer from large uncertainties, primarily due to the limited number of in situ primary production (PP) measurements available for their validation. In addition, the most widely used algorithms lead to different, sometimes even contradictory, results. Along with measurements of chlorophyll a concentration (Chla) and phytoplankton biomass (C_phyto), here we present in situ measurements of PP using ¹⁴C uptake and ¹³C isotope tracing, as well as O₂ and ¹⁸O₂ evolution inside incubation bottles, across the transition zone from the coastal Canary Eastern Boundary Upwelling System (CanEBUS) to the open ocean waters of the Cape Verde Frontal Zone (17–23°N; 16–26°W). We also calculate assimilation numbers ($P_{\mathrm{opt}}^b)$ and growth rates (μ) from in situ measurements. First, we compared in situ PP estimates measured concurrently using the four abovementioned techniques. We then tested the performance of four widely-used models including the Vertically Generalized Production Model (VGPM) and its variant based on Eppley's description of the growth function (Eppley), the Carbon-based Productivity Model (CbPM), and the Carbon, Absorption and Fluorescence Euphotic-resolving model (CAFE), along with the satellite-derived input variables that feed these algorithms. We found that the Chla-based VGPM and Eppley models were significantly correlated with in situ estimates, regardless of the satellite source used as input data. As for models based on C_phyto, only the CbPM from the Visible Infrared Imaging Radiometer Suite (VIIRS) data demonstrated performance comparable to that of the Chla-based models. In all other cases, C_phyto-based models were uncorrelated with in situ PP estimates. Our results indicate that the bias associated with the VGPM and Eppley models is primarily due to the algorithms' inability to accurately assess $P_b^{\mathrm{opt}}$. Meanwhile, the retrieval of both satellite-derived C_phyto and μ leads to a poor estimate of NPP by the CbPM. Our findings suggest that enhancing the accuracy of NPP estimates derived from satellite-based models necessitates the refinement of the methodology employed in deriving the input data and their subsequent validation, rather than developing increasingly complex models.