AUTHOR=Rivas David , Fransner Filippa , Koseki Shunya , Keenlyside Noel 

TITLE=Physical drivers and reconstruction of the interannual variability of satellite-derived chlorophyll-a in key regions of the tropical and south Atlantic

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 12 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2025.1528489

DOI=10.3389/fmars.2025.1528489

ISSN=2296-7745

ABSTRACT=Understanding drivers of variability in oceanic primary productivity is essential to increase our understanding of the functioning of marine ecosystems and biogeochemical cycles. Here, interannual variability of satellite-derived chlorophyll-a (CHL) and its underlying oceanographic processes are analyzed in six coastal regions of the tropical and south Atlantic. Along the South American coast, sea-surface height (SSH) and alongshore velocity, proxies for surface flows, were identified as the main drivers. Along the African coast, variations in sea-surface temperature (SST) and SSH related to coastal upwelling, were the dominant drivers. Important links to the Tropical Southern Atlantic, Dipole Mode Index, Western Hemisphere Warm Pool, and Southern Oscillation Index indices were identified, indicating potential role of teleconnections in the CHL-variability. The identified driver-linked variables were used to reconstruct the regional CHL series using multi-linear regressions and a neural-network model. The multi-linear models were able to reproduce significant fractions of the observed CHL variance. In particular, a model based on eigenvalues from an empirical orthogonal function decomposition of SST, outperformed the others. The neural-network model shows the highest performance reproducing most of the CHL variance (> 70%), but it presents difficulty to deduce the relative importance of individual drivers. Beyond this fitting/training period, the multi-linear model show better results respect to the neural-network model, especially that based on oceanographic variables. These CHL-reconstruction models present the possibility to reproduce CHL in periods when its observation is unavailable and even to predict it in multi-year climate projections.