High-resolution temporal dynamics of diatoms in a large and well-mixed tropical estuary

We conducted a high-resolution analysis of diatom populations in the microphytoplankton size range using data collected at 30-min intervals over a 20-month period by an automated imaging system deployed near the mouth of Baía de Todos os Santos (BTS), Brazil. Seven diatom taxa were identified and quantified through automated classification using a Convolutional Neural Network (CNN). Frequency-domain analysis revealed distinct environmental drivers acting across different temporal scales. At high-frequency scales (53 h), solar radiation was the predominant factor influencing diatom abundances. At intermediate to monthly scales (53 h–13 days, neap-spring cycles of 13–15 days, and monthly scales), canonical correspondence analysis (CCA) indicated that dissolved oxygen, temperature, and salinity were the primary environmental drivers. Multiple linear regression (MLR) models highlighted colored dissolved organic matter (CDOM) and the north-south wind component as key predictors for Coscinodiscus wailesii abundances. K-strategist marine taxa, including Rhizosolenia robusta and the Rhizosolenia–Proboscia complex, exhibited peak densities during neap tides, coinciding with stronger intrusion events of oligotrophic oceanic waters into the bay. Conversely, r-strategist coastal and estuarine taxa, including C. wailesii, Bacteriastrum-Chaetoceros complex, and Guinardia striata, reached maximum abundances during spring tides, associated with enhanced river discharge and pronounced ebb flow conditions. These taxon-specific distribution patterns demonstrate the influence of environmental forcing across multiple temporal scales on diatom populations. Our findings show the effectiveness of frequency-domain analytical approaches in resolving the complex interactions between environmental variability and phytoplankton dynamics, enhancing understanding of bottom-up regulatory processes and inter-taxa ecological interactions in coastal tropical ecosystems.