Machine Learning Assessment of Dredging Impacts on the Phytoplankton Community on the Brazilian Equatorial Margin: A Multivariate Analysis

Abstract

Using a machine learning approach, we assessed the impact of dredging on the phytoplankton community structure in a tropical estuarine macrotidal complex. The data spans before, during, and after dredging events in 2015, 2017, and 2020, with volumes ranging from 657,641 to 25,000 m3. As a result, Model 1, which employs the Random forest (RF) algorithm, highlighted that solid particulate matter (SPM) and total phosphorus influence phytoplankton abundance. Model 2 revealed the role of salinity and dissolved iron in reducing species diversity, whereas Model 3 revealed that river discharge and salinity influenced phytoplankton biomass. Generalized additive models (GAMs) indicated that chlorophyll-a responded positively to increasing salinity and negatively to NH4+ and dissolved aluminum. Species diversity decreased with SPM, whereas abundance was correlated with dissolved iron and manganese. Despite the identification of 283 phytoplankton species, 22 centric diatoms persisted across dredging events. Our results highlight the importance of dredging duration and intensity in shaping phytoplankton community dynamics.

Keywords: random forest, Generalized additive models (GAMs), microalgae, Brazilian equatorial margin, Port area.