Guanabara Bay is a tropical estuarine ecosystem that receives massive anthropogenic impacts from the metropolitan region of Rio de Janeiro. This ecosystem suffers from an ongoing eutrophication process that has been shown to promote the emergence of potentially pathogenic bacteria, giving rise to public health concerns. Although previous studies have investigated how environmental parameters influence the microbial community of Guanabara Bay, they often have been limited to small spatial and temporal gradients and have not been integrated into predictive mathematical models. Our objective was to fill this knowledge gap by building models that could predict how temperature, salinity, phosphorus, nitrogen and transparency work together to regulate the abundance of bacteria, chlorophyll and Vibrio (a potential human pathogen) in Guanabara Bay. To that end, we built artificial neural networks to model the associations between these variables. These networks were carefully validated to ensure that they could provide accurate predictions without biases or overfitting. The estimated models displayed high predictive capacity (Pearson correlation coefficients ≥0.67 and root mean square error ≤ 0.55). Our findings showed that temperature and salinity were often the most important factors regulating the abundance of bacteria, chlorophyll and Vibrio (absolute importance ≥5) and that each of these has a unique level of dependence on nitrogen and phosphorus for their growth. These models allowed us to estimate the Guanabara Bay microbiome's response to changes in environmental conditions, which allowed us to propose strategies for the management and remediation of Guanabara Bay.
Keywords: Artificial neural networks; Estuary; Eutrophication; Machine learning; Pollution; Time-series; Tropical.
Copyright © 2019 Elsevier B.V. All rights reserved.