Self-Organizing Maps for Evaluation of Biogeochemical Processes and Temporal Variations in Water Quality of Subtropical Reservoirs

• Autor(a) principal: Melo, Darllene S. [UNESP]
• Data de Publicação: 2019
• Outros Autores: Gontijo, Erik S. J. [UNESP], Frascareli, Daniele [UNESP], Simonetti, Vanessa C. [UNESP], Machado, Leila S. [UNESP], Barth, Johannes A. C., Moschini-Carlos, Viviane [UNESP], Pompêo, Marcelo L., Rosa, André H. [UNESP], Friese, Kurt
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1029/2019WR025991; http://hdl.handle.net/11449/198266
• topic: Kohonen neural network; limnology; reservoir management; trophic state index (TSI)

Water quality in reservoirs is often compromised in many regions worldwide by nutrients and trace metals. This demands continuous monitoring; however, analyses of large data sets collected during regular monitoring remain a difficult task. Multivariate techniques offer a fast and robust approach for interpreting complex results. The objective of this study was to check the efficacy of self-organizing maps (SOMs) as a tool to investigate biogeochemical processes. This tool can also help to illustrate influences of land use patterns on the water quality of reservoirs. Here we use the Itupararanga Reservoir in Brazil as a subtropical example. Vertical profiles were sampled from seven sites in the reservoir in a total of seven campaigns over 24 months. Next to physicochemical parameters in the water column (dissolved oxygen, Eh, pH, and temperature), levels of nutrients (NO3 −, NH4 +, and PO4 3−), transition and trace metals (Al, Ba, Ca, Cr, Cu, Fe, K, Mg, Mn, Na, Ni, and Zn), and chlorophyll-a (Chla) were measured. These variables were correlated with land use using SOM. With this technique samples were classified into 17 distinct groups that showed distinct influences of spatial heterogeneity and seasonality. The analyses helped to reveal a seasonal stratification period, where Fe, Mn, and P were released from sediments. Nutrients and some metal inputs (Al and Fe) were related to agricultural, urban, and grass/pasture areas around the reservoir. Our approach also helped to explain physical and biogeochemical seasonality in the reservoir.