Modeling the species richness and abundance of lotic macroalgae based on habitat characteristics by artificial neural networks: a potentially useful tool for stream biomonitoring programs

Nenhuma Miniatura disponível




Rocha, José Celso [UNESP]
Peres, Cleto K.
Buzzo, José Leonel L. [UNESP]
de Souza, Vinícius [UNESP]
Krause, Eric A. [UNESP]
Bispo, Pitágoras C. [UNESP]
Frei, Fernando [UNESP]
Costa, Lucas S. M. [UNESP]
Branco, Ciro C. Z. [UNESP]

Título da Revista

ISSN da Revista

Título de Volume



One of the major challenges in stream ecology is the development of computational models that can predict aspects of the community structure of organisms from these ecosystems when they are subject to natural or artificial environmental fluctuations. To contribute towards this aim, we conducted a study whose main goal was to evaluate the efficiency and accuracy of different architectures of multilayer artificial neural networks (ANNs) in predicting the species richness and abundance of macroalgae based on environmental variables of tropical streams. We used data from 82 streams located in southern Brazil, where species richness, macroalgal abundance, and environmental parameters were measured. A set of 20 environmental parameters measured directly in the stream was used as explanatory variables. The performance of the ANN architectures was assessed using two different pieces of software (random combinatorial and exhaustive) and the coefficient of determination (R2) and mean-squared error (MSE). For both species richness and macroalgal abundance, the best ANN architectures were obtained using random combination software and the performance parameters showed a combination of high R2 and very low MSE. Our results suggest that computational models that are constructed based on ANN frameworks can be efficient and accurate in predicting the species richness and abundance of stream macroalgae from environmental data. Therefore, considering that models based on linear relationships have often failed, we recommend the application of ANNs as a tool to estimate species richness and abundance of lotic macroalgae from environmental data, in the management, conservation, and biomonitoring programs of tropical stream ecosystems.



Artificial neural networks, Environmental distribution, Predictive models, Species richness and abundance, Stream biomonitoring programs, Stream macroalgae

Como citar

Journal of Applied Phycology, v. 29, n. 4, p. 2145-2153, 2017.