Application of neural networks to model changes in fish community biomass in relation to pressure indicators and comparison with a linear approach

Neural networks (NN) are considered well suited to modelling ecological data, especially nonlinear relationships, and were applied here to investigate which pressures best model changes in the fish community of the Grand Bank, Northwest Atlantic. Nine fishing and environmental pressures were used to...

Full description

Bibliographic Details
Published in:Canadian Journal of Fisheries and Aquatic Sciences
Main Authors: Dempsey, Danielle P., Pepin, Pierre, Koen-Alonso, Mariano, Gentleman, Wendy C.
Format: Article in Journal/Newspaper
Language:English
Published: Canadian Science Publishing 2020
Subjects:
Online Access:http://dx.doi.org/10.1139/cjfas-2018-0411
http://www.nrcresearchpress.com/doi/full-xml/10.1139/cjfas-2018-0411
http://www.nrcresearchpress.com/doi/pdf/10.1139/cjfas-2018-0411
Description
Summary:Neural networks (NN) are considered well suited to modelling ecological data, especially nonlinear relationships, and were applied here to investigate which pressures best model changes in the fish community of the Grand Bank, Northwest Atlantic. Nine fishing and environmental pressures were used to simultaneously model the biomasses of six fish functional groups before and after the collapse of fish biomass in the region and over the full data series. The most influential pressures were identified, and the fit and predictive power were evaluated. The analysis was repeated with different types and lengths of time delay imposed on the pressures. Results were compared with a similar analysis using a multivariate linear regression (MLR) approach. MLR generally resulted in better fit, although the ecological implications of the approaches were typically similar. Findings show that both top-down and bottom-up pressures influenced the fish community over the past several decades, over short and long time scales. NN may have useful forecast potential, although future work is required to improve the forecasts shown here before they can directly inform management.