Modelling habitat suitability of the Indo-Pacific humpback dolphin using artificial neural network: The influence of shipping

Abstract

The distribution of cetaceans is generally studied on the basis of their visual locations. However, the absence of observations does not exclude the presence of dolphins and not allow to distinguish habitats favourable to the species but where it would be currently absent due to anthropic disturbances. The modelling of ecological niches represents a powerful alternative choice and intensive computer modelling has been increasingly used to reveal the complexity of the relationships between cetaceans and their habitat. Here, we predicted the presence/absence of the Indo-Pacific humpback dolphin (Sousa chinensis), an endangered species, using the artificial neural network model of back-propagation (BP-ANN) with eight environmental variables. The BP-ANN model had a higher success rate for correct prediction (74%) compared to linear discriminant analysis (67%), especially for the prediction of the presence of S. chinensis (63% to 31%), indicating its potential application in cetacean habitat research. In the model output map, three suitable habitats were predicted without S. chinensis sightings identified. However, only one was confirmed by subsequent field surveys, the other two being located in a strong shipping area. Therefore, we suggest that the traditional assessment of the baseline habitat based on visual sighting may miss the identification of some suitable habitats due to anthropogenic disturbance. We have also highlighted the importance of ecological modelling research for cetacean conservation. In addition, among the eight environmental variables studied, distance from shore, fish abundance and salinity proved to be the most important factors for the distribution of S. chinensis, indicating that coastal construction, sea recovery and overfishing would be key constraints for its conservation.