A comparison of artificial neural networks and random forests to predict native fish species richness in Mediterranean rivers

Abstract

Key-words:Artificial neural networks, random forests, native fish, species richness, Mediterranean rivers Machine learning (ML) techniques have become important to support decision making in management and conservation of freshwater aquatic ecosystems. Given the large number of ML techniques and to improve the understanding of ML utility in ecology, it is necessary to perform comparative studies of these techniques as a preparatory analysis for future model applications. The objectives of this study were (i) to compare the reliability and ecological relevance of two predictive models for fish richness, based on the techniques of artificial neural networks (ANN) and random forests (RF) and (ii) to evaluate the conformity in terms of selected important variables between the two modelling approaches. The effectiveness of the models were evaluated using three performance metrics: the determination coefficient (R 2 ), the mean squared error (MSE) and the adjusted determination coefficient (R 2 adj ) and both models were developed using a k-fold crossvalidation procedure. According to the results, both techniques had similar validation performance (R 2 = 68% for RF and R 2 = 66% for ANN). Although the two methods selected different subsets of input variables, both models demonstrated high ecological relevance for the conservation of native fish in the Mediterranean region. Moreover, this work shows how the use of different modelling methods can assist the critical analysis of predictions at a catchment scale. RÉSUMÉUne comparaison des réseaux de neurones et des forêts aléatoires pour prédire la richesse en espèces de poissons indigènes dans les rivières méditerranéennes Les techniques d'apprentissage automatique (ML) sont devenues importantes pour aider à la décision dans la gestion et la conservation des écosystèmes aquatiques d'eau douce. Étant donné le grand nombre de techniques ML pour amélio-rer la compréhension de l'utilité des ML en écologie, il est nécessaire de réaliser des études comparatives de ces techniques comme analyse préparatoire pour des applications de modèles futurs. Les objectifs de cette étude étaient : (i) de comparer la fiabilité et la pertinence écologique de deux modèles prédictifs pour la richesse de poisson, basé sur les techniques de réseaux de neurones artificiels (ANN) et les forêts aléatoires (RF) et (ii) d'évaluer la conformité en termes de sélection des variables importantes entre les deux approches de modélisa-tion. L'efficacité des modèles a été évaluée au moyen de trois indicateurs de Article published by EDP Sciences E.J. Olaya-Marín et al.: Knowl. Managt. Aquatic Ecosyst. (2013) 409, 07 performance : le coefficient de détermination (R 2 ), l'erreur quadratique moyenne (MSE) et le coefficient de détermination ajusté (R 2 adj ) et les deux modèles ont été développés en utilisant une procédure de validation croisée k-fold. Selon les résul-tats, les deux techniques ont des performances de validation similaires (R 2 = 68 % pour RF et R 2 = 66 % pour ANN). Bien que les deux...

Random forests to evaluate biotic interactions in fish distribution models

“…Moreover, the area under ROC curve (AUC) and the true skill statistic (TSS), which can also be considered independent of prevalence (Vaughan and Ormerod, 2005;Maggini et al, 2006), suggested good to excellent model performance (Pearce and Ferrier, 2000;Allouche et al, 2006). 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 19 fish abundance may provide valuable additional information (Fukuda et al, 2012;Olaya-Marín et al, 2013). Table 1.…”

Spatial prediction of demersal fish diversity in the Baltic Sea: comparison of machine learning and regression-based techniques

“…Notwithstanding, a review of the literature showed a relatively low number of ML applications in ecology compared with other scientific fields (Olden et al, 2008). Several studies have highlighted the urgent need for comparisons of the performance of different ML predictive techniques in ecology (Aertsen et al, 2011;Olaya-Mar ın et al, 2013).…”

“…This group of organisms is considered an effective indicator of aquatic ecosystem quality due to its sensitivity to anthropogenic disturbances (HELCOM, 2012;Smoli nski and Całkiewicz, 2015). Fish species richness and diversity are often used as a primary measures of ecological shifts and as a basis for planning protected areas (Knudby et al, 2010b;Olaya-Mar ın et al, 2013). Knowledge of the relationships between fish assemblages and environmental factors is important for effective conservation, and the application of novel statistical techniques can improve our understanding of these ecological processes (Olden et al, 2008).…”

“…Knowledge of the relationships between fish assemblages and environmental factors is important for effective conservation, and the application of novel statistical techniques can improve our understanding of these ecological processes (Olden et al, 2008). A number of predictive analyses on freshwater fish communities, employing ML approaches, can be found in the literature (Olaya-Mar ın et al, 2013). Most available predictive applications of ML in marine fish ecology refer mainly to coral reef-associated assemblages in shallow waters (Pittman et al, 2007;Moore et al, 2009;Knudby et al, 2010a;Pittman and Brown, 2011).…”

Integration of habitat models to predict fish distributions in several watersheds of Northern Spain

“…Moreover, it would be interesting to include other chemical and physical variables that were not available for this study (e.g. the percentage of a certain type of mesohabitat could be important to simulate fish species richness; Olaya‐Marin et al., ).…”