Chapter 12
Utility of Sensitivity Analysis by Artifieial Neural
Network Models to Study Patterns of Endemie
Fish Speeies
M. Gevrey . S. Lek· T.Oberdorff
12.1
Introduction
Artificial Neural Networks (ANNs) have become more and more frequently used
in ecology in the last decade, essentially to resolve forecasting problems (Cornuet
et al. 1996; Recknagel et al. 1997; Guegan et al. 1998; Clair and Ehrman 1998;
Ozesmi and Ozesmi 1998; Maier and Dandy 1999; Laberge et al. 2000).
Many studies have shown that this tool has a better predictive power than the
classieal linear methods (Lek et al. 1996b; Brey et al. 1996; Ramos-Nino et al.
1997; Starrett et al. 1997; Huntingford and Cox 1997; Paruelo and Tomasel1997;
Whitehead et al. 1997). Nevertheless ANNs are known as a "black box" type
model, unable to clarify the contribution of the explanatory variables to the
dependent one. However, even if the prediction capacity is of prime importance, it
is also necessary, particularly in ecology, to evaluate the way explanatory
variables contribute to the explanation of the ecological processes involved.
Several authors have thus focused on the analysis of output variables sensitivity
with respect to the input (Garson 1991; Goh 1995; Lek et al. 1996a,b; Maier and
Dandy 1996; Balls et al. 1996; Scardi 1996; Seginer 1997; Dimopoulos et al.
1995; 1999). In this paper, two algorithms developed by Lek (Lek et al. 1996a,b)
and Dimopoulos (Dimopoulos et al. 1995; 1999) are evaluated and compared.
These two methods reinvestigate a previous study analysing patterns of
endemic riverine fish species richness at the global scale (Oberdorff et al. 1999).
In this study, the endemie species riehness was directly dependent on species
diversity (e.g. total species riehness). Total species richness was itself influenced
by factors related to components of river size (i.e. surface area of the drainage
basin) and to a lesser extent, energy availability (i.e. net primary productivity).
This study comprises four sections corresponding to: i) the development of the
two methods, ii) the presentation of the dataset, iii) the investigation of results
obtained for predietions and those obtained with the contribution methods, iv) the
discussion of the ecologieal significance of the results obtained.
Utility of Sensitivity Analysis by Artifieial Neural
Network Models to Study Patterns of Endemie
Fish Speeies
M. Gevrey . S. Lek· T.Oberdorff
12.1
Introduction
Artificial Neural Networks (ANNs) have become more and more frequently used
in ecology in the last decade, essentially to resolve forecasting problems (Cornuet
et al. 1996; Recknagel et al. 1997; Guegan et al. 1998; Clair and Ehrman 1998;
Ozesmi and Ozesmi 1998; Maier and Dandy 1999; Laberge et al. 2000).
Many studies have shown that this tool has a better predictive power than the
classieal linear methods (Lek et al. 1996b; Brey et al. 1996; Ramos-Nino et al.
1997; Starrett et al. 1997; Huntingford and Cox 1997; Paruelo and Tomasel1997;
Whitehead et al. 1997). Nevertheless ANNs are known as a "black box" type
model, unable to clarify the contribution of the explanatory variables to the
dependent one. However, even if the prediction capacity is of prime importance, it
is also necessary, particularly in ecology, to evaluate the way explanatory
variables contribute to the explanation of the ecological processes involved.
Several authors have thus focused on the analysis of output variables sensitivity
with respect to the input (Garson 1991; Goh 1995; Lek et al. 1996a,b; Maier and
Dandy 1996; Balls et al. 1996; Scardi 1996; Seginer 1997; Dimopoulos et al.
1995; 1999). In this paper, two algorithms developed by Lek (Lek et al. 1996a,b)
and Dimopoulos (Dimopoulos et al. 1995; 1999) are evaluated and compared.
These two methods reinvestigate a previous study analysing patterns of
endemic riverine fish species richness at the global scale (Oberdorff et al. 1999).
In this study, the endemie species riehness was directly dependent on species
diversity (e.g. total species riehness). Total species richness was itself influenced
by factors related to components of river size (i.e. surface area of the drainage
basin) and to a lesser extent, energy availability (i.e. net primary productivity).
This study comprises four sections corresponding to: i) the development of the
two methods, ii) the presentation of the dataset, iii) the investigation of results
obtained for predietions and those obtained with the contribution methods, iv) the
discussion of the ecologieal significance of the results obtained.