Modelling a pike (Esox lucius) population in a lowland river using a cellular automaton I.S. Pauwels a, b, ⁎, A.M. Mouton a, b , J.M. Baetens c , S. Van Nieuland a , B. De Baets c , P.L.M. Goethals a a Ghent University, Laboratory of Environmental Toxicology and Aquatic Ecology, J. Plateaustraat 22, B-9000 Ghent, Belgium b Research Institute for Nature and Forest (INBO), Kliniekstraat 25, B-1070 Brussels, Belgium c KERMIT: Research Unit Knowledge-based Systems, Ghent University, Coupure links 653, B-9000 Ghent, Belgium abstract article info Article history: Received 8 July 2011 Received in revised form 3 April 2012 Accepted 29 April 2012 Available online 12 May 2012 Keywords: Cellular automaton Fish Pike Habitat suitability Ecological knowledge Spatio-temporal model Cellular automata (CAs) allow for transparent modelling of complex systems based on simple transition rules and are flexible in incorporating individual differences and local interactions. They may therefore be partic- ularly suited to answer river management questions that could not be addressed by existing habitat suitabil- ity models, such as the optimal distance between spawning grounds. This study explores the usability of CAs for spatio-temporal modelling of a pike population to support river management. Specifically, we evaluated the usability of the CA model by analyzing its sensitivity to three model parameters: the number of pike in the grid, the initial pike distribution and the grid resolution. The model includes habitat characteristics and basic expert knowledge on the ecology of pike and was tested on a 10 km stretch of the river Yser in Flanders (Belgium). Simulation results showed that the model converged to a realistic pike distribution over the study area only at high pike density and low grid resolution, irrespective of the initial pike distribution. Pike density and grid resolution affected the sensitivity to the initial pike distribution in the grid. Specifically, the sensitivity was high at low pike density and high grid resolution, and absent when pike density was high. This analysis indicated that initial conditions and cell size may have a severe impact on the model output, illustrating the importance of firstly analyzing this impact before conducting further analyses. Depending on the outcome of such analyses, CAs can be a promising modelling technique to evaluate and predict the effect of river resto- ration on pike populations. © 2012 Elsevier B.V. All rights reserved. 1. Introduction Sampling campaigns in Flanders revealed a poor condition of pike populations in rivers since 1950 (www.vis.milieuinfo.be). In 1949, pike was widespread, but population viability decreased in all rivers and by the mid 1970s pike was no longer observed. Due to the improvement of the water quality, pike was observed again since 1990, although significantly less abundant than before (De Nayer and Belpaire, 1997; Goethals et al., 2006; Vandenabeele et al., 1998). In an attempt to rehabilitate pike populations, reintroduction programmes were started. These programmes were only moderately successful due to the poor water quality, the loss of suitable habitat (Maeckelberghe, 2002; Vandenabeele et al., 1998), and the obstruction of fish migration (Coeck, 2002; Knaepkens et al., 2004). Insights into the spatio-temporal dynamics of pike populations are thus essential to better predict the impact of pike introduction and conservation actions. Specifically, a successful, integrated management not only requires information on the specific habitat requirements, but also on the spatial distribution of the suitable habitats and on the migration of pike between these habitats during its life cycle. Although species distribution models may reveal new insights into the ecology of pike and their functioning within ecosystems (Glasbergen, 2001; Guisan and Zimmermann, 2000; Inskip, 1982; Kerle et al., 2001; Mouton, 2008; Zarkami, 2008), they are often limited to either temporal or spatial fluctuations in pike distribution. Consequent- ly they do not provide information about the spatial dynamics of pike, although this species typically needs different, spatially separated habitats to successfully complete its life cycle. Geographical Informa- tion Systems (GIS) and mathematical paradigms for spatially explicit and dynamic modelling such as partial differential equations (PDEs), individual-based models (IBMs) and cellular automata (CAs) were introduced in ecological modelling to overcome this problem. Recently, Chen et al. (2011) reviewed these modelling paradigms, their usability and shortcomings. They point out that, although GIS and remote sens- ing are powerful tools for spatial analysis, these give an inherently static view of the world and are unable to capture and model dynamic pro- cesses. PDEs can describe the abundance of a species both spatially and temporally, but neglect spatial heterogeneity, local interactions and individual differences, since they rely on mean-field assumptions Ecological Informatics 17 (2013) 46–57 ⁎ Corresponding author at: Ghent University, Laboratory of Environmental Toxicolo- gy and Aquatic Ecology, J. Plateaustraat 22, B-9000 Ghent, Belgium. Tel : +32 9 264 37 76; fax: +32 9 264 41 99. E-mail address: Ine.Pauwels@ugent.be (IS. Pauwels). 1574-9541/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.ecoinf.2012.04.003 Contents lists available at ScienceDirect Ecological Informatics journal homepage: www.elsevier.com/locate/ecolinf