Ecological modelling with self-organising maps S Shanmuganathan a , P Sallis a and J Buckeridge b b Environmental and Oceanic Sciences Centre, a Auckland University of Technology, New Zealand. subana.shanmuganathan@aut.ac.nz Abstract: Old and new ecological models can be classified into two basic categories: Those aimed at (i) gaining more insight into ecological systems and (ii) producing predictive models of ecosystem behaviour. Many of the models successfully applied to ecological modelling are borrowed from other disciplines such as engineering, mathematics and in recent times from intelligent information processing systems motivated by neuro-physiological understandings i.e. 1 artificial neural networks (ANNs). The use of ANNs in ecological modelling is becoming a popular method with considerable success in elucidating the complexity in ecosystem processes. We critically analyse some ecological modelling applications with self-organising maps (SOMs), within the connectionist neural computing paradigms. These are used to unravel the non-linear relationships in highly complex and often cryptic ecosystems from northern New Zealand. A need to accurately predict an ecosystems response to daily increasing human influences on the environment and its biodiversity is considered to be absolutely vital to preserve natural systems. The example illustrated shows SOM abilities to extract more knowledge from the ecological monitoring data of complex matrices with numeric values of environmental and biological indicators, compared to the conventional data analysis methods. Conventional methods are seen as of little use in exploring the non-linear relationships within the data. Keywords; Ecological modelling; Self-organising maps; Ecological data 1 An ANN is a biologically inspired computational model, which consists of processing elements (called neurons), and connections between them with coefficients (weights) bound to the connections. These connections constitute the neuronal structure and attached to this structure are the training and recall algorithms. (Kasabov 1995). The recent ANN models are referred to as ‘connectionist neural computing paradigms’. 1. INTRODUCTION Over the last few decades ecological modelling techniques borrowed from other disciplines provided a useful means to analyse natural systems with considerable success. Such models of ecology, based on engineering, statistical and mathematical concepts permitted ecologists to gain more insight into ecosystem structure and functioning. However, all of them demonstrated limitations, such as inability to produce conclusive results in environmental impact analysis (i.e. whether an impact was caused either by human influence or natural causes) and to predict long-term environmental effects for management purposes. The old and the recent ecological models, i.e. Before-After-Control- Impact (BACI), Before-After-Control-Impact Paired Series (BACIPS), etc., with highly complex mathematical and statistical techniques are described to be ineffective due to the above- mentioned reasons. The shortcomings of traditional methods led ecologists to experiment with innovative approaches using the recent intelligent systems (ISs) of information processing methodologies i.e. artificial neural networks (ANNs), Fuzzy logic, etc. The recent use of ANNs in ecological modelling is seen as a popular method, successfully applied to unravel ecosystem complexity using the widely available ecological monitoring data alone. The use of different SOM methods for modelling complex ecosystems, north of Auckland in New Zealand is elaborated upon. 2. CONVENTIONAL MODELS The following two conventional models of ecology are discussed herein: (i) The River Thames models (ii) BACI series models. 2.1. Simulation models: River Thames A class of simulation models (defined by partial differential equations) designed and implemented produced considerable success in cleaning up the 759