Application of natural computing (computational intelligence) methods to water resources and environmental modelling

Preface Application of natural computing (computational intelligence) methods to water resources and environmental modelling Over the last 10 to 15 years, natural computing methods have been applied extensively to the modelling and management of our natural and built environments. Data-driven approaches, such as artificial neural networks (ANNs), have been used for prediction and forecasting and evolutionary algorithms (EAs), such as genetic algorithms, have been applied to a range of optimisation problems. More recently, agent based models (ABMs) have been used to gain a better understanding of complex systems, as they enable the inclusion of social and behavioural aspects in the modelling process. Although ANNs and EAs have found wide acceptance in the research community, a number of obstacles have to be overcome before they can be used routinely by the wider modelling community. Firstly, their success has to be demonstrated for a range of realistic applications, including their advantages over more conventional techniques. Secondly, their transparency needs to be increased, which can be achieved by gaining a better understanding of how they operate and behave. Thirdly, guidelines need to be developed that help users to (i) choose an appropriate method from the ever-growing list of available options and (ii) implement the methods in an efficient manner. The papers in this issue make a significant contribution towards overcoming some of the obstacles mentioned above. The versatility of EAs for optimisation is highlighted in a number of papers, with applications including the calibration of urban drainage (Siriwardene and Perera), catalyst emission (Whitcombe et al.) and water quality (Gibbs et al.) models, the optimal design of water distribution systems (Zecchin et al.), determination of the optimal length of fishing seasons and the optimal investment in fishing vessel capital (Elliston and Cao), the optimal clustering of ecological data (Maier et al.) and the optimisation of behavioural rules (Purnomo and Guizol). The utility of ANNs for prediction and forecasting is also demonstrated by their successful application to the prediction and forecasting of river flow (Kingston et al.) and water quality in distribution systems (Bowden et al., Gibbs et al.), as well as their favourable performance compared with more traditional modelling approaches. In relation to the second point, the paper by Kingston et al. addresses the important issue of knowledge extraction from ANNs by introducing a novel stochastic approach for determining the relative contribution of each of the model inputs. The paper by Zecchin et al. …