Review of Neural Networks for Hydrological Modelling by Robert J. Abrahart, Pauline E. Kneale, and Linda M. See: Taylor & Francis, London, U.K.; 2004; ISBN 90-5809-619-x; 304 pp. Price: $119.95.

It is well documented that we need to use our water resources efficiently and operate our water systems optimally. The management of water systems requires forecasts of various hydrologic variables. Hydrological modeling has been the subject of research for a long time, and many models employing conventional tools of varying degrees of complexity and sophistication have been proposed. Neural networks (NNs) have been used for hydrological modeling since the late 1980s. However, their use has increased in the last five to ten years. This book, which is published in the form of an edited volume, is a collection of fifteen chapters written by selected researchers working in different subareas of hydrological sciences.

The first chapter starts by attempting to answer the question: Why use neural networks? A useful piece of information in chapter 1 is a table containing the details of software available for carrying out neural network simulations along with the Web site addresses. Chapter 2 presents NN basics such as network architectures, learning considerations, and commonly used NN models, e.g., back-propagation neural network (BPNN), radial basis function network (RBFN), and self-organizing feature map (SOFM), in a very concise manner. Anyone beginning to learn how to develop an NN model can go to chapter 3, which presents the NN model development as a six-stage process with the help of a case study. An important feature of the book is that it covers application of different types of neural networks in a wide variety of hydrological processes. For example, one can learn about hybrid neural network modeling, time delay neural networks (TDNNs), cascade correlation neural networks (CCNNs), and partial recurrent neural networks (PRNNs) in addition to the more popularly employed BPNNs, RBFNs, and SOFMs. River flow modeling using hybrid neural networks is presented in chapter 4, wherein broader issues encountered in practice are discussed with focus on training issues and uncertainties in modeling. Chapter 5 presents application of TDNNs to river level forecasting; chapter 6 discusses application of CCNNs to river flow forecasting; and chapter 7 describes the use of PRNNs for modeling of autoregressive dynamic hydrologic systems. Chapter 8 presents a flood forecasting application using three different types of neural networks using a web-based environment called RLF/1. Chapter 9 includes a state-of-the-art on rainfall runoff modeling using neural networks and fundamental issues on rainfall runoff relationships. Chapter 10 presents a neural network application for rainfall forecasts in an urban environmental for use in effective flood warning systems. Chapter 11 covers an overview of the NN studies in the areas of water quality and ecological management in freshwater. Chapter 12 discusses mechanisms of sediment supply and transfer in a catchment, erosion, and sediment yield assessment and presents some NN studies on the use of neural networks of sediment modeling. Chapter 13 describes the use of meteorological satellite image data in Nowcasting and Numerical Weather Prediction (NWP) and why neural networks have the potential to deal with the complex patterns present in such data in real-time operational forecasting applications. Chapter 14 focuses on the potential of feed-forward neural networks as a tool for land cover mapping using supervised digital image classification of remotely sensed imagery. The final chapter argues that although neural network solutions have been found that are either superior or comparable to the process-based and other models in use by various water agencies, neural network solutions have not found favors with the administrators, water managers, and policymakers. This chapter proposes a solution to this problem by way of a five-stage research agenda for neurohydrologists to pursue in the next decade. The stages considered include improvements of existing neural network models, comparison of neural network solutions with process-based modeling solutions, development of meaningful criteria for model evaluation, improvement of model understanding, and building dedicated hydrological neural network software packages.

The book is well organized; chapters are nicely written and easy to follow. The book bridges the gap between research and application of NN hydrologic models. The length of the individual chapters seems to have been limited, probably because the book covers many applications in hydrology. This prevents the authors from dealing with the subject matter in greater detail. Focusing on fewer application areas provides the authors freedom of space to describe intricacies in modeling a particular physical system. The book can be used by a researcher beginning work on hydroinformatics and will also be a valuable addition to the self-library of an experienced scientist working in the areas covered.