ABSTRACT
Artificial neural networks (NN) are now commonly used in many fields, ranging from social sciences to engineering, where in comparison to more conventional statistical or theoretical approaches, such tools are providing better and/or faster solutions to different types of problem. Despite this overall success, the uptake of NN technologies in the hydrological sciences has been much slower although we are now witnessing an increasing momentum in the reported use of NN and other AI technologies. Throughout this book there are clear illustrations of areas in which hydrological science could benefit from the application of NN including: rainfall-runoff modelling (Chapters 3 to 9); rainfall forecasting (Chapter 10); water quality prediction (Chapter 11); sediment modelling (Chapter 12); and applied remote sensing (Chapters 13 and 14). The examples presented in these chapters all demonstrate that neurocomputing can produce models of similar or superior performance, but to date there is little evidence that the available technologies are being transferred into an operational environment, which is the next crucial step forward, if the widespread potential of these tools is to be successful. To facilitate a process of technological transfer, neurohydrologists will need to pursue a research agenda that continues to tackle questions about improvements in modelling mechanics, and also address the concerns of traditional hydrologists who prefer to use process-based models
and to exclude data-driven approaches. This chapter suggests five general directions in which research in the next decade could be pursued:
• improvement to existing NN hydrological models including the investigation of current NN hydrological problems;
• more emphasis on their comparison with operational and process-based models; • the development and construction of more powerful and more efficacious modelling
evaluation criteria; • further research into understanding their internal workings and the real-world meaning
of each component; and • the construction of integrated and hybrid NN software for hydrological research.
