Hydrological models are so good, do we still need data? R.P. Silberstein * CSIRO Land and Water, Private Bag No. 5, PO Wembley, WA 6913, Australia Received 4 August 2004; received in revised form 6 April 2005; accepted 6 April 2005 Available online 5 October 2005 Abstract Our ability to numerically model natural systems has progressed enormously over the last 10e20 years. During the last decade computational power has increased to the stage where we can now have a super-computer on our desk, and the detail and fine scale processes that can be included in models are fantastic. The computational tools available for analysis and display have opened doors beyond the dreams of our forebears. However, as modelling power has increased there has been a concurrent reduction in ‘‘data power’’, particularly in the collection of hydrological data. While we undoubtedly have access to large datasets of extraordinary technological finesse such as the remotely sensed data from satellites, our collection of more basic and traditional datasets suffers. We can read car number plates from outer space, but we still, in the main, measure rainfall by putting a bucket out in a paddock. This paper discusses the growth in sophistication of hydrological modelling through the last hundred years. The concept of valida- tion or verification of models is questioned, and the role of data in modelling discussed. It is argued that modelling in the absence of adequate data is not science, unless it is to develop hypotheses that are to be tested by observation. Several modelling case studies with and without adequate testing data are discussed. It is also argued that improvement in the management of our environment and water resources will not come with improved models in the absence of improved data collection because we cannot manage what we do not measure. Ó 2005 Elsevier Ltd. All rights reserved. Keywords: Modelling; Hydrology; Monitoring; Data collection; TOPOG; LASCAM; IQQM; TOPMODEL 1. Introduction Modelling is now a common tool in many fields of scientific endeavour. Physical scale models have been used to study the static and dynamic behaviour by engi- neers for many years, to ensure their bridges would stay up, or their breakwaters would not wash away. Children construct scale models of racing cars and aeroplanes to explore the effect of high speed crashes and glide angles of fighters without engine power. Mathematical modelling is generally not visually im- pressive but it allows exploration of real behaviour that would be difficult to observe or measure in nature. Mathematical models are used in economics, politics and policy, finance, commerce and the behavioural sci- ences. Modelling these systems is generally impossible without simplifying the representation of the real system simulated. Statistical and behavioural models are used to predict horse races, stock markets, ecological systems and social systems. These models are used to try to pre- dict how the systems will develop over time, usually so that some return may be maximised for the modeller e this may be financial, political return, or in order to improve social harmony. For scientists, the aim is to construct a model repre- senting some component of the real world. For some, the model development is sufficient in its own right, such as Game Theoreticians or economists, who explore the impact of various rules on system outcomes. These * Fax: C61 8 9333 6211. E-mail address: richard.silberstein@csiro.au 1364-8152/$ - see front matter Ó 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.envsoft.2005.04.019 Environmental Modelling & Software 21 (2006) 1340e1352 www.elsevier.com/locate/envsoft