Received 19 April 2007 Accepted 24 April 2007 Beven’s (2006) commentary on the uncertainty in hydrological modelling seems to have touched a bit of a nerve in the hydrological community. However, while there may be different definitions for uncertainty (Montanari, 2007), different approaches to quantifying it (Todini and Mantovan, in press), and calls for more effort in evaluating it (Hall et al., 2007), there is no denial that uncertainty is an inherent and unwelcome attribute of all predictions. This conversation about adapting to the rude reality of uncertainty could be misconstrued as an admission that we have reached a limit in hydrology, where improvements in the value of our predictions are less a function of theoretical advances or of further field observations and experimentation and more a function of advances in numeric methods or computing horsepower. Furthermore, young hydrologists overhearing this discussion may be excused if they mistook the reaction to Beven’s call for strategically advancing the science and ethics of hydrology as an argument for devising ever more elegant and sophisticated apologies for shortcomings in our hydrological predictions. This current focus on how to maximize the usefulness of models in the context of uncertainty is pragmatic. However, optimizing the present value of model output does not preclude a vision of the future where models are not only useful but are also truthful. Uncertainty is frustrating hydrologists in a way that is similar to the way that microbes frustrated medical science before the discovery of antibiotics. We need to find ways to move beyond mere treatment of symptoms to resolution of underlying pathologies. Like most hydrologists, I believed that I used parameters responsibly. I discovered my addiction to poorly defined parameter ranges almost by accident. A colleague wrote an R script, which made it easy to produce dotty-plots of my calibrations showing parameter identifiability. My first reaction was denial, that there must be a problem with the objective function, but gradually I have come to the realization that I have a history of parameter-abuse. Accepting that I suffer from equifinality was the first step on the long road to recovery. The signs of addiction are subtle but recognizable. Hydrographs that are just a little too good, where the uncertainty in model predictions is less than the uncertainty in the driving force data. Axis-scaling that is a little too compressed, hiding inconsistencies in the model fit. You may recognize these symptoms in your colleagues but deny that you too, keep a flask of parameter ranges in a desk drawer that you sip generously from when faced with a difficult hydrograph. The wrack and ruin of parameter abuse is wide-spread. We need to be particularly concerned about the effect our example has on young hydrologists who are just now trying to learn how to calibrate responsibly. We need to provide alternatives to parameter abuse. A priori parameter identification (e.g. Wagener and Wheater, 2006) and development of observation systems that will support increased understanding and the development and testing of hydrologic theories and organizing principles (e.g. McDonnell et al., 2007) are the basis for the first of the four-pillars approach of Prevention, Treatment, Harm
Read full abstract