Tall tales from the hydrological crypt: are models monsters?

"Bizarre," "monstrous": in society as well as in science, this is the way we are used to describing objects that deviate from an expected standard. Hydrology is no exception. The bizarre or the monstrous describes every object that has a low probability of occurring, or that our models fail to represent. Actually, the bizarre or the monstrous is often a demonstration of the limits of our models rather than an intrinsic characteristic of the objects we study. This article provides a reflection on the definition of bizarre and monstrous in the context of hydrology. We base our reflection on 60 years of experience in hydrometeorological operational management and applied research at the French national electricity company (E1DF-DTG). First, we describe several classical a priori models or conceptions trusted by hydrologists, sometimes erroneously. These include classical rainfall or streamflow measurement issues, certain limits of the watershed concept or problems in the spatialization of local measurements. Then we attempt to show how the misuse of statistical models can generate bizarre or monstrous results. We give examples related to outliers and to the homogeneity and stationarity hypotheses. We show how difficult it may be for operational forecasters to anticipate and believe that extreme (monstrous) events will occur in the near future. Finally, we wish to show that the bizarre or the monstrous should not be rejected in hydrology, but instead is something to study in greater depth. We believe that this type of analysis offers new opportunities to improve the explanatory and predictive capacity of our models.