Information, Predictability and Fitness for Purpose

Abstract:

Decisions in catchment management are increasingly based on the predictions of catchment models. But hydrological models are subject to epistemic uncertainties that are difficult to model as simple aleatory variables. This will include the effects of model structure and data inconsistencies or disinformation that will result in non-stationarity of error characteristics in model calibration. This suggests that there will be limits to predictability and that a different approach to testing models as hypotheses about catchment response might be needed. Following Beven and Smith (ASCE JHE 2015) an event based approach is suggested for both checking data for consistency and assessing the information contributed by individual events to model conditioning. This raises issues of the effect of disinformative events on how well a model can predict subsequent events, and what happens in prediction when we will not know beforehand whether an event is informative or disinformative. The impact on assessing models as fit for purpose within the decision or policy making context will also be discussed.