Observational uncertainties in hypothesis testing: investigating the hydrological functioning of a tropical catchment

Hypothesis testing about catchment functioning with conceptual hydrological models is affected by uncertainties in the model representation of reality as well as in the observed data used to drive and evaluate the model. We formulated a learning framework to investigate the role of observational uncertainties in hypothesis testing using conceptual models and applied it to the relatively data-scarce tropical Sarapiqui catchment in Costa Rica. Observational uncertainties were accounted for throughout the framework that incorporated different choices of model structures to test process hypotheses, analyses of parametric uncertainties and effects of likelihood choice, a posterior performance analysis and (iteratively) formulation of new hypotheses. Estimated uncertainties in precipitation and discharge were linked to likely non-linear near-surface runoff generation and the potentially important role of soils in mediating the hydrological response. Some model-structural inadequacies could be identified in the posterior analyses (supporting the need for an explicit soil-moisture routine to match streamflow dynamics), but the available information about the observational uncertainties prevented conclusions about other process representations. The importance of epistemic data errors, the difficulty in quantifying them and their effect on model simulations was illustrated by an inconsistent event with long-term effects. Finally we discuss the need for new data, new process hypotheses related to deep groundwater losses, and conclude that observational uncertainties need to be accounted for in hypothesis testing to reduce the risk of drawing incorrect conclusions. Copyright (c) 2015 John Wiley & Sons, Ltd.