The role of process inclusion for the prediction of daily flow values

A comparative study of 372 catchments across the continental United States is conducted to explore the influence of several process conceptualisations for the prediction of daily discharge values. Catchments are studied using a split sample test of the streamflow simulations for a collection of lumped conceptual models. These models range from a simple base model, consisting of an unsaturated, fast and groundwater reservoir, to ones that include interception, riparian runoff, timing, snowmelt and accumulation, and phenology. Changes in performance of different models are associated with the importance of processes for the catchment conceptualisation. Results indicate that performance is strongly connected to the aridity of the catchment. A routing module improves the performance of almost all catchments. A snow module improves the performance in most catchments that have significant snowfall, but the crude representation of snow accumulation and snowmelt causes exceptions. Inclusion of phenology improves the performance for more than half of the catchments and interception and riparian runoff improve only a small fraction of the catchments, again the improvements are likely to be hampered by the simple lumped conceptualisation of the processes. Overall it appears to be difficult to produce generalisations of which processes should be included for an improved prediction of the daily flow values as the occurrence of several processes does not always coincide with the improved performance if the process is included in a simple lumped model.