This work evidences that inconsistencies may persist between the complexity of hydrological models and available data for model documentation and application. For example, the integrated hydrological models handle the whole water dynamics over a watershed, but are only conditioned on data that incompletely record the dimensions of the flow. It is suggested to reduce this type of model by aggregating the physical background to diminish its Euclidean dimension. Paradoxically, the complexity in the physics of a model may also result in some reduction. For example, handling a flow by relying upon a dual continuum approach conceals the structural heterogeneity of the reservoir in the model equations. The parameterization at the scale of the aquifer becomes much simpler and the model reduction is here associated with diminishing the effort to condition the model onto data.