The rate of progress in quantitative modelling since the 1950s has been
such that application of sophisticated computer models to a wide range
of geoscientific problems is now routine. It is generally held that by
making such models more physically (physics) based, their explanatory
power and predictive reliability are enhanced. This formulation, a
model-theoretic approach, assumes accurate knowledge of the properties,
states and relationships between all of the objects that are known to
matter within the system of interest but, simultaneously, an incomplete
understanding of the totality that this knowledge creates. In
hydrological modelling, this translates into a severe dependence upon
the data models that are needed to make a hydrological model work. The
opposite extreme is a model-data approach in which measurements become
the basis of generic relationships. Even in the most heavily
data-derived cases (eg, neural network forecasting of river flows)
these data models can be shown implicitly to have a theoretical
content. Thus, both model-theoretic and model-data approaches sit
within a general class of modelling, best labelled as `data-theoretic'.
Here, we illustrate this point and advocate an approach that is
knowledge-theoretic rather than data-theoretic, to capture the much
richer sources of knowledge available to the modeller. These sources
include third-party reports, personal recollections and diaries, old
photographs and press articles, opinions, etc, which are, by
convention, either excluded from analysis, or simply added into
descriptions of model results at the point of dissemination and
consultation of model findings. We conclude by noting that this
approach to hydrological modelling fits into current thinking that the
process by which publics engage with knowledge must be moved upstream.
Here, the production of scientific knowledge comes to include not just
scientists and specialists, but also those people for whom model
predictions make a material difference.