Geoscience Reference
In-Depth Information
be transformed simply into likelihood measures (Figure B7.1.3). The simplest transformation
would use a triangular function, analogous to a triangular fuzzy measure, but other forms of
transformation are also easily implemented according to the characteristics of particular ob-
servations. Note that the resulting likelihood measures apply to single observations. They need
to be combined in some way to form an overall likelihood weight for that model simulation.
Figure B7.1.3 Conversion of scores to likelihoods.
B7.1.5 Qualitative Measures for Model Evaluation
Model evaluation need not be based solely on quantitative measures of model fit to observa-
tions. Some qualitative or “soft” measures might also be useful (e.g. Seibert and McDonnell,
2002; Vache and McDonnell, 2006). An example would be in evaluating the predictions of a
distributed model. The distributed model might, with different parameter sets, be able to obtain
good fits to observed discharges using different runoff generation mechanisms, e.g. infiltration
excess runoff alone, saturation excess runoff alone, subsurface stormflow alone or a mixture of
them. If, however, observations suggest that there is negligible overland flow generated on the
hillslopes on a catchment, then models that give good simulations based on such mechanisms
should be rejected, even though those models might actually give the best simulations. Other
soft measures might be used based on acceptability in reproducing various hydrological
 
Search WWH ::




Custom Search