My experience in water quality modelling is quite limited: just an application on marine dispersion based on bidimensional grid but with only one quality variable in it, a generic “pollutant” relative concentration. Really barbaric, in some way, but effective too: it could easily give at least an idea of the plumes generated by some generic dissolved, conservative compound. The model was not calibrated too. Really, really rude. Fortunately, the expected result was not a deterministic time-space forecast of chemical distributions, but just an idea of some time-scale magnitudes, therefore we were quite satisfied.
Models are rarely wrong, it is the modeller’s (or the politician’s) expectations that make them inconsistent.
But what about the real water quality models? I take an example from the huge technical reports base of the US Army Corps of Engineers Environmental Laboratory.
Well, I did not read the whole document (888 pages, even if text is less than 10%). I started from the abstract, continued throughout the calibration graphs and stopped. The 15 (!) water quality variables do a lot of funny things in these figures, compared to the measured data points, and it made me ask to myself if such a big and magnificent model could really tell us something interesting.
I must admit that later the authors are very equilibrated and precise in delimiting the results meanings and extracting prudent conclusions from the whole job. If people just could learn from this wise modus operandi… especially when talking about global climate models!