How complex a model is necessary?

The model, as used, had a high degree of sophistication in terms of construction and data inputs. As an ``after-the-fact'' analysis of how much of this complexity is needed, figure 31 shows the observed vs predicted salinity using a very simplistic model, The model contained one squared sided lake cell, one depth flow and one volume flow.

  
Figure 31: Observed vs predicted for very simplistic model.

Given the results of the sensitivity analysis (figure 30), it is clear that the only data that really needs to be collected throughly and carefully for an effective lake model are, in decreasing importance,

• Rainfall.
• River flow.
• Evaporation.

Groundwater flow is clearly of neglicable effect. The shape and area of cells, and the shape and roughness of the joining channels is unimportant.

The differences in salinity between cells a,b,c and between cells d and e are relatively small and are not significant biologically. The model could, without loss of usefulness be reduced to two square sided (north lake d&e/south lake a,b&c) cells.

Has the time spent in making the model so sophisticated been wasted? No! The sophistication has helped in answering precisely the question of which sophistication is needed.

Any future model should need only the following information...

• Flow in the major rivers.
• Rainfall.
• Evaporation.

The lake should be made up of two square sided cells.

The following parameters should be fitted...

Total area of lake

Allowing this to be fitted provides a ``fudge'' factor to allow for unmeasurable things such as ground water flow, surface flow from the edges of the lake etc.

Ratio of the area of cell one to cell two

 

Average depth of cell

 

Mouth flow to lake level conversion factor

 

This gives rise to 5 fitted parameters, all of which have fairly good starting estimates available.