Constrained_CA
The research in this project is aimed at gaining experience with the application of Constrained Cellular Automata to coastal management and if possible to increase its use by solving a number of difficulties concerning:
Geographical resolution
Time scale
Uncertainty
Recognising and comparing geographical patterns
Calibration of CA models
Transition rules
The problem of [i]Geographical resolution and Time scale[/i] studied relates to the problems of choosing the right scales to represent a problem, and further of combining processes running at different geographical and temporal scales in one and the same model. The solutions worked out are rather pragmatic at this moment and require much more (fundamental) research work. A framework is provided with sufficient flexibility to deal with a wide range of processes.
The issue of [i]Uncertainty[/i] addressed in the project is different from what was originally intended at the outset of the project. During the project more and more the awareness rose that uncertainty is inherent to any prediction for systems that display complex behaviour. The approach taken therefore was not so much focussed on reducing the uncertainty in the model, but to learn more about the uncertainty in the real world problem represented by the model. This was facilitated by the development of a tool and a step-by-step procedure to implement scenarios, enabling the exploration of different policies under particular scenarios. The whole range of consequences and risks for the system visualised during this exercise can be of much help to the decision maker. The Decision Support System developed does not provide a complete turnkey solution; it still leaves the user with the difficult task of developing scenarios that are integrated, complete and consistent. Yet, the tools support greatly in the analysis of possible evolutions and tendencies in the modelled system.
In order to evaluate results generated by any high resolution spatial model, it is paramount to be able to [i]recognise and compare the geographical patterns[/i] generated. A cell-by-cell comparison is easy to apply; yet it is not the most useful for finding qualitative similarities. In the project three methods, each with its own useful application domain, have been developed and tested. Clearly recognising and comparing maps is an essential step in calibration and validation of models equally well. The project did not get very far on this subject.
[i]Calibration and Validation[/i] of Cellular Automata models are difficult problems. This is not in the least so because of their non-linear complex behaviour. Manual calibration has shown to be feasible but t is very tedious. Moreover there is little certainty that the best possible set of parameters for the transition rules is found at the end of the exercise. Yet, the analyst carrying out the calibration is applying a particular technique: he is focussing his attention on areas on the maps generated where the CA model is clearly wrong. Next he tries to correct these mistakes by adjusting the rule set till the model is generating acceptable results. In the project a method has been developed which is automating this kind of behaviour: it is adjusting the model in function of the largest deviations between the calculated map and the map with the goal state and proceeds onwards from there till the model produces a map that is similar to the goal state. This work has led to an automatic calibration procedure in embryonic form. Further research will be required to develop it into a practical calibration instrument.