The need for input parameter optimisation in environmental modelling is a long-known and very time-consuming task. However, to avoid tragedy, disaster propagation predictions have to satisfy hard real-time constraints. Especially small disaster control centres with limited computing resources require fast and efficient calibration methods to deliver reliable predictions in time. The combination of a clustering method together with a Genetic Algorithm is used as parameter optimisation technique in forest fire spread prediction. We formalise and demonstrate the potential of the resulting Evolutionary Intelligent System's architecture to solve the complex problem of input parameter calibration on restricted simulation conditions.