**4. Potential prediction models**

The prediction of flood occurrence is the ultimate goal of modelling. Prediction models are classified into two categories: real-time direct forecasting (similar to weather forecasting – e.g. Doswell III, 1996, 1998) and flowchart-type modelling or scenario building (similar to climate change prediction). The models of the first type of forecasting do not seem useful in flash flood prediction in our case since no appropriate monitoring system exists in the study area. Applying the second approach scenarios or flowcharts can be designed for the most endangered catchments in the region (e.g. Alkema, 2003). Flowchart-type modelling takes advantage of the results of the aforementioned rapid screening analysis and risk assessment. The most probable environmental change scenarios are generated with preselected boundary conditions. The boundary conditions incorporated in the model include soil moisture content, relief, surface storage, canopy cover, cumulative rainfall and rainfall intensity. To validate the suitability of this model type we need to verify it with hindcast modelling, i.e. we perform simulation backward in time to see whether it reconstructs the observed event.

For a flowchart analysis data on preceding rainfall events have to be collected to see whether the previous rainfall event was followed by flood warning. Precipitation data originates from meteorological data usually with a 3-hour lead time. These rainfall prediction schemes determine whether a heavy convective, or a prolonged and relatively low-intensity rainfall is expected. All the scenarios in a flow chart model contain a unique code. An analytical software investigates the resemblance of the present scenario to all the predetermined scenarios and finally selects the most adequate output scenario. Finally, it supports the decision of authorities on issuing a flood warning or not.
