In most general terms, there are two alternative ways to obtain design flood estimates and to relate them to some probability (frequency) of occurrence:

- statistical analysis of observed stream flow data, and
- rainfall-runoff modelling.

*Statistical analysis of observed flows* is the basic approach for estimation of flood-frequency relationship when adequate data on observed flows are available. If only a short record of streamflow is available, it is sometimes possible to establish a regional relationship with flows from neighbouring stations and to use such relationship to extend the existing short record in order to perform statistical analysis. Application of flood frequency analysis is limited to gauged catchments, i.e. to gauged stream profiles. When flow rates of certain probability of occurrence are required at ungauged sites or along a stream (e.g. for flood zone mapping), other methods are needed for flood estimation.

In case when observed flow data is not available or is inadequate, the alternative is to obtain flood estimates using a *rainfall-runoff model*. In general, rainfall-runoff models provide flood runoff estimates for a given rainfall input. Depending on the type of rainfall input, there are two possible approaches involving rainfall-runoff modelling.

*Design storm approach* is the approach in which a design storm represents the input for rainfall-runoff model, and the design flow is the model output. Design storms themselves are obtained by statistical analysis of observed rainfall. The key issue in this approach is the assumption that the frequency of the design flow is equal to the frequency of the design storm. However, having in mind that the rainfall-runoff relationship is neither linear nor straightforward, such an assumption is not always very realistic.

In the *observed storms approach* rainfall-runoff model is applied to observed rainfall data in order to obtain corresponding simulated flows. The simulated flows are then treated as a substitute for observed flows, and they are statistically analyzed in order to obtain flood-frequency relationship. This approach can be *event-based*, when separate storms are used as input to runoff models, or *continuous*, when the complete rainfall record is used for continuous runoff modelling. The disadvantage of the event-based rainfall-runoff modelling is that it requires estimation of initial conditions at the beginning of each event, while continuous rainfall-runoff modelling is quite demanding concerning data and model calibration.