This paper proposes a method for identifying the parameters and the structure of spatially distributed runoff models for a particular catchment using patterns of process information. The aim is to accurately represent the runoff processes in every single pixel to maximise the accuracy of the flood forecasts for a range of hydrological situations. The proposed method consists of five steps: (1) setting a priori parameter values; (2) fine tuning of the spatial patterns by spectral unmixing; (3) parameter calibration (fine tuning); (4) fine tuning of the model structure; (5) plausibility check of the simulated spatial patterns. The feasibility of the method is demonstrated for runoff modelling in the context of developing a flood-forecasting system in the Kamp catchment, Austria.