An adaptive model for on-line flood predictions using a piecewise uniformity framework

An adaptive rainfall-runoff model for real-time flood forecasting in large basins is presented. It is based on a framework characterized by an ensemble of spatially homogeneous units, with the transformation from effective rainfall to direct runoff described by the Clark translation-routing procedure. Evolution in time of infiltration is estimated, in terms of saturated hydraulic conductivity, Ks, and sorptivity parameter, S, through a representation of its pre-ponding and post-ponding stages. For rainfall rates less than soil infiltration capacity, a temporal redistribution of rainfall is also incorporated. Variability of infiltration in space is represented by the corresponding rainfall variability and using spatially "equivalent" values for Ks and S. The on-line correction of flow forecast involves the updating of the S parameter and of a runoff-scaling factor (C). The model performance was evaluated by using errors on peak runoff ε and values of a persistence coefficient, V. The latter is defined so that 0 < V ≤ 1 denotes usefulness in model application and its maximum value represents perfect model performance. The model was applied to actual events observed on a large Italian basin (area 4,147 km2) and the flow forecasts for lead-times (L) up to 6 h generally compared sufficiently well with the observations. For example, for L = 6 h, the mean values of ε (in magnitude) and V were 10.0% and 0.67, respectively. Furthermore, the structure of the model seems to be reasonably appropriate, as shown by analyzing the temporal evolutions of S and C for each of the events considered.