Abstract:
For reliable prediction of urban pluvial flooding it is essential to have reliable spatial
and temporal rainfall prediction at an appropriate scale. Radar data are considered
to be the most reliable. However, many urban catchments do not have access to
radar data. This paper presents a new methodology for rainfall forecasting based on
a network of raingauges. The methodology predicts rainfall at each raingauge
location based on the Support Vector Machine (SVM) technique improved with
Singular Spectrum Analysis (SSA). The prediction of the spatial distribution of rainfall
is based on interpolation techniques. The forecasted rainfall fields are then used as
inputs for simulation of drainage systems to obtain a short-term flood prediction. The
proposed methodology is tested using real data from a case study in Coimbra,
Portugal and the results obtained showed that it is possible to predict the water level
30 minutes in advance when using this methodology.
