FORECASTING OF STORM RAINFALL BY COMBINED USE OF RADAR, RAIN GAUGES AND LINEAR-MODELS

An integrated approach to real-time prediction of point rainfall is pr esented. This is based on the assumption that hourly rainfall at a sta tion can be predicted by a Multivariate AutoRegressive Integrated Movi ng Average (MARIMA) process. The real-time calibration of the multivar iate model is performed by combining radar maps and data from rain gag es. Accordingly, radar maps provide the basic information for a storm tracking procedure which enables to detect the direction and the speed of storm movement. Storm tracking is used to select those stations wh ich are characterized by the highest Lagrangian cross-correlation of o bserved precipitation, and which are therefore best suitable for appli cation of the multivariate model. The parameters of the multivariate m odel are finally estimated using only observed rainfall at the selecte d stations throughout the current event. Preliminary results of an app lication to some events which occurred in northern Italy show that the combined use of radar and rain gages allows for an increased efficien cy of the MARIMA model performances, as compared with empirical select ion of stations to be considered by the multivariate model. The multiv ariate approach performs better also when it is compared with simple n owcasting procedures based on rain gage data or on radar data used sep arately. Finally, some considerations are issued in view of a systemat ic use of this technique to nowcast rainfall intensity in small urban or natural catchments, with a response time of less than 1 or 2 h.