A neural network algorithm for the nowcasting of severe convective systems

Mesoscale Convective Systems (MCSs) are often correlated with heavy rainfall, thunderstorms and hail showers, frequently causing significant damages. The most intensive weather activities occur during the maturing stage of MCSs. Improving knowledge of the convective system phase and the forecast of its evolution is essential in giving better support to assistance in several fields as transports and civil protection where alert giving is an ordinary task. Different neural network techniques have been used at the Italian Meteorological Service of the Air Force to solve this problem. Nowcasting was focused on binary evolution status, only two phases. Best performance has been achieved with a system of back propagation network with a learning error designed respect the convective cell characteristic evolution phase. This system of neural networks has been learned to classify the status of the convective system and to forecast the evolution of each convective cells inside this. The input of the model is given by EUMETSAT satellite data with a time sampling of 15 minutes. This model processing the last Meteosat Second Generation (MSG) imagery in absorption and window channels evaluates some characteristic parameters of the cloud system and on this base forecasts the next phase of the cell with efficiency of 89% for the next 15 minutes and of 87% for the next 30 minutes.