A neural network model for the automatic detection and forecast of convective cells based on meteosat second generation data
Abstract
The 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 the development, which can be found in the case of a multi-cell storm in the centre of the convective complex systems. These convective systems may occur in several different unstable air mass; in a cold air mass behind a polar cold front, in the frontal zone of a polar front and in warm air ahead of a polar warm front. To understand the meteorological situation and apply the best conceptual model, the knowledge of the convective cluster is often not enough. In many cases the forecasters need to know the distribution of the convective cells in the cloudy cluster. A model, running in operational mode at the Italian Air Force Meteorological Service (UGM/CNMCA), for the automatic detection and forecast of the convective cells, is here proposed. The application relays on the Meteosat Second Generation infrared (IR) windows (10.8 μ m, 7.3 μ m) and the two water vapour (WV) channels (6.2 μ m and 7.3 μ m), giving as output the detection of the convective cells and their evolution for the next 15 and 30 minutes. The format of the output of the product is the last IR (10.8 μ m) image where the detected cells, their development and their tracking are represented. This multispectral method, based on a variable threshold method during the detection phase and a neural network algorithm during the forecast phase, allowed us to define a model able to detect the convective cells present in a convective cluster, plot their distribution and forecast the evolution of them for the next 15 and 30 minutes with a good efficiency. For analysing the performance of the model with the Meteosat Second Generation data, different error functions have been evaluated for various meteorological cloud contexts (i.e. high layer and cirrus clouds). Some methods for avoiding false detection in the case of inhomogeneous high layer clouds are also presented.
- Publication:
-
35th COSPAR Scientific Assembly
- Pub Date:
- 2004
- Bibcode:
- 2004cosp...35.2637P