The Development, Operation and Evaluation of Two Years of Real-Time Short-Term Precipitation Forecasting Procedure.

Digital weather radar data have been used with a simple pattern recognition technique to automatically generate precipitation forecasts in the zero to three hours range. Forecasts performed during a two year real time experiment are evaluated and compared with other basic procedures. The verification process has led to a preliminary "radar climatology" for the Montreal region in the form of maps showing areas of predominant growth or decay of precipitation patterns. The desirability to extend both the spatial and temporal scales of precipitation forecasts necessitated the incorporation of satellite data into the system. Thus, an algorithm yielding probability of rain from GOES visible and infrared imagery and simultaneous radar data is applied over an area the size of eastern Canada. Satellite data is mapped by means of a conic projection on a constant resolution Cartesian grid to facilitate overlay with synoptic charts. A pattern recognition technique is applied to 16 sub-areas of the entire map and it has proved successful in tracking the displacement of the probability of rain contours. The potential of the satellite data for hydrological purposes is also discussed.