Automated weather forecasts for all regions of Canada have been available for quite some time based on the SCRIBE expert system. The system has until now been restricted to using guidance from the numerical weather prediction (NWP) models and their derived statistical outputs either Perfect Prog (PP) or updateable Model Output statistics (UMOS). In practice, this means that the generated forecasts are solely based on model data, without any explicit observation data.

A system is under development that will merge the SCRIBE concepts with the latest local observations, using a statistical approach. This paper describes the architecture of the system and the different approaches that will be used for the very short term projection of the observation data extracted from hourly surface observations, from radar, from satellite and from lightning detection network. A prototype system has been developed to ingest hourly observations. This assimilation subsystem queries the local database at close regular intervals and extracts the relevant weather observations that are occurring at each surface observation station to update the SCRIBE weather matrices that contain the numerical forecast data. Using simple algorithms, the numerical content of the matrices is replaced with the appropriate observed values, when and where available. Further processing of the matrices up to the forecast outputs then proceeds in the regular fashion. Results from this prototype system show that 75% of the automated forecasts require updating from recent observations. Verification results will be used to assess the magnitude of the impact of ingesting the latest observation data.