Early Detection of Extreme Events through NWP Plugins

Early detection of extreme meteorological events in Numerical Weather Predictions (NWP) brings an intrinsically high societal and economical value by providing precious lead time for putting effective countermeasures in place. For this reason, an increasing number of initiatives are arising to develop the underlying technological solutions. One prominent drive in Europe is the EU flagship initiative Destination Earth (DestinE) to develop highly accurate digital models of the Earth at global scale (Digital Twins).

Under the DestinE initiative, the European Centre for Medium-Range Weather Forecasts (ECMWF) is developing a mechanism that allows extending data processing functionalities of a NWP model "on-the-fly" (while the model data still resides in memory). This mechanism is based on a software called Plume that allows the development of model functionalities as plugins. Plume plugins are separate software libraries that are loaded dynamically at the beginning of the NWP simulation, access the model data in memory and perform data processing tasks like extreme event detection. This very early access of data in memory through plugins has many advantages, including reducing the detection lead time, avoiding writing output data to disk unnecessarily and promoting collaborative development of plugins to be run with a NWP model.  

The object of this presentation is the development of a Plume plugin that performs Tropical Cyclone detection and tracking. The underlying detection algorithm is based on a Machine Learning model that analyses NWP model data in memory and performs the Tropical Cyclone detection at each iteration of the model. In addition, the plugin notifies an external notification system (Aviso) whenever an extreme event is detected (according to pre-defined criteria, for example a tropical cyclone with intensity above a threshold). This whole mechanism could be used for notifying and triggering appropriate downstream workflows in response of detected events. The Tropical Cyclone detection plugin prototype is demonstrated within the ECMWF Integrated Forecasting System (IFS).  

In summary, the Plume plugin system can add on-the-fly data processing capabilities to a NWP model through plugins and can be used to implement various data processing tasks including early extreme event detection. Furthermore, a plugin architecture preserves a modular structure of additional NWP data processing functionalities  and guarantees a collaborative development environment.