2.6 SASSE: Object-Oriented Analysis, Classifying and Nowcasting of Convective Storms in Finland

Monday, 8 January 2018: 11:45 AM
Room 7 (ACC) (Austin, Texas)
Roope Tervo, Finnish Meteorological Institute, Helsinki, Finland; and J. Karjalainen, S. Oksman, and P. Rossi

Severe convective weather causes significant harm for electricity supply in Finland. The hazards produced by intense winds, lightning and tornadoes have significant social impacts and cause remarkable liability for damages for distribution companies.

Since an individual storm may have a lifespan of a few tens of minutes and the spatial extent of a few kilometres, continuous real-time measurements are needed to identify and track convective storms, to categorise them and to assess their potential impacts to clients’ electricity network. Remote sensing instruments such as weather radars and lightning location systems are capable of fulfilling the requirements for high temporal and spatial resolution.

Storms are identified by contouring weather radar reflectivity based on pre-defined thresholds. Object-oriented storm tracking, which follows the movement of individual identified storms and then extrapolates the storms based on the tracking information, is used to track and forecast storms paths.

Storm cells are classified based on several input parameters. Currently radar reflectivity, lightning data and client’s electricity outage information are used in the process. In the future at least echo top height are going to be added to the classification. Outage information is used both in real-time classification and as a training data.

Four-step classification (from no damage to severe damage) is used at the moment. Clients’ outage data archive is used for training and real-time outage information to the classification. The training is done by a random forests algorithm supervised with percentage of outages from whole electricity network. A classification of the latest observed time step of the storm cell is used to forecast the damage.

UI consists of pannable and zoomable map view of storm cells over clients’ electricity network. User can get a graph of forecasted damage in a certain point by clicking the map. The graph contains also an estimation of total cost of liability for damage.

OpenCV-library is used for contouring the storm cells from weather radar data. Identified storm cells are handled in object-oriented form and stored in PostGIS-database. SciPy-library is used for the training and classification. End results are visualized by SmartMet Server WMS and OpenLayers UI library.

No proper validation has been done yet, but initial results are very promising. The system is under constant development. After validation, adding more parameters to the classification is on a roadmap. Utilizing more sophisticated classification methods is also an interesting topic along with applying the methodology to other weather information like EPS-forecasts.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner