J32.3 Humans over the Forecast Loop at The Weather Company, an IBM Business

Wednesday, 15 January 2020: 9:00 AM
157C (Boston Convention and Exhibition Center)
James Lidrbauch, The Weather Company, Andover, MA

At The Weather Company, an IBM Business, we are evolving the role of the human in forecast generation. Forecasters have transitioned from a gating, synchronous step in forecast creation to an asynchronous, continuous improvement model that enhances the forecast independently from machine-driven processes, like NWP model runs. This new role is facilitated by the Human Over The Loop platform, or HOTL (pronounced “hotel”). With model updates occurring at hourly or sub-hourly frequency and with forecasts available at more than 2 billion locations on earth, it is impractical to allow human input to block forecast publication. HOTL is a graphical platform that allows human input to be specified asynchronously from automated forecast generation. The HOTL user draws geographic polygons to specify the locations to apply rules for modifying the forecast of a particular parameter over a span of time. The rules are applied whenever a forecast is requested for the relevant parameters, locations, and times. The initial purpose was to provide quality control and the addition of nuanced information to the 15-day consumer forecast. However, HOTL has been extended to new missions, including managing flight plan guidance for aviation enroute hazards, removing non-precipitating echoes from radar mosaics, and publishing the seasonal and subseasonal forecasts of temperature and precipitation anomalies. The HOTL software also serves as a meteorological workstation, allowing the users to display and animate geonavigated raster data, point data, and bulletin renderings and to use point queries show tabular or time series graphs of collections of parameters over specified times. This talk will describe how the HOTL platform changes the role of the human in the modern forecast process.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner