5.2 Point Blend in NWS Operations

Tuesday, 24 January 2017: 4:15 PM
608 (Washington State Convention Center )
Thomas J. LeFebvre, GSD, Boulder, CO; and K. L. Manross

As part of our Forecast Decision Support Environment (FDSE) work defined by the National Weather Service’s (NWS) Weather Ready Nation Roadmap, the Global Systems Division has developed a gridpoint-oriented blending technique in order to improve the quality and reliability of the short-term gridded forecasts generated at Weather Forecast Offices (WFOs). Point Blend archives all relevant gridded observational and short-term model forecast guidance and then assesses the history each model’s performance at each forecast grid point in an attempt to minimize forecast error. Where a particular model performs poorly, those grid points are weighted lower than others that have performed better. By weighting and then blending point by point, spatial differences in performance can be measured and accounted for, resulting in a more accurate forecast when compared to other techniques where each model is weighted evenly over the forecast spatial domain. In addition, since the archive database is updated in real-time, changes in model performance are automatically accounted for with no human intervention. Results have shown that, on average, this technique produces gridded forecasts with lower errors than past model blending efforts. This implementation of model grid blending occurs at the WFO and is configurable to include any model guidance available within that office and with minimal latency.

With development and testing complete, the Point Blend system has been deployed quasi-operationally to a limited number of WFOs located in Montana in order to gather feedback in an operational environment. Preliminary results indicate that Point Blend accuracy is consistent with that measured during testing. We will present some preliminary performance results from the field including short-term forecast accuracy when compared to other model guidance for several forecast variables, system resource consumption, and future plans for the Point Blend framework.

- Indicates paper has been withdrawn from meeting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