145 Development of a short-range probabilistic precipitation forecast algorithm based on radar and numerical prediction model input

Thursday, 29 September 2011
Grand Ballroom (William Penn Hotel)
David H. Kitzmiller, NOAA/NWS, Silver Spring, MD; and W. Wu and S. Wu
Manuscript (1.2 MB)

An algorithm for creating 0-6 hour probabilistic and deterministic quantitative precipitation forecasts (QPF) from a combination of remote sensor and operational numerical weather prediction model forecasts has been developed for the conterminous United States. The forecast package has been developed to assist in short-term hydrologic forecast operations at National Weather Service field offices, particularly those involving fast-reacting stream basins. The algorithm incorporates a unique approach to blending these forecast information sources, accounting for strengths and weaknesses in each, and it generates a unique product suite.

The QPF system generates probabilities of precipitation exceeding various threshold amounts from 0.25 mm to 75 mm, and deterministic precipitation amount, at all points on a 4-km grid mesh. The QPF values are obtained from weighted combinations of QPFs from radar- and satellite-based extrapolation forecasts, and forecasts of precipitation and humidity from the Rapid Update Cycle 2 (RUC2) model. Radar input is from the National Mosaic and Quantitative Precipitation Estimation algorithm suite developed by the National Severe Storms Laboratory. Satellite input is from the operational infrared-based GOES Hydroestimator algorithm. The forecast equation weights were generated by statistical regression between historical forecasts from the 2009-2011 period and operational gauge-radar precipitation estimates prepared at National Weather Service River Forecast Centers. Forecasts can be created immediately prior to the start of the valid period.

Verification results from within the development period indicate that the algorithm suite produces forecasts with lower root-mean squared errors and higher correlations with observed precipitation than do the constituent forecast inputs. The forecasts are generally comparable in quality to the operational forecasts produced by the National Centers for Environmental Prediction's Hydrometeorological Prediction Center. The operational and experimental forecasts both yield mean absolute errors ranging from 20-25 mm for points at which the observed precipitation was ≥ 25 mm; the differences in mean absolute errors between the two sets of forecasts range +/- 2 mm. Therefore we are confident the automated forecasts will provide useful new input to the human forecast process. Short-term plans are for real-time generation of the forecasts in a prototype framework and continued evaluation of their quality.

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