180 likes | 275 Vues
Explore the development and performance of a Multisensor Precipitation Nowcaster (MPN) for providing automated forecast guidance and leading to improved flash flood warnings. The analysis evaluates forecast accuracy based on rain rate predictions without gage data mosaicking. Various algorithm configurations are tested across multiple cases to compare forecast statistics against observed data. Results show improvements in forecast bias, RMSE, POD, FAR, and CSI relative to a persistence method.
E N D
6R.4 PRELIMINARY RESULTS FOR THE 0-1 HOUR MULTISENSOR PRECIPITATION NOWCASTER Shucai Guan and Feng Ding RS Information Systems/Hydrology Laboratory Richard Fulton and David Kitzmiller Hydrology Laboratory Office of Hydrologic Development National Weather Service, NOAA Silver Spring, Maryland 32nd Conference on Radar Meteorology26 October 2005, Albuquerque, NM
Outline • Introduction • Description of the Multisensor Precipitation Nowcaster (MPN) • Analysis Method and Results • Conclusions
Introduction • NWS mission includes warning operations for flash flooding conditions, currently the greatest storm-related threat to life in the United States. • MPN is developed for NWS Weather Forecast Offices to provide additional automated forecast guidance and lead-time for issuance of flash flood warnings. • The purpose: evaluate the accuracy of a scaled-down MPN (no gage data and no mosaicking) forecasts of rainrate and establish a baseline of performance.
Description of MPN • 4-km resolution, updated every 5 min; forecast 1-hour accumulated precipitation and 0-1 hour rain rates. • Can use rain gauge data to adjust the radar rainrates. • Mosaics regional radar data before making the forecast. • Uses a standard local pattern-matching scheme to estimate storm motion. • Three options for the smoothing: 1) no smoothing, 2) adaptable smoothing using the Flash Flood Potential method, or 3) a method proposed by Bellon and Zawadski (1994) (hereafter called BZ94). • Growth/decay of local rain rates.
7 flash flood cases in the MD-VA-PA region are investigated. • Six statistics (Bias, RMSE, COR, POD, FAR, and CSI) are used to evaluate and compare the accuracy of the parameter tests. • There are13 algorithm configurations for each case: 2(growth/decay; N or G) X 3 smoothing (none, FFP method, BZ94 method; N or F or B) X 2 (local vs. area-averaged storm motion; L or A) + persistence (PRS). For example, NFL is test with turning off growth/decay, using FFP smoothing and local storm motion.
Example of observed and forecasted 60-minute rain rate and one-hour accumulation images for June 13, 2003 (MPN with option NFL)
Growth/decay option as implemented causes positive bias in forecasts Smoothing option reduces bias in forecasts
Turning off growth/decay option results a perceptible improvement on RMSE after the 30 minute forecast Smoothing option reduces RMSE
The smoothing option increases correlation Turning on the growth/decay option has negligible improvement on correlation
Turning on the growth/decay option and smoothing option improve POD The effect of the growth/decay option is much larger than that of the smoothing option at 60 minutes into the forecast
Turning on the growth/decay option increases FAR The smoothing produces notable improvement on FAR after the 30 minute forecast
The smoothing improves CSI The growth/decay option has small mixed effect on CSI
-10% -10% -11% +71% +67% +32% +32% +54% +54%
Conclusions • MPN substantially improves all six statistics relative to persistence method.The progressive spatial smoothing creates major improvement for all six statistics. • Comparing with persistence, MPN: • Reduces RMSE by 24%. • Raises POD by 71% for rainrate > 5 mm/h. • Raises POD by 32% for rainrate >15 mm/h. • Decreases FAR by about 10%. • Increases CSI by 67% for rainrate > 5 mm/h. • Increases CSI by 54% for rainrate > 15 mm/h.