3.2 Improving Deterministic Time-Lagged Ensemble Forecasts By Applying Blended Nowcasts

Monday, 23 January 2017: 4:15 PM
Conference Center: Skagit 2 (Washington State Convention Center )
Yunsung Hwang, Pukyong National University, Busan, Korea, Republic of (South); and D. I. Lee, A. J. Clark, and C. H. You

Handout (9.6 MB)

Improving short-term (0-8 hours) forecasts (nowcasts) is important for the prevention of accidents in the fields of aviation, agriculture, construction, and emergency management, etc. Predicting possible locations of convective cells can lower the economic costs associated with rescheduling or rerouting commercial flights in aviation. Previously, a new blending method (Saliency-based cross-dissolve, SAL) was developed and tested by applying different weights to extrapolation and model forecasts (High Resolution Rapid Refresh, HRRR) based on intensities and forecast times using echo-top heights from Weather Surveillance Radar 1988 Doppler (WSR-88D) in Continental United States (CONUS).

In order to represent probability of severe weather in CONUS (up to 8 hours), time-lagged ensembles are suggested by combining the deterministic maps of the mosaics of column max (CMAX) from rapidly updated HRRR, extrapolation (EXT) and blending of HRRR and EXT (SAL). Three probabilities of 23 and 40 dBZ are predicted using different numbers of previous forecasts (i.e., 1 to 7 previous forecasts) and initializations (00 to 23 UTC) from the data of mid-May to mid-June 2015. The skill scores of the ensemble forecasts (Brier scores, reliability diagrams, areas under Relative Operating Characteristics curves, AUCs) of SAL and EXT (probabilities of over 23 and 40 dBZ) showed better results compared to HRRR (especially 1500 to 1900 UTC for SAL). Ensemble forecasts of SAL and HRRR showed improved skills (i.e., lower Brier scores) as number of previous forecasts increase while those of EXT showed worse results (i.e., higher Brier scores) in 1 to 3 hours forecasts. Reliability diagrams showed the best skills of ensemble forecasts of SAL except forecast 1 hour forecasts (EXT showed the best skill scores). Mean AUCs of accumulated ensemble forecasts of SAL represented highest scores throughout the forecast hours while those of EXT and HRRR showed lower skill scores. The improved results of ensemble forecasts of blended nowcasts (EXT and HRRR) showed possible applications to decision making by providing probabilities of severs weather in the short-term forecasts (especially, forecast hours 1 to 5 hours).


This work was funded by the Korea Meteorological Industry Promotion Agency under Grant KMIPA 2015-5060.

This work was financially supported by the BK21 plus Project of the Graduate School of Earth Environmental Hazard System.

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