Answers are grounded in this paper and cite their sources.
• The CBPR-based method improves probabilistic prediction of extreme precipitation. • The method is for the Meteorological Ensemble Precipitation Processor of the US NWS. • The improvement is larger for heavier precipitation and shorter lead times. • The method objectively trades Type I error for reduced Type II error. The River Forecast Centers (RFC) in the US use the Meteorological Ensemble Forecast Processor (MEFP) to generate bias-corrected ensemble precipitation forecasts for hydrologic forecasting. Operational experience and verification indicate that the MEFP tends to under-forecast heavy-to-extreme precipitation significantly. To address the above, a conditional bias-penalized regression (CBPR) method has been developed as an enhancement to the ordinary least squares regression method currently used in the MEFP in the bivariate normal space. The two methods are then comparatively evaluated for 277 locations in 13 RFCs for the wettest consecutive 2 months using the ensemble mean precipitation forecasts from the Global Ensemble Forecast System version 12 reforecast dataset as input. The results show that the proposed method improves probabilistic forecasts for heavy-to-extreme precipitation by widely-varying margins for all 13 RFCs. The margin of improvement as measured by the continuous ranked probability skill score ranges up to about 5 to 40 % for Day-1 and Day-2 precipitation exceeding 50.8 mm for 24-hr amounts for different RFCs, and is generally larger for heavier precipitation, shorter lead times, the western RFCs and 0–96 hr precipitation. This improvement in conditional performance, however, is achieved at the expense of slight deterioration in unconditional performance and moderate wet bias. With urbanization and climate change, addressing Type-II error and conditional bias is an increasingly important topic in hydrologic forecasting. For the above, CBPR is particularly appealing in that its positive impact is likely larger for larger events.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: