Comparison and Verification of Point‐Wise and Patch‐Wise Localized Probability‐Matched Mean Algorithms for Ensemble Consensus Precipitation Forecasts. Issue 12 (24th June 2020)

• Record Type: Journal Article
• Authors: Snook, Nathan; Kong, Fanyou; Clark, Adam; Roberts, Brett; Brewster, Keith A.; Xue, Ming
• Is Part Of: Geophysical research letters. Volume 47:Issue 12(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-06-24
• Subjects: Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2020GL087839
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: When applied to precipitation on large forecast domains, the probability‐matched ensemble mean (PM mean) can exhibit biases and artifacts due to using distributions from widely varying precipitation regimes. Recent studies have investigated localized PM (LPM) means, which apply the PM mean over local areas surrounding individual points or local patches, the latter requiring far fewer computational resources. In this study, point‐wise and patch‐wise LPM means are evaluated for 18–24‐hr precipitation forecasts of a quasi‐operational ensemble of 10 Finite‐Volume Cubed‐Sphere (FV3) forecast members. Point‐wise and patch‐wise LPM means exhibited similar forecast performance, outperforming PM and simple means in terms of fractions skill score and variance spectra while exhibiting superior bias characteristics when light smoothing was applied. Based on the results, an LPM mean using local patches of 60 × 60 km and calculation domains of 180 × 180 km is well suited for operational warm‐season precipitation forecasting over the contiguous United States. Plain Language Summary: Weather and rainfall are often predicted using ensembles of numerical weather forecast models. The skill of the ensemble consensus is often better than any individual forecast, and valuable information about the range of possible outcomes and model uncertainty is gained. In this study, different methods for implementing a localized probability‐matched mean (LPM mean) are examined. The LPM mean is designed to produce a more accurate consensus from a forecast ensemble while retaining local structures that other consensus methods fail to capture. Two LPM variations were examined for predicting accumulated precipitation—one computed at every model grid point and another on patches containing many nearby points. Both methods produced similar results and outperformed traditional ensemble consensus algorithms. The patch‐based method took 1 to 2 orders of magnitude less time to compute. Operational weather providers should consider using the patch‐based LPM mean algorithm to efficiently compute ensemble rainfall forecasts. Key Points: The localized probability‐matched mean has distinct advantages over other ensemble consensus products for precipitation forecasts The computational performance of the localized probability‐matched mean can be greatly improved by using a patch‐based algorithm Localized probability‐matched mean forecasts show good objective skill and earn high subjective ratings from human forecasters