The influence of pbl parameterization on the practical predictability of convection initiation during the mesoscale predictability experiment (MPEX).

摘要

This study evaluates the influence of planetary boundary layer (PBL) parameterizations on short-range (0-15 h) forecasts of convection initiation (CI) within convection-allowing ensembles that utilize sub-synoptic-scale observations collected during the Mesoscale Predictability Experiment (MPEX). Running five thirty-member ensembles with the Advanced Research Weather Research and Forecasting Model (WRF-ARW) with each differing only in the chosen PBL parameterization, forecast skill, PBL sensitivity on the environment in which CI occurred, and the variability within are examined. Three MPEX cases, 19-20 May 2013, 31 May-1 June 2013, and 8-9 June 2013 are considered, each characterized by a different large-scale flow pattern to analyze a wider spectrum of events. Using an object-based method to verify and analyze the forecasts of CI, it was found that none of the Five PBL schemes analyzed significantly improved the forecast skill. The non-local mixing PBL schemes, MYJ and QNSE, had in all cases higher probability of detection (POD) but consequently had a higher false alarm ratio (FAR) resulting from the models overproducing the number of CI objects, with all PBLs, and thus resulting in relative high bias scores as well. The CSI showed only subtle changes between PBL schemes suggesting no one PBL scheme drastically outperforms the other. The temporal distribution of errors associated with the "hits" in the CI object matching showed an approximate normal distribution around a mean of 0-s suggesting little systematic timing bias. While the spatial distribution of errors yielded skewed distributions with on average a mean (median) distance error of just over 44-km (28-km). Analysis of cumulative distribution functions (CDFs) of the "hits" highlighted limits to increased forecast skill beyond temporal and spatial thresholds of 60-min and 100-km. Mean error (ME) plots computed for surface features as well as vertical profiles in pre-convective environments highlighted biases in both the initial conditions as well as between ensembles. In agreement with previous studies, it was found that non-local mixing PBL schemes tend to produce PBLs that are too shallow, cool, and moist while local mixing schemes tend to be deeper, warmer, and drier as a function of the stronger (weaker) vertical mixing within the local (non-local) PBL schemes. Relative to the analysis of the vertical profiles, it was seen that the model has an inherent inability to accurately represent strong capping inversions in models across all PBL schemes suggesting an issue with the handling of vertical diffusion within the PBL and the implicit damping associated with the discretization schemes used within WRF.