Evaluating Predictions of Convective Environments using the Climate Forecast System Version 2

The prospect for skillful long-term predictions of atmospheric conditions known to directly contribute to the onset and maintenance of severe convective storms remains unclear. Increasing levels of interest in extended-range predictions of severe weather continue to motivate our ongoing assessment of the NCEP Climate Forecast System Version 2 (CFSv2) forecasts for periods coinciding with climatological peaks in tornadoes, large hail and severe winds. Our assessment of the CFSv2 utilizes both real-time model runs along with reforecast data and is verified against analysis output from the Climate Data Assimilation System (CDAS). Comparisons are generated between the CDAS and select upper-air sounding locations for additional examination of the model output. Furthermore, a 20-year climatology from the CFSv2 Reanalysis has been generated as a baseline upon which predictions can be made and scrutinized with regard to the long-term mean.

Our particular focus is on the prediction of convective environments, which we represent in the form of environmental parameters. Because environmental convective available potential energy (CAPE) and deep layer vertical wind shear can be used to distinguish an atmosphere conducive to intense rotating convection from one supportive of primarily non-severe ‘ordinary' convection, we have limited our assessment to these two parameters. Each parameter is addressed individually to avoid the potential of one robust parameter masking the weakness of another. Connections between model verification and reports of severe weather are investigated.

Utilizing a series of methodologies designed to analyze the CFSv2 output from different perspectives, early results from Fall (SON) 2013 through Spring (AMJ) 2015 indicate potential value in CFSv2 forecasts over periods as long as multiple weeks. With a focus on periods exhibiting significant anomalous behavior in both CAPE and deep layer shear, trends in root-mean-squared difference (RMSD) and Spearman rank correlation coefficient demonstrate noteworthy skill. Larger RMSD errors tend to be more closely correlated with parameter magnitude and/or number of days exceeding a predetermined parameter threshold (e.g. 1000 J kg-1 of surface based CAPE or 20 m s-1 of surface–500 hPa shear) than widespread errors in geographical coverage. Conceivable methods for probabilistic outcomes, including aggregates of CFSv2 forecasts and evaluations of model performance in conjunction with various modes of the Madden Julian Oscillation (MJO) and El Nino-Southern Oscillation (ENSO), will be presented from the results of the research with an overarching goal of iterating towards an operationally useful product.