Ensemble Sensitivity Analysis Applied to a Southern Plains Convective Event

Abstract

Forecast sensitivity of an April 2012 severe convection event in northern Texas is investigated with a highresolution Weather Research and Forecasting (WRF) Model-based ensemble Kalman filter (EnKF). Through ensemble sensitivity analysis (ESA), which relates a forecast metric to initial and early forecast errors by linear regression, features of the flow are revealed that reflect dynamical relationships with the forecast convection. Results indicate that ESA can be successfully applied to high-resolution forecasts of convection, and the most important features are related to the synoptic-scale flow such as positioning of an upper-level low and lower-level thermodynamic characteristics of air masses. Comparisons of the maximum and minimum convectively active members in the region of interest show that the fields generated by ESA are consistent with the actual evolution of the event: members with more eastward progression of the synopticscale system produced a stronger convection forecast. The forecast metric of interest is modified in several ways to further evaluate the strength of the results of the sensitivity analysis. Three different variables acting as convection proxies (reflectivity, vertical velocity, and precipitation) are tested along with changing the location of the forecast metric and its spatial size. These additional tests highlight the same synoptic features of the flow with the only major differences reflecting the importance of magnitude versus position of the convective forecast.

Sensitivity of Dryline Convection Forecasts to Upstream Forecast Errors for Two Weakly Forced MPEX Cases

“…The results for this study have some intriguing implications for CI forecasts along the dryline during days without any clear feature to focus CI. Similar to previous sensitivity studies, convective forecasts appear to be sensitive to thermodynamic fields and boundaries; however, unlike these previous studies, the position of these boundaries does not appear to be tied to the position of larger-scale synoptic features that might be expected to modulate the environment (e.g., Bednarczyk and Ancell 2015;Torn and Romine 2015;Berman et al 2017;Hill et al 2016). Moreover, these convective forecasts appear to be most sensitive to the water vapor above the boundary layer; however, the results here suggest that the evolution of water vapor is not merely the advection of initial condition uncertainty.…”

“…As a consequence, this study employs the maximum vertical kinetic energy [( 1 /2)w 2 max , where w max is the maximum vertical motion in a column) forecast metric averaged over a geographic area and time period (hereafter J VKE ). 1 This metric has been used in previous studies (e.g., Berman et al 2017) and unlike precipitation or reflectivity-based metrics, which have been used in other sensitivity studies (e.g., Bednarczyk and Ancell 2015;Torn and Romine 2015;Hill et al 2016), it is relatively straightforward to separate convection from extensive stratiform precipitation. Moreover, the time period of the greatest variability in this metric occurs between 22 and 25 h (2200-0100 UTC); therefore, this time period is used in the metric calculation.…”

“…There are numerous methods to estimate forecast sensitivities, including adjoint (e.g., Errico et al 2003) and ensemble-based methods (e.g., Martin and Xue 2006;Melhauser and Zhang 2012;Bednarczyk and Ancell 2015;Torn and Romine 2015;Berman et al 2017;Hill et al 2016). Ensemble-based methods may be better posed for convective applications because they employ a set of nonlinear forecast trajectories to estimate sensitivities, rather than assuming linear forecast error growth (i.e., adjoints).…”

“…5). Given the proximity of this metric to the dryline and previous studies that demonstrate the sensitivity of convection forecasts to moisture boundary locations (e.g., Bednarczyk and Ancell 2015;FIG. 4.…”

Seeding Chaos: The Dire Consequences of Numerical Noise in NWP Perturbation Experiments

“…A linear regression of a chosen metric valid later in the forecast window, such as accumulated precipitation, is performed onto the perturbed initial-condition variable (e.g., soil moisture). Numerous ESA studies have shown significant and useful linear relationships between initial-condition variables and forecast variables, even when the chosen metric involves storm-scale variables and substantial nonlinearity (Bednarczyk and Ancell 2015;Hill et al 2016). Figure 9 shows an example of how ESA can discriminate real dynamical processes from unrealistic ones created through chaos seeding.…”