Butterflies and bow echoes: addressing poor forecasts with ensemble simulations
Bow-echo structures, a subset of mesoscale convective systems (MCSs), are associated with damaging wind and hail, and are often poorly forecast within deterministic numerical weather prediction model simulations. Among other things, this may be due to inherent low predictability associated with bow echoes, error in the initial conditions ICs), and inadequate parameterization schemes (model error). Ensemble simulations account for, and measure, these uncertainties in a forecast.
A study of two bow-echo cases, simulated with multiple ensemble configurations, find that location and timing variation of the simulated systems reduce when certain parameters are fixed (i.e., ICs; microphysical parameterizations). However, variations in convective mode remain substantial. Results suggest the MCS positioning is influenced primarily by ICs, but its mode is most sensitive to the model-error uncertainty.
A modification to the Structure Amplitude Location (SAL) method identifies and compares objects in both forecast and observed composite reflectivity fields. Both the original and modified SAL methods are used to evaluate daily 12-km North American Model (NAM) forecasts during the summer of 2015 for a central United States domain. SAL using reflectivity reveals a diurnal cycle of skill, with minimum skill occurring early-to-late afternoon (local time), and maximum skill occurring just before sunrise.
The modified SAL method is then deployed to evaluate the effect of finer resolution on both ensemble spread and the character of bow-echo development. Due to the increased prominence of noise close to the truncated scale, we expect a larger ensemble spread as horizontal grid spacing decreases. Two ensemble forecasts were generated using the Weather Research and Forecasting (WRF) model: one used a single domain with 3-km horizontal grid spacing, and another nested a 1-km domain inside the 3-km parent domain with two-way feedback. Ensemble members were then generated from the control with a stochastic kinetic-energy backscatter scheme, with identical initial and lateral-boundary conditions. Results show that the increase in grid resolution reduces both spread and skill, and that the nested ensemble produces a faster bow echo and stronger cold pools. The latter two are most likely due to increased (fractal) cloud surface area within the nested ensemble, which allow more entrainment of dry air and hence increased evaporative cooling.
Finally, we address the poor performance in previous bow-echo studies by evaluating a WRF hindcast dataset designed to capture numerous MCSs in the Great Plains. We may expect the skill of the hindcasts to be dictated by (a) inherent synoptic-scale predictability (i.e., ensemble spread), and (b) the skill of the NAM forecast dataset providing initial and lateral-boundary conditions to the WRF hindcast. However, there is no obvious relationship between the accuracy of MCS convective mode and either factor. When the MCS dataset is confined to cases containing bow echoes, we find that serial bow echoes (i.e., line-echo wave patterns) are better forecast by the WRF hindcasts than progressive bow echoes. Furthermore, stronger rising motion is linked with the propensity for bow echoes to be serial rather than progressive. We therefore speculate that the skill of storm-scale forecasts may inherit only limited characteristics of the large-scale predictability, perhaps due to rapid downscale cascade and growth of initially trivial errors in the initial-condition dataset.
In summary, as model errors appear random (not systematic), and the reduction of IC error yields only diminishing returns, we deem it likely that poor bow-echo forecasts stem from inherent low predictability. This demands the use of well-calibrated ensemble systems, accounting for both model and IC error, to properly gauge the probabilities of bow-echo events and their associated hazards.