Model initial condition sensitivity of downslope wind storms

With the advancement of numerical weather prediction models, forecasting mesoscale phenomenon such as downslope wind storms and mountain waves is common place. While these models are quite capable of producing realistic and plausible solutions, their actual ability produce valid forecasts of detailed mesoscale structure remains largely unexplored. One important unanswered question is, what is the sensitivity of the forecasted feature to the initial model conditions? If a model solution is extremely sensitive to initial conditions, then the phenomenon is thought to have limited predictability. In this study we use an ensemble of 70 equally likely members to examine the sensitivity of downslope wind storms, drawn from the T-REX data set, to initial model conditions.

An Ensemble Square-root Kalman Filter is used to generate the ensemble, which is then integrated forward with COAMPS at horizontal resolutions down to 3 km. With the 70 ensemble members it is possible to find a statistically-significant, linear relationship between the modeled downslope winds at a given forecast time and the initial conditions. As an example, during T-REX IOP 6 the ensemble shows that the strength of downslope wind speeds in the Owens Valley is strongly anti-correlated with the upstream upper-level wind speed. Increasing the wind speed in the initial conditions at the point of maximum anti-correlation by a mere 3 m/s leads to downslope wind speed decreases exceeding 10 m/s in a short 6 hour forecast. This rapid error growth suggests an inherent lack of predictability in at least some downslope flows.