Constraining Short-Term Storm-Scale Forecasts of Convective Initiation in OSSEs with Dense Surface Observations

Ensemble forecast skill for the convective initiation (CI) process is assessed in idealized OSSE experiments where a dense network of surface observations is assimilated.  Specifically,  CM1 ensemble simulations are performed for an environment where CI occurs due to boundary-layer processes alone.  These forecasts are cycled, using an ensemble Kalman filter to assimilate simulated observations of 2-m temperature, 2-m specific humidity, 10-m u- and v-winds, and surface pressure and at spatial densities from 16 km to 1 km.  The effect of these observations on constraining the CI process is quantified. We find that surface observations of at least 4-km horizontal spacing---and particularly with 1-km horizontal spacing---are able to spatially constrain storm-scale forecasts when compared to a control ensemble without assimilation that produces storms in random locations.  This forecast improvement is only possible when observations are assimilated within 1-2 hours of the onset of precipitation in the simulated storms, agreeing with previous estimates of convective-scale predictability.  Additional aspects of convective-scale data assimilation, including non-Gaussian ensemble estimates of the cloud field and the requirement for well-calibrated localization radii are also discussed.