Ensemble based multi-scale assimilation of Radar and In-situ Observations to Forecast Nocturnal Convection Initiation

Nocturnal convection has long been known to occur most frequently after sunset over the Great Plains (Wallace 1975). Most of the high precipitation events are associated with mesoscale convective systems (MCSs) and are accompanied by severe weather including flash floods, damaging winds, and hail. Many previous studies (e.g., Wheatley et al. 2014) have shown the benefits of radar data assimilation (DA) through improved forecasts in the structure and evolution of MCSs. However, convection initiation (CI) of these storms is notoriously difficult to forecast during the evening hours (e.g., Weisman et al. 2008) in numerical models due to the storms often involving interactions across many scales. Since nocturnal convection often initiates off of surface boundaries and cold pools, improvements in forecasting CI can be made by better resolving reflectivity and wind fields present at the time of analysis through the use of radar DA. In this study, a multi-scale, GSI-based EnKF that is extended to assimilate both radar and in-situ observations (Johnson et al. 2015) is used to address the following: (1) the impact of radar DA for nocturnal CI forecasts; (2) the optimal model and DA configuration for nocturnal CI forecasting; (3) the understanding of mechanisms that lead to nocturnal CI.

During the afternoon of June 24, 2013, isolated convection was initiated in southwest Kansas along a dry line and differential heating boundary. Operational forecasts did not note any threat for continued development into the evening. However, convection continued to initiate with elevated storms developing in northwestern Kansas by 0300 UTC June 25. As the storms propagated to the northeast, CI continued until the storms merged into a small squall line by 0700 UTC. A multi-scale WRF-ARW (v3.5.1) simulation is performed with hopes to forecast the CI in northwestern Kansas. On a 12 km outer domain, synoptic and mesoscale observations are assimilated every three hours to better analyze the pre-storm environment. The inner convection permitting domain assimilates radar observations (reflectivity and radial velocity) every five minutes to accurately analyze reflectivity and outflow structures of the storms in southern Kansas.

Our results so far have shown that the radar data assimilation has positive impacts on the prediction of the nocturnal CI. It is clear that the convection of interest is developing off of forcing from an outflow boundary produced by the initial southern storm. A control forecast without any data assimilation showed a poor representation of the southern Kansas storms with no convection occurring until close to 0700 UTC. The final mean EnKF analysis shows a similar reflectivity field as that which was observed. A deterministic forecast initialized from the final ensemble mean initiates convection at 0130 UTC, though it moves with a more northerly component than observed. Another small group of storms attempts to form in the area of interest at 0500 UTC, though they quickly die off. Work that is ongoing and will be presented at the conference include better tuning of the data assimilation parameters, upgrading to a double-moment microphysics scheme to better resolve the initial cold pool and impact of further assimilating in-situ surface observations on the inner domain.