The Impact of High-Resolution Surface Observations on Convective Storm Analysis with Ensemble Kalman Filter

A series of observing system simulation experiments (OSSEs) are performed using the ARPS model and its EnKF (ensemble Kalman filter) data assimilation system to investigate the impact of surface observations on the analysis and forecast of convective storms in addition to routine Doppler radar observations. A truth simulation is created for a supercell storm using the May 20, 1977 Del City, Oklahoma sounding at a 2 km horizontal resolution and the Lin et al 3-ice microphysics scheme. This storm is sampled using a radar emulator that in the vertical uses a Gaussian-shaped power weighting function, with a single WSR-88D-type radar located at different distance from the storm. Due to the earth curvature effect, the low-level coverage of the radar data decreases as the radar distance increases, causing the loss of coverage on important low-level features including the cold pool and gust front. In most experiments, the WSR-88D radar is located southwest of the model storm and the surface network has a uniform spacing. When radar is far (> 100 km) from the convective storm, the lack of the low-level data is alleviated by mesonet-like surface observations of 20 km spacing, which improves the storm analysis and forecast. Through the vertical background error covariance estimated by the EnKF, surface observations not only correct the low-level errors, but also improve the mid- and high-level fields although to a different extent for different state variables. Sensitivity experiments are performed that examine the relative impacts of different surface observation parameters. It is found that the wind observations give the largest impact. Different radar locations and surface observation network densities are also studied. When radar is closer to the storm, the impact of the surface network becomes smaller although the impact increases when the surface network density increases. The effects of model error and the error in the storm environment are also investigated. The model error is simulated by using different microphysics scheme(s) in the assimilation than that used in the truth simulation. In this case, it is found that the radar data are more effective in correcting the additional error in the state variables introduced by the microphysics error, apparently because most of such errors originate from the precipitation regions where the radar observes. It is also found that the EnKF analysis is generally improved when three different microphysics schemes are used by the ensemble members over the analyses obtained using a single, wrong, microphysics scheme. In both cases, the scheme used by the truth simulation is not included in the EnKF analysis. When moisture error exists in the low-level storm environment, surface observations display a larger impact than the corresponding case where the environment has no error.