Impact of radar data assimilation and orography on predictability of deep convection

Abstract

Deep convection represents a classic example of limited predictability on the convective scale. We investigate the potential impact of assimilating radar reflectivity and velocity observations on the predictive skill of precipitation in short-term forecasts (up to 6 hr) using the operational COSMO-KENDA ensemble data assimilation and forecasting system in an idealized set-up. Additionally, the role of a Gaussian-shaped mountain providing a permanent source of predictability for the location of convective precipitation is examined with and without data assimilation.

Using a hierarchy of quality measures, we found a long-lasting beneficial impact of radar data assimilation throughout the entire forecast range of 6 hr. The up-scaled normalized RMS error and the Fractions Skill Score show that precipitation forecasts based on initial conditions including the assimilation of radar data are skilful on scales larger than 40 km at a lead time of 6 hr and thus are better than a reference ensemble without any data assimilation at lead times of less than 1 hr. The presence of orography strongly increases the predictability of precipitation throughout the forecast range, particularly within the immediate area and where no radar data are assimilated.

This remarkable impact of radar data assimilation exceeding 6 hr is larger and longer-lasting than in many real modelling systems. While this is partly related to the idealized set-up assuming a perfect forecast model, perfect large-scale boundary conditions and a perfect radar forward operator, our study demonstrates the potential impact that could be achieved for radar data assimilation if the systematic model and operator deficiencies, as well as boundary condition errors, could be reduced. Furthermore, our results highlight the important role of orography in structuring the precipitation field, especially if no observations are assimilated.