Radar rainfall assimilation and short-range QPF in a high-resolution ensemble prediction system: a case study of summer convection
Daniel LEUENBERGER1, M. STOLL,1 Christian KEIL2 and Andrea ROSSA3
1MeteoSwiss, Zurich, Switzerland
2DLR, Institute of Atmospheric Physics, Oberpfaffenhofen, Germany
3Centro Meteorologico di Teolo, ARPA Veneto, Italy
Abstract
Summertime convection is generally hard to predict by deterministic regional Numerical Weather Prediction (NWP) systems due to model errors, incorrect initial and boundary conditions and the intrinsic chaotic nature of convection. Assimilation of radar data in high-resolution models is an efficient way to trigger convection at the right time and location and can improve particularly the initial conditions for quantitative precipitation forecasts (QPF). However, experience suggests that the introduced radar information is only successfully retained in the free forecast, if the mesoscale environment supports convection. Of particular importance is the low-level representation of temperature, humidity and wind.
A 9-year comparison between real-time radar precipitation estimates and ground observations of a high-resolution gauge network reveals large improvements achieved by modifications during the past decade. Particular attention was paid to the definition of a set of meaningful and robust descriptors of uncertainty.
In this study we examine the role of the environment in the success of high-resolution radar rainfall assimilation and short-range QPF in a summer convection case using a two-step ensemble approach: Mesoscale ensemble forecasts are produced using the COnsortium for Small-scale MOdelling Limited-area Ensemble Prediction System (COSMO-LEPS), in which the global ECMWF EPS provides initial and boundary conditions for the high-resolution non-hydrostatic Lokal-Modell (LM; Δx=7km). This ensemble drives a high-resolution (Δx=2.2km) LM ensemble wherein conventional data and radar-derived surface rainfall are assimilated with the Latent Heat Nudging technique during the first 10 hours, followed by 8 hours of free forecast. Every 7km member provides a different meso-scale representation of the convective environment and allows exploring the relation between the quality of the driving member and the skill of the model rainfall during assimilation and short-range forecast of its nested high-resolution counterpart. While at analysis time 3 of 10 members depict realistic precipitation, in the free forecast 2 of them carry the information beyond 4 hours.
Implications for a possible next-generation ensemble forecasting system with a best member selection based on remote sensing data are discussed. Moreover, the benefit of the ensemble approach over a deterministic forecast is examined.