An efficient modular volume‐scanning radar forward operator for NWP models: description and coupling to the COSMO model

Abstract

Since radar observations are highly dense in spatial and temporal resolutions, they have been often used to improve short‐term numerical weather prediction (NWP) by means of detailed model verification and 3D radar data assimilation. However, the observed quantities are not directly comparable to the prognostic variables of NWP models (e.g. hydrometeor densities, wind vector, temperature, pressure, etc.), so a common approach to facilitate this comparison is to derive synthetic radar observations from model variables; this is the so‐called ‘radar forward operator’. In the present article, a new Efficient Modular VOlume scanning RADar Operator (EMVORADO) for Doppler velocity and reflectivity is introduced. Although it has been developed in the COSMO model framework, it can be also coupled online to any other NWP model. Comprehensive physical aspects of radar measurements (e.g. beam bending/broadening/shielding, Doppler velocity with fall speed and reflectivity weighting, attenuated reflectivity, detectable signal, etc.) have been implemented in a modular way, using state‐of‐the‐art methods with different levels of approximation and numerical costs that can be optionally chosen. The reflectivity derivation from the prognostic model variables is as ‘model consistent’ as possible and carefully honours the uncertainties associated with partially melted particles. Efficiency and applicability on supercomputers (MPI‐parallelism) is a major design criterion, which allows us to simulate entire networks of 3D volume‐scanning meteorological radars within one model run and makes EMVORADO well suited for operational applications. This article aims to give a thorough description of the EMVORADO and to provide a first insight to the performance of different modules by some selected case‐studies.

From model to radar variables: a new forward polarimetric radar operator for COSMO

“…In our operator, similarly to Caumont et al (2006) and Zeng et al (2016), we set W (r 0 − r) = 1 if r ∈ r 0 − cτ 4 , r 0 + cτ 4 and W (r 0 −r) = 0 otherwise. Indeed since the model resolution (1-2 km) is about one order of magnitude larger than the typical gate length of a modern radar (80-250 m), effects related to the finite receiver bandwidth can be neglected.…”

“…w j and z j are respectively the weights and the roots of the Hermite polynomial of order K (for elevational integration) and w k and z k are the weights and roots of the Hermite polynomial of order K (for azimuthal integration). For the integration in the radar operator, default values of J = 5 and K = 7 are used according to Zeng et al (2016). The quadrature points thus correspond to separate sub-beams with different azimuth and elevation angles that are resolved independently.…”

“…Finally, Zeng et al (2016) developed a forward radar operator for the COSMO model. The operator is designed for operational purposes (assimilation and validation) with an emphasis on performance and modularity.…”

Representation of Model Error in Convective‐Scale Data Assimilation: Additive Noise, Relaxation Methods, and Combinations

“…In addition, the DWD operates a network of 17 C‐band Doppler radars over Germany and part of neighboring countries (see Figure 1 of Zeng et al, ), which can provide reflectivity and radial wind measurements in interval of 5 min. So far, the LHN technique has been applied to each ensemble member to assimilate radar‐derived precipitation rates (Schraff et al, ; Stephan et al, ).…”

“…So far, the LHN technique has been applied to each ensemble member to assimilate radar‐derived precipitation rates (Schraff et al, ; Stephan et al, ). In this work, radar reflectivity observations are assimilated by the LETKF, using the efficient radar forward operator EMVORADO (Zeng et al, , ). Radial wind observations are not assimilated and serve as an independent data source for verification.…”