A radar reflectivity operator with ice-phase hydrometeors for variational data assimilation (version 1.0) and its evaluation with real radar data

Shizhang Wang,Zhiquan Liu

doi:10.5194/gmd-12-4031-2019

Abstract

Abstract. A reflectivity forward operator and its associated tangent linear and adjoint operators (together named RadarVar) were developed for variational data assimilation (DA). RadarVar can analyze both rainwater and ice-phase species (snow and graupel) by directly assimilating radar reflectivity observations. The results of three-dimensional variational (3D-Var) DA experiments with a 3 km grid mesh setting of the Weather Research and Forecasting (WRF) model showed that RadarVar was effective at producing an analysis of reflectivity pattern and intensity similar to the observed data. Two to three outer loops with 50–100 iterations in each loop were needed to obtain a converged 3-D analysis of reflectivity, rainwater, snow, and graupel, including the melting layers with mixed-phase hydrometeors. It is shown that the deficiencies in the analysis using this operator, caused by the poor quality of the background fields and the use of the static background error covariance, can be partially resolved by using radar-retrieved hydrometeors in a preprocessing step and tuning the spatial correlation length scales of the background errors. The direct radar reflectivity assimilation using RadarVar also improved the short-term (2–5 h) precipitation forecasts compared to those of the experiment without DA.

Highlights

IntroductionOver the past several decades, radar reflectivity observations have been used in many data assimilation (DA) studies (Borderies et al, 2018; Caumont et al, 2010; Gao and Stensrud, 2012; Hu et al, 2006; Jung et al, 2010, 2008a; Liu et al, 2019; Putnam et al, 2014; Snook et al, 2012, 2015; Sun and Crook, 1997; Sun and Wang, 2013; Tong and Xue, 2005; Wang et al, 2013b; Wang and Wang, 2017; Wattrelot et al, 2014; Xiao et al, 2007; Xue et al, 2006) and they have demonstrated that assimilating this radar reflectivity improves the initial conditions of the convective scale and benefits the subsequent forecasts
As the first attempt to implement and apply RadarVar in WRF data assimilation (WRFDA), this study focused on the quality of the analysis using the univariate three-dimensional variational (3D-Var) data assimilation (DA) method in terms of the root mean square difference (RMSD) against the observed reflectivity and the similarity between the observed reflectivity distributions and the analysis
Spurious echoes appear over unobserved areas in the analyses at both 00:00 and 01:00 Z and most likely resulted from the spatial correlations in the background error covariance; these correlations allow the propagation of information from observed areas to unobserved areas both horizontally and vertically

Summary

Introduction

Over the past several decades, radar reflectivity observations have been used in many data assimilation (DA) studies (Borderies et al, 2018; Caumont et al, 2010; Gao and Stensrud, 2012; Hu et al, 2006; Jung et al, 2010, 2008a; Liu et al, 2019; Putnam et al, 2014; Snook et al, 2012, 2015; Sun and Crook, 1997; Sun and Wang, 2013; Tong and Xue, 2005; Wang et al, 2013b; Wang and Wang, 2017; Wattrelot et al, 2014; Xiao et al, 2007; Xue et al, 2006) and they have demonstrated that assimilating this radar reflectivity improves the initial conditions of the convective scale and benefits the subsequent forecasts. It is necessary to transform the model’s prognostic variables (e.g., rainwater, snow, and graupel) to the observed radar reflectivity To perform this transformation, early studies (e.g., Sun and Crook, 1997; Xiao et al, 2007) used the Marshall–Palmer distribution of raindrop size (Z–R relationship). Early studies (e.g., Sun and Crook, 1997; Xiao et al, 2007) used the Marshall–Palmer distribution of raindrop size (Z–R relationship) This relationship is only valid in precipitation areas without ice-phase species; its applications (e.g., Schwitalla and Wulfmeyer, 2018) are often limited to layers lower than 4 km or 8 km above ground level (a.g.l.). Several studies (e.g., Gao and Stensrud, 2012; Wang and Wang, 2017) have demonstrated that involving these ice species in the reflectivity operator improves the analysis of hydrometeors in terms of their spatial distribution, especially in the vertical direction

Objectives

Results

Conclusion