**1. Introduction**

There have been higher requirements put forward for the prediction of convective systems precipitation and its related disasters in recent years. Improving the accuracy of precipitation forecast has long been a challenge for Numerical Weather Prediction (NWP) model researchers and operational communities [1,2]. Kryza and Werner et al. [3] forecasted several short and intensive rainfalls over the SW area of Poland using the Weather Research and Forecasting Model (WRF) with different parameterization and spatial resolution. The results show that none of the experimented model configurations was able to reproduce a local intensive rainfall properly. Hamill and Thomas [4] applied ensemble prediction systems to describe the performance of the WRF precipitation forecasts. Although ensemble forecast can reflect the predictability or reliability of real atmosphere to some extent, it cannot improve the physical mechanism of the model. Among many considerable causes that could lead to the inherent low predictability of convective precipitation forecasting

 

 Liu, J.; Li, C.; Yu, F.;Wang, W. Effect of the Assimilation Frequency of Radar Reflectivity on Rain Storm Prediction by Using WRF-3DVAR. *Remote Sens.* **2021**, *13*, 2103. https://doi.org/10.3390/ rs13112103

Academic Editor: Yongqiang Zhang

Received: 23 April 2021 Accepted: 24 May 2021 Published: 27 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

using an NWP system, ambiguous description of the atmosphere initial state is one of them. Rainfall features are always generated in an inaccurate manner, regarding location, initiation, timing, and intensity, especially of convective storms [5]. As a result, accurately obtaining the initial state of a storm for the regional model is the key issue for a successful prediction of the convective system.

Several studies have confirmed it is possible to remedy defects through data assimilation into the NWP model. Among the numerous data assimilation methods, those commonly used by researchers include the optimal interpolation methods, the threedimensional variational (3DVAR) [6,7], the four-dimensional variational (4DVAR) [8,9], and the Kalman filter [10] approaches. Thereinto, 3DVAR is practicable in terms of computational efficiency, thus is adopted the most frequently [5]. Variational data analysis system was first developed by Sun and Juanzhen et al. [11], which is called Variational Doppler Radar Analysis System (VDRAS) and began this pioneering work. Subsequently, it was expanded by Sun and Crook [12] to use for short-term forecast initialization during convective precipitation. Yang and Duan et al. [13] discussed if 3DVar data assimilation could potentially improve the rainfall forecast in the WRF model, an obvious improvement of the event with even rainfall in temporal distribution was found regarding the northeastern Tibetan Plateau area. Physical initialization combined with the three-dimensional variational data assimilation method (PI3DVAR\_rh) was designed by Gan and Yang et al. [14], through which the spatial pattern forecasting of radar reflectivity and precipitation were improved based on WRF. Vedrasco and Sun et al. [15] certified that the 3DVAR analysis with the constraint introduced into WRF could improve the initial state of the model and guide the convective characteristics exhibited by six summer convection cases across Brazil. These studies show that 3DAVR plays a positive role in rainfall forecast, and this system, which is widely used in data assimilation, is also adopted in this study.

With high-resolution observation data becoming increasingly rich, it has been a hot research topic worldwide in recent years to improve the short-term quantitative precipitation prediction level by using high-resolution observation. High-resolution observations like Doppler radar and GTS data, can provide huge amounts of detailed information with high spatial and temporal resolution, which makes it possible to further improve the convective rainstorm forecast. Doppler radar can make millions of measurements of precipitation with a spatial resolution of a few kilometers and a temporal resolution of a few minutes [5]. Compared with existing meso-scale measure platforms, such a significant advantage of spatial and temporal resolution has grea<sup>t</sup> potential for improving small-scale and short-term rainfall prediction.

Many researchers combine high-resolution observational data with numerical models, which plays a positive role in promoting the development of convective rainfall prediction, improving the initial state of numerical models, as well as alleviating the imbalance caused by interpolation [16–19]. Routray and Mohanty [20] believed that the assimilation of radar data (radial velocity and reflectivity factor) has a positive impact on enhancing performance of the WRF-3DVAR (WRF-three-dimensional variational) system for the Indian region. Gvindankutty and Chandrasekar et al. [21] investigated how the 3DVAR assimilation of DWR radial wind and GTS data positively affected the precipitation intensity as well as spatial distribution. Abhilash and Sahai et al. [22] demonstrated improvement in spatial pattern of rainfall of convective systems precipitation by assimilating relevant parameter obtained by the WRF-3DVAR system, including DWR radial velocity and reflectivity as well as GTS data. Osuri and Mohanty et al. [23] used the WRF-3DVAR system to assimilate DWR with GTS, thus conducting 24 cases to predict tropical cyclones of Bay of Bengal, and the results show that this method helps to provide a positive impact on the credibility of prediction. Sugimoto and Crook et al. [5] indicated that in the 3DVAR framework and the storm case, assimilating the radial velocity as well as reflectivity can achieve the best performance, applied on short-range precipitation forecasting. A cycling data assimilation improves the regional models' initial state on the one hand, and meanwhile introduces observation data to mitigate the imbalance caused by interpolation of prediction on the

other hand [24]. The data assimilation system allows higher-resolution observations to be used as the background through update-cycling procedure [25,26], which makes the data assimilation system rely on a specified combination of error statistics to obtain the optimal short-term forecast analysis.

Although data assimilation is capable of improving the NWP (i.e., WRF) convective precipitation in an effectively manner, the quality of data assimilation creates uncertainty about results. Some studies have shown that the quality of assimilated data is critical to forecast results. Liu and Tian et al. [27] indicated that data assimilation via WRF-3DVar could potentially improve the rainfall forecasting in northern China, with GTS data, radar reflectivity, and radial velocity assimilated every 6 h. From their study, it is concluded that it is the effective information in the assimilated data that exhibited more significant results rather than the volume of data. Tian and Liu et al. [28] further explored the effect of assimilating radar data from different height layers on the improvement of the NWP rainfall accuracy. The results showed that the accuracy of the forecasted rainfall deteriorated with the rise of the height of the assimilated radar reflectivity. In the process of data assimilation, the frequency of assimilation determines the amount of effective data assimilated. Considering the time cost, most operation departments choose a 6 h assimilation interval [3,27,29]. In later developments, many researchers and principal operational centers upgraded the regional NWP systems with a 3 h assimilation time interval [30]. For highly convective storms, more frequent data assimilation with a shorter time interval is found to be more effective to produce reliable predictions [31]. Li and Wang et al. [32] demonstrated that the assimilation of radial velocity every half an hour could enhance the intensity analyses and forecasts of rainfall compared to results without assimilating radar data. Kawabata and Seko et al. [33] applied four-dimensional variational (4DVAR) with a horizontal resolution of 2 km and 1 h length of the assimilation window to forecast heavy rainfall at the central part of Tokyo. In most cases, researchers tend to assimilate observations as they are originally obtained, rather than choose an appropriate assimilation frequency or time interval.

Despite the wide application of data assimilation in enhancing precipitation forecasts, the sensitivity of data assimilation frequency has not ye<sup>t</sup> gained enough attention. This study mainly aims at evaluating how the WRF-3DVAR with different assimilation frequencies affects the accuracy of the forecast precipitation. Four typical rain storms that occurred in semi-humid and semi-arid area of northern China were chosen as study objects. Doppler radar and GTS data were assimilated in four designed experiments with the time intervals of 6 and 1 h by WRF-3DVAR. The results would be helpful to improving data assimilation efficiency with WRF-3DVAR, and provide guidance for the development of a similar basin rainstorm forecast system. At the same time, in order to study the sensitivity of data assimilation to rainfall forecast, the quality of radar data is also analyzed.
