Wenru Shi , Hishn Chen , Xin-Zhong Ling ,

a Key Laboratory of Meteorological Disaster, Ministry of Education (KLME)/Joint International Research Laboratory of Climate and Environment Change(ILCEC)/Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters (CIC-FEMD), Nanjing University of Information Science and Technology, Nanjing, China

b Earth System Science Interdisciplinary Center, University of Maryland, 5825 University Research Court, Suite 4001, College Park, MD, USA

Keywords:Tropical cyclone Ensemble simulation Physical parameterization schemes Regional downscaling

ABSTRACT To evaluate the downscaling ability with respect to tropical cyclones (TCs) near China and its sensitivity to the model physics representation, the authors performed a multi-physics ensemble simulation with the regional Climate-Weather Research and Forecasting (CWRF) model at a 30 km resolution driven by ERA-Interim reanalysis data. The ensemble consisted of 28 integrations during 1979-2016 with varying CWRF physics configurations.Both CWRF and ERA-Interim can generally capture the seasonal cycle and interannual variation of the TC number near China, but evidently underestimate them. The CWRF downscaling and its multi-physics ensemble can notably reduce the underestimation and significantly improve the simulation of the TC occurrences. The skill enhancement is especially large in terms of the interannual variation, which is most sensitive to the cumulus scheme, followed by the boundary layer, surface and radiation schemes, but weakly sensitive to the cloud and microphysics schemes. Generally, the Noah surface scheme, CAML(CAM radiation scheme as implemented by Liang together with the diagnostic cloud cover scheme of Xu and Randall(1996)) radiation scheme, prognostic cloud scheme, and Thompson microphysics scheme stand out for their better performance in simulating the interannual variation of TC number. However, the Emanuel cumulus and MYNN boundary layer schemes produce severe interannual biases. Our study provides a valuable reference for CWRF application to improve the understanding and prediction of TC activity.

1. Introduction

Tropical cyclones (TCs) are one of the most disastrous weather events. Powerful TCs are usually accompanied by strong wind and heavy rain, which cause huge economic loss and casualties. Accurate shortterm forecasting and seasonal prediction of TC activity has been a big challenge in meteorological operational work and the research community. During the past 20 years, great efforts have been devoted to the simulation and prediction of TC activity. Vitart and Stockdale (2001) performed seasonal forecasting of Atlantic tropical storms by using a coupled global climate model, indicating that the model could basically capture the main features of TC activity. The potential usage of numerical models in TC prediction has also been explored ( Vitart et al., 2003 ;Camargo and Barnston, 2009 ). Recently, more and more studies have carried out simulations and predictions of TC activity by using both regional climate models (RCMs) ( Landman et al., 2005 ; Camargo et al.,2007 ; Au-Yeung and Chan, 2012 ; Liang et al., 2013 ; Wu and Gao, 2019 )and high-resolution global climate models ( Chen and Lin, 2011 , 2013 ;Vecchi et al., 2014 ; Murakami et al., 2015 , 2016 ). Several studies have explored the future changes of TCs in different basins with global atmospheric models ( Murakami and Wang, 2010 ; Murakami et al., 2011 ;Murakami et al., 2013 ).

The western North Pacific (WNP) is a well-known TC-active basin.The simulation and prediction of TCs over this region have drawn increasing attention. For example, Au-Yeung and Chan (2012) simulated the TC activity in the WNP during 1982-2000 with RegCM3, showing that the model can capture the seasonal cycle of TC genesis but has evident bias in simulating the tracks of TCs. Liang et al. (2013) evaluated the performance of the PRECIS model of the Hadley Center in simulating the TC activity in the WNP region. Wu and Gao (2019) found that RegCM4 can reproduce the seasonal cycle of TCs but evidently underestimates the frequency and intensity of TCs in the WNP region. Jin et al. (2016) evaluated the performance of six RCMs in simulating the TC activity over East Asia. Overall, the regional models showed potential in the seasonal prediction of TCs over the WNP region. However,their capabilities need to be comprehensively evaluated and improved.In addition, physical processes play important roles in TC activity. Recent studies have investigated the sensitivity of TC genesis, track and intensity to model physical parameterization schemes. Cumulus schemes,microphysical schemes, and planetary boundary layer (PBL) schemes have been proven to be important in improving the simulation of TC activity ( Ha et al., 2009 ; Lai et al., 2010 ; Raju et al., 2011 ; Osuri et al.,2012 ; Pattanayak et al., 2012 ). Notably, there are still very few studies that have focused on the sensitivity of seasonal TC simulation and prediction to physical parameterization schemes, which deserves further investigation for the purpose of seasonally predicting TC activity.

The regional Climate-Weather Research and Forecasting (CWRF)model was developed by Liang et al. (2012) . The model offers massive improvements with respect to the land-atmosphere-ocean, convectionmicrophysics, and cloud-aerosol-radiation interactions, as well as the system consistency ( Liang et al., 2004 , 2005a , 2005b , 2005c , 2005d ;Choi et al., 2007 ; Liang et al., 2006 ; Choi and Liang, 2010 ; Liang and Zhang, 2013 ) throughout all process modules, compared to the original WRF model ( Skamarock et al., 2008 ). CWRF has been widely applied in both short-term weather prediction and long-term climate prediction( Yuan and Liang, 2011 , Liang et al., 2012 ; Chen et al., 2016 ; Qiao and Liang, 2016a , 2016b ). More recently, a high-resolution ensemble simulation and prediction system has been developed based on CWRF and has been used in seasonal climate prediction in China ( Liang et al., 2019 ).However, the performance of the CWRF-based ensemble prediction system in simulating TC activity still needs to be further evaluated before its application to TC seasonal prediction in China. Since operationally the seasonal prediction of TC activity focuses mainly on the TC occurrence or number, our study emphasizes the simulation of TC number only.

In this study, we investigated the performance of the CWRF model in simulating the TC activity and its sensitivity to model physics based on a 28-member ensemble simulation. The model description and experimental design are presented in section 2 , and section 3 describes the observation data and TC tracking method. Section 4 evaluates the model performance and section 5 explores the capability of the CWRF ensemble simulation and its sensitivity to model physics. Finally, a summary is given in section 6.

2. Model description and experimental design

The model used in this study, i.e., CWRF, integrates a comprehensive ensemble of alternate parameterization schemes for each of the key physical processes, including cumulus, microphysics, radiation, PBL,surface, and cloud, which can facilitate the use of an optimized physics ensemble approach to improve weather or climate prediction. A more detailed and systematic description of CWRF’s physical processes can be found in Liang et al. (2012) .

Fig. 1. (a) Annual cycle and (b) interannual variation of TC number during 1982-2016.

The model domain was centered at (35.18°N, 110°E), with 232 ×172 grid points in the horizontal, which covers China and its adjacent regions (8.37°-48.14°N, 58.4°-137.39°E), and the 14 grids along the four edges of the domain were the buffer zones. The horizontal resolution was 30 km and 36 terrain-following levels were used in the vertical direction. We performed 28 runs with different physical scheme configurations ( Table 1 ) forced by the ～79 km grids and 60 levels of ERA-Interim 6 h reanalysis data ( Dee et al., 2011 ), and all runs were integrated for the period from 1 October 1979 to 31 December 2016 (with the first two months used for spinup). This paper analyzes all the TCs that occurred in the model domain, with the GFDL Vortex Tracker ( Biswas et al., 2018 ) used for TC detection.

3. Comparison between the CWRF control run and ERA-Interim

We firstly evaluated the performance of CWRF in simulating the TC number by comparing the results of the control (CTL) run (E0 experiment) with the JTWC (Joint Typhoon Warning Center) and ERA-Interim reanalysis data. It is noted that both CWRF and ERA-Interim can basically capture the seasonal cycle of the TC number ( Fig. 1 (a)). Both the model and ERA-Interim can reproduce the peaks of the TC number during July to October (JASO) and bear significant correlation with the JTWC data ( Table 2 ). However, both the model and ERA-Interim evidently underestimate the TC number, especially in the TC-active period(JASO). ERA-Interim seriously underestimates the occurrence of the TCs in JASO. Compared to ERA-Interim, the downscaling by CWRF evidently reduces the bias, with the RMSE reducing from 0.71 to 0.29 ( Table 2 ). In terms of the interannual variation of TC activity, both ERA-Interim and the CTL run capture the interannual variation of the TC number during 1982-2016 ( Fig. 1 (b)). Similarly, the model performs better than ERAInterim in simulating the TC time series. ERA-Interim can capture about 2/3 of the JTWC TC number and seriously underestimates the number of TCs, bearing a correlation of 0.55 with the JTWC data ( Table 2 ). Compared to ERA-Interim, CWRF can produce the TC number much closer to the JTWC result by reducing the RMSE from 7.48 to 3.23 and improving the correlation from 0.55 to 0.84 ( Table 2 ). Generally, CWRF CTL can basically capture the annual and interannual variations of the TCs near China, and the downscaling by the regional model adds significant value to using ERA-Interim reanalysis data to simulate the TC activity.

4. Ensemble simulation and sensitivity to model physics

To further evaluate the capability of the ensemble simulation in TC simulation, we compared the ensemble mean (CWRF-Avg) and simulations of 28 members (CTWF-ensemble) with the ERA-Interim and JTWCRepresents the same physical schemes as used in the CTL(E0) run.data ( Fig. 2 ) in terms of both the seasonal cycle and interannual variations of TC number near China. We also present their correlations and RMSEs with the JTWC data in Table 2 for quantitative analysis. From Fig. 2 (a), CWRF-Avg and CWRF-ensemble show similar performance to the CTL run in their simulation of the seasonal cycle. However, CWRFAvg can improve the simulation of CTL by reducing the RMSE from 3.23 to 3.14 and increasing the correlation from 0.84 to 0.90 ( Table 2 ). The above results suggest that the ensemble simulation does not show evident advantages in simulating the TC seasonal cycle, but can effectively improve the performance in simulating the interannual variations of TC activity, indicating a potential application of the ensemble simulation in the seasonal prediction of TC activity. In addition, it is noted that all individual members of the 28-member ensemble simulation obviously outperform ERA-Interim, which demonstrates the potential advantages of the downscaling by CWRF.

Table 1 Description of the experimental design.

Table 2 RMSE and correlations between the ERA-Interim/model results and the JTWC TC number in the period 1982-2016.

Fig. 2. (a) Annual cycle and (b) interannual variation of TC number during 1982-2016. Shading represents the ensemble results of the 28 members.

We also examined the results of each member of the ensemble simulation to explore the sensitivity of the TC simulation to the model physics( Fig. 3 ). Overall, there is no big difference ( Fig. 3 (a, b)) in the spread of the correlation between the individual members and the JTWC data(all around 0.99), with the RMSEs ranging from 0.27 to 0.36 ( Table 2 ), which indicates that the TC seasonal cycle is not very sensitive to the model physics. Certainly, it is partly because the model has better skill to simulate the seasonal cycle. Among all the physical schemes, the NSAS [New Simplified Arakawa-Schubert scheme] cumulus scheme, Lin microphysics scheme, and CAM radiation scheme stand out for their better performance in seasonal cycle simulation by comprehensively evaluating both the RMSE and correlation between the JTWC. In contrast,the simulation of the interannual variations of the TC number are very sensitive to the model physics, which can be concluded from the relatively distinct difference in both RSMEs and correlations among different physics and the same kind of physical schemes ( Fig. 3 (c, d)). The spreads of the ensemble simulation in the correlation between the individual members and the JTWC data range from 0.77 to 0.89, and the RMSEs vary between 2.95 and 4.01 ( Table 2 ). Generally, the simulation of the interannual variation of TC number is most sensitive to the cumulus scheme, followed by the boundary layer, surface and radiation schemes, but is weakly sensitive to the cloud and microphysics schemes. Generally, the Noah surface scheme, CAM and CAML (CAM radiation scheme as implemented by Liang together with the diagnostic cloud cover scheme of Xu and Randall (1996)) radiation schemes,prognostic CC[Could Cover] cloud scheme, and Thompson microphysics scheme stand out for their better performance in simulating the interannual variation ( Fig. 3 (d)); whereas, the Emanuel cumulus scheme and MYNN boundary layer schemes produce serious bias in terms of the simulated interannual variations of TC activity.

Fig. 3. Differences (each member minus the ensemble average) in (a, c) RMSE and (b, d) correlation of the (a, b) annual cycle and (c, d) interannual variation of TC number between the simulations by the 28 CWRF members and the JTWC from 1982 to 2016.

Based on our results, the multi-physics ensemble has the potential to be applied in the seasonal prediction of TC number near China. Besides the ensemble strategy, it is very important to consider the sensitivity of the simulation to the model physics and select the optimal physical schemes for the specific purpose. For simulation of the interannual variation of TC number near China and its seasonal prediction with CWRF,configuration of the model physics with the ECP (Ensemble cumulus parameterization modified from G3) cumulus, Thompson microphysics,CAM/CAML radiation, CAM3 boundary layer, Noah surface, and prognostic CC cloud schemes is recommended.

5. Discussion and conclusion

In this study, we performed a multi-physics ensemble simulation for the period 1979-2016 by using CWRF to evaluate the performance of the downscaling simulation of TCs that occurred over mainland China and its coastal oceans and to understand their sensitivity to the model physical parameterization schemes.

Results show that both CWRF and ERA-Interim can basically capture the seasonal cycle and interannual variation of the TC number. However, both the model and ERA-Interim evidently underestimate the TC number near China. The downscaling by CWRF evidently reduces the bias and can significantly improve the simulation of TC occurrences, especially in terms of the interannual variation. The downscaling by CWRF adds significant value in using ERA-Interim reanalysis data to simulate TC activity. It was found that the multi-physics ensemble simulation of CWRF (CWRF-Avg) can significantly improve the simulation of the interannual variation of TC number, which has the potential to be used in the seasonal prediction of TC number near China.

Further analysis shows that the simulated seasonal cycle of TCs seems to be insensitive to the model physics; however, simulation of the interannual variations is very sensitive to the model physics -mostly to the cumulus scheme, followed by the boundary layer, surface and radiation schemes, but is weakly sensitive to the cloud and microphysics schemes. Generally, the Noah land surface scheme, CAM and CAML radiation schemes, prognostic CC cloud scheme, and Thompson microphysics scheme stand out for their better performance in terms of the simulation of the interannual variation. However, the Emanuel cumulus scheme and MYNN boundary layer schemes produce serious bias in simulating the interannual variations of TC activity.

Our study can provide valuable information for the application of CWRF and ensemble simulations in the simulation and prediction of TC activity near China. We have also provided a preliminary analysis of the sensitivity of the simulation of TC number over this region, which can help in configuring models used in the regional simulation of TC activity.Certainly, further evaluation is still needed in the future, which will help towards a better understanding of the underlying physical mechanisms involved and improvements in the simulation of TCs and their seasonal prediction.

Funding

This study was supported by the National Climate Center of China under Grants 2211011816501.

Acknowledgments

The authors thank the Maryland Advanced Research Computing Center (USA) and the Wuxi National Supercomputing Center (China) for their support in terms of computer resources.