1 Introduction

As one of the worst natural disasters, tropical cyclones (TCs) have immense destructive power along the track region, bringing strong wind, heavy rain, and storm surges that can cause substantial losses in both human lives and the economy, particularly when making landfall. The western North Pacific (WNP) region experiences the greatest TC activities and most frequent landfalls of all ocean basins (Maue 2011). It is critical to improve the forecast accuracy of TCs as well as to understand their genesis, motion, intensification, and activity. As the global climate changes, TC climatology has become an important subject of ongoing research (Emanuel 2005; Knutson et al. 2010; Walsh et al. 2016). Many previous studies have examined the climatology of TC activity over different regions using global or regional climate models (GCMs and RCMs, respectively) (Kim et al. 2014; Manganello et al. 2012; Wu et al. 2012; Zhao et al. 2009).

Although GCMs have been used to simulate TC activity, their coarse resolution makes it difficult to reproduce realistic TC intensity. High-resolution GCMs simulate more realistic intensities, and thus are more suitable, but require high computing capacity (Barcikowska 2013; Manganello et al. 2014; Walsh et al. 2010). By providing finer model resolution, more detailed physical processes, and topographic information, RCMs have been widely adopted to investigate inter-annual changes in TCs (Jin et al. 2015). Many studies have shown that RCMs have improved skill in simulating the genesis and movement of TCs, and can improve our understanding of the key processes and mechanisms of TC activity (Cha et al. 2011; Diro et al. 2014; Feser et al. 2011; Knutson et al. 2007; Wu et al. 2014). The RCMs can also reproduce realistic inter-annual TC variability (Jin et al. 2015; Nguyen and Walsh 2001). For example, by using an 18-km-grid regional model, Knutson et al. (2007) captured a recent increasing trend in Atlantic hurricanes with a high correlation coefficient between simulations and observations.

Driven by global reanalysis data or GCM output, RCMs can generate added value for certain variables and locations, particularly those influenced by specific regional factors (Feser et al. 2011). However, RCM simulations may be compromised if the model degrades the large-scale information prescribed by the global analysis. To prevent the RCM deviating from the large-scale driving field, spectral nudging (SN) methods have been widely applied in regional climate downscaling and TC simulations (Feser and Barcikowska 2012; Hong and Chang 2012; Jin et al. 2015; von Storch et al. 2000; Wang et al. 2013). The general terms of SN are to impose the time-variable large-scale information from driving field on the RCM without affect the regional and small-scale disturbance development of RCM (von Storch et al. 2000). Many studies have shown that improved RCM simulations are obtained when SN method is used. For example, Song et al. (2011) reported that the SN method used in the MM5 RCM led to an improved precipitation distribution over East Asia.

While SN has positive effects on many aspects of RCM research, it is unclear whether SN can improve simulations of TC activity. Cha et al. (2011) suggested that continuous SN weakens TC intensity by suppressing the intrinsic intensification. Many researches show that SN can decrease the exaggerated rate of TC genesis in RCM simulations (Feser and Barcikowska 2012; Knutson et al. 2007). For instance, Knutson et al. (2007) showed that SN improved the inter-annual variations in Atlantic hurricanes activities, and the contrast between active and inactive TC seasons. Previous studies have also suggested that the SN effect on simulated TC activity is highly sensitive to details of the nudging (e.g., the nudging variables and nudging coefficient) (Cha et al. 2011; Knutson et al. 2007). In addition, the conclusions of previous studies are based mainly on a single TC season and their applicability to long-term simulations has not been investigated in detail.

In addition to the nudging scheme, physical parameterizations have been important factors in simulations of TCs. Among the physics schemes, the radiation scheme plays a key role in long-term regional climate modeling, but few studies have considered the radiation effect on long-term simulations of TC genesis and activities. Caron et al. (2013) showed that a modified radiation scheme induced significant changes in the simulated Atlantic TC activity. In the present study, we use two most common radiation schemes, Community Atmosphere Model (CAM) and Rapid Radiative Transfer Model (RRTM), in long-term regional climate simulations to investigate how the radiation scheme influences the TC activity in long-term TC climatology simulations.

In this study, we present results from a series of sensitivity experiments conducted with the Weather Research and Forecasting (WRF) model (Skamarock et al. 2008) driven by European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis data (ERAIN; Dee et al. 2011) for the East Asia domain specified in the East Asia Coordinated Regional Downscaling Experiment (CORDEX; Giorgi et al. 2009). To assess the sensitivity of long-term TC simulations to the SN (also include nudging variables and coefficient) and radiation parameterization schemes, eight 20-year (1989–2008) regional climate simulations with different model configurations are performed. The remainder of this paper is arranged as follows. In Sect. 2, the experimental design, data, and the TC detection method are described. Section 3 gives a detailed analysis of the model skill in simulating the structure and climatology of TCs, their inter-annual and seasonal cycles, and the large-scale environment associated with TC activity. A discussion of the effect of the SN method and different radiation schemes on TC genesis and activity is provided in Sect. 4. Finally, Sect. 5 provides summary and conclusion remarks of the TC simulations using different schemes.

2 Model, experiment design, data, and TC tracking method

2.1 Model and experiment design

The regional climate model used here is the WRFv3 model with the Advanced Research WRF (ARW) dynamic core (Skamarock et al. 2008), which has been widely used in TC simulations (Jin et al. 2015; Ma et al. 2012; Nguyen et al. 2014) and regional climate modeling (Bukovsky and Karoly 2009; Leung and Qian 2009). The model domain is centered at (25°N, 120°E), and has 232 × 196 grid points in the zonal and meridional directions with a horizontal grid spacing of 50 km. This domain, which corresponds to the CORDEX East Asia domain (Fig. 1), covers East Asia and the West Pacific, where TCs are generated and move northwest. There are 28 terrain-following sigma levels in the vertical direction with model top at 50 hPa. The WRF model is run continuously for 20 years, from 1 Jan 1989 to 31 Dec 2008, driven by the ERA-Interim reanalysis (ERAIN), which is at 0.75° × 0.75° resolution (Dee et al. 2011). Sea surface temperatures (SSTs) are prescribed by the Optimum Interpolation Sea Surface Temperature (OI SST) weekly data from the National Ocean and Atmospheric Administration (Reynolds et al. 2002).

Fig. 1
figure 1

The WRF model simulation domain and topography (color shading, m)

Full size image

To investigate the effects of the SN method and radiation scheme on TC climatology simulation, eight WRF experiments (Table 1) are designed, corresponding to the possible combinations of the SN method [with SN (include nudging variables and coefficient) or without SN (NOSN)] and two radiation schemes (CAM, Collins et al. 2004; RRTM, Iacono et al. 2008). The control experiments, which use the traditional nudging through the lateral boundaries (Davies 1976) with 15-point sponge zone (NOSN), combined with two difference radiation schemes are referred as CAM and RRTM experiment, respectively. The SN experiments perform nudging not only in the lateral boundaries but also in the interior of model domain. The effect of SN can be modified by modulating the factors such as the nudging components and nudging coefficient. Two types of nudging variables are used: one nudges only the horizontal winds, and the other nudges both the horizontal winds and temperature. Two nudging coefficients are used in the SN experiments: one uses nudging coefficient of 3.0 × 10−4 with corresponding time scale of 1 h, and the other uses a smaller value of 2.5 × 10−5 with corresponding time scale of 12 h. The experiments named by SN-UV in Table 1 represent the simulation with SN applied only to horizontal wind and nudging time scale of 1 h. Experiments SN-UVT are the same as SN-UV, except that temperature is also nudged by the SN. Compared to SN-UVT experiments, experiments SN-UVT-12h adopt a longer relaxation time scale of 12 h. When the spectral nudging is conducted, only the waves with wavenumbers less than the SN setting are nudged. In this study, the CORDEX-EA domain size is about 11,000 km × 8000 km in zonal and meridional directions respectively. Hence, the wavenumber 3 is employed in both directions, relaxing the WRF outputs toward the wavelengths that are resolved in ERAIN (>2500 km here). And the SN is only applied above the planetary boundary layer (PBL) to maximize the freedom of mesoscale circulation development (Liu et al. 2012; Otte et al. 2012).

Table 1 Summary of the parameterization combinations used in the WRF simulations
Full size table

The other physical parameterizations used in this study include the WRF Single-Moment 5-class (WSM5) microphysical parameterization (Hong et al. 2004), the new Kain–Fritsch convective parameterization (Kain 2004), the Yonsei University (YSU) PBL scheme (Hong et al. 2006), and the Noah land surface model (Chen and Dudhia 2001).

2.2 Data

The dataset of observed TC best tracks is produced by the Regional Specialized Meteorological Centers (RSMC) Tokyo Typhoon Center, based on the International Best Track Archive for Climate Stewardship (IBTrACS; Knapp et al. 2010). It consists of 6-h best-track data and contains information such as the central location of named TCs and the associated maximum wind speed (MSW) and minimum sea level pressure for the WNP basin. To compare with the WRF-simulated TCs, only TCs with intensity at least 17 m s−1, which is corresponding to 10-min sustained surface MSW, in the best-track data are included. The TC best track dataset covering the period 1989–2008 is used to evaluate the WRF simulations.

To assess the effect of different physical parameterizations on TC simulation, the ERAIN dataset on a 0.75° × 0.75° grid is used to evaluate the model performance in simulating large-scale circulation and moisture.

2.3 TC detection and tracking scheme

The TC tracking algorithm employed in this study is that proposed by Nguyen and Walsh (2001) and used in earlier studies by Stowasser et al. (2007) and Wu et al. (2012). With 6-h model outputs, the main variables used for TC tracking (the detection criteria) are vorticity at 850 hPa, 10-m maximum wind speed, and a mid-tropospheric temperature anomaly calculated over the center of the storm. Several slight modifications of the criteria are made to identify a TC in the WRF simulations:

  1. 1.

    The relative vorticity at 850 hPa must be higher than 5 × 10−5 s−1.

  2. 2.

    There must be a closed pressure minimum within a radius of 250 km, which is taken as the TC center. The minimum surface pressure must be at least 2 hPa lower than the average surface pressure over the surrounding 5 × 5 grid boxes.

  3. 3.

    The maximum 10-m wind speed within 250 km from the center must be greater than 17 m s−1.

  4. 4.

    The mean wind speed at 850 hPa around the center must be higher than that at 300 hPa.

  5. 5.

    The total tropospheric temperature anomaly, calculated by summing temperature anomalies at 700, 500 and 300 hPa around the center, must be greater than zero.

  6. 6.

    The storm must form at latitudes south of 30°N.

TC tracking is performed on those grid points satisfying the above detection criteria at each 6-h sample. A TC must last at least 2 days, and each TC snapshot is checked to see whether there are cyclones during the following 6-h period within a distance of 250 km. If any such cyclones are detected and satisfy the previous criteria, the closest cyclone is chosen as belonging to the same track as the initial one.

3 Results

3.1 Simulated TC structure

The strongest TCs in each WRF run at the peak time of the life cycle have been picked out to verify the general ability of the WRF to simulate the TC structure. The maximal wind speed in these selected case varies from 42 to 47 m s−1, which is generally moderate intensity TCs. Figure 2 shows the vertical distribution of azimuthally averaged tangential wind and temperature anomalies (the mean temperature is calculated at a radius of 1000 km) over the radial distance from TC center in the control simulations (Fig. 2a, b) and SN experiments (Fig. 2c–h). All experiments reproduce the well-defined tangential wind and warm core thermodynamic structure of moderate intensity TCs. The maximum of tangential wind appears at the 850 hPa level, about 100–200 km from the TC center, and weakens with height and radial distance in all simulations. The center of the eyewall represented by the high tangential wind speed zone is well realistic. However, the eyewall locates slightly farther from the center in the simulations, and has less vertical development compare to its wide radial range, which may be attributed mainly to the coarse model resolution. There are well defined warm-core structure of TCs located at middle and upper tropospheric layers and inside the maximal tangential wind zone in all TC cases. The positive temperature anomalies extend from the top of tropospheric down to the surface and strong warming located in the center of TCs.

Fig. 2
figure 2

Simulated TC structure from eight WRF simulations, showing tangential wind (contours) and temperature anomaly (color shading)

Full size image

3.2 TC climatology

To assess the capability of the WRF in simulating TC activity, the TC climatology of the WRF experiments is compared with observed TC best tracks from RSMC. First, we examine the simulated TC intensity to evaluate whether the WRF model can reliably reproduce TC strength. Figure 3 shows the relationship between the maximum surface wind speed and minimum central surface pressure, i.e., the TC wind–pressure relationship. All simulations have the similar wind-pressure relation with observation and generate considerable intensity for the minimum central surface pressure. But the simulated TCs only reach severe typhoon scale (41.5–50.9 m s−1) in terms of maximum wind speed, which is similar to the results in previous studies (Jin et al. 2015; Kim et al. 2014; Knutson et al. 2007). In general, the WRF model run with a 50 km horizontal resolution generates insufficient strong TCs with maximal surface winds of 45 m s−1 or above (Jin et al. 2015). Finer horizontal resolutions (<10 km) or more reasonable surface drag of model may be needed to simulate strong TCs (Jin et al. 2014; Suzuki-Parker 2012). The maximum surface wind in the simulated TCs reaches 47 m s−1 and the corresponding central pressure drops to 915 hPa in the RRTM experiment, which agrees well with observations and is consistent with the results of Wu et al. (2014). The wind-pressure relation gradually deviates from observation when the maximal wind speed up to typhoon scale (>33 m s−1) implied by red dash line in Fig. 3. The distorted wind-pressure relation in high wind condition have been discussed by Knutson et al. (2007) and Jin et al. (2015), which can mainly attribute to the deficiencies in surface flux parameterizations at high wind conditions.

Fig. 3
figure 3

Maximum surface wind speed versus minimum central surface pressure for observed (black) and simulated WNP TCs (1989–2008). The solid lines are least squares quadratic best-fit lines and the red dashed line divides the TCs into tropical storms and typhoons, according to the maximum surface wind speed

Full size image

The spatial distribution of TC tracks from WRF simulations is shown in Fig. 4b–i, while the observed tracks from RSMC is represented in Fig. 4a. Figure 5 presents the annual mean number of TC occurrences within 5° × 5° grid boxes from the observations and eight WRF experiments. All the simulated TC tracks show realistic features, such as starting from the western Pacific tropical region and moving westwards or northwest. The SN-UV runs (Fig. 4d, g) and the NOSN runs (Fig. 4b, c) simulate comparable typhoon numbers, which implied by the numbers of red line track. On the contrary, the SN-UVT runs, with additional temperature nudged (Fig. 4e, f, h, i), have too little typhoon compare to observation or other runs. The TC occurrence, also called as the TC track density, is calculated by counting the number of TCs within a 5° × 5° box at 6 h intervals in this study. Figure 5 clearly shows that the spatial patterns of TC occurrence frequency are reasonably well simulated by the WRF experiments, with maxima concentrations at similar latitudes and from the South China Sea (SCS) to 150°E as observed (Fig. 5a). However, there are positive biases of TC occurrence frequency in the SN-UV and SN-UVT-12h runs (Fig. 5d, g, f, i) compared to the observation, especially in the SN-UVT-12h runs with the overestimation of the center frequency about 1.4 times. The SN-UVT runs (Fig. 5e, h) can produce comparable occurrence frequency. The track density in the SN-UVT and SN-UVT-12h runs have less northward/northeastward extension, which implies that the track type of northwards turning or recurving is less than the westward track. In addition, the differences between the SN-UVT runs with two radiation schemes have been reduced significantly, which suggests that the SN can reduce the internal variation of regional climate model (Wu et al. 2012). The RRTM experiment (Fig. 5c) underestimates the occurrence frequency significantly, while the location of maximal frequency in the CAM experiment (Fig. 5b) is about 5° west of the observations.

Fig. 4
figure 4

Observational and model-simulated tracks for all storms in the WNP over the period 1989–2008. Storm intensity is indicated by line color (blue tropical storm; red typhoon)

Full size image
Fig. 5
figure 5

TC occurrence frequencies in each 5° × 5° grid box averaged over 1989–2008 for a observation and b–i WRF simulations

Full size image

Figure 6 shows the spatial distributions of TC genesis frequency during 1989–2008 from the WRF simulations (Fig. 6b–i) and the RSMC best-track data (Fig. 6a). In the RSMC best track data, two main genesis centers are located in the north of SCS and east of the Philippines, extending eastward. All WRF experiments reproduce the maximum band of TC formation between 5 and 20°N. The NOSN and SN-UV runs (Fig. 6b–d, g) well simulate the genesis distribution in the WNP but have southward location bias in SCS, while the CAM experiment (Fig. 6b) simulates too much TC genesis, almost doubled the genesis frequency, at 135°E. The SN-UV runs (Fig. 6d, g) reproduce the genesis frequency in the SCS, especially in the RRTM-SN-UV experiment with a well-defined center located in the middle of SCS though somewhat south. All SN runs (Fig. 6d–i) reproduce the genesis center of WNP, though located farther east as previous studies (Wu et al. 2012; Wu et al. 2014). The TC genesis frequency over WNP in the SN-UVT runs is as well as the realistic, while the SN-UVT-12h runs significantly overestimate the frequency. The center of TC genesis in SCS is poorly simulated in the SN-UVT and SN-UVT-12h runs. For the TC genesis frequency, CAM-SN-UVT experiment has less genesis frequency than the CAM-SN experiment, while RRTM-SN-UVT experiment shows higher central frequency than the RRTM-SN experiment.

Fig. 6
figure 6

As for Fig. 5 but for TC genesis

Full size image

3.3 Inter-annual variability and seasonal cycle of downscaled TC activity

In this section, we examine the inter-annual variability and seasonal cycle of TC activity over the WNP, which are dynamically downscaled by the WRF model with different configurations. As intense TCs have the greatest socio-economic impact, the analysis considers the number of typhoons in addition to the total number of TCs. Figure 7 shows the inter-annual variability in TC and typhoon numbers from observations and WRF experiments during the period 1989–2008. Table 2 lists the linear trend (Trend) and correlation coefficient (Corr) of TC and typhoon numbers between WRF simulations and the observation. In all eight WRF simulations, the inter-annual variation in TC numbers is comparable to the observations (Fig. 7a), with correlation coefficients all reaching ~0.56 (significance at the 0.01 level). The contrast between the inactive (1998) and active (1994, 2004) TC years are reproduced well in all the runs. The SN-UVT-12h and SN-UV runs obviously overestimate the annual TC numbers through the whole simulation period. For example, the annual mean TC number in the CAM-SN-UV experiment is 35, but there are only 25 TCs in the observation (Table 2). Although with the overestimation of the TC numbers, WRF experiments with SN show higher temporal correlation with the observations comparing to the NOSN runs, especially in the CAM-SN-UVT-12h experiment which has the highest correlation at 0.79. Among all the SN runs, the RRTM-SN-UVT has the least improvement on the simulation of TC inter-annual variation. By comparing the SN-UVT to the SN-UVT-12h, strong nudging coefficient (3.0 × 10−4) can reduce the simulated TC numbers, which was discussed by Knutson et al. (2007). The SN-UVT runs have less TC numbers comparing to the SN-UV runs, which may suggest that the temperature nudging suppresses the genesis of simulated TCs.

Fig. 7
figure 7

Inter-annual variability in downscaled a TC and b typhoon numbers from observation and eight WRF simulations for the period 1989–2008

Full size image
Table 2 Linear decadal trend (Trend) and correlation coefficient (Corr) of TC and typhoon numbers between WRF simulations and observations from 1989 to 2008
Full size table

The RSMC data show a clear decreasing trend in TC number (–3.5/decade) from 1989 to 2008 (Table 2). All WRF runs reproduce realistic decreasing trend, while the SN runs have an obvious overestimation on the amplitude, especially in the SN-UVT and SN-UVT-12h runs that doubled the decreasing slope. The overestimation of decreasing trend is mainly caused by the excessive genesis before the year 1998, especially in the runs with additional temperature nudging. The shift of simulated TC numbers relative to the observation in the late twentieth century may relate to the ERA-Interim driving field. There are almost an order of increased assimilated observations throughout the entire simulated period, especially some new satellite data that may contribute to improve the tropical atmospheric reanalysis at the beginning of the twenty-first century (Dee et al. 2011).

Typhoon is a mature TC with maximum sustained surface winds larger than 33 m s−1 in the WNP basin, with three different classifications according to its intensity: typhoon, severe typhoon, and super typhoon. Figure 7b shows the inter-annual variability and linear trend of typhoon numbers from observations and WRF simulations. All the WRF runs have the ability to simulate the typhoon inter-annual variation, with the temporal correlation above 0.3, except for the RRTM-SN-UVT experiment. Almost all WRF runs overestimate the 20-year averaged annual mean typhoon numbers, except the SN-UVT runs (Table 2). While the SN-UV runs generate much more typhoons comparing to the NOSN runs, the SN-UVT-12h runs have roughly equivalent typhoon numbers to the NOSN runs. The WRF runs with nudged horizontal wind and temperature (SN-UVT) underestimate the typhoon activity, with averaged annual typhoon numbers of only 11, and the underestimation is apparent after the year of 2000. The underestimation of typhoon in the SN-UVT runs suggests that temperature nudging can suppress the intensification of TCs. The observational typhoon numbers have a slight decreasing trend in the simulation period (−1.7/decade), and this decreasing trend is captured by almost all WRF simulations, except for the RRTM runs that have a slightly increasing trend (0.4/decade). Similar to the TCs trend, the WRF runs with SN overestimate the decreasing slope except that the CAM-SN-UVT experiment has an equally trend (−1.9/decade) with the observation. In general, the WRF experiments using the CAM radiation scheme can broadly capture the inter-annual variability and trend in typhoon numbers.

Figure 8 shows the seasonal cycle of TC and typhoon numbers over WNP for the period 1989–2008 from the simulations and the observation. The realistic seasonal cycles of TC numbers are well reproduced in all WRF simulations (Fig. 8a). The TC count increases from January to the most active season (August or September) and then declines until December. The CAM-SN-UV, RRTM-SN-UV and CAM-SN-UVT-12 h experiments overestimate the TC numbers from May to November, while the RRTM experiment underestimates it from May to August. Almost all WRF runs overestimate the activity of TCs from September to November. The observational typhoon numbers have similar seasonal variation with TCs but the peak month appears in September (Fig. 8b). The simulated seasonal cycles of typhoon numbers are broadly consistent with the observations, but all runs overestimate the typhoon count in the typhoon season except the SN-UVT runs. For instance, the CAM-SN experiment yields ~1.5 times the observed typhoon numbers from June to November. Comparing the seasonal cycle of TC numbers to that of typhoons, the overestimation of TC numbers is mainly due to the overestimation of typhoon numbers in the typhoon season.

Fig. 8
figure 8

Climatological mean seasonal variability in monthly TC (a) and typhoon (b) numbers from the eight simulations and observation averaged over the period 1989–2008

Full size image

3.4 Large-scale environment

The formation and development of TCs are strongly associated with environmental conditions, such as the enhanced mid-tropospheric relative humidity and conditional instability that support deep convection as well as the strengthened low level cyclonic vorticity. Convergence and small vertical wind shear are the dynamical requirements for tropical cyclogenesis (Emanuel 2003; Wang et al. 2007). To further investigate the possible causes for the errors arose in the downscaled TC activity, the large-scale environmental conditions that affect TC activity are analyzed, including zonal winds and relative vorticity at 850 hPa, mid-tropospheric relative humidity at 600 hPa and vertical wind shear between 200 and 850 hPa for the typhoon season (June to November).

Because of providing a vorticity-rich low level environment, the monsoon trough has been traditionally considered as a favorable condition to support the genesis and development of TCs. The convergence of lower tropospheric westerly monsoon wind and easterly trade can significantly affect the intensity and location of monsoon trough and then change the distribution and frequency of TCs genesis (Jin et al. 2015; Wu et al. 2014). Fields of 850 hPa zonal wind (contour) and relative vorticity (shading) (Fig. 9) show that the WRF model can simulate realistic features of low-level circulation as in the ERAIN reanalysis. The intertropical convergence zones (ITCZ) at 850 hPa simulated by all the WRF SN runs agree with the observation, with westerly monsoon wind extending from SCS to WNP. However, the convergence zone in the NOSN runs (Fig. 9b, c, contour) only extend to 140°E, which can be attributed to the stronger east wind located in the subtropical area. Meanwhile, WRF SN experiments (Fig. 9d–i, contour) can well simulate the spatial pattern of convergence zone, but overestimate the westerly wind intensity in the east of 150°E. The enhanced west wind to the east of 150°E may extend convergence zone in the monsoon trough to the east, which eventually led to the excessive TCs genesis in SN runs. The SN-UVT and SN-UVT-12h runs produce weak westerly wind in the SCS, which may result in the underestimation of TCs genesis in SCS (Jin et al. 2015).

Fig. 9
figure 9

Climatological mean (1989–2008) zonal wind (contours) and relative vorticity (shading) at 850 hPa for the typhoon season (June–November) from ERAIN (a) and WRF simulations (b–i)

Full size image

The relative vorticity at 850 hPa characterizes the monsoon trough, the dynamical parameter for TC genesis. A long strips of positive vorticity are located in the tropics, extending from SCS to WNP basin, corresponding to the monsoon trough in the observation (Fig. 9a, shading). The magnitude of positive relative vorticity over the tropical region and SCS extends less east in the NOSN runs (Fig. 9b, c, shading) corresponding to the less extension of lower tropospheric westerly monsoon wind, especially in the RRTM experiment (Fig. 9c, shading). However, the NOSN runs simulate strong vorticity in the east of Philippines, which contributes to the overestimation of the genesis center at the corresponding location. All SN runs show a comparable spatial distribution of positive relative vorticity over tropical regions, although the location of positive vorticity zone shifted slightly south of the observations. However, WRF experiments with SN simulate an unrealistic maximum centered east of 150°E, which results in excessive TCs genesis in this region. In addition, all simulations show a negative bias of relative vorticity in the SCS, particularly in the SN-UVT and SN-UVT-12h runs, which is adverse to the formation of TCs. The SN-UV and SN-UVT-12h runs also underestimate the vorticity of the east of Philippines. The well corresponding relation between TCs genesis location and monsoon trough suggests that the WRF simulated locations of TCs genesis have strong dependence on the simulation of intertropical convergence zone or the monsoon trough represented by the low-level vorticity and convergence.

Fig. 10
figure 10

As for Fig. 9, but for relative humidity at 600 hPa

Full size image

The overestimation of TCs genesis in the east of WNP can be also associated with a high moisture in mid-troposphere. Figure 10 shows the climatology of relative humidity at 600 hPa during the TC season. The wet tongue, which carries the warm-moist by the westerly monsoon wind, extends from SCS to WNP in tropic in the observation. All WRF runs reproduce the wet tongue, but overestimate the relative humidity over the tropical region east of 150°E, which can contribute to the excessive genesis of TCs. The SN-UV and SN-UVT-12h runs underestimate the humidity over the east of Philippines, while the SN-UVT runs overestimate it and produce an unrealistic maximum of humidity close to the equator in 135°E. In the SN-UV and SN-UVT-12h runs, the reduced frequency of TC genesis in the east of Philippines may attribute to the underestimation of humidity.

The strong vertical shear of horizontal wind have adverse effect on the formation of TCs, and small vertical shear is required for TC development (Emanuel 2003; Gray 1968). Figure 11 shows the vertical wind shear between 200 and 850 hPa. The weak vertical wind shear is located in the monsoon trough zone in both the observations (Fig. 11a) and simulations (Fig. 11b–i), which is a favor of the genesis of TCs.

Fig. 11
figure 11

As for Fig. 9, but for vertical wind shear and shear vector between 200 and 850 hPa

Full size image

Previous studies have shown that the SN can improve the pattern correlations of large-scale circulation between simulations and observations (Cha et al. 2011; Feser and Barcikowska 2012), and therefore decrease the systematic errors of the model. Figure 12 shows the inter-annual variation of the pattern correlations of geopotential height at 500 hPa, relative humidity, and zonal and meridional winds at 850 hPa between the WRF simulations and the ERAIN reanalysis. The correlations of zonal and meridional winds (Fig. 12c, d) in the SN runs are apparently higher than those in the NOSN runs, and the SN runs also improve the simulation of inter-annual variability in geopotential height at 500 hPa. The low-level humidity, which is more strongly influenced by surface processes and subject to the model physics, shows a weaker pattern correlation with the observations and less relation with nudging and radiation scheme. The better pattern correlation in the SN runs is expected to improve the simulated inter-annual variation of TCs activity.

Fig. 12
figure 12

Inter-annual variation in the pattern correlations between simulations and ERAIN reanalysis of a geopotential height at 500 hPa, b 850 hPa relative humidity, and c zonal and d meridional winds at 850 hPa

Full size image

In general, the underestimation of TCs in the SCS appears to be associated with the weaker low-level convergence and relative vorticity in the SN-UVT and SN-UVT-12h runs, whereas the eastward extensional monsoon trough, combined with the overestimation of moisture and vorticity, may contribute to the over-prediction of TCs in the tropical region east of 150°E in the SN runs.

4 Discussion

Different from the former studies, the SN runs produce excessive TCs and typhoons activity than NOSN in this study. Knutson et al. (2007) suggests that model without nudging generates excessive TCs attribute to the too unstable vertical thermodynamic profile and wind only nudging has less effect on the inhibition of unstable thermodynamic state. To investigate the excessive TCs development in our simulations, the analysis of model thermodynamic over the tropical sub-region (0–25°N, 100–180°E) is perform, which is shown in Fig. 13. The NOSN runs and SN-UV runs have large deviations of temperature at the low troposphere levels from the ERAIN driving field. The thermodynamic unstable in the SN-UV runs is more obvious than NOSN runs, which is contributed to the excessive TCs formation and intensification. The SN-UVT and SN-UVT-12h runs decrease the thermodynamic deviation significantly, while the SN-UVT-12h still has slightly warming in the lower troposphere. The reduction of thermodynamic unstable in SN-UVT experiments may suppress the genesis and intensity of TCs.

Fig. 13
figure 13

Mean tropical temperature profile difference between the eight model runs and ERAIN averaged over 20 simulated years

Full size image

The corrected thermodynamic state in the SN-UVT runs decreases the total TCs numbers compare to that in the SN-UV or SN-UVT-12h, but has more remarkable effect in suppressing the TCs intensification. Figure 14 shows the normalized maximum intensity distribution of simulations and observations. The observational maximal wind speed has a uniform distribution, but the model runs concentrate the distribution and overestimate the frequency of TCs with moderate intensity significantly. This distribution of maximum wind speed of TCs is similar to the result of others (Bender et al. 2010; Jin et al. 2015; Wu et al. 2014). The unreasonable concentration on the more weak intensity in the simulated TCs is mainly due to the coarse model resolution and the deficiencies in the surface momentum flux parameterization, which have been discussed by Jin et al. (2015). The SN-UVT and SN-UVT-12h runs lower the peak intensity of TCs about 5 m s−1 and reduce the frequency of TCs above typhoon intensity (>33 m s−1) comparing to the NOSN and SN-UV runs. This suggests that nudging with temperature weakens the intensity of the simulated TCs significantly.

Fig. 14
figure 14

Normalized maximum intensity histogram from eight model runs and RSMC best track

Full size image

In addition to the unstable thermodynamic state, the high average of TCs numbers in our nudging runs may also relate with the convection parameterization and the nudging only performed above the boundary layer in our simulations, which have positive contribution on the formation and intensification of TCs. Knutson et al. (2007) shows that NOSN experiment with a convection parameterization produces more TCs comparing to without convection parameterization.

In both the NOSN and SN-UV runs, the WRF experiments with CAM scheme produce a too unstable atmosphere due to the warming in the lower level and cooling in the upper level comparing to those with the RRTM radiation scheme. The unstability of vertical thermodynamic is expected to increase overall TC genesis and intensity. The difference in the two radiation schemes is significantly reduced by nudging the wind and temperature simultaneously.

5 Summary and conclusions

In this study, we investigated the sensitivity of long-term simulations of TC activity to the SN method and radiation schemes in the WNP basin over the CORDEX-EA domain. Eight 20-year (1989–2008) WRF experiments, forced by initial and lateral boundary conditions from the ERAIN, were conducted with 50 km horizontal resolution. The WRF model performed reasonably well in simulating the climatology of key TC features such as the location/frequency of genesis and tracks, and generally reproduces the inter-annual variations and seasonal cycle of TC counts. Compared to the NOSN runs, the SN runs have some advantages in simulating the inter-annual variability of TCs, but have side effect on the intensity of TCs. In addition, the spatial distributions of local genesis maxima of TCs in the SN runs have an eastward shift in the WNP, and the local maximum of TC genesis in the SCS is not well reproduced. The large-scale environment associated with TC genesis also exhibits an eastward shift, mainly due to the extended low-level intertropical convergence zone, positive relative vorticity and the overestimation of relative humidity over the tropical region east of 150°E. The underestimation of TC genesis in the SCS is mainly caused by the weaker westerly and relative vorticity in the lower troposphere.

Although performing better in simulating inter-annual variability than the NOSN runs, the SN runs produce stronger TC activity, unlike in previous studies. A detailed analysis of the thermodynamic state related to TC activity suggests that this excessive TC activity is linked to the SN parameters. The nudging with wind only cannot correct the unstable thermodynamic profile appearing in the NOSN runs. Nudging both horizontal winds and temperature improves the low-level thermodynamic deviation significantly and suppresses the TCs formation and intensification. Smaller nudging coefficients significantly increase the TCs activity even with both wind and temperature nudged. This suggests that the parameters (including nudging variables and coefficients) are critical to long-term simulation of TC climatology when applying the SN method to nudge the large-scale driving field. As noted by Feser and Barcikowska (2012), spectral nudging is still worth advocating for long-term modeling of TCs, but the SN parameters should be considered carefully to match the specific aims of research. The runs with the CAM radiation scheme produce more TC activity compared with the runs with the RRTM scheme. The vertical thermodynamic distribution in the tropical region reveals a more unstable atmospheric stratification in the CAM runs, which is expected to increase TC genesis and activity.