1 Introduction

Cumulus parameterization is crucial to climate models (Arakawa 2004). Both the general circulation model and regional climate model results are sensitive to the convection schemes (e.g. Liang et al. 2004; Chen et al. 2010). In climate models, there are some schemes in which the convection process is mainly driven by buoyancy. The buoyancy-dependent cumulus schemes may overestimate the occurrence of convection, because in nature some unfavorable conditions of large-scale environment, i.e., the dry condition of lower troposphere over subtropical high, may inhibit the formation of deep convection (Markowski et al. 2006).

To include the effects of large-scale environment, some convection suppression criteria have been employed to improve the performance of cumulus parameterization scheme (Krishnamurti et al. 1983; Emori et al. 2001; Chow et al. 2006). The convection suppression criterion based on the relative humidity (RH), in which the convection is shut down if the mean RH between the cloud top and cloud base is lower than a threshold value, is the most common usage. This type of convection suppression criterion has been applied to Arakawa–Schubert cumulus scheme (Emori et al. 2001), Emanuel scheme (Chow et al. 2006), and Tiedtke scheme (Wang et al. 2003). These studies suggest that the inclusion of convection suppression criterion significantly improves the simulation of Asian summer monsoon precipitation.

The Grell scheme (Grell 1993) is one of the buoyancy-dependent cumulus schemes. In this scheme, the convection is activated when a lifted parcel becomes buoyant. Zou and Zhou (2011) applied the convection suppression criterion based on the relative humidity (RH) to Grell scheme in a regional ocean–atmosphere coupled model and found that the cold bias of sea surface temperature (SST) is significantly improved over WNP region. However, the impacts of Grell scheme with convection suppression criterion in stand-alone simulation forced by observed SST has not been investigated.

Regional climate model (RegCM), which is developed in Abdus Salam International Centre for Theoretical Physics (ICTP), has been widely used in dynamical downscaling and regional climate change studies (Pal et al. 2007). RegCM with Grell scheme generally shows the systematic cold bias of near-surface air temperature, i.e., over Europe (Giorgi et al. 2004), over East Asia (Zhang et al. 2007), over Caribbean Region (Martinze-Castro et al. 2006) and over South Atlantic Ocean (Reboita et al. 2010). The cold bias of simulated 2 m air temperature is one of the systematic errors of RegCM3 and may be attributed to the problems in the physical parameterization schemes, i.e., the intensified vertical heat and moisture transport in non-local boundary layer parameterization (Giorgi et al. 2004), too efficient on the heat and moisture redistribution in Grell convective scheme (Giorgi et al. 2004; Martinze-Castro et al. 2006). The effect of Grell convection may be more important in the warm season, when cumulus convection is more active. However, how to reduce the cold bias is less well known.

In this study, the sensitivity of stand-alone RegCM3 simulation, which is forced by observed SST, to the convection suppression criterion, is investigated with focus on the western North Pacific summer monsoon. Our results suggest that performance of regional climate model is significantly improved, and the cold bias of simulated 2 m air temperature is significantly reduced.

2 Model, experiments, and data description

The model used in this study is regional climate model version 3 (RegCM3), which is developed in Abdus Salam International Centre for Theoretical Physics (Pal et al. 2007). It is the advanced version of RegCM2 (Giorgi et al. 1993a, b). The mass flux scheme of Grell (1993) with Fritsch–Chappell closure assumption (Fritsch and Chappell 1980) is employed to calculate the convective precipitation. The following physics schemes are also employed: the subgrid explicit moisture scheme (SUBEX) (Pal et al. 2000), the radiation package of National Center for Atmospheric Research (NCAR) community climate model version 3 (CCM3) (Kiehl et al. 1996), the non-local planetary boundary layer (Holtslag et al. 1990), the biosphere–atmosphere transfer scheme of Dickinson et al. (1993), the ocean–atmosphere flux algorithm proposed by Zeng et al. (1998).

The model domain covers 0°N–40°N, 105°E–160°E. The horizontal resolution of RegCM3 is 45 km and there are 18 vertical levels with a model top at 50 hPa. The initial and lateral boundary condition are derived from National Center for Environmental Prediction/Department of Energy (NCEP/DOE) reanalysis 2 (R2) (Kanamitsu et al. 2002) and are updated every 6 h. The SST forcing is from the weekly optimally interpolated sea surface temperature (OISST) data (Reynolds et al. 2002).

The model is integrated starting from 25 April through 31 August of 1998. Two experiments are performed. One is the control experiment (termed as “RegCM3_CTRL” in following discussion), which no convection suppression criterion is used. Another is the convection suppressed sensitivity experiment (termed as “RegCM3_RH70” in following discussion) in which the convection is activated when the relative humidity (RH) averaged from the cloud top to cloud base is larger than a critical value. In present study, the threshold value is set to 0.70, which was considered adequate after a series of sensitivity experiments. The results covering (5°N–35°N, 110°E–155°E), which exclude the buffer zone of regional atmospheric model, are analyzed. The regional average means an average within the ocean region of (5°N–35°N, 110°E–155°E) in the following discussion.

Because the WNPSM in 1998 is much weaker than the climatology, to examine whether the modified scheme is case-dependent, we have done another case simulation of 2002, when the WNPSM is normal. All the simulation was done with single member.

The following datasets are used to validate the model results: (1) the convective and stratiform rainfall derived from Tropical Rainfall Measurement Mission (TRMM) 3A12 version 6 (Kummerow et al. 2001); (2) the daily rainfall with a resolution of 0.25° × 0.25° derived from TRMM 3B42 (Huffman et al. 2007); (3) the 2 m air temperature derived from the objectively analyzed air–sea heat fluxes (OAFlux) version 3 (Yu et al. 2008); (4) the circulation fields from NCEP/DOE R2 (Kanamitsu et al. 2002).

3 Results

3.1 Rainfall

The spatial distribution of rainfall averaged from May to August of 1998 over WNP region is shown in Fig. 1. The major rainband is located over the equatorial region and the northwestern flank of the western Pacific subtropical high in the TRMM 3B42 data (Fig. 1a). An elongated band of excessive precipitation extending northeastwardly from north of South China Sea to south of Japan is found in RegCM3_CTRL (Fig. 1b). The RegCM3_CTRL has a spatial pattern correlation coefficient (SCC) of 0.19 with TRMM 3B42 data, and the root mean square error (RMSE) is 5.69 mm/day. When compared with the RegCM3_CTRL, the too excessive rainfall is reduced in the RegCM3_RH70. The SCC (RMSE) of RegCM3_RH70 is 0.40 (3.61 mm/day) with TRMM 3B42, both are significantly improved in comparison to the RegCM3_CTRL.

Fig. 1
figure 1

Spatial distributions of rainfall (mm/day) averaged from May to August of 1998 for a TRMM 3B42, b RegCM3_CTRL, and c RegCM3_RH70

Full size image

Suppressing the occurrence of convection may reduce the simulated convective percentage, which is defined as the percent of total rainfall that is convective. The spatial distributions of convective percentages of rainfall averaged from May to August of 1998 are shown in Fig. 2. The convective percentage for most oceanic region is about 60–80 % in TRMM 3A12 data (Fig. 2a). The corresponding percentage is overestimated in the RegCM3_CTRL in which the convective percentage is larger than 80 % over most oceanic region (Fig. 2b). When convection suppression criterion is applied, the simulated convective percentage is reduced (Fig. 2c), especially over south of 20°N. The regional averaged convective percentage is 63.8, 85.2, 61.3 % in TRMM data, RegCM3_CTRL and RegCM3_RH70, respectively. The reasonable performance of RegCM3_RH70 in simulating the regional averaged convective percentages is the reason that we set the threshold value of RH to 0.70. The deficiency is that RegCM3_RH70 produces greater percentage of convective precipitation in the mid-latitudes, which is opposite to the observational pattern. This is because, first, the precipitation events occurred in the mid-latitudes is related in part to the low level circulation changes and associated water vapor flux changes, while the convection suppression criterion is solely based on the relative humidity; secondly, changes in the distribution of other variables over the mid-latitude due to low-level circulation changes are, in fact, relatively smaller than those over sub-tropical lower latitude region (Fig. 4), so the circulation change may not have played a noticeable role in changing the distribution of other variables, including water vapor flux in the mid-latitudes (north of 25°N), resulting in little reduction in convective rainfall percentage through modified scheme.

Fig. 2
figure 2

Spatial distribution of convective percentages (%, defined as the percent of total rainfall that is convective rainfall) of rainfall averaged from May to August of 1998 for a TRMM 3A12, b RegCM3_CTRL, and c RegCM3_RH70

Full size image

3.2 Monsoon circulations

The atmospheric circulations over WNP, which is considerably associated with rainfall, are also improved in the RegCM3_RH70. The spatial distributions of wind vector at 850 hPa and geopotential height with 5,880 gpm averaged from May to August of 1998 are shown in Fig. 3. When compared with the reanalysis, the low-level southwesterly wind associated with the Meiyu rainband is much stronger and shifts southward in the RegCM3_CTRL. The simulated western North Pacific subtropical high (WNPSH) also exhibits southward displacement. A weak cyclonic circulation is located over north of Meiyu rainband in the RegCM3_CTRL.

Fig. 3
figure 3

Spatial distributions of wind vector at 850 hPa and 5,880 gpm isoline at 500 hPa averaged from May to August of 1998 for a NCEP2, b RegCM3_CTRL, c RegCM3_RH70. The distributions of wind speeds at 850 hPa are also shown as shaded

Full size image

When the convection suppression criterion is applied, the simulated atmospheric circulations are closer to the reanalysis (Fig. 3c). The bias of the low-level wind in the RegCM3_CTRL is prominently reduced in terms of the intensity and spatial pattern. The low-level weak cyclone is not found in the RegCM3_RH70. The 5,880 gpm isoline is used to define the scope and north edge of WNPSH. Over east of 130°E, the north edge of WNPSH is found along 18°N in the RegCM3_CTRL, while it is found along 22°N in the RegCM3_RH70, which is close to the reanalysis.

The changes in circulation and rainfall are closely related. The differences of convective rainfall, wind vector at 850 hPa, surface wind speed at 10 m and latent heat flux averaged from May to August of 1998 between RegCM3_RH70 and RegCM3_CTRL are shown in Fig. 4. The experiment with convection suppression criterion reduces the convective rainfall over north of South China Sea and the ocean south of Japan, with a central value of larger than 12 mm/day (Fig. 4a). The suppressed convective heating simulates a low-level anomalous anticyclone to its north (Fig. 4b) due to a baroclinic Rossby wave response. This heating induced response is similar to the response to Meiyu–Baiu heating with a linear baroclinic model (Sampe and Xie 2010). The anomalous northeasterly at 850 hPa weakens the intensity of low-level southwesterly flow. The weakening of low-level southwesterly flow reduces the wind speed at 10 m (Fig. 4c), which decreases the evaporation at sea surface (Fig. 4d). The less supply of water vapor from underlying ocean in turn favors less convective rainfall. These features indicate the positive feedback over the ocean between convective precipitation, low-level southwesterly flow, and surface evaporation.

Fig. 4
figure 4

Differences of a convective rainfall (mm/day), b wind vector at 850 hPa (m/s), c surface wind speed at 10 m (m/s), and d latent heat flux (W/m2) averaged from May to August of 1998 between RegCM3_RH70 and RegCM3_CTRL

Full size image

3.3 Near surface air temperature

The spatial distribution of the bias of simulated 2 m air temperature averaged from May to August of 1998 by RegCM3_CTRL is shown in Fig. 5a. The cold bias is evident over the entire WNP region in the RegCM3_CTRL (Fig. 5a). The largest bias is located over east of the Philippines, with the central value of −3 °C. The regional averaged cold bias is −2.03 °C.

Fig. 5
figure 5

a Spatial distribution of the difference of sea surface 2 m air temperature (°C) between RegCM3_CTRL and OAFlux averaged from May to August of 1998. b The difference between RegCM3_RH70 and RegCM3_CTRL

Full size image

The inclusion of convection suppression criterion reduces the cold bias of simulated 2 m air temperature (Fig. 5b), especially over east of the Philippines. The regional averaged bias of 2 m air temperature simulated by the RegCM3_RH70 is −1.55 °C.

Why is the cold bias of simulated 2 m air temperature reduced after applying the convection suppression criterion? The RegCM3_CTRL exhibits cold bias in the lower troposphere and warm bias in the upper troposphere (Fig. 6), due to the overestimation of convection frequency (Fig. 2b) that leads to more stable atmospheric stratification as compared to the reanalysis (Fig. 6). The experiment with realistically suppressed convection improves the simulation of atmospheric stratification and reduces the cold bias (warm bias) in the lower (upper) troposphere (Fig. 6). Thus, the cold bias of simulated 2 m air temperature is also reduced in the RegCM3_RH70.

Fig. 6
figure 6

Vertical profiles of biases of regional averaged air temperature (°C) derived from RegCM3_CTRL and RegCM3_RH70 averaged from May to August of 1998. The regional average means an average within the ocean region of (5°N–35°N, 110°E–155°E)

Full size image

4 Case simulation of 2002 when the WNPSM is normal

In 1998 summer, sea surface temperature over the western North Pacific (WNP) was anomalously high when compared with the climatology, and the WNPSM in 1998 is much weaker than the climatology. To examine whether the modified scheme based on 1998 is case dependent, we have conducted another two simulations for the year 2002 with and without the convection suppression criterion. Note we chose the case of 2002 because the WNPSM is normal in this year.

The spatial distributions of the observed and simulated rainfall averaged from May to August of 2002 are compared in Fig. 7. The simulation with default scheme highly overestimates the rainfall over east of the Philippines. The SCC (RMSE) is 0.03 (4.95 mm/day) with the TRMM 3B42 data. The simulation with modified scheme improves the simulation of rainfall, with an SCC (RMSE) of 0.24 (3.84 mm/day).

Fig. 7
figure 7

Spatial distributions of rainfall (mm/day) averaged from May to August of 2002 for a TRMM 3B42, b RegCM3_CTRL, and c RegCM3_RH70

Full size image

The improvements of the simulation of rainfall in 2002 are also due to the reduction in positive feedback between convective rainfall, low-level southwesterly flow, and sea surface evaporation. Figure 8 shows the differences of convective rainfall, wind vector at 850 hPa, surface wind speed at 10 m and latent heat flux averaged from May to August of 2002 between the control run and the simulation with modified scheme. As in the case of 1998, the suppressed convective heating over east of the Philippines (Fig. 8a) simulates a low-level anomalous anticyclone to its north (Fig. 8b). The anomalous northeasterly at 850 hPa weakens the intensity of low-level southwesterly flow, and then reduces the wind speed at 10 m (Fig. 8c). The associated evaporation at sea surface is decreased (Fig. 8d). The less supply of water vapor from underlying ocean in turn favors less convective rainfall.

Fig. 8
figure 8

Differences of a convective rainfall (mm/day), b wind vector at 850 hPa (m/s), c surface wind speed at 10 m (m/s), and d latent heat flux (W/m2) averaged from May to August of 2002 between RegCM3_RH70 and RegCM3_CTRL

Full size image

The cold biases of simulated 2 m air temperature in 2002 are also reduced in the simulation with modified scheme (Fig. 9). The regional averaged cold bias is −1.37 °C in the control run, while it is −0.96 °C in the simulation with modified scheme. Figure 10 shows the vertical profile of the biases of regional averaged air temperature. As in the case of 1998, the simulation of 2002 with suppressed convection reduces the cold bias (warm bias) in the lower (upper) troposphere, and then improves the simulation of atmospheric stratification.

Fig. 9
figure 9

a Spatial distribution of the difference of sea surface 2 m air temperature (°C) between RegCM3_CTRL and OAFlux averaged from May to August of 2002. b The difference between RegCM3_RH70 and RegCM3_CTRL

Full size image
Fig. 10
figure 10

Vertical profiles of biases of regional averaged air temperature (°C) derived from RegCM3_CTRL and RegCM3_RH70 averaged from May to August of 2002. The regional average means an average within the ocean region of (5°N–35°N, 110°E–155°E)

Full size image

Above the results demonstrate that the modified scheme based on the case of 1998 is also suitable to the case of 2002, and the mechanisms that improve the simulation of WNPSM are identical between two cases. These results imply that the modified scheme will also suitable to other cases, which will be tested in the future.

5 Summary and discussions

5.1 Summary

This study demonstrates that the inclusion of convection suppression criterion based on the averaged relative humidity from cloud base to cloud top improve the simulation of western North Pacific summer monsoon by RegCM3 with Grell cumulus parameterization.

When the convection is realistically suppressed, the intensity and spatial pattern of mean rainfall is significantly improved in terms of spatial pattern correlation coefficient and root mean square error. The simulated convective percentages are comparable to those derived from the TRMM data.

The simulated circulations are also improved in the experiment with convection suppression criterion through the positive feedback between convective rainfall, low-level southwesterly flow, and sea surface evaporation. A reduction in convection rainfall favors a low-level anomalous anticyclone to its north. The associated anomalous northeasterly reduces the intensity of low-level southwesterly flow and surface wind speed at 10 m. The decrease in surface wind speed at 10 m reduces the sea surface evaporation, which in turn favors less convective rainfall.

The cold bias of 2 m air temperature, which is one of systematic errors of RegCM3, is also partly reduced over WNP through the improvement of simulated atmospheric stratification.

5.2 Discussions

For the case of 1998, the effects of this technique over WNP in stand-alone RegCM3 and corresponding regional air–sea coupled model could be compared based on the current study and the study by Zou and Zhou (2011), since in two studies, the model configuration and the experimental design are exactly the same except for the inclusion of local air–sea coupling in Zou and Zhou (2011).

In both studies, the inclusion of convection suppression criterion improves the simulation of rainfall and reduces the cold bias of simulated surface air temperature, highlighting the usefulness of this technique. In addition, no matter the local air–sea coupling is included or not, the simulated regional averaged convective percentages are quite similar and comparable to that derived from TRMM data [Fig. 9 in Zou and Zhou (2011)].

In stand-alone RegCM3 simulation and ocean–atmosphere coupled simulation, the physical processes responsible for the improvement of simulations are different. In regional air–sea coupled model, the reduction in cold bias of simulated SST is mainly due to an increase in downward solar shortwave radiation at sea surface, which is followed by a decrease in column-averaged cloud cover (Zou and Zhou 2011). However, in stand-alone RegCM3 simulation, since the underlying SST is prescribed, the changes in surface downward solar shortwave radiation cannot induce any change of SST. The improvements of rainfall and monsoon circulation are mainly due to the reduction in positive feedback between convective rainfall, low-level southwesterly flow, and sea surface evaporation.