1 Introduction

The Himalayas are a natural barrier to water vapor transport from South Asia into the Tibetan Plateau (TP). Water vapor transport in this region directly affects the spatial and temporal distribution of precipitation in the TP and even in East Asia (Yanai and Wu 2006; Xu et al. 2008; Feng and Zhou 2012; Zhang et al. 2012). The water vapor entering the TP from the southern boundary is mainly controlled by the low-pressure troughs and the meso- and micro-scale low-pressure systems (Li et al. 2017; Wang et al. 2011), which can bring more water vapor to the TP than the dominant South Asian High during the summer monsoon season (Sugimoto et al. 2008). Meanwhile, water vapor transported by these weather systems is the main source of extreme precipitation events which have direct impacts on natural disasters, human economic life and ecological environment (Bohlinger et al. 2017; Dong et al. 2018). The formation of meso- and micro-scale weather systems are closely related to complex terrain, which can increase wind shear, directly affect local convective activities (Lindzen et al. 1983; Fujinami and Yasunari 2001; Fujinami et al. 2005), and provide favorable conditions for generating precipitation.

The Central Himalayas (CHM) is characterized by steep and complex terrain with ridges and valleys crisscrossed. Multi-scale land-air interaction due to complex terrain is the main cause of convection during the monsoon season (Houze et al. 2007). However, CHM is one of the most difficult areas for satellite remote sensing and meteorological simulations. The widely used precipitation satellites (i.e., TRMM and GPM) have very low detection ability for precipitation events in high altitude areas (Xu et al. 2017; Wang and Lu 2016). Meanwhile, both global and regional climate models show systematic deviations in the simulated precipitation and precipitable water vapor in the complex terrain region (Mueller and Seneviratne 2014; Su et al. 2013; Wang and Zeng 2012; Gao et al. 2015; Wang et al. 2017). Mueller and Seneviratne (2014) founded that the simulated precipitation by Coupled Model Inter-comparison Project Phase 5 (CMIP5) shows significant deviation in Himalayas and its surrounding areas, with wet biases in the high altitude areas of the TP and dry biases in low altitude area of CHM (Nepal and India, etc.). Even a regional climate model, such as the Weather Research and Forecasting (WRF) model using a high spatial resolution, can still suffer from significant wet bias over north Himalayas (Gao et al. 2015; Lin et al. 2018).

One of the key factors contributing to the simulation biases is the inappropriate representation of complex terrain for moisture transport and precipitation process in the Himalayas and its surroundings (Ma et al. 2015; Zhou et al. 2017; Lin et al. 2018). Higuchi et al. (1982) pointed out that the precipitation at the Himalayan peaks and ridges is 4–5 times as much as that in the valleys, and a coarse horizontal resolution will lead to large errors in precipitation simulation. Regional climate models at a high-spatial-resolution are particularly useful to depict the terrain more realistically, and thus describe the impacts of orography on the wind speed and pathway of moisture more accurately (Ménégoz et al. 2013). Further, it has the potential to reduce the errors of moisture transport into the TP, and improves the simulation of water cycle over the TP (Ma et al. 2015; Walther et al. 2013) though the computing efficiency will be debased.

A sub-grid topography parameterization scheme can also be used to improve the description of orography effects (Xue et al. 2011; Choi and Hong 2015; Sandu et al. 2016; Zhong et al. 2012) without modifying the model horizontal resolution and save computing resources. Beljaars et al. (2004) designed a Turbulent Orographic Form Drag (TOFD) parameterization scheme for sub-grid terrain drag, which has been implemented into the WRF model (Zhou et al. 2018). This scheme is found to reduce the excessive moisture transport from South Asia to the TP, and improve the simulation ability of wind field and summer precipitation over the TP. Further, Zhou et al. (2019) found that the TOFD scheme also improves the simulation of diurnal cycle of winter wind field and western TP snowfall.

These studies show that using high horizontal resolution or sub-grid orography drag parameterization schemes can improve the representation of complex terrain effect in regional climate models. However, the terrain drag parameterization scheme was originally designed for large-scale models, and was mainly applied to the climate models of large scale (> 10 km). Due to the complexity of CHM topography, the topographic variability is very drastic, even at a small scale (~ 1 km). Thus, it is not clear whether TOFD is needed or not for precipitation simulation at a high horizontal resolution (several kilometers).

Based on the latest precipitation observation data from a steep valley in CHM, this study attempts to use WRF model at a high horizontal resolution of several kilometers (0.03°) and TOFD scheme to examine their impacts on the simulations of precipitation over the CHM, and then analyze the errors in the high and low altitudes of the CHM. Section 2 introduces the data, methods and model configurations used in this study. Section 3 compares the simulated precipitation with observation. Section 4 discusses the simulation effects of multi-scale terrain on moisture transport and wind. Section 5 summarizes the main findings of this study.

2 Data and method

2.1 In-situ observations

In this study, the precipitation over the Yadong Valley was simulated by WRF and validated against in situ observations. This valley is a typical south-north valley of CHM and its longitude is about 89–90° E. Its main body is located above 2800 m above sea level (a.s.l). The elevation difference between its north and south is larger than 2 km. This valley is one of the channels through which moisture enters the TP. The warm and moist air flow from the Indian Ocean brings abundant rainfall to this region and facilitates the growth of vegetation and agriculture in the Himalayas and its surroundings.

To study the influence of complex terrain on water vapor transport and precipitation, 14 rain-gauges (named RG01, RG02 … RG14) were set up along the Yadong Valley from the south (about 2800 m a.s.l) to the north (about 4500 m a.s.l) side of the CHM (Table 1). Generally, RG01-09 is located on the south slope of CHM, and RG10-14 is located on the north side (Table 1). The sampling interval of precipitation observation is 1 h. The observation period covers the whole summer in 2018. In this study, a typical monsoon period of 2018.07.21–08.10, when all the stations had strong precipitation events, was used to validate the simulated precipitation in three WRF experiments and explore the multi-scale orography impacts.

Table 1 The geographic locations, elevations and precipitation amounts of 14 rain-gauge stations in Yadong Valley during 2018.07.21–08.10
Full size table

2.2 ASTER elevation data

The ASTER elevation data is used to show the complexity of the terrain in the central Himalaya. The ASTER GDEM (The Advanced Space borne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM) is one of the most commonly used high-precision terrain data at a spatial resolution of 1 arc second. The data download website is http://www.gisat.cz/content/en/products/digital-elevation-model/aster-gdem. Using this data, the terrain variability (i.e. elevation standard deviation) was calculated in the grids of 0.09° and 0.03°, according to the following formula:

$$\sigma_{z} = \sqrt {\frac{1}{M}\mathop \sum \limits_{i = 1}^{M} \left( {Z_{i} - \bar{Z}} \right)^{2} } ,$$
(1)

where \(\sigma_{z}\) is the elevation standard deviation, M is the total ASTER DEM point number in a grid of 0.09° and 0.03°, Zi is the terrain height (units: m) at the ith point in the grid, Z (units: m) is the average terrain height of the grid.

As shown in Fig. 1a, there is great topographic variability in 0.09° (~ 10 km) grid, especially along Himalaya ridge (around 3000 m contour line), where the standard deviation of topography is larger than 500 m. Even in the smaller grid, such as 0.03° (~ 3 km), the spatial variability of terrain is also remarkable along this region. As shown in Fig. 1b, the standard deviation of terrain is larger than 200 m in most grids and even larger than 500 m in the altitudes around 3000 contour line. Therefore, even if a climate model uses a high spatial resolution of 0.03°, the sub-grid microscale orographic impacts should not be neglected.

Fig. 1
figure 1

Standard deviation of terrain complexity (elevation standard deviation, unit: m) in the grid of 0.09° (a) and 0.03° (b), based on the ASTER GDEM data at horizontal resolution of 1″. The black line denotes that the terrain height of 3000 m a.s.l

Full size image

2.3 WRF configuration

In this study, we take the topography at the scale of 3–10 km as the mesoscale orography, and the scale less than 3 km as the microscale orography (Rontu 2006; Koo et al. 2018). In order to represent impacts of the mesoscale and microscale orography on the moisture transport and precipitation over the CHM, a control experiment and two sensitivity simulation experiments using WRF model were designed. The control case is WRF3TOFD, in which a horizontal resolution of 0.03° is used to describe mesoscale orographic effects and the TOFD parameterization scheme of Beljaars et al. (2004) was switched on to describe the sub-grid microscale orographic impacts. The TOFD scheme was developed by Beljaars et al. (2004) and firstly implemented in WRF to improve the TP climate simulation by Zhou et al. (2018). This scheme directly adds the orographic turbulent form drag on each atmospheric layer, whereas the WRF default scheme, as developed by Jimenez and Dudhia (2012), adds the orographic drag only at the first model layer. The details of the TOFD scheme can be found in Beljaars et al. (2004) and Zhou et al. (2018). One sensitivity simulation is WRF9, in which a horizontal resolution of 0.09° is used and the TOFD scheme was switched off. In this case, both meso- and micro-scale orographic effects are ignored. Another sensitivity simulation is WRF3, in which a horizontal resolution of 0.03° is used to describe mesoscale orographic effects but the TOFD scheme was switched off to ignore the microscale topographic effects in the model.

For all the three WRF experiments, the simulation period is July 19–August 10, 2018. The first 48 h are treated as the spin-up time, and the later 20 days are the analysis period. The WRF9 experiment contains the whole TP and South Asia (Fig. 2a), while the WRF3 and WRF3TOFD experiments cover the CHM and its surroundings (Fig. 2b). The domain includes several high mountains, e.g., Mt. Everst, and low valleys, e.g. Yadong Valley. The initial condition and lateral boundary driving the WRF9 experiment is from ERA-interim (https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era-interim), which is the commonly used climate reanalysis produced by ECWMF (European Centre for Medium-Range Weather Forecasts). Then, the simulation results of WRF9 are used as the initial field and lateral boundary condition of WRF3 and WRF3TOFD simulations. In all the three simulations, the New Thompson microphysics scheme (Thompson et al. 2008), short-wave radiation scheme developed by Dudhia (1989), Rapid Radiative Transfer Model (RRTM, Mlawer et al. 1997), and unified Noah land-surface model (Chen and Dudhia 2001) were selected. These setups are identical to those for generating the High Asia Refined (HAR) data (Maussion et al. 2011, 2014), with the exception that the Yonsei University scheme (YSU, Hong et al. 2006) for the planetary boundary layer parameterization is used in our study. In addition, no convective parameterization scheme is used in all the WRF experiments. The configurations of WRF experiments are also shown in Table 2.

Fig. 2
figure 2

a The terrain height (unit: m) and the domain region in the WRF9 simulation, with the red quadrangle representing the nested region of WRF3 and WRF3TOFD. b The terrain height (unit: m) and the domain region in the WRF3 and WRF3TOFD simulations, with the blue rectangular being the area where the 14 rain-gauge stations are located. c The terrain height (unit: m) in Rain Gauge (RG) area. The color-bar at the bottom is shared for all the panels and the black triangles indicate the location of the in situ stations

Full size image
Table 2 The configuration of WRF model for each numerical experiment
Full size table

2.4 Methodology

Some common evaluation metrics, such as correlation coefficient (CC), relative deviation (RB) and relative root mean square error (RRMSE), were used to validate the three WRF experiments, against the observations of the 14 rain-gauge stations. These metrics are defined as follows:

$$CC = \frac{{\sum \left( {P_{obs,i} - \overline{{P_{obs} }} } \right)\left( {P_{wrf,i} - \overline{{P_{wrf} }} } \right)}}{{\sqrt {\sum \left( {P_{obs,i} - \overline{{P_{obs} }} } \right)^{2} \sum \left( {P_{wrf,i} - \overline{{P_{wrf} }} } \right)^{2} } }},$$
(2)
$$Bias = \frac{{\mathop \sum \nolimits_{i = 1}^{N} \left( {P_{wrf,i} - P_{obs,i} } \right)}}{N},$$
(3)
$$RB = \frac{Bias}{{\overline{{P_{obs} }} }},$$
(4)
$$RMSE = \sqrt {\frac{{\mathop \sum \nolimits_{i = 1}^{N} (P_{wrf,i} - P_{obs,i} )^{2} }}{N}} ,$$
(5)
$$RRMSE = \frac{RMSE}{{\overline{{P_{obs} }} }},$$
(6)

where \(P_{obs,i}\) is the total observed precipitation at station i during the analysis period; \(P_{wrf,i}\) is the total simulated precipitation at station i in WRF simulations; \(\overline{{P_{obs} }}\) is the average of the observed precipitation at all the 14 stations; \(\overline{{P_{wrf} }}\) is the average of the simulated precipitation value at the 14 corresponding WRF grids; Bias is the difference between simulated precipitation and observed precipitation; RMSE is the Root Mean Square Error in the simulated precipitation; N is 14, which is the total number of the rain-gauge stations; i is the station number.

3 Results

First, the simulated precipitation in three WRF experiments are compared with the in situ observed precipitation over the Yadong Valley. Second, the importance of the high horizontal resolution and TOFD scheme in improving the WRF’s ability of simulating precipitation are examined. Finally, their impacts on the simulated precipitation in the high and low altitudes of the CHM are shown.

3.1 Evaluation of WRF simulated precipitation against the observations

According to the observations (Table 1; Fig. 3), the precipitation over the south slope of the CHM (RG01-09) is much larger than that over the north side. The relatively low precipitation in the north side of the CHM (RG09-12) is caused by the moisture attenuation over the CHM south slope (Wang et al. 2019). The simulations of all three WRF experiments can approximately reproduce the spatial distribution of the precipitation, but with significant biases, especially over the south slope of the CHM (Table 3; Fig. 3). WRF9 shows the largest wet bias, with an averaged RB of 385% and RRMSE of 537%. These large errors are surprising, as a ~ 10 km resolution is usually considered a high horizontal resolution for simulating the huge area of the whole Tibetan Plateau. When the model horizontal resolution increases from 0.09° to 0.03° (WRF3), the RB value of the simulated precipitation is 127%, which is more than 250% lower than that in WRF9. Meanwhile, the value of RRMSE also shows a significant decrease by about 400% in WRF3. Moreover, the spatial correlation between simulation and observations also increased greatly from WRF9 to WRF3, with the CC values from 0.39 to 0.94. The wet bias and the RRMSE value of the simulated precipitation is further reduced by about 50% and 40%, respectively, when the TOFD scheme was turned on in the high-spatial-resolution WRF experiment (WRF3TOFD), though the CC value is not further improved. Actually, the improvement of precipitation simulation at resolution of 0.09° is also seen at almost all stations when the TOFD scheme is switched on (not shown). A potential risk in the evaluation is the scale-mismatch between the observations and the coarse-resolution (e.g. 0.09°) modeling for this terrain-complex region. By assuming that the spatial representativeness of the simulated precipitation at 0.03° resolution is comparable to that of the station data, the WRF3-simulated precipitation is upscaled to 0.09° grids, and then compared with the WRF9-simulated precipitation. It is found that their differences at most of the stations are minor, and do not change our evaluation results (not shown).

Fig. 3
figure 3

Comparisons of precipitation between the WRF simulations and the in situ observation. The RG01-09 stations are located at the south slope of the CHM, while the RG10-14 stations are located at the north side of HM. The precipitation in WRF9 simulation at RG02 site is too heavy, more than 500 mm

Full size image
Table 3 The comparison of error metrics among the three WRF cases
Full size table

In addition, we also evaluated the temporal correlation between the observed and the simulated precipitation at daily scale, and found that the correlation is improved when using high resolution and the TOFD scheme in WRF (not shown). However, the correlation at daily scale is low, indicating it is still a big challenge to simulate the short-term variability of precipitation over the complex terrain.

Therefore, over the CHM, the mesoscale orography seems to play a dominant role in precipitation simulation, and the microscale orography impacts should not be ignored either.

3.2 Impacts of meso- and micro-scale orography on precipitation simulation

In summer, the warm moisture from Indian Ocean and Bay of Bengal is transported by the South Asian monsoon to the CHM, which is the main vapor source of precipitation over CHM region. The steep and complex topography results in the unique precipitation pattern over this region, where precipitation over the south is more than ten times higher than that over the north. Besides, the precipitation distribution between the peaks/ridges and valleys is also quite distinct. What’s more, the precipitation gradient with elevation in the south slope of HM is very large from low to high altitude regions (Fig. 4a).

Fig. 4
figure 4

Spatial distribution of simulated precipitation (unit: mm) averaged from July 20 to August 10, 2018. a The WRF3TOFD simulated result; b difference between the WRF3 and WRF3TOFD; c difference between the WRF9 and WRF3. The thin solid black contour line in a–c indicates the terrain height of 3000 m a.s.l. The bold solid black lines in b and c denote the major area with positive biases, and the bold dashed black lines denote the major area with negative biases

Full size image

In flat regions, the influence of sub-grid orographic drags may be neglected when model horizontal resolution is finer than 10 km. But over the CHM region, the topography spatial variability in grid of 0.03° is still in the order of 100 m or larger. When the microscale topographic drag is ignored in the WRF3, the simulated precipitation in the high altitude area of the CHM is much larger than that in the WRF3TOFD. This effect is even extended to the southern TP (or the north to HM). In contrast, the precipitation in the WRF3 is less than that in the WRF3TOFD in the low altitude area of CHM (Fig. 4b). When the mesoscale topography (grid scale < 10 km) impact is not considered, the effect is similar: the simulated precipitation in WRF9 is more (less) than that in WRF3 in the high (low) altitude area of the CHM (Fig. 4c). It indicates that the inadequate consideration of the impacts by the meso- and micro-scale topography is one of the key factors that cause the bias in the simulated precipitation over the region with complex terrain.

Either improving spatial resolution or using sub-grid TOFD scheme can reduce the wet (dry) bias in the high (low) altitude region in the WRF model. As shown in Fig. 5, the simulated precipitation in the WRF9 and WRF3 in the low altitudes (south to 26.7° N) is less than that in the WRF3TOFD. Especially for WRF9, the precipitation difference to the WRF3TOFD is larger than 200 mm. By contrast, the two simulations yield more precipitation than the WRF3TOFD, and the difference can be as large as 500 mm. Moreover, the place where the peak of precipitation occurs is also very different among the three simulations. WRF3TOFD yields two peaks of precipitation, with the main peak occurring near 26.5° N. The precipitation then decreases significantly toward the top of the mountain. However, WRF3 yields a main peak at a higher and more northern place (near 27° N), and the precipitation difference over the low altitude (south to 26.7° N) between the two cases is approximately 300 mm. One reason for the differences is the microscale terrain fluctuation, which is neglected in the WRF3, can cause a lot of precipitation (Bookhagen and Burbank 2006; Shrestha et al. 2012; Yang et al. 2018; Zhou et al. 2018; Lin et al. 2018; Wang et al. 2019) in the region of abundant moisture, i.e., the low altitude areas of CHM (south of 26.7° N). WRF9 produces a major precipitation peak at an even higher and farther northern location (near 27.2° N). One possible reason is that the WRF9 smooths the mesoscale topography over the south slope of the CHM, which suppresses precipitation in low altitudes. Thus, the moist air can be transported to higher altitudes (farther north) before water vapor condenses into precipitation, resulting in much more precipitation in the WRF9 than the other two in the high altitude (farther north) region of the CHM (27–27.5° N). Actually, WRF9 can simulate 500 mm more precipitation than the other two cases. Furthermore, the difference between WRF3 and WRF3TOFD simulated precipitation is very different in the low altitude area of the CHM but not so in the high altitude area, showing the impact of the TOFD scheme is more evident in the low altitude area of the CHM.

Fig. 5
figure 5

Distribution of simulated precipitation (unit: mm) in three WRF experiments, averaged over the 88.5–90.5°E. The terrain height is from WRF3TOFD experiment, and the WRF9 precipitation is interpolated into the grids of WRF3TOFD experiment

Full size image

3.3 Simulated precipitation frequency and intensity

The three simulations yield not only different precipitation amount but also different precipitation frequency and intensity (Fig. 6). The simulated precipitation features at the high CHM (> 2800 m) and the low CHM (< 2000 m) in the region (26–28.5° N, 88.5–90.5° E) are investigated in this section.

Fig. 6
figure 6

The precipitation amounts (a, c) and frequency (b, d) in different precipitation intensities of the three experiments over the high altitudes of CHM (a, b) and low altitudes of CHM (c, d), respectively. Note both the precipitation amount and frequency are averaged in all the grids of high or low CHM during the whole analysis period

Full size image

Figure 6b, d show that light rain (0.1–2.5 mm h−1) is most likely to happen in both the low and high CHM areas. However, the contribution by heavy precipitation (> 10 mm h−1) to the total precipitation amount cannot be ignored and is much more than that by light rain (Fig. 6a, c), especially in the low CHM. In addition, the impacts of complex terrain on precipitation frequency are distinctly different between the high and low CHM. In the low CHM, WRF3TOFD yields more frequent precipitation than WRF3, which yields more frequent precipitation than WRF9. The order is inversed for the high CHM.

In conclusion, using the high horizontal resolution to reflect mesoscale topography and the TOFD scheme to reflect the sub-grid microscale topography can effectively reduce wet biases of WRF simulated precipitation over the high CHM and dry biases over the low CHM.

4 Discussion

By improving model horizontal resolution or using sub-grid TOFD schemes, the surface friction is added in the model. Consequently, the near-surface wind speed and water vapor flux into the TP are reduced (Lin et al. 2018; Zhou et al. 2018, 2019). This section discusses how the multi-scale terrain over the CHM affects the total amount of moisture flux and the vertical structure of moisture transport.

4.1 Impacts of the multi-scale terrain on the total moisture flux

In summer, the moisture conveyed by monsoon from the Bay of Bengal and Indian Ocean is the main source of precipitation in the CHM and its surrounding region. The moisture flux of south-north direction plays a dominate role during summer (Fig. 7a). The moisture attenuates seriously over the south slope of the CHM, which not only brings large amount of precipitation to the CHM south slope, but also causes the formation of the dry-belt in the north side of CHM (Wang et al. 2019).

Fig. 7
figure 7

a Spatial distribution of moisture flux (unit: 10−2 kg m−2 s−1) in the whole layer; b similar to Fig. 5, but for the northward moisture transport (unit: 10−2 kg m−2 s−1)

Full size image

Different scales of orography can cause distinct water vapor attenuation along the south slope of CHM. WRF3TOFD considers the impacts of both the meso- and micro-scale orography, and it yields a greater attenuation rate of moisture flux at the lower part (South to 26.5° N) of the south slope than WRF3 and WRF9, while it is opposite in the upper part (Fig. 7b). The WRF3 includes the effect of meso-scale orography but not micro-scale orography, so more moisture flux can reach further north and higher altitude areas in this case. The WRF9 simulation even ignores the impact of mesoscale orography, so the moisture in this case can be transported to even further north and higher areas than that in the WRF3. This water vapor attenuation pattern is consistent with the precipitation pattern over the CHM, i.e., WRF3TOFD yields more precipitation in the lower part and less precipitation in the higher part (Fig. 5).

4.2 Multi-scale orographic impacts on the vertical structure of the moisture transport

The complex terrain not only affects the total moisture flux, but also influences the vertical structure of the moisture transport. As shown in Fig. 8a, considering both the meso- and micro-scale orographic drags, the WRF3TOFD simulates strong vertical velocity due to the uplifting effect by strong near-surface drag over the slope, which causes moist air in the upper layers (Fig. 8a).

Fig. 8
figure 8

Distribution of simulated relative humidity (color; unit: %) and wind (vector sum of zonal and vertical wind components; unit: m/s), averaged over the 88.5–90.5°E during daytime: a in the WRF3TOFD experiment, b difference between the WRF3 and WRF3TOFD experiment, c difference between the WRF9 and WRF3. The z-component wind was magnified ten times. The grey area denotes the missing value at each pressure level due to the high terrain

Full size image

Compared to the WRF3TOFD, the WRF3 simulates stronger meridional winds below 500 hPa, which is favorable for moisture air reaching high altitudes and entering the plateau. Therefore, the relative humidity (RH) in the lower atmosphere over the high CHM increases significantly, while the RH in the upper atmosphere decreases (Fig. 8b). The WRF9 excludes both the meso-and micro-scale orographic impacts, lead to an even stronger moisture transport in the lower atmosphere but weaker in the upper layers toward the TP (Fig. 8c). In a word, a drag force on the near-surface air can cause less water vapor transport in the lower layer but more in the upper layer.

4.3 Physical concept of orographic effects on water vapor transport and precipitation

The terrain over the Himalayas is very complex (Figs. 1, 9a), but it is greatly smoothed in climate/weather models, even at high horizontal resolution of about 3 km (Figs. 2b, 9b). Based on above analyses, both meso- and micro-scale orography can play important roles in the water vapor transport and the formation of precipitation. As shown in Fig. 9a, when air crosses the Himalayas, the complex orography will exert a strong drag force on the airflow, which retards the upslope winds and water vapor transport in the lower atmosphere layers. Therefore, the convergence of air including water vapor in the lower part of CHM will be enhanced and cause more precipitation. Meanwhile, the water vapor transport to the upper layer will smaller. Consequently, the moisture will almost be depleted over the low CHM, and only a small part of moisture can arrive to the high CHM (Fig. 9a). However, once the complex terrain was smoothed in a model, the drag force caused by the complex terrain will be diminished, and thereby more moisture can reach the high CHM, causing more precipitation over there (Fig. 9b) and in the interior of the TP.

Fig. 9
figure 9

Schematic diagram of how complex terrain influences the processes of precipitation and moisture transport. The black line indicates a ASTER DEM, and b terrain smoothed by model. In a and b, the dark to light blue arrow denotes the moisture transport and the red arrow denotes the orographic drag

Full size image

5 Summary

The terrain over the CHM is uniquely complex, even at a horizontal resolution of 0.03°. In this study, three WRF experiments (WRF9, WRF3 and WRF3TOFD) were conducted to explore the impacts of mesoscale (> 3 km) and microscale (< 3 km) orography on water vapor transport and precipitation over the CHM.

Compared with the unique rain gauge observations, the simulated precipitation in the WRF9, which ignored the meso- and micro-scale orographic drag effects, has the largest wet bias (RB: 385%) over the high CHM. After considering the mesoscale orographic impacts, the wet bias of simulated precipitation in the WRF3 is decreased to 127%. And when exerting both the meso- and micro-scale orographic impacts, the wet bias in WRF3TOFD simulated precipitation is further reduced to 87%. It indicates that both the meso- and micro-scale orography impacts in the climate models need to be considered in realistically simulating precipitation in this complex terrain region. The meso- and micro-scale orography drag forces can directly diminish the wind and moisture flux into the TP. Meanwhile, they can also adjust the vertical distribution of moisture transport, and further control the spatial pattern of precipitation. What’s more, either high horizontal resolution (0.03°) or sub-grid TOFD scheme can reduce the wet bias over the high CHM or dry bias over the low CHM. Conclusively, both of them play key roles for the heavy rain events (> 10 mm h−1).

The insufficient description of multi-scale orographic impacts in climate/weather models is a key factor that leads to the wet (dry) bias over the high (low) CHM in the simulated precipitation. Our work shows that both increasing the model horizontal resolution and including sub-grid orographic drag parameterization can significantly reduce the simulated precipitation biases over the regions of complex terrain. Therefore, the multi-scale orographic impacts should be taken into consideration in the climate models.