Simulation of the dipole pattern of summer precipitation over the Tibetan Plateau by CMIP6 models

The datasets used included the following: (a) monthly precipitation grid dataset of 0.5° × 0.5° horizontal resolution for the period of 1961–2014, distributed across TP regions provided by the Climate Data Center, China Meteorological Administration. (b) Monthly reanalysis data for geopotential height, wind, vertical velocity, specific humidity and evaporation on 2.5° × 2.5° horizontal resolution for 1961–2014 period, obtained from the National Center for Atmospheric Research (NCEP/NCAR) (Kalnay et al 1996), and 1° × 1° horizontal resolution in 1979–2014 period provided by the European Center for Medium-Range Weather Forecasts (ERA-Interim, Dee et al 2011). The observational and reanalysis datasets are called 'observation' in the present study. (c) The monthly historical simulation for 18 CMIP6 models from 1961–2014 is used in this study. The model outputs are interpolated into the same resolution to compare with the observation datasets. More detail information for the 18 CMIP6 models is in table 1 and can also be found at https://esgf-node.llnl.gov/search/cmip6/.

The regional precipitation is generally connected with evaporation and moisture transport. The anomalous moisture budget (Chou et al 2009, Seager et al 2010) is expressed as follows:

$$\begin{equation}P^{^{\prime}} = E^{^{\prime}} - {\partial _{\text{t}}}\langle q\rangle ^{^{\prime}} - \langle \mathbf{V} \cdot {\nabla _{\text{h}}}q\rangle^{^{\prime}} - \langle \omega {\partial _{\text{p}}}q\rangle^{^{\prime}},\end{equation} \tag{ 1 }$$

where q is the specific humidity, P is the precipitation, E is the evaporation, V is the horizontal wind vector (u, v), $\omega$ is the vertical pressure velocity. $\left\langle {} \right\rangle$ represents a vertical integration from surface to 300 hPa. The time variation of specific humidity term ${\partial _{\text{t}}}\langle q\rangle ^{^{\prime}}$ can be neglected on the seasonal mean timescale.

A variable is decomposed into its time mean (overbar) and interannual anomalies (prime) and non-interannual components, the anomalous interannual moisture transport term in (1) can be expressed as:

$$\begin{align} - \left\langle {{\mathbf{V}} \cdot {\nabla _{\text{h}}}q} \right\rangle ^{^{\prime}} & = - \left\langle {{\mathbf{V}}^{^{\prime}} \cdot \nabla \bar q} \right\rangle - \left\langle {{{\bar{\mathbf{V}}}} \cdot \nabla q^{^{\prime}} } \right\rangle \nonumber\\[4pt] & \quad- \left\langle {{\mathbf{V}}^{^{\prime}} \cdot \nabla q^{^{\prime}} } \right\rangle \; + \;{\text{residue}},\end{align} \tag{ 2 }$$

$$\begin{align} - \langle\omega {\partial _{\text{p}}}q\rangle^{^{\prime}}& = - \langle\omega ^{^{\prime}} {\partial _{\text{p}}}\bar q\rangle - \langle\bar \omega {\partial _{\text{p}}}q^{^{\prime}}\rangle \nonumber\\ & \quad - \langle\omega ^{^{\prime}} {\partial _{\text{p}}}q^{^{\prime}} \rangle + {\text{residue}},\end{align} \tag{ 3 }$$

where term $- \langle{\mathbf{V}} \cdot {\nabla _{\text{h}}}q\rangle^{^{\prime}}$ and term $- \langle\omega {\partial _{\text{p}}}q\rangle^{^{\prime}}$ represent the horizontal and vertical moisture advection. The dynamic terms in horizontal and vertical terms are $- \left\langle {{\mathbf{V}}^{^{\prime}} \cdot \nabla \bar q} \right\rangle$ and $- \langle\omega ^{^{\prime}} {\partial _{\text{p}}}\bar q\rangle$. The thermodynamic terms are $- \left\langle {{{\bar{\mathbf{V}}}} \cdot \nabla q^{^{\prime}} } \right\rangle$ and $- \langle\bar \omega {\partial _{\text{p}}}q^{^{\prime}}\rangle$, respectively. Reside is represent the nonlinear interaction part. Term $- \left\langle {{\mathbf{V}}^{^{\prime}} \cdot \nabla q^{^{\prime}} } \right\rangle$, $-\langle \omega ^{^{\prime}} {\partial _{\text{p}}}q^{^{\prime}}\rangle$ and residue are very small terms and, hence, are ignored in this study.

Figure 1(a) plots the first empirical orthogonal function (EOF) pattern of the observation. It demonstrates a dipole pattern in the south and north TP with a 21% variance being explained. The EOF1 pattern of the CMIP6 models are shown in figures 1(b) and (s). There are large differences among the CMIP6 models and the variances range between 14% and 23%. According to the pattern correlation coefficients of EOF1 between the observation and models, the models can be categorized into two groups. (a) Better group (BG) models, with correlation coefficients larger than 0.5, comprising CAMS-CSM1-0, CESM2, EC-Earth3, EC-Earth3-Veg, FIO-ESM-2-0, GISS-E2-1-G, KACE-1-0-G, MIROC6, NESM3, and SAM0-UNICON. (b) Worse group (WG) models, with pattern coefficients smaller than 0.5, involving ACCESS-CM2, ACCESS-ESM1-5, BCC-CSM2-MR, FGOALS-f3-L, GISS-E2-1-H, INM-CM5-0, MPI-ESM1-2-LR, and NorESM2-MM. Besides the pattern correlation coefficients of EOF1, we also consider the latitude of the Tanggula Mountains separating the south and north TP at 33° N–34° N. All the BG models reproduce the dipole pattern and, in each model, the center of the pattern is located over the central-eastern TP, similar to observation. However, the EOF1 pattern shows strong differences across the WG models. For example, the centers are simulated in the western TP in ACCESS-CM2 and BCC-CSM2-MR, and the separating latitude is north of the Tanggula Mountains.

Zoom In Zoom Out Reset image size

Figure 2 displays the variation of summer precipitation averaged over the south TP (27° N–33° N, 85° E–105° E) and north TP (34° N–39° N, 85° E–105° E) by each model and the ensemble means of the BG models (figure 2(a)) and WG models (figure 2(b)). Both the south and north TP precipitation manifest an interannual variation during 1961–2014. In the BG models, the precipitation over the south TP and north TP demonstrates the opposite variation with a correlation coefficient of −0.46, consistent with the EOF1 pattern. However, the correlation is only −0.11 for the WG models, much lower than the significant test. The results verify the well simulation of the dipole pattern of TP summer precipitation in the BG models.

Zoom In Zoom Out Reset image size

The interannual variation of precipitation over the TP is dominantly influenced by the moisture transport from the western boundary, which is associated with the NAO (Gao et al 2013, Liu et al 2015, Wang et al 2017). In this section, we focus on the teleconnection between the NAO and TP precipitation variation and assess the models' performance when simulating this relationship.

Figures 3(a) and (c) illustrate the regression fields of sea level pressure (SLP) anomalies against the PC1 (time series of principal component) of TP summer precipitation in observation, ensemble-mean BG models and WG models, respectively. In observation, there are high SLP anomalies in Iceland and low SLP anomalies in Azores. This north–south opposite pattern over the North Atlantic is related to the negative NAO phase (Watanabe et al 1999, Pan 2005). Meanwhile, low pressure anomalies manifest over the south TP, which favors the enhancement of local precipitation. The BG models could successfully reproduce the dipolar SLP pattern over the North Atlantic sector exactly consistent with the observation. However, in the WG models, the SLP anomalies are inverse to the observation and BG models. Although weaker low anomalies appear in the TP, they may not be associated with the NAO.

Zoom In Zoom Out Reset image size

To further demonstrate the impact of NAO on TP summer precipitation, figures 3(d) and (f) depict the regression anomalies of geopotential height and wind at 200 hPa against PC1. In observation, there is an evident Rossby wave train propagating from northwestern Europe to southern TP, which resembles that stimulated by the negative NAO phase (Liu et al 2015, Wang et al 2017). Such wave train causes a strong anticyclone over the south TP, enhancing the divergence in the upper troposphere and providing beneficial conditions for abnormal precipitation. This wave train connects the NAO and TP precipitation, suggesting the importance of western boundary downstream moisture transport. The regression anomalies of the SLP pattern and associated Rossby waves at 200 hPa are also clearly seen by using the ERA-interim reanalysis data for the observation (figure S1 (available online at stacks.iop.org/ERL/16/014047/mmedia)). In the BG models, the Rossby wave related to NAO anomalies is closely similar to that of the observation, indicating the successful simulation of the NAO–TP relationship. In contrast, the wave trains in the WG models are much weaker and are not reproduced as the NAO phase.

In general, the BG models could simulate the dipolar SLP anomalies and Rossby waves associated with NAO anomalies, which contribute to reproduce the dipole pattern of TP precipitation. However, this process is not captured in the WG models. The above analysis confirms that the simulation of the dipole pattern of summer TP precipitation is exactly related to the reproduction of the teleconnection relationship between the NAO and TP.

A realistic NAO in terms of structure and variability may be vital for reproducing the TP summer precipitation pattern. Thus, we have checked the summer NAO pattern in the BG and WG models (figure S2). The spatial extent and intensity of the summer NAO are weaker than the winter NAO, and the center is located further north to the Western Europe (Folland et al 2009). In the BG model, the NAO pattern is well reproduced, which is similar to the observation. However, the SLP anomalies exhibit a tripolar pattern over North Atlantic regions in the WG models. That means the WG models fail to capture the summer NAO pattern, and thus the unrealistic impact of NAO on TP precipitation.

It is interesting to note that the weak low surface pressure and high pressure at 200 hPa in the WG models are also favorable for precipitation over TP. However, they are not related to NAO and downstream moisture transport. It also implies that the NAO may not the only reason for the anomalous precipitation over TP. According to the moisture budget in equation (1), the regional precipitation anomalies may both influenced by the remote moisture transport and local evaporation. For the dipole pattern of TP precipitation, the horizontal moisture transport is dominantly related to the NAO in observation and BG models. However, the unrealistic anomalous local circulation and evaporation could also cause the precipitation changes, which may affect the failed simulation in the dipole pattern of TP precipitation for the WG models. Further detailed analysis will be discussed in the next section.

We diagnose the atmospheric moisture budget to demonstrate the dynamic and thermodynamic process contributions to the dipole TP summer precipitation. In observation, the dynamic component could explain large majority variance of the TP moisture transport, whereas the thermodynamic and evaporation contributions are the much smaller on interannual timescale (Wang et al 2017, Zhang et al 2017a). This implies that the variations in atmospheric circulation play a more important role in the water vapor transport of summer TP precipitation than the specific humidity changes.

Figure 4(a) shows the regression anomalies of the horizontal components of moisture convergence against PC1 averaged over the south TP (27° N–33° N, 85° E–105° E) for observation, BG ensemble-mean and WG ensemble-mean. In observation, the dipolar pattern of TP precipitation is dominated by the horizontal dynamic term ($- \left\langle {{\mathbf{V}}^{^{\prime}} \cdot \nabla \bar q} \right\rangle$), which is substantially linked to the NAO influence and western boundary moisture transport. The horizontal thermodynamic term ($- \left\langle {{{\bar{\mathbf{V}}}} \cdot \nabla q^{^{\prime}} } \right\rangle$) contributes much less to TP precipitation. Thus, we focus on the dynamic part of the moisture transport. In the BG models, the horizontal dynamic term resembles that of the observation. However, the WG models simulate much smaller values than the observation and BG models. In addition, the correlation between the ensemble-mean PC1 and horizontal dynamic term in the BG models can be reached to 0.63 at 99% significant level (figure 4(b)), while it is only 0.18 in the WG models (figure 4(c)). This suggests that the WG models could not accurately simulate the dynamic process of moisture transport related to the dipole pattern of TP precipitation, consistent with their inability to reproduce the NAO–TP relationship. Better simulation of the dynamic process indicates a realistic reproduction of atmosphere circulations affecting the water vapor transport, and thus the dipolar pattern of the TP summer precipitation.

Zoom In Zoom Out Reset image size

The vertical components of moisture convergence and evaporation anomalies are given in figure 4(d). In the observation and BG models, the vertical process and evaporation anomalies have less impact on the water vapor transport than the horizontal moisture transport. In contrast, they are simulated to be much more significant in the WG models. Though the individual models may behave different performance, the WG models simulate much larger biases of local vertical circulation and evaporation anomalies, especially in the south TP regions (figures S3 and S4). The correlation between the ensemble-mean PC1 and vertical dynamic term ($- \langle\omega ^{^{\prime}} {\partial _{\text{p}}}\bar q\rangle$) is only 0.24 in the BG models (figure 4(e)), while it is twice as large (0.48) in the WG model (figure 4(f)). Furthermore, the evaporation anomalies in the WG models also have larger deviation than those in observation and BG models, especially in the south TP regions. That means the major contribution of the TP precipitation anomalies in the WG models can be attributed to the anomalous local vertical circulation and evaporation. The large deviations in the horizontal and vertical dynamic components of moisture convergence suggest the false simulation in water vapor transport of the summer TP precipitation, which further affects the reproduction of precipitation over the south and north TP in the WG models.