-
The outputs from the historical simulation (1995–2014) and the SSP5-8.5 experiment (2081–2100) of 20 CMIP6 models are used. The SSP5-8.5 represents a combined scenario of a high energy-intensive, socioeconomic developmental path (i.e., SSP5) with strong radiative forcing peaking at 8.5 W m–2 by 2100 (O'Neill et al., 2016). The 20 CMIP6 models used for the AMME and the models used for the BMME and WMME are shown in Table 1. The members selected for the BMME and WMME were based on their respective performances on the observed climatology and interannual variability of precipitation during 1995−2014. Further details are provided in Yang et al. (2021).
Model name DJF JJA BMME WMME BMME WMME ACCESS-CM2 √ ACCESS-ESM1-5 √ BCC-CSM2-MR √ CanESM5 CESM2 √ CESM2-WACCM EC-Earth3 √ EC-Earth3-Veg √ FGOALS-g3 √ GFDL-CM4 √ GFDL-ESM4 INM-CM4-8 √ √ INM-CM5-0 √ IPSL-CM6A-LR MIROC6 MPI-ESM1-2-HR MPI-ESM1-2-LR √ √ MRI-ESM2-0 NorESM2-LM NorESM2-MM Table 1. 20 CMIP6 models used for all-model ensemble (AMME) and the members (marked by √ ) used for “highest-ranked” model ensemble (BMME) and “lowest-ranked” model ensemble (WMME).
The CN05.1 precipitation data (Wu and Gao, 2013) and the ERA5 reanalysis data (Hersbach et al., 2020), each having a 0.25° × 0.25° resolution, are also employed (hereafter referred to as “observation”). The variables from the EAR5 reanalysis include evaporation (E), specific humidity (q), meridional wind (v), zonal wind (u), geopotential height (Z), sea level pressure (SLP), surface pressure (Ps), and vertical velocity (ω). Because of the varying resolutions of the data, we used the bilinear interpolation method to interpolate all the data to 1° × 1° grid resolution. To avoid artificial increases in geopotential height due to the influence of global warming (He et al., 2015; Huang et al., 2015, 2016; Zhang et al., 2021), the eddy geopotential height, defined as the departure from its zonal mean, was adopted for the analysis of projected changes.
-
We attempt to diagnose the cause for the differences among the three ensembles in the simulation and projection of winter (December to February, DJF) and summer (June to August, JJA) precipitation in terms of the atmospheric circulation and moisture budget. When considering precipitation climatology, moisture variation with time
$ \left( {{\partial _t}\left\langle q \right\rangle } \right) $ is negligible (Bao and Feng, 2016; Li et al., 2017). Thus, the vertically integrated moisture equation is shown in Eq. (1) (Trenberth and Guillemot, 1995), which can be simplified by Eq. (2):where the operator
$\left\langle . \right\rangle = \frac{1}{g}\int_{{P_{\text{s}}}}^{100} {{\rm{d}}p}$ ,$ - \left\langle {{V_h} \cdot {\nabla _h}q} \right\rangle $ and$ - \left\langle {\omega {\partial _p}q} \right\rangle $ represent horizontal moisture advection and vertical moisture advection, respectively; E indicates evaporation, and Res is the residual term.Referencing Chou et al. (2009), the moisture budget for the projected change of precipitation (dP) at the end of the 21st century (2081–2100) relative to the reference period (1995–2014) is estimated as:
in which the overbar operator, ¯ , is the climatology of the reference period and the prime operator, ' , denotes the future change relative to the reference period; dTH, dDY, and dNL represent the thermodynamic contribution related to the change in specific humidity, the dynamic contribution associated with the change in atmospheric circulation, and the nonlinear term influenced by changes in both specific humidity and atmospheric circulation, respectively.
-
First, we present the biases of winter and summer precipitation as simulated by the three ensembles for 1995–2014. As shown in Fig. 1, the BMME outperforms the AMME and WMME in simulating the spatial pattern of observed winter precipitation. However, all the ensembles generally show an overestimation in northern China, especially in Northwest China (NWC, 36°–46°N, 75°–111°E). One exception is the presence of negative biases in Xinjiang for the BMME simulation (Fig. 1e), which differs from that simulated by the AMME (Fig. 1c) and WMME (Fig. 1g). For summer precipitation, the BMME simulation shows less improvement compared to the other two ensembles. Jiang et al. (2020) compared the performances of CMIP5 and CMIP6 models in capturing the climatological precipitation over China. Their study revealed that the performances have improved from the CMIP5 to the CMIP6. Lun et al. (2021) also indicated that the overestimation of precipitation is reduced from the CMIP5 to the CMIP6 in the ensemble of optimal models.
Figure 1. Spatial distributions of (a, b) CN05.1 climatology (units: mm) and (c, d) AMME, (e, f) BMME, and (g, h) WMME simulation bias (units: %) for winter (left panel) and summer (right panel) precipitation during 1995–2014. The thick black lines represent the boundaries of subregions of northern China: Northeast China (NEC, 39°–54°N, 119°–134°E), North China (NC, 36°–46°N, 111°–119°E), and Northwest China (NWC, 36°–46°N, 75°–111°E).
When regionally averaged, the BMME simulation exhibits the smallest deviation from the observation for both the winter and summer precipitation, followed by the AMME simulation and then the WMME simulation. Specifically, the percentage-based wet biases for winter (summer) precipitation over NWC, North China (NC, 36°–46°N, 111°–119°E), and Northeast China (NEC, 39°–54°N, 119°–134°E) in the BMME simulation are 462%, 185%, and 132% (51%, 25%, and 8%), which increase to 1433%, 267% and 187% (55%, 26%, and 15%) in the AMME simulation and further to 2763%, 473%, and 326% (78%, 27% and 16%) in the WMME simulation, respectively (Table 2). Note that larger percentage-based wet biases in winter compared to summer do not reflect the larger biases of the absolute values due to the different climatology of seasonal precipitation in northern China (Table 2).
Subregion Winter (DJF) Summer (JJA) Observation AMME BMME WMME Observation AMME BMME WMME NC 7 267 (18.69) 185 (12.95) 473 (33.11) 267 26 (69.42) 25 (66.75) 27 (72.09) NEC 12 187 (22.44) 132 (15.84) 326 (39.12) 335 15 (50.25) 8 (26.8) 16 (53.6) NWC 5 1433 (71.65) 462 (23.10) 2763 (138.15) 95 55 (52.25) 51 (48.45) 78 (74.10) Table 2. CN05.1 observed winter and summer precipitation (units: mm), and the AMME, BMME, and WMME simulated relative biases (units: %) and absolute biases (shown in parentheses; units: mm) over the subregions of northern China.
-
Large-scale atmospheric circulations provide an important background for the occurrence of precipitation. In general, the precipitation in NWC is influenced by the westerly circulation, while precipitation over eastern China is primarily influenced by monsoon circulations. The EAWM circulations are characterized by the Siberian high and the Aleutian low in SLP, the prevalent northerly over eastern China in the lower troposphere (Fig. 2a), the East Asian trough in the middle troposphere, and the East Asian westerly jet in the upper troposphere (Fig. 2b). The EASM circulations are characterized by the prevailing southwesterly winds over eastern China in the lower troposphere (Fig. 3a), the western Pacific subtropical high in the middle troposphere, and the East Asian westerly jet, located around 40°N, in the upper troposphere (Fig. 3b).
Figure 2. Spatial distributions of (a, b) ERA5 climatology and simulation biases of (c, d) AMME, (e, f) BMME, and (g, h) WMME for (left panel) sea level pressure (shading; units: hPa) and 850 hPa winds (vectors; units: m s–1) as well as (right panel) 500-hPa geopotential height (contours; units: gpm) and 300-hPa zonal wind (shading; units: m s–1) during winters from 1995–2014. The gray shading represents the Tibetan Plateau.
The three ensembles can reasonably reproduce the basic features of the EAWM and EASM circulations. However, compared with observations, the AMME simulated SLP is slightly lower in the mid-high latitudes and higher around the Tibetan Plateau in winter (Fig. 2c), indicating an overestimation of the south-north meridional pressure gradient. Hence, a slightly stronger westerly flow is introduced, which contributes to a wet bias in NWC. The BMME simulated SLP is also somewhat lower in the mid-high latitudes; however, the positive SLP bias around the Tibetan Plateau is reduced compared to the AMME simulation (Fig. 2e). Thus, the westerly deviation simulated by the BMME is smaller than that of the AMME, reducing the wet bias over NWC in the BMME simulation. In the middle and high troposphere, a widespread negative bias in the 500-hPa geopotential height and a southward shift of the East Asian westerly jet are produced in the AMME and BMME (Figs. 2d, f). The WMME simulation bias behaves differently from that of the AMME and BMME. For the WMME simulation (Figs. 2g, h), the simulated Aleutian low, East Asian trough, and East Asian westerly jet are consistently weaker than the observation. The biases in the EAWM circulations are significantly larger than that simulated by the AMME and BMME, making the WMME performance on winter precipitation inferior to that of the AMME and BMME.
During the summer, all the ensembles simulate an increased mid-latitude zonal pressure gradient, which allows for anomalous southwesterlies or westerlies to occur in the eastern part of northern China (Figs. 3c, e, g). The northwestern part of northern China is dominated by westerly anomalies. In addition, cyclonic and anticyclonic circulation anomalies at 850 hPa are prevalent to the south and north of 30°N in the western Pacific, respectively (Figs. 3c, e, g), indicating a relatively northward location of the western Pacific subtropical high. For the 500-hPa geopotential height, there are negative biases in the AMME and BMME simulations over the region where the observed 5880-gpm contour that characterizes the intensity of the western Pacific subtropical high is located (Figs. 3d, f). In contrast, a systematic positive bias is noted in the WMME simulation, indicating an overestimation of the western Pacific subtropical high (Fig. 3h). In the upper troposphere (Figs. 3d, f, h), the simulated zonal wind increases north of 40°N and weakens south of 40°N, suggesting a northward displacement of the East Asian westerly jet, which is conducive to the excessive precipitation in northern China (Lu, 2004; Huang et al., 2013; Ren et al., 2017; Zhou et al., 2018a). The BMME simulation for the upper-tropospheric zonal winds is generally similar to that of the AMME and WMME, which is partly responsible for the fact that the BMME simulated summer precipitation is not saliently improved.
-
Figure 4a shows the observed climatological values for the winter moisture budget terms in each of the subregions (NWC, NC, NEC) of northern China. The results indicate that all three subregions are characterized by the outward transport of moisture in winter, mainly due to the influence of the vertical moisture advection, and that the evaporation is balanced by precipitation. For the AMME simulation of winter precipitation in NWC (Fig. 4g), the wet bias mainly comes from the overestimation of the evaporation (571%) and vertical moisture advection (445%). The simulated evaporation bias in the WMME increases to 2160%, while it decreases to –162% in the BMME. In contrast, the bias of the simulated vertical moisture advection is comparable among the three ensembles. This means that different behaviors of the BMME, AMME, and WMME in simulating winter precipitation over NWC are mainly attributed to their simulated biases in evaporation.
Figure 4. Climatology of the observed (a) winter and (b) summer moisture budget (units: mm d–1) in the three subregions of northern China during 1995–2014 and (c–h) simulation biases of AMME, BMME, and WMME for (left panels) winter and (right panels) summer moisture budget (units: %, percentage anomalies relative to 1995–2014). The boxes and error bars indicate the ensemble mean and the range of ±0.5 standard deviations, respectively.
Similarly, the overestimation of the evaporation also contributes to the wet biases of winter precipitation in NC and NEC. Among the three ensembles, the BMME presents the smallest evaporation biases (29% and 29%) in both NC and NEC, which increase to 112% and 93% in the AMME simulation and are further exaggerated to 451% and 242% in the WMME simulation, respectively (Figs. 4c, e). The BMME simulated advection of vertical and horizontal moisture is improved in NEC compared with the simulations of the AMME and WMME (Fig. 4e). It is worth noting that the residual term is somewhat large in NC and NEC (Figs. 4a, c, e), which may be a consequence of the sub-monthly transient eddies and the moisture imbalance between the models and the reanalysis, leaving the moisture budget analysis in these regions subject to a certain degree of uncertainty.
Figure 4b indicates that the contribution to the observed summer precipitation in NC (NEC and NWC) is mainly from the evaporation, followed by the horizontal moisture advection (vertical moisture advection). For simulations in NWC (Fig. 4h), the AMME, BMME, and WMME show biases of 2%, 14%, and 15% for the evaporation, –27%, –57%, and –20% for the vertical moisture advection, and –19%, –25%, and –2% for the horizontal moisture advection, respectively. Compared with the AMME and WMME, the BMME does not show notable improvement in the simulation of any given terms. Thus, the slight improvement of the BMME simulation for summer precipitation over this region is considered to be the result of a mutual offset between the overestimation and underestimation.
For the simulations in NC (NEC) (Figs. 4d, f), the evaporation biases of 13%, 12%, and 13% (6%, 5%, and 2%), respectively, from the AMME, BMME, and WMME are approximately the same. The biases of the simulated horizontal moisture advection among the three ensembles are also comparable. The slight improvement from the WMME to the BMME in simulating summer precipitation is mainly reflected in the enhancement of the performance on the vertical moisture advection. Compared with the WMME bias of 34% (18%) for the vertical moisture advection in NC (NEC), the bias decreases to 11% (7%) in the AMME simulation and decreases further to –2% (1%) in the BMME simulation.
In brief, the wet winter bias in northern China in the three ensembles mainly results from overestimating the evaporation. Due to a significant reduction of the evaporation bias in the BMME, it performs better than the AMME and WMME in reproducing winter precipitation over northern China. In contrast, the enhancement of the BMME performance on the vertical moisture advection contributes to reducing the wet summer bias in NC and NEC.
Model name | DJF | JJA | |||
BMME | WMME | BMME | WMME | ||
ACCESS-CM2 | √ | ||||
ACCESS-ESM1-5 | √ | ||||
BCC-CSM2-MR | √ | ||||
CanESM5 | |||||
CESM2 | √ | ||||
CESM2-WACCM | |||||
EC-Earth3 | √ | ||||
EC-Earth3-Veg | √ | ||||
FGOALS-g3 | √ | ||||
GFDL-CM4 | √ | ||||
GFDL-ESM4 | |||||
INM-CM4-8 | √ | √ | |||
INM-CM5-0 | √ | ||||
IPSL-CM6A-LR | |||||
MIROC6 | |||||
MPI-ESM1-2-HR | |||||
MPI-ESM1-2-LR | √ | √ | |||
MRI-ESM2-0 | |||||
NorESM2-LM | |||||
NorESM2-MM |