Figure 1 shows the annual mean surface air temperature differences relative to APHRODITE. First note that although they share some common surface measurements, JRA55 and APHRODITE have major differences over the Tibetan Plateau and Southeast Asia regions (Fig. 1a), while the CPC data show large differences (relative to APHRODITE) over the Sichuan province in southwestern China (Fig. 1b). We should not expect an atmospheric numerical model, which is only constrained at the surface by ocean temperature, to have a better agreement with the observational benchmark (APHRODITE here) than the reanalysis product (JRA55), which is fully constrained by both ground and atmospheric observations. Thus, only the regions over which the model–APHRODITE difference is larger than the JRA55–APHRODITE difference are considered “significant” and stippled in Fig. 1c–f.

Similar to the difficulty in capturing the high-elevation temperature in JRA55, there is a warm bias over eastern and southern China and a cool bias over Tibet in the CAM5 simulations (Fig. 1c). The cool bias over the Tibetan Plateau and warm bias over the foothill regions (both the Indian side and the northern edge of the Tibetan Plateau) are a long-standing bias in many global and regional climate models. We note that the bias over the Sichuan province appears to be muted in the high-resolution version (Fig. 1e, f), illustrating the promise of further enhancing the resolution.

Moreover, CAM6α reduces the temperature bias over southern China (Fig. 1c, d). Over the entire domain considered in Fig. 1, the root mean square difference (RMSD) relative to APHRODITE is 2.3^∘ for CAM5-1^∘, 2.4^∘ for CAM6α-1^∘, 2.2^∘ for CAM5-0.25^∘ and 2.3^∘ for CAM6α-0.25^∘ (note that RMSD for JRA55 is 1.5^∘). The RMSD as a regional average quantity is not sufficient to characterize the performance of models that vary at finer scales. Because of this limitation, in the following evaluation for precipitation, we divide East Asia into several regional boxes (Fig. 2).

Figure 2The difference of 1980–2004 annual mean precipitation (mm d⁻¹) between APHRODITE and (a) JRA55, (b) MERRA2, (c) CAM5-1^∘, (d) CAM6α-1^∘, (e) CAM5-0.25^∘ and (f) CAM6α-0.25^∘. The values at the lower left of each panel indicate the root mean square difference (RMSD) relative to APHRODITE. Grid points in panels (c–f) are stippled when the absolute value of the difference between the model and APHRODITE is larger than that between JRA55 and APHRODITE. All the data were interpolated to 1^∘ resolution. The boxes with different colors in Fig. 2c are as follows. (1) Tibet (blue box): 27–37^∘ N, 79–99^∘ E; (2) southwestern China (red box): 28.5–35.5^∘ N, 100–105^∘ E; (3) Korea (black box): 34–40^∘ N, 124.5–129.5^∘ E; (4) Japan (gold box): 31–43^∘ N, 130–144^∘ E; (5) the Maritime Continent (green box): 9.75^∘ S–19.75^∘ N, 90–150^∘ E. The India average is entirely within mainland India. Northern China and southern China are also defined in Fig. 2c: northern China is north of Qin Mountain and the Huai River at 32.8^∘ N, and southern China is south of this. The western boundary of northern China and southern China is a straight line called the Hu–Huanyong line between Heihe (50.2^∘ N, 127.5^∘ E) and Tengchong (24.5^∘ N, 98.0^∘ E).

Figure 2 illustrates the climatological precipitation biases relative to APHRODITE. MERRA2 is considered here to be the third “data” source in addition to JRA55 and APHRODITE as an estimate of the large uncertainty of observational datasets (Herold et al., 2015, 2016a, b). All four CAM simulations have a dry bias over southern China and a wet bias over the rest of China, especially the Sichuan basin (near the eastern edge of the Tibetan Plateau) and the Himalayan mountain range that defines the southern edge of the Tibetan Plateau. The precipitation biases in those regions are particularly important for two reasons: (a) many major rivers in East Asia, South Asia and Southeast Asia have their headwaters in those regions; and (b) these regions are also prone to natural hazards such as landslides.

Overall, the CAM6α-1^∘ performance is slightly better than CAM5-1^∘, and CAM6α-0.25^∘ falls in between the 1^∘ resolution models (RMSD is 1.83 mm d⁻¹ for CAM5-1^∘, 1.62 mm d⁻¹ for CAM6α-1^∘ and 1.71 mm d⁻¹ for CAM6α-0.25^∘). The CAM5-0.25^∘ RMSD (1.42 mm d⁻¹) is better than that of CAM6-0.25^∘ due to the lower bias over the Himalayas and Kalimantan (Fig. 2e, f).

Because of large spatial heterogeneity, eight regions are selected to evaluate precipitation. Five domains are shown as colored boxes in Fig. 2c: (1) Tibet: 27–37^∘ N, 79–99^∘ E; (2) southwestern China: 28.5–35.5^∘ N, 100–105^∘ E; (3) Korea: 34–40^∘ N, 124.5–129.5^∘ E; (4) Japan: 31–43^∘ N, 130–144^∘ E; (5) the Maritime Continent: 9.75^∘ S–19.75^∘ N, 90–150^∘ E. Only the land within these boxes is considered in this study. The other three are India, northern China and southern China. The India average is entirely within mainland India. Northern China and southern China are also defined in Fig. 2c: northern China is north of Qin Mountain and the Huai River at 32.8^∘ N, and southern China is south of this. The western boundary of northern China and southern China is a straight line called the Hu–Huanyong line between Heihe (50.2^∘ N, 127.5^∘ E) and Tengchong (24.5^∘ N, 98.0^∘ E).

Figure 3 shows the RMSD and mean bias (taking the domain average of a respective model simulation or data products minus APHRODITE) of annual mean precipitation for nine selected regions. Note that over a few selected regions, the model performance is as good as the reanalysis (such as over Japan, Korea and southern China) and thus we will not further investigate model improvements. Among regions where CAM versions perform poorly compared with reanalyses (southwestern China, Tibet, India, northern China and the Maritime Continent), the CAM6α-1^∘ performance (RMSD) is better (lower) than CAM5-1^∘ for Tibet, southwestern China and northern China, but it gets worse for the Maritime Continent. Notably, CAM6α-0.25^∘ is closer to observations over southwestern China and northern China, but it also gets considerably worse for the Maritime Continent (with a large RMSD of 7.2 mm d⁻¹ and a bias of 4.0 mm d⁻¹). Both CAM versions with higher resolution simulate the climatological precipitation over northern China better. Increasing model resolution decreases the RMSD and bias of CAM5 over Tibet and southwestern China, while it increases those of CAM6 (Fig. 3).

Figure 3The root mean square difference (RMSD) and bias of annual mean precipitation (PRECT, mm d⁻¹) relative to APHRODITE for JRA55, MERRA2, CAM5-1^∘, CAM6α-1^∘, CAM5-0.25^∘ and CAM6α-0.25^∘ over six regions: Tibet: 27–37^∘ N, 79–99^∘ E; southwestern China: 28.5–35.5^∘ N, 100–105^∘ E; Korea: 34–40^∘ N, 124.5–129.5^∘ E; Japan: 31–43^∘ N, 130–144^∘ E; Maritime: 9.75^∘ S–19.75^∘ N, 90–150^∘ E; Asia: 5–55^∘ N, 60–140^∘ E. Three other domains follow the geographical boundary and climatic zones for India, northern China and southern China as in Lin et al. (2018). The India average is entirely within mainland India. The western boundary of northern China and southern China is a straight line (shown in Fig. 2c) called the Hu–Huanyong line between Heihe (50.2^∘ N, 127.5^∘ E) and Tengchong (24.5^∘ N, 98.0^∘ E). The separation of northern China and southern China (shown as the straight line in Fig. 2c) is along the latitude of Qin Mountain and the Huai River (32.8^∘ N).

Download

We will explore the details that might lead to the progressive improvement over southwestern China and northern China and the poor performance of CAM6-0.25^∘ over the Maritime Continent in Sect. 5.

We next assess the model performance in simulating convective and large-scale precipitation components separately. The ratio of convective to large-scale precipitation is a useful diagnostic because both convective activity and large-scale instability can lead to precipitation in this model. Most atmospheric models use convective parameterizations to represent the effects of sub-grid-scale convective processes, with reduced-complexity microphysics (Zhang and McFarlane, 1995; Kooperman et al., 2016). The convective precipitation in CAM5 includes shallow and deep convective precipitation. In CAM6 the shallow convective regime is handled by CLUBB coupled to stratiform microphysics, and hence shallow convective precipitation is prognostic and part of the large-scale precipitation.

The latent heating required by the atmosphere imposes an important constraint on the amount of mean rainfall, but model horizontal resolution and parameter settings dictate the timescales (Gustafson Jr. et al., 2014). CAM6 estimates shallow convective precipitation from the prognostic calculations in CLUBB (which have memory between time steps of turbulent motion) rather than diagnostically representing the effects of sub-grid-scale convective processes at each location and time step. One of the reasons why the simulated rainfall intensity is expected to improve when the model is run at higher resolution is that the variance of sub-grid-scale humidity and thermodynamics drops, and the parameterized sub-grid-scale processes (such as sub-grid-scale turbulence with CLUBB) are better separated into regimes (Kopparla et al., 2013).

Figure 4 illustrates the ratio of convective (PRECC) to large-scale (PRECL) precipitation. This ratio (PRECC ∕ PRECL) is greater over the ocean than over the land, as expected. CAM6α-1^∘ has a larger ratio over the tropics compared to CAM5-1^∘. CAM6α-0.25^∘ simulated a lower ratio than CAM6α-1^∘ (Fig. 4b) and a similar pattern as CAM5-0.25^∘ (Fig. 4c).

Figure 4The 1980–2004 convective to large-scale precipitation ratio (unitless) for (a) CAM5-1^∘, (b) CAM6α-1^∘, (c) CAM5-0.25^∘ and (d) CAM6α-0.25^∘.

Higher-horizontal-resolution models tend to simulate higher vertical velocities (Gettelman et al., 2018) and a lower ratio of convective to total rainfall. A larger fraction of precipitation can be resolved as a consequence of large-scale flow, limiting the need to invoke sub-grid convective schemes. Increasing resolution also better resolves topographic and surface effects, and it separates regimes as sub-grid-scale variance is reduced, particularly in the thermodynamic variables. With a fixed amount of precipitable water, more condensation caused by the stratiform scheme means less is available for convective precipitation, so the ratio becomes lower.

The compensation above is a feature of the physical parameterization suite in CAM due to the timescale. Large-scale liquid condensation by resolved-scale cloud schemes (CLUBB and microphysics) instantaneously condenses all vapor in excess of liquid saturation to cloud liquid. In contrast, the deep convective parameterization has a timescale that produces mass flux and precipitation at a defined rate. Note that as the time step gets shorter, the mass flux and precipitation over a time step will decrease, which is the major reason for the decrease in deep convective precipitation in CAM6. The large-scale condensation (including shallow convection and cloud microphysics) does more as the time step changes, while the deep convective parameterization does less (Gettelman et al., 2018).

Figure 5 uses Taylor diagrams (Taylor, 2001) to evaluate the simulation of annual and seasonal surface air temperature, mean precipitation and two metrics of extreme precipitation. Red circles show the results from CAM5-1^∘, with blue for CAM6α-1^∘ and green for CAM6α-0.25^∘. Movement closer to the point of (1.0,1.0) in the Taylor diagrams indicates improvement of the simulation relative to APHRODITE. For example, the model circles are very close to the blue circles representing JRA55 for near-surface (2 m) temperature (TREFHT) (Fig. 5a), which indicates good performance for surface air temperature for all three models. The annual and seasonal precipitation (PRECT) correlations of the two CAM6α simulations are around 0.8, while the annual correlation of CAM5-1^∘ is 0.7 (Fig. 5b). The CAM6α-0.25^∘ root mean square (RMS) is bigger than that of CAM6α-1^∘ (green circles relative to blue circles in Fig. 5b). The maximum daily precipitation during each month of a year (RX1day) of CAM6α-1^∘ performs better than CAM5 (blue circles relative to red circles in Fig. 5c). The number of heavy precipitation days (R10) has large differences among the observational and reanalysis data (black and gold circles in Fig. 5d). CAM6α has higher correlation coefficients with APHRODITE (∼0.8) than CAM5 (∼0.7).

Figure 5Taylor diagrams for (a) surface air temperature, (b) precipitation, (c) RX1day (the maximum daily precipitation during each month of a year; mm d⁻¹) and (d) R10 (number of days with precipitation more than 10 mm; units: days). These two extreme indices are suggested by the Expert Team for Climate Change Detection and Indices (Zhang et al., 2011). Note that R10, by design, only has an annual mean but not a seasonal mean. Spatial correlations and normalized RMS (root mean square) are calculated for ANN (1, annual), DJF (2, winter), MAM (3, spring), JJA (4, summer) and SON (5, fall). APHRODITE is used as the benchmark over Asia for 1980–2004. All the data were interpolated to 1^∘ resolution.

Download

Overall, CAM6α versions perform better than CAM5-1^∘ for the mean precipitation (Fig. 5b), maximum daily precipitation and number of heavy precipitation days (Fig. 5c, d). Although CAM6α-0.25^∘ performs worse than 1^∘ for the mean precipitation and RX1day (green circles relative to blue circles in Fig. 5b and c), mostly due to the bias over the Maritime Continent (defined as 9.75^∘ S–19.75^∘ N, 90–150^∘ E), CAM6α-0.25^∘ performs better for R10 (Fig. 5d). All three model versions show that precipitation bias in JJA is bigger than other seasons (Fig. 5b, c).

In this section, we evaluate simulated rainfall variability across a wide range of timescales. This serves as a useful test of model performance because it is difficult to improve model performance in simulating temporal variability, as opposed to the climatological average, based on the simple tuning of one or more parameters. Moreover, it is expected that CAM6 will be widely used to study climate variability at various timescales.

4.1 Variability of precipitation due to PDO, ENSO and East Asian summer monsoon index

The Pacific Decadal Oscillation (PDO) represents variability in the tropical and extratropical North Pacific at interdecadal timescales that significantly impact climate (Mantua et al., 1997; Meehl et al., 2013). Figure 6 shows a regression of observed and simulated annual precipitation against the PDO index, derived from the leading principal component of monthly SST anomalies in the North Pacific Ocean (Mantua et al., 1997). A common feature as revealed in the observational records is the drying tendency over Indochina (e.g., Thailand) during positive PDO phases. This feature is well represented in all four CAM versions. Similarly, a wet anomaly over most of India during positive PDO is also captured in all CAM versions. In contrast, the wet anomaly over southern China is simulated well in the CAM6-0.25^∘ version, while the opposite anomaly is seen in CAM6α-1^∘ and CAM5-0.25^∘.

Figure 6Regression of annual mean precipitation (mm d⁻¹) onto observed PDO indices for 1980–2004.

The El Niño–Southern Oscillation (ENSO) plays a central role in ocean–atmosphere coupled interannual variability. CAM5 has been widely used to study ENSO impacts over Asia (Chen et al., 2018; Hoell et al., 2016). Figure 7a–g show the regression coefficients between annual mean precipitation and ENSO. The ENSO index is the cold tongue index following Deser and Wallace (1987). Both 1^∘ CAM5 and CAM6α (Fig. 7d, e) capture the observed wet anomaly over Pakistan and Afghanistan and the dry anomaly over Indonesia. However, we find that the drying tendency over southern China during El Niño years (the upper row of Fig. 7) is completely missing in CAM5 but starts to emerge in CAM6α-1^∘ and gets better in CAM6α-0.25^∘. The influence of ENSO on boreal summer monsoon rainfall over Asia is well known. Figure 7h–n show the regression coefficients between JJA precipitation and ENSO. Those patterns are similar when the analysis is done with annual mean precipitation but with a weaker correlation of annual precipitation and ENSO. Our results here thus call into question the fidelity of previous ENSO studies on hydroclimate over southern China using CAM5.

Figure 7Regression of annual mean (a–g) and summer (JJA, h–n) precipitation onto the observed ENSO index (mm d⁻¹) for 1980–2004. Top color bar is for panels (a–g), bottom color bar is for panels (h–n).

Next, we examine seasonal precipitation variability associated with the East Asian summer monsoon (EASM). The EASM is an important climate system over eastern China. Climate models are widely used for EASM studies (Zhou and Li, 2002; Chen et al., 2010; Zhou et al., 2013), although model biases for the EASM are a long-lasting problem from CMIP3 (Coupled Model Intercomparison Project phase 3) to CMIP5 with limited improvement (Song and Zhou, 2015; Kusunoki and Arakawa, 2015). The East Asian summer monsoon index (EASMI) is a unified dynamic normalized seasonality (DNS) monsoon index defined by Li and Zeng (2002, 2003):

$$\begin{array}{}\text{(1)}& \text{EASMI}={\displaystyle \frac{∥\stackrel{\mathrm{‾}}{{\mathit{V}}_{\mathrm{1}}}-{\mathit{V}}_i∥}{∥\stackrel{\mathrm{‾}}{\mathit{V}}∥}}-\mathrm{2},\end{array}$$

where $\stackrel{\mathrm{‾}}{{\mathit{V}}_{\mathrm{1}}}$ and V_i are the January climatological and monthly wind vectors for a grid, respectively, and $\stackrel{\mathrm{‾}}{\mathit{V}}$ is the mean of January and July climatological wind vectors for the same gird. The constant 2 on the right-hand side of the formula is the determinant criterion. The double vertical line indicates the normalized value. EASMI can be used to depict both the seasonal cycle and interannual variability of the EASM. The EASMI summer average is considered here.

Figure 8 shows regression coefficients between summer (JJA) precipitation in the east of China and EASMI. There is an apparent negative relationship between the EASMI and summer rainfall in the middle and lower reaches of the Yangtze River in China for the observational and reanalysis data (Fig. 8a–c). All three CAM versions capture the positive correlation over southern China (Fig. 8d–g) found in the observations and reanalyses (Fig. 8a–c), although CAM6α at both 1 and 0.25^∘ and CAM5-0.25^∘ has the northern edge of the positive correlation more northward (about 2^∘ latitude) distributed than that of observational and reanalysis data (Fig. 8e–g).

Figure 8Regression of summer (JJA) precipitation onto EASMI (mm d⁻¹) for 1980–2004.

4.2 Seasonal cycle

Due to the monsoonal influence on both East Asia and South Asia, summer (JJA) precipitation dominates the annual total. Thus, the summer–winter contrast in precipitation serves as a useful quantity for model evaluation. JRA55 has less of a summer–winter contrast than APHRODITE (Fig. 9a). Compared to APHRODITE, MERRA2 has almost the same annual cycle over China (Fig. 9b). In CAM5, the JJA precipitation appears to be too large compared to the rest of the year. CAM6α-1^∘ simulations reduce the bias in the annual cycle over Pakistan (Fig. 9c, d), although models continue to overestimate seasonal variability over arid regions in western China. CAM6α-0.25^∘ shows a bigger bias than CAM5-0.25^∘ over the Himalayas (Fig. 9e, f). The seasonal variability is significantly improved in CAM6α-0.25^∘, especially in Himalayan mountain regions. Southern China, again, is a region where the CAM6-0.25^∘ performance is improved substantially. This is further diagnosed in the next figure.

Figure 9The difference of summer minus winter (JJA − DJF) precipitation (mm d⁻¹) between APHRODITE and (a) JRA55, (b) MERRA2, (c) CAM5-1^∘, (d) CAM6α-1^∘, (e) CAM5-0.25^∘ and (f) CAM6α-0.25^∘. The values at the lower left of every panel indicate the root mean square difference (RMSD) relative to APHRODITE. Grid points in panels (c–f) are stippled when the absolute value of the difference between the model and APHRODITE is larger than that between JRA55 and APHRODITE.

In Fig. 10, we show the time series of zonal mean precipitation between 100 and 125^∘ E to evaluate model performance in simulating the annual cycle of EASM-related precipitation. Figure 10 shows that precipitation mainly occurs from May to September, rapidly shifts to northern China around June and continues to September (Su et al., 2017). It is clear that APHRODITE, JRA55 and MERRA2 all depict such a northern shift (Fig. 10a, b, c). Both versions of CAM6α capture this northern shift and its persistence from June to September, especially CAM6α-0.25^∘ (Fig. 10e, f), while CAM5-1^∘ illustrates persistence from June to July only (the area of continuous yellow shading in Fig. 10d is smaller than those of CAM6). High-resolution CAM5-0.25^∘ captures that shift and its persistence from June and September (Fig. 10g). This explains the CAM5-1^∘ deficiency as shown in Fig. 9.

Figure 10Annual cycle of the regional mean precipitation rate (mm d⁻¹) for 1980–2004 within a part of Asia (between 100 and 125^∘ E): (a) APHRODITE, (b) JRA55, (c) MERRA2, (d) CAM5-1^∘, (e) CAM6α-1^∘, (f) CAM6α-0.25^∘ and (g) CAM5-0.25^∘.

Download

4.3 Daily precipitation frequency and diurnal cycle of precipitation

Figure 11 illustrates the daily rainfall frequency distribution from reanalyses, observations and models. Despite large uncertainties among observational datasets, daily rainfall intensity for CAM6-0.25^∘ (green line) is closer to observational values over southwestern China than other simulations (Fig. 11b). CAM5-0.25^∘ overestimates the frequency of light precipitation (0.1–10 mm d⁻¹) over southwestern China (purple line in Fig. 11b). CAM5-1^∘ simulates a higher frequency of light precipitation (0.1–1.0 mm d⁻¹) over Korea, Japan, northern China and southern China (Fig. 11a, c, d, g, h). New physical schemes in CAM6-1^∘ capture the observed distribution over Korea and Japan (Fig. 11c, d). Realistic rainfall intensity is not energetically necessary because more frequent weak events can produce the same latent heating as less frequent but more intense rainfall events. CAM6 improves the light large-scale precipitation over Korea, Japan and northern China (not shown).

Figure 11Frequency distribution of daily precipitation (mm d⁻¹) over (a) Tibet, (b) southwestern China, (c) Korea, (d) Japan, (e) the Maritime Continent (see the boxes in Fig. 2c), (f) India, (g) northern China and (h) southern China for 1980–2004. Black lines (shading) are for APHRODITE (the maximum and minimum of APHRODITE, JRA55 and MERRA2). Red solid lines are for CAM5-1^∘; blue solid lines are for CAM6α-1^∘; green solid lines are for CAM6α-0.25^∘; purple solid lines are for CAM5-0.25^∘. All data were interpolated to 1^∘ resolution before regional averaging.

Download

To look into the heavy precipitation more carefully, Figure 12 shows precipitation percentiles from 90 % to 99.99 %, which captures the heaviest precipitation events (Kooperman et al., 2016). CAM6 with new physics modules (blue) has a better performance than CAM5 over five of eight selected regions (Tibet, southwestern China, Japan, India, northern and southern China). Higher horizontal resolution in CAM6 and CAM5 (green and purple) better simulates intensities over the Maritime Continent (Fig. 12e), but the results of CAM6 (CAM5) degrade (upgrade slightly) for the heaviest precipitation events over India, northern China and southern China (Fig. 12f–h). No significant differences are found between CAM5 and CAM6 over Tibet, southwestern China, Japan and the Maritime Continent.

Figure 12Similar to Fig. 11, but for daily precipitation rates (mm d⁻¹) as a function of percentile.

Download

Figure 13 shows the diurnal cycles of June precipitation rate from TRMM satellite observations and model simulations. Satellite observations show that the diurnal peak in precipitation is around 20:00 (LT) over Tibet, India and the Maritime Continent, while the peak over southern China is in the afternoon (15:00 LT) (Fig. 13a). CAM6α reproduces many features of the TRMM observations (Fig. 13c, e), while CAM5-1^∘ and CAM5-0.25^∘ simulate an earlier-in-the-day peak precipitation over India, southern China and the Maritime Continent (Fig. 13b, d). CAM6α-0.25^∘ simulates the diurnal cycle better than CAM6α-1^∘ in southern Tibet (Fig. 13d) along the Himalayas. In general, CAM6 at all resolutions has significant improvements over CAM5 in simulating the diurnal cycle of precipitation in East Asia, and CAM5 with higher resolution improves the diurnal cycle only slightly.

Figure 13Diurnal cycle of precipitation in June (6-year average for 1999–2004) of (a) TRMM, (b) CAM5-1^∘, (c) CAM6α-1^∘, (d) CAM5-0.25^∘ and (e) CAM6α-0.25^∘. The local time peak of the diurnal cycle is shown in color on the color wheel. The intensity of the color is the amplitude of precipitation.

Download

The CLUBB parameterization appears to be the reason for this improvement. CLUBB has prognostic moments that provide memory to facilitate the initiation of shallow convection in the midmorning and early afternoon. It is important to note that improved simulation of the diurnal cycle by CLUBB is a robust feature in every coupled and atmosphere-only simulation with CAM6α. The CLUBB unified parameterization is able to prevent the deep convective scheme from firing off too early and better simulate a gradual transition of these regimes successfully (Bogenschutz et al., 2018).

Here, we quantitatively contrast climate variables over southwestern China, northern China and the Maritime Continent, since the model with new physical modules or high horizontal resolution simulates a better climatology over southwestern China and northern China but does the opposite over the Maritime Continent (Fig. 3). We attempt to attribute changes to either physical parameterizations or resolution. Additionally, we investigate whether the improvement due to resolution is dependent on the CAM version. Table 2 illustrates simulation differences due to physical parameterizations (CAM6α-1^∘ minus CAM5-1^∘) and higher horizontal resolution (CAM6α-0.25^∘ minus CAM6α-1^∘ and CAM5-0.25^∘ minus CAM5-1^∘).

New physical modules and higher horizontal resolution in CAM6α perform better over southwestern China by decreasing convective precipitation (by −0.38 and −0.81 mm d⁻¹, respectively). New physical modules simulate larger surface latent heat flux but less vertically integrated humidity, which leads to a decrease in convective precipitation over southwestern China (Table 2). Higher resolutions in CAM6 and CAM5 both decrease the surface latent heat flux and convective precipitation over southwestern China (Table 2). However, higher resolution leads to an opposite change in surface sensible heat flux (total cloud amount) between CAM5 and CAM6α by 12.7 and −7.4 W m⁻² (−3.9 % and 0.5 %), respectively.

Newer physics parameterizations in CAM6α simulate a stronger solar flux reaching the surface in northern China (Table 2). This may be due to improvements in the diurnal cycle of precipitation (Li et al., 2008). The stronger solar flux leads to larger latent heat release, although convective precipitation over northern China is not changed by the new physics modules. Better representation of topography at 0.25^∘ resolution leads to increased downward shortwave flux at the surface in northern China. Less convective precipitation associated with increased resolution is seen over northern China in Table 2 (−0.31 mm d⁻¹ for CAM6 and −0.33 mm d⁻¹ for CAM5). Note that CAM6 simulates a decrease in the surface latent heat flux with a higher resolution, while CAM5 does not. We also examine the moisture budget using diagnostics for precipitation changes (Chou and Lan, 2012). The moisture budget analysis defines the mass conservation of water substance in an atmospheric column as

where P is precipitation, q is specific humidity, u(v) is zonal (meridional) wind and E is evaporation into the atmosphere. $<X>$ is a mass-weighted vertical integral and $\stackrel{\mathrm{‾}}{X}$ denotes a temporal average. The horizontal advection can be further decomposed into the stationary and transient terms based on

where $\stackrel{\mathrm{‾}}{\left[X\right]}$ and $\left({\stackrel{\mathrm{‾}}{X}}^{*}\right)$ are the climatological zonal mean (stationary) eddy, and X^′ is transient variation. See Yao et al. (2017) and Chen et al. (2018) for more details.

No significant difference in evaporation (E) is seen between the model and JRA55 data (Fig. 14). The results indicate that CAM6 (both resolutions) simulates a moisture budget closer to JRA55 than CAM5-1^∘, and the model precipitation bias appears mostly in the zonal moisture flux convergence term $(-{\partial }_x(\stackrel{\mathrm{‾}}{uq}\left)\right)$ over southwestern China (Fig. 14a). Although a large residual over northern China (Fig. 14b) may result from water vapor transport by the surface vertical movement induced by terrain slope (Trenberth and Guillemot, 1995; Seager et al., 2010), all four versions of CAM and JRA55 show that the zonal mean of specific humidity eddy transport $\left[\stackrel{\mathrm{‾}}{q}\right]$ dominates northern China precipitation (Fig. 14b). CAM5-0.25^∘ simulates a similar moisture budget as CAM6-0.25^∘, while the corresponding results are different between CAM5-1^∘ and CAM6-1^∘ over southwestern China and northern China (Fig. 14a, b).

Figure 14Moisture budget over (a) southwestern China, (b) northern China and (c) the Maritime Continent (see the boxes in Fig. 2c) for 1980–2004, including precipitation, evaporation, zonal and meridional moisture flux convergence, and the residual and (right to dashed line) major processes contributing to moisture flux convergence.

Download

Next, we explore the differences in simulated variables due to higher horizontal resolution over the Maritime Continent (Fig. 3). As seen in Table 2, the higher horizontal resolution in CAM6 not only increases the vertically integrated total cloud cover over the Maritime Continent, but also leads to more shortwave flux reaching the surface, which tends to release more latent heat. Both CAM5 and 6 versions with 0.25^∘ resolution reduce the convective precipitation and increase the large-scale precipitation relative to 1^∘ resolution, which leads to an overestimation of total precipitation. The two 0.25^∘ resolution CAM versions simulate the same surface air temperature, surface energy terms (surface latent and sensible heat flux, downwelling solar flux, and longwave flux at the surface) and vertically integrated humidity change but simulate a different vertically integrated total cloud change and downwelling clear-sky solar flux at the surface (Table 2). With a fully coupled Community Climate System Model version 4 (CCSM4), an earlier version of CAM (CAM4), Shields et al. (2016) have shown that higher horizontal resolution tends to decrease convective precipitation and increase large-scale precipitation. The moisture budget analysis shows that the meridional specific humidity eddy transport is the main factor leading to the bias over the Maritime Continent (Fig. 14c). Note that the analysis in this study is focused on the Maritime Continent land area only. Johnsan et al. (2016) carried out a moisture budget analysis for the whole Maritime Continent (including the ocean) and found that increased resolution causes increased moisture convergence and precipitation on the windward (southern) side of the orography, which leads to decreased moisture availability on the leeward (northern) side, reducing precipitation.

The model code of CESM2 is released at http://www.cesm.ucar.edu/models/cesm2/, last access: 1 June 2019 (Hurrell et al., 2013). Readers can download the current release code using the following command: git clone -b release-cesm2.1.0 (https://github.com/ESCOMP/cesm, last access: 1 June 2019).

The model data are archived at https://doi.org/10.5281/zenodo.2548255 (Lin, 2019). Details about MERRA-2 can be found at https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/, last access: 1 June 2019. Details about APHRODITE can be found at http://www.chikyu.ac.jp/precip/, last access: 1 June 2019. The observational EASMI is from http://ljp.gcess.cn/dct/page/65577, last access: 1 June 2019. The ENSO index is from http://research.jisao.washington.edu/data/cti/, last access: 1 June 2019. The PDO index is from http://research.jisao.washington.edu/pdo/PDO.latest, last access: 1 June 2019. TRMM is from https://pmm.nasa.gov/data-access/downloads/trmm, last access: 1 June 2019. The CPC dataset is from https://www.esrl.noaa.gov/psd/data/gridded/data.cpc.globaltemp.html, last access: 1 June 2019.

WD and CW designed the study, LL led the analysis, AG performed CESM simulations, and NR and SCB performed the CAM5-0.25^∘ simulation. LL, YX and ZW prepared the paper with contributions from all coauthors.

The authors declare that they have no conflict of interest.

This research has been supported by the The National Key Research and Development Program of China (grant no. 2016YFA0602701).

This paper was edited by Paul Ullrich and reviewed by two anonymous referees.