-
The dynamical core of GRIST solves the primitive equations but has an option to restore the vertical acceleration term for nonhydrostatic modeling (Zhang et al., 2019, 2020). This layer-averaged, dry-mass, unstructured-mesh dynamical core allows both uniform-resolution and VR modeling within a unified global framework (Zhang et al., 2019; Zhou et al., 2020). It conserves a dry air mass to within machine accuracy and conserves the “total energy” within a magnitude that is suitable for global climate modeling.
The physics-dynamics coupling of GRIST supports the application of multiple physics suites (Zhang et al., 2020). In this study, the hydrostatic core is coupled with a physics suite that was originally taken from CAM5.3. A pure operator-splitting approach is used for the physics-dynamics coupling (Li et al., 2022). It inherits several CAM5-derived parameterization schemes, including the UW moist PBL turbulence scheme (Bretherton and Park, 2009), the Morrison–Gettelman microphysics and Park macrophysics parameterizations (Gettelman et al., 2010; Park et al., 2014), and the RRTMG (Rapid Radiative Transfer Model for GCMs, Iacono et al., 2008). The convective parameterization uses a double-plume scheme (Chu et al., 2022; Li et al., 2022), with revisions in the trigger and closure of deep convection described in the next section. At the surface, the Noah-MP land model (Niu et al., 2011) is used. The fluxes over the ocean and sea ice use the formulation from CAM3 (Collins et al., 2004).
-
The default double-plume convective parameterization uses a pair of entraining/detraining plumes to describe deep and shallow convection independently. The plumes share the same buoyancy-sorting bulk mass flux cloud function but use different lateral mixing rates and trigger and closure assumptions. Entrainment occurs at the lateral boundary of the plume by turbulent mixing. The lateral mixing rate is set to a constant value of 3 km–1 for shallow convection. For deep convection, a mixing rate dependent on relative humidity is applied to inhibit deep convection in hostile environments (Chu et al., 2022). Detrainment includes the turbulent mixing at the lateral boundary, constrained detrainment at the top of a plume, and the effect of anisotropy that occurs at the level where the buoyancy starts to decrease with height. The precipitation formation and re-evaporation are calculated following Park and Bretherton (2009) but using different condensate threshold values (1 g kg−1 for deep convection and 0.5 g kg−1 for shallow convection).
Shallow convection is triggered if the turbulent kinetic energy (TKE) overcomes the convective inhibition (i.e., a CIN-TKE trigger). The trigger of deep convection is further determined by a dynamic CAPE production similar to that used in Wang and Randall (1994), i.e., deep convection is triggered when CAPE is generated at the previous time step, and sufficient TKE is present to overcome the CIN. The CAPE is calculated with a dilute plume approximation (Neale et al., 2008). Accordingly, a closure based on a dynamic CAPE quasi-equilibrium assumption is used for the deep plume, i.e.,
${{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}}/{\partial t}|}_{\mathrm{L}}={{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}}/{\partial t}|}_{\mathrm{c}\mathrm{u}}$ , assuming that the CAPE generated by non-convective processes (L) is balanced by the deep convective (cu) consumption (Zhang, 2002; Bechtold et al., 2014). The CAPE tendency due to convection is approximated by the heating through compensating environmental subsidence:where
$ {M}_{\mathrm{b}} $ is the actual cloud base mass flux,$ {M}_{\mathrm{b}}^{*} $ is a prescribed cloud base mass flux (set as 0.1 m2 s–1), and$ {M}^{*} $ is the initial vertical mass flux profile obtained from the updraft computation based on$ {M}_{\mathrm{b}}^{*} $ ,$\overline{Tv}$ is grid-scale virtual temperature, and$ {c}_{p} $ is the specific heat at constant pressure. The cloud base mass flux is expressed as:where the superscript ' denotes the atmospheric state after the calculation of the deep plume at the previous time step (regardless of whether it is triggered or not).
GRIST can produce an overall reasonable mean climate of circulation, precipitation, and radiation budget using this physics suite (Li et al., 2022). However, the peak of daytime precipitation occurs too early over land, thus revealing its deficiency in capturing the diurnal cycle of precipitation (cf., Fig. 8). We speculate that the initiation of deep convection is overly coupled to the solar diurnal cycle if solar radiation, surface heating, and PBL turbulence are considered to contribute to the onset of convection. Following Xie et al. (2019), the trigger of deep convection can be revised so that convection occurs only when the large-scale advection makes a positive contribution to the existing positive CAPE and the TKE overcomes the CIN. The threshold value for CAPE production by the dynamical core is 0 J kg–1 h–1 (denoted as dCAPE0). Cui et al. (2021) investigated the sensitivity of the dCAPE trigger to its threshold value (denoted as dCAPEN, N > 0 J kg–1 h–1) in CAM5. They suggested that the diurnal phase of rainfall over the land can be further improved by setting the dCAPE threshold value as ~55 J kg–1 h–1, at the expense of degraded mean precipitation (especially over the ocean) and a reduction in the diurnal amplitude of precipitation. Considering that the purpose of dCAPEN trigger is to relax the overly strong coupling between deep convection and surface heating, dCAPEN is only used over the land in this study to avoid the potential degradation over the ocean.
Figure 8. Annual cycle of mean precipitation (units: mm d–1) for GPCP, ERA5, and each model averaged over the five sub-regions denoted in Fig. 2: (a) Tibetan Plateau, (b) Western Plain, (c) Southwest China, (d) Eastern Plain, and (e) South China.
The key to improving the closure to avoid potential reduction of the diurnal amplitude of precipitation is to re-quantify the CAPE production by the non-convective processes under the more restricted trigger condition, i.e., whether to update
$ {{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}}^{\text{'}}}_{} $ in Eq. (2) when deep convection does not trigger. Here, we allow other physical processes, especially the shallow convection, to consume CAPE, that is if the CAPE at the current (n) time step is less than${\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}}'$ , then${\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}}'(n+1)=\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}$ (Fig. 1). Otherwise,${\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}}'$ remains unchanged, which means that the generated CAPE will be accumulated and consumed by deep convection once the revised dCAPE trigger is satisfied. It potentially allows the model to capture the transition of convection from shallow to deep, which is an important element for the diurnal cycle of precipitation (Zhang and Klein, 2010; Rio et al., 2019). The deep and shallow plumes are coupled in a parallel fashion and share the same cloud function, thus favoring the transition. -
A VR model configuration (110–35 km) is used in this study, within which the highest resolution cells cover a part of continental East Asia and the surrounding islands and seas (Fig. 2). The VR configuration is chosen to examine the revised convective parameterization for two reasons. First, earlier modeling tests suggested that the VR model reduces the wet biases of mean precipitation over the West Pacific and East Asia. Thus, we may further explore improvements in the diurnal cycle of precipitation based on this added value of higher resolution for mean precipitation. Second, the VR setup provides a more challenging environment as the modification in convective parameterization may introduce additional sensitivities to the transition region due to the continuous changes in mesh size.
Figure 2. Surface elevation over East Asia at 110–35 km resolution (units: m). The gray contours show the grid scales in VR modeling. The five rectangles correspond to the selected sub-regions over China for regional analysis: Tibetan Plateau (28°–35°N, 103°–108°E), Western Plain (28°–35°N, 103°–108°E), Southwest China (23°–27°N, 100°–105°E), Eastern Plain (28°–35°N, 112°–120°E), and South China (23°–27°N, 112°–118°E).
A series of AMIP-type sensitivity simulations were conducted (Table 1). The simulations were initiated in May 2000 and continuously integrated for 5.5 years using the pre-processed sea surface temperature and sea-ice concentration data (Taylor et al., 2000). The model output for 2001–05 was used to assess its representation of mean state and precipitation variability. The vertical resolution consists of a 30-full-level Lorenz grid with a model top at 2.25 hPa. The model (physical) time step (dt) was set to 300 s, and the dynamical core is sub-cycled with dt = 100 s within each model time step. The simulation using the default trigger and closure assumption of deep convection is referred to as CTRL. The two sensitivity experiments, dCAPE0 and dCAPE60, use the revised dCAPE trigger and closure with dCAPE threshold values set to 0 and 60 J kg–1 h–1, respectively.
Name Resolution Timestep Trigger of deep convection Closure of deep convection CTRL 110–35 km, refined over East Asia 300 s (1) Sufficient TKE to overcome the CIN; (2) CAPE is generated at the previous time step, ${\left.\dfrac{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }{\partial t}\right|}_{ {\rm L} }{\approx \left.\dfrac{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}-{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }' }{\Delta t}\right|_{\rm L} } > 0$. ${\dfrac{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }{\partial t}\Bigg|}_{\mathrm{L} }=-{\dfrac{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }{\partial t}\Bigg|}_{\mathrm{c}\mathrm{u} }$, the dynamic CAPE is built up by all the non-convective processes at the previous step. dCAPE0 110–35 km 300 s (1) Sufficient TKE to overcome the CIN; (2) CAPE is generated by dynamical core at the previous time step, i.e., dCAPE > 0 J kg–1 h–1. The dynamic CAPE is represented by the net CAPE generation during the period that deep convection does not trigger. dCAPE60 110–35 km 300 s Same as dCAPE0, but increasing the threshold of CAPE production by dynamical core over the land, i.e., dCAPE > 60 J kg–1 h–1. Same as dCAPE0. Table 1. Description of experiments
The satellite-based observational and reanalysis precipitation datasets used in this study include the products from the Global Precipitation Climatology Project (GPCP, 1979–2010, Adler et al., 2018) and the monthly mean ERA5 reanalysis (2001–20, Hersbach et al., 2020). These datasets evaluate the global mean precipitation and its seasonal variation and the 30-min Integrated Multi-satellite Retrievals for Global Precipitation Measurement (GPM) (IMERG, 2001–05, Huffman et al., 2018), as well as the hourly ERA5 datasets are used for assessing the diurnal cycle. Five sub-regions over China are chosen for further evaluation of the variability of precipitation at various temporal scales (Fig. 2, black boxes). These include the Tibetan Plateau (28°–35°N, 103°–108°E), Western Plain (28°–35°N, 103–108°E), Southwest China (23°–27°N, 100°–105°E), Eastern Plain (28°–35°N, 112°–120°E), and South China (23°–27°N, 112°–118°E).
Name | Resolution | Timestep | Trigger of deep convection | Closure of deep convection |
CTRL | 110–35 km, refined over East Asia | 300 s | (1) Sufficient TKE to overcome the CIN; (2) CAPE is generated at the previous time step, ${\left.\dfrac{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }{\partial t}\right|}_{ {\rm L} }{\approx \left.\dfrac{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}-{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }' }{\Delta t}\right|_{\rm L} } > 0$. | ${\dfrac{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }{\partial t}\Bigg|}_{\mathrm{L} }=-{\dfrac{\partial \mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E} }{\partial t}\Bigg|}_{\mathrm{c}\mathrm{u} }$, the dynamic CAPE is built up by all the non-convective processes at the previous step. |
dCAPE0 | 110–35 km | 300 s | (1) Sufficient TKE to overcome the CIN; (2) CAPE is generated by dynamical core at the previous time step, i.e., dCAPE > 0 J kg–1 h–1. | The dynamic CAPE is represented by the net CAPE generation during the period that deep convection does not trigger. |
dCAPE60 | 110–35 km | 300 s | Same as dCAPE0, but increasing the threshold of CAPE production by dynamical core over the land, i.e., dCAPE > 60 J kg–1 h–1. | Same as dCAPE0. |