Search

Atmosphere modelling applications become increasingly memory-bound due to the inconsistent development rates between processor speeds and memory bandwidth. In this study, we mitigate memory bottlenecks and reduce the computational load of the GRIST dynamical core by adopting the mixed-precision computing strategy. Guided by a limited-degree of iterative development principle, we identify the equation terms that are precision insensitive and modify them from double- to single-precision. The results show that most precision-sensitive terms are predominantly linked to pressure-gradient and gravity terms, while most precision-insensitive terms are advective terms. The computational cost is reduced without compromising the solver accuracy. The runtime of the model's hydrostatic solver, non-hydrostatic solver, and tracer transport solver is reduced by 24 %, 27 %, and 44 %, respectively. A series of idealized tests, real-world weather and climate modelling tests, has been performed to assess the optimized model performance qualitatively and quantitatively. In particular, in the high-resolution weather forecast simulation, the model sensitivity to the precision level is mainly dominated by the small-scale features. While in long-term climate simulation, the precision-induced sensitivity can form at the large scale. [ABSTRACT FROM AUTHOR]

A double-plume convective parameterization scheme is revised to improve the precipitation simulation of a global model (Global-to-Regional Integrated Forecast System; GRIST). The improvement is achieved by considering the effects of large-scale dynamic processes on the trigger of deep convection. The closure, based on dynamic CAPE, is improved accordingly to allow other processes to consume CAPE under the more restricted convective trigger condition. The revised convective parameterization is evaluated with a variable-resolution model setup (110−35 km, refined over East Asia). The Atmospheric Model Intercomparison Project (AMIP) simulations demonstrate that the revised convective parameterization substantially delays the daytime precipitation peaks over most land areas, leading to an improved simulated diurnal cycle, evidenced by delayed and less frequent afternoon precipitation. Meanwhile, changes to the threshold of the trigger function yield a small impact on the diurnal amplitude of precipitation because of the consistent setting of dCAPE-based trigger and closure. The simulated mean precipitation remains reasonable, with some improvements evident along the southern slopes of the Tibetan Plateau. The revised scheme increases convective precipitation at the lower levels of the windward slope and reduces the large-scale precipitation over the upper slope, ultimately shifting the rainfall peak southward, which is in better agreement with the observations. [ABSTRACT FROM AUTHOR]

We investigated the resolution sensitivity of the Global‐to‐Regional Integrated forecast SysTem global nonhydrostatic model characterized by explicit dynamics–microphysics coupling using varying uniform resolutions (120, 60, 30, 15, and 5 km). The experiments followed the DYnamics of the Atmospheric general circulation Modeled On Non‐hydrostatic Domains (DYAMOND) winter protocol, which covers a 40‐day integration. These simulations did not activate parameterized convection. One 120 km test with parameterized convection was performed as a coarse‐resolution reference. Other model configurations for different simulations were kept as consistent as possible. Our results showed that the model gradually improved its representation of the fine‐scale features as the resolution increased. The 5 km simulation was overall close to a 3.75 km simulation during the first 12 days of the DYAMOND winter. With respect to the mean climate, the 5 km simulation had a more realistic rainfall distribution than the lower resolution explicit convection simulations. Cloud water and the related physical fields (e.g., shortwave cloud radiative forcing) had a large resolution sensitivity. The tropical rainfall frequency–intensity spectra became more realistic in the 5 km explicit convection simulation, but the 120 km run with parameterized convection showed a more realistic mean climate. As the resolution increases, the mean bulk effect of finely resolved model convection gradually converges to that of parameterized convection. The mean climate of this storm‐resolving model has slightly higher rainfall biases than a parameterized convection coarse‐resolution model, highlighting the importance of balancing resolved‐ and under‐resolved model convection for developing a unified multiscale global model. Key Points: The resolution sensitivity of the Global‐to‐Regional Integrated forecast SysTem nonhydrostatic model with an explicit‐convection setup is assessed from O (100 km) to O (km) scalesThe model gradually improves its representation of fine‐scale fluid structures with increasing resolutionThe interaction between cumulus ensemble and large‐scale flow of a Global storm‐resolving model overall converges to that of a parameterized convection model [ABSTRACT FROM AUTHOR]

Convective parameterization can drastically regulate the mean climate and tropical transient activity of a General circulation model (GCM). In this study, the physics suite of the NCAR Community Atmosphere Model, version 5 (CAM5) was first ported to the Global‐to‐Regional Integrated Forecast System model. Then, the original convective parameterization of CAM5—with a separate representation of deep convection Zhang–Mcfarlane (ZM) and shallow convection University of Washington (UW)—was replaced by a double‐plume (DP) scheme. This DP scheme adopts a quasi‐unified representation of shallow and deep convection within a single framework. Results demonstrate that the new scheme brings about several improvements in the modeled climate. The differences in the trigger and closure assumptions, lateral mixing rate, and cloud model for the deep convection result in systematic regional differences in the simulated precipitation pattern, cloud vertical structure, and the associated radiative forcing. Compared with ZM‐UW, DP reduces the biases in precipitation over the Indian Ocean, ameliorates the "high‐frequency and low‐intensity" problem of tropical precipitation, and leads to an improved representation of tropical variability, including the Madden–Julian Oscillation. Double‐plume reduces low clouds and increases high clouds in the tropics, due to its internal parallel‐split convective processes and smaller cumulus cloud fraction. Discussions related to parametric tuning of convective parameterization are also presented. Plain Language Summary: Cumulus convection plays an important role in most of the interactions among the processes in a General circulation model (GCM). In this work, two convective parameterizations were compared and evaluated within a GCM to understand the uncertainties brought about by the convective parameterizations. Various aspects of the modeled climate can be regulated by the different configurations of convective parameterization. The double‐plume scheme reduces the biases in precipitation over the Indian Ocean and improves the frequency–intensity structure of tropical precipitation, as well as an improved representation of tropical variability. Key Points: A double‐plume (DP) convective parameterization was evaluated and compared with the original scheme of Community Atmosphere Model, version 5 in the Global‐to‐Regional Integrated Forecast System modelThe DP scheme ameliorates the "high‐frequency and low‐intensity" problem in tropical precipitation and improves the representation of tropical variabilityDifferences in the trigger and closure assumptions, lateral mixing rate, and cloud model result in different model behaviors [ABSTRACT FROM AUTHOR]

A global model formulation for unified weather‐climate forecast is evaluated, with emphasis on the climate simulations at typical hydrostatic resolutions. The internal sensitivity is explored by considering different dynamical configurations (resolution, solver type, transport scheme). After a basic assessment of the global mean climate, a detailed analysis of precipitation characteristics is extended to East Asia. The model shows a reasonable mean state, seasonal variation, frequency–intensity structure, and diurnal phase time. The artificial rainfall around the steep slopes of the Tibetan Plateau can be improved through choices in the dynamical configuration. The regional features characterized by "afternoon versus nocturnal‐to‐early‐morning peaks" are properly distinguished. The hourly climatic features are comparable to super‐parameterized CAM5. Different dynamical configurations demonstrate unique sensitivities related to underling physical mechanisms, which are studied from the perspective of the diurnal cycle for three representative regions. Over South China, the higher‐resolution models decrease the weak‐precipitation while increase intense rainfall, thus reducing the dry biases. This is contributed by enhanced grid and sub‐grid scale motions associated with daytime convection progression. Over central western China, the variable‐resolution model better simulates the eastward propagating episodes characterized by a transition from convective to stratiform rainfall along the eastern slope of the Plateau. This reduces the positive biases at the high topography of the Plateau and alleviates the negative biases at the lower foot. Over central eastern China, the model replicates the dominant role of large‐scale governing factors in regulating the early morning rainfall peaks, and produces stratiform heating patterns. Plain Language Summary: We examine the statistical simulation of a model developed within the Global‐to‐Regional Integrated forecast SysTem, using Atmospheric Model Intercomparison Project style integration. This is among an effort in exploring a unified system for weather and climate modeling. Because of the evolutionary nature of model development, the absolute model performance can change. Thus, special attention is paid to the model sensitivity related to dynamical configurations, which helps to gain understanding that can support continuous research and development. Precipitation is one of the most important variables in a general circulation model as it encapsulates the synergistic effects of resolvable and non‐resolvable processes, the hydrological cycle, and radiative balance. Simulating precipitation faces the challenges in representing the moist physics and its interaction with dynamics. We use precipitation as performance metrics for a better understanding of the model behavior, focusing on a unique region–East Asia. The diurnal cycle is given particular emphasis because it reveals the direct physical mechanism of precipitation, and can help to disentangle the interactions between model dynamics and physics. We show that the models produce hourly rainfall features that are comparable to a global multiscale modeling framework that has two interactive dynamical solvers, and support some beneficial sensitivity. Key Points: To evaluate a model formulation for unified weather‐climate forecast using the Atmospheric Model Intercomparison Project‐style experimentTo understand model precipitation characteristics over East Asia by analyzing hourly scale apparent sources/sinks and grid‐scale tendenciesTo understand model sensitivity related to dynamical configurations, including grid resolution, moisture transport scheme and solver type [ABSTRACT FROM AUTHOR]

To understand the effect of the time step on the physics–dynamics interaction in a model, we used an idealized tropical cyclone test to evaluate the sensitivities to the physics time step in the Community Atmosphere Model Version 5 (CAM5). The investigated time steps were 450, 900 and 1800 s at a resolution of 1°, and 225, 450, 900 and 1800 s at a resolution of 0.25° in the corresponding ensemble simulations. We found that the intensity and precipitation of the simulated tropical cyclone and the physics parameterizations are fairly sensitive to the time step. These sensitivities are affected by the dynamical core and the physics–dynamics coupling strategy and vary with the horizontal resolution. In low-resolution runs, the intensity of the simulated tropical cyclone varies little with physics time step in the finite volume (FV) dynamical core, but it tends to weaken with decreasing time steps in the spectral element (SE) dynamical core. The horizontal circulation of the tropical cyclone in both the FV and SE simulations increases as the length of the time step decreases in high-resolution runs, where large-scale condensation dominates. The sensitivities in the physical parameterizations to time step play an important role in regulating the impact of time step on the physics–dynamics interaction, especially in high-resolution simulations. Compared with the sequential coupling approach (ftype1) with a sudden adjustment at each physics time step in the SE core, the dribbling coupling strategy (ftype0) that adjusts the state more gradually weakens the effect of the physical parameterizations. [ABSTRACT FROM AUTHOR]

A 3D dynamic core of the non-hydrostatic model GRAPES (Global/Regional Assimilation and Prediction System) is developed on the Yin-Yang grid to address the polar problem and to enhance the computational efficiency. Three-dimensional Coriolis forcing is introduced to the new core, and full representation of the Coriolis forcing makes it straightforward to share code between the Yin and Yang subdomains. Similar to that in the original GRAPES model, a semi-implicit semi-Lagrangian scheme is adopted for temporal integration and advection with additional arrangement for cross-boundary transport. Under a non-centered second-order temporal and spatial discretization, the dry nonhydrostatic frame is summarized as the solution of an elliptical problem. The resulting Helmholtz equation is solved with the Generalized Conjugate Residual solver in cooperation with the classic Schwarz method. Even though the coefficients of the equation are quite different from those in the original model, the computational procedure of the new core is just the same. The bi-cubic Lagrangian interpolation serves to provide Dirichlet-type boundary conditions with data transfer between the subdomains. The dry core is evaluated with several benchmark test cases, and all the tests display reasonable numerical stability and computing performance. Persistency of the balanced flow and development of both the mountain-induced Rossby wave and Rossby-Haurwitz wave confirms the appropriate installation of the 3D Coriolis terms in the semi-implicit semi-Lagrangian dynamic core on the Yin-Yang grid. [ABSTRACT FROM AUTHOR]