Atmosphere Model

AMWG Priorities and Near-Term Development

Below is a list of near-term (12-24 months) development activities planned for the CESM Atmosphere Model Working Group (AMWG) and the Community Atmosphere Model (CAM).

We encourage further contributions detailing AMWG development activities.

AMWG co-chairs

Rich Neale, NCAR
Mark Taylor, DOE-Sandia
Minghua Zhang, Stony-Brook, NY

A. Coupling across components and understanding interactions

Using CAM with the spectral element dynamical core (CAM-SE) is becoming more mature and we continue to advance testing within CAM5. Over the coming year testing will include long control coupled simulations to examine important coupled modes of variability including ENSO, sea-ice variability and other decadal modes of variability. Testing will be performed with both a 1-deg and 0.25 deg CAM-SE atmosphere and land coupled to 1-deg ocean and ice models. The interplay between cloud-aerosol interactions and climate sensitivity will also be analyzed in coupled 20th-century 1-deg CAM-SE simulations. In anticipation of different atmosphere-land interactions between CAM and CLM, AMIP simulations will be performed using CAM-SE and CLM4.5.

B. New Parameterizations/processes

The Unified Convection Scheme (UNICON) is a candidate for replacing both the current deep and shallow convection schemes in CAM5. UNICON continues to be developed and climate simulations show significant improvements. Over the next year it is expected to become more stable and tested in a more rigorous range of scenarios including coupled simulations and different vertical and horizontal resolutions. UNICON has been developed as a prototype scale-aware parameterization and is expected to demonstrate minimal sensitivity to resolution changes in CAM.

A version of CAM is being developed that uses a Cloud Resolving Model (CRM) embedded into each grid-box of the standard low resolution model grid. This super-parameterized version of CAM (CAM-SP) enables a more explicit representation of the currently parameterized physics by running the CRM at cloud resolving scales and then integrating the resulting prognostic fields back to the resolved CAM grid to be advected by the CAM dynamics. This code is currently available, but requires further development and compatibility testing before being made more formally available on the CAM code trunk.

Efforts are being applied towards PDF closure methods for the cloud-physics in CAM. The Cloud Layers Unified by Binomials (CLUBB, Bogenschutz et al. 2013) parameterization scheme is a candidate for replacing most or all of the current moist physics parameterizations in CAM5. It is based on an assumed PDF closure method. The plan is to move CAM-CLUBB to the CAM trunk and further test its fidelity in climate simulation replacing all of the moist physics, and replacing all of the moist physics except the deep convection. Mean climate simulations are performing well and we will continue to focus on variability metrics, performance timing, scalability and scale sensitivity, as well as the 'emergent' properties of CAM-CLUBB: feedbacks and indirect effects. Further activity using PDF cloud schemes will focus on sensitivity to PDF truncation and the inclusion of mixed-phase and ice cloud representations.

A new version of the Morrison-Gettelman (MG) cloud microphysics is in development. The main features are portable refactored code that is easy to modify, read and accelerate and modified sub-stepping with cloud condensation. Science advancements include prognostic precipitation, mixed phase ice nucleation and a revised ice cloud closure. We plan to integrate and test these developments over the next year. Efforts will also be made to compute vertical velocity variance from Turbulent Kinetic Energy (TKE) for use in aerosol activation and cloud macrophysics; a more consistent representation than the isotropic turbulence currently used in CAM5.

CAM5 currently uses the 3-mode version of the Modal Aerosol Model (MAM3). However, CAM5 with MAM3 has been shown to be significantly deficient in black carbon burdens at high latitudes. A four-mode version of MAM is currently under development that represents a primary carbon mode and improves the simulation of black carbon at high-latitudes. CAM with MAM4 will be examined in climate simulations to assess any changes to cloud-aerosol interactions and indirect effects. Current MAM efforts also include improved dust optics and aerosol convective transport and scavenging. A number of development activities are also planned with the more complex seven mode version (MAM7). These include the addition of a nuclei mode; ion-induced nucleation; stratospheric application of MAM for volcano and geoengineering studies; speciated dust optics, ice-nucleation, primary organic matter (POM) and primary marine aerosol; aerosol effects on shallow and deep convection; advance thermodynamics and nitrate aerosols; and a volatility basis set (VBS) model for secondary organic aerosol (SOA).

C. High-resolution and new dynamical cores

The focus for high horizontal resolution simulations using CAM5 continues to be a 0.25-deg (25-km) resolution configuration. This is intended to be a defined high-resolution version of CAM that will be rigorously tested for climate fidelity over the next year. Although a 0.125-deg (12.5-deg) version of CAM5-SE has been integrated in limited length simulations it is felt that the computational, storage and processing overheads associated with doing climate length integrations would inhibit progress significantly at this point in time.

The number of vertical levels in CAM has been 30 or less for the last 10 years. In order to maintain a reasonable aspect ratio with 0.25 deg configurations we intend to significantly increase the number of vertical levels. Currently we have been investigating 60 levels in CAM and intend to further investigate the climate response to different configurations of the added levels, with more levels in the boundary layer and an elevated model top from 2 mb to 0.2mb to resolve the stratopause and lower mesosphere. We plan to work closely with the WAWG and CCWG to develop complementary model versions aimed at simulating the middle atmosphere.

A regionally refined version of CAM-SE enables high-resolution (0.25 deg) to be targeted at a specific geographic region while maintaining low-resolution (1 deg) elsewhere, all within the context of a global simulation. Initial simulations have, as expected, shown significant sensitivity of the current parameterized physics to this regional refinement. Development work over the next year is aimed at further diagnosing these sensitivities both in idealized aqua-planet and AMIP type simulations, in order to identify the most sensitive physics parameterizations. Furthermore, initial efforts will begin to reduce these sensitivities by enabling existing and development parameterizations to be more scale-aware than they currently are. Specific code improvements include improved hyperviscosity treatment with a better CFL restriction requiring less tuning of the coefficients and the development of an open source unstructured mesh generation package (replacing the commercial CUBIT) with improved mesh quality metrics for spherical geometry.

The Conservative Semi-LAgrangian Multi-tracer scheme (CSLAM) transport algorithm has been implemented in an offline configuration of the spectral element (SE) core. CSLAM will enable accurate and efficient transport of multiple tracer species; a major requirement for complex chemistry-climate applications using CAM, CAM-chem and WACCM. It remains to couple CSLAM with SE dynamics which may require a switch to the flux-form formulation of CSLAM. The CSLAM transport scheme operates on an equiangular gnonomic grid that is more isotropic than the SE grid. Work will focus on adding the capability to run CAM physics on an equiangular gnonomic grid that is either finer, coarser or co-insides with the CSLAM grid. CSLAM will also be extended to support unstructured variable resolution grids.

Further development of the SE core will include fully implicit and linearly implicit time stepping options in addition to the inclusion of condensate loading effects on surface pressure and pressure gradient force which has potentially significant effects at resolutions of 25 km and finer. In CAM-FV a specified dynamics option is being heavily used for chemistry applications and it is planned to include a similar functionality in CAM-SE. Work b eing done in the SE core, which can be used in CAM, includes further development of the discontinuous Galerken dynamical core and the development of non-hydrostatic options for both spectral elements and discontinuous Galerken. Additional development efforts also continue with the Model for Prediction Across Scales (MPAS) dynamical core; a non-hydrostatic dynamical core using CAM5 physics. Non-hydrostatic capability will be required for CAM resolutions of 10 km and finer.

D. Addressing biases/shortcomings

To a certain extent the proposed new and updated parameterization schemes are intended to address many of the shortcomings of the CAM climate. For example, the inclusion of UNICON in CAM has demonstrated success in improving the amplitude and phase of the diurnal cycle of precipitation over land as well as the Madden Julian Oscillation (MJO); both long-standing biases in the model and crucial for capturing the observed high-frequency behavior in CAM. Inclusion of the development version ofMG microphysics is expected to improve the representation of mixed-phase clouds at high latitudes. Efforts will also focus on providing a more rigorous energy conservation formulation for CAM5.

A persistent bias that remains is the double ITCZ and strong northern ITCZ in CAM5 which is exacerbated at high resolution and upon coupling to an interactive ocean. Efforts to assess and reduce these errors are planned for the coming year. Specific tools to address these and other biases will include the evolution of errors in Cloud Associated Parameterization Tested (CAPT) initialized simulations as well as in nudged simulations. The CAPT analysis shows the evolution of errors over a period of a few days and nudging shows how the parameterized physics respond to the correct large scale dynamical forcing.

Research has shown increasing evidence that the response of the physical parameterizations in the model is ordering and timestep dependent. Research into these sensitivities aims to determine the largest contributors and identify potential methodologies for reducing the impact on CAM simulations.

The Data Assimilation Research Testbed (DART) continues to be an invaluable tool for ensemble data assimilation within CAM. It is used to provide atmospheric state nudging increments that are required to push the model to be more consistent with observations. Over the coming year integrating DART into fully coupled CESM will be a top priority. This enables multi-component (each component assimilates its own observations) and cross-component (an individual component assimilates across other components) assimilation. This will be a further method with which to diagnose fully coupled model biases and efficiently initialize simulations for decadal prediction. The ability to use DART in CAM-SE with refined resolution grids will also be developed. We plan to establish a standardized diagnostic package using DART to test development versions of CAM.

In order to more objectively diagnose CAM systematic biases three sets of Uncertainty Quantification (UQ) tasks are being undertaken in the near term. The first set utilizes so-called perfect model calibration tests to assess the effects of mild to severe structural deficiencies that are intentionally added to model-generated observational targets. Perfect model calibration tests provide a finely-controlled testbed for analyzing structural deficiencies because the answer is known in advance. The second set involves statistical significance tests and other calculated measures of discrepancy between distributions to determine the degree to which atmospheric phenomena are tunable and need structural changes. Given an ensemble of perturbed parameter CAM simulations and observations with uncertainties, the statistical tests and discrepancy measures provide a simple framework to detect structural errors. A final set of UQ tasks is directed toward characterizing model discrepancies, with the goal of better assessing the potential to quantify and apply post-bias corrections.

E. Software development

CAM continues to rapidly evolve its software infrastructure. Much of the proposed future developments are meant to facilitate as wide a range of parameterization schemes and dynamical cores as is feasible by generalizing interface software that point to code externals. This is necessary since CAM is currently in a mid-release period where there are many possible candidates for a future CAM6 and the demand for flexibility to have a wide range of parameterizations interacting with dynamical cores is ever increasing. This flexibility is also being extended to backwards compatibility of resolutions in order to provide configurations for researchers with more limited computing resources.

One specific planned development that more completely facilitates the use of finer resolution dependent parameterizations (e.g., CAM-CLUBB and CAM-SP) is the implementation of atmospheric sub-columns. This development essentially provides the functionality to specify an arbitrary decomposition of the resolved grid-scale into a number of sub-column representations of the prognostics quantities (temperature, humidity etc.) for use by a specific parameterization. We expect a variety of different efforts to use this infrastructure over the next year.

Software development is also targeted at enabling CAM to exploit current and future computing architectures to the greatest extent possible. Currently this is focused on enabling the CAM-SE dynamical core and a more advanced version of the Rapid Radiative Transfer Model (RRTM) radiation scheme to take full advantage of Graphics Processing Units (GPU). The focus is on RRTM and the SE core since they are the most costly components of CAM and have the greatest potential for exploiting GPUs. Development over the next year is expected to be with RRTM and the SE core in isolation, but integration of the architecture into the full CAM is expected soon thereafter.