Modeling

Our NGEE Arctic modeling and model-scaling approach is founded on a simple concept that distinguishes between models at different scales on the basis of which processes are treated explicitly, which processes are treated as sub-grid parameterizations, and the model boundary condition requirements. This approach also provides a natural point of integration for new knowledge emerging from measurements, observations, and experimentation at multiple spatial and temporal scales. The ability to perform up-scaling depends on explicit representation of some process or processes at a finer scale, which are treated implicitly, or as sub-grid parameterizations, at a coarser scale. The ability to perform useful down-scaling depends on the existence of boundary conditions in a finer-scale model that can be assigned using outputs from a coarser-scale model.

P.Thornton (ORNL) will lead this team.

To help clarify our model-scaling approach, we summarize the basic steps for the simple case of two models at different scales: a climate-prediction scale model, and a higher resolution process-resolving model (Figure 8). Initial model development is required for both modeling scales to accommodate the information passing and parameterization steps required by the scaling framework. As models are being developed in parallel, early model testing can begin without communication between them. A first scaling step is the up-scale coupling of information from the fine-scale model to inform parameterizations (set parameter values) in the climate-scale model. A second scaling step is to run the climate-scale model with new parameters and pass boundary constraints to the process-resolving model. This leads to subsequent simulations at fine and coarse scales in an iterative process through which final parameter estimates for the coarse scale model are expected to converge to their optimal values. The model-development and model-application phases shown here also represent periods during which integration of new process knowledge from observation and experimentation will occur.

Our scaling framework engages models at three spatial scales: the climate-grid scale, an intermediate scale, and a fine scale, with larger domain and coarser grid resolution for the climate-scale model and smaller domains and finer grid resolution for the intermediate- and fine-scale models. Observations, experimental results, and new process understanding will be integrated with modeling components at scales appropriate to the measurements and mechanisms. Our scaling approach assumes that the finer-scale modeling domains are nested within the larger-scale domains, and that these domains are centered on the NGEE Arctic study sites. Our initial implementation for NGEE Arctic Phase 1 will have a climate-scale domain on the order of 50 to 100 km (x and y horizontal dimensions), with grid resolution of approximately 10 × 10 km. Our intermediate model domains will consist of single climate-scale model grid cells with resolution of approximately 100 × 100 m, with the potential for several intermediate-scale domains within the climate domain. Our fine=-scale model domains will consist of single intermediate-scale model grid cells with resolution of approximately 1 × 1 m. We expect to have multiple fine-scale model domains represented within the each intermediate domain, to assess process sensitivity to variability in local topographic setting. In addition, at the fine-scale and intermediate-scale domain sizes, we will use both atmospheric models and subsurface modeling tools to characterize and quantify the coupled processes controlling the fluxes of mass species and energy.

Our climate-scale model will be derived from the current CLM4, and will share all of CLM4’s explicit process representations, including one-dimensional (1D) (column-based) mass and energy balance, permafrost structure, and active layer dynamics. New development will add thermokarst-driven changes in surface elevation as an explicit 1D process. Mass balance equations currently include explicit carbon and nitrogen biogeochemistry, and new plant functional types representing real Arctic vegetation will be introduced. The model will include an explicit representation of surface flow in a static drainage network, replacing RTM with a more sophisticated approach that maintains hydrologic connectivity information at the sub-grid scale (Section III.2.4). Sub-grid parameterized processes added to the model to accommodate the NGEE Arctic scaling approach will include overland flow across sub-gridunits and transport to the static drainage network as well as surface and subsurface flow between landunits, including dissolved-phase biogeochemistry and sediment transport. The distribution and dynamics of PFTs are handled as sub-grid parameterizations in the current CLM4, and this functionality will be retained. Shifts in sub-grid area from one type to another (e.g., polygonal ground to thaw lake, or thaw lake to DTLB) will be added as a new sub-grid process. Boundary conditions for the climate-scale model include near-surface weather and the imposed structure of the static drainage network and large-scale topography.

Our intermediate-scale subsurface model will be based on an existing three-dimensional (3D) `reactive transport model architecture [either PFLOTRAN or the Advanced Simulation Capabilities for Environmental Management (ASCEM) and Amanzi], using 3D finite volume computational methods for subsurface processes, and two-dimensional (2D) surface mesh representations for overland flow. All processes treated explicitly in the climate-scale model are also treated explicitly in the intermediate-scale model. Additional explicit processes include thaw lake dynamics, dynamic drainage network organization, and surface and subsurface lateral flow. The subsurface thermal hydrology processes will be represented at a level of detail appropriate for the spatial scale of interest and may thus consider approximate representations and sub-grid parameterization for some processes. Processes represented through sub-grid parameterization include polygonal ground and ice wedge dynamics as well as PFT distributions. Boundary conditions for the intermediate-scale model include near-surface weather and surface and subsurface water and energy inflow at domain boundaries. Regardless of the architecture selected, a substantial amount of new model development and integration of existing model components is required to arrive at a fully functional intermediate-scale model. Further details on model technical requirements and computational components will be developed early in Year 1 of the project.

The High-Gradient Applications Model (HIGRAD), an atmospheric hydrodynamics model, will be used to study the near-surface atmospheric dynamics at fine spatial scales. A critical aspect of HIGRAD is its capability to model the interaction between atmospheric flows and heterogeneous vegetation and topography, including aerodynamic drag and turbulence, energy exchange, and gas species transport. A variety of HIGRAD horizontal resolutions ranging from 0.1 × 0.1 m to 100 × 100 m will be used with appropriately scaled vertical resolutions in order to adequately understand the effects of resolving various atmospheric motions and landscape elements and to verify that sub-grid models of interaction with heterogeneities in topography and vegetation are capturing critical behaviors.

The processes treated explicitly for the intermediate-scale model will also be treated explicitly in the fine-scale model. However, a more mechanistic representation (Painter 2011) of thermal hydrology of freezing soil will be used. Additional explicit representations will include ice wedge polygons and dynamic microtopography. The need to model evolving microtopography associated with degrading ice wedges places considerably more demand on the computational meshing infrastructure than in the intermediate-scale case, requiring fully unstructured grids that can conform to features of interest such as material interfaces and, depending on the approach used to track material properties as deformation occurs (e.g., Lagrangian meshes vs Eulerian meshes across which material properties are advected), mesh movement and remeshing. Boundary conditions are formulated in the same way as for the intermediate-scale model. Finally, we expect that even for the fine-scale model we will not be able to resolve the explicit dynamics of individual plants as they grow, reproduce, and die, and so PFT dynamics will continue to be represented as a sub-grid parameterized process at this scale.

No existing highly parallel codes implement a freezing soil thermal hydrology model. Moreover, there are no existing codes that address the interactions among subsurface thermal hydrology of freezing/thawing soils, overland flow, and topographic evolution caused by mechanical deformation processes, which are the key process couplings at the fine scale. The proposed path forward is to evaluate the parallel subsurface flow/transport codes PFLOTRAN and Amanzi/ASCEM and select one of these codes, or a combination of the two, as the computational platform for the fine- and intermediate-scale modeling. This selection decision will be made by October 2012.

Given these three modeling scales, and following the general approach shown in Figure 8, we propose a more detailed research plan relating model development, parameterization, application, up-scaling, down-scaling, and integrative analysis across scales. Figure 9 provides a schematic for our more detailed plan, and describes dependencies among tasks at multiple modeling scales. The “time” axis can be roughly interpreted to cover all of NGEE Arctic Phase 1 and extending into Phase 2 as the iterative scaling process develops. Specific milestones and associated timelines are provided for each task in the following sections. The gray “Analysis” blocks represent the many points of integration with existing and newly collected measurements and observations from field and laboratory research. The orange “Initial Parameterization” blocks are additional critical points of integration between models and observations, where multiscale characterizations of the research sites from landscape-scale distributions of fluxes and geomorphological features down to the measured properties of individual permafrost cores or plant communities will be used to constrain models. In the following sections we describe the specific modeling tasks required to construct our scaling framework, apply that framework to guide observational and experimental strategies, and integrate new knowledge emerging from process studies in the hydrology/geomorphology, biogeochemistry, and vegetation dynamics areas. For clarity, we have organized these modeling tasks by scale, but these tasks are closely related across scales, as described above (see Appendix XIV.1).

Climate-Scale Modeling Tasks

A sophisticated sub-grid scaling framework already exists within the CLM4 architecture; however, significant modifications are required to accommodate the current best understanding of Arctic tundra processes and to accommodate connections to intermediate and fine-scale models with more explicit process representation. The following tasks define CLM4 development, application, and evaluation, which will lead us toward our goal of a process-rich ecosystem model in which the evolution of Arctic tundra can be modeled in the context of the coupled Earth System.

Baseline CLM4 simulations. We will conduct a series of “baseline” simulations using the current operational CLM4 code, applied at high spatial resolution (10 × 10 km) over a region of approximately 100 × 100 km around the Barrow Environmental Observatory (BEO). These simulations will serve as a point of comparison as modifications are made, first to the CLM data structures and process representations, and later to CLM parameterizations as the result of up-scaling and down-scaling iterations. The series of simulations will include spinup to a pre-industrial steady-state and a transient simulation from 1850 to present with observed climate system forcings. We will also include a simple set of modeled manipulation experiments: enhanced CO₂, warming, and drought. The purpose of these simulated manipulations is to evaluate the influence of new model dynamics on predicted experimental outcomes, as a rapid feedback mechanism to observational and experimental groups.

Structural modifications to CLM4 required for NGEE scaling approach.

Implement multiple vegetated landunits. The NGEE Arctic scaling approach requires that CLM sub-grid elements be represented with explicit knowledge of spatial location and hydrologic setting, a departure from the current approach in CLM4, which ignores location within the grid cell and carries only fractional area information. The NGEE Arctic sub-grid treatment uses hydrologically distinct sub-basins to define landunits. This task generates the preprocessing code that defines the new landunits from sub-grid resolution DEMs, and modifies the CLM surface dataset file to carry new information on landunit geography and its relationship to hydrologic network topology.

Implement improved surface and subsurface flow processes. River-routing information in CLM4 is currently treated on a 0.5° × 0.5° grid, which is too coarse to accommodate the integration with intermediate- and fine-scale model outputs for the NGEE Arctic scaling approach. This task generates surface flow networks from digital elevation data at 50 m or finer resolution and provides network topology information that is incorporated in surface datasets to describe the hydrologic connections between landunits. One specific sub-task of CM2.2 is the introduction of a hydrologic head–based formulation for subsurface hydrology, replacing the current volumetric water content–based scheme in CLM4. Subsurface processes in CLM4 are treated as 1D (only gravity driven), which is a major limitation for NGEE Arctic modeling. Thus, an additional sub-task is development of new code to allow lateral flows between landunits. Since lateral flows will be largely determined by sub-landunit topography and subsurface frost table elevations associated with polygonal ground and DTLB depressions, response functions for sub-landunit lateral flows will be developed to generate partitioning of snowmelt and rainfall between on-site inundation and runoff available for lateral transfer between CLM landunits. This task is linked to the Hydro-Geomorphology tasks.

Implement multiple vegetated soil columns per landunit. The current CLM sub-grid scheme defines only a single vegetated soil column with a single landunit. The NGEE Arctic scaling approach requires multiple soil columns to coexist within a single landunit so that the full range of geomorphological types that are observed to occupy relatively small tundra regions can be represented. A primary mechanistic connection between climate-scale and intermediate- and fine-scale modeling will be the representation of a dynamic surface elevation for each soil column, which will be used to represent the influence of permafrost degradation on surface topography and associated changes in hydrologic state. Introduction of new code allowing dynamic shifts in soil column area, which is currently implemented only for the plant functional type level of the CLM4 sub-grid hierarchy. These modifications to the column sub-grid level in CLM are necessary first steps that allow subsequent improvements in treatment of Arctic tundra biophysics (e.g., snow pack and snow albedo effects, surface temperature, and permafrost dynamics) and biogeochemistry.

Implement realistic Arctic plant functional types. Current arctic vegetation in CLM4 is represented by a small set of PFTs, with only bare ground, shrubs, and an “arctic grass” representing the complexity of arctic ecosystems. We will build on the existing CLM4 structures to add new herbaceous, woody, and bryophyte PFTs. This approach will involve more realistic representation of physiology, nutrient interactions (e.g., N fixation, organic N uptake), photosynthetic controls, subsurface interactions between roots and permafrost, and radiation/energy balance. We will begin exploring the plant competitive processes and environmental thresholds responsible for determining PFT distributions under a changing climate.

Implement realistic Arctic biogeochemistry on soil column. For climate-scale biogeochemical modeling in NGEE Arctic which is slated for release in December 2012. That model version will include vertically resolved belowground C and N pools; simplified representations of nitrification, denitrification, and leaching; and a CH₄ submodel. We will work with the modeling teams at the fine and intermediate scales and the biogeochemistry observations to improve climate-scale model representations of microbial community composition and function; aerobic and anaerobic C decomposition and carbon use efficiency; N mineralization, nitrification, and denitrification; plant inorganic and organic N uptake; exudation; mineral N competition between consumers; sorption impacts on C and N availability; impact of freezing on CH₄ ebullition; CH₄ oxidation; and vertical species transport.

Incremental CLM simulations with NGEE Arctic modifications. As new model versions are available from each of the climate-scale model development tasks, we will rerun the series of preindustrial, historical transient and simple manipulation simulations, and evaluate the incremental influence of new model structure and new process representation on predictions of thermal and hydrologic dynamics, vegetation dynamics, and biogeochemical dynamics.

Preliminary CLM scaling study—hydrologic routing. As a first demonstration of how the NGEE Arctic scaling framework interacts with the modified CLM sub-grid hydrologic routing scheme, we will evaluate the default CLM hydrologic outflow estimates against measurements of streamflow made at multiple locations within the BEO (observations described in Hydro-Geomorphology). We will then repeat these evaluations using the explicit sub-grid hydrologic routing framework introduced. We hypothesize that the explicit sub-grid treatment of hydrologically distinct landunits in a topologically connected network will lead to improved prediction of mean streamflow and seasonal variation, compared to the implicit routing scheme in the default CLM.

Preliminary CLM scaling study—soil saturation state. As an early demonstration of how the NGEE Arctic scaling framework interacts with modified CLM soil column representation, we will evaluate the default CLM water table height and soil saturation state against site level measurements taken in the BEO (observations described in Hydro-Geomorphology). We will then repeat these evaluations using the multiple soil column implementation with dynamic surface height, introduced. We hypothesize that the explicit surface height and multiple soil column approach will improve prediction accuracy for depth to water table, soil saturation state above the water table, and estimates of fractional inundated area at the grid cell level.

Preliminary CLM scaling study—soil biogeochemistry. We will apply new climate-scale biogeochemical representations within the modified CLM4 structure We will evaluate the impact of new CLM sub-grid representation on soil biogeochemistry dynamics, quantifying the effects of representing landscape units at finer spatial scales within CLM. We will assess the influence of new surface and subsurface hydrologic routing schemes on net fluxes and transport of biogeochemical species. We will examine the interactions of new soil biogeochemistry representations with new Arctic plant functional types.

Preliminary CLM scaling study—vegetation dynamics. We will demonstrate the influence of new sub-grid hydrologic routing and soil column hydrology representations  on vegetation transitions as predicted by new Arctic-specific PFT representations, including interactions with nutrient dynamics as predicted by a more mechanistic representation of soil biogeochemistry. We will evaluate predictions of vegetation structure and function from the default CLM against observations across multiple geomorphological types within the BEO. We hypothesize that the new PFTs and competitive process representations, coupled with improved hydrology and biogeochemistry boundary conditions, will result in improved predictions of vegetation structure and community composition when compared with the default CLM.

CLM parameterization from finer-scale models (up-scaling). We will use results generated by intermediate-scale modeling over subsets of the BEO as training targets in the development of new soil-column and landunit parameterizations for CLM. Specifically, we will use intermediate-scale model predictions of surface height, water table depth, saturation states, and hydrologic outflow as data assimilation targets to define parameter settings in the soil column and landunit levels of the modified CLM sub-grid hierarchy. We expect that changes in surface elevation can be related empirically to changes in temperature, moisture content, and associated phase changes by depth in the soil column and that fractional saturated area, water table depth, and hydrologic outflow can be related in turn to changes in surface height. This hierarchical approach to parameter estimation will be supported by established data assimilation methodologies. New up-scaled parameter settings will be used to generate full CLM simulations over the BEO, and results will be compared with integrative observations to evaluate changes in prediction accuracy at the climate scale.

CLM boundary conditions passed to finer-scale models (down-scaling). We will use the up-scaling results to provide updated boundary conditions to the intermediate scale model over subsets of the BEO, which will in turn provide more finely resolved boundary conditions to the fine-scale model over smaller subsets of the BEO spatial domain.  Specifically, we will pass hydrologic outflow as predicted by the explicit sub-grid hydrologic routing component of CLM as boundary forcing input to the intermediate model surface hydrology. We will also pass water table depth, soil water saturation state, and soil temperature from the CLM landunit level as additional intermediate-scale boundary conditions.

Intermediate- and Fine-Scale Modeling Requirements

Our intermediate- and fine-scale modeling efforts will benefit from ongoing efforts that have generated flexible and extensible surface and subsurface flow and reactive transport models. As an early NGEE Arctic effort, we will evaluate the capabilities of two such modeling systems (Amanzi and PFLOTRAN), for the purpose of selecting a framework for the further development of NGEE Arctic–specific functionality at intermediate and fine scales. It is clear from the outset that neither of these existing frameworks meets all of the NGEE Arctic scaling framework needs. Our approach is to identify the most appropriate elements from each approach and to move forward with an NGEE Arctic–specific development effort that draws these elements together and introduces new capabilities as needed.

Amanzi (Moulton et al. 2012) is an open-source, modular, object-oriented simulator written in C++ and based on the Trilinos parallel toolkit. Amanzi is under development as part of the DOE Environmental Management-supported ASCEM project. Amanzi accommodates fully unstructured grids, includes computational mesh updating toolkits to allow for dynamic meshes and has a sophisticated multiphysics coordinator to address the coupling among different processes.

PFLOTRAN (Mills et al. 2009) is an open-source (GNU lesser general public licensed) code for simulation of multiscale, multiphase, multicomponent flow and reactive transport phenomena in porous media on computers ranging from laptops to leadership-scale supercomputers. It is a modular, object-oriented code (mostly written in modern Fortran 95/2003) built on top of the Portable, Extensible Toolkit for Scientific Computation (PETSc) parallel framework, which provides access to cutting-edge scalable solvers and infrastructure for “composable” multiphysics simulations. It supports a comprehensive suite of biogeochemical reactions, implicit or operator-split time stepping, finite-volume and mimetic finite-difference discretizations, structured adaptive mesh refinement (experimental), structured and unstructured grids, and parallel input/output.

The following capabilities are required of any computation infrastructure that will address both fine- and intermediate-scale: adequate parallel performance, modularity, extensibility, flexibility in process coupling, flexibility in gridding, and dynamic mesh updating.

Parallel performance: Fine-scale simulations resolving scales of individual ice wedge polygons and related vegetation heterogeneities and spanning domains of sizes 100 × 100 × 10 m are estimated, for example, to require approximately 50 million grid cells for subsurface calculations, assuming optimized unstructured grids. This is well within the capability of simple hydrologic codes built on the Trilinos or PETSc parallel frameworks. PFLOTRAN exhibits strong parallel scaling to hundreds of thousands of cores in simulations involving complex reactive chemistry. The Amanzi code exhibits strong parallel scaling to thousands of cores, the limit of its current testing. Based on performance of other Trilinos-based codes, we expect that it will be possible to achieve acceptable scaling in Amanzi to at least tens of thousands of cores. However, performance and required resources for the significantly more complex simulations that include the key couplings required for an Arctic simulator is uncertain for both PFLOTRAN and Amanzi. These uncertainties come from several sources: the possible need for mesh smoothing and associated conservative remapping of the unknown variables, which likely induces a large computational burden (discussed below); uncertainty in the level of detail required in representations of hydrologic and biogeochemical processes; and the novelty of coupled simulations of overland flow, subsurface reactive transport, and freezing and thawing in geologic media.

Modularity and Extensibility: Modularity and extensibility are design goals for both PFLOTRAN and Amanzi. Amanzi takes advantage of C++ language features and is fully object oriented and thus highly extensible. PFLOTRAN is largely written in modern Fortran (with some C++ interface code) and, through judicious use of features from Fortran 95/2003, follows an object-oriented paradigm. It is built on top of the highly object-oriented PETSc parallel framework (Balay et al. 2011).

Flexibility in process couplings: The fine-scale and intermediate-scale models will couple several different physical, chemical and ecological processes. A range of computational strategies is available for enforcing the couplings, depending on the strength and time scales of interactions. Clearly, fully implicit global coupling among all the processes is not needed. However, significant numerical experimentation will be required to determine optimal coupling strategies, and it is likely that mixed strategies for time stepping (implicit to treat stiff components and explicit to treat non-stiff components) will need to be employed. Thus, flexibility in process coupling is a key requirement for the simulator. In Amanzi, such flexibility is provided through the use of a multiprocess coordinator, which controls the execution of individual process kernels and their interactions. This capability allows coupling between process kernels to be modified with little or no code changes. In PFLOTRAN, this flexibility can be provided using the composable multiphysics infrastructure that has been developed in PETSc over the last few years.

Flexibility in gridding: Amanzi is fully unstructured and is based on advanced discretization methods specifically designed for computational cells based on polyhedra of any order. PFLOTRAN originally supported only static, structured grids, but now offers some support for structured adaptive mesh refinement (SAMR) using the SAMR infrastructure framework as well as a newly implemented unstructured grid capability that is currently being tested and is expected to be performing well at scale before September 2012. The unstructured grid capability in PFLOTRAN supports a mixture of distorted low-order polyhedral elements (prisms, tetrahedra, pyramids, hexahedra, and combinations of these types).

Evolving surface and subsurface geometries: The capability of the NGEE Arctic landscape simulator to adaptively track an evolving surface topography associated with thermokarst formation is a critical piece of computational infrastructure. This can be done using Lagrangian or Eulerian approaches. In a Lagrangian approach, the computational mesh follows material deformations. For small displacements of the surface elevation caused by thawing, simple Lagrangian mesh motion calculations would be adequate. However, for the relatively large displacements associated with degrading ice wedges, mesh motion will quickly lead to mesh entanglement and invalid meshes. Thus, mesh smoothing—the movement of mesh nodes independently of mechanical displacements to maintain a mesh of sufficient quality—is required. Mesh smoothing must be paired with a remapping of the solution-state variables from the old to the new mesh. Such remappings should conserve water mass and total heat content. Access to multiple mesh toolkits is a primary design requirement for Amanzi and full capability for mesh movement. Smoothing and conservative remapping are anticipated in Amanzi or derivative products by September 2012. Thus far in its development, PFLOTRAN has not targeted problems in which mesh motions are required. However, one avenue for supporting complicated mesh operations in PFLOTRAN is to use the Sieve mesh infrastructure in PETSc, an abstraction that allows complicated mesh operations to be carried out with relative ease. Sieve has been used successfully in the tectonics code PyLith (Aagaard et al. 2011) to represent complicated changes in mesh geometry and topology associated with processes such as fault development.

We note that, precisely because of the difficulties associated with large mesh movements and remeshing, Eulerian approaches—in which the computational grid remains fixed in space while deforming material flows through it—are sometimes preferred for problems involving large deformations, despite reduced accuracy in tracking material interfaces and the computational cost of tracking material movement within the grid. It may be worthwhile to explore such approaches in the NGEE Arctic intermediate- and fine-scale models.

Intermediate-Scale Model Development and Tasks

The following key processes are required for the intermediate-scale model: lateral water flow on the surface and in the subsurface, reactive transport of solutes, subsurface heat transport, surface energy balance processes, soil biogeochemistry, and plant dynamics. The intended spatial resolution for the subsurface model is approximately 100 × 100 m in the horizontal direction, with resolutions ranging from 10 × 10 to 100 × 100 m for the atmospheric process model. Processes represented through sub-grid parameterization include polygonal ground and ice wedge dynamics as well as PFT distributions.

The intermediate-scale simulator differs from the fine-scale simulator in three important aspects. First, the effect of microtopography associated with polygonal ground (and vegetation heterogeneity in the case of the near-surface atmospheric calculations) will be a sub-grid parameterization and not represented directly. Second, with topography averaged to the 100 m scale, it is not necessary to represent the full complexity of topographic evolution. Thus, the computationally difficult issue of an adaptive mesh may be avoided at the intermediate scale. Third, it is not necessary to represent the full complexity of the thermal hydrology of freezing/thawing soils or fine-scale turbulence at the intermediate scale. In particular, more approximate and computationally tractable models may be used to represent lateral subsurface thermal flows and effective mixing.

Development of intermediate-scale model requirements. Detailed specifications of the mathematical model and algorithmic approaches appropriate at the intermediate scale will be developed. Simple vertical subsidence calculations and associated small-displacement mesh motion may be required to represent thaw subsidence, but the more difficult issues of mesh smoothing and mesh remapping are avoided at the intermediate scale. The requirements document will specify the model processes to be represented as well as algorithm options and required computational infrastructure. Research including prototype development is required to identify appropriate approximations and algorithms for two key processes: representations of subsurface lateral flow and representations of thaw lakes. Variants of a nonlinear Boussinesq equation with a moving lower boundary will be explored as a way of routing water laterally above the seasonally varying frost table. Thaw lakes are not adequately represented by shallow overland flow representations and research is needed to identify an appropriate representation.

Work plans for Amanzi and PFLOTRAN development. Work plans describing the effort required to produce the intermediate-scale simulator from Amanzi and PFLOTRAN will be developed. These two work plans will support the decision on foundational architecture for the intermediate scale simulator. Each plan will describe the code modifications required and a plan for testing those modifications. Estimates of level of effort required to complete the work will also be provided. A workshop on the requirements for the intermediate-scale subsurface simulator and work plans for its development will be held by September 2012. Potential users of the community code for intermediate-scale simulation will be invited. In addition, external computational specialists will be asked to participate to provide feedback on the proposed approaches.

Decision on the intermediate-scale simulator. A decision on foundational architecture for the intermediate-scale subsurface simulator will be made by October 2012. The decision will be based on an objective evaluation of the existing capabilities of the Amanzi and PFLOTRAN models and their potential for extension, as referenced against the requirements described under Tasks IM1 and IM2. Decisions will be made by consensus of the modeling team. Possible outcomes include adoption of one or the other modeling framework as the foundational architecture for the NGEE intermediate-scale simulator, or the selection of appropriate components from each framework as merged foundational elements for new development.

Construction of the intermediate-scale simulator. Work plans will be executed, following model requirements as modified by workshop input. Intermediate-scale model development will begin prior to the decision point on foundational architecture, as many model components and algorithms can be developed as modules with well-defined interfaces. Upon completion of Version 1, the code developers will conduct a tutorial workshop addressing code use, the underlying technical basis for each process representation, and preliminary results for test problems.

Intermediate-scale simulation test cases. A series of simulation cases will be developed, including simple verification/validation tests that exercise individual process models, synthetic but realistic cases for testing model couplings, and site-specific cases that exercise the full capabilities of the simulator. These cases will be executed concurrently with development of Version 1 of the intermediate-scale simulator.

Near-surface atmosphere simulations. We will begin a series of intermediate scale-simulations ranging from 10 × 10 m resolution to 100 × 100 m horizontal resolution, with appropriately scaled vertical resolutions, characterizing the implications of reduced resolution for mixing processes. These simulations will assist in the development of appropriate up-scaling parameterizations for the connection between subsurface and climate-scale atmospheric flows, providing guidance in quantifying up-scaling uncertainties.

Development and testing of an intermediate-scale biogeochemistry framework. We will develop a computationally efficient solution for the complex biogeochemistry of Arctic tundra by generalizing relationships between biogeochemical cycles and environmental variables. The modeling will be performed in the intermediate-scale simulator framework emerging. We will begin by integrating the intermediate-scale surface characterization; hydrological and thermal sub-models; and vegetation sub-models with a subsurface biogeochemistry submodel. Simulations will be performed for individual geomorphological units (e.g., low-centered, transitional, and high-centered polygons) found across drained lake basins in the BEO. Model predictions will be compared to measurements of surface trace-gas fluxes, subsurface C and N concentrations, and NO_3– concentrations in stream water and lakes.

Initialization of intermediate-scale model domains. This task will use the output from landscape characterization and hydro-geomorphology tasks associated with characterization and classification of drainage network, hydraulic geometry and connectivity, surface and frost table topography, and storage capacity to define the topographic, thermal and permafrost hydrology components of several intermediate-scale simulation domains over the BEO. This effort includes the explicit characterization of low centered, transitional and high centered polygonal ground found across drained lake basins of different ages as well as interstitial terrain. Results from this task also provide constraints on climate-scale sub-grid parameterization of landunits and soil columns, ensuring consistency in landscape characterization across scales, which is crucial to unbiased up-scaling and down-scaling.

Intermediate-scale parameter estimation and up-scaling simulations. We will use results from fine-scale model simulations, generated over subsets of the BEO as training targets in the parameterization of polygonal ground and ice wedge dynamics in the intermediate-scale model. These processes will be represented explicitly in the fine-scale model, but empirically in the intermediate-scale model. We will employ formal data assimilation methods to determine optimal parameter values, capturing as much of the fine-scale model variability as possible, given constraints imposed by uncertainty in fine-scale model results and observation-based a priori estimates for intermediate-scale parameters. Vegetation dynamics as predicted by the fine-scale model will also be used to parameterize rates of change in PFT distributions as functions of mean thermal, hydrologic, and biogeochemical states. After estimating optimal parameters based on up-scaling, we will perform intermediate-scale simulations over subsets of the BEO. These results serve as input and will also be subjected to evaluation against integrative observations.

Intermediate-scale boundary constraints and down-scaling simulations. The intermediate-scale model integration will be repeated with updated boundary conditions generated under task CM7, and the results will be compared with arbitrary and field-based boundary conditions imposed in earlier simulations. Boundary conditions passed to the intermediate model will include hydrologic outflow, water table depth, soil water saturation state, and soil temperature. Results will be compared with intermediate-scale observations of subsurface state, surface flow, and land-atmosphere flux measurements. This task will also generate new boundary conditions to be passed to the fine-scale model, including surface and subsurface flows as influenced by the dynamic drainage network and thaw lake dynamics of the intermediate-scale model.

Fine-Scale Model Development and Tasks

The following key processes are required for the fine-scale model: water migration in freezing/thawing soils, overland flow, reactive transport of solutes and colloids, subsurface heat transport, surface energy balance processes, topographic evolution caused by thaw-induced mechanical deformation processes, soil biogeochemical/microbial processes that result in carbon release, and plant dynamics. Support for a multicontinuum reactive transport formulation may also be necessary to represent mobile/immobile regions that coexist within a grid cell. The intended spatial resolution for the subsurface model is fine enough to resolve topography and subsurface heterogeneity within single ice wedge polygons (approximately 0.1 m in the horizontal and 0.02 m to 0.1 m in the vertical). Horizontal domain size will be on the order of 100 × 100 m. Simulating these coupled processes at the spatial resolution and spatial scale required is computationally demanding and software suitable for execution on large parallel computers is required.

The proposed path forward is to develop a set of code requirements for the fine-scale model and then to evaluate PFLOTRAN and Amanzi and derivative products against these requirements to reach a decision on the preferred computational platform. Subsequent development will add any needed additional capabilities, optimize code for computers to be used by NGEE Arctic, undertake a sequence of “confidence building” test simulations, and integrate into the scaling framework.

Development of fine-scale model requirements. Detailed specifications of the mathematical model and algorithmic approaches appropriate at the fine scale will be developed. Although the key processes that are required of the fine-scale simulator are understood, detailed specifications of the mathematical models, algorithmic approaches, and computational infrastructure are still needed. For example, computational approaches for topographic evolution may be based on Lagrangian or Eulerian approaches. Computational infrastructure requirements for a Lagrangian approach are relatively well understood. The alternative Eulerian approach based on a fixed grid through which material is propagated has reduced computational infrastructure requirements but also has reduced accuracy due to material mixing at material interfaces. Some prototype development and testing are warranted to refine the Eulerian approach and to evaluate its advantages and limitations in NGEE Arctic context.

Work plans for Amanzi and PFLOTRAN development. Work plans describing the effort required to produce the fine-scale simulator from Amanzi and PFLOTRAN will be developed by September 2012. Each plan will describe the code modifications required and a plan for testing those modifications. Estimates of level of effort required to complete the work will also be provided. These two work plans will support the decision on foundational architecture for the fine-scale simulator. A workshop on the requirements for the fine-scale subsurface simulator will be held by September 2012. Potential users of the community code for fine-scale simulation will be invited. In addition, external computational specialists will be asked to participate to provide feedback on the proposed approach. This workshop will be held jointly with the requirements workshop for the intermediate-scale model (Task IM2).

Decision on the fine-scale subsurface simulator. A decision on the foundational architecture for the fine-scale subsurface simulator will be made by December 2012. The decision will be based on an objective evaluation of the existing capabilities of the Amanzi and PFLOTRAN models and their potential for extension, as referenced against the requirements described under Tasks FM1 and FM2. Decisions will be made by consensus of the modeling team matrix leads (see Section VI). Possible outcomes include adoption of one or the other modeling framework as the foundational architecture for the NGEE fine-scale simulator, or the selection of appropriate components from each framework as merged foundational elements for new development.

Construction of the fine-scale simulator. Work plans from Task FM2 will be executed, following model requirements from Task FM1 as modified by input from the fine-scale workshop. Fine-scale model development will begin prior to the decision-point on foundational architecture (Task FM3), as many model components and algorithms can be developed as modules with well-defined interfaces. Upon completion of Version 1, the code developers will conduct a tutorial workshop addressing code use, underlying technical basis for each process representation, and preliminary results on test problems.

Fine-scale simulation test cases. A series of simulation cases will be developed, including simple verification/validation tests that exercise individual process models, synthetic but realistic cases for testing model couplings, and site-specific cases that exercise the full capabilities of the simulator. These cases will be executed concurrently with development of Version 1 of the fine-scale simulator.

Simulate fluvial landscape evolution. An important functionality of the fine-scale landscape simulator is to predict changes in topography and drainage patterns as the result of permafrost thaw and thermokarst formation. Existing fluvial landscape evolution modeling approaches will be evaluated for inclusion in the fine-scale simulator. Simulations will explore several idealized cases such as low-slope domains, and flat domains with random initial elevations, and realistic domains based on LiDAR topography from polygonal tundra environments on the BEO. We will also perform exploratory modeling analyses using existing erosion/sedimentation code (Erode, http://csdms.colorado.edu/wiki/Model:Erode) to simulate the process of terrain modifications associated with melting of buried ice wedges and accompanying thermal and mechanical erosion. We will use this exploratory modeling framework to simulate thermokarst propagation and effects on lateral drainage and drying of polygonal networks.

Subsurface biogeochemical reaction network. A critical function of the fine-resolution landscape simulator will be to simulate the complex subsurface biogeochemical environment, where multiple functional microbial and fungal groups perform a range of C and N transformations. These biological reactions occur in an environment with various forms of soil organic matter (SOM), mineral surfaces, thermal states, hydrological states, redox states, and competition with plants. The goal of this task will be to develop, test, and apply a biogeochemical reaction network in a 1D reactive transport numerical model that explicitly includes processes relevant to high latitudes. For Phase 1, we will synthesize, from our previous work and from the literature, a reasonable reaction network that includes bacteria and fungal activities, characterization of multiple functional groups and their dependencies on substrates and other environmental variables, freeze and thaw cycles, nitrification, denitrification, plant nutrient uptake, abiotic reactions, and transport. We will use the proposed NGEE chemostat, column, and genomic observations to test the reaction network. Uncertainty and sensitivity analysis of the fine-scale model will be conducted to identify the key parameters and model structures that contribute to model predictions and behavior.

Initialization of fine-scale model domains. This task will use the output from landscape characterization, hydro-geomorphology, and biogeochemistry tasks associated with characterization and classification of drainage network, hydraulic geometry and connectivity, surface and frost table topography, soil carbon and ice content, and storage capacity to define the topographic, thermal, and permafrost hydrology components of several fine-scale simulation domains over the BEO. This effort includes the explicit characterization of low-centered, transitional and high-centered polygonal ground and the subsurface distribution of ice wedges that structure the microtopography in these landscapes. Results from this task also provide constraints on intermediate-scale sub-grid parameterization.

Fine-scale simulations for up-scaling. We will exercise Version 1 of the fine-scale model over subsets of the BEO, using model parameters; boundary conditions; and initial model thermal, hydrologic, and biogeochemical states. Results from these simulations serve as input to the intermediate-scale model up-scaling parameterization. The fine-scale model includes mechanistic treatment of ice wedge dynamics and dynamic microtopography. We will employ formal data assimilation methods to determine optimal parameter values and initial conditions, capturing as much of the observed fine-scale variability as possible, given constraints imposed by uncertainty in measurements. Simulation results will be evaluated against independent integrative observations.

Fine-scale boundary constraints and down-scaling simulations. The fine-scale model integration will be repeated with updated boundary conditions generated, and the results will be compared with field-based boundary conditions imposed in earlier simulations. Boundary conditions passed to the fine-scale model will include surface and subsurface flow and the thermal and hydrologic effects of thaw-lake dynamics. Results will be compared with fine-scale observations of subsurface state, surface flow, and land-atmosphere flux measurements .

Near-surface atmosphere simulations. We will begin a series of fine-scale-simulations ranging from 0.1 × 0.1 m to 10 × 10 m horizontal resolution, with appropriately scaled vertical resolutions, characterizing the implications of reduced resolution for mixing processes. These simulations will assist in the development of appropriate up-scaling parameterizations for the connection between subsurface and climate-scale atmospheric flows, providing guidance in quantifying up-scaling uncertainties.

Continued development of fine-scale simulator. Version 1 of the NGEE fine-scale simulator will represent a significant step forward in the process-resolving representation of Arctic ecosystem processes, but there are important dimensions of the fine-scale simulation problem that we will not be able to incorporate early enough in NGEE Phase 1 to allow a full integration with the up-scaling and down-scaling framework. The purpose of Task FM12 is to continue several lines of development in parallel with the up-scaling and down-exercises, to provide a more capable Version 2 of the fine-scale simulator in time for application in NGEE Phase 2. These efforts will focus on advanced grid representations for landscape deformation and increasingly sophisticated coupling of soil biogeochemical processes with vegetation dynamics.

Phase 1 Deliverables

• New version of Community Land Model with explicit sub-grid representations of geomorphological landunits controlling mass and energy fluxes in Arctic ecosystems.
• Operational fine-scale and intermediate-scale process resolving models of Arctic ecosystems, through which climate-scale parameterizations can be quantified.
• Demonstrated ability to integrate new knowledge from process observations and experimentation into a multiscale modeling framework that results in improved predictions across scales.
• A comprehensive set of data-based metrics quantifying model prediction skill and uncertainty for processes related to Arctic ecosystem hydrology, geomorphology, biogeochemistry, and vegetation dynamics.