Modeling and Scaling Strategy

P. Thornton (ORNL) will lead this team, with primary responsibility for coordinating modeling efforts across scales and across process domains, as well as overall coordination between the modeling team and the other science teams. E. Coon (LANL), D. Harp (LANL), and S. Kara (LANL) will develop, apply and test new fine- to intermediate-scale eco-thermal-hydrology and landscape evolution models. C. Xu (LANL) will develop, apply and test new Arctic plant ecosystem demography models. C. Wilson (LANL), C. Xu (LANL) and J. Rowland (LANL) will participate in the design, implementation and analysis of regional to global model inter-comparisons related to shrub expansion and the wetting and drying of the Arctic. D. Nicolsky (UAF), an expert in permafrost modeling, will participate in the data collection and modeling activities related to permafrost distribution and properties. S. Panda (UAF), an expert in permafrost remote sensing and modeling, will participate in the data collection and modeling activities related to permafrost distribution. B. Bolton (UAF) will continue to develop the Alaska Thermokarst Model (ATM) and work with W. Riley (LBNL) to integrate the landscape transition concepts of the ATM into the ACME Land Model (ALM). B. Busey (UAF) will continue to work with the fine-scale landscape evolution model ERODE, evaluating mechanical and thermal degradation processes in a polygonal tundra environment. W. Riley (LBNL) will work on development, testing, and application of models for high-latitude carbon and nutrient biogeochemistry, permafrost and geomorphological dynamics, ecosystem dynamics (e.g., shrubification), and hydrology. C. Koven (LBNL) will work on climate-scale model development and analysis of coupled plant and soil biogeochemistry and ecosystem demographics (e.g., shrubification). N. Bouskill (LBNL) will work on explicit microbial models, their integration with climate-scale models, and analyses of nutrient impacts on carbon-climate interactions. J. Kumar (ORNL) and F. Hoffman (ORNL) will develop landscape characterizations as model inputs at multiple scales. G. Tang (ORNL) will work on integration of biogeochemical processes and thermal hydrology processes in fine, intermediate, and climate-scale models. S. Painter (ORNL) will lead the multi-institutional fine-scale modeling efforts, including permafrost modeling and integration of existing models in a unified fine-scale modeling framework.

The modeling approach employed in Phase 1 was driven by the recognition that current ESMs fail to capture many of the processes known to control distributions of carbon, water, nutrients, and vegetation communities in Arctic permafrost regions. A founding assumption for NGEE Arctic is that we can improve the predictive skill of climate-scale models by synthesizing existing knowledge at the spatial scales native to the fundamental driving processes and by supplementing that synthesis through new observation and experimentation to fill critical knowledge gaps. In this strategy, models serve as integration tools and are constructed in a way that allows explicit process representation at spatial scales amenable to testing in the field and laboratory. In the course of our investigations, we may discover new processes or new system connections which challenge current assumptions. Our multi-scale modeling system allows objective evaluation of new knowledge as it is uncovered, providing a quantitative and self-consistent framework for the formulation of new working hypotheses (Figure 13).

Figure 13. A scale-aware representation of model-observation-experimentation integration (ModEx), recognizing the central goal of new knowledge generation and the use of models and observations at multiple scales to drive the process of new knowledge discovery through uncertainty quantification.

A guiding principle for our project has been to foster a two-way interaction between process-level understanding and model development at multiple spatial scales up to and including the exchange of parameterizations and knowledge with climate-scale land components of ESMs. As an important outcome of our Phase 1 efforts, we have already demonstrated a successful pathway between new field-informed fine-scale modeling and global-scale process representation. The NGEE Arctic effort led to the coupling of a CLM with a highly resolved subsurface simulator for thermal hydrology and biogeochemistry (PFLOTRAN). This coupled CLM-PFLOTRAN code forms the basis for improved subsurface process integration into the ACME Land Model, or ALM. In Phase 2, we expect to benefit from climate-scale developments emerging from the ACME effort while continuing to drive future generations of ESM development through improved ecosystem process representation from bedrock through soils and vegetation canopy to the atmospheric boundary layer. Specifically, our Phase 2 efforts will adopt and extend ACME developments in the areas of subgrid heterogeneity and topographic down-scaling to inform and improve our ability to place new fine-scale Arctic process understanding in a climate-relevant context. The result will be a much more process-explicit representation operating at multiple subgrid scales within future generation ESM watershed-based grid cells. This coordination between NGEE Arctic and ACME is facilitated by overlap in the project teams and leadership and is recognized as a mutually beneficial approach to improved science delivery by the top-level leadership for both projects (see letter of collaboration from ACME PI David Bader, Section 10). Strong coordination with other BER modeling, observation, and experimentation is a priority for NGEE Arctic.

Based on assessment of uncertainties in current-generation ESMs and on our team’s experience during Phase 1, we have identified the following five critical effort areas for new scaling and modeling research for NGEE Arctic Phase 2: (1) landscape heterogeneity, (2) biogeochemistry, (3) plant trait variation and distribution, (4) shrub biogeography, and (5) watershed hydrology. We have also identified fine-scale permafrost modeling as an overarching modeling priority which provides essential context for each of the other five science areas. Modeling priorities for improved climate prediction have guided the identification of these areas and have informed the more specific science questions to be tackled in Phase 2. Rationale for the specific science questions is presented together with detailed tasking for field, laboratory, and modeling efforts in Section 5.C (“Integrated Research Plans”). In the remainder of this section we describe the scaling and modeling approach and infrastructure that underpin effort across all of the science questions. We include descriptions of some integrated modeling (IM) and scaling tasks that are necessary elements of our Phase 2 approach but which do not fit neatly into the plans for any single science question.

Arctic Landscape Heterogeneity

Heterogeneity of land surface structure and function presents a major scientific challenge for Earth system modeling. For the sake of computational efficiency, and also in the face of sparse observational data, ESMs need to simplify landscape heterogeneity while retaining the components of fine-scale variance that have the strongest influence on aggregated fluxes and states. One approach has been to define relatively large grid cells that represent subgrid heterogeneity by carrying information about the number, type, and fractional area of different land classes within each grid cell. The large grid cells form the basis for communication of fluxes and state information between land and atmosphere, while the subgrid fractions capture processes known to vary among different land classes. Current ESM land subgrid schemes differ in the definition of classes; in the topology of the subgrid (e.g., nested hierarchy or single vector); and in the degree of communication among subgrid elements.

A deficiency in the current generation of subgrid schemes is that although they capture the subgrid area of various land classes they ignore the geography of those classes – how the subgrid regions relate to each other in space and how they relate to regional landforms such as ridges, valleys, and drainage basins (Qian et al. 2010). These considerations are particularly important in complex terrain where near-surface weather and surface and subsurface hydrology are all strongly influenced by elevation, slope, aspect, and surface geology (see Section 5.C, Q1). Vegetation and biogeochemistry are similarly responsive to these landscape variations. Next-generation ESM development in the ACME project is exploring the use of automatically delineated watersheds to replace rectangular grid cells as the top-level organization in a nested spatial hierarchy. The characteristic scale of delineated watersheds is a selectable parameter allowing grids of varying average resolution.

The ACME approach treats subgrid heterogeneity as sub-watershed topographic units organized around variance in elevation, slope, and aspect (Tesfa et al. 2014). Watersheds and sub-watershed topographic units are delineated from high-resolution (30 m) digital terrain data available globally. A nested hierarchy is used, with topographic units themselves composed of multiple land types representing broad classes such as bare rock, glacier, lake, and unmanaged and managed vegetation. Each land type can itself be composed of multiple “columns,” representing states of mass and energy in a vertically oriented plant-soil subunit including snowpack and plant litter layers resting on the soil surface. Multiple soil columns are also used to capture the carbon cycle and climate consequences of managed and unmanaged disturbances (Wang et al. 2014b). A final level in the subgrid hierarchy describes variation in vegetation types with multiple plant types potentially sharing space on a single column. In this framework all levels below the watershed are spatially implicit meaning that areas are retained but landscape locations are not.

NGEE Arctic Phase 2 modeling will adopt the ACME multi-level subgrid hierarchy as a climate-scale component of our scaling framework and will achieve intermediate-scale process representation by increasing the number of subunits at each hierarchy level and by also retaining detail on the explicit location of subunits within the landscape, as well as lateral connections among subunits (e.g., Cresto Aleina et al. 2013; Shi et al. 2015).

Task IM1: Define climate and intermediate-scale modeling domains. We will use high resolution (5 m) digital terrain data to delineate watersheds and sub-watershed units for our study regions near Barrow and on the Seward Peninsula.

For our Phase 1 model scaling strategy, we included an intermediate-scale modeling effort that was connected to fine-scale and climate-scale modeling through parameter estimation and upscaling; this was conceived as an independent software system. With the new subgrid developments in ACME and our Phase 1 success in linking ALM-PFLOTRAN, we have the opportunity to use the same software system for both climate-scale and intermediate-scale simulation thereby improving consistency in knowledge migration across scales while reducing the complexity of the software system. Parameter estimation across scales remains a critical science focus for the project, and we will be able to devote more effort to that area by using the flexible subgrid scheme of ALM. Our fine-scale modeling efforts will continue, as in Phase 1 to be based on a 3D reactive transport framework with convergence toward a single fine-scale modeling framework for Phase 2 (see more details in this section under Fine-Scale Permafrost Modeling). As we look toward NGEE Arctic Phase 3 deliverables, we also see fine-scale representation of landscape heterogeneity strengthening the relevance of a multi-scale modeling framework for Arctic policy and decision-making (Kraucunas et al. 2015).

Arctic Soil Biogeochemistry

We will continue to evaluate the scaling hypothesis put forward under Phase 1 that soil biogeochemical processes are controlled by temperature, moisture, and mineralogy, with organic inputs from new plant production and decomposition of old organic matter in thawing permafrost and outputs as gaseous, dissolved, and particulate flows. This hypothesis implies that the biogeochemistry scaling approach requires spatial scaling of physical model components (thermal and hydrological fluxes and states) and plant dynamics (e.g., fine-root turnover, exudation, competition for nutrients). Special areas of focus in Phase 2 will include new vegetation-biogeochemistry feedbacks associated with position along hillslope flowpaths and up-slope inputs of nutrients, for example through N-fixing shrub communities, as well as modeling temperature responses of microbial processes and their sensitivity to mineralogy and redox state. The ALM-PFLOTRAN framework coupling thermal-hydrology and subsurface biogeochemistry is being incorporated by ACME as a global modeling capability. We propose to implement that capability in Phase 2 and to continue development of new process representations in that framework.

Task IM2: Initialize simulations with coupled thermal-hydrology and biogeochemistry. We will parameterize coupled physics-vegetation-biogeochemistry capability at watershed scales for intensive study sites (see Task IM1), providing initial framework upon which new Phase 2 biogeochemistry studies can build.

A major emphasis for NGEE Arctic research is on improved prediction of net carbon exchange in Arctic ecosystems with special attention to the partitioning among sources and sinks of CO₂ and CH₄ under conditions of thawing permafrost (see Q2). This is an area where our models depend very strongly on structural evaluation and parameterizations obtained through field and laboratory study. The NGEE Arctic models require information about the chemical composition of fresh plant litter inputs and multiple classes of soil organic matter (e.g., in the active layer and in permafrost), and they also require empirical rate constants describing the dynamics of various SOM decomposition pathways under a range of soil temperature, moisture, and nutrient availability conditions. A new research need emerging from our soil biogeochemistry modeling focuses on the question of time scales for temperature, moisture, and nutrient variation and their influence on microbial dynamics and decomposition. Current model parameterizations for base rates of decomposition and variation in rates associated with changing temperature and moisture are derived from incubation experiments in which temperature and moisture conditions are held constant for extended periods. In our models those same parameterizations are applied to variation on diurnal, synoptic, seasonal, and inter-annual time scales. This assumption of linear scaling in time stands as a testable hypothesis, and coordinated field and laboratory efforts are proposed under Q2 to address it.

Arctic Plant Traits

Earth system models represent the structure and function of vegetation, providing a critical biological link in the global energy budget and in the global cycles of water, carbon, and nutrients. Simplifying assumptions are required to accommodate the need for global simulation and in recognition of the uneven availability of parameterization data for plant physiological processes across the global range of vegetation types. For example, the most sophisticated ESMs currently include subgrid fractions of individual grid cells occupied by multiple plant functional types, but those types are parameterized with mostly static (one value per plant type, no time variation) representations of the numerous plant traits that influence energy, water, carbon, and nutrient fluxes. Recent efforts to quantify variation in multiple traits and co-variation among traits have led to modeling improvements at regional and global scales. For example, NGEE Arctic efforts under Phase 1 greatly improved the representation of key plant traits regulating photosynthesis for multiple Arctic vegetation types (Rogers 2014; Rogers et al., in preparation). In addition, new approaches for collecting plant traits using remotely sensed data offer the potential to dramatically increase the amount of plant trait information available for modeling activities (Serbin et al. 2012, 2014, 2015; Singh et al. 2015). Global-scale plant trait databases are also providing useful constraints for modeled plant traits (Kattge et al. 2011), but Arctic plant traits and belowground traits in general are under represented due to sparse data (e.g., Iversen et al. 2015). Early successes have highlighted the value of increased emphasis on realistic plant trait representation in ESMs including the need for plant trait prediction and introducing dynamic trait distributions in place of single, static values.

Task IM3: Initialize simulations with Arctic plant traits. We will bring the best current representations of Arctic plant traits from global databases and NGEE Arctic Phase 1 efforts into baseline simulations and intensive study sites (from Task IM1), providing an initial framework upon which new Phase 2 trait studies can build.

Our Phase 2 efforts will expand on previous work, bringing a major new focus on Arctic shrubs, which were not well represented at the Barrow field site but which play a crucial role in land-atmosphere interactions for energy, water, and greenhouse gases across the Arctic. For maximum benefit to global models, our emphasis will be on building up the sparsely populated Arctic plant trait categories in existing databases through new observations, focused experimentation, and frequent integration of new knowledge in models at fine, intermediate, and climate scales (see Q3 and Q4). A crucial aspect of the scaling strategy for vegetation modeling is to relate plant-scale or organ-scale measurements to plot-scale or canopy-scale properties. For example, we have an established canopy scaling approach that relates distribution of leaf mass, leaf area, nutrient concentrations, and photosynthetic rates to vertical canopy structure (Thornton and Zimmermann 2007). That approach relies on leaf-scale measurements of specific leaf area, leaf nutrient concentration, and photosynthetic potential at the top of the canopy (Rogers 2014) and on canopy-scale measurements of increase in specific leaf area with canopy depth. The modeling, observational, and experimental teams will work together to extend these scaling approaches to other physiological traits including a focus on the belowground system, which has received relatively little attention. We will coordinate our investigations into Arctic plant traits and trait-based modeling with related efforts under the NGEE Tropics project through participation in community workshops, frequent informal exchange of research findings, and through integration of Arctic-specific findings into the global trait-based modeling framework being developed under the ACME project.

Arctic Shrub Dynamics

Due to constraints from physical climate, Arctic tundra has relatively low biomass density and patchy plant distribution. Canopies with low leaf area, low vegetation stature, short growing seasons with rapid phonological transitions, and strong vegetation-snow interactions together produce a landscape energy balance that is uniquely sensitive to shifts in vegetation community composition. Major modes of variation in present day climate and predictions for future climate change depend strongly on the Arctic surface energy balance (Bekryaev et al. 2010; Winton 2006). Vegetation distributions and surface energy balance also drive significant interactions with surface and subsurface thermal hydrology and permafrost dynamics. With inputs to soil organic matter coming from above- and belowground plant litter, biogeochemical processes also interact strongly with vegetation distributions in this heterogeneous landscape. Shrubs play an especially important role in the landscape because they introduce variation in albedo, woody allocation, canopy height, nutrient dynamics, snow interactions, rooting distribution, and hillslope organization (see Q4). ESMs need to be able to predict the current and future distributions of Arctic shrubs and other vegetation types, but in addition to poorly constrained physiological trait parameterization (see Q3), current global models include only very simple approaches to prediction of dynamic vegetation distribution, and even those models have not been carefully evaluated in Arctic tundra landscapes.

Research is under way in the ACME project to introduce a global-scale ecosystem demography (ED) capability in ALM, and this is being integrated with trait-based modeling in NGEE Tropics. Facilitated by overlap in modeling teams, we will collaborate with these developers to bring new dynamic vegetation capability into our Phase 2 modeling framework.

Task IM4: Initialize simulations with Arctic vegetation dynamics. We will parameterize ED capability at watershed scales for intensive study sites (from Task IM1), providing an initial framework upon which new Phase 2 vegetation dynamics studies can build.

We will develop new predictive models representing both the current distribution of vegetation types and the expected shifts in distribution under a changing climate, taking into account interactions with disturbance (e.g., fire, extreme temperatures, wind); changing permafrost; topographic setting; snowpack; hillslope hydrology; and nutrient dynamics. In Phase 2, models will be developed in parallel with data synthesis and new data collection. A common modeling framework will ensure tight integration of these efforts with physiological trait modeling.

Arctic Watershed Hydrology

With improved representation of landscape heterogeneity in next-generation ESMs, NGEE Arctic has an opportunity to explore and improve the quality of prediction for subgrid spatial extent of permafrost and its interaction with surface and subsurface hydrology through hillslope flowpaths, vegetation and soil organic matter distribution, and variation in soil depth and mineralogy. Several of the science questions posed for our Phase 2 research plan engage aspects of watershed or hillslope hydrology with a particular emphasis on these dynamics in Q5. To ensure integration across the science question investigations, we will instantiate a full range of multi-scale modeling in each of the intensive study sites using watersheds, hillslopes, and sub-hillslope partitions as the distinct geomorphological units providing spatial organization to our scaling strategy. This approach will provide a common baseline of simulations at the intensive sites, at watershed, sub-watershed, and fine scales, available for use by our entire team over the Seward Peninsula study sites early in Phase 2. A similar strategy is already being deployed at our Barrow study sites, and that effort will continue in Phase 2.

This common modeling framework approach serves several purposes. First, it prescribes a rapid assessment at each site of data availability compared to a minimum set of modeling requirements. Missing data required for model initialization will be quickly identified and new data collection can be prioritized. Second, it provides measurement and experimental teams with a consistent set of model outputs that can be used to guide sampling strategy and experimental design, in an iterative hypothesis testing framework. Third, it delivers a baseline simulation capability at multiple spatial scales that can be customized for the modeling tasks associated with each science question and progressively refined as new knowledge is gained through those investigations. Fourth, it allows an assessment of the prediction uncertainty and computational cost for model deployment at various spatial scales over extended regions, be that over the satellite sites or over larger regional landscape extents.

Task IM5: Baseline watershed simulations. We will identify watersheds enclosing the intensive study sites from Task IM1 and will implement climate-scale, intermediate-scale, and fine-scale simulations at each site based on available site characterization data and preliminary, prioritized new measurements.

Climate-scale simulations will be based on watersheds as grid cells, with implicit geographic information for subgrid units (area fractions, but not locations within watersheds). Intermediate-scale simulations will be based on the same watersheds, but with explicit geographic information (both area and location), potentially also resolving a larger number of subgrid elements at each level in the nested subgrid hierarchy. Fine-scale models will be implemented as highly resolved (1 to 5 m horizontal resolution) 2D and 3D grids oriented along hillslope flowpaths.

Fine-Scale Permafrost Modeling

At the outset of Phase 1, we defined a model-based scaling strategy that used field and laboratory data to constrain fine-scale models of thermal, hydrological, biophysical, and biogeochemical dynamics in polygonal tundra and then we used these fine-scale modeling results to parameterize larger-scale models of the same processes, targeting new process representations and reduced uncertainties in models at the climate-prediction scale. The coupled fine-scale modeling had never been attempted, and to mitigate risk of failure we proposed and executed two efforts based on independent numerical and software engineering frameworks. We succeeded in producing first-ever high-resolution simulations of coupled thermal-hydrology surface-subsurface dynamics for the polygonal tundra environment. We also demonstrated the robust coupling of vegetation and biogeochemistry to thermal-hydrology simulations.

Although significant progress was made in Phase 1, a consensus conclusion emerging from our fine-scale modeling efforts is that existing numerical/software frameworks, as currently configured, are not able to generate fine-scale predictions of freeze-thaw dynamics at the spatial scales required to parameterize larger-scale models of Arctic tundra. After significant development we are able to make one-dimensional (vertical) simulations of freeze-thaw dynamics over multiple years with reasonably fast computation times. Two-dimensional simulations (horizontal transects with vertical resolution) have been demonstrated, but have proven to be more challenging, with simulations occasionally being forced to very slow computation. Three-dimensional problems in realistic terrain, with coupled surface and subsurface hydrology have so far been possible only for small domains (a few tens of meters in the horizontal) and for short time periods (one to several years).

Small simulation time steps have consistently been an issue, preventing us from carrying out fine-scale simulations over large domains, and hindering our progress on the up-scale migration of new knowledge to inform and improve climate-scale prediction. Small time step size is a direct result of the numerical representation of ice-water phase change, which causes mass and energy conservation equations to become extremely nonlinear in the vicinity of freeze/thaw fronts. In particular, a small change in temperature can create a large change in the energy residual at these critical junctures. As a result, it is difficult to numerically find pressures and temperatures that satisfy the mass and energy conservation equation in each grid cell. Typically, the nonlinear solver fails to converge for time steps of reasonable size, thus requiring the time step to be reduced until a numerical solution can be found. We see similar behavior in both the Arctic Terrestrial Simulator (ATS) and PFLOTRAN, which use different nonlinear solution algorithms.

Based on our experience in Phase 1, we understand the cause of this poor numerical performance and have identified multiple options to improve the time step and move to simulation over larger domains. The options include: modifying the nonlinear solution algorithms; developing more robust time stepping schemes based on solving part of the equation explicitly (implicit/explicit schemes); and making thoughtful approximations to the underlying physics to make the nonlinear system easier to solve.

Task IM6: Fine-scale permafrost modeling. We will assign a small team of specialists, drawn from the two Phase 1 fine-scale sub-teams, to work together over the first year of Phase 2 to solve this fundamental numerical and computational problem.

The problem is now well enough constrained that this experienced task team will work toward a solution at the pure numeric or functional unit level, outside of either the ATS or the PFLOTRAN software framework. The results of this effort will be applied to a single Phase 2 fine-scale modeling framework which merges the realized benefits of our two parallel Phase 1 fine-scale models. Specifically, we will adopt the multi-process coupling framework of ATS and combine that with the vegetation-biogeochemistry coupling currently implemented in ALM-PFLOTRAN.

Model benchmarking

Hypothesis testing, model evaluation, sensitivity analysis and uncertainty quantification play a crucial role in the cycle of knowledge discovery shown in Figure 13. Model outputs and hypotheses are evaluated against quantitative knowledge to establish new process study needs, a step which serves as both a filter on multiple competing hypotheses and also as a “ratchet”, moving understanding of the complex multi-scale system forward according to prediction skill and bias reduction in the face of quantified uncertainties. We will formalize this process in our Phase 2 work through a series of model-benchmarking iterations, starting with simulation results from integrated modeling tasks IM2-IM5, evaluated against observational datasets already available from the International Land Model Benchmarking project (ILAMB, supported by DOE’s BER Regional and Global Climate Modeling Program) and from our Phase 1 research. These early uncertainty quantification efforts will result in baseline metrics which will be revisited periodically as new observational, experimental, synthesis, and model development efforts proceed in Phase 2. Developments that result in broad improvements in metrics over baselines will be given more weight in deciding new process study priorities. Specific model benchmarking tasks are called out in Section 5.C, and we will contribute new Arctic-specific benchmarking datasets and evaluation metrics to the ILAMB effort for use by other modeling teams and projects.

Summary table of prioritized modeling needs

We have identified the most critical data needs as a common set of prioritized modeling requirements relevant to most or all of the integrated modeling tasks, and also to more specific modeling tasks described in Section 5.C (Table 1). It is recognized that filling these modeling requirements will be an iterative process, and that more complete and accurate information will become available as our Phase 2 studies proceed. Modeling requirement identifiers (MR1‑8) are used for easy reference in later proposal sections.

Table 1. List of high-priority modeling data requirements.

5.B   Site Selection and Sampling Strategy

A critical component of developing capabilities to simulate terrestrial ecosystem-climate feedbacks in carbon-rich permafrost environments is the selection of study sites and development of sampling strategies. Multiple datasets from field sites combined with remote sensing imagery that puts field-scale information into a regional context are required to quantify properties and processes that underpin the five Phase 2 science questions. Such information is also needed to develop and test scaling schemes and to initialize, parametrize, and validate models. Given our long-term goal (Phase 3) to deliver a predictive understanding of pan-Arctic ecosystems and their potential response to a changing climate, we propose in Phase 2 to extend our field, laboratory, and modeling efforts across a representative landscape and climate gradient in Alaska. This strategy will enable our team to compare and contrast knowledge gained across a strong north-south climatic and vegetation gradient and leverage modeling capabilities for regional and pan-Arctic simulations.

NGEE Arctic Phase 1 research focused on the BEO situated on the coastal plain of the North Slope of Alaska. At the largest scale, the hydrological basins of the North Slope comprise thaw lakes, drained thaw-lake basins, interstitial polygonal regions and drainage features with length scales of hundreds of meters. Nested within these land units are ice-wedge polygons of different types. This area was chosen to represent a cold, lowland, carbon-rich, low vegetation-biomass density and diversity, continuous permafrost site located at the northern extent of an Alaskan landscape and climatic gradient. Lowland landscapes such as the BEO comprise ~30% of the Arctic and sub-Arctic. These Alaska coastal plain, river valley, and delta regions have thin (<1 m) active layer soils and deep, ice-rich permafrost. This geometry leads to saturated and flooded conditions across large portions of the landscape throughout the thaw season (Kane et al. 2008), while long periods without precipitation lead to drought-like conditions.

A driving hypothesis of the NGEE Arctic Phase 1 efforts was that polygonal landforms control hydrological stocks and fluxes in the region and that moisture distribution would in turn greatly influence vegetation and soil microbiology and thus components of the carbon cycle and energy balance. As described in Section 4, research in Phase 1 verified this hypothesis, documenting the strong dependencies that exist among geomorphology and carbon cycle processes (Lara et al. 2015), hydrology and biogeochemistry (Hubbard et al. 2013; Newman et al. 2015; Roy Chowdhury et al. 2015), and developing new approaches to extrapolate linked properties that control carbon fluxes at landscape scales (Wainwright et al. 2015).

In Phase 2, we will build upon the success demonstrated in Phase 1 by establishing an observational gradient that includes sites on the North Slope but also on the Seward Peninsula (Figure 14). In contrast to Barrow, this warmer region occupies a highly dynamic transition between Arctic and boreal ecosystems (Epstein et al. 2004). It is characterized by warm, discontinuous permafrost, well-defined watersheds, extensive shallow bedrock, greater diversity in vegetation composition, and this region is experiencing increased frequent episodes of disturbance (Swanson 2010; Jones et al. 2012; Rocha et al. 2012). Approximately 40% of the Arctic can be classified as hilly, vegetated, soil-mantled landscapes with significant carbon stored in active layer soils and permafrost (Grosse et al. 2011; Schuur et al. 2015). Due to the topographic complexity and heterogeneity in permafrost distribution on the Seward Peninsula, we expect greater variability at significantly larger scales in vegetation, biogeochemistry, and permafrost dynamics than observed on the North Slope. A focus on the Seward Peninsula will allow us to compare and contrast knowledge gained on the North Slope, as well as to continue to refine our scaling and modeling approaches gained in Phase 1 to predict ecosystem-climate feedbacks. As described in Section 5A (Modeling and Scaling Strategy), the proposed Phase 2 watershed-centric effort in the Seward Peninsula is not only critical for the NGEE Arctic goal of representing landscapes with strong lateral flows, but it will also provide a natural point of connection to ACME, which treats subgrid heterogeneity as sub-watershed top

ographic units that are organized around variance in elevation, slope, and aspect.

In addition to the contrasting characteristics of the Seward Peninsula relative to the North Slope and their relevance to other Arctic regions, the logistics in the Seward Peninsula region are tractable. This is an important consideration given the need for teams to access the site and to deploy instrumentation in a manner that is safe, cost-effective, and with minimal disturbance to the fragile ecosystem. Our strategy in Barrow has enabled more than 70 scientists to conduct field research on the BEO each summer. While it is unlikely that we can fully replicate this on the Seward Peninsula, our experience suggests that multiple small teams can be deployed safely and efficiently, with larger teams working in intensive field research campaigns. The NGEE Arctic Leadership Team has already taken three trips to the Seward Peninsula (nine to 12 people at a time) to canvass and discuss a range of sites that could serve as Phase 2 test beds. Candidate sites have been identified in three regions of the Seward Peninsula. These include (1) a lowland wet and warm and relatively thin permafrost region with extensive thermokarst and drained lakes near Council; (2) hillslopes and watersheds near Kougarok where vegetation varies with hillslope position and aspect, some previously affected by fire and other disturbance; and (3) along the bedrock-rich areas along the Teller road where hillslope and watershed position, as well as wind, may influence vegetation patterns. In parallel with synthesis of existing research data and analysis of remote-sensing products for the Seward Peninsula, an additional campaign is scheduled for July/August 2015 to acquire samples and perform reconnaissance geophysical imaging at a number of candidate sites. These datasets will be considered together with the Phase 2 science questions in the final site selection process. A characterization tool has already been developed to aid in the visualization, description, and communication of site details within Phase 2 of the NGEE Arctic project.

In addition, a series of synthesis activities will compile relevant data collected at often used field sites that span the tundra area between Barrow and Council including Atqasuk, Ivotuk, Oumalik, Selawik, and Quartz Creek (e.g., McGuire et al. 2003). We will also undertake synthesis activities to compile and analyze data from other regions of Alaska including the North American Arctic Transect (Walker et al. 2008) and sites at Toolik Lake and along the Dalton Highway. We will complement these synthesis activities with representativeness analyses similar to Hoffman et al. (2013). In this way we will be able identify the spatial extent represented by each of the possible satellite sites and down-select in a rational and informed manner.

In conjunction with our site selection strategy, it will also be important to consider site design and sampling. We propose to deploy two different but coordinated sampling strategies during Phase 2. Analogous to the BEO Intensive Study Site, we will develop one site on the Seward Peninsula that will focus all observational and modeling efforts. Within that site, we will develop key intensive sampling transects, parallel and perpendicular to the axis of a chosen watershed. Three parallel transects traversing the watershed will give us replicated sampling of the catena. Plots located at regular intervals, and some stratified by landscape position and vegetation type, will be overlaid on spatially continuous geophysical and remote sensing observations on the transects. Experience at the BEO suggests that such a nested strategy, which includes intensive above- and belowground sampling along geomorphic transects (e.g., Hubbard et al. 2013), can lead to important insights about ecosystem behavior that are difficult to achieve using only stratified, anova-type designs. We envision co-acquisition of datasets along the intensive transects including thermal-hydrogeological, geophysical, vegetation, energy and biogeochemical data. In addition to the Intensive Study Site on the Seward Peninsula and associated intensive transects, we will develop a limited number of satellite sites and transects elsewhere for subgroups to perform specific process investigations related to Q1 through Q5. These sites will be selected to construct controlled (Jenny 1941) comparisons or gradients of, for example, fire chronosequences, permafrost conditions, vegetation types (e.g., alder) or soil mineralogy.