Climate change is profoundly transforming the carbon-rich Arctic tundra landscape, potentially moving it from a carbon sink to a carbon source by increasing the thickness of soil that thaws on a seasonal basis. However, the modeling capability and precise parameterizations of the physical characteristics needed to estimate projected active layer thickness (ALT) are limited in Earth system models (ESMs). In particular, discrepancies in spatial scale between field measurements and Earth system models challenge validation and parameterization of hydrothermal models. A recently developed surface–subsurface model for permafrost thermal hydrology, the Advanced Terrestrial Simulator (ATS), is used in combination with field measurements to achieve the goals of constructing a process-rich model based on plausible parameters and to identify fine-scale controls of ALT in ice-wedge polygon tundra in Barrow, Alaska. An iterative model refinement procedure that cycles between borehole temperature and snow cover measurements and simulations functions to evaluate and parameterize different model processes necessary to simulate freeze–thaw processes and ALT formation. After model refinement and calibration, reasonable matches between simulated and measured soil temperatures are obtained, with the largest errors occurring during early summer above ice wedges (e.g., troughs). The results suggest that properly constructed and calibrated one-dimensional thermal hydrology models have the potential to provide reasonable representation of the subsurface thermal response and can be used to infer model input parameters and process representations. The models for soil thermal conductivity and snow distribution were found to be the most sensitive process representations. However, information on lateral flow and snowpack evolution might be needed to constrain model representations of surface hydrology and snow depth.
In Arctic tundra, the thickness of the soil layer that reaches above
0
Previous efforts have been made to characterize ALT using field, lab, and numerical experiments (e.g., Osterkamp and Romanovsky, 1996; Romanovsky and Osterkamp, 1997). Site-specific properties of Arctic soils, such as porosity, bulk thermal conductivity, and water retention characteristics, have been measured in lab settings from samples taken in the field (Hinzman et al., 1991; Letts et al., 2000). Those field and lab measured properties were then used in ESMs in order to predict future ALT and permafrost conditions (Beringer et al., 2001; Lawrence and Slater, 2008; Subin et al., 2013). However, such regional- and global-scale projections are difficult to constrain by measurements of soil properties made at vastly smaller scales of observation. This scale gap between the governing fine-scale physical processes and large-scale simulations impedes direct model validation against measurements, which has motivated development of fine- to intermediate-scale hydrothermal models (e.g., Hinzman et al., 1998; Hansson et al., 2004; Daanen et al., 2007; Mckenzie et al., 2007; Painter, 2011; Karra et al., 2014; Endrizzi et al., 2014; Yi et al., 2014); for a review see Kurylyk and Watanabe (2013). Numerical experiments using high-resolution coupled hydrothermal models, which are calibrated against fine-scale measurements, can play a fundamental role in understanding the governing physical processes of ALT formation.
Simulating thermal hydrology in polygonal tundra systems is a challenging endeavor that requires simultaneous representation of multiple physical processes including phase change and highly nonlinear constitutive relationships (e.g., Painter, 2011). Soil thermal conductivity alone depends on volumetric water content, mineral composition, porosity, density, and temperature (Farouki, 1981). In soils experiencing freeze–thaw cycles, the phase of water also affects bulk thermal conduction (e.g., Johansen, 1977; Peters-Lidard et al., 1998). Latent heat of fusion and evaporation impart further control on the propagation of the freezing front and therefore thermal conduction. Thermally driven vapor transport can slowly change ice content and thus thermal conduction in partially and fully frozen soils (Grimm and Painter, 2009; Karra et al., 2014). Characterizing subsurface properties for modeling is further complicated due to variability in microtopography and cryoturbated soil that create a heterogeneous surface and subsurface in polygonal tundra systems. In addition, coupling of the soil to the atmosphere involves a balance among multiple energy transfer processes, which occur across interfaces of snow, water, ice and exposed ground. All of the above attributes describing soil structure, surface energy balances, and processes of phase change result in a tightly coupled hydrothermal system. Therefore, numerical experiments using high-fidelity representations of fine-scale processes require calibrated parameters that are able to effectively link dependent processes.
Lidar of site-C with the three observation locations mapped and the greater Barrow, AK, area (credit Garrett Altmann).
Schematic representation of a Model Observation/Experiment (ModEx) process involving traditional parameter estimation–calibration (inner loop) and model structural–conceptual refinement (outer loop). Observations inform simulation input and provide a starting point for a conceptual model. Both the conceptual and numerical model is then tested against observations. In successive ModEx iterations the model is then refined and at times re-drawn in order to elicit governing processes that shape model outcome to match observed and measured phenomena. Finally, model experiments and the identification of governing processes inform future observations as to which measurements are needed to assess the state of the system.
Despite the model gains of calibrating thermal properties (Romanovsky and Osterkamp, 1997; Nicolsky et al., 2009), relatively few hydrothermal modeling studies of Arctic systems have documented calibration procedures, with the noted exception of Tang and Zhuang (2011) and Jiang et al. (2012). Additionally, correct model structure representation, capable of representing the system based on known physical relationships while using plausible model parameters, is typically not known a priori. Calibration of a model with an inadequate model structure may result in over-fitting and unreliable forward simulations that incorrectly predict system behavior based on faulty process representation (e.g., Beven, 2005; Gupta et al., 2012). Therefore, when dealing with a coupled system of complex processes, it is imperative that the conceptual model is refined during the calibration process to increase model structure adequacy (Gupta et al., 2012).
Iterative modeling approaches that use repeated model runs with different combinations of parameters, governing mechanisms, or process representation can help fundamental system understanding (Clark et al., 2008; Kavetski and Fenicia, 2011; Fenicia et al., 2011; Larsen et al., 2014). Here we use an iterative procedure that integrates finely resolved models with field observations and measurements to develop a process-rich model with physical mechanisms and parameters consistent with measurements from the Department of Energy Office of Science – Next Generation Ecosystem Experiment (NGEE-Arctic) site, Barrow Environmental Observatory (BEO), Barrow, Alaska (Fig. 1). The iterative process of using field observations to inform model development and subsequent simulations to inform new data needs is referred to here as the model–observation/experiment or ModEx cycle (Fig. 2). Clearly, there is no unique way to approach iterative modeling procedures (Larsen et al., 2014), which is intrinsically subjective and highly dependent on expert knowledge. Well-documented examples of successful applications of model refinement are thus invaluable for building the required experience base. We use repeated calibration of model parameters against site-specific field measurements and iterative model adjustments of the model structure to reduce mismatch between model predictions and measurements in order to attain a viable model of thermal hydrological conditions.
In this paper we summarize our ModEx experience involving the detailed use of subsurface temperature and snow cover field data to develop and test process-rich simulations of ALT dynamics, such that observational data and necessary physical dynamics are incorporated into the model. In order to calibrate and refine model structure in a tractable fashion, the model development first focuses on a series of subsurface-only calibrations in Sect. 3 before moving onto a series of coupled surface energy balance and subsurface calibrations in Sect. 4. The end result is a set of calibrated thermal and hydrological parameters for moss, peat, and mineral soil layers, along with a consistent model structure, employed for various microtopographic positions characteristic of polygonal tundra. We demonstrate how the detailed calibration and model development effort informs understanding of the key processes that define the ALT in polygonal ground. We further complete the ModEx cycle by discussing how future data needs can reduce system uncertainty and refine our understanding of process behavior.
Our variant of the ModEx approach is shown schematically in Fig. 2. Starting with site identification and characterization, field observations and measurements begin to form the modeling activity by providing model parameter inputs and targets for the model calibration process. Standard model calibration – denoted by the inner loop – aims to match simulations to field measurements by varying parameters while keeping the model structure fixed. Here the ModEx procedure moves beyond the standard calibration by assuming the model itself is uncertain, but can be further constrained through successive comparison to observations (outer loop in Fig. 2). These improved model runs then inform the observation process by specifying the data needs, either through further calibration or through informal numerical experimentation. Such model refinement is not a unique process, and can be achieved through multiple avenues. For example, flexible modeling approaches have been used in understand structural errors by combining functional aspects of several models (Clark et al., 2008; Kavetski and Fenicia, 2011; Fenicia et al., 2011). We implement ModEx model refinement by evaluating the plausibility of calibrated parameters in addition to the mismatch between field measurements and simulated responses.
The calibration process uses a multi-dimensional response surface to
evaluate the plausibility of parameters and the degree of mismatch between
simulated results and observed data. Sets of parameter values are mapped to
the response surface with the respective mismatch between simulated results
and field observations/measurements, quantified by the root mean squared
error (RMSE), which determines the shape of the responses surface. RMSE is
given by
The ModEx process is facilitated by two software components. First, for calibrating a given model to determine an optimal match to measurements we use the parameter estimation software, PEST (Doherty, 2004), which implements the Levenberg–Marquardt algorithm (Marquardt, 1963). This method uses gradient descent to determine (from a high-dimensional space of calibration parameters) a set of parameters that (in a local sense) minimize the forward model's error in predicting observed data. Second, the ModEx process requires iterative exchange, comparison, and addition of process models, which is greatly facilitated by a dynamically configured model with many process options. Therefore, a framework that manages complexity and allows for rapid development of new physical representations is critical. To this end, we have implemented the Advanced Terrestrial Simulator (ATS), version 0.83, as a collection of physics modules managed by the Arcos multiphysics framework (Coon et al., 2015b). At runtime, Arcos dynamically forms a dependency graph where each variable identifies its data requirements, allowing the automation of model evaluation. Process kernels (i.e., a single PDE (partial differential equation), such as mass balance) are coupled to form complex systems of equations in which each term or component can easily be replaced. The ease of swapping and adding processes makes model verification and evaluation more tractable, and facilitates the ModEx process by allowing the model structure to be easily changed and extended.
Diagram of the three 1-D columns and the associated measured soil temperature depths.
The lowland, cold continuous permafrost tundra at BEO was established as the
end-member of the NGEE-Arctic sites, which follow a bioclimatic gradient
that extends to the warm discontinuous permafrost, shrub tundra environment
of the Seward Peninsula. The site supports the NGEE-Arctic goal to improve
climate model predictions through advanced understanding of coupled
processes in Arctic terrestrial ecosystems. NGEE-Arctic scientists are
collecting multiscale in situ field measurements and remote-sensing
observations of polygonal tundra. A range of polygon types including low
center polygons, which are surrounded by rims and, in some areas, shallow
troughs, and high center polygons with deep troughs as a result of ice-wedge
degradation. The focus of the model development chronicled here is
NGEE-Arctic site “area C” (Fig. 1), which is characterized
by
The ATS solves water and energy flow in variably saturated soils at
temperatures above and below freezing using the conservation equations
described by Karra et al. (2014) (see also Painter, 2011 and Coon et al.,
2015a). Liquid and ice partitioning is represented by the model of Painter
and Karra (2014). In this model liquid water can coexist with ice below
0
Valid parameter range for calibration sets.
*
Parameter value ranges for moss, peat, and mineral soils of Arctic tundra
systems were drawn from literature and field observations at the NGEE-Arctic
site (NGEE-Arctic data portal;
The ModEx cycle as applied here to subsurface thermal hydrologic system in freezing–thawing soils.
Panels
Our experience with the ModEx cycle applied to the coupled subsurface
hydrothermal system at the BEO is shown in process flow form in Fig. 4. In
this cycle the ATS model only included subsurface processes, and the
shallowest measurement of temperature (2
The calibration error from the measured values reported as the RMSE
The first subsurface calibration attempt used the BPC model (Fig. 4) and resulted in unrealistic parameters sets. The response surface of the center and rim columns resulted in calibrated peat porosities to move to the lower parameter boundary (Fig. 5). With a few exceptions, the thermal conductivities for peat in the center, rim, and trough calibrated outside the acceptable parameter range to the lower boundary for peat. The first calibration iteration produced unrealistic parameter values and indicated that the BPC model is not an adequate calibration tool for subsurface hydrothermal modeling.
In the second iteration of our model–data integration cycle, subsurface thermal conductivity was simulated using the MC model instead of the BPC model, which reshaped the calibration response surface such that calibrated porosities spread out across parameter space and away from the parameter boundary. Calibrating with the MC model generally kept the porosity parameters within the acceptable range and improved the thermal conductivity parameters; however, RMSE increased for all columns (Table 2). Yet, the MC model was selected for the remainder of the paper because calibrated parameters were reasonable.
Upscaled parameters for larger-scale models were calibrated by coupling all
three columns to find a single set of peat and mineral soil hydrothermal
parameters. The calibration was coupled by combining objective function
results from each microtopographical feature in the PEST Levenberg–Marquardt
algorithm to inform the next parameter update that is then applied to all 1-D
columns. The initial application of the coupled calibration resulted in
unrealistic parameter values and motivated a reformulation of the conceptual
model to include near-surface unsaturated conditions necessary for center
and trough simulations. The saturated condition response surface decreased
the
Thermal conductivity of peat throughout a year with different surface pressures. Percent liquid saturation is based off of summer time water liquid saturation, which changes during winter due to an increase in ice saturation. The change in thermal conductivity coincides with spring thaw, approximately Julian day 160 or early June, and fall freeze-up near Julian day 265 or late September.
The fourth iteration of the ModEx cycle allowed the surface pressure to be a
calibration parameter for the center and trough columns, which were
previously assumed fully saturated for the duration of the year. A surface
pressure less than atmospheric results in an unsaturated condition at the
top of the soil column, and introduces air with low thermal conduction,
creating a gradient of increasing
The eight calibration starting locations for the uncoupled column
calibration were then re-tested for the center and trough by calibrating
surface pressures (Fig. 4). Here we only tested unsaturated conditions
using the MC thermal model rather than posthumously retesting prior model
structural decisions, as the MC model was thought to be more physically
accurate. The new conceptual model with unsaturated conditions at the soil
surface became the second model refinement, which resulted in a reshaped
parameter response surface. More calibrated center porosity values were
within the acceptable parameter range when surface pressures were
calibrated, but more trough peat porosities calibrated to the upper peat
boundary. Both the center and trough had more calibrated
The subsurface un-calibrated and calibrated temperature
time series is compared to measured soil temperature time series to showcase
the improvement from the calibration process at 40
The ModEx cycle applied to the surface energy balance and moss parameters.
After the calibration of subsurface thermal properties, a 2
For the surface energy balance calibration each column was spunup over a
10-year loop using decadally averaged air temperature along with shortwave
radiation, relative humidity, and wind speed data from 1 October 1998 to 30 September 2009
at Barrow, AK, where meteorological data from each day in the 10 years were
averaged together. After spin-up, daily meteorological data from 2010 to 2013
were used to drive the model. This forcing data were compiled from several
sources; the incoming solar radiation is from the Atmospheric Radiation Measurement (ARM) Climate Research Facility (ARM, 1993, 1996); rainfall and
snowfall is from Barrow airport (Station GHND:USW00027502 National Weather
Service, National Atmospheric and Oceanic Administration); air temperature,
relative humidity and wind speed are from individual research projects at
the BEO (Liljedahl et al., 2011; Zona et al., 2014); landscape-averaged
end-of-winter snow depth is from the Circumpolar Active Layer Monitoring (CLAM)
Program (Shiklomanov et al., 2012). Daily rain- and snowfall were adjusted
for undercatch according to Yang et al. (1998). A second adjustment was
applied to the snowfall where the average ratio between the 1997–2006 CALM
observations and the undercatch-adjusted National Weather Service (NWS) snow accumulation was applied
to respective daily precipitation events. The simulation results from 2013
were then compared with measured subsurface temperature data, at a 2
Temperature profiles for a 2
The second set of ModEx cycle iterations is presented in Fig. 8 in process
flow form. The focus of the second set of ModEx cycles is process
identification and calibration of the moss layer and surface energy balance
parameters. The first iteration of the cycle coupled the surface energy
balance model and 2
Ice and liquid saturation are shown in
Forcing the subsurface thermal propagation through a surface energy balance in the second set of ModEx cycles attempts to capture variable surface thermal conductivities due to changing surface saturation states as pulses of precipitation enter the subsurface and subsequently dry from evaporation. Modeling studies that do not explicitly model surface energy balance processes may not adequately capture near-surface saturation states and have reported the greatest error during the summer when highly variable soil moisture states occur (Romanovsky and Osterkamp, 1997; Jiang et al., 2012). It is known that soil moisture influences soil temperature in addition to meteorological controls, by governing the amount of latent heat of fusion necessary to freeze–thaw and evaporate water from soils (Johansen, 1977; Farouki, 1981; Peters-Lidard et al., 1998; Subin et al., 2013). Consequently, the timing of the precipitation pulses and subsequent drying may have a significant impact on ALT because the highly variable saturation states coincide with summer soil warming. Therefore, the second set of ModEx cycles starts with a more detailed representation of transient soil moisture conditions, which is the third major model refinement. Simulation results showed that it is important to capture the freeze-up timing with the highly variable fall saturation state in order to set up near-surface ice content and thermal conductivity during winter (Fig. 10a). Properly representing the freeze-up with transient soil moisture is especially important giving that winter has the largest range of possible thermal conductivity values (Fig. 6) and therefore is highly variable from year to year.
Measured snow depth ranges were gathered from a compilation of 258
snow depth measurements taken on 2 May 2013 in the area encompassing all
three borehole temperature measurements. universal transverse mercator (UTM) coordinates: Northing
7910330-7910350 and Easting 585900-585930. Measured snow water equivalence
(SWE) ranges were calculated from measured snow depth and the measured
average snowpack density of 326
The ALT for all three columns are listed for each iteration of the
calibration process, also with the range of possible ALT from the observed
data. The observed ALT range was made by finding the deepest borehole
measurement for center rim and trough with a temperature above 0
Simulating the surface energy balance for each column resulted in varied
model fits to the measured 2
Final calibrated parameter table (referred to throughout the text).
*
The largest gains from calibrating the surface energy balance portion of the
model came from the fourth model refinement, which resulted from two
additional ModEx iterations (1) updating the conceptual and numerical model
to add snow depth variation informed by microtopography and (2) including a
depth hoar representation in the snowpack model. The snowpack at Barrow, AK,
is scoured relatively flat due to strong winds (Benson and Sturm, 1993;
Zhang et al., 1996) resulting in deeper snow in depressions such as troughs
and low centers. To match measured snow depths of the three topographical
features (Table 3), snowfall was increased for the center and trough columns
by 30 % (3.6
Without snow re-distribution or depth hoar representation the snowpack
evolved to a density of 410 to 440
Adjusting the snow accumulation due to topographically informed snow
distribution and including a depth hoar representation increased the
insulative effect of the snowpack and had a clear impact on winter near-surface temperatures (Fig. 9). In addition snow distribution and depth
hoar representation improved summertime ALT predictions (Table 4).
Summertime changes in ALT due to winter conditions highlights a memory trait
of the system and the necessity to capture dominant winter processes in
order to simulate transient thermal conditions in physically based models.
Research by Hinkel and Hurd (2006) showed that large snow drifts cause long-term deepening of the ALT, due in part from the additional insulation for
the snow and the loss of cold thermal propagation into the subsurface.
Timing of snowpack accumulation and thickness has also been shown to govern
permafrost formation (Zhang, 2005). However, at the scale of
microtopographical relief, where trough to rim vertical relief changes by
40
In the final ModEx iteration and model refinement, attempts to increase the
simulated summer surface (2
1-D thermal hydrology models of transient saturation and frozen states combined with a surface energy balance model were used to represent active layer dynamics in polygonal tundra at the Barrow Environmental Observatory. In the coupled model, surface water was allowed to pond to a specified maximum height but any additional water was removed (spill over condition). The surface model also includes a surface energy balance model for bare, snow-, ice- or water-covered ground. The model was used in combination with borehole temperature and snowpack field measurements in an iterative model-data integration (ModEx) framework to produce calibrated model parameters and refine constitutive models and process representations. The particular variant of the ModEx approach combined calibration with iterative refinement of the model structure; parameter feasibility and model–observation mismatch were used as metrics to achieve the objective of model development and identification of viable representations of key thermal hydrological processes.
The results demonstrate the effectiveness of using borehole temperature measurements to effectively develop and refine the model structure for hydrothermal models of permafrost-affected landscapes. Results also suggest that properly constructed and calibrated 1-D models coupled to a surface energy balance may be adequate for representing thermal response at a given location provided the maximum ponded depth (spill point) is known for that location. This suggests a multiscale modeling strategy that uses overland flow models to establish the spill point (maximum ponded depth) at each surface location in conjunction with a set of thermal hydrology simulations. Further evaluations of the 1-D representations against 3-D model representations are needed to identify additional process representation and the appropriate level of model complexity to capture scale dependencies of thermal dynamics. In addition, it is important to note that the largest discrepancy between model and field measurements occurred during early summer in the troughs and that mismatch is likely indicating model structural error with inflow of water from upstream locations and/or unique surface energy balance conditions. Observations of water fluxes such as evapotranspiration, lateral flow, and snowmelt at the sub-polygon scale would help model representation and, in particular, the role of advective lateral heat transport. However, the temperature mismatch was brief and confined to the trough location, and is thus not expected to have large consequences for integrated results such as thaw depth.
The model refinement process identified the representation of thermal conductivity – specifically the dependence of bulk thermal conductivity on porosity, water content and ice content – as a constitutive model that affects model performance. Thus, field and laboratory work to better constrain hydrothermal representation and the governing model parameters would help reduce uncertainty in model projections. Further modeling efforts focusing on uncertainty analysis and environmental parameter sensitivity can provide information regarding which parameters govern model outcome and thus inform future observational efforts. Similarly, snowpack properties and snow distribution were found to be important. Investigations similar to Benson and Sturm (1993), Zhang et al. (1996) and Tape et al. (2010) that better define the relationship between depth hoar, microtopography and wind slab formation would help reduce uncertainty in projections. For example, snowpack dynamics and density profile observations at the NGEE-Arctic site will inform models of how the snowpack develops and how snow will distribute across microtopography.
More generally, these results demonstrated the utility of one particular approach to merging observations and models in environmental applications. In this particular iterative approach, formal parameter estimation methods are used iteratively. Each calibration run – the inner loop in Fig. 2 – minimizes mismatch between data and models with fixed model structure. The “reasonableness” or feasibility of the calibrated parameters and the RMSE are performance metrics for the calibrated model. Model structural adjustment, the outer loop in Fig. 2, is initiated when calibrated parameters fall outside reasonable bounds. Although structural model adjustments were done in an ad hoc manner guided by experience and knowledge of the system being modeled, the resulting refinements have produced robust representation of system response. Such an approach combining structural model adjustments drawing from literature, field observations and formal calibration exercises is likely to be useful in other environmental applications.
Farouki (1981) reviewed methods for calculating the thermal conductivity of soils and concluded that a modification to a method by Johansen (1977) was superior to other models in most conditions. Peters-Lidard et al. (1998) provide a clear summary of the modified Johansen approach. Following Painter (2011), we further modify the approach to a form convenient for a three-phase model and to more accurately represent thermal conductivity of peat and organic-rich soils.
Thermal conductivity in unfrozen soils is often written as (Johansen, 1977;
Farouki, 1981; Peters-Lidard et al., 1998)
For soils that are frozen and with no liquid water content, the
corresponding equation is
For a general-purpose three-phase code, thermal conductivity is needed as a
function of both
A variety of empirical fits have been used to relate the Kersten numbers to
saturation indices for ice and liquid (see, e.g., Farouki, 1981, for a
summary). A simple power-law function is assumed here as a convenient model
(Painter, 2011)
For saturated conductivity, geometric means are often used (Johansen, 1977)
We denote the model specified by Eqs. (A3), (A4), (A5) and (A8) with input
parameters,
An alternative model, which we denote the MC model, is obtained by relating
To better represent
In summary, two thermal conductivity models are available. The BPC model
uses the following parameters: thermal conductivity of dry soil, saturated
thermal conductivity in unfrozen conditions, the exponents
The surface energy balance model is a coupled mass and energy balance
simulator used to deliver energy fluxes and any water associated with
snowmelt or precipitation to the ground surface simulated by the Advanced
Terrestrial Simulator. The surface energy simulator is split into two
parts depending on whether a snowpack is present or absent. If a snowpack is
present, the surface energy balance solves for the snow surface temperature
(
Components of the energy balance model that do not depend on the surface
temperature are computed initially,
The
If
The albedo of the four possible surfaces is listed in Table B1.
Albedo values and parameter range.
The
Once the incoming radiation components of the energy balance model are
computed, evaporative resistance (
The stability function (
Once the energy balance is calculated, then the water fluxes to the ground
surface are calculated. In the case of snow, if the snow surface temperature
(
Snow water equivalence (SWE),
Values for hydrothermal properties of moss were gathered from Hinzman et al. (1991), Letts et al. (2000), Quinton et al. (2000), Price et al. (2008), O'Donnell et al. (2009), and Zhang et al. (2010). Large-scale simulations including a moss layer were also considered and informed valid parameters ranges (Beringer et al., 2001). Peat properties were found in Hinzman et al. (1991, 1998), Letts et al. (2000), Quinton et al. (2000, 2008), Nicolsky et al. (2009), Zhang et al. (2010) and the accompanying larger-scale simulations (Beringer et al., 2001; Lawrence and Slater, 2008). Mineral soil properties were gathered from Hinzman et al. (1991, 1998), Beringer et al. (2001), Overduin et al. (2006), Lawrence and Slater (2008), Nicolsky et al. (2009). van Genuchten parameters were fitted to the published soil water characteristics curves (Hinzman et al., 1991).
The Advance Terrestrial Simulator (version 0.83) is a suite of physics
modules managed within the Arcos metaphysics framework that couples multiple
model components at runtime. ATS, Arcos and the host software AMANZI is
developed by Los Alamos National Labs and the source code is available upon
request (
This work was supported by the Los Alamos National Laboratory, Laboratory Direction Research and Development project LDRD201200068DR and by the Next Generation Ecosystem Experiment (NGEE-Arctic) project. NGEE-Arctic is supported by the Office of Biological and Environmental Research in the DOE Office of Science. We thank two anonymous reviewers and Nina Kirchner for their helpful comments to improve this manuscript. We are also dearly indebted to all field personal, in particular Andy Chamberlain, William Cable and Robert Busey, who braved freezing temperatures, polar bears, and mosquito swarms to provide the necessary field measurements to develop our models. Edited by: N. Kirchner