Improved modeling of permafrost dynamics in a GCM land-surface scheme

Authors


Abstract

[1] Global climate models (GCM) are frequently used to understand and predict future climate change, but most GCMs do not attempt to represent permafrost dynamics and its potentially critical feedbacks on climate. In this paper, we evaluate the Community Land Model (CLM3), which is a land-surface scheme, against observations and identify potential modifications to this model that improve fidelity of permafrost and soil temperature simulations. These modifications include increasing the total soil depth by adding new layers, incorporating a surface organic layer, and modifying the numerical scheme to include unfrozen water dynamics and more realistic phase change representation.

1. Introduction

[2] Recently, the Arctic Climate Impact Assessment report [Arctic Climate Impact Assessment, 2004] concluded that climate change is likely to significantly transform the present natural environments, particularly across extensive areas in the Arctic and sub-Arctic. Among the highlighted potential transformations are changes in soil temperature regime and associated changes in permafrost. Soil warming can potentially drive an increase in the active layer thickness and degradation of permafrost as well as have broader impacts on soil hydrology, northern ecosystems and infrastructure. At present, permafrost is widely distributed with permafrost-affected areas covering about 25% of the land surface in the Northern Hemisphere [Brown et al., 1997].

[3] At present, there are two common approaches to simulate soil temperature and permafrost dynamics on regional and global scales. The first approach is well-developed and can be classified as a post-processing approach. In this approach, climatic variables computed by global climate models are used as input parameters for stand-alone permafrost models [Anisimov and Nelson, 1997; Sazonova et al., 2004]. Given the correct parameters, these stand-alone models are accurate and can be used for a quantitative analysis. The second approach involves simulation and prediction of permafrost dynamics within a coupled global climate model [Stendel and Christensen, 2002; Lawrence and Slater, 2005; Mölders and Romanovsky, 2006]. This approach permits the integration of potentially important interrelationships between permafrost, hydrology, and climate. However, in order for these benefits to be fully realized, improvements to ground temperature dynamics in present generation GCMs are required. In this paper, we demonstrate that correcting a few common simplifications in GCM land surface schemes can significantly improve the simulation of permafrost dynamics. Specifically, we evaluate and suggest improvements to the CLM3 [Oleson et al., 2004] which is the land-surface scheme in the Community Climate System Model [Collins et al., 2006].

2. Sensitivity Analysis

[4] Throughout this paper, we analyze the soil temperature dynamics computed by CLM3 in its offline mode and compare it with in-situ data collected in Alaska. In all our numerical experiments, we use atmospheric forcing data that is based on NCEP/NCAR reanalysis data. To simplify the analysis and remove the influence of trends, we compute 100 years of temperature dynamics by repeatedly forcing the model with the 1998 forcing data that is provided by NCAR with the CLM3 code. The soil temperature initial conditions are specified according to the present day measured temperatures at locations where the temperature dynamics are computed. In the following sections, we describe a number of modifications to CLM3 that together significantly improve simulated permafrost dynamics.

2.1. Soil Layer Depth

[5] The CLM3 simulates the ground temperature by a 10-layer soil model with the total simulated thickness of the soil column equal to 3.43 meters. Under natural conditions at the Alaska Coastal Plane, the so-called heat waves related to multi-decadal surface temperature variability decay at a depth of approximately 100 meters. The geothermal heat flux can be used as the lower boundary condition at this depth [Lachenbruch et al., 1982]. This suggests that in order to capture multi-decadal development of the temperature regime and permafrost thickness, a deeper soil column is required and that the specification of the bottom boundary condition needs to be reconsidered. For calculation of temperature dynamics in the upper 20–30 meters of soil and for time intervals of around a century, the geothermal heat flux at the depth of 100 meters can be ignored and the heat flux at the bottom of the lowest soil layer can be assumed to be zero. This assumption can be particularly valid under the present warming conditions due to the substantially decreased temperature gradient in the upper 30–100 meters of the ground material [Osterkamp and Romanovsky, 1999; Romanovsky and Osterkamp, 2001; Osterkamp, 2005].

[6] We illustrate the importance of an increased total thickness of the soil with an example simulation of the seasonal temperature dynamics at Deadhorse, Alaska. Typical active layer depths at Deadhorse under present day climate conditions are around 0.6 meters and do not exceed 1 meter. The CLM3 simulates an active layer that is too deep (see the simulated ice content dynamics in Figure 1, bottom). The impact of increasing the model depth to 80 m is seen in Figure 1 (top). To get to 80 m, we added 10 layers to the bottom of the existing vertical grid in the original CLM3 model with exponentially increasing thickness of each layer (hereafter CLM3/80). Because of the very low hydraulic permeability of the added permafrost layers, this modification does not lead to significant changes in soil moisture conditions. As a result of this modification, the active layer depth is simulated more realistically, although the simulated active layer is still deeper than that observed due primarily, as we show in the next section, to the lack of a surface organic soil layer. A systematic sensitivity analysis of the temperature dynamics with respect to the total soil thickness is beyond the scope of this paper, but it will be presented in a separate work [Alexeev et al., 2007].

Figure 1.

The air temperature and snow cover used to compute the volumetric ice content at Deadhorse for (top) 80 m and (bottom) 3.43 m thick soil columns without an organic layer.

2.2. Thermal Conductivity

[7] A surface layer of organic material of 0.2–0.3 meter depth is prevalent in many permafrost regions and plays a dominant role in heat balance and soil temperature regime because of its distinct thermal properties [Walker et al., 2002]. The CLM3 does not include the effect of soil organic material on the thermal and hydraulic properties of the soil. The thermal conductivity and heat capacity of each soil layer are directly parameterized by its sand and clay content, and the entire soil column is treated as a mineral soil only. Consequently, thermal conductivity values simulated in the CLM3 are typically significantly larger than measured ones for both soil types (mineral and organic) [Romanovsky and Osterkamp, 1997; Mölders and Romanovsky, 2006].

[8] We illustrate the importance of including an organic layer by analyzing the soil temperature dynamics computed at Nome (grid cell 165.6° ± 0.3°W, 65.2° ± 0.2°N) on the Seward Peninsula in Alaska. Permafrost at Seward Peninsula can be characterized as discontinuous and its temperature is near 0°C. Therefore, in order to estimate the current permafrost distribution in Seward Peninsula and its evolution it is necessary to prescribe thermal properties accurately. In our numerical experiments, we compute long-term temperature dynamics over 100 years. Recall that in all our simulations the atmosphere forcing is given by repeating 100 times the 1988 NCEP/NCAR reanalysis forcing data. Initial soil temperatures are prescribed according to observed present-day conditions. Results from two CLM3/80 simulations are shown in Figure 2. In the first simulation, the organic layer is not included and original CLM3 thermal properties, as listed in Table 1, are used. In this simulation, initially frozen soil becomes thawed after 100 years, i.e. there is no ice at 3 meter depth and the soil temperature stays above 0°C. In the second simulation, we represent the surface organic layer by reducing thermal conductivities throughout the upper 3 meters (see Table 1). In this simulation, the temperature at 3 meter depth stays below 0°C and permafrost exists, as it should for this tundra site with a well-developed organic soil horizon. Therefore, these simulations indicate that incorporation of an organic layer in CLM3 is likely crucial to obtain a realistic permafrost distribution, especially in areas of discontinuous permafrost where soil temperature is close to 0°C.

Figure 2.

The air temperature and snow cover used to compute the soil temperature dynamics at Nome, Seward Peninsula, Alaska for 80 m thick soil columns (top) with and (bottom) without an organic layer.

Table 1. Thermal Conductivity of Frozen and Thawed Soils
Soil HorizonsDepthCLM3Typically Measured
  • a

    Thawed/frozen soil.

Organic layer0.0–0.21.8/3.3a0.4/0.7
Mineral-organic mix0.2–1.01.8/3.30.8/1.3
Mineral soil1.0–3.02.1/3.31.2/1.9

[9] We also compare measured soil temperature dynamics at the West Dock site (148°33.13′W, 70°22.47′N, a cold permafrost site near Deadhorse, Alaska) with results simulated by CLM3 (grid cell 148.4° ± 0.06°W, 70.0° ± 0.06°N) and its modifications: CLM3/80 and CLM3/80+“organic layer”. We note that for 1998 year the NCEP surface air temperature (SAT) at this specific grid cell reflects reasonably well the measured SAT at the West Dock site except for the shoulder seasons. Since the West Dock site is located near a coastal line, the measured SAT is influenced by the Arctic Ocean, whereas in the CLM, the climate at the West Dock is modeled as more continental. For instance, in June and July,1998 the measured SAT at 2 m height is colder by 7 ± 2°C than the SAT computed by CLM at the same height. With this discrepancy in mind, the measured and simulated soil temperature dynamics at 1m depth are shown in Figure 3. Once again, the inclusion of an organic layer and deeper soil layers significantly improve soil temperature simulations, particularly during the summer and fall. In winter, the simulated soil temperatures are much colder than the measured ones. This winter colder bias may be partially explained by a thinner snow layer in the CLM3 (≈0.11 m) than the one observed at this site in 1998 (≈0.25 m). Another potential reason for the cold winter soil temperature bias in the CLM3 is the colder NCEP SAT (by 10 ± 5°C) compared to the measured SAT in early winter.

Figure 3.

Measured and computed soil temperature at 1 m depth at West Dock, near Deadhorse, Alaska.

[10] The results in this section indicate that inclusion of an organic layer has a significant bearing on the soil temperature simulations and should be included in future versions of the model. D. Lawrence and A. Slater (Incorporating the effect of soil organic material into a global climate model: Impacts on soil temperature and climate, submitted to Journal of Climate, 2007) describe a method of incorporating globally organic soil and its impact on soil thermal and hydraulic properties through the use of global soil carbon data provided by the Global Soil Data Task [2000].

2.3. Numerical Scheme of the Soil Heat Transfer

[11] The CLM3 simulates typical seasonal soil temperature T dynamics at different depths by the following procedure [Oleson et al., 2004]. First, neglecting the phase change, the heat equation

equation image

where C is the heat capacity and λ is the thermal conductivity is solved numerically to calculate the soil temperatures. At the second step, an energy conservation principle is used to adjust the temperature, water, and ice content for each soil layer independently (not taking into account temperature of the adjusted soil layers). One of the consequences of this two-step procedure is that the region where the phase change occurs can be artificially stretched, leading to inaccuracies in the simulation of active layer depth.

[12] In order to calculate temperature dynamics near 0°C more accurately, and hence to compute the active layer dynamics more precisely, it is preferable to solve heat diffusion and phase change simultaneously. We propose to employ the enthalpy formulation of the heat equation [Samarskii and Vabishchevich, 1996].

[13] First, evaluate the apparent heat capacity

equation image

where θw = θw(T) is a volumetric liquid water content at the temperature T. One of the common parameterizations of the liquid water content θw is proposed by Anderson et al. [1978]:

equation image

where the temperature T is in °C, the quantity n is the volumetric liquid water content at the moment when freezing starts; T* < 0°C is referred to as the temperature of the freezing point depression, and b < 0 is a constant.

[14] After evaluating the apparent heat capacity Capp, the next step is to use Capp instead of C in the original code that solves (1). Since the apparent heat capacity Capp is a rapidly changing function of the temperature near 0°C, iteration of the soil temperature T calculation is required.

[15] Note that this two-step procedure allows explicit evaluation of the unfrozen water content θw at any temperature. In many in-situ measurements, it is observed that some liquid water exists even at soil temperatures significantly below 0°C [Romanovsky and Osterkamp, 2000]. In the CLM3, liquid water can co-exist with ice only at 0°C. Physically, unfrozen water introduces a spatially distributed latent heat and changes thermal properties which retard the thermal response of an active layer or permafrost. Unfrozen water in the freezing and frozen active layer and near-surface permafrost protects the ground from rapid cooling and creates a strong thermal gradient at the ground surface that increases the heat flux out of the ground. This enlarged heat flux also increases the insulating effect of the snow cover [Romanovsky and Osterkamp, 2000]. To capture these effects in the CLM3, we need to include the unfrozen liquid water content in the CLM3, e.g. by (2).

[16] Finally, we compare two numerical schemes by analyzing soil temperatures calculated by CLM3 and its modification (the enthalpy formulation of the heat equation). To highlight differences between two numerical schemes, we consider soil layers which have constant thermal properties and porosity. Also, just for this comparison we modify the original code to remove thermal effect of precipitation, moisture evaporation and incident solar radiation. In our simulations, the soil is initially thawed and its temperature is 4°C. When the soil surface temperature falls below freezing temperature, the soil starts to freeze from the top down, and its temperature dynamics is determined by upper and lower boundary conditions. The boundary conditions are calculated directly by the CLM3. On the soil surface, the upper boundary condition is computed by the CLM3 from the energy balance principles, where the atmosphere temperature is set to be a linear, rapidly decreasing function of time. The lower boundary condition is also determined by the CLM3 and is zero heat flux at 3.43 m.

[17] Results of two simulations are shown in Figure 4. The soil temperature profiles computed by CLM3 have sharp corners and artificially stretched regions where the phase change occurs. Therefore, the thickness of developing frozen soil can not be computed accurately, but rather estimated up to thickness of adjusted layers. The proposed numerical scheme gives a better physical behavior of temperature, and allows an accurate prediction of the soil freezing depth. Further details regarding stability and accuracy of the proposed numerical scheme are given by Voller and Swaminathan [1990] and Alexiades and Solomon [1992].

Figure 4.

Calculated temperature profiles by the original (dotted line) and modified (solid line) numerical schemes.

3. Conclusions

[18] (1) Introducing a deeper soil column, and the associated heat reservoir, into CLM3 results in more realistic annual mean and seasonal soil temperatures.

[19] (2) The thermal properties of the organic and organically enriched mineral soil layer play an important role for correct simulation of the temperature regime both in winter and summer.

[20] (3) Realistic treatment of unfrozen liquid water at temperatures below 0°C with a modification to the numerical scheme improve the simulation of permafrost dynamics, particularly at soil temperatures near 0°C.

Acknowledgments

[21] We would like to thank S. Marchenko and N. Mölders. This research was funded by National Science Foundation (OPP-0120736, IARC-NSF CA ARC-0327664), NASA, UAF CIFAR Student Award, and by the State of Alaska.

Ancillary

Advertisement