Representing and evaluating the landscape freeze/thaw properties and their impacts on soil impermeability: Hydrological processes in the community land model version 4

Authors

  • Mingjie Shi,

    1. Department of Geological Sciences, John A. and Katherine G. Jackson School of Geosciences, University of Texas at Austin, Austin, Texas, USA
    Search for more papers by this author
  • Zong-Liang Yang,

    Corresponding author
    1. Department of Geological Sciences, John A. and Katherine G. Jackson School of Geosciences, University of Texas at Austin, Austin, Texas, USA
    • Corresponding author: Z.-L. Yang, Department of Geological Sciences, John A. and Katherine G. Jackson School of Geosciences, University of Texas at Austin, 1 University Station C1100, Austin, TX 78712, USA. (liang@jsg.utexas.edu)

    Search for more papers by this author
  • Felix W. Landerer

    1. Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA
    Search for more papers by this author

Abstract

[1] Snow cover at high latitudes is an excellent natural insulator that can maintain the underlying ground at a higher temperature than the overlying atmosphere. Soil impermeability usually varies when snow cover accumulates, which is closely related to soil and landscape freeze/thaw status. How snow cover affects the landscape frozen fraction and soil impermeability and how this impermeability regulates hydrological processes in cold regions have not been fully assessed and quantified. In order to understand these processes, this study performed a series of experiments by using the Community Land Model version 4 (CLM4). We first simulated the top-soil-layer ice, snow ice, and canopy ice to calculate the landscape frozen fraction, which was evaluated based on the Special Sensor Microwave/Imager (SSM/I) observed landscape freeze/thaw earth system data record (FT-ESDR) in two selected regions at high latitudes. Then two soil impermeability parameterizations were validated against various in situ and satellite observations. The results suggest the following: (1) compared to SSM/I FT-ESDR, CLM4 can capture the overall landscape freeze/thaw status in the regions north of 60°N in boreal winter and spring; (2) as the snow cover fraction approaches unity, the CLM4-simulated landscape frozen fraction is mainly controlled by the snow ice amount, resulting in step changes between SSM/I FT-ESDR observed and CLM4-simulated landscape frozen fractions; and (3) in most of the cold regions, the timing of the boreal spring runoff simulations is improved by reducing the impermeable area in high landscape frozen fraction regions.

1 Introduction

[2] The landscape freeze/thaw status is closely related to the surface energy budget, hydrological activity, vegetation dynamics, terrestrial carbon budgets, and land-atmosphere trace gas exchanges [Kim et al., 2011]. Soil freezing warms up the surface in winter and autumn as a result of heat release, while soil thawing cools down the surface in spring at high latitudes [Poutou et al., 2004]. Frozen soil reduces infiltration of snowmelt or rain and promotes overland flow [Johnsson and Lundin, 1991; Hardy et al., 2001; Niu and Yang, 2006]. As such, soil freeze/thaw affects soil moisture content and its seasonality, which, in turn, regulates the vegetation carbon density in permafrost-dominated regions [Beer et al., 2007]. Increased soil freezing also affects root and microbial mortality, cycling and loss of nutrients, chemistry of drainage waters, and soil-atmosphere trace gas fluxes [Groffman et al., 2001].

[3] As another land feature, snow is also important to thermal and hydrological characteristics in midlatitude and high-latitude regions. Snow has high surface albedo, which reduces the radiative energy absorption at the surface; snow has low thermal conductivity, which insulates the subsurface from rapid and large temperature variations [Marshall et al., 1994]. Snow seasonality is important to the soil moisture budget and freshwater resources, because runoff from snow melting is essential to many drought-prone regions in spring and summer [Su et al., 2011].

[4] Despite its importance, the soil freeze/thaw status and its dependence on the overlying snowpack depth and coverage are not well understood. By using Special Sensor Microwave/Imager (SSM/I) data, Zhang and Armstrong [2001] detected the near-surface soil freeze/thaw status over the snow-free land area of the contiguous United States but failed to capture the soil freeze/thaw status under the snow cover. In order to quantify freeze/thaw status dynamics over vegetated land areas for the global domain, Kim et al. [2011] developed a landscape freeze/thaw earth system data record (FT-ESDR) through the use of SSM/I. They also pointed out the limitations of applying FT-ESDR to soil freeze/thaw dynamics, especially under snow or thick surface organic layers. However, it is widely known that because of its low thermal conductivity, snow cover can maintain the underlying ground at a higher temperature than the overlying atmosphere [e.g., Goodrich, 1982; Zhang et al., 1996]. Soil freezing in the northern latitudes is usually reduced by the seasonal snowpack, which insulates the soil surface from the atmosphere; a lack of snow or a late accumulating snowpack usually causes the soil to freeze deeper and for a longer duration than a snowpack that forms early in winter [Hardy et al., 2001]. Soil under the seasonal snow cover may eventually thaw even though the soil freezes before the snow cover develops [Zhang and Armstrong, 2001].

[5] The soil freeze/thaw, which is influenced by snow cover, is associated with global climate change; warmer winter temperatures, as a result of climate change, may result in less snowfall in temperate forests [Hardy et al., 2001]. By analyzing 19 years of North American snow cover climatology, Karl et al. [1993] also found large-scale systematic decreases of snow cover extent as temperature increases. A decrease in snow cover or late development of the snowpack, either of which may occur in a warmer climate, may eventually lead to increased soil freezing because of less insulating snow cover and changes in soil water dynamics during the snowmelt period [Groffman et al., 2001; Hardy et al., 2001].

[6] Improved modeling work can advance our understanding of the mechanisms that relate the soil freeze/thaw status, hydrological processes, and biogeochemical cycles at high latitudes. Randerson et al. [2009] showed that current land surface models underestimate the magnitude of net carbon uptake during the growing season in boreal forest regions. This deficiency is partially due to inadequate modeling treatment of frozen soil hydrology [Lawrence et al., 2011]. Tang and Zhuang [2010] suggested that coupling hydrological and soil thermal dynamics in Alaskan terrestrial ecosystems is important for modeling ecosystem dynamics. However, most land surface models, such as the Community Land Model version 4 (CLM4) [Oleson et al., 2010; Lawrence et al., 2011] and the Community Noah Land Surface Model [Ek et al., 2003], do not properly address the snowpack insulating effects.

[7] This study aims to improve our understanding of the relationships between snow cover fraction (SCF), soil freeze/thaw status, and hydrological processes. To this end, we will (1) investigate landscape freeze/thaw processes at high latitudes and (2) clarify how soil impermeability regulated by soil freeze/thaw processes affects soil hydrological processes. Following Hardy et al. [2001], we postulated two soil impermeability parameterizations: (1) the unsaturated soil under the snow cover is permeable; and (2) impermeable areas are regulated by an exponential term that decreases as the snow cover area increases in size and as the temperature in the top soil layer rises. We first validated model-simulated landscape freeze/thaw status against observations from microwave remote sensing. In addition, a group of discharge/runoff observations and the quality of two meteorological forcing data sets were investigated. Finally, a series of numerical experiments were carried out based on the two parameterizations.

[8] The structure of this paper is as follows. Section 2 describes the data, model, and model parameterizations. Section 3 is an analysis of model-simulated landscape freeze/thaw status and runoff. Section 4 discusses the weaknesses and limitations of this study and the possible ways to improve the hydrological simulations in high-latitude regions and summarizes the main findings of this study.

2 Methodologies

2.1 Data Sets

2.1.1 Atmospheric Forcing Data

[9] To force the model at a global scale, we used the Modern Era Retrospective-Analysis for Research and Applications (MERRA) meteorological data (0.66° latitude × 0.33°longitude) for 2003–2009 [Rienecker et al., 2011] and the National Centers for Environmental Prediction/National Center for Atmospheric Research reanalysis (1.915° latitude × 1.875° longitude) for 2000–2004 [Qian et al., 2006]. These forcing data are observation-derived fields including precipitation, air temperature, air pressure, specific humidity, shortwave radiation, and wind speed. We used the two data sets to force CLM4 and compared model outputs associated with the hydrological cycle. The model outputs obtained from MERRA were eventually used to evaluate the hydrological processes of the default model and the parameterizations in high-latitude regions. MERRA was chosen because it has significant improvement in precipitation data quality and covers the same time period as other available data sets that are used for validating the model. These data sets include SSM/I FT-ESDR, in situ observations of discharge from the ArcticRIMS projects, discharge derived from terrestrial water storage variations (TWS) measured by the Gravity Recovery and Climate Experiment (GRACE) satellites and net precipitation from an atmospheric reanalysis data set (precipitation minus evapotranspiration, or P − ET) [Landerer et al., 2010]. Detailed information on these data sets is summarized in sections 2.1.2 and 2.1.3.

2.1.2 Special Sensor Microwave/Imager (SSM/I) Observed Landscape Freeze/Thaw Earth System Data Record (FT-ESDR)

[10] The daily SSM/I FT-ESDR is derived from radiometric brightness temperature (Tb) time series at 37 GHz (V-pol) frequency, which is acquired from SSM/I on board the Defense Meteorological Satellite Program (DMSP) polar orbiting satellite series [Kim et al., 2011]. This data set spans the period from 1988 to 2007 with 25 km × 25 km spatial resolution. It was developed for quantifying the freeze/thaw dynamics over vegetated land areas, where ecological processes are primarily constrained by seasonal frozen temperatures. Basically, FT-ESDR uses specific digital numbers (DN) to describe the landscape status: frozen, nonfrozen, transitional (A.M. frozen and P.M. thawed), and inverse transitional (P.M. frozen and A.M. thawed) (Table 1). The daily SSM/I FT-ESDR was used in this study to validate the landscape freeze/thaw status represented by CLM4.

Table 1. Landscape Freeze/Thaw Classification From SSM/I FT-ESDR
ClassificationFT DN
Frozen (A.M./P.M. frozen)0
Thawed (A.M./P.M. thawed)1
Transitional (A.M. frozen and P.M. thawed)2
Inverse Transitional (P.M. frozen and A.M. thawed)3
Masked (permanent ice, nonvegetated and urban area)254
100% open water255

2.1.3 River Basin Discharge

[11] This study used three sources of data to validate the discharge simulations; our discussions follow the timeline of data acquisition. The first data set is the New Hampshire-Global Runoff Data Center (UNH-GRDC) monthly composite runoff, which combined observed river discharge with output from a water balance model driven by observed meteorological data. The GRDC climatology runoff not only preserves the accuracy of the discharge measurements but also maintains the spatial and temporal distribution of simulated runoff [Fekete et al., 2000; Niu and Yang, 2006]. The second data set is in situ observations of discharge from the ArcticRIMS project (available at http://rims.unh.edu/data.shtml), which has near-real time monitoring of river discharge to the Arctic Ocean. The ArcticRIMS project does not have records on the Churchill-Nelson and Amur River basins. The third source of discharge data came from GRACE TWS variations [Landerer et al., 2010] and the Japanese 25 year Re-Analysis (JRA-25) derived net precipitation (P − ET) [Onogi et al., 2006], hereafter referred to as GRACE/JRA-25. Detailed information on this discharge data set was discussed in Landerer et al. [2010].

2.2 Model Descriptions

[12] CLM4 [Oleson et al., 2010; Lawrence et al., 2011] is the land component of the Community Earth System Model [Gent et al., 2011]. The substantial improvements since CLM3 [Dickinson et al., 2006] are summarized in Lawrence et al. [2011]. CLM4 includes three modules: (1) CLM4SP, which is CLM4 with satellite phenology, (2) CLM4CN, which is CLM4 with explicit carbon and nitrogen balances, and (3) CLM4CNDV, which is CLM4CN integrated with a global dynamic vegetation model. All three CLM4 modules use the same hydrological parameterizations, based on a Simple TOPMODEL-Based Runoff Scheme (SIMTOP) [Niu et al., 2005].

[13] In SIMTOP, surface runoff consists of overland flow as a result of saturation excess (Dunne runoff) and infiltration excess (Hortonian runoff) mechanisms [Oleson et al., 2010]:

display math(1)

where qliq, 0 is liquid precipitation reaching the ground plus any melt water from snow (kg m−2 s−1) and qinfl, max is the maximum soil infiltration capacity (kg m−2 s−1). The fractional impermeable area (fsat) is expressed as

display math(2)

where fmax is the maximum saturated fraction for a grid cell, which is defined as the discrete cumulative distribution function of the topographic index when the mean water table depth of the grid cell is zero [Oleson et al., 2010], fover is the decay factor, z is the grid-cell-mean water table depth, and the impermeable fraction (ffrz,i) in each soil layer is parameterized as a function of soil ice content in a layer [Niu and Yang, 2006; Swenson et al., 2012]:

display math(3)

where θsat and θice are soil porosity and partial volume of ice, and α (= 3.0) is an adjustable scale-dependent parameter. Equation (1) suggests that when qliq, 0 is smaller than qinfl, max (the Hortonian runoff is equal to zero), model-simulated surface runoff will decrease as a result of reduced fractional impermeable area. However, when Hortonian runoff is greater than zero, runoff depends on fsat and qinfl, max. Here, qinfl, max is related to three factors: fsat, the partial volume of liquid water, and the effective soil porosity [Oleson et al., 2010].

[14] In CLM4, surface runoff is associated with infiltration into the surface soil layer, which is defined as

display math(4)

where qseva is the evaporation of liquid water from the top soil layer [Oleson et al., 2010]. Therefore, infiltration into the surface soil layer will increase with decreased surface runoff, and vice versa.

[15] Using the models discussed above, this paper is an investigation of the impacts of fsat on high-latitude hydrological cycles in CLM4SP. The model resolution used in this study is 0.9° latitude × 1.25° longitude.

2.3 Landscape Freeze/Thaw Status Calculations

[16] CLM4 calculates soil ice (Qice, soil, kg m−2), snow ice (Qice, snow, kg m−2), and canopy interception (Qintr, mm). In order to investigate the landscape freeze/thaw status simulated by the model, the landscape frozen fraction (LFF; ffrozen, land, %) was calculated based on the normalized soil ice in the first soil layer (NQice, soil, 1), snow ice (NQice, snow), canopy interception (NQintr, ice) when the vegetation temperature is lower than 273.15 K, and leaf area index (LAI, NLAI):

display math(5)

where fveg is fractional vegetated area and fsno is the SCF in each grid cell. Generally, the frozen lake/wetland (i.e., the lake/wetland on land) without snow cover is also a component of frozen landscape. Based on the CLM4 surface data sets, the total area of lake and wetland takes about 1.2% of the global land area. Therefore, we did not consider the frozen lake/wetland without snow cover in equation (5). CLM4 calculates SCF on frozen lake/wetland; hence, snow ice over lake/wetland is also considered in this study.

[17] The LFF obtained by equation (5) was validated against the landscape freeze/thaw status observed by SSM/I FT-ESDR. Since SSM/I FT-ESDR uses specific values (Table 1) rather than percentage values to represent the landscape freeze/thaw status, it cannot be directly compared with the CLM4 calculations. CLM4 was run with an hourly time step, and the mean ffrozen, land values were computed from 12:00 A.M. to 11:00 A.M. (hereafter referred to as ffrozen, land, am), from 12:00 P.M. to 11:00 P.M. (hereafter referred to as ffrozen, land, pm), and from 12:00 A.M. to 11:00 P.M. (hereafter referred to as ffrozen, land, avg) of each day. We classified the ffrozen, land values into four groups: (a) frozen, ffrozen, land, avg > 50%; (b) thaw, 0% ≤ ffrozen, land, avg < 20%, or 20% ≤ ffrozen, land, avg < 50% and |ffrozen, land, am − ffrozen, land, pm| < 20%; (c) transitional, 20% ≤ ffrozen, land, avg < 50% and |ffrozen, land, am − ffrozen, land, pm| ≥ 20% and ffrozen, land, am ≥ ffrozen, land, pm; and (d) inverse transitional, 20% ≤ ffrozen, land, avg < 50% and |ffrozen, land, am − ffrozen, land, pm| ≥ 20% and ffrozen, land, am < ffrozen, land, pm. The ffrozen, land, avg values that are equal to 50% were treated as missing values. Based on this classification, the SSM/I FT-ESDR observed and CLM4-estimated LFF was compared at a global scale. In order to calculate SSM/I FT-ESDR observed frozen fraction values in the selected two regions discussed in the following sections, we used the number of grid cells with the frozen status to divide the total number of the grid cells in each selected region. We defined the results as the observed LFF. Consequently, the LFF based on the CLM4-simulation was validated against SSM/I FT-ESDR regionally.

2.4 Parameterization of the Snow Cover Area

[18] Besides the control simulations (hereafter referred to as CTL), we did another two numerical experiments to explore the relationship between soil impermeability, which is controlled by soil ice amount, and hydrological processes in cold regions. Hardy et al. [2001] found that in temperate hardwood forests, soil freezing increased in mild winters with low snowfall. In the first experiments of this study, we reduced soil impermeability by assuming that the impermeable fraction decreases linearly with the augmentation of the snow cover:

display math(6)

Basically, this modification is explained as follows: saturated areas can overlap with the snow cover, while unsaturated soil under the snow cover is permeable as a result of the snow cover insulating effects. In other words, the impermeable area of each grid cell includes the frozen area, which is snow free, and the saturated area. This experiment is named as IMPERF1.

[19] Since in cold regions soil will freeze before snow is accumulated on the ground, IMPERF1 may overestimate the permeable area. Therefore, we used both soil temperature in the top soil layer and SCF to regulate the impermeable area and put these two judgments into an exponential term. This experiment is named as IMPERF2:

display math(7)

where T(c,1) is the soil temperature in the first soil layer. In this experiment, when the top layer soil temperature is lower than the freezing point (273.15 K), the impermeable areas decrease as the snow cover area increases in size and as the temperature in the top soil layer rises; otherwise, ffrz, new follows IMPERF1.

3 Results

3.1 Landscape Freeze/Thaw Status

[20] We used SSM/I FT-ESDR to evaluate the CLM4-simulated landscape freeze/thaw status, which influences soil permeability. Based on surface vegetation distributions, two rectangular regions located at 60°N–75°N, 75°E–180°E (hereafter referred to as Region 1) and 45°N–60°N, 22.5°E–75°E (hereafter referred to as Region 2; Figure 1) were chosen to study the landscape freeze/thaw status in high latitudes. Generally, the CLM4 surface data suggest that Region 1 is dominated by boreal broadleaf deciduous shrub, needleleaf deciduous (larch) trees, and C3 arctic grass, and Region 2 is mainly covered by boreal and temperate needleleaf evergreen trees, C3 nonarctic grass, and boreal broadleaf deciduous trees.

Figure 1.

CLM4 vegetation types in cold regions: (a) boreal broadleaf deciduous shrub, (b) needleleaf deciduous (larch) trees, (c) C3 arctic grass, (d) boreal and temperate needleleaf evergreen trees, (e) C3 nonarctic grass, and (f) boreal broadleaf deciduous trees. The solid-line rectangles represent Region 1, and the dashed-line rectangles represent Region 2.

[21] In this study, 12 October, 1 November, 21 November, 11 December, and 26 December in 2003, and 10 January, 25 January, 9 February, 2 March, 22 March, 11 April, and 1 May in 2004 were selected to investigate the landscape freeze/thaw status. SSM/I FT-ESDR suggests that the frozen area increases from 12 October to 11 December, reaches the maximum values from 26 December to 9 February, and decreases from 2 March to 1 May for the two selected regions. Region 1 is frozen for a longer period of time than Region 2 as a result of latitude differences (Figure 2). On 12 October and 1 May, Region 2 is almost unfrozen (Figures 2a and 2h). On the selected days, an inverse transitional landscape type was seldom observed globally (Figure 2). The spatial distribution of landscape freeze/thaw status is slightly different year by year, but we can obtain the same conclusions in terms of the records on the same days in other years (figure not shown).

Figure 2.

SSM/I FT-ESDR obtained landscape freeze/thaw status on (a) 12 October 2003, (b) 1 November 2003, (c) 21 November 2003, (d) 11 December 2003, (e) 26 December 2003, (f) 10 January 2004, (g) 25 January 2004, (h) 9 February 2004, (i) 2 March 2004, (j) 22 March 2004, (k) 11 April 2004, and (l) 1 May 2004. Areas in white are outside of the FT-ESDR domain [Kim et al., 2011].

[22] According to equation (5), we calculated the LFF by using both MERRA and the forcing data discussed by Qian et al. [2006] (hereafter referred to as MCTL and QCTL, respectively). The results indicate that the total frozen area obtained by QCTL is smaller than that of MCTL at high latitudes, but the spatial and temporal distributions of QCTL and MCTL are similar (figure not shown). Actually, MCTL performs better in capturing runoff and TWS seasonalities, which are discussed in the following sections. Therefore, we investigated the LFF obtained from MCTL on the same days that selected for SSM/I FT-ESDR analyses.

[23] Compared to SSM/I FT-ESDR, CLM4 generally reproduces the landscape freeze/thaw pattern globally and properly captures the landscape freeze/thaw status on the four selected days in boreal winter. The frozen time period of SSM/I FT-ESDR is shorter than that of CLM4-simulation in both Region 1 and Region 2. Specifically, the CLM4-simulated landscape frozen area is larger than that of SSM/I FT-ESDR on 12 October and 1 November and is similar to that of SSM/I FT-ESDR on the other selected 10 days in Region 1. On 1 and 21 November, 11 December, 22 March, and 11 April, the CLM4-based landscape frozen area is larger than that of SSM/I FT-ESDR in Region 2. Generally, the CLM4-simulated transitional area on these 12 days is smaller than that of SSM/I FT-ESDR. The inverse transitional area obtained from the CLM4 simulation is very limited, which is similar to that from SSM/I FT-ESDR (Figures 2 and 3).

Figure 3.

CLM4-simulated landscape freeze/thaw status on (a) 12 October 2003, (b) 1 November 2003, (c) 21 November 2003, (d) 11 December 2003, (e) 26 December 2003, (f) 10 January 2004, (g) 25 January 2004, (h) 9 February 2004, (i) 2 March 2004, (j) 22 March 2004, (k) 11 April 2004, and (l) 1 May 2004.

[24] We also compared observation-based and CLM4-simulated LFF on the selected days in boreal fall and spring. In Region 1, the regional-averaged LFF values simulated by CLM4 gradually increase from 27% to 45% from 12 October to 11 December, while the SSM/I FT-ESDR values are 8%, 49%, 73%, and 72% on 12 October, 1 November, 21 November, and 11 December, respectively. The regional mean LFF values of the CLM4-simulation are 66%, 73%, 78%, and 75% on 2 March, 22 March, 11 April, and 1 May, respectively. Correspondingly, the SSM/I FT-ESDR values are 74%, 73%, 73%, and 68%, respectively on those four spring days. While the CLM4 and SSM/I FT-ESDR mean values of LFF in the boreal spring are 73% and 72%, respectively, their variations on the 4 days are different. The CLM4-simulated LFF value on 2 March is smaller than that on the following 3 days, while SSM/I FT-ESDR suggests a decreasing trend in the LFF on these 4 days. In Region 2, the CLM4-simulated LFF increases from 12 October to 11 December and decreases from 2 March to 1 May (Figure 3), similar to SSM/I FT-ESDR. CLM4 also demonstrates the nonfrozen landscape status on 12 October and 1 May (Figures 3a and 3i). The CLM4-simulated fraction values on those same dates in boreal spring are 59%, 49%, 25%, and 8%, respectively, in Region 2. These values for SSM/I FT-ESDR are 54% and 14% on 2 and 22 March, and they decrease to less than 0.3% after 11 April, suggesting that the landscape thawing process represented by SSM/I FT-ESDR is faster than that of CLM4. The SSM/I FT-ESDR LFF values in Region 2 after 11 April also imply the temporal uniform feature of this data set, which is mentioned by Kim et al. [2011] and further discussed in the following paragraphs. Even with some minor spatial distribution differences year by year, we can obtain the same conclusions from the model outputs on the same dates in other years (figure not shown).

[25] In this study, we also calculated the 5 year (2003–2007) mean daily snowfall (Psnow), normalized SCF (Nfsno), normalized snow ice (NQice, snow), and normalized snow depth (NDsnow) in Region 1 and Region 2, respectively (Figure 4). The seasonal variations of the four variables suggest that the CLM4-simulated SCF reaches peak values in November and starts to shrink in late April in Region 1 (Figure 4a). Actually, the SCF reproduced by the model is higher than that of MODIS observations in Region 1 (figure not shown). Therefore, snow ice is the dominant component of the frozen landscape during November to late April in this region (equation (5)). By evaluating NQice, snow in Region 1, we found that snow ice increased during October to early May, which had a quasi-linear relationship with NDsnow (Figure 4a). When SCF starts to reduce in late April, snow depth and snow ice are still growing as a result of snowfall increase starting from late April (Figure 4a). We speculate that as temperature increases, snowfall regions move northward after late April, and the shift of the snowfall region and temperature increase result in the northward shrinking of SCF. In addition, increased snow depth is associated with snowfall, and the growth of snow ice is associated with both snowfall and snow density. Actually, Niu and Yang [2007] showed that the SCF–snow depth relationship varies with seasons as a function of snow density. Therefore, we conclude that the snow events in Region 1 starting in late April result in the snow ice growth, which is a function of the CLM4-simulated LFF. In Region 2, Psnow, Nfsno, NQice, snow, and NDsnow begin to decrease in early March (Figure 4b) and vary similarly to the CLM4-simulated LFF.

Figure 4.

CLM4-simulated the 5 year (2003–2007) daily mean atmospheric snow (Psnow), normalized SCF (Nfsno), normalized snow ice (NQice, snow), and normalized snow depth (NDsnow) in (a) Region 1 and (b) Region 2.

[26] In order to further explore the relationship between the LFF and the SCF described by CLM4, we summarized the monthly (Figure 5) and daily variations (Figure A1) of these two variables in the two selected regions from 2003 to 2007. As discussed before, the SCF in Region 1 reaches maximum from November to late April in the next year, during which the LFF gradually increases from 30% to 85%. In the snow melting seasons (May and June), the LFF and the SCF have a linear relationship. Region 1 is almost unfrozen in July and August, and the landscape starts to freeze in September (Figures 5a and A1a). In addition, the freezing rate in September is higher than that in October (Figure 5a). We infer that this difference is associated with a quicker seasonal snow accumulation rate in September than in October (Figure 4a). In Region 2, the SCF reaches maximum in January and February, and the LFF grows from 40% to 75% in these 2 months. From March to April, the LFF and the SCF decrease simultaneously, which is earlier than that in Region 1. From May to September, Region 2 is almost ice free (Figures 5b and A1b). Overall, the LFF and SCF relationship shows a step-function pattern in Region 1 and tends to be linear in Region 2. This difference is associated with the snow ice increase when the SCF reaches its maximum values (Figure 4a). The increase of snow ice results in the LFF growth from November to late April in Region 1. Compared to Region 1, the latitude of Region 2 is lower, which is related to a relatively warmer temperature. Consequently, the snow accumulation season of Region 2 starts 1 month later than that of Region 1, and the snow ice varies linearly with the SCF (Figure 4b). Therefore, the LFF and the SCF vary in a quasi-linear pattern.

Figure 5.

CLM4-simulated the relationship between LFF and SCF in (a) Region 1 and (b) Region 2 during 2003–2007.

[27] The spatial distribution of SSM/I FT-ESDR (Figure 2) and the relationships between the CLM4-simulated and the SSM/I FT-ESDR observed LFF prove the spatial and temporal uniform feature of SSM/I FT-ESDR (Figure 6). Actually, the SSM/I FT-ESDR freeze/thaw patterns are more spatially and temporally uniform during winter (summer) than those in the freeze/thaw transitional period. In Region 1, the SSM/I FT-ESDR LFF starts to increase rapidly in October and reaches its maximum value in November. In contrast, the CLM4-simulated LFF increases slowly in October and November and grows quickly from December to March (Figure 6a). The variation differences between observation and model simulation can be explained by two reasons: (1) observations indicate the frozen status as soon as the brightness temperature in each model grid cell is below the freezing point, and (2) the CLM4-calculated snow ice continuously increases even though the SCF reaches its maximum values. In Region 1, both SSM/I FT-ESDR and CLM4 indicate the reduced LFF starting in April, and the thawing process represented by SSM/I FT-ESDR is faster than that of CLM4, which is also suggested by the LFF values. Actually, the snow/soil melting processes in CLM4 are determined by the snow/soil temperature and the excess/deficit of energy in each snow/soil layer [Oleson et al., 2010], and the numerical schemes used for calculating temperatures and energy fluxes may result in errors of snow/soil melting rates. In addition, SSM/I FT-ESDR also has uncertainties in representing the landscape freeze/thaw status Kim et al. [2011]. We infer that these two aspects are closely related to the melting rate differences of SSM/I FT-ESDR and CLM4. In Region 2, the CLM4-simulated and the SSM/I FT-ESDR observed LFF vary in a quasi-linear pattern during the snow accumulation seasons (October to March), and the CLM4-simulated landscape thawing rate of April in Region 2 is faster than that in Region 1 (Figure 6). These variations are associated with differences in climate conditions between Region 1 and Region 2.

Figure 6.

The relationship between CLM4-simulated and SSM/I FT-ESDR obtained LFF in (a) Region 1 and (b) Region 2 during 2003–2007.

3.2 Fractional Impermeable Area

[28] In order to understand soil impermeability variations, which are associated with soil freeze/thaw processes, we calculated the 7 year (2003–2009) mean of fsat obtained from CTL and from the modified cases (IMPERF1 and IMPERF2). The differences between IMPERF1 and CTL in March, April, May, and June (MAMJ) indicate that owing to the model-simulated high SCF in March, April, and May (MAM; figure not shown), IMPERF1 results in significant decrease of fsat (up to 100%) during this time period (Figures 7a, 7b, and 7c and equation (6)). Compared to CTL, soil impermeability obtained from IMPERF1 has some minor increase in May (Figure 7c), and it enlarges northward in June, with up to 8% fsat enlargement (Figure 7d). Most of the soil-impermeability-amplified areas calculated by IMPERF1 locate in the regions with the SCF less than 5% (figure not shown).

Figure 7.

Seven-year (2003–2009) mean fractional impermeable area (%) differences between (a) IMPERF1 and CTL in March, (b) IMPERF1 and CTL in April, (c) IMPERF1 and CTL in May, (d) IMPERF1 and CTL in June, (e) IMPERF2 and CTL in March, (f) IMPERF2 and CTL in April, (g) IMPERF2 and CTL in May, and (h) IMPERF2 and CTL in June.

[29] Compared to IMPERF1, IMPERF2 also simulates reduced soil impermeability in MAMJ. However, the total area with soil impermeability decrease is not as large as that of IMPERF1, and the fraction of the decrease is up to 60%. From March to May, the soil-impermeability-reduced area expands northward, while the soil-impermeability-amplified area develops southward (Figures 7e, 7f, and 7g). From May to June, the area with increased soil impermeability calculated by IMPERF2 is smaller than that of IMPERF1 (Figures 7c, 7d, 7g, and 7h). Overall, IMPERF1 and IMPERF2 enlarge/reduce soil impermeability in regions with the SCF less/greater than 5% in May and June.

3.3 Discharge in the Six Largest River Basins in Cold Regions

[30] In order to understand the hydrological impacts of soil impermeability, we evaluated the runoff simulated by the CLM4 control run, IMPERF1, and IMPERF2. We first compared the GRDC, ArcticRIMS, and GRACE/JRA-25 data sets in the six largest river basins in cold regions (Lena, Yenisei, Mackenzie, Ob, Churchill-Nelson, and Amur River basins) to study the uncertainties of observed runoff/discharge. In order to compare the GRDC runoff with the other two data sets, we used the UNH-GRDC observed runoff fields to mask river basins and converted runoff to discharge (with the volumetric unit; Figure 8). The results indicate that in the Mackenzie, Ob, Churchill-Nelson, and Amur River basins, the discharge magnitude of GRDC is considerably higher than that of GRACE/JRA-25 and ArcticRIMS (Figures 8c, 8d, 8e, and 8f). In the Mackenzie, Ob, and Churchill-Nelson River basins, the discharge obtained from the GRDC data set peaks earlier than that observed by ArcticRIMS and derived from GRACE/JRA-25. Therefore, the observational uncertainties are more significant in those four river basins than in the Lena and Yenisei River basins (Figures 8a and 8b). In most basins, the annual mean of GRACE/JRA-25 usually has a secondary discharge peak, which is absent in GRDC and ArcticRIMS.

Figure 8.

GRDC discharge (runoff based), ArcticRIMS discharge, and GRACE/JRA-2 discharge in the (a) Lena, (b) Yenisei, (c) Mackenzie, (d) Ob, (e) Churchill-Nelson, and (f) Amur River basins in cold regions. GRDC denotes the UNH-GRDC runoff climatology, ARCTICRIMS denotes a 7 year (2003–2009) mean ArcticRIMS river discharge (no data in the Churchill-Nelson and Amur River basins), and GRACE/JRA-25 denotes a 7 year (2003–2009) mean discharge derived from GRACE TWS variations and JRA-25.

[31] Owing to the uncertainty of the flow velocity used in the CLM4 river transport model (RTM) [Swenson et al., 2012], we did not validate the CLM4-RTM-simulated discharge (Appendix B). The observational data shown in Figure 8 are real discharge or discharge calculated from runoff. We also transformed CLM4-simulated runoff to discharge by using the same method for GRDC runoff processing, which does not consider the river transport processes. The “discharge” mentioned hereafter refers to the converted results, which are based on CLM4-simulated runoff.

[32] Two sets of CLM4-simulated discharge based on MERRA and the forcing data discussed in Qian et al. [2006] were referred to as D-MCTL and D-QCTL, respectively. Both D-MCTL and D-QCTL were validated against the mean of GRACE/JRA-25 and ArcticRIMS (OBS1) in the Lena, Yenisei, Mackenzie, and Ob River basins and against GRACE/JRA-25 in the Churchill-Nelson and Amur River basins from 2003 to 2004 (Figure 9). Compared to OBS1 and GRACE/JRA-25, D-QCTL has earlier peaked discharge in all the six river basins. In the Lena, Yenisei, Mackenzie, and, Ob River basins, D-MCTL has later peaked discharge than D-QCTL (Figures 9a, 9b, 9c, and 9d), while the timing differences between D-QCTL and D-MCTL are not significant in the Churchill-Nelson and Amur River basins (Figures 9e and 9f). Both D-QCTL and D-MCTL have higher discharge magnitudes than observations in the six river basins. The correlation coefficient indicates that D-MCTL can better capture the discharge seasonality than D-QCTL in all the six river basins except in the Churchill-Nelson River basin (Figure 9).

Figure 9.

Comparison of observed and modeled monthly discharge (2003–2004) in the (a) Lena, (b) Yenisei, (c) Mackenzie, (d) Ob, (e) Churchill-Nelson, and (f) Amur River basins in cold regions. OBS1 represents the mean of GRACE/JRA-25 and ArcticRIMS in the Lena, Yenisei, Mackenzie, and Ob River basins; GRACE/JRA-25 represents the discharge from GRACE/JRA-25 in Churchill-Nelson and Amur River basins; D-QCTL represents discharge from Qian's data forced CLM4; D-MCTL represents discharge from the MERRA reanalysis forced CLM4. Also shown are the correlation coefficients (R) between OBS1 (or GRACE/JRA-25) and each experiment.

[33] We also validated the discharge obtained from both MERRA forced control run and the modified runs (the parameterizations discussed in section 2.4) against the mean of GRDC, ArcticRIMS, and GRACE/JRA-25 (OBS2) in the Lena, Yenisei, Mackenzie, and Ob River basins and that of GRDC and GRACE/JRA-25 in the Churchill-Nelson and Amur River basins (Figure 10). The shaded areas denote the observational uncertainty, which was obtained from the maximum and minimum values of the three observations. Generally, the 7 year (2003–2009) mean discharge from CLM4 also has secondary peaks, and the possible reason for this pattern is that a considerable amount of runoff bypassing the streams is not considered by the installed gauges [Syed et al., 2007]. Compared to the mean of the observations, the control run has earlier peaked discharge in five out of the six river basins. The Yenisei River basin is the exception, which is located in the center of the Siberian region (50°N–66.5°N, 60°E–140°E). We infer the insignificant discharge variation in the Yenisei River basin is associated with the relatively high discharge magnitude and soil moisture and temperature in the top soil layer in April and May (figure not shown). The discharge magnitude of the control run is significantly higher than the observations in the Lena, Yenisei, Mackenzie, and Ob River basins (Figures 10a, 10b, 10c, and 10d). This result is similar to the monthly comparisons during 2003–2004.

Figure 10.

Seven-year (2003–2009) mean monthly discharge in the (a) Lena, (b) Yenisei, (c) Mackenzie, (d) Ob, (e) Churchill-Nelson, and (f) Amur River basins in cold regions. OBS2 denotes the mean of GRDC, ArcticRIMS, and GRACE/JRA-25 data sets (the mean of GRDC and GRACE/JRA-25 in the Churchill-Nelson and Amur River basins). D-MCTL denotes model's control run, and IMPERF1 and IMPERF2 denote the parameterizations discussed in section 2.4. The shaded areas denote the uncertainty of the observations. Also shown are the correlation coefficients (R) between observations and each experiment.

[34] By assuming the unsaturated soil under the snow cover is permeable (following IMPERF1), the discharge magnitude in MAM is reduced in all the six river basins, and the discharge peak is postponed in Lena, Mackenzie, Churchill-Nelson, and Amur River basins (Figures 10a, 10c, 10e, and 10f). The peaked discharge magnitude of IMPERF1 is lower than that of D-MCTL in the Lena and Churchill-Nelson River basins (Figures 10a and 10e). The discharge seasonality of IMPERF2 is similar to that of IMPERF1, and the peak values of IMPERF2 are slightly higher than IMPERF1 in all the six river basins. The correlation coefficient suggests that IMPERF1 performs better than IMPERF2. Overall, the modified tests reduce the magnitude and improve the timing of the boreal spring (MAM) discharge, and the improvements are significant in the Lena, Mackenzie, Churchill-Nelson, and Amur River basins (Figures 10a, 10c, 10e, and 10f).

[35] In CLM4, runoff affects infiltration (equation (4)), which is associated with soil liquid water variations. The two parameterizations in this study result in infiltration capacity increase in MAM in each river basin, which is in accordance with the surface runoff decrease and baseflow increase during the same time period (figure not shown). Soil moisture and soil temperature also increase as a result of the infiltration changes. Soil water increase is up to 0.012 mm3 mm−3 in the 2–2.5 m soil layers in May, while soil temperature raise is up to 0.1°C in the 20 cm soil layers or deeper in the same month. Since the soil moisture and temperature changes and their impacts are not significant, we did not further discuss them in this study. In this study, the CLM4-simulated water storage change was validated against the GRACE TWS anomalies, and the detailed information is included in Appendix C.

4 Discussions and Conclusions

[36] Since SSM/I FT-ESDR has its limitations in reflecting soil freeze/thaw status under snow or thick surface organic layers [Kim et al., 2011], we used this data set to validate the model-simulated landscape freeze/thaw status rather than soil freeze/thaw status. This is the first time that this satellite data set is used to evaluate a land surface model. Since the SSM/I FT-ESDR observed landscape freeze/thaw status cannot be directly compared with the CLM4-simulated LFF, we classified the model outputs into the four categories shown in Table 1. Even though SSM/I FT-ESDR has observational uncertainties and the classification method has its limitations, we obtained reasonable comparison in this study. The landscape freeze/thaw status is reasonably represented by CLM4 in the regions north of 60°N in boreal winter and spring. The model-simulated LFF increases in this region during March and April, which is associated with the model-simulated high SCF and snow ice amount. Generally, the increased snowfall lasting from late April to late May results in the growth of snow depth as well as snow ice in areas that are too cold for snow melting in the regions north of 60°N. In the regions with low snowfall rates during the snow melting seasons, the model-simulated LFF decreases in MAM along with the decreased snow depth and snow ice. In the regions north of 60°N, the LFF values of SSM/I FT-ESDR and CLM4 are in the same range in boreal spring, whereas the decreasing rates of the SSM/I FT-ESDR fraction values south of 60°N are higher than those from CLM4 in Region 2. The landscape freeze/thaw status from SSM/I FT-ESDR is more spatially and temporally uniform than that of CLM4, which is associated with the snow accumulation and melting processes of the model.

[37] This paper used both MERRA and Qian et al.'s forcing to simulate runoff and TWS, and the results prove that MERRA is better in capturing the hydrological cycles of CLM4. In addition, we compared the seasonality of GRDC, ArcticRIMS, and GRACE/JRA-25 and found that the uncertainties of runoff/discharge observations were significant in cold regions, especially in the Ob, Churchill-Nelson, and Amur River basins. Therefore, attention must be paid to the strengths and weaknesses of both meteorological forcing and observed data sets. In addition, the flow velocity in CLM4-RTM is lower than the values suggested by other studies [Decharme et al., 2010], especially in cold regions. Since realistic spatial distributions of flow velocities are absent, we validated CLM4-simulated runoff, rather than CLM4-RTM-simulated discharge, against the observations.

[38] Hardy et al. [2001] found negative correlations between snow depth and soil frost in the temperate forest site. Zhang and Armstrong [2001] also suggested that soil under the seasonal snow cover is frozen before the snow cover develops, and it could eventually thaw. By following these assumptions, we linearly reduced the impermeable area along with the growth of SCF and obtained postponed boreal spring runoff in CLM4. However, this parameterization results in zero values in soil impermeability in boreal winter at high-latitude regions, which is unrealistic due to the existence of permafrost and seasonally frozen ground [Lawrence et al., 2012]. The parameterization that the impermeable area exponentially decreases as the snow cover area increases in size and as the temperature in the top soil layer rises results in the northward soil-impermeability-reduced area expansion in MAM and southward soil-impermeability-amplified area developments from May to June. Based on section 3.1, the CLM4-simulated LFF increases in the regions north of 60°N from March to April and decreases northward in Region 2 during MAMJ. Since the CLM4-simulated LFF is mainly determined by the snow ice amount when the regions are totally covered by snow, this parameterization follows the suggestions in the previous studies [Hardy et al., 2001; Zhang and Armstrong, 2001]. We conclude that by gradually reducing the soil impermeability during the period of LFF increase in March and April (Figures 7e and 7f), the discharge magnitude in March and April is reduced and the early peaked discharge is postponed, especially in the Lena, Mackenzie, Churchill-Nelson, and Amur River basins (Figures 10a, 10c, 10d, and 10g). The increased soil impermeability along with the decreased LFF in May and June is associated with the growth of discharge magnitude. Actually, the river basins with increased discharge magnitude in May and June (e.g., Yenisei, Mackenzie, Ob, and Amur; Figures 10b, 10c, 10d, and 10f) locate in the regions where soil impermeability is amplified in these 2 months. Hence, we speculate that the hydrological cycles of CLM4 (e.g., discharge) would be improved if a soil ice parameterization properly accounting for the soil insulation effect of the snow cover is included into the model. In addition, the two parameterizations introduced in this study result in small but favorable soil water and temperature increase in the Siberia region. Soil moisture increases in April, May, and June in the 1 m soil layer or deeper, while soil temperature increases in late April to September in the 20 cm soil layer or deeper.

[39] This paper focused on the impacts of impermeable areas on discharge variation in cold regions. In order to obtain further improvements, it is necessary to consider other processes that have not been included in the model yet. For example, the ponding processes [Verseghy, 1991; Swenson et al., 2012], which prevent the melt water from becoming discharge immediately, can reduce the discharge in the transitional seasons (winter to spring) and postpone the early peaked discharge. Topography (e.g., slope heights and directions) also influences the discharge amount. In addition, Wood et al. [2011] suggested the importance of high-resolution modeling on terrestrial water and energy cycles evaluation. Therefore, if high-resolution topography data are available, discharge simulation would also be improved.

Appendix A: The Landscape Frozen Fraction and Snow Cover Fraction Variations

[40] Besides the scatter plots (Figure 5) showing the relationship between the LFF and SCF, we also included the evolution curves of LFF and SCF in this study (Figure A1).

Figure A1.

The daily variation of LFF and SCF obtained from CLM4 in (a) Region 1 and (b) Region 2 during 2003–2007.

Appendix B: CLM4-RTM-Simulated Discharge

[41] In CLM4-RTM, the river flow velocity is equal to 0.35, which is a global constant. However, Decharme et al. [2010] demonstrated that global average flow velocities lower than 0.5 m s−1 are not reasonable and they should be close to 0.5–1 m s −1, especially for boreal regions where, for instance, the maximum flow velocity is higher than 1–1.5 m s−1 in May. Decharme et al. [2010] also suggested limiting the flow velocity to 0.5 m s−1 in the Mackenzie and Ob River basins. In order to investigate the seasonality of flow velocity controlled discharge, we performed sensitivity tests of flow velocities in the CLM4-RTM. We applied a flow velocity of 1.0 m s−1 in the Lena, Yenisei, Churchill-Nelson, and Amur River basins and a flow velocity of 0.5 m s−1 in the Mackenzie and Ob River basins, and then we evaluated the CLM4-RTM-simulated discharge seasonality with ArcticRIMS and GRACE/JRA-25 (Figure B1). The increase of flow velocity results in hydrographs that peak earlier and more sharply, which is consistent with the conclusions of Decharme et al. [2010]. The earlier peaked discharge in the Lena, Mackenzie, and Ob River basins are postponed by IMPERF2 (Figures B1a, B1c, and B1d), and the improvements in the other three river basins are not significant. Since flow velocity is associated with channel cross-sectional area, wetted perimeter, and roughness [Dingman, 2002], which are not available globally, these experimental based sensitivity tests are tentative. Therefore, developing flow velocities that have realistic spatial distribution is important to discharge simulations. For the reasons discussed above, the CLM4-simulated runoff was evaluated in this study.

Figure B1.

Comparison of observed and CLM4-RTM-simulated 7 year mean monthly discharge (2003–2009) in the (a) Lena, (b) Yenisei, (c) Mackenzie, (d) Ob, (e) Churchill-Nelson, and (f) Amur River basins in cold regions. The model was forced with MERRA data. MCTL represents the control simulation. FV represents the experiments with changed flow velocities, which are equal to 1.0, 1.0, 0.5, 0.5, 1.0, and 1.0 in the Lena, Yenisei, Mackenzie, Ob, Churchill-Nelson, and Amur River basins, respectively. FV&IMPERF2 denotes IMPERF2 simulations based on the changed flow velocities, which are the same to that of FV.

Appendix C: Water Storage in the Six Largest River Basins in Cold Regions

[42] The GRACE TWS anomalies were used to evaluate the CLM4-simulated TWS anomalies in the six largest river basins in cold regions. This global gridded data set, which spans from 2003 to 2009 at 1° × 1° resolution, is derived from temporal gravity field variations, which are observed by the GRACE satellites. Landerer and Swenson [2012] not only assessed the accuracy of GRACE TWS estimates but also inferred relationships between different regional time series that represent various spatial scales according to TWS variations simulated by land-hydrology models. These relationships were then utilized to extrapolate the GRACE TWS estimates from some certain spatial resolution to finer spatial scales. The gridded GRACE TWS fields processed in this way are corrected for signal-leakage and therefore allow users to average over arbitrary regions and compare the results to other gridded data (e.g., hydrological models or groundwater data sets) without the need to apply the GRACE filtering process to the data in the spherical-harmonic domain. The gridded gain factor and error maps are provided along with the GRACE TWS observations (The data are available at http://grace.jpl.nasa.gov).

[43] In this study, the CLM4-simulated water storage changes were validated against the GRACE TWS variations. We first evaluated the monthly TWS variations from both MERRA reanalysis and Qian's data forced CLM4 from 2003 to 2004 (Figure C1), and these two simulations were named as TWS-MCTL and TWS-QCTL, respectively. The results demonstrate that the seasonality and magnitude of both TWS-MCTL and TWS-QCTL are similar to the GRACE TWS variations in five out of the six river basins (TWS-QCTL in the Amur River basin is the exception). Based on the correlation coefficient, TWS-MCTL is significantly superior to TWS-QCTL in the Lena and Amur River basins (Figures C1a and C1f). In other river basins, TWS-MCTL likewise performs better than TWS-QCTL in capturing TWS variations but not as noticeably as in the previously mentioned two basins.

Figure C1.

Comparison of GRACE TWS estimates and CLM4-simulated monthly TWS (2003–2004) in the (a) Lena, (b) Yenisei, (c) Mackenzie, (d) Ob, (e) Churchill-Nelson, and (f) Amur River basins in cold regions. GRACE represents the mean of GRACE TWS estimations, TWS-QCTL represents TWS from Qian's data forced CLM4, and TWS-MCTL represents TWS from the MERRA reanalysis forced CLM4. Also shown are the correlation coefficients (R) between GRACE and each model run.

[44] We have also validated the 7 year (2003–2009) mean TWS variations obtained from MERRA-based control run and the modified cases, which are discussed in section 2.4 (Figure C2). The results indicate that the 7 year averaged seasonality of TWS-MCTL is similar to that of the GRACE TWS variations in the Lena and Yenisei River basins (Figures C2a and C2b). In the Mackenzie, Ob, and Churchill-Nelson River basins, the trough of TWS-MCTL is usually in August, while that of GRACE TWS is usually in September or October (Figures C2c, C2d, and C2e). The TWS seasonality from the model is significantly different from that of GRACE in the Amur River basin, which has the lowest mean latitude among the six river basins (Figure C2f). The TWS variations from IMPERF1 and IMPERF2 are similar to each other, and their seasonalities are similar to that described by TWS-MCTL. In Lena and Amur, the correlation coefficients between the modified tests and the GRACE observations are about 7–8% higher than that between CTL and GRACE observations. Moreover, model-simulated TWS variations not only depend on soil liquid water, but also are related to soil ice, water in unconfined aquifers, snow depth, and intercepted water, which are unchanged in the parameterizations. Hence, the CLM4-simulated TWS variations are not significantly influenced by the parameterizations.

Figure C2.

Seven-year (2003–2009) mean monthly TWS in the (a) Lena, (b) Yenisei, (c) Mackenzie, (d) Ob, (e) Churchill-Nelson, and (f) Amur River basins in cold regions. GRACE represents GRACE TWS estimations, TWS-MCTL represents TWS from the MERRA reanalysis forced CLM4, and IMPERF1 and IMPERF2 denote the TWS based on the parameterizations discussed in section 2.4. Also shown are the correlation coefficients (R) between GRACE and each model run.

Acknowledgments

[45] This work was funded in part by the U.S. National Aeronautics and Space Administration under the Interdisciplinary Science Project NNX11AE42G. The computations were performed on Texas Advanced Computing Center and the National Center for Atmospheric Research computer resources. F.W.L.'s work was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. We appreciate D.K. Hall for the valuable suggestions, E. Podest for introducing SSM/I FT-ESDR, and S.C. Swenson for providing the processed MERRA data. We also appreciate the comments from three anonymous reviewers that led to significant improvements of this paper. The authors would like to acknowledge the language help from Patricia A. Bobeck and Adam R. Bowerman.

Ancillary