1 Introduction

Vegetation plays an important role in snow cover processes in vegetated environments. More than 19% of seasonal snow cover in the Northern Hemisphere (NH) overlaps with vegetation (Link and Marks 1999; Rutter et al. 2009). Snow cover processes in vegetated environments are complex and highly sensitive to vegetation structure (Essery et al. 2008; Musselman et al. 2008; Varhola et al. 2010). During the snow accumulation period, vegetation intercepts snowfall, reducing snow mass on the ground under the vegetation (Zheng et al. 2018). During the snow ablation period, both shortwave and longwave radiation are often the primary energy source for snowmelt under vegetation since vegetation attenuates turbulent heat flux transfer between the vegetation canopy and the snow surface (Marks et al. 2008; Varhola et al. 2010; Webster et al. 2017). Previous studies have shown that snow cover under vegetation receives less shortwave radiation due to absorption and reflection by the vegetation (Rutter et al. 2009; Webster et al. 2016), and the associated heating may increase the longwave radiation emitted from the vegetation canopy to the snow surface (Pomeroy et al. 2009; Rutter et al. 2009). Other studies have also indicated that longwave radiation, especially from vegetation, is a large component of the total incoming radiation for snowmelt over middle to high latitudes during early spring (Bernier and Swanson 1993; Essery et al. 2008; Lawler and Link 2011; Webster et al. 2017). However, much research has been focused on the impact of radiation on the snow surface in vegetated environments at very small spatial scales, even point scales, since field measurements can be conveniently made at these scales (Sicart et al. 2006; Essery et al. 2008; Pomeroy et al. 2009; Musselman and Pomeroy 2017). The contribution of radiation to snowmelt in complex vegetated environments at regional scales and beyond is not well understood partly due to the scarcity and unavailability of measurements (Sicart et al. 2006; Webster et al. 2016). Hence, the effects of vegetation on snow at large scales need to be quantitatively investigated to improve understanding of the role of land surface processes in the climate system.

Satellite remote sensing plays an important role in investigating the variability of snow cover at regional and global scales (Pu and Xu 2009; Hori et al. 2017; Huang et al. 2018). Many satellite-derived snow products have been widely used for snow cover monitoring (Frei and Robinson 1999; Estilow et al. 2015). The snow cover fraction (SCF) products from the Moderate-Resolution Imaging Spectroradiometer (MODIS) are some of the best-quality data, with high spatial resolution (e.g., 500 m) and temporal frequency (e.g., daily), which are used to examine detailed changes in snow cover extent (Klein and Barnett 2003; Pu et al. 2007; Marchane et al. 2015). This dataset has high-quality coverage in vegetated areas (Klein et al. 1998; Klein and Barnett 2003; Wang et al. 2017), where it can be used to inspect the effects of vegetation on snow cover. However, the snow cover products retrieved from satellite remote sensing data provide limited information on the physical processes and mechanisms of changes in snow cover (Rodell and Houser 2004; Dong 2018). Therefore, although remote sensing data are indispensable for investigating snow cover variability, additional tools are needed to perform in-depth analysis of changes in snow cover in vegetated regions (Sturm 2015).

To overcome the shortcomings of remote sensing data, land surface models (LSMs) are usually used to explore physical snow processes. These models provide a valuable tool for quantifying radiative fluxes between vegetation canopies and the snow surface (Essery et al. 2008; Rutter et al. 2009). Nevertheless, LSMs are often unable to realistically simulate the distribution of the components in the energy balance equation. One issue is that inaccurately simulated vegetation emissivity is used to calculate outgoing longwave radiation (Jin and Liang 2006; Mullens 2013). Many LSMs, such as the “big leaf” model (Deardorff 1978), the Noah-Multiparameterization land surface model (Niu et al. 2011), and the Community Land Model (CLM) (Oleson et al. 2010, 2013), conceptualize the canopy as a single vegetation layer and assume that vegetation emissivity is a function of the sum of the exposed leaf area index (ELAI) and exposed stem area index (ESAI). This scheme produces unrealistically low vegetation emissivity when the sum of ELAI and ESAI (LSAI) is less than 2.5 (Mullens 2013). Mullens (2013) demonstrated that the whole canopy vegetation emissivity in CLM version 4.0 ranges from 0.40 to 0.78, with LSAI between 0.5 and 1.5 for cropland. Ogawa et al. (2003) found that observed vegetation emissivity values of individual leaves range from 0.93 to 0.99 based on 31 vegetation samples. In addition, the vegetation emissivity for the whole canopy varies between 0.9 and 1.0 based on thermograph measurements with a thermal infrared imaging radiometer (Pomeroy et al. 2009; Webster et al. 2016). Moreover, vegetation emissivity for the whole canopy should be higher than that for individual leaves due to multiple internal reflections resulting from the vegetation geometry (Van De Griend and Owe 1993; Jin and Liang 2006). The lower vegetation emissivity resulting from LSMs leads to underestimated outgoing longwave radiation from vegetation, resulting in overestimated vegetation temperature (Jin and Liang 2006; Mullens 2013). Thus, the vegetation emissivity schemes in many existing LSMs are unable to reflect reality, especially when LSAI is lower.

The objective of this study is to more accurately simulate the surface energy budget by developing a new vegetation emissivity scheme in CLM, investigate the effects of vegetation on snow cover processes, and improve snow cover simulations for the NH. In this paper, Sect. 2 introduces the atmospheric forcing data that were used to drive CLM and the in situ observations and satellite remote sensing data that were applied to evaluate the quality of the atmospheric forcing data and the performance of CLM, respectively. Section 3 describes CLM version 4.5 and the methods for calculating longwave radiation and SCF. Section 4 presents the new vegetation emissivity scheme and its effects on snow processes, and a discussion and conclusions are given in Sects. 5 and 6, respectively.

2 Data

2.1 Climate forcing data

Data from the Climatic Research Unit-National Centers for Environmental Prediction (CRU-NCEP) reanalysis products were used to drive CLM. These data include air temperature, total incident solar and longwave radiation, total precipitation, surface pressure, wind speed, and specific humidity. The CRU-NCEP data combine observations and reanalysis products and have a spatial resolution of 0.5° latitude by 0.5° longitude with a six-hourly time step, covering the period from 1901 through 2016. All the CRU-NCEP forcing variables were interpolated to a half-hourly time step for our simulations in this study. The incident solar radiation was interpolated using a diurnal function based on the cosine of the solar zenith angle, which produced a smoother diurnal cycle of solar radiation. Precipitation was interpolated with the nearest algorithm, and the other forcing variables were interpolated linearly. These interpolation methods are embedded in the original CLM. In this study, we used data from 1995 through 2012. The quality of the CRU-NCEP data has been verified by many studies (Barman et al. 2014; Sidike et al. 2016; Guo et al. 2018).

2.2 In situ observations

We collected in situ observations from Surface Radiation Budget Monitoring (SURFRAD) stations (Augustine et al. 2000) to assess the quality of the CRU-NCEP climate forcing dataset. The SURFRAD network provides high-quality surface radiation and air temperature measurements at time steps of one to three min from six stations in the United States, and the measurements from three of those six stations were selected for this study. The SURFRAD network was established in 1993 under the support of the National Oceanic and Atmospheric Administration’s Office of Global Programs.

2.3 Satellite remote sensing data

In this study, we used MODIS emissivity products (Version 6) from the Terra monthly global 0.05° Climate Model Grids for model evaluation (Wan 2013). Many previous studies have validated the accuracy of MODIS emissivity products using ground-based measurements (Wang et al. 2005) and have proved that the monthly mean MODIS emissivity dataset has a very high quality (Jin and Liang 2006; Wang et al. 2007; Wang and Liang 2009). The retrieved MODIS emissivity values are from individual spectral bands, but LSMs need broadband emissivity, which is an integration over the longwave water vapor window regions of 8-14 μm (Sobrino et al. 2005; Jin and Liang 2006; Wang and Liang 2009). Thus, MODIS emissivity values for wavelengths of 8.40–8.70 μm (Band 29), 10.78–11.28 μm (Band 31), and 11.77–12.27 μm (Band 32) were used to calculate broadband emissivity according to the method described in Wang and Liang (2009). The calculated MODIS broadband emissivity data at a spatial resolution of 0.05° were then interpolated onto 0.5° grids coincident with the spatial resolution of the CRU-NCEP forcing variables and CLM output.

Moreover, MODIS monthly mean SCF data from Terra/Aqua (Riggs and Hall 2015) were used to examine the performance of CLM. The MODIS SCF data have been assessed at different spatiotemporal scales (Pu et al. 2007; Arsenault et al. 2014; Crawford 2015; Marchane et al. 2015) and applied to observational and modeling studies at continental and global scales (Pu and Xu 2009; Zhang et al. 2014). We combined the Terra and Aqua data to reduce the data gaps in the MODIS SCF products caused by cloud cover (Parajka and Blöschl 2008; Xie et al. 2009; Huang et al. 2018). We also interpolated the MODIS SCF data at a 0.05° spatial resolution onto 0.5° grids. During this interpolation, the highest-quality MODIS SCF data marked in Riggs and Hall (2015) were selected to compute the mean SCF. We chose only areas between 0°N and 62°N, because MODIS sensors cannot receive optical information effectively in NH regions beyond 62°N during winter due to the polar night (Zhu et al. 2017).

3 Model and methodology

CLM version 4.5 is embedded in the Community Earth System Model version 1.2.1 developed by the National Center for Atmospheric Research. CLM includes a snow scheme of up to 5 layers, a 10-layer soil scheme, and a single-layer vegetation scheme. The SCF parameterization is used to represent the fractional snow cover for a given value of snow water equivalent in snow accumulation and melting seasons (Swenson and Lawrence 2012). A maximum of 16 plant functional types (PFTs) (Bonan et al. 2002) per model grid is included in CLM to better resolve the surface heterogeneity for complex vegetated environments. The spatial distribution and seasonal climatology of those PFTs for CLM are derived from MODIS satellite land-surface data products (Lawrence and Chase 2007).

3.1 Methodology

3.1.1 Vegetation emissivity and longwave radiation for the ground

Emissivity is defined as the ratio of the energy emitted by an object (gray body) at a given temperature to that emitted by a black body at the same temperature. The emissivity is computed by:

$$\varepsilon_{\lambda } = \frac{{E_{\lambda } \left( T \right)}}{{B_{\lambda } \left( T \right)}}$$
(1)

where \(B_{\lambda } \left( T \right)\) and \(E_{\lambda } \left( T \right)\) are the thermal emissions from a black body and a gray body at wavelength \({\lambda }\) and temperature T, respectively. In CLM, the emissivity of the ground \(\varepsilon_{g}\) is calculated with snow emissivity (\(\varepsilon_{sno}\)) and soil emissivity (\(\varepsilon_{soi}\)) weighted by SCF as follows (Oleson et al. 2013):

$$\varepsilon_{g} = \varepsilon_{soi} \left( {1 - SCF} \right) + \varepsilon_{sno} SCF$$
(2)

where \(\varepsilon_{soi} = 0.96\) for soil and wetland, 0.97 for glacier and lake, and \(\varepsilon_{sno} = 0.97\). The emissivity of vegetation \(\varepsilon_{v}\) is parameterized with ELAI and ESAI (Oleson et al. 2013):

$$\varepsilon_{v} = 1 - e^{{ - \left( {ELAI + ESAI} \right)/\bar{\mu }}}$$
(3)

where \({\bar{\mu }} = 1\) is the average inverse optical depth for longwave radiation. For vegetated surfaces, the downward longwave radiation for the ground surface under the vegetation is (Oleson et al. 2013):

$$L_{g} \downarrow = \varepsilon_{v} \sigma T_{v}^{4} + \left( {1 - \varepsilon_{v} } \right)L_{atm} \downarrow$$
(4)

where \({\sigma } = 5.67 *10^{ - 8} {\text{W/m}}^{2} *{\text{K}}^{4}\) is the Stefan–Boltzmann constant; \(T_{v}\) is the vegetation temperature; and \(L_{atm} \downarrow\) is the atmospheric downward longwave radiation acquired from CRU-NCEP forcing data. The first term in the above equation is the longwave radiation emitted by the vegetation directly to the ground surface. The second term is the atmospheric longwave radiation that is transmitted through the vegetation and reaches the ground surface. The upward longwave radiation from the ground surface is (Oleson et al. 2013):

$$L_{g} \uparrow = \varepsilon_{g} \sigma T_{g}^{4} + \left( {1 - \varepsilon_{g} } \right)L_{g} \downarrow$$
(5)

where \(T_{g}\) is the ground temperature. The first term in the above equation is the upward longwave radiation emitted by the ground. The second term is the longwave radiation reflected by the ground. The net longwave radiation flux for the ground surface is (positive toward the ground):

$$\overrightarrow {{L_{g} }} = L_{g} \downarrow - L_{g} \uparrow = \varepsilon_{g} L_{g} \downarrow - \varepsilon_{g} \sigma T_{g}^{4}$$
(6)

Therefore, vegetation emissivity affects the net longwave radiation for the ground surface under vegetation by changing the longwave radiation emitted from and transmitted through the vegetation.

3.1.2 Snow cover fraction

The SCF is calculated using a scheme based on snow water equivalent (Swenson and Lawrence 2012). Changes in SCF are parameterized separately for snow accumulation and depletion due to the different processes governing snowfall and snowmelt:

$$SCF^{n + 1} = 1 - \left[ {\left( {1 - \tanh \left( {k_{accum} q_{sno} \Delta t} \right)} \right)\left( {1 - SCF^{n} } \right)} \right]$$
(7)

where \(k_{accum}\) is a constant whose default value is 0.1; \(q_{sno} \Delta t\) is the amount of new snow falling on the ground, which is related to snowfall directly through the vegetation canopy and snow being unloaded from the canopy; \(SCF^{n}\) is the SCF from the previous time step; \(SCF^{n + 1}\) is the updated SCF; and \({\text{n}}\) is the time step. When snowmelt occurs, SCF is parameterized through a depletion curve:

$$SCF = 1 - \left[ {\frac{1}{\pi }\cos^{ - 1} \left( {2\frac{W}{{W_{\hbox{max} } }} - 1} \right)} \right]^{{N_{melt} }}$$
(8)

where \({\text{W}}\) and \(W_{max}\) are the current snow water equivalent and the maximum accumulated snow water equivalent, respectively. \(N_{melt}\) is a parameter that depends on the subgrid topographic variability within a grid cell, which is defined from the standard deviation of topography, \({\sigma }_{topo}\), by:

$$N_{melt} = \frac{200}{{{ \hbox{max} }\left( {10,{\sigma }_{topo} } \right)}}.$$
(9)

3.2 Experimental design

Two sets of offline CLM simulations for the NH over the period of 1995–2012 were performed with the prescribed satellite-derived phenology, driven by the CRU-NCEP atmospheric forcing dataset. The first set was conducted with the original CLM vegetation emissivity scheme, defined as the control run (CTL). The second set was carried out with the newly developed vegetation emissivity scheme (discussed in detail in Sect. 4.1), defined as the new run (NEW). The spatial resolution of the two sets of simulations was 0.5°, the same as that of the CRU-NCEP forcing data. The simulations from the two runs for the period of 1995–2001 were discarded as model spin-up, and the rest of the simulations (December 2002 through May 2012) with monthly output were analyzed. In this study, December, January, and February are defined as winter, while March, April, and May are defined as spring, since seasonal snow cover in the NH occurs mainly in these two seasons (Estilow et al. 2015; Wang et al. 2017).

4 Results

4.1 Development of a new vegetation emissivity scheme

An unreasonably low vegetation emissivity is produced by Eq. 3 in areas of lower canopy density (\(0 < LSAI < 2.5\)) (Mullens 2013). We chose grid cells with 100% vegetation coverage with LSAI less than 2.5 over the NH (0°N–62°N) and re-calculated the vegetation emissivity with Eq. 3. The calculated vegetation emissivity over these grid cells with full vegetation coverage can represent surface emissivity to a large extent. Figure 1 shows the 10-year averaged seasonal MODIS surface emissivity (black line with squares) and simulated vegetation emissivity with snow on the canopy (blue line with dots; see Supplemental Material 1) over these grid cells. The figure illustrates that the vegetation emissivity produced by Eq. 3 is remarkably lower than the MODIS data, and the minimum value reaches 0.70 in April.

Fig. 1
figure 1

Time series of the observed MODIS surface emissivity (black line with squares) and simulated vegetation emissivities from CTL (blue line with dots) and NEW (red line with triangles) averaged over grid cells with 100% vegetation cover

Full size image

In this study, to improve the vegetation emissivity simulations, we developed a new scheme to replace the current scheme in CLM. The maximum emissivity of different vegetation PFTs, \(\varepsilon_{pft,max}\), was adopted from Mullens (2013) (Table 1). Based on \(\varepsilon_{pft,max}\) and vegetation indices, we formulated a new vegetation emissivity scheme for CLM as follows:

$$\varepsilon_{v} = \left( {\frac{ELAI + ESAI}{{LAI_{pft,\hbox{max} } + SAI_{pft,\hbox{max} } }}} \right)^{{{\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 k}}\right.\kern-0pt} \!\lower0.7ex\hbox{$k$}}}} *\varepsilon_{pft,\hbox{max} }$$
(10)

where \(LAI_{pft,\hbox{max} }\) and \(SAI_{pft,\hbox{max} }\) are the maximum leaf and stem area indices for different PFTs, respectively, which come from the input surface data and are shown in Table 1. Note that \(k\) is an adjustable unitless parameter. We used irrigated crop as an example to test this new scheme. When \(k\) is changed from 30 (pink line) to 60 (green line), the minimum crop emissivity increases from 0.847 to 0.912, and the maximum emissivity stays at 0.981 for all \(k\) values (Fig. 2). Olioso et al. (2007) indicated that when plants are dry, leaf emissivity is about 0.9. Studies also show that the emissivity of branches or stems of other vegetation types has a similar minimum value (Chen and Zhang 1989; Haverd et al. 2007; Rice 2007; López et al. 2013). We analyzed the different PFTs and found that when \(k\) is 50, the minimum emissivity is 0.9. Therefore, \(k\) is set to 50 to simulate vegetation emissivity in this study. The CLM simulations with NEW were produced for the same period as with CTL (Fig. 1). We can see that the emissivity simulations with the new scheme (red line with triangles) agree more closely [root-mean-square-error (RMSE): 0.03] with the MODIS data than those with the original scheme (RMSE: 0.23). Clearly, the new scheme can better reproduce the observed emissivity.

Table 1 Maximum emissivity, leaf and stem area indices of different PFTs in CLM
Full size table
Fig. 2
figure 2

The horizontal axis is for the leaf and stem area indices, and the vertical axis is for vegetation emissivity. k represents the adjustable parameter in Eq. 10

Full size image

4.2 Impact of vegetation emissivity on the longwave radiation budget

We compared the longwave radiation budget simulated with the new emissivity scheme to that with the original scheme. Figure 3a, b show the spatial distribution of the difference in ground surface net longwave radiation between the NEW and CTL simulations for winter and spring, respectively. The new scheme (Eq. 10) generates larger net longwave radiation to the ground surface across our study regions (red areas in Fig. 3a, b) than the original scheme, especially in areas with low to medium canopy density (0 < LSAI < 2.5; Fig. 4). The larger net longwave radiation implies that more downward longwave radiation came from the vegetation due to the higher emissivity in the new scheme. Significant differences occur in the Indian Peninsula, North America, the areas between the Sahara Desert and the Gulf of Guinea, central Eurasia, and western and northeastern China. Meanwhile, the hourly values of such differences can reach 150 W/m2. Areas of dense canopy (LSAI above 2.5; Fig. 4), such as equatorial regions, experience unnoticeable changes in net longwave radiation on the ground surface in the two simulations, since the two schemes produce similar emissivity values over these areas. Figure 3c, d show the net longwave radiation relative differences, which are defined as the ratios of the differences between the NEW and CTL simulations (NEW-CTL) over the CTL simulation. We can see that the ground surface net longwave radiation experiences a significant increase with the new scheme relative to that with the original scheme. Moreover, we explored the impact of the new emissivity scheme on the radiative and turbulent fluxes in Supplemental Material 2 (Figure S1).

Fig. 3
figure 3

Differences (W/m2) a, b and relative differences (%) c, d in ground surface net longwave radiation between the NEW and CTL simulations for winter and spring over 2003–2012, respectively

Full size image
Fig. 4
figure 4

Geographic distribution of the vegetation LSAI used in CLM for winter (a) and spring (b)

Full size image

4.3 Impact of vegetation emissivity on snowfall to the ground surface

In addition to producing larger longwave radiation to the ground surface, the new vegetation emissivity scheme also causes changes in snow accumulation. Figure 5 shows the differences and relative differences in snowfall on the ground surface (SNOW_G) between the NEW and CTL simulations over winter and spring. A larger SNOW_G is seen over the mid-latitude regions with medium canopy density, such as North America, central Eurasia, and northeastern China, over the two seasons, indicating that more snow reaches the ground surface with the new scheme. SNOW_G, the snow water equivalent with a unit of mm, is the sum of the solid throughfall to the ground surface after vegetation interception and the solid canopy drips off the vegetation (Oleson et al. 2013). The difference in SNOW_G results primarily from the difference in solid canopy drips between the NEW and CTL simulations, since their difference in solid throughfall is negligible (see Supplemental Material 4, Figure S4). Compared with those from the original emissivity scheme, the higher emissivity values from the new emissivity scheme result in more energy release from the vegetation, leading to a lower vegetation temperature (Figure S3a). Thus, evaporation from the wetted vegetation is weakened with the new scheme (Figure S3b) (Dickinson et al. 1993; Jacobs et al. 2006). In the meantime, the colder vegetation attracts more dew/frost to the vegetation surface, when the vegetation temperature is below the dew/frost point of the adjacent air (Kabela et al. 2009). These processes are consistent with those found in actual data (Burrage 1972; Sudmeyer et al. 1994; Hanisch et al. 2015). Less evaporation and/or more dew/frost on the vegetation surface with the new scheme produce more canopy water. When canopy water exceeds the maximum amount of water the canopy can hold, the water drips off the canopy to the ground surface. Thus, we see a larger SNOW_G during the cold season with the new emissivity scheme (see Supplemental Material 4). Furthermore, the relative differences in SNOW_G (Fig. 5c, d) indicate that the increase in SNOW_G is not negligible with the new scheme; it can reach more than 30% over North America, northern India, and northeastern China due to more solid drips off the canopy to the ground surface during winter and spring. In addition, we can also see in Fig. 5 that the difference in SNOW_G is larger in winter than in spring due to more solid drips in winter.

Fig. 5
figure 5

Differences (mm/season) a, b and relative differences (%) c, d in SNOW_G between the NEW and CTL simulations for winter and spring over 2003–2012, respectively

Full size image

4.4 Impact of vegetation emissivity on the SCF

For this study, we examined how the new emissivity scheme affects SCF simulations when compared to the original scheme. Figure 6 shows the geographical distribution of the SCF difference and relative differences between the NEW and CTL simulations over winter and spring. Here, we changed the unit of the simulated SCF from a fraction to a percentage to make it consistent with that for the MODIS SCF product. In winter (Fig. 6a), reduced SCF exists over the regions south of 40°N with the new scheme, suggesting that more energy from the canopy to the ground surface dominates the changes in SCF. However, we can see an increase in SCF with the new scheme above 40°N, which can be attributed to a larger SNOW_G, as discussed above. The increase in SCF occurs mainly in the northeastern United States, and in northeastern China and the surrounding areas. In spring (Fig. 6b), a significant SCF reduction is seen in North America, central Eurasia, and the eastern Tibetan Plateau, indicating that stronger downward longwave radiation from vegetation plays a more important role in affecting SCF than SNOW_G. The relative differences in SCF (Fig. 6c, d) indicate that SCF changes with the two vegetation emissivity schemes are noticeable, especially in northern China during winter, and in North America, central Eurasia, and the eastern Tibetan Plateau during spring. The largest absolute SCF relative difference can reach about 30% relative to the SCF with the original emissivity scheme.

Fig. 6
figure 6

Differences (%) a, b and relative differences (%) c, d in SCF between the NEW and CTL simulations for winter and spring over 2003–2012, respectively

Full size image

The spatial differences in SCF RMSE between the NEW and CTL simulations show improvements in the SCF simulations (Fig. 7a, b). When compared to the MODIS SCF data, the new emissivity scheme mostly decreases the simulated SCF errors (blue areas) in most of the study regions over the two seasons. Obvious improvements (up to 24%) are seen in regions such as North America, central Eurasia, and the eastern Tibetan Plateau during spring. However, some areas experience an increase in SCF RMSE with the new scheme in winter and spring (red areas in Fig. 7a, b). Additional results show that 67% of the total area with the SCF RMSE changes (280 thousands km2) for winter experiences a decrease in simulated SCF RMSEs with the new scheme, and 84% of the total area of 396 thousand km2 for spring (Fig. 7c).

Fig. 7
figure 7

Differences (%) a, b in SCF RMSE between the NEW and CTL simulations for winter and spring over 2003–2012, respectively. Average MODIS SCF data are used as observations. c Areas with SCF RMSE changes between the two vegetation emissivity schemes during winter and spring (grid cells with SCF RMSE changes of less than 0.5% are ignored)

Full size image

5 Model forcing data validation

Atmospheric forcing data play an important role in affecting snow cover simulations. Thus, we examined the quality of air temperature, downward shortwave radiation, and downward longwave radiation from the CRU-NCEP forcing dataset by comparing with in situ observations from three SURFRAD stations: Bondville in Illinois (BND), Fort Peck in Montana (FPK), and Sioux Falls in South Dakota (SXF). These stations were selected due to the larger SCF RMSEs with CLM with the new vegetation emissivity scheme in winter, and all are located in flat prairie with native grasses. Figure 8 shows the comparison between the CRU-NCEP forcing dataset and SURFRAD observations for six-hourly air temperature, downward shortwave radiation, and downward longwave radiation for the 2005 winter. Results indicate that the air temperatures derived from the CRU-NCEP forcing dataset and observations are very close, and the squared correlation coefficients (R2) are larger than 0.80 at each station. The downward shortwave radiation from CRU-NCEP is slightly greater than observations. The biases are 12.95, 2.12, and 8.96 W/m2 for BND, FPK, and SXF, respectively. The R2 of downward longwave radiation between CRU-NCEP and observations is larger than 0.60 for all stations, but the values of downward longwave radiation from CRU-NCEP are much smaller than the observations. The biases are − 30.22, − 16.45, and − 21.18 W/m2 for BND, FPK, and SXF, respectively. We re-ran the model with the new vegetation emissivity scheme by replacing CRU-NCEP air temperature, downward shortwave radiation, and downward longwave radiation one by one and all three SURFRAD variables together with observations. Table 2 summarizes the simulated SCFs driven by the CRU-NCEP forcing dataset with both emissivity schemes and by the observations with the new emissivity scheme for the winter of 2005. All the simulated SCFs for the three SURFRAD stations are closer to the MODIS SCF data when compared to those with CTL and NEW. Therefore, the quality of atmospheric forcing data is also a critical factor affecting snow cover simulations.

Fig. 8
figure 8

Six-hourly air temperature, downward shortwave radiation, and downward longwave radiation from CRU-NCEP forcing and three SURFRAD in situ observations (BND, FPK, and SXF) for the 2005 winter

Full size image
Table 2 Simulated SCFs driven by the CRU-NCEP forcing dataset with the two emissivity schemes and by the in situ observations with the new emissivity scheme over three SURFRAD stations for the 2005 winter
Full size table

6 Conclusions

In this study, CLM with the newly developed vegetation emissivity scheme improves simulations of vegetation emissivity and SCF for the NH over winter and spring compared to the original emissivity scheme. Results show that simulated vegetation emissivity (0.70–0.80) with the original scheme in CLM (Eq. 3) is lower than MODIS emissivity (~ 0.98), especially during winter and spring. Therefore, we developed a new vegetation emissivity scheme as a function of the maximum emissivity and leaf and stem area indices of vegetation, which more realistically simulates vegetation emissivity (~ 0.95) than the original scheme. Results show stronger longwave radiation released from vegetation with the new scheme, resulting in a lower vegetation temperature and a higher probability of snowmelt than with the original scheme. In addition, the lower vegetation temperature weakens vegetation interception loss and enhances dew/frost onto the vegetation surface, leading to more water dripping off the canopy and thus a larger amount of snow onto the ground surface. Moreover, compared to the original emissivity scheme in CLM, the new scheme better reproduces the MODIS SCF mainly in the middle and high latitudes, including North America, central Eurasia, and the eastern Tibetan Plateau. The RMSE of the simulated SCF using the new scheme reduces over 187 and 334 thousand km2 of the total area with the SCF RMSE changes in winter and spring, respectively. In some regions, the new emissivity scheme degrades the SCF simulation, which is most likely related to inaccuracies in the atmospheric forcing data. Moreover, the new emissivity scheme generates less total upward longwave radiation and more surface net radiation than the original scheme. The sensible and latent heat fluxes experience increases over the areas with low to medium canopy density with the new emissivity scheme. The newly developed vegetation emissivity scheme provides a more effective tool to investigate the effects of vegetation on snow cover processes, resulting in a better understanding of land surface processes.