Journal of Geophysical Research: Atmospheres

Evaluation of surface albedo and snow cover in AR4 coupled climate models



[1] Surface albedo (ALB), snow cover fraction (SCF) and snow water equivalent (SWE) of state-of-the-art coupled climate models are compared and validated against ground-based and remote-sensed climatologies. Most IPCC AR4 climate models predict excessive snow mass in spring and suffer from a delayed spring snow melt while the onset of the snow accumulation is generally well captured. This positive SWE bias is mainly caused by too heavy snowfall during the winter and spring season. Seasonal cycles of snow cover area (SCA) at continental scales are captured reasonably well by most participating models. Two models clearly overestimate SCA over both Eurasia and North America. Year-to-year variations are reasonably well captured over both Eurasia and North America in winter and spring. The most pronounced underestimation in the interannual SCA variability is generally simulated during snow melt. The pronounced negative SCA trend that has been observed from 1979 to 2000 is only partly reproduced in the AR4 model simulations. Furthermore, the computed trends show a large spread among the models. Results from time slice simulations with the ECHAM5 climate model suggest that accurate sea surface temperatures are vital for correctly predicting SCA trends. Simulated global mean annual surface albedos are slightly above the remote-sensed surface albedo estimates. The participating AR4 models generally reproduce the seasonal cycle of the surface albedo with sufficient accuracy while systematic albedo biases are predicted over both snow-free and snow-covered areas, with the latter being distinctly more pronounced. The study shows that the surface albedo over snow-covered forests is probably too high in various state-of-the-art global climate models. The analysis demonstrates that positive biases in SCA are not necessarily related to positive albedo biases. Furthermore, an overestimation of area-averaged SWEs is not necessarily related to positive SCA anomalies since the relationship between SWE and SCF is highly nonlinear.

1. Introduction

[2] Snow cover plays a key role in the climate system as it largely affects both the energy balance, through the high reflectivity of snow, as well as regional water budgets and, thus, evaporation and the hydrological cycle. Snow is a very useful diagnostic parameter since, in order to correctly model snow thickness, both the temperature and precipitation distribution need to be realistic. Snow depth exhibits a strong annual cycle with a distinct interannual variability at mid latitudes. Over 50% of Eurasia and North America can be seasonally covered by snow [Robinson et al., 1993]. Numerous studies have shown the importance of snow for weather forecasts as well as for climate simulations from local to global scales. Barnett et al. [1989] confirmed the sensitivity of the Indian monsoon to the Eurasian snow cover. Walsh and Ross [1988] tested the sensitivity of 30-day forecasts to continental snow cover and found large sensitivity over Eurasia. Snow cover is related to many feedbacks [Randall et al., 1994], the most obvious being the surface albedo feedback [Hall, 2004].

[3] Snow also acts as a water storage reservoir, which is released during snow melt in spring. Thus snow cover largely influences runoff, soil moisture, evaporation and, therefore, precipitation and the entire hydrological cycle as reported in many studies [e.g., Douville et al., 2002; Groisman et al., 2004].

[4] In climate models, snow cover fraction (SCF) is diagnostically derived from the snow water equivalent (SWE), which is a prognostic variable in most models. A correct simulation of the SCF is crucial for the computation of surface albedo during the winter season, and the literature presents several parameterizations for use in GCMs [Dickinson et al., 1993; Marshall et al., 1994; Sellers et al., 1996; Yang et al., 1997; Roesch et al., 2001].

[5] Surface albedo is closely related to snow cover since a large part of its interannual variability is caused by changes in the snow cover area (SCA). It is thus reasonable to discuss surface albedo and snow cover together. However, surface albedo is also subject to large variability in snow-free areas since the albedo also depends on the soil wetness of the uppermost soil layer [Culf et al., 1995], the solar angle [Verseghy et al., 1993], the solar spectrum [Briegleb and Ramanathan, 1982], and modifications in the type of the vegetation cover (e.g., through deforestation, overgrazing). As for snow, the effect of varying surface albedo on the surface climate may be further enhanced through positive feedbacks such as the vegetation feedback [Claussen, 1997]. Numerous studies have shown that climate models exhibit a significant sensitivity to surface albedo changes [Charney et al., 1977; Potter et al., 1981; Henderson-Sellers and Wilson, 1983; Roesch et al., 2002].

[6] This paper deals with the validation of the surface albedo and snow cover in a set of IPCC AR4 climate models and is organized as follows. The participating models are described in section 2. The observational data for both snow cover and surface albedo are detailed in section 3. Section 4 briefly introduces into the methodology. The results and discussions are presented in section 5 and section 6 for snow cover and surface albedo, respectively. Section 5 is divided into a separate validation of the SWE and SCA. Conclusions are drawn in section 7.

2. Models

[7] The climate simulations of the 20th century that have been used for this study were performed in the framework of the Fourth Assessment Review (AR4) of the Intergovernmental Panel on Climate Change (IPCC). The participating state-of-the-art coupled climate models are listed in Table 1 along with their originating groups and the availability of the parameters that are relevant within this study. Detailed documentations on all AR4 models are available at the IPCC model documentation web page ( Two climate models performed the 20th century runs at two different grid resolution (CGCM3.1 and MIROC3.2). In that case, the present study involves the results from the high-resolution runs only.

Table 1. AR4 Models With Their Originating Groups and the Availability of Relevant Parameters
CCSM3National Center for Atmospheric ResearchUSAxxxx
CGCM3.1 (T42)Canadian Centre for Climate ModellingCanadax xx
CGCM3.1 (T63)Canadian Centre for Climate ModellingCanadax xx
CNRM-CM3Centre National de Recherches MétéorologiquesFrancex xx
CSIRO-Mk3.0CSIRO Atmospheric ResearchAustraliaxxxx
ECHAM5/MPI-OMMax Planck Institute for MeteorologyGermanyx x 
FGOALS-g1.0LASG/Institute of Atmospheric PhysicsChinaxxxx
GFDL-CM2.0Geophysical Fluid Dynamics LaboratoryUSAx x 
GISS-AOMNational Center for Atmospheric ResearchUSAxxx 
GISS-EHNational Center for Atmospheric ResearchUSAxxxx
GISS-ERNational Center for Atmospheric ResearchUSAxxxx
INM-CM3.0Institute for Numerical MathematicsRussiaxxxx
IPSL-CM4Institut Pierre Simon LaplaceFrancexx  
MIROC3.2 (hires)Center for Climate ResearchJapanxxxx
MIROC3.2 (medres)Center for Climate ResearchJapanxxxx
MRI-CGCM2.3.2Meteorological Research InstituteJapanxxx 
PCMNational Center for Atmospheric ResearchUSAx   
UKMO-HadCM3Hadley Centre for Climate Prediction and ResearchUKx x 

[8] The ECHAM5/T106L31 data presented in this study are obtained from a time slice integration of the atmosphere-only ECHAM5 at T106L31 resolution (31 vertical levels at spectral horizontal resolution T106) over the period 1961 to 1990, using observed sea surface temperature (SST) and sea ice cover (SIC) provided by the EU project PRUDENCE.

3. Observational Data

3.1. Snow Cover

3.1.1. NOAA Visible Data

[9] Since 1966, the National Oceanic and Atmospheric Administration (NOAA) has prepared weekly snow charts for the Northern Hemisphere. NOAA charts are based on a visual interpretation of photographic copies provided by the Advanced Very High Resolution Radiometer (AVHRR) sensor in the visible band (0.58–0.68 μm). The data are given on a regular 1° × 1° grid [Robinson, 1993]. In general, the NOAA charts are considered to be the most accurate means of obtaining snow cover information on large regional to hemispheric scales. Furthermore, they compose the longest satellite-based record available and have been intensively used in former studies [Gutzler and Rosen, 1992; Iwasaki, 1991; Kukla and Robinson, 1981; Masuda et al., 1993; Robinson et al., 1993].

[10] Data that are used in this study span the time period from 1979 to 2001. Data prior to 1979 have been omitted because of inhomogeneities in the time series caused by different satellite generations [Roesch and Roeckner, 2006].

3.1.2. SSM/I Microwave Data

[11] Microwave-frequency data acquired from the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I) since July 1987 have been used to estimate snow cover. The algorithm to estimate snow cover was developed by Chang et al. [1987]. This algorithm uses the difference between the 37-GHz and 19-GHz channels to derive a snow depth/brightness temperature relationship for a uniform snow field. Monthly data are provided for the period 1987–1998 at a regular 1° grid.

3.1.3. MODIS Monthly Snow Cover Fraction Product

[12] In this analysis, a preliminary version of the gridded global monthly snow cover product MOD10CM at 0.05° resolution from March 2000 to present is used. The most challenging task compiling monthly files is the correct handling of missing values in the daily product MOD10 Level 2 from which the monthly values are derived.

3.1.4. USAF/ETAC Snow Depth Climatology

[13] For the validation of simulated snow depth (SD), the global SD climatology of the U.S. Air Force Environmental Technical Application Center (USAF/ETAC) as documented by Foster and Davy [1988] is used. This data set provides a midmonthly mean SD climatology with the highest spatial resolution currently available (1 × 1° equal-angle grid), using a comprehensive set of station data for the months of September through to June. The USAF data are generally considered to be one of the most reliable and accurate snow depth climatologies available [Douville et al., 1995] and is used in several studies for the validation of snow models [Douville et al., 1995; Marshall et al., 1994; Foster et al., 1996].

3.2. Albedo

[14] Satellite observations constitute the only available means for global monitoring of the surface. They allow the establishment of a global climatology of surface albedo at homogeneous resolution [Li and Garand, 1994]. Validation of the AR4 models is achieved using three remote-sensed surface albedo climatologies as well as the ERA40 reanalysis.

3.2.1. MODIS

[15] The MODIS BRDF/Albedo product is generated with data acquired by MODIS on the Terra satellite platform. The MODIS BRDF/Albedo Product MOD43B provides both white-sky albedo (WSA, bihemispherical reflectance) and black-sky albedo (BSA, directional hemispherical reflectance at local solar noon) for 7 spectral bands and 3 broadbands (0.3–0.7 μm, 0.7–5.0 μm,0.3–5.0 μm) at 0.05° resolution [Schaaf et al., 2002]. Data used within this study are based on version V004 data, which are available from November 2000 to February 2004. For model comparisons, the 0.05° product for the spectral band between 0.3 μm and 5.0 μm was aggregated to the T106 grid using area weighting.

3.2.2. PINKER

[16] The PINKER surface albedo climatology has been compiled from the version 2.1 of the surface albedo algorithm developed at the University of Maryland [Pinker, 1985; Pinker and Laszlo, 1992]. Inputs are based on the International Satellite Cloud Climatology Project (ISCCP) data D1 for July 1983 to December 1998 at 2.5° resolution, as provided by the Goddard Institute for Space Studies (GISS).

3.2.3. ISCCP-FD

[17] The International Satellite Cloud Climatology Project (ISCCP) has produced a new 18-year (1983–2000) global radiative flux data product called ISCCP-FD [Zhang et al., 2004]. Among various other parameters, global gridded surface albedos are available as monthly mean values at a resolution of 2.5° × 2.5°.

3.2.4. ERA40

[18] ERA40 is computed with the reanalysis data assimilation system at the European Centre for Medium-Range Weather Forecast (ECMWF). ERA40 produces a physically consistent space-time interpolation of the available historical observational data from 1958 to 2001. Further information are available at

3.3. Surface Temperature

[19] CRU: The Climate Research Unit (CRU) provides high-quality gridded data sets for various climate parameters. For this study, the mean surface temperature climatology for the period 1960–1999 was computed from the CRU TS 2.1 data covering the global land surface at 0.5° × 0.5° resolution from 1901 to 2002 [Mitchell and Jones, 2005].

3.4. Precipitation

[20] GPCC: The Global Precipitation Climatology Centre (GPCC) provides monthly global land precipitation for the 50-year period from 1951 to 2000 [Rudolf and Schneider, 2005]. It is based on quality tested observations of 9343 stations with long homogeneous records. The data are available at 0.5°, 1.0° and 2.5°. For this study, the mean climatology for the period 1960–1999 was derived from the 0.5° resolution.

4. Methodology

4.1. Surface Albedo

[21] Surface albedos are computed as the ratio between the reflected shortwave (SW) radiation and the global radiation. Regional surface albedos are calculated as the ratio between the area-weighted reflected SW and the area-weighted downward SW. As MODIS does not provide surface radiation fluxes, regional surface albedos were computed through the area-weighted mean of WSA albedos.

4.2. SWE/SCA

[22] AR4 models provide either SWE or SD as a prognostic output variable. As the majority of the models provide SWE rather than SD, SWE has been selected for the intercomparison. 10 out of the 15 AR4 models also provide SCF (Table 1). Note that no attempt has been made to derive missing monthly SCF from monthly SWE according to the model's specific parameterization because of the nonlinear relationship between SCA and SWE [Roesch and Roeckner, 2006, section 4.3.3]. SCA is computed by summing the product of each grid cell's area and the respective SCF over all cells within the region of interest.

[23] In order to compare the simulated SWE, simulated SD estimates from USAF/ETAC were transformed to SWE using the density of snow ρs following [Verseghy, 1991]:

equation image

where b1 = 188.82 kgm−3, b2 = 419.0 kgm−2. This relationship accounts for mechanical compaction but ignores temperature induced metamorphism which might lead to significant density changes at the end of the snow season. As a result, equation (1) probably underestimates snow density in spring since the snow density increases as the snow pack melts. SWE derived from SD is thus presumably overestimated during snow melt periods. Despite this limitation, the above relationship has been reasonably well confirmed by using observed values of SD and SWE at six Russian sites from 1978 to 1983 [Robock et al., 1995; Roesch, 2000].

[24] The GISS-AOM (see Table 1) applies a special procedure to compute SWE. Snow is only accumulated up to a value of SWEmax = 91.66 kgm−2. If more snow accumulates, 10% of the total snow mass is compacted into ice (G. L. Russell, GISS/NASA, personal communication, 2005). To bring this model in line with the other participating AR4 models, the current month's surface ice mass minus the August surface ice mass was added to the snow mass.

[25] To facilitate the intercomparison, all data given at grids other than T106 have been mapped to T106 resolution using area weighting. This paper focuses on the investigation of the present day climate. Thus this study is based on the simulations of the 20th century climate, with focus on the last 40 years (1960–1999). Deviations from this standard period will be clearly stated.

5. Results and Discussion: Snow Cover

[26] Snow cover can be characterized by either SWE or SCA. Accurate simulation of SWE is crucial for the hydrological cycle and runoff processes, while SCA strongly affects the surface energy balance through the high reflectivity of snow. It is thus important to investigate the models' performance with respect to both SWE and SCA. In the following discussion, snow mass is used synonymously with SWE.

5.1. Snow Water Equivalent

5.1.1. Annual Cycle

[27] SWE is a fundamental measure for the quality of climate models since the correct simulation of SWE requires the accurate simulation of both surface temperature and precipitation. In the following, the seasonal SWE cycle, as simulated in the IPCC AR4 models, are compared with ground-based data compiled by USAF/ETAC (see section 3.1).

[28] Most state-of-the-art climate models significantly overestimate the snow mass on the Northern Hemisphere (Figures 1a–1d), particularly in spring. On a hemispheric scale, CSIRO-Mk3.0, ECHAM5/MPI-OM and INM-CM3.0 are closest to the observations. In Eurasia, the models ECHAM5/MPI-OM and CSIRO-Mk3.0 most closely match the observation, while in North America, UKMO-HadCM3 and ECHAM5/MPI-OM are the most accurate models (Figure 1c). Restricting the analysis to the zone between 40–60°N, ECHAM5/MPI-OM shows the best agreement to observations out of all participating climate models. Mean Eurasian SWE predicted by the GISS-AOM is clearly higher than in most other climate models, which might be partly related to the snow/ice transformation algorithm (section 4) and problems in how snow is melted (G. L. Russell, personal communication, 2005). Mean Eurasian SWE simulated by GISS-AOM for the period from February to April, is approximately twice that derived from ground measurements. Furthermore, AR4 models produce a wide range of predicted peak snow accumulation. Figure 1 shows that the models tend to significantly overestimate the peak snow accumulation, mainly in Eurasia. The observed Eurasian SWE peaks in February (Figure 1b) and reaches its minimum in August, whereas maximum SWE in North America is reached in March (Figure 1c). Most models predict a delayed peak snow accumulation. Only CSIRO-Mk3.0 is in line with the observed SWE maxima in February, while the other models do not peak until March. Excessive snow amount and a delayed snow melt in spring was also found by Foster et al. [1996]. Therefore it is obvious that various state-of-the-art climate models still suffer from a delayed retreat of the snowline and excessive snow amount in spring. It should be mentioned, however, that there are several deficiencies in the USAF/ETAC data as well. It is, e.g., challenging to construct isonivals in areas with relatively few data points such as the boreal forests. In addition, sparse measurements in mountainous areas might bias the SD to too low values.

Figure 1.

Monthly mean SWE (1960–1999) for 14 AR4 models and the USAF/ETAC climatology. (a) Northern Hemisphere (land) without Greenland; (b) Eurasia, north of 20°N; (c) North America, north of 20°N; and (d) 40–60°N.

[29] In order to investigate the main reasons for the overestimated SWE, simulated surface temperature and snowfall are compared with the most accurate currently available data sets for global land surface temperature (CRU) and precipitation (GPCC) (sections 3.3 and 3.4).

[30] Figure 2 shows a scatterplot of the simulated SWE deviations and the models' temperature bias for both March and April. Figure 2 reveals that there is no distinct relationship between a warm or cold bias and SWE in March. In April, the AR4 models generally show a slight cold bias over both Eurasia and the 40–60°N land region. This might be related to a positive snow/albedo feedback that could partly contribute to the delayed snow melt in spring. However, the coupled climate models generally produce a slight warm bias over North America in April (not shown) along with a positive SWE bias. So temperature biases are not the primary driver for the overly thick snow deck as simulated in (late) winter and spring.

Figure 2.

Scatterplot of SWE biases and surface temperature biases (model minus observation). Biases are displayed for (a and b) Eurasia (north of 20°N) and (c and d) the land region between 40 and 60°N for both March and April. Color scale is as in Figure 1: black diamonds, CCSM3; black triangles, CGCM3.1 (T63); and black squares, CSIRO-Mk3.0. Observation are taken from CRU (surface temperature) and USAF/ETAC (section 3). All 10 AR4 models providing both SCA and SWE are shown.

[31] Accumulated snowfall might also contribute to the poor SWE simulation in many coupled atmosphere-ocean climate models. In order to test this hypothesis, accumulated snowfall was computed from the GPCC precipitation, assuming solid precipitation for CRU temperatures below 0°C. The comparison between the SWE biases and biases in accumulated snowfall reveals that most AR4 models produce too much snowfall during the winter and spring season (Figure 3). High positive SWE biases (such as simulated in GISS-AOM) are generally related to high positive snowfall anomalies. Assuming some snow melt and sublimation, the positive snowfall anomaly could easily explain most of the positive SWE bias. The results are barely sensitive to the threshold temperature for the transition between rain and snow (e.g., 1°C instead of 0°C).

Figure 3.

As in Figure 2 but for SWE and accumulated snowfall. Accumulated snowfall for March and April corresponds to the total snowfall from October to March and October to April, respectively. Observed snowfall has been computed by assuming solid precipitation for CRU temperatures below 0°C. Color scale is as in Figure 2: black diamonds, CCSM3; black triangles, CGCM3.1 (T63); and black squares, CSIRO-Mk3.0.

[32] It is thus likely that the positive SWE biases are primarily caused by an overly high snowfall rate. This finding also suggests that, for a correct SWE simulation, an accurate simulation of the large-scale circulation and the precipitation rate is probably more fundamental than the improvement of the implemented snow model.

[33] All AR4 models produce the SWE in late autumn/early winter reasonably well. Thus the onset of the snow season is better captured than the snow melt by current climate models. The difficulty in predicting the snow melt might be caused not only by biases in temperature and precipitation, but also in a poor parameterization of the snow melt processes. For example, ECHAM5, the latest ECHAM version, is superior to the previous version 4 in simulating the timing of the spring snow melt by profiting from an improved representation of the snow melt processes [Roesch and Roeckner, 2006].

[34] Excessive SWE is found in all AR4 models in the Himalayas (not shown). However, uncertainties in snow measurements in high mountainous areas may be afflicted with significant errors and tend to underestimate the area averaged snow height since measurements are sparse and biased to lower situated regions (valleys). In some models, a positive snowfall bias at the southern slopes of the Himalayas probably further contributes further to the excessive Himalayan snow amount [Hagemann et al., 2006].

[35] Ten out of the 15 investigated climate models predict SD (Table 1). This allows a direct comparison with the USAF/ETAC SD climatology without applying equation (1). The comparison clearly shows that, at least on larger (temporal and spatial) scales, the percentage SD biases are only slightly lower than the respective SWE biases (Table 2). This provides confidence in the SWE validation of the AR4 models and the snow density as reported by Verseghy [1991].

Table 2. SWE and SD Biases (Percentage Differences): Nine-Model Mean Minus USAF/ETACa
 Percentage Difference
  • a

    The model mean is computed from the nine AR4 models providing both SWE and SD (CGCM3.1, CNRM-CM3, FGOALS-g1.0, GISS-AOM, GISS-EH, GISS-ER, INM-CM3.0, MIROC3.2, and MRI-CGCM2.3.2). CCSM3 and CSIRO-Mk3.0 have been omitted because of questionable SD data. Percentage deviations are averaged over the period January to March.

Northern Hemisphere
Eurasia (>20°N)
North America (>20°N)
Land, 40–60°N

[36] In summary, most of the AR4 models suffer from an excessive snow amount in spring and a delayed peak snow accumulation and snow melt. The spread among the models is largest in spring while the onset of the snow season is reasonably well captured by all participating models.

5.1.2. Frequency Distribution

[37] The frequency distribution of observed and predicted SWE is shown in Figure 4 for spring (Figures 4a and 4c) and autumn (Figures 4b and 4d). Grid cells with spring (MAM) SWE exceeding 10 cm are too frequent in most AR4 models (Figures 4a and 4c). According to the USAF snow depth climatology 12.0%, 15.3%, 5.9% of all grid elements are covered with SWE above 10 cm in spring for Eurasia, North America and the zone 40–60°N, respectively. The corresponding numbers for the mean of the 14 AR4 models are distinctly higher and amount to 23.0%, 21.1%, and 14.9% for the respective regions. On the other hand, most models underestimate the number of grid boxes in Eurasia and North America that are covered with a SWE between 2 cm and 10 cm.

Figure 4.

Frequency distribution of SWE for 14 AR4 models (average from 1960 to 1999) and the USAF/ETAC climatology. Frequencies are given for 2 cm bins (classes). The class for snow-free pixels and thin snow decks (0–2 cm) has been omitted. Eurasia (a and b) north of 20°N and (c and d) 60–70°N excluding Greenland in MAM (Figures 4b and 4d) and October/November (Figures 4a and 4c).

[38] For both Eurasia and North America, the maximum SWE frequencies for the analyzed AR4 models are generally between 5 cm and 15 cm in spring, with a wide spread among the models, whereas the class frequencies for USAF/ETAC generally decrease with increasing SWE (Figure 4a). The CSIRO-Mk3.0 shows superior SWE frequency distribution compared to all other AR4 models. This is clear from either direct visual inspection of Figure 4 or by comparing the skewness S of the various frequency distribution curves. For Eurasia, e.g., S is only positive for USAF/ETAC (S = 0.12) and CSIRO-Mk3.0 (S = 0.27) while the corresponding skewness is negative for all other coupled climate models. For the 60–70°N land zone, simulated SWE peak frequencies in spring are generally between 10 cm and 15 cm, with the highest value predicted by GISS-AOM with approximately 19 cm (Figure 4c). Note that frequencies for the first bin, ranging from 0 cm to 2 cm have been omitted since this gives, as a first approximation, the percentage of snow free pixels in the specified domain. This class contains more than half of all grid boxes for all presented domains excluding the zone 60–70°N.

[39] In late autumn (Figures 4b and 4d), the frequency distribution of SWE is captured reasonably well in all AR4 models. The number of grid boxes with moderate snow covers (SWE between 2 cm and 10 cm) is generally overestimated in North America by the AR4 models (not shown). This feature agrees with a slight overestimation in the mean snow mass over North America in late spring as produced by most AR4 models (Figure 1c).

5.2. Snow Cover Area

5.2.1. Annual Cycle

[40] The snow cover extent is vital in determining the surface albedo and thus the amount of energy available to turbulent and radiant energy exchange. In order to detect the predominant discrepancies between simulated and observed snow cover extent, simulated annual SCA cycles at hemispheric, continental and regional scales are compared with both visible and microwave remote-sensed observations. Because of the highly nonlinear relationship between SWE and SCA, it is of great importance to validate SWE and SCA separately as positive biases in SWE over a specific domain do not necessarily lead to overestimated SCAs.

[41] The seasonal cycle of observed and simulated SCA over four different domains is displayed in Figure 5. The selected areas span a total of 75.2 · 106 km2 (Northern Hemisphere), 47.2 · 106 km2 (Eurasia), 19.1 · 106 km2 (North America) and 6.5 · 106 km2 (Europe). According to the NOAA climatology, approximately 59%, 60%, 70% and 54% of the total area is snow-covered during the midwinter months (DJF) in the Northern Hemisphere, Eurasia, North America and Europe, respectively. Estimates from the other two remote-sensed climatologies are 5–10% lower. This can be attributed to both different algorithms and different lengths in the underlying time periods. Figures 5 and 6show that eight out of 10 AR4 models providing SCA are quite well in line with the observation. Only two out of them, FGOALS-g1.0 and CSIRO-Mk3.0 are generally above the satellite climatologies while CCSM3 is, during DJF, clearly below the observation. For FGOALS-g1.0, the main reason leading to the overly high snow cover extent, the model predicts 76% (89%) of the Northern Hemispheric land (Europe) to be snow covered in January, can be attributed to restricting the SCF to either the value 0 or 1. Remnants of snow are thus sufficient to produce complete snow-covered grid boxes which shifts the snowline too far south. The reason for the positive SCA deviations in CSIRO-Mk3.0 are not as obvious. SCF is computed from the prognostic SWE as follows

equation image

where SWE is in cm and SWEcr = 2.0 cm. This results in a SCF of only 33% and 50% for a SWE equal to 1 cm and 2 cm, respectively, which is at the lower end of currently used SCF parameterizations. A closer investigation reveals that the snowline probably protrudes too far south during the relevant seasons.

Figure 5.

Seasonal snow cover extent, averaged over 1960–1999, for 10 IPCC AR4 models providing SCA. (a) Northern Hemisphere without Greenland, (b) Eurasia north of 20°N, (c) North America north of 20°N, and (d) Europe (bounded by the 35°E meridian).

Figure 6.

Scatterplot of AR4 model biases (model minus observation, Eurasia >20°N, 40-year mean: 1960–1999) for (a and b) SCA versus SWE and (c and d) SCA versus albedo for March (Figures 6a and 6c) and April (Figures 6b and 6d). Observed SCA are based on NOAA visible data, and observed surface albedo is the arithmetic mean from PINKER, ISCCP-FD and ERA40 (excluding MODIS because of its short data record). Color scale as in Figure 2: black diamonds, CCSM3; black triangles, CGCM3.1 (T63); and black squares, CSIRO-Mk3.0.

[42] The lowest SCA during late spring is generally produced by the Russian climate model INM-CM3.0. This is somewhat surprising as this model incorporates the parameterization in equation (2), but with SWEcr = 0.4 cm. This means that for SWE = 0.4 cm (2 cm), 50% (83%) of the grid box will be snow covered. This is clearly higher than in the other analyzed climate models and those derived from ground and remote-sensed observations [Yang et al., 1997; Roesch et al., 2001]. In North America and Europe, the low SCA coincides with rather low SWE for the Russian model (Figures 1b and 1c). A closer analysis shows that the snowline in INM-CM3.0 is probably situated too far north, mainly in Europe and along the US west coast. The negative bias might be further exacerbated by the coarse grid resolution (5° × 5°), resulting in an underestimation of the mountainous snow cover extent due to a poor orographical representation of mountain ridges. The negative SCA bias in CCSM3 for DJF can be probably related to the SCA parameterization which computes rather low SCAs even for thick snow packs (e.g., SCA = 80% for SD = 40 cm).

[43] It is of great interest to correlate the models' SWE biases with SCA deviations (Figures 6a and 6b). This scatterplot for 10 AR4 models clearly reveals that positive SWE biases do often not correspond to positive SCA biases. The main reason is the highly nonlinear relationship between SCA and SWE. SCA is barely sensitive to SWE changes for thick snowpacks while thin snowpacks may experience extreme changes in SCA for moderate changes in SWE. Further, the applied parameterization that computes the diagnostic SCF from the prognostic SWE (or SD) differs strongly between the models.

[44] Summarizing, most of the AR4 models capture the seasonal cycle of the snow cover extent with sufficient accuracy. The two climate models FGOALS-g1.0 and CSIRO-Mk3.0, however, clearly overestimate the SCA in both North America and Eurasia. CCSM3 generally simulates the lowest snow-covered areas in DJF, which may be related to its diagnostic SCF parameterization. Positive biases in the SCA need not necessarily be related to positive deviations in the SWE because of the highly nonlinear relationship between the prognostic SWE and diagnostic SCF. Finally, it should be stressed that remote-sensed climatologies are also afflicted with errors because of reasons such as incomplete knowledge of the atmospheric state, cloud screening problems or snow-masking effects of forests. However, reasonable agreement between the various satellite products suggest that remote-sensing is a suitable tool for portraying snow cover extents and for the validation of climate models.

5.2.2. Interannual Variability

[45] Besides the annual cycles and mean values it is crucial for an overall good performance of a GCM to correctly capture the interannual variability. High year-to-year variability is usually found in regions with thin snow decks where frequent melt is observed. In contrast, SCF variations in areas with thick snowpacks are generally low and the SCF remains close or equal to one.

[46] Figure 7 shows the normalized interannual variability of monthly SCAs for ten IPCC AR4 models and the remote-sensed NOAA and SSM/I time series. The regions shown are identical to those in Figures 5a and 5d. Normalized year-to-year variations are low in winter while the normalized variability increases toward spring and peaks in summer/early autumn when only remnants of snow are found in Siberia and the Canadian Archipelago. This is reasonable as the winter snowpack is quite stable over extended areas in Northern Eurasia and North America, with temperatures clearly below freezing point. In contrast, regions close to the snowline are exposed to substantial interannual variations. Figure 7 reveals that the interannual variability is generally captured fairly well on both the hemispheric (Figure 7a) and continental scale in winter and spring. This is partly in contrast to Frei et al. [2003] who found a general underestimation of the interannual variability in the majority of the 15 AMIP-2 models. FGOALS-g1.0 predicts distinctly too low interannual variabilities over all domains and seasons. This can be related to missing intermediate SCF values between 0 and 1, leading to a drastic underestimation of the interannual variability. The Russian model INM-CM3.0, in contrast, simulates overly high variability mainly in spring. This may be due to problems in simulating the large-scale circulation in the spring season or problems in capturing the interannual variability in surface temperature and/or precipitation.

Figure 7.

Interannual variation of SCA for (a) Northern Hemisphere without Greenland and (b) Europe (bounded by the 35°E meridian). Standard deviations are normalized with the corresponding snow cover extent. Remote-sensed data (NOAA and SSM/I) are shown as thick red lines.

[47] Over Europe (Figure 7b), the observed normalized variations are clearly higher than for the entire Eurasian landmass because of frequent accumulation and snow melt periods in major areas of Europe during the winter season. The simulated interannual SCA variations over Europe are mostly underestimated from November to March. Displaying normalized SCA standard deviations averaged over 10° latitude bands clearly reveals that the models' underestimation of the interannual variability is most pronounced during the spring snow melt (not shown). Some of the underestimation may be due to the coarse horizontal resolution in some of the analyzed climate models. In summer and autumn, when SCAs are low, the normalized standard deviations are less reliable and meaningful as moderate changes in total SCA may drastically change the normalized variations.

[48] In summary, the IPCC AR4 models generally underestimate the interannual variability in Eurasia in winter and early spring while they capture year-to-year changes over North America with sufficient accuracy. The underestimation in the interannual SCA variability is most pronounced during the snow melt period.

5.2.3. Trends

[49] Several studies report a significant decrease in hemispheric SCA during the last three decades [Brown, 2000; Dye, 2002]. On the basis of the monthly gridded NOAA snow cover data for the time interval 1979–1999, significant negative trends in the SCF were computed for the Northern Hemispheric spring (Figure 8a). Data before 1979 were ignored because of an inhomogeneity caused by launching the Advanced Very High Resolution Radiometer (AVHRR) in November 1978. Figure 8 reveals that SCA depletion is most pronounced in regions where the snow deck is relatively shallow and the potential snow albedo feedback is large. The most pronounced negative trend has been observed in August (70–80°N) and May (60–70°N and 50–60°N). These months can be characterized by a rapid snow melt and moderate snow packs in the respective zones. The maximum negative trends in these latitudinal bands are −0.267 × 106 km2/10y (−7.1%), −0.52 × 106 km2/10y (−4.3%) and −0.56 × 106 km2/10y (−3.9%), respectively, with the numbers in brackets giving the percentage of the total land area in the individual zones. Over Eurasia and North America (not shown), the MAM SCA decreased by −7.8 · 105 km2/10y and −1.9 · 105 km2/10y, respectively. These values agree well with results given by Brown [2000] who investigated trends in the SCA from 1915–1997. A statistically significant reduction in the SCA has occurred in large parts of Europe in February/March, often exceeding 10% during the last two decades of the 20th century.

Figure 8.

Decadal trends of snow cover extent for 10° latitude bands: (a) 70–80°N, (b) 60–70°N, (c) 50–60°N, and (d) 40–50°N. IPCC AR4 models are plotted as thin lines, and trends derived from NOAA and ECHAM5/T106L31 (time slice simulation) are plotted as thick red lines. Unit is 106 km2/10 years. Total land areas are 3.75, 12.00, 14.34, and 16.00 million square kilometers for 70–80°N, 60–70°N, 50–60°N, and 40–50°N, respectively.

[50] Comparing the trends in the NOAA data with the AR4 models reveals that negative trends are also generally found in the model simulations (Figure 8). However, the participating models, except for INM-CM3.0 and CCSM3, tend to underestimate the decrease in the snow cover extent during the last two decades of the 20th century. Some of the models, such as CGCM3.1, CSIRO-Mk3.0FGOALS-g1.0, and GISS-EH even produce positive SCA trends. Note that the ECHAM5 time slice simulation that is driven by prescribed SST obviously captures the observed SCA trends quite well.

[51] According to the 3rd IPCC report, observations show that the multidecadal correlation between increases in the Northern Hemisphere spring land temperature and a reduction in the Northern Hemisphere spring snow cover are highly significant since data have been available (1966). It is important to verify how well the AR4 coupled climate models can reproduce these observed changes. Figure 9 displays a scatterplot of simulated decadal SCA trends and decadal temperature trends for four regions in spring (MAM). North America shows the expected pattern: positive temperature trends are related to decreasing SCA (Figure 9b). However, some AR4 models neither capture the observed decreasing snow extent nor the increasing surface temperature. On a hemispheric scale (Figure 9d), most AR4 models simulate a positive decadal temperature trend of approximately 0.1°C–0.2°C which is in good agreement with the observation. However, more than one third of the models fail in capturing the right sign of spring snow cover changes. Similar results are found for the Eurasian land mass and the latitude band between 40 and 60°N (Figures 9a and 9c). From this it can be concluded, that the coupled AR4 models reproduce the observed temperature increase in the last two decades of the last century with sufficient accuracy. However, apart from North America, they often fail in capturing the observed relation between decreasing snow cover extent and increasing temperatures.

Figure 9.

Scatterplot showing decadal trends of AR4 models, spring (MAM), 1979–1999: temperature trends versus SCA trends. (a) Eurasia, >20°N; (b) North America, >20°N; (c) 40–60°N latitude land band; and (d) Northern Hemispheric land, without Greenland. Color scale is as in Figure 2: black diamonds, CCSM3; black triangles, CGCM3.1 (T63); and black squares, CSIRO-Mk3.0.

6. Results and Discussions: Surface Albedo

6.1. Global Annual Mean

[52] Annual global mean surface albedo values in AR4 models for the period 1960–1999 and the remote-sensed climatologies PINKER and ISCCP-FD are presented in Figure 10. The mean annual surface albedo of the 15 AR4 models amounts to 0.140 with a standard deviation of 0.013. All AR4 models are slightly above the mean of PINKER (0.124) and ISCCP-FD (0.121). However, on a global scale, differences among the models, as well as the biases between the models and the remote-sensed climatologies, are small. Three (MRI-CGCM2.3.2, INM-CM3.0, and CSIRO-Mk3.0) out of the 15 AR4 models are more than 1 standard deviation above the all-model mean; two models (GISS-EH and PCM) are more than 1 standard deviation below the all-model average. The mean surface albedos predicted by both the AR4 models and the satellite-derived composites are distinctly lower than what has been reported in early studies: Robock [1980] and Hummel and Reck [1979] estimate the global mean albedo to be 0.170 and 0.154, respectively. The main reason for this discrepancy is, that these authors did not weight the albedo values by incoming radiation. Assuming the albedo to be a surface property independent of actual irradiation leads to an overweighting of polar regions with frequent snow and ice cover and the related high surface albedo values.

Figure 10.

Global mean surface albedo in a total of 15 GCMs from IPCC AR4 and 2 remote-sensed surface albedo climatologies (ISCCP-FD and PINKER).

[53] Some further statistics for mean annual surface albedos on a regional scale are given in Table 3. The regional albedo averages predicted by the AR4 models are consistently higher than the observation over the selected regions (global land, global sea, Northern (NH) and Southern Hemisphere (SH)). Averages over NH are a little higher than for SH because of the unequal land/sea distribution. Variations among the AR4 models are significantly higher over the SH than over the NH. This may be related to different parameterizations in the zenith angle dependence of the water albedo and different sea-ice extensions in Antarctica. In addition, most of the participating GCMs do not distinguish between the diffuse and direct beam albedo which may lead to substantial biases in the total broadband surface albedo. In the polar regions north and south of 60°N and 60°S, respectively, the mean annual surface albedo increases to approximately 0.5 in both the models and the observations. The differences between the AR4 models in seasonally snow-covered regions are more pronounced than in snow-free regions. This is due to (1) differences in the simulated SWE, (2) differences in the snow albedo parameterization, and (3) different approaches in the parameterization of the SCF.

Table 3. Statistics for Mean Surface Albedos for Some Selected Areasa
RegionMean (AR4 Models)STDDEV (AR4 Models)Mean (ISCCP/PINKER)
  • a

    Note that surface albedos are unimportant during polar night. First column, region; second column, average over all 15 IPCC AR4 models; third column, standard deviation from all models, and fourth column, average of ISCCP and PINKER data.

Global land0.2600.0120.230
Global sea0.0930.0150.084
Northern Hemisphere0.1460.0110.128
Southern Hemisphere0.1230.0160.105
60–90°N, 60–90°S0.5380.0320.494
20°S–20°N, land0.1690.0150.162

6.2. Annual Cycles

[54] Annual albedo cycles of some selected AR4 models, providing both reflected and incoming radiation, are displayed in Figure 11 for six selected domains. The annual cycles of the models and the satellite products evolve in a similar manner for all domains. This again means that the differences among the various models are, at least on large spatial scales, quite constant throughout the year. Therefore the investigated models capture the seasonal cycle (at a hemispheric scale) quite well. This also applies for the tropics where the surface albedo varies little throughout the year in both the AR4 models and the remote-sensed surface albedo climatologies (Figure 11f). The three models with the highest mean annual surface albedos (MRI-CGCM2.3.2, INM-CM3.0 and CSIRO-Mk3.0) are thus generally above the other participating models, the remote-sensed products and the ERA40 estimates during the entire year. On a global scale, surface albedo peaks twice a year: in April/May and November to January (Figure 11a). The first peak can be attributed to the maximum albedo in the NH due to snow and ice cover at higher latitudes which are no longer masked by the polar night (note that the albedo does not play a role during polar nights). The second maxima can be assigned to both the extended snow cover at middle and high northern latitudes and the peak of Southern Hemispheric sea albedo in November (Figure 11e). This can also be related to the bright Antarctica and adjacent sea ice that is no longer masked by polar night, but is favored by high global radiation fluxes. Figure 11b suggests that ISCCP might be too low over land for the Northern summer.

Figure 11.

Mean (thick black solid line) and standard deviation (vertical black lines) for all participating AR4 climate models. The remaining thin lines represent the mean annual cycles of surface albedos for 6 selected AR4 models. The selection of these models is determined by analyzing the highest three and lowest three values of mean annual global surface albedo. The two remote-sensed climatologies and the ERA40 estimates are shown as thick red lines. (a) global; b) global, land; (c) global, sea; (d) Northern Hemisphere; (e) Southern Hemisphere; and (f) 20°S–20°N, land.

[55] Monthly zonal ALB averages for 10° latitude land bands are shown in Figure 12. The scatter among the AR4 models becomes significantly higher than for global or hemispheric scales. Absolute albedo differences among the models and between the models and the observations are highest in snow-covered regions and become modest in (sub)tropical regions with a standard deviation of approximately 0.02 throughout the entire year (Figure 12d). The standard deviation for the latitude band 20°S–20°N is only slightly above 0.01 (not shown).

Figure 12.

(a–d) Seasonal cycle of mean surface albedo for selected AR4 climate models (as in Figure 11) and 4 remote-sensed climatologies (thick red lines). Average of all 15 AR4 models (thick solid black line) and monthly standard deviations (vertical black lines) for four 10° latitude land bands between 20 and 80°N are shown. The band 20–30°N (Figure 12d) is restricted to 20°W–50°E. Missing values are due to polar night in some part of the domain.

[56] The highest absolute albedo differences of up to 0.3 are found between 50 and 70°N in winter and 70–80°N in summer. The high variations in winter/early spring in primarily snow-covered regions are, in addition to errors in SWE, due to uncertainties in (1) the SCF parameterization, (2) the snow albedo and (3) the snow masking of forests. If restricting the domain to the boreal forests in Eurasia and Canada, differences among the models are up to 0.4 in winter and early spring (not shown). For example, the mean albedo as simulated with the model INM-CM3.0 is close to 0.7 over the boreal forest in January and February. This is far above the albedo for snow-covered forests of approximately 0.2 that has been suggested in the literature and from surface based observation [Pomeroy and Dion, 1996; Betts and Ball, 1997; Roesch et al., 2001]. The positive albedo bias of snow-covered forests simulated by the INM-CM3.0 GCM is also reflected in the high albedo values in Figures 12b and 12c. In partly snow-covered regions, surface albedos at the lower end of the AR4 model suite are thus likely to be more reliable. The ISCCP albedos are probably too high over snow-covered and/or forested areas while PINKER agrees well with MODIS [Roesch et al., 2004] (Figures 12a–12c). ERA40 provides further justification for the overestimation inherent in the ISCCP data. ERA40 surface albedo estimates are even below MODIS and PINKER between 50 and 70°N in spring.

[57] During the snow-free season, the AR4 model simulated surface albedos, in general, do not significantly differ. The monthly standard deviation in the various models is generally between 0.01 and 0.02. However, Figure 12d reveals that the models' desert albedo also differ quite substantially: For the region 20–30°N and 20°W–50°E, CCSM3 is approximately 0.1 below the albedo simulated by PCM throughout the entire year. This is clearly above the absolute accuracy requirements of 0.02 that have been proposed for land surface albedo in climate models [Henderson-Sellers and Wilson, 1983; Sellers et al., 1995].

[58] It is strongly expected that, at least on a local scale, positive SCA biases will produce positive albedo biases. However, on a local scale this does not apply for all vegetation types such as forests. For the large scale, the self-evident correlation between positive albedo and SCA biases no longer applies (Figures 6c and 6d). From this it can be concluded that, in addition to reasonably simulated SCAs, it is also crucial to correctly capture the snow (and snow-free) surface albedo as well as the forest fraction and its albedo under snow-covered conditions.

7. Conclusions and Outlook

[59] Surface albedo has long been recognized as one of the key surface parameters in climate models through its direct effect on the energy balance. Nevertheless, the parameterization of surface albedo is still oversimplified in most current climate models. Further, large uncertainties are inherent in the parameterization of the SCF which is fundamental for an accurate prediction of the surface albedo in snow-covered areas.

[60] This study presents a model intercomparison of simulated snow cover (in terms of SWE and SCA) and the surface albedo. Further, the climate models are validated against remote-sensed and ground-based observations. Positive biases in SWE are not necessarily related to overestimated snow extents since the relationship between SWE and SCF is highly nonlinear. It is thus necessary to validate snow mass and snow cover extent separately.

[61] Most AR4 models predict an excessive snow mass in spring. Various models fail in resolving the end of season snowpack. The spread among the participating models is most pronounced in spring, while the onset of the snow accumulation is generally well captured. These model deficiencies manifest into a distinct overestimation of the area that is covered with SWE above 10 cm in spring. The excessive snow mass in winter and spring is likely to be produced by too heavy snowfall during these seasons.

[62] The seasonal cycle of the SCA is captured reasonably well by most participating models. FGOALS-g1.0 and CSIRO-Mk3.0, however, clearly overestimate SCA in both Eurasia and North America, but for different reasons. Simulated interannual SCA variations are sufficiently in line with the observations for Eurasia and North America in winter and spring. During snow melt, most models predict too low interannual SCA variations.

[63] The pronounced negative trends from 1979 to 2000 as found in the observations are only partly reproduced in the AR4 model simulations of the 20th century. Further, computed trends show a large spread among the models. The majority of the models, however, reproduce the observed increase in Northern Hemispheric land temperature and declining snow cover extent. The time slice simulation using ECHAM5 with prescribed SST and sea-ice suggests that accurate SSTs and ice coverage are fundamental for a correct prediction of SCA trends.

[64] The annual mean surface albedo of the AR4 models is 0.140 with a standard deviation of 0.013. All climate models are slightly above the average derived from the PINKER and ISCCP climatology. The participating models all capture the large-scale seasonal cycle of the surface albedo quite well. However, pronounced systematic biases are predicted in some areas. Highest differences between the models are found over snow-covered forested regions. The winter surface albedo of CNRM-CM3, averaged over the latitude zone from 50 to 70°N, is nearly 0.3 lower than in MIROC3.2 and INM-CM3.0. Comparisons with ground-based and remote-sensed data reveal that most AR4 models predict positive biases over primarily forested areas during the snow period. These substantial deviations are still far too high to meet the required accuracy of surface albedos in GCMs.

[65] Compared to snow-covered regions, the models' albedo vary less over snow-free regions. However, the spread among the 15 climate models is not negligible. The mean annual albedo for deserts between 20–30°N and 20°W–50°E ranges from slightly above 0.25 in CCSM3 to approximately 0.35 in PCM. This reveals that the estimation of desert albedo is still afflicted with some uncertainties. The maximum albedo difference among the AR4 models amounts to approximately 0.05 for the tropical region between 20°S and 20°N.


[66] The research reported herein was performed at the Swiss Federal Institute of Technology, Institute for Climate Research, under the sponsorship of NCCR Climate. We acknowledge the international modeling groups for providing their data for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the model data, the JSC/CLIVAR Working Group on Coupled Modelling (WGCM) and their Coupled Model Intercomparison Project (CMIP) and Climate Simulation Panel for organizing the model data analysis activity, and the IPCC WG1 TSU for technical support. The IPCC Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science, U.S. Department of Energy. The (modified) monthly SWE for the GISS-AOM model has been provided by Gary L. Russell from GISS/NASA since these data are not available through the PCMDI web site.