1. Introduction
Accurate simulations of the snow cover strongly impact on the quality of weather and climate prediction as the absorption of solar radiation at the land–atmosphere interface is modified by a factor up to 4 in response to the presence of snow (albedo effect) (Viterbo and Betts 1999). Changes in snow cover modify the surface albedo, which then feedbacks to surface temperature (Groisman et al. 1994). In northern latitudes and mountainous regions, snow also acts as an important energy and water reservoir (Immerzeel et al. 2010). Therefore, a correct representation of snow mass and density is crucial for predicting the snow thermal insulation, especially in climate change conditions, with direct consequences for soil hydrology and ground temperature (Bartlett et al. 2004, 2005; Lynch-Stieglitz 1994; Rangwala et al. 2010; Stieglitz and Smerdon 2007). This thermal process also plays an important role in controlling soil moisture freezing, with impact on the stable boundary layer (Viterbo et al. 1999).
The role of snow cover in the climate system has been a topic of research for a long time, including early field observations reported by Sverdrup (1935). On a modeling framework, Williams (1975) was one of the first to explore the influence of snow cover on atmospheric circulation and climate change. Snow models have been developed for many applications, such as snow process studies, runoff, and avalanche forecasting, and to be used in atmospheric models. Brun et al. (2008) compiled a detailed review of existing models and their applications. The snow schemes designed for atmospheric models (both global and regional) for applications in numerical weather prediction and climate modeling share a similar basic purpose: to provide lower-boundary conditions to the atmosphere over snow-covered regions.
This study evaluates the performance of three snow schemes of different complexity coupled to the EC-EARTH climate model (Hazeleger et al. 2010). EC-EARTH was recently developed starting from the European Centre for Medium-Range Weather Forecasts (ECMWF) seasonal forecast system motivated by the seamless prediction concept of bringing weather forecast and climate change studies into a single framework. The land surface component of EC-EARTH is the Hydrology Tiled ECMWF Scheme of Surface Exchanges over Land (HTESSEL; Balsamo et al. 2009). The snow schemes evaluated include an old snow scheme used at ECMWF until 2009 [control (CTR)], a new snow scheme in operations at ECMWF and the default on EC-EARTH [operational (OPER)], and a multilayer version of OPER (ML3). Dutra et al. (2010, hereafter DU10) describes OPER and its validation against CTR in offline mode. This study is a natural extension of DU10, evaluating OPER in coupled atmosphere simulations. Additionally, a more complex snow scheme (ML3) is also evaluated.
The ML3 snow scheme is validated in offline simulations over several site locations. This is an important step to verify the behavior of the newly developed snow scheme. A three-member 30-yr-long ensemble was carried out with each one of the snow schemes activated in EC-EARTH. The simulations were performed with prescribed sea surface temperature and sea ice cover from the ECMWF reanalysis. This simulation setup guarantees an identical coupling of the different snow schemes with the atmosphere and soil. The evaluation of the performance of the three snow schemes in EC-EARTH is performed by comparing the simulations with observed snow cover, surface albedo, and 2-m temperature, as in other studies that validate climate models (e.g., Roesch 2006). The simulated surface water and energy fluxes are also evaluated for the three snow schemes. Special attention is given to the role of snow thermal insulation in controlling the land–atmosphere coupling (Cook et al. 2008; Ge and Gong 2010). Dutra et al. (2011b) investigated the role of interannual snow-cover variability in controlling the land–atmosphere coupling and its relation with near-surface and soil temperature by performing EC-EARTH simulations with the OPER snow scheme. The authors found that snow cover and depth interannual variability explained almost 60% of winter near-surface interannual variability in predominantly snow-covered regions.
A brief description of the snow schemes is presented in the following section. The validation of the ML3 snow scheme and comparison with CTR and OPER in offline site simulations is presented in section 3. The EC-EARTH simulations in section 4 are validated against remotely sensed data of snow-cover extent and surface albedo. Additionally, the 2-m temperature is compared against the Climate Research Unit (CRU) dataset and the snow–albedo feedback (SAF) with observational estimates from Hall and Qu (2006, hereafter HQ06). Changes in upper air temperature and specific humidity are also discussed. Section 5 summarizes the main conclusions of this study.
2. Snow schemes
The three snow schemes coupled to HTESSEL are compared in Table 1 with respect to the number of snow layers, liquid water representation, snow density, and snow albedo parameterizations. CTR (ECMWF 2011) was the snow scheme used at ECWMF from the late ‘90s until 2009, including the 40-yr ECMWF Re-Analysis (ERA-40) (Uppala et al. 2005) and ERA-Interim (ERAI) (Dee et al. 2011) reanalyses. In 2009, a new snow scheme, OPER, was introduced in the integrated forecast system at ECMWF, and its description, validation, and comparison against CTR in offline mode is presented by DU10. A multilayer version of OPER (ML3) was developed for this study and its main characteristics are presented in the appendix. The three snow schemes are identically coupled with the overlying atmosphere and underlying soil within HTESSEL.
Summary of the main differences between the three snow schemes coupled to HTESSEL.
The snow scheme represents one (CTR and OPER) or up to three (ML3) additional layers on top of the upper soil layer, with independent prognostics thermal and mass contents. The net energy flux at the top of the snowpack is the residual of the skin energy balance from the snow-covered tiles in HTESSEL. The basal heat flux is given by a diffusion formulation between snow temperature (CTR and OPER and lower snow layer in ML3) and soil (uppermost layer) temperature. The absorbed energy is used to change the snow temperature or melt the snow when snow temperature exceeds the melting point. CTR assumes dry snow; in OPER, liquid water is diagnosed from snow temperature, mass, and density; while ML3 represents liquid water as a prognostic for each layer following a bucket-type approach. In both OPER and ML3 the liquid water coexists in the snowpack leading to internal phase changes (freezing–melting), and rainfall can be intercepted. The heat capacity of the snow layer is a function of its depth, density, and presence of liquid water. Snow thermal conduction changes with changing density (different for each layer in ML3). Snow density changes because of overburden and thermal metamorphisms (Anderson 1976), independently for each layer in ML3 and for the entire snowpack in OPER. Both snow-cover fraction and snow albedo are parameterized similarly in OPER and ML3. In CTR, snow density and albedo follow closely the formulation proposed by Douville et al. (1995).
3. Site simulations
Data from the micrometeorological site in Col de Porte (CdP) was used to force, validate, and intercompare the three snow schemes. CdP is operated by the Center of Snow Studies in Grenoble, France, and is located at 45°N, 6°E at an altitude of 1320 m. Data from this site has been widely exploited in model validation (e.g., Boone and Etchevers 2001) and was one of the sites that participated in the first Snow Model Intercomparison Project (SnowMIP; Etchevers et al. 2004). Offline simulations of snow mass (SWE), density, and runoff for the winter seasons of 1996/97 and 1997/98 were validated against CdP observations (Fig. 1). From November to mid-January, in both years, all the simulations properly represent the snow accumulation. From February to May both CTR and OPER tend to overestimate SWE, whereas ML3 is closer to the observations. The final ablation is delayed in all simulations with a stronger overestimation by CTR. The mean absolute error (MAE) of SWE was 98, 82, and 42 kg m−2 during the 1996/97 winter in CTR, OPER, and ML3, respectively, and 128, 84, and 40 kg m−2 during the 1997/98 winter. On average for the two winter seasons, the MAE of SWE reduced 25% from CTR to OPER and 64% from CTR to ML3. Snow density evolution is correctly captured by OPER and ML3. CTR shows a fast increase and saturation at 300 kg m−3, similar to what was found by DU10, and by Bartlett et al. (2006) in other site locations, when using the CTR snow density parameterization. Pronounced density changes due to snowfall events are evident in the observations. In mid-April 1998, the sharp increase of SWE is accompanied by a significant decrease of snow density that is poorly simulated by OPER but correctly captured by ML3.
Simulation results for CTR (gray), OPER (thin black), and ML3 (thick black) for the (a)–(c) 1996/97 and (d)–(f) 1997/98 winter seasons at Col de Porte site: (a),(d) snow mass, (b),(e) snow density, and (c),(f) runoff. Observations are represented by open circles. Runoff was accumulated since 1 Dec of each year and is defined as liquid precipitation and snowmelt that is in excess of the snow-cover holding capacity.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
The accumulated runoff (Figs. 1c,f) is defined as the liquid precipitation and snowmelt that is in excess of the snow-cover holding capacity. Its temporal evolution gives an integrated view of the snowpack water balance. The difficulties of the three snow schemes in simulating short-duration melting events during late winter and spring are evident. These melting episodes are mainly driven by synoptic conditions, and are poorly represented by CTR and OPER. On the other hand, ML3 partially reproduces those episodes with melting in the first layer. These results clearly show one of the main drawbacks of a single snow layer associated with the large thermal inertia of thick snowpacks (Marks et al. 2008). The multilayer scheme has an enhanced accuracy in simulating midseason sporadic melting events, since the top snow layer has a reduced thermal inertia.
Surface albedo and surface temperature simulation are compared against observations for the 1997/98 winter season at CdP in Fig. 2. The overestimation of SWE and delayed melting in CTR is explained by its overestimation of snow albedo and limitations in representing the albedo decay between snowfall events. On the other hand, both OPER and ML3 (which share the same snow albedo parameterization) are able to represent the albedo decay between snowfall events, especially during spring. However, in the beginning of April 1998, there was a sharp decrease of surface albedo that both OPER and ML3 did not simulate. This was a critical point for an accurate final ablation, which would have triggered melting because of the high available energy. The simulations of surface temperature (skin temperature) are in good agreement with the observations, except during late January and early February where all the simulations underestimate surface temperature, particularly ML3.
Simulation results for CTR (gray), OPER (thin black), and ML3 (thick black) for the 1997/98 winter season at Col de Porte site of (a) surface albedo and (b) surface temperature. Observations are represented by open circles in both panels.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
Other site locations were tested (from SnowMIP2; Rutter et al. 2009) and showed similar results with significant improvements from CTR to OPER (as presented by DU10). Figure 3 compares the root-mean-square error of SWE simulations in open and forest plots with different configurations of the model. In open sites there is an improvement from CTR to OPER and to ML3, while in forest sites ML3 is similar to OPER. In open sites, there is also an improvement when increasing the number of snow layers from two (ML2) to three (ML3), but with five layers (ML5) the results are similar to ML3. Changing the minimum snow layer depth (see appendix) from 0.2 to 0.1 m with three or five snow layers (ML3* and ML5*) worsens the results because of the time-implicit coupling with the atmosphere. The reduction of the minimum snow layer depth leads to a reduction of the thermal capacity of the uppermost snow layer and its associated time scale in response to the energy forcing. Further analysis is needed to fully understand this impact and its relation with the algorithmic details of the implicit coupling to the atmosphere. From OPER to ML3 the improvements are mainly restricted to deep snowpack locations, while for shallow snowpacks and highly forested locations ML3 returns very similar results to OPER. These findings suggest that over forested regions other processes, such as radiation interaction with the canopy and snow interception by the canopy, should be also addressed in the model formulation.
Boxplot summaries describing the normalized RMSE of SWE for different model configurations, combined at all SnowMIP2 locations at (a) open sites and (b) forest sites (nine winter simulations each; see Rutter et al. 2009 for a list and description of the sites). The boxes have horizontal lines at the lower quartile, median, and upper quartile and the whiskers extend from min to max. The configurations ML2 and ML5 are similar to ML3, but with max of snow layers set to two and five, respectively. The configurations ML3* and ML5* are similar to ML3 and ML5 but change minimum snow layer depth from 0.2 to 0.1 m (see appendix).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
4. Atmospheric simulations
The atmospheric model of EC-EARTH version 2 (Hazeleger et al. 2010) is based on ECMWF’s Integrated Forecasting System (IFS), corresponding to the current seasonal forecast system of ECMWF (system 3). The model runs at TL159 horizontal spectral resolution (about 125 km on the gridpoint space) with 62 vertical levels that extend up to 5 hPa. Some aspects of newer IFS cycles have been implemented, particularly the new land surface scheme HTESSEL. EC-EARTH also has ocean and sea ice model components, but in this study atmosphere-only runs were performed. The boundary conditions of sea surface temperature and sea ice cover are prescribed as time-varying monthly fields derived from the ECMWF reanalysis ERA-40 up to December 1988 and ERA-Interim onward. This setup was adopted over coupled atmosphere–ocean simulations to avoid the long-term spinup necessary for the ocean, and to confine the impacts of the different snow schemes to the atmospheric evolution and variability.
Both CTR and OPER snow schemes are available in EC-EARTH—the latter being the default—and ML3 was implemented. A three-member 30-yr-long ensemble (January 1979–December 2008) was performed for each of the snow schemes. Each ensemble member differs in its initial conditions: 1 November 1977, 1 December 1977, and 1 January 1978, taken from ERA-40. The first months of the simulation (until December 1978) were discarded from the analysis. The results will be presented as the ensemble mean unless otherwise stated. Throughout the discussion the statistical significance of the differences between the mean of the simulations (monthly or seasonal means) is assessed using a two-sample two-tailed Student’s t test assuming equal variances (Wilks 2006). In the following sections, results are analyzed over the Northern Hemisphere landmasses poleward of 40°N, excluding Greenland (hereafter NH40). This region was selected as representative of the area where seasonal snow cover plays an important role in controlling the surface water and energy fluxes.
a. Snow cover and surface albedo
Simulated snow cover is compared with the weekly snow-cover data from the National Snow and Ice Data Center (NSIDC) (Armstrong and Brodzik 2007). The weekly files consist of gridded binary counts of snow or no snow at 25-km resolution available from 1979 to 2006. This dataset has been used to validate climate model simulations (e.g., Roesch 2006) and to analyze the spatial and temporal patterns of the snow season length during the last three decades of the twentieth century (Choi et al. 2010). The weekly data were averaged into monthly means and aggregated to the model grid.
The simulated Northern Hemisphere snow-covered area (SCA) is compared against NSIDC in Fig. 4. The bias was calculated for each month (synchronous in time) and normalized by the mean NSIDC SCA. CTR underestimates SCA throughout the cold season, with particular problems during spring because of early melting. OPER significantly reduces the biases, especially during spring. ML3 shows similar results during early spring, but in May and June is outperformed by OPER. The mean NSIDC snow cover and the model biases during spring are represented in Fig. 5. The early melting in CTR is marked in central Canada, eastern Europe, Scandinavia, and north of the Himalayan range. Over those regions, both OPER and ML3 significantly reduce the early melting bias of CTR.
Northern Hemisphere SCA biases normalized by the NSIDC snow-covered area. For each month (January 1979–December 2006 three-member ensemble) the simulated SCA biases were evaluated and are represented as box plots. The vertical boxes represent lower and upper quartiles, the circle the median, and vertical lines extend from percentiles 5 to 95. The mean annual cycle of snow-covered area from NSIDC is represented by square symbols. The vertical axis represents area-relative differences for the box plots and SCA (×106 km2) for NSIDC.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
Mean observed snow cover during spring derived from (a) NSIDC weekly snow-cover extent and simulated differences of (b) CTR, (c) OPER, and (d) ML3 with respect to NSIDC. Only differences significant at p < 0.05 are represented. Note the different color scale of (a).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
The Moderate Resolution Imaging Spectroradiometer (MODIS) albedo product MCD43C3 provides data describing both directional hemispheric reflectance (black-sky albedo) and bihemispherical reflectance (white-sky albedo) in seven different bands and aggregated bands. Data from the Terra and Aqua platforms are merged in the generation of the product that is produced every 8 days, with 16 days of acquisition, and available on a 0.05° global grid. The accuracy and quality of the product has been studied by several authors in different locations (e.g., Roman et al. 2009; Salomon et al. 2006; Stroeve et al. 2005). The MODIS product has already served as reference for model validation (e.g., DU10; Wang and Zeng 2010; Zhou et al. 2003). As for the NSIDC snow cover, the MODIS albedo was averaged to monthly fields and aggregated to the simulations grid.
In this study the mean annual cycle of MODIS white-sky broadband shortwave albedo (2001–09) is compared against the model simulations (1979–2008) and is represented in Fig. 6 for the NH40 region. OPER and ML3 have an almost coincident mean annual cycle on the hemispheric scale. This is expected since OPER and ML3 share the same snow albedo parameterization, and their evolution of snow cover is also similar (see Fig. 4). From July to November all simulations overestimate surface albedo, whereas CTR is closest to MODIS from October to December. This overestimation of OPER and ML3 was also identified by DU10 in OPER and is mainly related with the snow albedo parameterization, since such a signal is not visible on snow cover. From March to June, OPER and ML3 are similar to MODIS while CTR shows an underestimation that is associated with the early melting discussed before. The mean spring surface albedo maps from MODIS and simulation differences are represented in Fig. 7. The strongest underestimation of spring surface albedo in CTR is mainly localized over snow-covered regions poleward of 60°N. Both OPER and ML3 reduce those biases. However, over the Himalayan range and Rockies, OPER and ML3 reinforce the overestimation of surface albedo that was already visible in CTR. This overestimation of surface albedo in complex terrain regions is associated with an overestimation of snow cover (see Fig. 5). These results highlight a current deficiency of the model in representing snow cover over complex terrain regions. This is not surprising, since none of the snow schemes include parameterizations for subgrid-scale processes. For such processes, the multilayer representation of the snowpack does not improve the simulations, indicating that further improving the model might need to address the representation of other important processes such as subgrid-scale variability.
Mean annual cycle of surface albedo simulated by CTR (gray), OPER (thin black), ML3 (thick black), and observed from MODIS (open circles) averaged over NH40. The spatial averages of simulations included only the grid points where MODIS data was available.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
Mean observed surface albedo during spring derived from (a) MODIS and simulated differences of (b) CTR, (c) OPER, and (d) ML3 with respect to MODIS. Note the different color scales.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
The early spring snowmelt and underestimation of snow albedo in CTR are closely related, but it is not possible to clearly identify a cause–effect: low snow albedo leading to excessive energy absorbed on the snowpack resulting in an early snow ablation, or the early snow ablation exposing the land surface (with an albedo lower than a snow surface) leading to the underestimation of surface albedo. It is most likely to result from the conjunction of the two factors associated with the SAF.
b. Surface water balance
The mean annual cycle of the surface water fluxes averaged over NH40 are presented in Fig. 8a for the CTR simulation. Figures 8b–e illustrate the differences of OPER and ML3 with respect to CTR for each component of the surface water balance. The sign of the fluxes follow the model convention with positive for incoming water (or energy) to the surface and negative for outgoing water (or energy) from the surface. The runoff and terrestrial water storage variation (TWSV) exhibit a distinct behavior in spring and summer in both OPER and ML3 when compared with CTR. There is a significant reduction of runoff during spring (peak in April) matching an increase of soil water storage that is also evident in the mean soil moisture (see Fig. 9b). During summer there is the opposite signal with higher soil moisture depletion in OPER and ML3 than in CTR owing to higher runoff rates.
(a) Mean annual cycles of surface water fluxes simulated by CTR averaged over NH40: total precipitation (thin black), evapotranspiration (gray), terrestrial water storage variation (TWSV thick black), and runoff (diamond symbol). The fluxes’ sign follow the model’s convention: positive—incoming water to the surface; negative—outgoing water from the surface. (b)–(e) The mean annual cycle of simulated differences between ML3 and CTR (dashed) and OPER and CTR (solid) of (b) precipitation, (c) evapotranspiration, (d) runoff, and (e) TWSV. The symbols mark if the monthly differences are significant at p < 0.01 (two-tailed t test) between ML3 and CTR (squares) and OPER and CTR (circles).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
Mean annual cycle of (a) simulated snow mass and (b) top 1-m soil moisture content averaged over NH40 by CTR (gray), OPER (thin black), and ML3 (thick black).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
OPER and ML3 also show a higher snow mass peak in winter and delayed snowmelt (Fig. 9a) that contribute to the reduction of runoff. The delayed snowmelt was discussed in the previous section in terms of snow cover. The delayed depletion of snow mass in OPER and ML3 is also associated with the interception of rainfall by the snow schemes (Table 1). The increased soil moisture availability during summer leads to higher evapotranspiration rates in OPER and ML3. Following the increased moisture supply to the atmosphere, there is also an increase of the precipitation rates due to local–regional recycling and to the soil–precipitation feedback (Schar et al. 1999). Compared with OPER, ML3 has a slightly higher spring and summer soil moisture content, which explains the higher evapotranspiration rates and, consequently, precipitation changes in respect to CTR.
Figure 10 compares the mean annual cycle of precipitation of the coupled simulations against ERA-Interim and several observational-based global datasets: Climate Research Unit (CRU version 3.1 updated by Mitchell and Jones 2005), the Global Precipitation Climate Centre (GPCC version 4.0; Schneider et al. 2010; see http://gpcc.dwd.de/). the Global Precipitation Climatology Project [GPCP version 2.1 (Huffman et al. 2009) and version 2.2] and the Climate Prediction Centre (CPC) Merged Analysis of Precipitation (CMAP; Xie and Arkin 1997). The mean annual cycles were evaluated for the period 1979–2006 in all the datasets and for different regions: the Himalayas, Rocky Mountains, and NH40. In the NH40 region (Fig. 10c) the uncertainty between the different datasets reaches about 25% of the amplitude of the annual cycle. In this region, the coupled simulations are in good agreement with the datasets, apart from an overestimation during spring. It is also possible to identify the increase of precipitation in OPER and ML3 (discussed before in Fig. 8). In both the Himalayan and Rocky Mountains regions, the coupled simulations tend to overestimate precipitation during winter and underestimate it during summer. However, these results highlight the large uncertainty among these global datasets; its comparison with the coupled simulations is not trivial and is beyond the scope of the present work. Moreover, this comparison does not evaluate the partitioning between solid and liquid phases, which is of paramount importance in snow-cover simulations (Dutra et al. 2011a). Nevertheless, the coherent signal of overestimation of precipitation over the Himalayan and Rockies regions partially explains the overestimation of snow cover (see Fig. 5) and surface albedo (see Fig. 7) during spring.
Mean annual cycle of precipitation over the (a) Himalayans (22°–47°N, 65°–110°E, land only), (b) Rocky Mountains (30°–48°N, 125°–100°W, land only), and (c) NH40. The panels compare different observational datasets: CRU3.1 (green), GPCCv4 (gray), CMAP (cyan), GPCPv2.1 (dashed red), and GPCPv2.2 (solid red) with the ECMWF ERAI reanalysis (blue) and the coupled simulations of CTR (thick black), OPER (dashed black), and ML3 (thin black).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
c. Surface energy balance
Following the analysis of the water balance components in the previous section, the mean annual cycles of the surface energy fluxes, averaged over NH40, are represented in Fig. 11. There is a significant reduction of surface solar radiation (SSR—downward minus upward) from March to May in both OPER and ML3. This reduction is mainly associated with the increased surface albedo and delayed snowmelt (see Figs. 6 and 9a). There is a reduction of both downward and upward thermal radiation resulting from the cooling of the atmosphere and surface from CTR to OPER and ML3 (see Fig. 12). This results in an increase of surface thermal radiation (STR—downward minus upward), which is more pronounced during winter. The decrease (positive differences) of the surface sensible heat flux (SSHF) in OPER and ML3 is also associated with the cooling of the atmosphere and with the reduction of SSR. The surface latent heat flux (SLHF) shows a significant increase (negative differences) in summer due to increased evapotranspiration rates (see Fig. 8c). The surface net radiation reflects the combination of the above-mentioned changes, with a warming of the soil and cooling of the atmosphere during winter in OPER and ML3 when compared with CTR.
(a) Mean annual cycle of the surface energy fluxes simulated by CTR averaged over NH40: SSR (star symbol), STR (diamond symbol), SSHF (gray line), SLHF (black line), and surface net energy (SNET; thick black). The fluxes sign follow the model’s convention: positive—incoming energy to the surface; negative—outgoing energy from the surface. (b)–(f) The mean annual cycle of simulated differences between ML3 and CTR (dashed) and OPER and CTR (solid) of (b) SSR, (c) STR, (d) SSHF, (e) SLHF, and (f) SNET. The symbols mark if the monthly differences are significant at p < 0.01 (two-tailed t test) between ML3 and CTR (squares) and OPER and CTR (circles).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
Mean annual cycle of (a) simulated T2M, (b) skin temperature, and (c) soil temperature (first soil layer between 0- and 7-cm depth) averaged over NH40. T2M is represented as the differences between simulations and CRU, while skin and soil temperature are the differences of OPER and ML3 in respect to CTR. Note that the horizontal axis starts in September.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
The mean annual cycle of 2 m (T2M, or air temperature), skin, and soil temperature (surface to 7-cm depth) in Fig. 12 are consistent with the changes in surface fluxes presented in Fig. 11. The changes in surface fluxes, as well as temperature, are larger in ML3 than in OPER when compared with CTR. These changes are mainly due to the increased snow thermal insulation of the atmosphere from the underlying soil, which is much stronger in ML3. In CTR, heat exchanges between the atmosphere and the surface during winter led to colder soil temperature and warmer air temperature. OPER simulates lower snow densities (see DU10 for details), leading to a stronger insulation of the atmosphere from the soil. This effect is stronger in ML3 since the multilayer snow scheme also accounts for the thermal insulation within the snowpack (i.e., the upper layer thermally insulates the lower layers).
The simulated T2M is compared against the CRU Time Series version 3.0 (TS3.0) global dataset of monthly temperature (updated from Mitchell and Jones 2005) for the period 1979–2006. Over NH40 (see Fig. 12a) the T2M mean bias during winter was reduced from +2.2 K in CTR to 0.9 K in OPER and −0.1 K in ML3. The mean winter differences between CTR and CRU T2M (Fig. 13a) show a pronounced warm bias, reaching 8 K over snow-covered regions. Both OPER and ML3 lead to a significant cooling over snow-covered regions. Over the Himalaya and Rockies, CTR presents a negative T2M bias that was intensified in OPER and ML3. Part of the cold bias can be associated with an overestimation of snow cover and surface albedo (see Figs. 6 and 7) that is related with an overestimation of precipitation during winter in those regions (see Fig. 10).
Simulated winter 2-m temperature (K) biases of (a) CTR, (b) OPER, and (c) ML3 compared against CRU, and differences between (d) OPER and CTR, (e) ML3 and CTR, and (f) ML3 and OPER. Only differences significant at p < 0.05 are represented. Note the different color scales between (a)–(c) and (d)–(f).
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
d. Snow–albedo feedback
HQ06 proposed the calculation of (Δαs/ΔTair) in the seasonal cycle by taking climatological changes in the Northern Hemisphere from one month to another. They also estimated an observed (Δαs/ΔTair) in the seasonal cycle context, taking changes from April to May values from the satellite-based International Satellite Cloud Climatology Project (ISCCP) and T2M from reanalysis data. In 17 climate models used for Fourth Assessment Report (AR4) climate change experiments, the ratio of (Δαs/ΔTair) evaluated as changes from present to future were very similar to the changes from April to May in present climate. The models showed a strong variability of the ratio (Δαs/ΔTair) with differences up to a factor of 2 when compared with the observational estimates. HQ06 also found that the intermodel variation of (∂αp/∂αs) was small because of the convergence of the models in representing the atmosphere’s interaction with upwelling solar photons (Qu and Hall 2006).
The (Δαs/ΔTair) SAF component estimated from the EC-EARTH simulations and the observational estimates from HQ06 are presented in Table 2. CTR had a reduced sensitivity of surface albedo to surface temperature changes—almost 50% less than the estimates of HQ06. Both OPER and ML3 have similar SAF sensitivity values that are closer to the HQ06 estimates. The increased values of OPER and ML3 when compared with CTR were mainly due to an increased change in surface albedo from April do May. This is due to the later spring melting and to a stronger sensitivity of surface albedo to surface temperature changes in both OPER and ML3 when compared with CTR.
Sensitivity of surface albedo to surface temperature changes from April to May (Δαs/ΔTair) averaged over the Northern Hemisphere landmasses poleward of 30°N for the three simulations and the observational-based estimates from HQ06. The values of surface albedo were weighted by April incoming insulation prior to averaging. The values between brackets represent the std dev of the mean for CTR, OPER, and ML3 and the error estimate for HQ06.
e. Upper air
The changes in upper air temperature and specific humidity of OPER and ML3 with respect to CTR are presented in Fig. 14 as the mean annual cycle of time–pressure cross-section average of NH40. Both OPER and ML3 show a cooling of the lower troposphere up to 700 hPa during winter. During spring the cooling extends further in altitude, reaching 300 hPa. Although the surface cooling in OPER and ML3 has their strongest signal during winter, the vertical propagation is constrained by the strong stability of the planetary boundary layer. As for the surface temperature, the cooling in ML3 is stronger than in OPER (cf. Figs. 14a,b).
Mean annual cycle of time-pressure cross section of (a),(b) temperature and (c),(d) specific humidity differences of (a),(c) OPER and (b),(d) ML3 in respect to CTR averaged over NH40. Solid (dashed) contours represent negative (positive) differences with an interval of 0.3 K and 2 × 10−5 kg kg−1 for temperature and specific humidity, respectively. Dark (light) shading indicates regions where the negative (positive) differences are significant at p < 0.05.
Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-072.1
The air temperature cooling has a direct impact on the reduction of the atmosphere water content at saturation. Both OPER and ML3 show a reduction of specific humidity in the lower troposphere during winter and spring (see Figs. 14c,d). In addition, the near-surface cooling in OPER and ML3 (in comparison with CTR) implies a reduction of moist convection, which in turn would reduce the moistening of the atmospheric column. However, during summer there is an increased water availability in the lower troposphere from CTR to OPER and ML3—higher in the latter. This is coherent with the increased evapotranspiration (see Fig. 8c) that ultimately led to an increase of precipitation. The changes in upper air temperature and specific humidity in ML3 are stronger, both in spatial distribution and intensity, than in OPER when compared with CTR. This is consistent with the previous analysis to the surface water and energy fluxes, showing that the snow thermal insulation plays an important role in controlling the land–atmosphere coupling.
There is a reduction of the geopotential height in OPER and ML3 as a result of the atmospheric cooling during winter and spring (not shown). The mean sea level pressure presents a positive anomaly (mainly over Siberia; not shown) in OPER and ML3 when compared with CTR that is associated with the cooling of the atmosphere. However, the changes in mass distribution are difficult to analyze because of their reduced statistical significance. Further analysis of changes in local and/or hemispheric circulation patterns would require an increase of the ensemble size to properly disentangle the changes due to the different snow schemes from those linked with atmospheric internal variability, which is out of the scope of the present study.
5. Conclusions
The offline simulations of the three snow schemes at Col de Porte and SNOWMIP2 sites were evaluated as a validation of the ML3 snow scheme. ML3, the multilayer version of OPER, improved the snowpack simulations mainly over deep snowpack conditions. This was expected, since a bulk snowpack has a large thermal inertia, with difficulties in representing sporadic melting events during winter and early spring. On shallow snowpacks under high vegetation, ML3 had similar results to OPER. These results show that developments associated with the increased complexity of the snowpack vertical structure should be also associated with a revision of other processes such as subgrid-scale variability, radiative effects of the canopy, and coupling with the atmosphere via the turbulent heat fluxes, among others. However, ML3 has a very simple vertical structure (only three layers with a minimum depth of 0.2 m in each layer). Although sharing a similar construction to other multilayer snow schemes used within climate models (e.g., Boone and Etchevers 2001; Sun et al. 1999) but with a different vertical grid, there are more complex formulations for the vertical structure and physical process of the snowpack available in the literature (e.g., Anderson 1976; Brun et al. 1989). Nevertheless, the implementation and validation of such a complex snow scheme within EC-EARTH is beyond the scope of this study.
EC-EARTH climate simulations of snow cover and surface albedo were compared against remote sensing products. ML3 simulations of snow cover and surface albedo were analogous to OPER. This was foreseen, since both OPER and ML3 share the same parameterizations for snow-cover fraction and snow albedo. The early spring snowmelt in CTR associated with an underestimation of surface albedo was partially reduced with the OPER snow scheme. This underestimation of surface albedo can be due to an early melting (exposing surfaces with lower albedos) and/or an underestimation of snow albedo in the model. The early melting could be also related with an underestimation of snowfall in the coupled model. Dutra et al. (2011a) found that in offline simulations the quality of atmospheric forcing (both total precipitation and the partitioning between liquid and solid phases) is very important for an accurate snow cover, while the model parameterizations were more important for simulating surface albedo. While it was possible to clearly identify the problems in the CdP simulations (overestimation of snow albedo in critical time periods), the general underestimation of surface albedo in the coupled simulations in northern latitudes during spring is associated with a series of interconnected processes. The improved performance of OPER over CTR was also found in offline simulations by DU10. The resemblance between the offline (DU10) and atmosphere-coupled simulations highlight the reliability of the offline strategy to validate land surface models and the importance of the model physics in simulating snow cover.
Over the Himalayan range and Rockies, OPER and ML3 overestimate snow cover and surface albedo during winter and spring (see Figs. 5 and 7). This bias was already present in CTR, but was intensified in OPER, and a similar signal was found in offline simulations with HTESSEL (Dutra et al. 2011a). Over those areas there is a cold bias of T2M (see Fig. 13). Part of the bias can be directly associated with the excess snow cover and surface albedo that is reflecting part of the incoming solar radiation. This explanation is supported by the overestimation of precipitation in those regions during winter (see Fig. 10). However, other atmospheric process such as cloud cover can also be responsible for the cold bias. Furthermore, the subgrid-scale distribution of snow cover over complex terrain is not taken into account in EC-EARTH, with a significant impact on snow-cover evolution and associated energy and moisture fluxes (Liston 2004). These results suggest that the impact of snow schemes over high–complex orography deserves dedicated studies that involve the capacity of resolving fine resolution scales (Bernier et al. 2011). This is also associated with the uncertainty of precipitation simulations in such regions. For example, Minder et al. (2011) found that the choice of microphysical parameterizations is an important source of uncertainty in simulating mountain snowfall with a mesoscale numerical model.
The increased snow thermal insulation from CTR to OPER and ML3 lead to warmer soil temperatures during winter and spring. This reduced soil freezing, allowing snowmelt and rainfall to infiltrate the soil instead of generating surface runoff. The increased soil moisture content during late spring and summer is then responsible for higher evapotranspiration rates. This, in turn, resulted in an increase of summer precipitation due to the increased moisture in the lower troposphere and an intensification of the soil–precipitation feedback. The sensitivity of surface albedo to surface temperature changes from April to May was strongly underestimated by CTR, while OPER and ML3 display similar values, closer to the estimates of HQ06. This chain of processes started during the cold season and ended during summer. These findings highlight the importance of snow cover and of its representation in climate models on both surface water and energy budgets.
Furthermore, the increased snow thermal insulation in OPER and ML3 reduced the T2M warm bias over snow-covered regions during winter and spring. Hazeleger et al. (2010) found similar improvements in coupled ocean–atmosphere simulations when running EC-EARTH with CTR and OPER snow schemes. Cook et al. (2008) reported similar results when investigating the sensitivity of surface climate to the treatment of snow thermal conductivity. In their experiments, high versus low insulation cases led to soil cooling of up to 20 K in winter and T2M warming of 6 K. Compared with this study, CTR would be the low-insulation case, OPER an intermediate case, and ML3 a high insulation. While Cook et al. (2008) demonstrated the importance of snow thermal conductivity by prescribing its value in a set of experiments, in this study we achieved similar results by using different snow schemes.
Acknowledgments
The authors are grateful for the valuable support of the EC-EARTH consortium members. The National Snow and Ice Data Center is acknowledged for providing the snow-cover data. We also thank the MODIS albedo science team and the Land Processes Distributed Active Archive Center (LP DAAC) for providing the MCD43C albedo data. The data from Col De Porte station was kindly provided by Samuel Morin and Pierre Etchevers, Meteo-France. We also thank Xin Qu for providing the SAF estimates and SnowMIP2 data providers. We acknowledge two anonymous reviewers for their valuable suggestions. This work was supported by the Portuguese Foundation for Science and Technology (FCT) under project AMIC PTDC/AAC-CLI/109030/2008 cofinanced by the European Union under program FEDER. E. Dutra acknowledges the financial support of FCT under Grant SFRH/BD/35789/2007 and ECMWF for travel support. IDL is funded by FCT under Grant Pest-OE/CTE/LA0019/2011-IDL.
APPENDIX
Multilayer Snow Scheme
The ML3 snow scheme is a multilayer version of OPER for the energy and mass balance within the snowpack. ML3 has the same snow-cover fraction, surface albedo, and snow density parameterizations (see DU10 for further details)—the latter evaluated for each snow layer separately. This multilayer scheme follows closely the formulation of the Interactions between Soil, Biosphere, and Atmosphere–explicit snow (ISBA-ES) snow scheme (Boone and Etchevers 2001) with some adaptations required for HTESSEL, and with a different snow-layering procedure.
The adopted time step numerical procedure is the following:
skin energy balance is solved to return the residual energy to the snowpack (HTESSEL component);
updated snow mass with snowfall and snow sublimation (first snow layer);
the energy balance for each layer is evaluated by solving the linearized system of equations simultaneously;
water–ice phase changes are evaluated from the updated temperature profile. In the case of melting in the first layer, Tsn,1 is set to the freezing point, and point 3) is repeated again;
phase changes and layer runoff are evaluated to update the profiles of Tsn,i, Si, and Sl,i;
snow density (for each layer) and surface snow albedo are updated; and
reset of vertical grid according to Eq. (A4) for the depth of each layer, conserving mass and energy to obtain new profiles of Tsn,i, Si, Sl,i, and ρsn,i. This last point leads, in some degree, to a vertical mixing of properties, especially during snowmelt conditions.
REFERENCES
Anderson, E. A., 1976: A point energy and mass balance model of a snow cover. NOAA Tech. Rep. NWQ 19, 150 pp.
Armstrong, R. L., and Brodzik M. J. , 2007: Northern Hemisphere EASE-Grid weekly snow cover and sea ice extent version 3. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://nsidc.org/data/nsidc-0046.html.]
Balsamo, G., Viterbo P. , Beljaars A. , Van den Hurk B. , Betts A. K. , and Scipal K. , 2009: A revised hydrology for the ECMWF model: Verification from field site to terrestrial water storage and impact in the Integrated Forecast System. J. Hydrometeor., 10, 623–643.
Bartlett, M. G., Chapman D. S. , and Harris R. N. , 2004: Snow and the ground temperature record of climate change. J. Geophys. Res., 109, F04008, doi:10.1029/2004JF000224.
Bartlett, M. G., Chapman D. S. , and Harris R. N. , 2005: Snow effect on North American ground temperatures, 1950–2002. J. Geophys. Res., 110, F03008, doi:10.1029/2005JF000293.
Bartlett, P. A., MacKay M. D. , and Verseghy D. L. , 2006: Modified snow algorithms in the Canadian land surface scheme: Model runs and sensitivity analysis at three boreal forest stands. Atmos.–Ocean, 44, 207–222, doi:10.3137/ao.440301.
Bernier, N. B., Belair S. , Bilodeau B. , and Tong L. , 2011: Near-surface and land surface forecast system of the Vancouver 2010 Winter Olympic and Paralympic Games. J. Hydrometeor., 12, 508–530.
Boone, A., and Etchevers P. , 2001: An intercomparison of three snow schemes of varying complexity coupled to the same land surface model: Local-scale evaluation at an Alpine site. J. Hydrometeor., 2, 374–394.
Brun, E., Martin E. , Simon V. , Gendre C. , and Coleou C. , 1989: An energy and mass model of snow cover suitable for operational avalanche forecasting. J. Glaciol., 35, 333–342.
Brun, E., Yang Z. , Essery R. , and Cohen J. , 2008: Snow-cover parameterization and modeling. Snow and Climate, R. L. Armstrong and E. Brun, Eds., Cambridge University Press, 125–180.
Cess, R. D., and Potter G. L. , 1988: A methodology for understanding and intercomparing atmospheric climate feedback processes in general circulation models. J. Geophys. Res., 93 (D7), 8305–8314, doi:10.1029/JD093iD07p08305.
Choi, G., Robinson D. A. , and Kang S. , 2010: Changing Northern Hemisphere snow seasons. J. Climate, 23, 5305–5310.
Cook, B. I., Bonan G. B. , Levis S. , and Epstein H. E. , 2008: The thermoinsulation effect of snow cover within a climate model. Climate Dyn., 31, 107–124.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/qj.828.
Douville, H., Royer J. F. , and Mahfouf J. F. , 1995: A new snow parameterization for the Meteo-France climate model. 1. Validation in stand-alone experiments. Climate Dyn., 12, 21–35.
Dutra, E., Balsamo G. , Viterbo P. , Miranda P. M. A. , Beljaars A. , Schar C. , and Elder K. , 2010: An improved snow scheme for the ECMWF land surface model: Description and offline validation. J. Hydrometeor., 11, 899–916.
Dutra, E., Kotlarski S. , Viterbo P. , Balsamo G. , Miranda P. M. A. , Schär C. , Bissolli P. , and Jonas T. , 2011a: Snow cover sensitivity to horizontal resolution, parameterizations and atmospheric forcing in a land surface model. J. Geophys. Res., 116, D21109, doi:10.1029/2011JD016061.
Dutra, E., Schär C. , Viterbo P. , and Miranda P. M. A. , 2011b: Land-atmosphere coupling associated with snow cover. Geophys. Res. Lett., 38, L15707, doi:10.1029/2011gl048435.
ECMWF, cited 2011: IFS Documentation CY28r1 IV: Physical processes. [Available online at http://www.ecmwf.int/research/ifsdocs/CY28r1/Physics/index.html.]
Etchevers, P., and Coauthors, 2004: Validation of the energy budget of an alpine snowpack simulated by several snow models (SnowMIP project). Ann. Glaciol., 38, 150–158, doi:10.3189/172756404781814825.
Ge, Y., and Gong G. , 2010: Land surface insulation response to snow depth variability. J. Geophys. Res., 115, D08107, doi:10.1029/2009JD012798.
Groisman, P. Ya., Karl T. R. , Knight R. W. , and Stenchikov G. L. , 1994: Changes of snow cover, temperature, and radiative heat balance over the Northern Hemisphere. J. Climate, 7, 1633–1656.
Hall, A., and Qu X. , 2006: Using the current seasonal cycle to constrain snow albedo feedback in future climate change. Geophys. Res. Lett., 33, L03502, doi:10.1029/2005gl025127.
Hazeleger, W., and Coauthors, 2010: EC-Earth: A seamless earth-system prediction approach in action. Bull. Amer. Meteor. Soc., 91, 1357–1363.
Huffman, G. J., Adler R. F. , Bolvin D. T. , and Gu G. , 2009: Improving the global precipitation record: GPCP Version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009gl040000.
Immerzeel, W. W., van Beek L. P. H. , and Bierkens M. F. P. , 2010: Climate change will affect the Asian water towers. Science, 328, 1382–1385, doi:10.1126/science.1183188.
Liston, G. E., 2004: Representing subgrid snow cover heterogeneities in regional and global models. J. Climate, 17, 1381–1397.
Lynch-Stieglitz, M., 1994: The development and validation of a simple snow model for the GISS GCM. J. Climate, 7, 1842–1855.
Marks, D., Reba M. , Pomeroy J. , Link T. , Winstral A. , Flerchinger G. , and Elder K. , 2008: Comparing simulated and measured sensible and latent heat fluxes over snow under a pine canopy to improve an energy balance snowmelt model. J. Hydrometeor., 9, 1506–1522.
Minder, J. R., Durran D. R. , and Roe G. H. , 2011: Mesoscale controls on the mountainside snow line. J. Atmos. Sci., 68, 2107–2127.
Mitchell, T. D., and Jones P. D. , 2005: An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol., 25, 693–712.
Qu, X., and Hall A. , 2006: Assessing snow albedo feedback in simulated climate change. J. Climate, 19, 2617–2630.
Rangwala, I., Miller J. R. , Russell G. L. , and Xu M. , 2010: Using a global climate model to evaluate the influences of water vapor, snow cover and atmospheric aerosol on warming in the Tibetan Plateau during the twenty-first century. Climate Dyn., 34, 859–872, doi:10.1007/s00382-009-0564-1.
Roesch, A., 2006: Evaluation of surface albedo and snow cover in AR4 coupled climate models. J. Geophys. Res., 111, D15111, doi:10.1029/2005jd006473.
Roman, M. O., and Coauthors, 2009: The MODIS (Collection V005) BRDF/albedo product: Assessment of spatial representativeness over forested landscapes. Remote Sens. Environ., 113, 2476–2498.
Rutter, N., and Coauthors, 2009: Evaluation of forest snow processes models (SnowMIP2). J. Geophys. Res., 114, D06111, doi:10.1029/2008JD011063.
Salomon, J. G., Schaaf C. B. , Strahler A. H. , Gao F. , and Jin Y. F. , 2006: Validation of the MODIS Bidirectional Reflectance Distribution Function and Albedo retrievals using combined observations from the Aqua and Terra platforms. IEEE Trans. Geosci. Remote Sens., 44, 1555–1565.
Schar, C., Luthi D. , Beyerle U. , and Heise E. , 1999: The soil–precipitation feedback: A process study with a regional climate model. J. Climate, 12, 722–741.
Schneider, U., Becker A. , Meyer-Christoffer A. , Ziese M. , and Rudolf B. , 2010: Global precipitation analysis products of the GPCC. Global Precipitation Climatology Centre DWD Tech. Rep., 12 pp. [Available online at http://gpcc.dwd.de.]
Stieglitz, M., and Smerdon J. E. , 2007: Characterizing land–atmosphere coupling and the implications for subsurface thermodynamics. J. Climate, 20, 21–37.
Stroeve, J., Box J. E. , Gao F. , Liang S. L. , Nolin A. , and Schaaf C. , 2005: Accuracy assessment of the MODIS 16-day albedo product for snow: Comparisons with Greenland in situ measurements. Remote Sens. Environ., 94, 46–60.
Sun, S. F., Jin J. M. , and Xue Y. K. , 1999: A simple snow-atmosphere-soil transfer model. J. Geophys. Res., 104 (D16), 19 587–19 597.
Sverdrup, H. U., 1935: Scientific results of the Norwegian-Swedish Spitsbergen expedition in 1934. Part IV. The ablation on Ischsen’s Plateau and on the 14th of July glacier. Geogr. Ann., 17, 145–218.
Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131, 2961–3012, doi:10.1256/qj.04.176.
Viterbo, P., and Betts A. K. , 1999: Impact on ECMWF forecasts of changes to the albedo of the boreal forests in the presence of snow. J. Geophys. Res., 104 (D22), 27 803–27 810.
Viterbo, P., Beljaars A. , Mahfouf J.-F. , and Teixeira J. , 1999: The representation of soil moisture freezing and its impact on the stable boundary layer. Quart. J. Roy. Meteor. Soc., 125, 2401–2426, doi:10.1002/qj.49712555904.
Wang, Z., and Zeng X. B. , 2010: Evaluation of snow albedo in land models for weather and climate studies. J. Appl. Meteor. Climatol., 49, 363–380.
Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences. Academic Press, 648 pp.
Williams, J., 1975: The influence of snowcover on the atmospheric circulation and its role in climatic change: An analysis based on results from the NCAR global circulation model. J. Appl. Meteor., 14, 137–152.
Xie, P., and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539–2558.
Zhou, L., and Coauthors, 2003: Comparison of seasonal and spatial variations of albedos from Moderate-Resolution Imaging Spectroradiometer (MODIS) and Common Land Model. J. Geophys. Res., 108, 4488, doi:10.1029/2002jd003326.