• Balsamo, G., , A. Beljaars, , and K. Scipal, 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, 623643, doi:10.1175/2008JHM1068.1.

    • Search Google Scholar
    • Export Citation
  • Balsamo, G., and Coauthors, 2015: ERA-Interim/Land: A global land surfaces reanalysis data set. Hydrol. Earth Syst. Sci. Discuss., 19, 389407, doi:10.5194/hess-19-389-2015.

    • Search Google Scholar
    • Export Citation
  • Bonan, G. B., , K. W. Oleson, , M. Vertenstein, , S. Levis, , X. Zeng, , Y. Dai, , R. E. Dickinson, , and Z.-L. Yang, 2002: The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model. J. Climate, 15, 31233149, doi:10.1175/1520-0442(2002)015<3123:TLSCOT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Brasnett, B., 1999: A global analysis of snow depth for numerical weather prediction. J. Appl. Meteor., 38, 726740, doi:10.1175/1520-0450(1999)038<0726:AGAOSD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., , and P. W. Mote, 2009: The response of Northern Hemisphere snow cover to a changing climate. J. Climate, 22, 2124–2145, doi:10.1175/2008JCLI2665.1.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., , and B. Brasnett, 2010: Canadian Meteorological Centre (CMC) Daily Snow Depth Analysis Data. National Snow and Ice Data Center Distributed Active Archive Center, Boulder, CO, doi:10.5067/W9FOYWH0EQZ3.

  • Brown, R. D., , and C. Derksen, 2013: Is Eurasian October snow cover extent increasing? Environ. Res. Lett., 8, 024006, doi:10.1088/1748-9326/8/2/024006.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., , C. Derksen, , and L. Wang, 2010: A multi-data set analysis of variability and change in Arctic spring snow cover extent, 1967–2008. J. Geophys. Res., 115, D16111, doi:10.1029/2010JD013975.

    • Search Google Scholar
    • Export Citation
  • Brun, E., , P. David, , M. Sudul, , and G. Brunot, 1992: A numerical-model to simulate snow-cover stratigraphy for operational avalanche forecasting. J. Glaciol., 38, 1322.

    • Search Google Scholar
    • Export Citation
  • Brun, E., , V. Vionnet, , A. Boone, , B. Decharme, , Y. Peings, , R. Valette, , F. Karbou, , and S. Morin, 2013: Simulation of northern Eurasian local snow depth, mass, and density using a detailed snowpack model and meteorological reanalyses. J. Hydrometeor., 14, 203219, doi:10.1175/JHM-D-12-012.1.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and Coauthors, 1996: Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101, 72517268, doi:10.1029/95JD02165.

    • Search Google Scholar
    • Export Citation
  • De Lannoy, G. J. M., , R. H. Reichle, , P. R. Houser, , K. R. Arsenault, , N. E. C. Verhoest, , and V. R. N. Pauwels, 2010: Satellite-scale snow water equivalent assimilation into a high-resolution land surface model. J. Hydrometeor., 11, 352369, doi:10.1175/2009JHM1192.1.

    • Search Google Scholar
    • Export Citation
  • Derksen, C., , and R. Brown, 2012: Spring snow cover extent reductions in the 2008–2012 period exceeding climate model projections. Geophys. Res. Lett., 39, L19504, doi:10.1029/2012GL053387.

    • Search Google Scholar
    • Export Citation
  • Drewitt, G., , A. A. Berg, , W. J. Merryfield, , and W. Lee, 2012: Effect of realistic soil moisture initialization on the Canadian CanCM3 seasonal forecast model. Atmos.–Ocean, 50, 466474, doi:10.1080/07055900.2012.722910.

    • Search Google Scholar
    • Export Citation
  • Dutra, E., , G. Balsamo, , P. Viterbo, , P. M. A. Miranda, , A. Beljaars, , C. Schär, , and K. Elder, 2010: An improved snow scheme for the ECMWF land surface model: Description and offline validation. J. Hydrometeor., 11, 899916, doi:10.1175/2010JHM1249.1.

    • Search Google Scholar
    • Export Citation
  • Dutra, E., , P. Viterbo, , P. M. A. Miranda, , and G. Balsamo, 2012: Complexity of snow schemes in a climate model and its impact on surface energy and hydrology. J. Hydrometeor., 13, 521538, doi:10.1175/JHM-D-11-072.1.

    • Search Google Scholar
    • Export Citation
  • Jeong, J., , H. W. Linderholm, , S. Woo, , C. Folland, , B. Kim, , S. Kim, , and D. Chen, 2013: Impacts of snow initialization on subseasonal forecasts of surface air temperature for the cold season. J. Climate, 26, 19561972, doi:10.1175/JCLI-D-12-00159.1.

    • Search Google Scholar
    • Export Citation
  • Koren, V., , J. Schaake, , K. Mitchell, , Q.-Y. Duan, , F. Chen, , and J. M. Baker, 1999: A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys. Res., 104, 19 56919 585, doi:10.1029/1999JD900232.

    • Search Google Scholar
    • Export Citation
  • Koster, R. D., , and M. J. Suarez, 1994: The components of a ‘SVAT’ scheme and their effects on a GCM’s hydrological cycle. Adv. Water Resour., 17, 6178, doi:10.1016/0309-1708(94)90024-8.

    • Search Google Scholar
    • Export Citation
  • Koster, R. D., , M. J. Suarez, , A. Ducharne, , M. Stieglitz, , and P. Kumar, 2000: A catchment-based approach to modeling land surface processes in a general circulation model: 1. Model structure. J. Geophys. Res., 105, 24 80924 822, doi:10.1029/2000JD900327.

    • Search Google Scholar
    • Export Citation
  • Koster, R. D., and Coauthors, 2010: Contribution of land surface initialization to subseasonal forecast skill: First results from a multi-model experiment. Geophys. Res. Lett., 37, L02402, doi:10.1029/2009GL041677.

    • Search Google Scholar
    • Export Citation
  • Langlois, A., and Coauthors, 2009: Simulation of snow water equivalent (SWE) using thermodynamic snow models in Québec, Canada. J. Hydrometeor., 10, 14471463, doi:10.1175/2009JHM1154.1.

    • Search Google Scholar
    • Export Citation
  • Liang, X., , D. P. Lettenmaier, , E. F. Wood, , and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 41514 428, doi:10.1029/94JD00483.

    • Search Google Scholar
    • Export Citation
  • Mudryk, L. R., , P. J. Kushner, , and C. Derksen, 2014: Interpreting observed Northern Hemisphere snow trends with large ensembles of climate simulations. Climate Dyn., 43, 345359, doi:10.1007/s00382-013-1954-y.

    • Search Google Scholar
    • Export Citation
  • NOAA National Geophysical Data Center, 1988: Data Announcement 88-MGG-02, Digital relief of the Surface of the Earth. NOAA, National Geophysical Data Center, Boulder, CO. [Available online at http://www.ngdc.noaa.gov/mgg/global/etopo5.html.]

  • Orsolini, Y. J., , R. Senan, , G. Balsamo, , F. J. Doblas-Reyes, , F. Vitart, , A. Weisheimer, , A. Carrasco, , and R. E. Benestad, 2013: Impact of snow initialization on sub-seasonal forecasts. Climate Dyn., 41, 19691982, doi:10.1007/s00382-013-1782-0.

    • Search Google Scholar
    • Export Citation
  • Reichle, R. H., , R. D. Koster, , G. J. M. de Lannoy, , B. A. Forman, , Q. Liu, , S. P. P. Mahanama, , and A. Touré, 2011: Assessment and enhancement of MERRA land surface hydrology estimates. J. Climate, 24, 63226338, doi:10.1175/JCLI-D-10-05033.1.

    • Search Google Scholar
    • Export Citation
  • Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 36243648, doi:10.1175/JCLI-D-11-00015.1.

    • Search Google Scholar
    • Export Citation
  • Rodell, M., , and H. K. Beaudoing, 2007a: GLDAS CLM Land Surface Model L4 3 Hourly 1.0 × 1.0 degree subsetted, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 5 August 2014, doi:10.5067/83NO2QDLG6M0.

  • Rodell, M., , and H. K. Beaudoing, 2007b: GLDAS Mosaic Land Surface Model L4 3 Hourly 1.0 x 1.0 degree subsetted, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 5 August 2014, doi:10.5067/DLVU8VOPKN7L.

  • Rodell, M., , and H. K. Beaudoing, 2007c: GLDAS Noah Land Surface Model L4 3 Hourly 1.0 × 1.0 degree subsetted, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 10 January 2014, doi:10.5067/7J0CPW2PF9N9.

  • Rodell, M., , and H. K. Beaudoing, 2007d: GLDAS VIC Land Surface Model L4 3 Hourly 1.0 × 1.0 degree, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 5 August 2014, doi:10.5067/Z7SA0KYUTBQS.

  • Rodell, M., , and H. K. Beaudoing, 2013: GLDAS Noah Land Surface Model L4 3 Hourly 1.0 × 1.0 degree, version 2.0. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 23 January 2014. [Available online at http://disc.sci.gsfc.nasa.gov/datacollection/GLDAS_NOAH10_3H_V020.shtml.]

  • Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381394, doi:10.1175/BAMS-85-3-381.

    • Search Google Scholar
    • Export Citation
  • Rutter, N., and Coauthors, 2009: Evaluation of forest snow processes models (SnowMIP2). J. Geophys. Res., 114, D06111, doi:10.1029/2008JD011063.

  • Sheffield, J., , G. Goteti, , and E. F. Wood, 2006: Development of a 50-year high-resolution global dataset of meteorological forcings for land surface modeling. J. Climate, 19, 3088–3111, doi:10.1175/JCLI3790.1.

    • Search Google Scholar
    • Export Citation
  • Stieglitz, M., , A. Ducharne, , K. Koster, , and M. Suarez, 2001: The impact of detailed snow physics on the simulation of snow cover and subsurface thermodynamics at continental scales. J. Hydrometeor., 2, 228242, doi:10.1175/1525-7541(2001)002<0228:TIODSP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sturm, M., , J. Holmgren, , and G. E. Liston, 1995: A seasonal snow cover classification system for local to global applications. J. Climate, 8, 12611283, doi:10.1175/1520-0442(1995)008<1261:ASSCCS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Takala, M., , K. Luojus, , J. Pulliainen, , C. Derksen, , L. Lemmetyinen, , J.-P. Kärnä, , and J. Koskinen, 2011: Estimating Northern Hemisphere snow water equivalent for climate research through assimilation of space-borne radiometer data and ground-based measurements. Remote Sens. Environ., 115, 3517–3529, doi:10.1016/j.rse.2011.08.014.

    • Search Google Scholar
    • Export Citation
  • Tong, J., , S. J. Déry, , P. L. Jackson, , and C. Derksen, 2010: Testing snow water equivalent retrieval algorithms for passive microwave remote sensing in an alpine watershed of western Canada. Can. J. Remote Sens., 36, S74S86, doi:10.5589/m10-009.

    • Search Google Scholar
    • Export Citation
  • Vaughan, D. G., and Coauthors, 2013: Observations: Cryosphere. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 317–382.

  • Wang, Z., , X. Zeng, , and M. Decker, 2010: Improving snow processes in the Noah land model. J. Geophys. Res., 115, D20108, doi:10.1029/2009JD013761.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Trends in NH SWM for the five principal datasets over 1981–2010 (colors as marked). The average trend of the four reanalysis-derived datasets over alpine regions has been added to the trend of the GlobSnow product in order to compare trend magnitudes. Also shown are trends for CMC (1999–2010, dotted); GLDAS-1–Noah (1981–2010, dotted–dashed); and MERRA-Land (1983–98, long dash; 1999–2010, short dash).

  • View in gallery

    (a) Time series of anomalous NH SWM from MERRA-Land. (b) Difference in climatological SWE between 1999 and 2010 and 1983 and 1998 periods for MERRA-Land.

  • View in gallery

    (a) Climatological total NH SWM for individual datasets and the multidataset mean for the 1981–2010 period. The average of the NH alpine SWM time series from the four reanalysis datasets is added to the GlobSnow time series in order to appropriately compare total NH SWM. (b) Stacked bar chart of SWM differences between individual datasets and the multidataset mean over pairs of months for various regions. Each individual dataset is colored as in (a) and for a given season, shows a sequence of three bars corresponding to NA, EU, and NH regions. Separate contributions from the three land types are stacked vertically using different shades for midlatitudes, Arctic, and alpine regions; the total difference summed over all land types on a given continent is shown by the bar with the black outline. Crocus and MERRA differences have been multiplied by three; GLDAS-2, ERA-Interim/Land, and GlobSnow differences are unaltered.

  • View in gallery

    Spread among climatologies for NH SWM by region over the 1981–2010 period.

  • View in gallery

    (a) Climatology of multidataset mean SWE for February–March over 1981–2010 period. (b) Ratio of climatological SWE to spread among the five component datasets calculated for February–March over 1981–2010 period. The black contour delineates the 1:1 ratio.

  • View in gallery

    Anomalous total NH nonalpine SWM for the five individual datasets (individually detrended) and the multidataset mean for (top)–(bottom) 1981–1991, 1991–2001, and 2001–2011. Gray shading marks the range of anomalies (spread) among the datasets on a given day.

  • View in gallery

    Correlations of daily (NDJFMA) SWM time series (1981–2010) between pairs of datasets. Correlations are shown for detrended time series of NH (black), NA (red), and EU (blue) SWM as well as nondetrended NH (gray) SWM. Also shown are mean values of the SWE pattern correlation between dataset pairs (green circles), calculated daily and averaged over the NDJFMA season and 1981–2010 period.

  • View in gallery

    Spatial map of the mean correlation between pairs of daily (NDJFMA) SWE time series (1981–2010). Correlations of detrended time series are performed for a given grid cell location between all available pairs of datasets and averaged together.

  • View in gallery

    (a) Spread in anomalous SWM among detrended datasets (black, solid), nondetrended datasets (black, dotted), and interannual variability of multidataset mean (gray, solid) over the 1981–2010 period. (b) Spread in anomalous SWM attributed to each of the five component dataset time series (1981–2010) and the multidataset mean. Total spread (black) is measured on left axis. Spread attributed to individual datasets is measured on the right axis (2 times smaller). A smoothing window of 30 days has been applied to all curves for clarity. See text for definitions of spread and attributed spread.

  • View in gallery

    (a) NH interannual variability of SWM for each dataset. (b) Attributed spread calculated from standardized SWM for each dataset. A smoothing window of 30 days has been applied to all curves for clarity.

  • View in gallery

    (a) Ratio of absolute anomaly size for multidataset mean to dataset spread, calculated daily and averaged over February–March and 1981–2010 period. (b) As in (a), but only for the MERRA, ERA-Interim/Land, and Crocus datasets (group 1, see section 3c). The black contour delineates the 1:1 ratio.

  • View in gallery

    (a) Spread in total NH SWM for group 1 (MERRA, Crocus, and ERA-Interim/Land) plus GLDAS-2 (gray shading). Climatologies for the four GLDAS-1 products (group 2) are shown in the labeled curves. The single GLDAS-2 product is also shown for comparison. (b) Mean pairwise correlation among different groups of SWM time series (2001–10). For each of the three groupings, the average correlation is calculated for NA (red), EU (blue), and NH (black) SWM time series. For each landmass and group, four averages are shown in sequence for SWM summed over all land types, only midlatitude, only the Arctic, and only alpine regions. Mean correlations of group 1 (circles) and group 2 (diamonds) total NH SWM time series (all land types) with GlobSnow are also shown. Correlations with GlobSnow over individual land types are omitted for clarity.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 112 112 55
PDF Downloads 88 88 33

Characterization of Northern Hemisphere Snow Water Equivalent Datasets, 1981–2010

View More View Less
  • 1 Department of Physics, University of Toronto, Toronto, Ontario, Canada
  • 2 Climate Research Division, Environment Canada, Toronto, Ontario, Canada
  • 3 Department of Physics, University of Toronto, Toronto, Ontario, Canada
  • 4 Climate Research Division, Environment Canada, Ouranos, Montreal, Quebec, Canada
© Get Permissions
Full access

Abstract

Five, daily, gridded, Northern Hemisphere snow water equivalent (SWE) datasets are analyzed over the 1981–2010 period in order to quantify the spatial and temporal consistency of satellite retrievals, land surface assimilation systems, physical snow models, and reanalyses. While the climatologies of total Northern Hemisphere snow water mass (SWM) vary among the datasets by as much as 50%, their interannual variability and daily anomalies are comparable, showing moderate to good temporal correlations (between 0.60 and 0.85) on both interannual and intraseasonal time scales. Wintertime trends of total Northern Hemisphere SWM are consistently negative over the 1981–2010 period among the five datasets but vary in strength by a factor of 2–3. Examining spatial patterns of SWE indicates that the datasets are most consistent with one another over boreal forest regions compared to Arctic and alpine regions. Additionally, the datasets derived using relatively recent reanalyses are strongly correlated with one another and show better correlations with the satellite product [the European Space Agency (ESA)’s Global Snow Monitoring for Climate Research (GlobSnow)] than do those using older reanalyses. Finally, a comparison of eight reanalysis datasets over the 2001–10 period shows that land surface model differences control the majority of spread in the climatological value of SWM, while meteorological forcing differences control the majority of the spread in temporal correlations of SWM anomalies.

Corresponding author address: L. R. Mudryk, Department of Physics, University of Toronto, 60 St. George Street, Toronto, ON M5S 1A7, Canada. E-mail: mudryk@atmosp.physics.utoronto.ca

Abstract

Five, daily, gridded, Northern Hemisphere snow water equivalent (SWE) datasets are analyzed over the 1981–2010 period in order to quantify the spatial and temporal consistency of satellite retrievals, land surface assimilation systems, physical snow models, and reanalyses. While the climatologies of total Northern Hemisphere snow water mass (SWM) vary among the datasets by as much as 50%, their interannual variability and daily anomalies are comparable, showing moderate to good temporal correlations (between 0.60 and 0.85) on both interannual and intraseasonal time scales. Wintertime trends of total Northern Hemisphere SWM are consistently negative over the 1981–2010 period among the five datasets but vary in strength by a factor of 2–3. Examining spatial patterns of SWE indicates that the datasets are most consistent with one another over boreal forest regions compared to Arctic and alpine regions. Additionally, the datasets derived using relatively recent reanalyses are strongly correlated with one another and show better correlations with the satellite product [the European Space Agency (ESA)’s Global Snow Monitoring for Climate Research (GlobSnow)] than do those using older reanalyses. Finally, a comparison of eight reanalysis datasets over the 2001–10 period shows that land surface model differences control the majority of spread in the climatological value of SWM, while meteorological forcing differences control the majority of the spread in temporal correlations of SWM anomalies.

Corresponding author address: L. R. Mudryk, Department of Physics, University of Toronto, 60 St. George Street, Toronto, ON M5S 1A7, Canada. E-mail: mudryk@atmosp.physics.utoronto.ca

1. Introduction

The seasonal cycle of terrestrial snow cover and snow mass has a notable influence on the Northern Hemisphere energy budget, water balance, and geochemical cycles. Snow water equivalent (SWE) is expected to respond in a complex way to projected temperature and precipitation changes with the magnitude and sign of the response varying with climate regime and elevation (Brown and Mote 2009). Verification of such responses in climate models and the initialization of snow in seasonal to decadal prediction systems requires a gridded, observational SWE dataset with well-characterized uncertainty (De Lannoy et al. 2010). For snow cover extent, intercomparison of existing data has led to estimation of uncertainties in snow cover extent anomalies and trends (Brown et al. 2010; Derksen and Brown 2012) as well as improved documentation and understanding of systematic differences and inhomogeneities (Brown and Derksen 2013; Mudryk et al. 2014).

A similar quantitative understanding of uncertainties in the Northern Hemisphere is lacking for SWE datasets apart from some more limited comparisons cited below. To address this gap, we compare an ensemble of daily, gridded datasets in order to fully characterize inter-dataset spread and produce a multidataset mean. Our intent is to make available the mean and spread of the SWE datasets analyzed here on the National Center for Atmospheric Research Climate Data Guide (https://climatedataguide.ucar.edu/). All of the datasets include observations (e.g., satellite measurements and observed inputs to reanalysis) as at least a component of the data generation but otherwise draw from a variety of sources including remote sensing, station data, land surface assimilation systems, and reanalysis-driven snow models of varying complexity. In particular we use 1) the European Space Agency (ESA)’s Global Snow Monitoring for Climate Research (GlobSnow), version 2, analysis, combining satellite-based passive microwave retrievals and ground-based weather station data (Takala et al. 2011); 2) the Global Land Data Assimilation System (GLDAS), version 2, product (Rodell et al. 2004); 3) the European Centre for Medium-Range Forecasts interim land surface reanalysis (ERA-Interim/Land), which uses a simple snow scheme (Balsamo et al. 2015); 4) the Modern-Era Retrospective Analysis for Research and Applications (MERRA), which uses an intermediate complexity snow scheme (Rienecker et al. 2011); and 5) SWE from the Crocus snow scheme, a detailed physical snowpack model driven by meteorology from ERA-Interim (Brun et al. 2013).

Evaluations of the above datasets or of the land surface models and snow schemes used to produce them have been conducted at specific locations (e.g., Wang et al. 2010; Dutra et al. 2012; Brun et al. 1992; Langlois et al. 2009; Stieglitz et al. 2001); however, such local validations do not necessarily represent the datasets’ hemisphere-wide fidelity. Indeed, for coarsely gridded datasets such as these, a meaningful hemisphere-scale evaluation with ground measurements is not a trivial undertaking. Single-point climate station or snow survey measurements are inappropriate for the validation of coarse grid cell SWE datasets, and there are insufficient in situ SWE observations available at high enough spatial resolution to develop reliable gridded in situ SWE estimates for continental-scale validation. Surface-based validation in alpine areas, whose gradients in elevation and associated snow properties fall well below the grid scale for current hemispheric-scale products, are particularly challenging. Efforts to address such scaling challenges are ongoing but are outside the scope of this study.

Instead, the objective of this analysis is to exploit the use of multiple datasets to robustly characterize the spatial and temporal agreement in SWE climatologies and interannual anomalies at the hemispheric scale. While the climate modeling community has long recognized the strength in using output from a large number of climate models, such an ensemble-based approach has been less readily adopted by the observational community. Data assessments and intercomparisons have typically focused on identifying the best product. This approach can produce the potentially misleading impression that a single dataset is capable of characterizing the observational truth for all regions and seasons. In reality, variables like SWE are particularly challenging to characterize with coarse-resolution gridded datasets due to significant subgrid heterogeneity in horizontal (i.e., snow depth) and vertical (i.e., snow stratigraphy) properties. Recent intercomparisons limited only to snow models (Rutter et al. 2009) reinforce this perspective, demonstrating that most models are good at simulating certain aspects of observed snow conditions but less good at simulating others. Therefore, we argue that the analysis of multiple datasets is necessary to understand uncertainties and inconsistencies within current SWE datasets. In turn, this understanding of uncertainty can enhance the use of observational snow analyses. For instance, dataset spread can inform the assignment of uncertainty to observations necessary for land surface data assimilation in numerical weather prediction and seasonal forecast applications (Orsolini et al. 2013; De Lannoy et al. 2010; Koster et al. 2010; Jeong et al. 2013; Drewitt et al. 2012). Consideration of spread also reduces the sensitivity of climate model evaluation to the selection of a single dataset for evaluation. This becomes an important issue when disagreement among datasets is large enough to affect the significance of SWE biases depending on the choice of observational data.

In section 2, we provide details of the SWE datasets and how they are combined into a multidataset mean. Several metrics are used to compare individual datasets with one another as well as the multidataset average in section 3. Section 4 contains a summary and discussion of key results.

2. Data and methods

a. Datasets

The SWE datasets used in this study are chosen based on two main criteria: complete Northern Hemisphere spatial coverage (with the exception of an alpine mask applied to GlobSnow) and continuous availability through the satellite era (we use 1981–2010 as our analysis period). We also require relatively homogenous SWE time series for the entire analysis period, which we diagnose from trends in the time series of global snow mass (see section 2c). The component datasets analyzed in this study are described below and summarized in Table 1.

Table 1.

Summary of products referred to in this study. The first five datasets are used in developing the multidataset mean over the 1981–2010 period. The four GLDAS-1 datasets are suitable for analysis over the restricted, 2001–10 period (see section 3c).

Table 1.

The GlobSnow, version 2, product (Takala et al. 2011; www.globsnow.info) is the only satellite-based product in our analysis; however, it also uses ground-based weather station data in the SWE retrieval. Estimates of snow grain size are first derived for grid cells containing weather station snow depth measurements by optimizing agreement between microwave snow model simulations and observed satellite passive microwave brightness temperatures at 19 and 37 GHz. These local estimates of grain size are interpolated via kriging across the Northern Hemisphere and used in a second round of emission model simulations for which grain size is fixed and SWE is optimized. Resolution of the product is 25 km. GlobSnow retrievals over complex terrain defined in section 2b are masked from the standard product due to well-known uncertainties related to subgrid heterogeneity in snow properties and microwave signatures (Tong et al. 2010).

ERA-Interim/Land is a reanalysis product (Balsamo et al. 2015) that diagnoses SWE using the Hydrology Tiled ECMWF Scheme of Surface Exchanges over Land (HTESSEL; Balsamo et al. 2009) driven by meteorological forcing from the atmospheric ERA-Interim. The snow scheme used is a single-layer scheme described in Dutra et al. (2010). The precipitation values used to force the land model are corrected using the Global Precipitation Climatology Project (GPCP) dataset, version 2.1. The product resolution is ° × °.

MERRA (Rienecker et al. 2011) is a National Aeronautics and Space Administration (NASA) atmospheric reanalysis product generated with the Goddard Earth Observing System Model, version 5.2.0 (GEOS-5) atmospheric general circulation model and atmospheric data assimilation system (ADAS). SWE is diagnosed from a hydrological catchment-based land surface model (called Catchment; Koster et al. 2000). Catchment uses an intermediate complexity snow scheme with up to three snow layers describing snow accumulation, melting, refreezing, and compaction in response to surface meteorological conditions (Stieglitz et al. 2001). Product resolution is 1/2° × 1/2°. In section 2c, we also analyze the related product MERRA-Land. This second product is produced by rerunning only the land surface component of MERRA forced by atmospheric data from the standard MERRA product except for precipitation that is forced by the gauge-based National Oceanic and Atmospheric Administration (NOAA)’s Climate Prediction Center unified (CPCU) precipitation dataset. The canonical reference for the MERRA-Land product is Reichle et al. (2011); however, the current working product differs in several ways from the preliminary version described therein, including in the choice of precipitation forcing that Reichle et al. (2011) list as the GPCP dataset. Henceforth, unless explicitly referred to as MERRA-Land, all references to MERRA data refer to the standard Rienecker et al. (2011) product.

The Crocus SWE dataset is from the Interactions between Soil, Biosphere, and Atmosphere (ISBA) land surface model driven by ERA-Interim meteorology. The Crocus snow scheme (Brun et al. 2013) is embedded in ISBA in place of the usual snow scheme. Crocus is a detailed snowpack model with multiple historical snow layers possible, each representing a separate snowfall event. Each layer is described by the thickness, temperature, dry density, liquid water content, and grain type (dendricity, sphericity, size, and age). Resolution of the data is 1° × 1°.

GLDAS, version 2 (GLDAS-2; Rodell et al. 2004; Rodell and Beaudoing 2013), is another NASA product that estimates SWE based on the NOAA/National Centers for Environmental Prediction–Oregon State University–Air Force–NOAA/Hydrologic Research Laboratory (Noah) land surface model, version 3.3 (Chen et al. 1996; Koren et al. 1999), constrained by assimilated observations and land surface parameters. This version is forced using the Princeton meteorological forcing dataset (Sheffield et al. 2006). We also use four different GLDAS, version 1 (GLDAS-1), products (Rodell and Beaudoing 2007a,b,c,d) that have been forced using the Global Data Assimilation System (GDAS) with precipitation adjustments from the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP). Because of changes in the forcing data in GLDAS-1 between 1979 and 2001, the products are unsuitable for analysis over the full 1981–2010 period. However, there are no changes to the forcing meteorology after 2001, and we analyze the products over the 2001–10 period in section 3c. The four GLDAS-1 products each use different land surface models to diagnose SWE: the Variable Infiltration Capacity (VIC) land surface model (Liang et al. 1994); the Community Land Model (CLM), version 2.0 (Bonan et al. 2002); the Mosaic land surface model (Koster and Suarez 1994); and the Noah land surface model, version 2.7. The corresponding snow schemes implemented in these models range from simple single-layer schemes in the Noah and Mosaic models to intermediate complexity schemes in CLM and VIC. Resolution of all GLDAS products used is 1° × 1°.

b. Methods

For each dataset we acquired daily SWE at the native resolution for the 1981–2010 period. Between 1981 and 1987, GlobSnow is only available approximately every second day with occasional gaps of longer duration. We linearly interpolated any temporal gaps in the data using the two nearest dates with available SWE, assuming SWE is uniformly zero between 30 June and 7 September.

Each dataset was interpolated to a regular 1° × 1° longitude–latitude grid. Before interpolating we excluded snow from land ice/glaciers and large lakes based on the MERRA land fraction mask, which specifies the fractional area occupied by land ice or lakes in a given grid cell. We upscaled the MERRA land fraction mask to the native resolution of the other datasets and removed snow from any grid cell containing a nonzero fraction of land ice. Snow over lakes in MERRA was removed in proportion to the percentage of the grid cell area occupied by lakes. The remaining datasets already mask snow over large lakes, which we considered sufficient for our purposes.

To treat alpine and nonalpine regions separately, we also upscaled the 25-km-resolution binary alpine mask applied in the GlobSnow data processing chain to the 1° grid. Topographic information for this mask was derived from ETOPO5 data (NOAA National Geophysical Data Center 1988) that contain global elevation information at a resolution of 5 arc minutes, an appropriate resolution given the 25-km scale of the GlobSnow product. Grid cells are considered mountainous if the standard deviation of the elevation within is larger than 200 m. Once interpolated to the native resolution of the remaining SWE datasets, the mask represents the fraction of a given grid cell that is of a nonalpine land type. Multiplying the regridded SWE data by the fractional alpine mask provides a second SWE field consisting only of nonalpine SWE; the difference between the two fields represents alpine-only SWE. Separating the alpine and nonalpine SWE in this simple manner is equivalent to evenly partitioning SWE based on the proportion of a given land type over grid cells that are not 100% alpine or 100% nonalpine. Separating alpine and nonalpine SWE was necessary first because the alpine mask was applied to GlobSnow but also in order to isolate the relative uncertainty/spread in the datasets over complex terrain (which poses unique challenges due to topographic variability) compared to nonalpine regions. Because of the alpine mask applied to GlobSnow, there is only nonalpine SWE available for this product.

We construct the dataset mean from the first five datasets listed in Table 1 by averaging the datasets available over a given grid cell accounting for land type as follows:
e1
where represents the SWE value for a particular product at the grid cell, f is the alpine fraction of the grid cell, N is the total number of data products, and refers to the GlobSnow product. This procedure reduces to an unweighted average of four products (all but GlobSnow) over strictly alpine grid cells and an unweighted average of all five products over strictly nonalpine cells.

We also perform a time series analysis over three landmasses: the entire Northern Hemisphere landmass (referred to as NH), the North American continent (NA), and the Eurasian continent (EU). We exclude Greenland from all three definitions. Where possible we further decompose the time series resulting from each of the landmasses into three mutually exclusive land types: midlatitudes (nonalpine land regions below 60°N), Arctic (nonalpine land regions above 60°N), and alpine. We determine the total snow water mass (SWM) by summing the equivalent volume of snow water per grid cell over the appropriate spatial domain and converting to mass using the density of pure water. We use these daily time series to calculate climatologies, anomalies, standard deviations, and linear trends. Henceforth, we reserve the use of SWM to describe SWE that has been spatially aggregated over a given region to form a time series. When discussing the spatially varying field, we will use the term SWE.

c. Diagnosis of temporal discontinuities

For the inclusion of a reanalysis-based dataset in this comparison over the 1981–2010 period, we require a relatively homogenous time series of NH SWM. An acceptable level of homogeneity is empirically diagnosed by comparing the sign and magnitude of trends of NH SWM. On this basis, we reject three established datasets (MERRA-Land, CMC, and GLDAS-1) for which we find spurious trends. Figure 1 shows how NH SWM trends from these three datasets compare with those from GlobSnow, GLDAS-2, MERRA, ERA-Interim/Land, and Crocus. While the five latter datasets exhibit a range of negative SWM trends (solid lines in Fig. 1), the trends of the three remaining datasets are exceedingly different (broken lines in Fig. 1). To illustrate how trends are indicative of inhomogeneities in global SWM, we show a time series of NH SWM for MERRA-Land in Fig. 2a. Except for the first two snow seasons (which have exceedingly large positive anomalies compared to the rest of the time series), anomalies are almost exclusively negative before 1998, with a negative trend, and positive after 1999, with a weak, positive trend. Examining separate time series for EU and NA (not shown) shows some covariability (12%) but that Eurasian variability dominates the total (64%) compared to NA (12%). Figure 2b shows a spatial map of the difference in climatologies between 1983 and 1998 and 1999 and 2010. The latter period has additional SWE over most of Eurasia. Comparing the 1981/82 average to the 1983–2010 climatology also shows additional SWE uniformly across Eurasia (not shown). Further analysis indicates the additional SWE is a result of discontinuities present in the precipitation data used to force the MERRA-Land model. Similar evidence of temporal discontinuities can be found in the other two datasets with spurious trends. In the case of CMC, changes in the spatial resolution of the precipitation forcing that was used to drive the analysis (specifically, increased resolution after 2006) result in a noticeable discontinuity in SWM anomalies at that time (not shown). The discontinuity produces the positive trend in SWM seen in Fig. 1 because the higher-resolution precipitation forcing better resolves deep alpine snow. In GLDAS-1 (trend shown in Fig. 1 for the Noah LSM only), a discontinuity occurs at the end of 1998 and results in a strongly negative winter season trend roughly 3–4 times larger than that found in any of the other datasets.

Fig. 1.
Fig. 1.

Trends in NH SWM for the five principal datasets over 1981–2010 (colors as marked). The average trend of the four reanalysis-derived datasets over alpine regions has been added to the trend of the GlobSnow product in order to compare trend magnitudes. Also shown are trends for CMC (1999–2010, dotted); GLDAS-1–Noah (1981–2010, dotted–dashed); and MERRA-Land (1983–98, long dash; 1999–2010, short dash).

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

Fig. 2.
Fig. 2.

(a) Time series of anomalous NH SWM from MERRA-Land. (b) Difference in climatological SWE between 1999 and 2010 and 1983 and 1998 periods for MERRA-Land.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

Comparing the five principal datasets of our analysis, Crocus, ERA-Interim/Land, and MERRA have more strongly negative NH SWM trends than GlobSnow, stemming from more strongly negative trends over Eurasia with comparable or more weakly negative trends over North America (continental breakdown not shown). GLDAS-2 shows much weaker trends through the entire snow season than the other four datasets. There is some consensus for decreasing NH winter SWE over the 1981–2010 period that is consistent with in situ trends to less snow cover in many regions of the NH (Vaughan et al. 2013).

3. Results

a. Climatological snow water equivalent

Total NH SWM climatologies are shown in Fig. 3a for the five SWE datasets and the multidataset mean. A key finding is the large amount of spread in climatological SWM among the products; during the seasonal peak, they vary by as much as 50%. To display total NH SWM for GlobSnow alongside the other datasets, we use the average of the non-GlobSnow datasets in alpine regions so that the additional SWM seen in the GlobSnow climatology stems from differences over nonalpine regions.

Fig. 3.
Fig. 3.

(a) Climatological total NH SWM for individual datasets and the multidataset mean for the 1981–2010 period. The average of the NH alpine SWM time series from the four reanalysis datasets is added to the GlobSnow time series in order to appropriately compare total NH SWM. (b) Stacked bar chart of SWM differences between individual datasets and the multidataset mean over pairs of months for various regions. Each individual dataset is colored as in (a) and for a given season, shows a sequence of three bars corresponding to NA, EU, and NH regions. Separate contributions from the three land types are stacked vertically using different shades for midlatitudes, Arctic, and alpine regions; the total difference summed over all land types on a given continent is shown by the bar with the black outline. Crocus and MERRA differences have been multiplied by three; GLDAS-2, ERA-Interim/Land, and GlobSnow differences are unaltered.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

To better determine which datasets and regions account for the spread among the individual climatologies, we decompose the differences between the individual datasets and the multidataset mean by continental domain and land type (Fig. 3b). Three seasonal periods are presented (calculated as 2-month averages) corresponding to before-peak (December and January), near-peak (February and March), and after-peak (April and May) SWM. GLDAS-2 shows the largest difference from the multidataset mean with less SWM over both continents during all month pairs and for all land types. The ERA-Interim/Land product has the largest positive bias relative to the multidataset mean, stemming primarily from differences with the other datasets over Eurasia (especially in the Arctic). SWE in GlobSnow peaks earlier in the season than the other datasets. The MERRA and Crocus datasets show the smallest differences from the mean, although for Crocus the similarity is in part because of opposite-signed differences in Arctic and alpine regions; Crocus has the most SWM of all the datasets in the alpine regions of both continents, but it has less SWM in the Eurasian Arctic than any of the other datasets except for GLDAS-2.

A close examination of Fig. 3b also shows that alpine and Arctic regions generally contribute the most to the differences in climatological SWM, which is consistent with our a priori assumptions that SWE is most poorly constrained in these regions. These regions have poor agreement among precipitation data, sparse observing networks, and complex snow processes that occur, including snow redistribution in the Arctic and complex elevational gradients in alpine regions, all of which will increase the uncertainty for the datasets. We show this result more explicitly in Fig. 4, which shows the total spread (range of dataset climatologies) in NH SWM according to land type. The spread over Arctic and alpine regions is comparable to one another but roughly 2–3 times larger than that over midlatitude regions.

Fig. 4.
Fig. 4.

Spread among climatologies for NH SWM by region over the 1981–2010 period.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

We also examine the spatial distribution of the multidataset mean and its spread in Fig. 5. The climatology compares well with known features of the observed SWE distribution (Brown and Mote 2009). Examining the ratio of climatological SWE to the spread among the component datasets provides a measure of the signal-to-noise ratio. Regions with a ratio greater than one coincide approximately with the boreal forest regions of North America and Eurasia. That we have the best agreement in the datasets over these regions is reasonable for several reasons. These are regions where snow cover has well-defined start and end points to the season and where midlatitude winter cyclones are well-represented in numerical weather prediction models. Farther south of this zone, differences in air temperature will have increasingly important impacts on accumulation through differences in snow on dates and rain/snow separation. Farther north of this zone uncertainty in the precipitation forcing increases. Furthermore, because the taiga snowpack is generally spatially homogeneous (Sturm et al. 1995), wind redistribution and topographic effects are minimal. The relative consistency of the datasets across the NH boreal forest is further supported by other measures of agreement among the datasets (discussed later; see Figs. 8 and 11).

Fig. 5.
Fig. 5.

(a) Climatology of multidataset mean SWE for February–March over 1981–2010 period. (b) Ratio of climatological SWE to spread among the five component datasets calculated for February–March over 1981–2010 period. The black contour delineates the 1:1 ratio.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

b. Snow water equivalent variability

We present time series of anomalous total NH nonalpine SWM in Fig. 6 (anomalies calculated separately for each dataset using its own climatology and individually detrended). Some years show good agreement in the sign and evolution of SWM anomalies over the winter (e.g., 1986/87, 1990–92, and 2001/02), while other years show larger spreads in the anomalies (1988/89, 1996/97, and 2009/10). Note the datasets evident as an outlier for a given time period varies among all five component datasets (i.e., GLDAS-2 in 1989, Crocus in 1993, GlobSnow in 1994, MERRA in 1999, and ERA-Interim/Land in 2003).

Fig. 6.
Fig. 6.

Anomalous total NH nonalpine SWM for the five individual datasets (individually detrended) and the multidataset mean for (top)–(bottom) 1981–1991, 1991–2001, and 2001–2011. Gray shading marks the range of anomalies (spread) among the datasets on a given day.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

To determine the strength of agreement between datasets, we present pairwise correlations between individually detrended SWM time series in Fig. 7. Each value is determined by correlating the detrended SWM anomaly time series for a given pair of datasets for all winter days [November–April (NDJFMA)] over the entire 1981–2010 period (such that each time series for a given pair contains 181 × 30 days). Calculated as such, correlations reflect both intraseasonal and interannual covariance. Slightly different approaches were also evaluated: 1) determine the intraseasonal correlation of winter season days (NDJFMA) for each year separately and average the 30 resulting correlations, and 2) calculate the interannual correlation for each calendar day and average the correlations over the winter season (NDJFMA). We find that the results are largely insensitive to which approach was used (correlations affected at most by 0.05), and the ranking of which datasets correlate best with one another shifts only slightly among pairs involving GLDAS-2.

Fig. 7.
Fig. 7.

Correlations of daily (NDJFMA) SWM time series (1981–2010) between pairs of datasets. Correlations are shown for detrended time series of NH (black), NA (red), and EU (blue) SWM as well as nondetrended NH (gray) SWM. Also shown are mean values of the SWE pattern correlation between dataset pairs (green circles), calculated daily and averaged over the NDJFMA season and 1981–2010 period.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

Because correlations involving GlobSnow only consider nonalpine snow, the datasets have been ranked by their correlations over nonalpine regions only, which differ slightly from those that consider all land types. Except for GLDAS-2 with MERRA and GLDAS-2 with Crocus, the difference in correlation between detrended and nondetrended time series is less than 0.05, and these differences in correlation stem from differences over Eurasia rather than North America (not shown). Examining separately the time series for each continent, we note that all datasets correlate better over NA than over EU. GLDAS-2 shows the weakest correlation with other datasets. GlobSnow has slightly higher correlations, and the remaining three reanalysis-based datasets, that is, MERRA, Crocus, and ERA-Interim/Land, are correlated with one another at ~0.8. An increased correlation among these three datasets is expected since MERRA and Crocus both use the same forcing meteorology and the ERA-Interim meteorology is itself well correlated with that of MERRA. This point is made in Rienecker et al. (2011) and further investigated in section 3c. Finally, we also show the mean SWE spatial pattern correlation for each dataset pair in Fig. 7 (calculated daily and averaged over all wintertime days of the 30-yr period). The mean pattern correlation is lower than the corresponding temporal correlation of total snow mass. This result may be due to the presence of opposite-signed spatial biases that cancel when spatially aggregated into a SWM time series.

Figure 8 presents a spatial map of the temporal correlation. For this figure, rather than spatially aggregating snow water before correlating, we calculate the temporal correlation between pairs of SWE datasets for each grid cell individually. Values plotted are the average among all possible pairs of datasets (for alpine regions, the correlations shown are the average of the 6 dataset pairs that exclude GlobSnow, and for nonalpine regions they are average of all 10 pairs). Interestingly, alpine regions do not exhibit noticeably weaker correlations than neighboring regions despite the comparatively large climatological spread indicated in Fig. 4. Correlations are lower in Arctic regions and the marginal snow zones and peak over the taiga, consistent with the high signal-to-noise present in the climatology (Fig. 5b).

Fig. 8.
Fig. 8.

Spatial map of the mean correlation between pairs of daily (NDJFMA) SWE time series (1981–2010). Correlations of detrended time series are performed for a given grid cell location between all available pairs of datasets and averaged together.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

Despite the reasonable spatial and temporal correlations shown in Figs. 7 and 8, the inter-dataset spread of SWM anomalies is comparable to the interannual variability. This result can be approximately assessed from an examination of Fig. 6 but is shown explicitly in Fig. 9a, where for each day over the 1981–2010 period, anomaly spread was calculated as the difference between the datasets with the maximum and minimum SWM anomaly. This comparability in magnitude means that on average at least one of the datasets fails to agree on the sign of the anomaly. When the same analysis is applied to nondetrended data, it is apparent that differences in trends among the datasets are not responsible for the majority of the spread.

Fig. 9.
Fig. 9.

(a) Spread in anomalous SWM among detrended datasets (black, solid), nondetrended datasets (black, dotted), and interannual variability of multidataset mean (gray, solid) over the 1981–2010 period. (b) Spread in anomalous SWM attributed to each of the five component dataset time series (1981–2010) and the multidataset mean. Total spread (black) is measured on left axis. Spread attributed to individual datasets is measured on the right axis (2 times smaller). A smoothing window of 30 days has been applied to all curves for clarity. See text for definitions of spread and attributed spread.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

We also quantify the relative contribution of each dataset to the total spread, defined as the attributed spread. To calculate this quantity for each grid cell on a given day and year, we partition the total spread between just two of the five datasets: those with the maximum and minimum anomaly on that day. The magnitude of spread attributed to each is the absolute difference between each dataset’s anomaly and the multidataset mean. This definition allows a particular dataset to accrue more of the total spread for the given day if it is farther from the mean value. The remaining datasets are attributed no spread for the given day but may be attributed spread on other days. This definition may be extended to a time series comprising any number of datasets as long as only two component datasets are attributed the full spread on a given day (as expected in the case with only two datasets, each would always receive half of the total spread). By definition, the sum of attributed spread across the total number of component datasets is equal to the total spread. Figure 9b shows the amount of spread attributed to each of the five component datasets averaged over the 1981–2010 period. To account for alpine regions in GlobSnow, the mean of the remaining four datasets was used to fill in alpine SWM in GlobSnow. This accounting means that spread is attributed to GlobSnow only when it stems from SWE differences in nonalpine regions.

Ranked in sequence, MERRA is attributed the least spread, followed by Crocus, ERA-Interim/Land, and GLDAS-2, while GlobSnow is attributed the most. While the first three datasets constitute roughly half of the total spread, GLDAS-2 and GlobSnow together are responsible for the remaining half, each contributing approximately one-quarter of the total. Note that ERA-Interim/Land’s attributed spread peaks during the spring and that GlobSnow has relatively low attributed spread during the spring but a large amount during the fall and early winter. We may understand these two results by considering that attributed spread convolves the frequency that a given dataset is an outlier with its distance from the mean anomaly. Figure 10a demonstrates that ERA-Interim/Land’s standard deviation (which will be related to its distance from the mean dataset anomaly) peaks during the spring, while that of GlobSnow peaks during the fall and early winter. Each of these two datasets are also more frequently outliers during these seasons (as opposed to Crocus, which shows a similar spring peak in interannual variability but is not a frequent outlier). The combination of these two traits means each dataset accrues more spread during the time of year that it shows increased variability. It is possible to account for these separate effects by standardizing the time series (i.e., by dividing each dataset’s time series by its interannual standard deviation for that calendar day) and examining the spread in the standardized time series (Fig. 10b). The spread now attributed to each dataset is more constant throughout the snow season (total spread of all datasets is around two standard deviations) and more closely reflects the proportion of days and years that the particular dataset is an outlier. Calculated as such, GlobSnow, Crocus, MERRA, and ERA-Interim/Land all contribute approximately equally to the total spread; by contrast, GLDAS-2 contributes approximately twice as much to the total spread.

Fig. 10.
Fig. 10.

(a) NH interannual variability of SWM for each dataset. (b) Attributed spread calculated from standardized SWM for each dataset. A smoothing window of 30 days has been applied to all curves for clarity.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

The spatial distribution of signal to noise for SWE anomalies (defined as the ratio of absolute deviation of the multidataset mean to the spread among the datasets) is shown in Fig. 11a. Consistent with the relative magnitudes of SWM anomalies and spread shown in Fig. 9a, the signal-to-noise ratio of the local SWE distribution is less than one nearly everywhere. It is larger in the boreal forest region than elsewhere, consistent with results for climatological SWE and for temporal correlations of daily SWE. The signal-to-noise ratio among only the MERRA, ERA-Interim/Land, and Crocus datasets is substantially larger, indicating better agreement among these three datasets (Fig. 11b). These datasets are referred to as group 1 datasets in the figure, for reasons described next.

Fig. 11.
Fig. 11.

(a) Ratio of absolute anomaly size for multidataset mean to dataset spread, calculated daily and averaged over February–March and 1981–2010 period. (b) As in (a), but only for the MERRA, ERA-Interim/Land, and Crocus datasets (group 1, see section 3c). The black contour delineates the 1:1 ratio.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

c. Relative influence of land model and meteorological forcing on reanalysis-based datasets

Finally, we examine the relative influence of differences in the precipitation forcing versus differences in the treatment of snow processes in the land model on the resulting correlation and spread of SWM time series. For this analysis we use two separate groups of reanalysis-based datasets. The first of these two groups (group 1) contains the MERRA, ERA-Interim/Land, and Crocus datasets. We consider these datasets to form a single group because they use meteorological forcing from either the MERRA or ERA-Interim datasets, which have been shown to be well-correlated with one another (Rienecker et al. 2011). The second group (group 2) contains four GLDAS-1 products. Each of these datasets has been forced using the same GDAS forcing data; however, each one uses a different land model as described in section 2 and Table 1. Because the forcing data for GLDAS-1 contain changes over the 1981–2010 period, we restrict our analysis to the 2001–10 period over which the forcing data are consistent.

Figure 12a shows that the spread in climatologies within each of the two groups is affected by the particular land models that are included. The spread shown in the gray shading (group 1 plus GLDAS-2) is comparable to that found among the group 2 products and remains comparable with the exclusion of the Noah-associated products from both groupings (but is decreased by about half). Note that both versions of GLDAS that use the Noah land surface model are outliers with similar (but especially low) SWM climatologies, despite using different meteorological forcing. This shows that it is possible to obtain similar climatologies from the same land surface model, while using different meteorological forcing. We also see that using the same meteorological forcing but different land models (group 2) results in a large climatological spread. These results imply that differences among the land models generate the majority of the climatological spread. We examine the effect on the correlation of the SWM time series in Fig. 12b. Datasets within each of the two groups correlate very well with one another over all continental domains and over all land types. However, the correlations of SWM time series between the two groups are substantially lower, especially over alpine and Arctic regions. This result suggests that differences in the meteorological forcing exert a larger influence on the resulting SWM correlation than differences in the details of the land model used to produce the data. We also note that agreement with the GlobSnow product (an independent estimate of SWM) is higher for group 1 datasets (circles, right column) than for group 2 datasets (diamonds, right column). Since the meteorological forcing data for group 1 are more recent, this may represent an improvement in the accuracy of the more recent reanalyses.

Fig. 12.
Fig. 12.

(a) Spread in total NH SWM for group 1 (MERRA, Crocus, and ERA-Interim/Land) plus GLDAS-2 (gray shading). Climatologies for the four GLDAS-1 products (group 2) are shown in the labeled curves. The single GLDAS-2 product is also shown for comparison. (b) Mean pairwise correlation among different groups of SWM time series (2001–10). For each of the three groupings, the average correlation is calculated for NA (red), EU (blue), and NH (black) SWM time series. For each landmass and group, four averages are shown in sequence for SWM summed over all land types, only midlatitude, only the Arctic, and only alpine regions. Mean correlations of group 1 (circles) and group 2 (diamonds) total NH SWM time series (all land types) with GlobSnow are also shown. Correlations with GlobSnow over individual land types are omitted for clarity.

Citation: Journal of Climate 28, 20; 10.1175/JCLI-D-15-0229.1

4. Conclusions

We have presented a comparison of five, daily, gridded, Northern Hemisphere SWE datasets over the 1981–2010 period. Our intent is to make available the multidataset mean and corresponding spread of both the climatology and anomalies as part of the National Center for Atmospheric Research (NCAR) Climate Data Guide (https://www.climatedataguide.ucar.edu/). Our analysis has shown that the individual datasets exhibit a large amount of spread in their total snow water mass (SWM) climatologies (Figs. 3, 4) as well as their anomalies (Fig. 9a). Despite the large spread, the SWM time series show moderate to good correlations with one another, approximately 0.85 for the three datasets using modern reanalysis meteorological forcings; these correlations are higher over North America than over Eurasia (Fig. 7). Boreal regions not only have the lowest relative amount of spread (highest signal to noise) for both climatological SWE (Fig. 5) and SWE anomalies (Fig. 11), but these regions also have the highest temporal correlation among the SWE datasets (Fig. 8). We have also examined the relative influence of the particular land surface model compared to the choice of meteorological forcing using a suite of reanalysis-derived datasets. The former accounts for the majority of the spread in the climatologies, while the latter exerts a larger influence on the resulting SWM correlation (Fig. 12). More modern reanalysis-derived datasets also show improved correlation with the satellite product, GlobSnow, compared to the previous generation (Fig. 12).

This analysis of gridded SWE data has yielded important insights on the amount of spread and the strength of spatial and temporal correlation between independent SWE datasets. At this point, the relatively good agreement between some groups of datasets (i.e., modern reanalysis-derived datasets such as MERRA, ERA-Interim/Land, and Crocus) does not imply lower bias compared to the ground truth, only that these datasets are generally consistent with each other. The agreement between these three datasets is not surprising given the commonalities in how SWE is derived: modern reanalysis meteorology and high-quality precipitation forcing coupled with state of the art land surface models. A product like GlobSnow adopts an entirely different approach by blending in situ snow depth observations and satellite passive microwave measurements through the use of a microwave snow emission model. Going forward the challenge to the community is how to combine these unique perspectives with their respective strengths and sources of error.

Given the spread in climatology, these results highlight the sensitivity of climate analysis to the selection of an individual SWE dataset for model evaluation from the available pool of data. For example, evaluation of simulated SWE with ERA-Interim/Land (highest premelt SWE) versus GLDAS (lowest premelt SWE) could lead to entirely different interpretations of model performance and bias. There are also seasonal and regional differences that are also important for users to consider. For example, ERA-Interim/Land has very different biases over NA compared to EU and Crocus has different biases for Arctic versus alpine snow, while GlobSnow exhibits different seasonality than the multidataset mean.

For applications like climate model evaluation, it is straightforward to see the value in using an ensemble of SWE datasets for the evaluation of a multimodel ensemble of simulations. This approach would illustrate the overlap (or lack thereof) in simulated climate realizations against the uncertainties in characterizing our current climate, which is evident when a single dataset is used to characterize reality (for instance, Derksen and Brown 2012). For land surface data assimilation applications, an ensemble approach could be used to statistically characterize observational uncertainty, an important requirement for the assimilation of the model first guess with observations.

What remains to be determined is how to select the SWE datasets for inclusion in an observational ensemble. Thresholds based on attributed spread or detrended anomaly correlations could be the basis for selecting a subgroup of datasets, but this must be done with caution. In the absence of an evaluation with representative ground measurements, agreement between a subset of datasets does not necessarily represent better overall accuracy in representing reality. Comparisons with independent reference measurements in order to determine the dataset bias relative to ground truth are ultimately necessary and currently underway within the European Space Agency’s Satellite Snow Product Intercomparison and Evaluation Experiment (SnowPEX). In advance of a comparison with high-quality in situ reference datasets, quantifying the spread between different available products, as was accomplished in this study, is an important step in informing users of the level of spatial and temporal agreement between products and the relationship of individual datasets to the multidataset mean.

Acknowledgments

We acknowledge funding from the Natural Sciences and Engineering Research Council of Canada’s Climate Change and Atmospheric Research initiative via the Canadian Sea Ice and Snow Evolution Network. We also thank Eric Brun for providing data from the Crocus snowpack model for this analysis. CMC SWE data were provided by author Ross Brown. The remaining datasets were provided by the NASA Goddard Earth Sciences Data and Information Services Center (MERRA, MERRA-Land and GLDAS), the ECMWF Public Datasets web interface (ERA-Interim/Land), and the Finnish Meteorological Institute’s FTP server (GlobSnow).

REFERENCES

  • Balsamo, G., , A. Beljaars, , and K. Scipal, 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, 623643, doi:10.1175/2008JHM1068.1.

    • Search Google Scholar
    • Export Citation
  • Balsamo, G., and Coauthors, 2015: ERA-Interim/Land: A global land surfaces reanalysis data set. Hydrol. Earth Syst. Sci. Discuss., 19, 389407, doi:10.5194/hess-19-389-2015.

    • Search Google Scholar
    • Export Citation
  • Bonan, G. B., , K. W. Oleson, , M. Vertenstein, , S. Levis, , X. Zeng, , Y. Dai, , R. E. Dickinson, , and Z.-L. Yang, 2002: The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model. J. Climate, 15, 31233149, doi:10.1175/1520-0442(2002)015<3123:TLSCOT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Brasnett, B., 1999: A global analysis of snow depth for numerical weather prediction. J. Appl. Meteor., 38, 726740, doi:10.1175/1520-0450(1999)038<0726:AGAOSD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., , and P. W. Mote, 2009: The response of Northern Hemisphere snow cover to a changing climate. J. Climate, 22, 2124–2145, doi:10.1175/2008JCLI2665.1.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., , and B. Brasnett, 2010: Canadian Meteorological Centre (CMC) Daily Snow Depth Analysis Data. National Snow and Ice Data Center Distributed Active Archive Center, Boulder, CO, doi:10.5067/W9FOYWH0EQZ3.

  • Brown, R. D., , and C. Derksen, 2013: Is Eurasian October snow cover extent increasing? Environ. Res. Lett., 8, 024006, doi:10.1088/1748-9326/8/2/024006.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., , C. Derksen, , and L. Wang, 2010: A multi-data set analysis of variability and change in Arctic spring snow cover extent, 1967–2008. J. Geophys. Res., 115, D16111, doi:10.1029/2010JD013975.

    • Search Google Scholar
    • Export Citation
  • Brun, E., , P. David, , M. Sudul, , and G. Brunot, 1992: A numerical-model to simulate snow-cover stratigraphy for operational avalanche forecasting. J. Glaciol., 38, 1322.

    • Search Google Scholar
    • Export Citation
  • Brun, E., , V. Vionnet, , A. Boone, , B. Decharme, , Y. Peings, , R. Valette, , F. Karbou, , and S. Morin, 2013: Simulation of northern Eurasian local snow depth, mass, and density using a detailed snowpack model and meteorological reanalyses. J. Hydrometeor., 14, 203219, doi:10.1175/JHM-D-12-012.1.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and Coauthors, 1996: Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101, 72517268, doi:10.1029/95JD02165.

    • Search Google Scholar
    • Export Citation
  • De Lannoy, G. J. M., , R. H. Reichle, , P. R. Houser, , K. R. Arsenault, , N. E. C. Verhoest, , and V. R. N. Pauwels, 2010: Satellite-scale snow water equivalent assimilation into a high-resolution land surface model. J. Hydrometeor., 11, 352369, doi:10.1175/2009JHM1192.1.

    • Search Google Scholar
    • Export Citation
  • Derksen, C., , and R. Brown, 2012: Spring snow cover extent reductions in the 2008–2012 period exceeding climate model projections. Geophys. Res. Lett., 39, L19504, doi:10.1029/2012GL053387.

    • Search Google Scholar
    • Export Citation
  • Drewitt, G., , A. A. Berg, , W. J. Merryfield, , and W. Lee, 2012: Effect of realistic soil moisture initialization on the Canadian CanCM3 seasonal forecast model. Atmos.–Ocean, 50, 466474, doi:10.1080/07055900.2012.722910.

    • Search Google Scholar
    • Export Citation
  • Dutra, E., , G. Balsamo, , P. Viterbo, , P. M. A. Miranda, , A. Beljaars, , C. Schär, , and K. Elder, 2010: An improved snow scheme for the ECMWF land surface model: Description and offline validation. J. Hydrometeor., 11, 899916, doi:10.1175/2010JHM1249.1.

    • Search Google Scholar
    • Export Citation
  • Dutra, E., , P. Viterbo, , P. M. A. Miranda, , and G. Balsamo, 2012: Complexity of snow schemes in a climate model and its impact on surface energy and hydrology. J. Hydrometeor., 13, 521538, doi:10.1175/JHM-D-11-072.1.

    • Search Google Scholar
    • Export Citation
  • Jeong, J., , H. W. Linderholm, , S. Woo, , C. Folland, , B. Kim, , S. Kim, , and D. Chen, 2013: Impacts of snow initialization on subseasonal forecasts of surface air temperature for the cold season. J. Climate, 26, 19561972, doi:10.1175/JCLI-D-12-00159.1.

    • Search Google Scholar
    • Export Citation
  • Koren, V., , J. Schaake, , K. Mitchell, , Q.-Y. Duan, , F. Chen, , and J. M. Baker, 1999: A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys. Res., 104, 19 56919 585, doi:10.1029/1999JD900232.

    • Search Google Scholar
    • Export Citation
  • Koster, R. D., , and M. J. Suarez, 1994: The components of a ‘SVAT’ scheme and their effects on a GCM’s hydrological cycle. Adv. Water Resour., 17, 6178, doi:10.1016/0309-1708(94)90024-8.

    • Search Google Scholar
    • Export Citation
  • Koster, R. D., , M. J. Suarez, , A. Ducharne, , M. Stieglitz, , and P. Kumar, 2000: A catchment-based approach to modeling land surface processes in a general circulation model: 1. Model structure. J. Geophys. Res., 105, 24 80924 822, doi:10.1029/2000JD900327.

    • Search Google Scholar
    • Export Citation
  • Koster, R. D., and Coauthors, 2010: Contribution of land surface initialization to subseasonal forecast skill: First results from a multi-model experiment. Geophys. Res. Lett., 37, L02402, doi:10.1029/2009GL041677.

    • Search Google Scholar
    • Export Citation
  • Langlois, A., and Coauthors, 2009: Simulation of snow water equivalent (SWE) using thermodynamic snow models in Québec, Canada. J. Hydrometeor., 10, 14471463, doi:10.1175/2009JHM1154.1.

    • Search Google Scholar
    • Export Citation
  • Liang, X., , D. P. Lettenmaier, , E. F. Wood, , and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 41514 428, doi:10.1029/94JD00483.

    • Search Google Scholar
    • Export Citation
  • Mudryk, L. R., , P. J. Kushner, , and C. Derksen, 2014: Interpreting observed Northern Hemisphere snow trends with large ensembles of climate simulations. Climate Dyn., 43, 345359, doi:10.1007/s00382-013-1954-y.

    • Search Google Scholar
    • Export Citation
  • NOAA National Geophysical Data Center, 1988: Data Announcement 88-MGG-02, Digital relief of the Surface of the Earth. NOAA, National Geophysical Data Center, Boulder, CO. [Available online at http://www.ngdc.noaa.gov/mgg/global/etopo5.html.]

  • Orsolini, Y. J., , R. Senan, , G. Balsamo, , F. J. Doblas-Reyes, , F. Vitart, , A. Weisheimer, , A. Carrasco, , and R. E. Benestad, 2013: Impact of snow initialization on sub-seasonal forecasts. Climate Dyn., 41, 19691982, doi:10.1007/s00382-013-1782-0.

    • Search Google Scholar
    • Export Citation
  • Reichle, R. H., , R. D. Koster, , G. J. M. de Lannoy, , B. A. Forman, , Q. Liu, , S. P. P. Mahanama, , and A. Touré, 2011: Assessment and enhancement of MERRA land surface hydrology estimates. J. Climate, 24, 63226338, doi:10.1175/JCLI-D-10-05033.1.

    • Search Google Scholar
    • Export Citation
  • Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 36243648, doi:10.1175/JCLI-D-11-00015.1.

    • Search Google Scholar
    • Export Citation
  • Rodell, M., , and H. K. Beaudoing, 2007a: GLDAS CLM Land Surface Model L4 3 Hourly 1.0 × 1.0 degree subsetted, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 5 August 2014, doi:10.5067/83NO2QDLG6M0.

  • Rodell, M., , and H. K. Beaudoing, 2007b: GLDAS Mosaic Land Surface Model L4 3 Hourly 1.0 x 1.0 degree subsetted, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 5 August 2014, doi:10.5067/DLVU8VOPKN7L.

  • Rodell, M., , and H. K. Beaudoing, 2007c: GLDAS Noah Land Surface Model L4 3 Hourly 1.0 × 1.0 degree subsetted, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 10 January 2014, doi:10.5067/7J0CPW2PF9N9.

  • Rodell, M., , and H. K. Beaudoing, 2007d: GLDAS VIC Land Surface Model L4 3 Hourly 1.0 × 1.0 degree, version 001. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 5 August 2014, doi:10.5067/Z7SA0KYUTBQS.

  • Rodell, M., , and H. K. Beaudoing, 2013: GLDAS Noah Land Surface Model L4 3 Hourly 1.0 × 1.0 degree, version 2.0. Goddard Earth Sciences Data and Information Services Center (GES DISC), NASA/GSFC/HSL, accessed 23 January 2014. [Available online at http://disc.sci.gsfc.nasa.gov/datacollection/GLDAS_NOAH10_3H_V020.shtml.]

  • Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381394, doi:10.1175/BAMS-85-3-381.

    • Search Google Scholar
    • Export Citation
  • Rutter, N., and Coauthors, 2009: Evaluation of forest snow processes models (SnowMIP2). J. Geophys. Res., 114, D06111, doi:10.1029/2008JD011063.

  • Sheffield, J., , G. Goteti, , and E. F. Wood, 2006: Development of a 50-year high-resolution global dataset of meteorological forcings for land surface modeling. J. Climate, 19, 3088–3111, doi:10.1175/JCLI3790.1.

    • Search Google Scholar
    • Export Citation
  • Stieglitz, M., , A. Ducharne, , K. Koster, , and M. Suarez, 2001: The impact of detailed snow physics on the simulation of snow cover and subsurface thermodynamics at continental scales. J. Hydrometeor., 2, 228242, doi:10.1175/1525-7541(2001)002<0228:TIODSP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sturm, M., , J. Holmgren, , and G. E. Liston, 1995: A seasonal snow cover classification system for local to global applications. J. Climate, 8, 12611283, doi:10.1175/1520-0442(1995)008<1261:ASSCCS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Takala, M., , K. Luojus, , J. Pulliainen, , C. Derksen, , L. Lemmetyinen, , J.-P. Kärnä, , and J. Koskinen, 2011: Estimating Northern Hemisphere snow water equivalent for climate research through assimilation of space-borne radiometer data and ground-based measurements. Remote Sens. Environ., 115, 3517–3529, doi:10.1016/j.rse.2011.08.014.

    • Search Google Scholar
    • Export Citation
  • Tong, J., , S. J. Déry, , P. L. Jackson, , and C. Derksen, 2010: Testing snow water equivalent retrieval algorithms for passive microwave remote sensing in an alpine watershed of western Canada. Can. J. Remote Sens., 36, S74S86, doi:10.5589/m10-009.

    • Search Google Scholar
    • Export Citation
  • Vaughan, D. G., and Coauthors, 2013: Observations: Cryosphere. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 317–382.

  • Wang, Z., , X. Zeng, , and M. Decker, 2010: Improving snow processes in the Noah land model. J. Geophys. Res., 115, D20108, doi:10.1029/2009JD013761.

    • Search Google Scholar
    • Export Citation
Save