• Armstrong, R. L., , and M. J. Brodzik, 2001: Recent Northern Hemisphere snow extent: A comparison of data derived from visible and microwave satellite sensors. Geophys. Res. Lett., 28, 36733676.

    • Search Google Scholar
    • Export Citation
  • Bitner, D., , T. Carroll, , D. Cline, , and P. Romanov, 2002: An assessment of the differences between three satellite snow cover mapping techniques. Hydrol. Processes, 16, 37233733.

    • Search Google Scholar
    • Export Citation
  • Blyth, E., and Coauthors, 2006: JULES: A new community land surface model. Global Change Newsletter, No. 66, International Geosphere–Biosphere Programme, The Royal Swedish Academy of Sciences, Stockholm, Sweden, 9–11. [Available online at http://www.igbp.net/documents/NL_66-3.pdf.]

    • Search Google Scholar
    • Export Citation
  • Brasnett, B., 1999: A global analysis of snow depth for numerical weather prediction. J. Appl. Meteor., 38, 726740.

  • Brubaker, K. L., , R. T. Pinker, , and E. Deviatova, 2005: Evaluation and comparison of MODIS and IMS snow-cover estimates for the continental United States using station data. J. Hydrometeor., 6, 10021017.

    • Search Google Scholar
    • Export Citation
  • Cameron, J., 2003: Comparison of Unified Model snow and sea-ice fields with NESDIS and USAF data sets. Forecasting Research Tech. Rep. 416, Met Office, 40 pp.

    • Search Google Scholar
    • Export Citation
  • Davies, T., , M. J. P. Cullen, , A. J. Malcolm, , M. H. Mawson, , A. Staniforth, , A. A. White, , and N. Wood, 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 17591782.

    • Search Google Scholar
    • Export Citation
  • Drusch, M., , D. Vasiljevic, , and P. Viterbo, 2004: ECMWF’s global snow analysis: Assessment and revision based on satellite observations. J. Appl. Meteor., 43, 12821294.

    • Search Google Scholar
    • Export Citation
  • Essery, R., , E. Martin, , H. Douville, , A. Fernández, , and E. Brun, 1999: A comparison of four snow models using observations from an alpine site. Climate Dyn., 15, 583593.

    • Search Google Scholar
    • Export Citation
  • Hall, D. K., 1988: Assessment of polar climate change using satellite technology. Rev. Geophys., 26, 2639.

  • Hall, D. K., , G. A. Riggs, , J. L. Foster, , and S. V. Kumar, 2010: Development and evaluation of a cloud-gap-filled MODIS daily snow-cover product. Remote Sens. Environ., 114, 496503, doi:10.1016/j.rse.2009.10.007.

    • Search Google Scholar
    • Export Citation
  • Helfrich, S. R., , D. McNamara, , B. H. Ramsay, , T. Baldwin, , and T. Kasheta, 2007: Enhancements to, and forthcoming developments in the Interactive Multisensor Snow and Ice Mapping System (IMS). Hydrol. Processes, 21, 15761586.

    • Search Google Scholar
    • Export Citation
  • Maurer, E. P., , J. D. Rhoads, , R. O. Dubayah, , and D. P. Lettenmaier, 2003: Evaluation of the snow-covered area data product from MODIS. Hydrol. Processes, 17, 5971.

    • Search Google Scholar
    • Export Citation
  • Nolin, A. W., , and S. Liang, 2000: Progress in bidirectional reflectance modelling and applications for surface particulate media: Snow and soils. Remote Sens. Rev., 18, 307342.

    • Search Google Scholar
    • Export Citation
  • Ramsay, B. H., 1998: The Interactive Multisensor Snow and Ice Mapping System. Hydrol. Processes, 12, 15371546.

  • Rodell, M., , and P. R. Houser, 2004: Updating a land surface model with MODIS-derived snow cover. J. Hydrometeor., 5, 10641075.

  • Romanov, P., , G. Gutman, , and I. Csiszar, 2000: Automated monitoring of snow cover over North America with multispectral satellite data. J. Appl. Meteor., 39, 18661880.

    • Search Google Scholar
    • Export Citation
  • Romanov, P., , D. Tarpley, , G. Gutman, , and T. Carroll, 2003: Mapping and monitoring of the snow cover fraction over North America. J. Geophys. Res., 108, 8619, doi:10.1029/2002JD003142.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    IMS NH snow and ice chart for 15 Dec 2009, where snow cover is shown in white and sea ice is shown in yellow (data are available online at http://www.natice.noaa.gov/ims/).

  • View in gallery

    Relationship between snow areal density and fractional cover for a masking depth of 0.2 m2 kg−1.

  • View in gallery

    Schematic diagram showing how the snow analysis works, where fc = fractional cover, Sb = background snow amount, and Sa = analyzed snow amount.

  • View in gallery

    A snow analysis performed on 10 Nov 2008 over Europe, showing (a) the observed fractional cover, (b) the background snow amount field, (c) the analyzed snow amount field, and (d) the snow amount increments.

  • View in gallery

    Average snow cover extents, in terms of the frequency of snow cover throughout the month, for the months marked, based on Rutgers Global Snow Laboratory analysis of IMS daily snow maps (data are available online at http://climate.rutgers.edu/snowcover/).

  • View in gallery

    Time series of number of grid points for which snow was present in both background and analysis for (a) December 2006 and (b) March–May 2007. The solid line represents the background, and the dashed line shows the analysis.

  • View in gallery

    Number of snow (a) additions and (b) removals made to each grid box throughout December 2006.

  • View in gallery

    As in Fig. 7, but for April 2007.

  • View in gallery

    (a) Analysis increments and (b) model background snow amount for 26 Apr 2007, from the spring experiment.

  • View in gallery

    (a) Analyzed snow amounts, (b) NOHRSC snow depth (data are available online at http://www.nohrsc.noaa.gov/nsa), and (c) background snow amounts for (top) 13 and (middle) 24 Dec 2006 and (bottom) 14 Apr 2007 over the United States.

  • View in gallery

    Positions of SYNOP stations in (a) Europe and (b) North America for which reports were received between 0500 and 0700 UTC 7 Dec 2006. Marked in red are those stations that have reported on “state of ground.”

  • View in gallery

    Percentage of model grid points, in Europe, with snow presence in agreement with SYNOP snow reports within them for (a) 1–14 Dec 2006 and (b) 10–23 Apr 2007. The control run, without snow analysis, is shown by the solid line, and the experiment run is shown by the dashed line.

  • View in gallery

    As in Fig. 10, but for North America on 30 Nov 2006.

  • View in gallery

    (a) Schematic diagram showing how the improved snow removal method works, where fc = fractional cover, Sb = background snow amount, and Sa = analyzed snow amount. (b) Analyzed snow amounts for 30 Nov 2006, using the improved method.

  • View in gallery

    Impacts on surface variables during the December experiment. Forecast mean (upper panel of group) and RMS (lower panel of group) error verified against observations for surface RH over (a) Europe and (b) North America and verified against analysis for (c) snow depth over Europe. The experiment run (with snow analysis) is shown by a dashed line, and the control run (with no snow analysis) is shown by a solid line.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 67 67 12
PDF Downloads 52 52 9

Using Satellite-Derived Snow Cover Data to Implement a Snow Analysis in the Met Office Global NWP Model

View More View Less
  • 1 Met Office, Exeter, United Kingdom
© Get Permissions
Full access

Abstract

Snow cover and amount are important components in the interaction between the land surface and atmosphere, but until recently there has been no observational snow information incorporated into the Met Office global numerical weather prediction (NWP) model. This paper presents a Northern Hemisphere (NH) snow analysis, based on satellite-derived observations of snow cover, that has recently been implemented in the operational global Unified Model (UM). The analysis uses daily snow maps from the NOAA National Environmental Satellite Data and Information Service Interactive Multisensor Snow and Ice Mapping System (IMS) to modify the UM background snow amount. Assimilation experiments carried out during the NH snow accumulation and ablation periods show improvements in analyzed snow cover, both qualitatively and quantitatively, when compared with unmodified control runs. Although the effect on forecast accuracy of introducing the snow analysis is largely neutral, there is some evidence of small improvements in screen-level temperature and humidity forecasts. Retention of information introduced by the snow analysis is hard, and so the limited impact on forecast accuracy is not unexpected. Several methods of validation and verification are presented, and their results and implications are discussed. The NH snow analysis delivers a basic snow assimilation system that can be expanded and developed further in future iterations to improve the representation of snow cover and amount in global and regional forecasting models.

Corresponding author address: Samantha Pullen, Met Office, FitzRoy Rd., Exeter EX1 3PB, United Kingdom. E-mail: samantha.pullen@metoffice.gov.uk

Abstract

Snow cover and amount are important components in the interaction between the land surface and atmosphere, but until recently there has been no observational snow information incorporated into the Met Office global numerical weather prediction (NWP) model. This paper presents a Northern Hemisphere (NH) snow analysis, based on satellite-derived observations of snow cover, that has recently been implemented in the operational global Unified Model (UM). The analysis uses daily snow maps from the NOAA National Environmental Satellite Data and Information Service Interactive Multisensor Snow and Ice Mapping System (IMS) to modify the UM background snow amount. Assimilation experiments carried out during the NH snow accumulation and ablation periods show improvements in analyzed snow cover, both qualitatively and quantitatively, when compared with unmodified control runs. Although the effect on forecast accuracy of introducing the snow analysis is largely neutral, there is some evidence of small improvements in screen-level temperature and humidity forecasts. Retention of information introduced by the snow analysis is hard, and so the limited impact on forecast accuracy is not unexpected. Several methods of validation and verification are presented, and their results and implications are discussed. The NH snow analysis delivers a basic snow assimilation system that can be expanded and developed further in future iterations to improve the representation of snow cover and amount in global and regional forecasting models.

Corresponding author address: Samantha Pullen, Met Office, FitzRoy Rd., Exeter EX1 3PB, United Kingdom. E-mail: samantha.pullen@metoffice.gov.uk

1. Introduction

Snow is an extremely important component of the land surface system, substantially affecting the radiative and hydrological properties of the surface and consequently the way it interacts with the atmosphere. Most important is that snow cover dramatically increases the land surface albedo from between 0.05 and 0.4 (typical for bare soil and vegetation) to up to 0.9 for pure snow (Nolin and Liang 2000), which has a huge effect on diabatic heating. In hydrological terms, water is accumulated in the snowpack during the winter season and then released during snowmelt, with important effects on variables such as soil moisture and surface runoff. There are also effects on the near-surface air temperatures due to the insulating properties of snow and the latent heat needed for snowmelt. About 98% of global seasonal snow cover can be found in the Northern Hemisphere (NH), where it affects between 7% and 40% of the land surface during the annual cycle (Armstrong and Brodzik 2001; Hall 1988), so it is clear that accurate representation of snow cover in numerical weather prediction (NWP) models is essential for calculations of surface exchange fluxes and subsequent forecasts of atmospheric variables. An accurate knowledge of surface emissivities, which are affected by snow cover, is also important to enable the assimilation of satellite sounding data from surface-affected channels. Increases in the number of usable sounding channels could potentially yield considerable improvements to forecast accuracy, and so efforts to improve the surface representation are also important from the point of view of satellite data assimilation.

Until 2008 no observational snow information was used in the Met Office global NWP model. The surface snow variable in the Unified Model (UM; Davies et al. 2005) is snow depth in units of kilograms per meter squared. This is a prognostic variable in the model but is internal to the UM, generated by snow precipitation. This work presents a snow analysis developed to assimilate satellite-derived observations of NH snow cover into the global UM. The primary aim is to improve the global model snow analysis, but improvements to analyzed and forecast screen-level temperatures and humidity are also a possible benefit of this work.

Section 2 describes the observational data that have been used and their validation and use by other centers. Section 3 discusses the options for incorporating these data into the UM and presents the development and implementation of the analysis scheme chosen. The assimilation experiments carried out and their validation and verification are discussed in section 4, with a summary in section 5.

2. Observational data

Observations of snow cover can be retrieved from satellite, with global coverage and high temporal and spatial resolution. At this time the most widely used satellite data–derived snow cover product for large-scale NWP is the Interactive Multisensor Snow and Ice Mapping System (IMS) from the National Oceanic and Atmospheric Administration (NOAA) National Environmental Satellite Data and Information Service (NESDIS) (Ramsay 1998). Although only currently available operationally for the NH, it is freely available with timeliness appropriate for use in NWP.

The IMS data consist of a daily map of NH snow cover and sea ice extent, which is drawn up by analysts on workstations that display data products and satellite imagery from a variety of sources, using the map from the previous day as the initial state. The primary data sources are visible imagery from polar-orbiting and geostationary satellites, these being NOAA’s Polar Orbiting Environmental Satellite (POES) and Geostationary Operational Environmental Satellite (GOES), European geostationary (Meteosat) and polar-orbiting (MetOp) satellites, the Moderate Resolution Imaging Spectroradiometer (MODIS), and the Japanese Multifunctional Transport Satellite (MTSAT) series. In addition, ground weather observations and microwave products from the Advanced Microwave Sounding Unit (AMSU) and Special Sensor Microwave Imager (SSM/I) are also incorporated, allowing detection in cloudy conditions or in conditions of low solar illumination. The analyst also has access to a weekly sea ice analysis from the National Ice Center, the U.S. Air Force Snow and Ice Analysis Product, and snow products from the National Operational Hydrologic Remote Sensing Center, as well as several automated snow detection tools developed by NESDIS and the National Centers for Environmental Prediction (NCEP) (Helfrich et al. 2007). The IMS data are produced on a 6144 × 6144 grid at approximately 4-km resolution. They are in polar stereographic projection, with the central meridian at 80°W and the standard parallel at 60°N. Figure 1 shows the IMS snow and ice map for 15 December 2009.

Fig. 1.
Fig. 1.

IMS NH snow and ice chart for 15 Dec 2009, where snow cover is shown in white and sea ice is shown in yellow (data are available online at http://www.natice.noaa.gov/ims/).

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

A number of intercomparisons and validation studies have been done using IMS snow cover data. Romanov et al. (2000) made a quantitative comparison of snow cover products from IMS, SSM/I, and a blended GOES and SSM/I automated product. They found an average difference in derived snow fraction between IMS and the blended product of only 3%. However, the SSM/I-only product showed significant deficiencies over forested areas and melting snow. Brubaker et al. (2005) compared IMS with both ground station and MODIS snow cover data. They found that IMS performed well against both ground measurements and MODIS data. Although snow detection rates are low (less than 40%) at the start of the winter, they increase as the snow accumulates, to 95% in December, against MODIS data. Detection rates are lower during the snowmelt but are better than those during snow accumulation. “No snow” detection rates are very high during snow accumulation and ablation (near 100%) but are slightly lower at the peak of the winter.

IMS was originally created for use in NCEP forecasting models as an initial state of the surface cryosphere and has been used as such for a number of years (Romanov et al. 2003). It has also been adopted by the European Centre for Medium-Range Weather Forecasts (ECMWF) as an additional constraint in their snow analysis system (Drusch et al. 2004) and serves an important function as a long-term climate record, with 40 years of continuous snow cover data. At the Met Office, IMS has been considered for use as input data to a snow analysis for some time. Preliminary investigations into the suitability of the data were performed by Cameron (2003). Cameron compared 20 months of IMS (24-km resolution) snow cover maps with UM snow cover. He found that snow cover formed slightly too quickly in the UM at the onset of winter and melted far too rapidly in spring when compared with the IMS data. His comparisons also revealed a tendency for the UM to overestimate the temporal variability in the snow cover. Over the two winter seasons analyzed, the UM tended to underestimate snow cover in the United States, Europe, and north of the Caspian Sea but overestimate snow cover in parts of the Far East, particularly in the 2002/03 winter. The prominent early springtime melt of UM snow is mainly attributable to deficiencies in the model’s snow physics scheme. This represents snow as a single layer of uniform snow cover within the model grid box, which melts much more quickly than a more realistic pattern of patchy snow cover of varying depths and densities. Other more localized anomalies observed may be due to a combination of model temperature and precipitation biases and to orographic effects.

In comparisons of IMS snow cover with ECMWF operational snow water equivalent (SWE) analyses, prior to their assimilation of IMS, Drusch et al. (2004) noted that the ECMWF model snow was also melted too rapidly in the spring. They found that the operational ECMWF analysis overestimated snow cover systematically, with the most pronounced differences at the snow edge. They also found systematic overestimation of snow extent over the Tibetan Plateau.

3. Development of a snow analysis

a. Snow analysis options

While the IMS data provide a simple binary diagnosis of snow cover (fully covered or no snow), the UM snow variable is snow amount in kilograms per meter squared, or areal density. The presence of snow cover indicates that some snow amount exists but gives no information about how much, and so there is no continuous relationship between the model states and the observations. The methods of assimilation available are therefore limited and unsophisticated, as explained by Rodell and Houser (2004), and simple update methods are commonly adopted. The presence of snow can be compared in model and observations, and, where discrepancies arise, a set of rules can be followed to determine how to modify the model snow field.

It is easy to remove model snow to agree with zero snow cover conditions in the IMS data, but addition of snow to the model field is harder because the IMS data contain no information on nonzero snow amounts. In the ideal situation, amounts added should have a minimal impact on the local hydrology balance but be large enough to give snow residence times that are sufficiently long to affect the albedo. From reports by others on using snow cover observations to update model snow depths or snow amounts, it would appear that an SWE of 10 mm, which is equivalent to an areal density of 10 kg m−2, is accepted as a sensible quantity to add.1 At ECMWF, 100-mm snow depth is added to snow-free model points for which 100% snow cover is obtained from IMS data (Drusch et al. 2004). Rodell and Houser (2004) tested a snow analysis using MODIS snow cover data to update the “Mosaic” model, for use in the Global Land Data Assimilation System (GLDAS). Where the model SWE was zero but the MODIS data indicated greater than 40% snow cover, they added 5-mm SWE. However, they found that this thin layer was often melted immediately by the model.

Experiments were performed to determine the effects of adding different quantities of snow to the UM. The UM snow amount was updated using IMS snow cover data, and the model was then run for 2 days, for a summer and winter case. Screen-temperature error statistics were calculated, using surface synoptic (SYNOP) observations as ground truth. Snow was removed from the UM where the IMS data denoted no snow. Where IMS data denoted snow but model snow amount was zero, initially (called Exp1) between 10 and 50 kg m−2 was added, depending on the mean snow amounts in the latitude band in question. Mean statistics were calculated separately for Canada and Siberia for the summer case (June 2003) and for the United States and Eurasia for the winter case (February 2004). For each region, the RMS error and bias in screen temperature increased for cases in which snow had been added, relative to an unmodified control run, but decreased in cases in which snow had been removed. These increases in error were attributed to the UM quickly melting added snow rather than retaining it, leading to anomalous perturbations of the surface heat flux. In a second experiment (Exp2), a constant snow amount of 10 kg m−2 was added where IMS denoted snow cover while the model was snow free. The effect on the temperature statistics of this method was less detrimental, suggesting that additions of smaller quantities of snow during assimilation may be less damaging to forecasts of related screen-level variables, such as temperature. Results of these experiments are shown in Table 1 in terms of the change in RMS error and change in the absolute value of the bias on modifying snow amounts. Positive changes denote an increase in error, or degradation, and negative changes denote a decrease in error, or improvement.

Table 1.

Mean statistics showing the impact on UM screen-level temperatures of modifying the UM snow amount field using IMS data, verified against SYNOP observations. Shown are the changes in RMS error and in the absolute value of the bias in screen temperature (K) when snow is modified relative to an unmodified run, where positive changes are increases in error and negative changes are decreases in error. The bias did not change sign in any of the cases. In Exp1, snow was added in variable amounts (10–50 kg m−2), whereas in Exp2 added snow was of a constant amount (10 kg m−2). Results are shown for all cases and separately for cases in which snow was added and removed. The number of cases contributing to the statistics is given by N.

Table 1.

While monitoring snow cover fraction over North America, Romanov et al. (2003) showed that there is a close correlation between snow fraction and snow depth in nonforested areas. This can be simply explained by considering that the deeper the snowpack becomes, the more vegetation it will completely cover, and therefore the higher the fraction of cover “seen” by satellite will be. As the vegetation canopy increases in height, the less likely complete snow cover is to be seen from space, and the correlation becomes less noticeable. However, they did show that the snow fraction remains sensitive to changes in the snow depth in lightly forested areas, and the correlation is still statistically significant even in the most densely forested regions.

These results suggest that there may be a way of extracting information about snow amounts from the IMS snow cover data, if they can be converted to a fractional cover product. Several NWP models derive fractional snow cover from a gridbox value of snow-water equivalent (e.g., Drusch et al. 2004), and a relationship such as this is already used in the UM for specifying albedo as an interpolation between the snow-free and snowy albedos, weighted by the fractional snow cover (Essery et al. 1999):
e1
where a = actual albedo, a0 = snow-free albedo, as = snowy albedo, S = snow areal density, and D = masking depth of vegetation. Fractional cover is represented by 1 − exp(−DS).
The inverse of the relation can be used to relate fractional cover fc to areal density in the following way:
e2
where D is set to 0.2 m2 kg−1 in the UM.

Figure 2 illustrates the relationship in (2) between snow areal density and fractional cover for a masking depth of 0.2 m2 kg−1. Use of this value of masking depth is consistent with that already in use in the UM for specifying snowy albedos. Variation of the masking depth depending on the land cover type could be explored in future developments to the snow analysis scheme.

Fig. 2.
Fig. 2.

Relationship between snow areal density and fractional cover for a masking depth of 0.2 m2 kg−1.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

b. Data processing

IMS data for the previous day are retrieved from NCEP at 0200 UTC for use in the 0600 UTC assimilation cycle. Although each IMS dataset contains both sea ice concentration and snow cover records, only the snow cover record is extracted for use. Discrepancies between model and IMS land and sea classification are resolved, with only snow cover data corresponding to land on the UM grid accepted for further processing.

The full-resolution snow cover observations are then reprojected to latitude–longitude projection and converted to a fractional cover product on the UM grid: the number of IMS snow cover points falling within each UM grid box is calculated, and a fractional cover value is computed from those points. Quality control is carried out on these fractional cover data to identify, and allow exclusion of, unrealistic IMS reports of snow-covered land. A minimum fractional cover threshold of 0.03 is applied, below which a check is performed against a maximum UM surface temperature threshold of 283.15 K. [Romanov et al. (2003) report that snow often exists in satellite-imagery pixels for which the brightness temperature exceeds freezing level by 10 K.] These tests aim to identify incorrectly specified IMS data points. By imposing the surface temperature threshold, exclusion of legitimate isolated snow cover points representing the edge of a snow field should be avoided. This method does not account for incorrect snow-free IMS points, but isolated errors of this type would only give a slight reduction in fractional cover for the grid box, which would affect the model much less than a transition from no snow to snow.

c. The snow analysis scheme

The snow analysis combines information from both the fractional cover observational product, described above, and the UM snow amount short-range forecast (the model background) from the previous model cycle to produce an analysis of snow amount. The analyzed snow field is initialized to the model background snow amount field, and the analysis is performed for all points for which the fractional cover product has passed quality control and for which the UM background does not contain land ice.

For this scheme, the analyzed snow field is calculated in the following way:

  1. Where fractional cover is zero, analyzed snow amount is set to zero.
  2. Where fractional cover is nonzero but UM background snow amount is zero, analyzed snow amount is calculated according to the relation in (2), up to a maximum value of 10.0 kg m−2.
  3. Where both fractional cover and UM background snow amounts are nonzero, no change is made.

The main elements of the scheme are illustrated in Fig. 3. The analyzed snow field is used to update the model daily at 0600 UTC. Should no IMS data be available, the model background snow amount field is used unchanged. A demonstration of the scheme is illustrated in Fig. 4 where a snow analysis has been performed over Europe. Additions of areas of snow over the Alps, northern Spain, and southern Sweden are clearly shown in Figs. 4c and 4d where no snow was present in the background snow field but was present in the observations.

Fig. 3.
Fig. 3.

Schematic diagram showing how the snow analysis works, where fc = fractional cover, Sb = background snow amount, and Sa = analyzed snow amount.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

Fig. 4.
Fig. 4.

A snow analysis performed on 10 Nov 2008 over Europe, showing (a) the observed fractional cover, (b) the background snow amount field, (c) the analyzed snow amount field, and (d) the snow amount increments.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

Related surface and near-surface variables are also assimilated in the global NWP model: screen temperature and humidity from SYNOP observations are assimilated every 6 h, and soil temperature nudging is performed, in which the first-atmospheric-level temperature increments are added to the surface skin temperature and topmost-level soil temperature, except where snow is present.

4. Assimilation experiments

Global model assimilation experiments have been run during the two main seasons that are affected by snow in the NH. A 1-month experiment was run for December 2006, during snow accumulation, and a 3-month experiment was run for March–May 2007, encompassing the majority of the snowmelt season. Controls for the experiments consisted of equivalent model runs without execution of the snow analysis. Figure 5 shows monthly snow maps, based on IMS, from the Rutgers, the State University of New Jersey, Global Snow Laboratory to illustrate the snow extent during the experiment periods.

Fig. 5.
Fig. 5.

Average snow cover extents, in terms of the frequency of snow cover throughout the month, for the months marked, based on Rutgers Global Snow Laboratory analysis of IMS daily snow maps (data are available online at http://climate.rutgers.edu/snowcover/).

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

a. Behavior of the snow analysis

Figure 6 shows time series of the snow extent in analysis and background, for the NH winter and spring experiments. For the winter experiment (Fig. 6a) snow was removed from approximately 3 times as many grid boxes as it was added to by the analysis, and substantially more was removed after day 20 (18 December). The background contained more snowy points than the analysis throughout the entire period; in other words, there was net removal of snow by the analysis each day. The background and analysis differed as to the presence of snow by an average of approximately 1% of NH grid points. For the spring experiment (Fig. 6b), the snow extent decreased steadily throughout the experiment. Initially snow was added to and removed from approximately the same number of grid points, but after day 34 (4 April) the snow extent in background and analysis diverged as snow was added to increasingly large numbers of grid points and snow removal decreased. This net addition of snow by the analysis as the spring progresses is consistent with previous reports (Cameron 2003) that the UM melts snow too rapidly in the spring.

Fig. 6.
Fig. 6.

Time series of number of grid points for which snow was present in both background and analysis for (a) December 2006 and (b) March–May 2007. The solid line represents the background, and the dashed line shows the analysis.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

Figure 7 (December 2006) and Fig. 8 (April 2007) show how many snow additions, (Figs. 7a and 8a) and removals (Figs. 7b and 8b) were made to each grid box throughout the month. This gives an indication of the spatial behavior of the snow analysis during the two seasons and highlights areas of persistent model bias. During December 2006 snow cover was initially fairly steady overall, with a rapid increase westward across eastern Europe after 18 December, as seen in Fig. 6a. Changes made by the snow analysis were either at snow field edges or in regions of complex terrain. Snow was consistently added in the Sierra Nevada and Rocky Mountain ranges of North America, as seen in Fig. 7a. Snow was removed every day in the region of the Tibetan Plateau, indicating a persistent model bias in this area (see Fig. 7b). Fairly frequent snow removals over western Europe, shown in Fig. 7b, suggest that the model built up snow too early in the season, particularly over southern Europe, where snow was removed for more than one-half of the days of the month in some cases.

Fig. 7.
Fig. 7.

Number of snow (a) additions and (b) removals made to each grid box throughout December 2006.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

Fig. 8.
Fig. 8.

As in Fig. 7, but for April 2007.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

In the spring, the pattern of snow additions followed the northeastward retreat of the snow field in both North America and Eurasia, with large areas of snow reinstated by the analysis each day to correct premature melting of the snow field by the model, as seen in Fig. 8a. Figures 9a and 9b show the analysis increments and model background snow field, respectively, on 26 April 2007, clearly illustrating the position of the model snow field relative to bands of snow addition in North America and Eurasia. Snow removal in April was over much smaller areas, but it is interesting that there was a large area of fairly frequent removals in southern Scandinavia and south of Scandinavia, shown by Fig. 8b, as in December, which could point to a persistent model snow overestimation in this region.

Fig. 9.
Fig. 9.

(a) Analysis increments and (b) model background snow amount for 26 Apr 2007, from the spring experiment.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

It is clear from Figs. 7 and 8 that many of the changes made are not retained by the model and must be made again by the subsequent analysis. This is particularly so for additions of snow to the model; if the model surface temperature is above freezing, the added snow quickly melts. As discussed in section 3a, the same result has been found by others, and the hydrology balance could be adversely affected if a much greater quantity of snow were added to try to force some to remain unmelted. This lack of retention of assimilated information makes significant impacts on forecast accuracy unlikely.

The UM will soon use the Joint U.K. Land Environment Simulator (JULES; Blyth et al. 2006) as its land surface model. JULES includes an improved multilayer snow scheme, and there are plans to allow partial snow cover in future developments. Both of these changes should improve model springtime snow retention, meaning that the snow assimilation should not have to work so hard and may be more successful at making lasting snow additions to the model background.

b. Validation of snow analysis experiments

There are two aspects to validation of the snow analysis experiments. As stated in section 1, the main aim of implementing a snow analysis is to improve model analyzed snow fields, and so the validation first concentrates on verification of the analyzed snow fields against alternative observations. Although no large impacts on forecast skill are anticipated overall, it is important to establish that the snow analysis has not degraded forecast skill significantly. Forecast skill impacts are therefore calculated for the main prognostic variables, with particular attention to screen-level temperature and humidity forecasts, which could be beneficially affected by an improved snow cover analysis.

1) Qualitative verification

The National Operational Hydrologic Remote Sensing Center (NOHRSC) produces operational daily snow maps, using ground-based, airborne, satellite, and snow model data for the coterminous United States and Alaska. Although not entirely independent, the NOHRSC products are often used to evaluate the accuracy of satellite-derived snow cover (e.g., Bitner et al. 2002; Maurer et al. 2003) and provide a convenient daily source of qualitative validation data in these experiments.

Comparison of NOHRSC daily snow depth analyses with UM background (before modification) and analyzed (after modification) snow amounts, in terms of presence or absence of snow, shows that there is generally good agreement between the NOHRSC product and UM analyses for snow coverage. This agreement is often, but not always, better than that between NOHRSC and UM background. In the winter experiment, over the United States, the snow analysis increments were generally small and the UM background tended to represent the snow cover well, with some notable exceptions. Figure 10 shows comparisons for three dates that illustrate interesting features. The plots for 13 December 2006 show two points of interest at which the analysis has improved the comparison with the NOHRSC product. The analysis has added snow in the Sierra Nevada range that is absent in the background, and this change was seen frequently in both experiment seasons. The analysis has also captured snow cover around the Great Lakes better than did the background, although the extent may be overestimated. Changes made by the analysis often led to small degradations in the snow cover representation, for instance on 24 December 2006 where the analysis has exacerbated a positional error in the background where the snowband across the mid–United States intercepts the Great Lakes. However, in the spring experiment, when the UM background tended to exhibit prematurely depleted snow fields, the snow analysis made huge improvements in the North American snow cover representation. The background snow field shows practically no snow cover over the mountainous western parts of the United States on most days after 14 April. It is clear from the NOHRSC analysis that there is significant, albeit patchy, coverage over these regions throughout the period, and the cover is well captured by the analyzed snow field.

Fig. 10.
Fig. 10.

(a) Analyzed snow amounts, (b) NOHRSC snow depth (data are available online at http://www.nohrsc.noaa.gov/nsa), and (c) background snow amounts for (top) 13 and (middle) 24 Dec 2006 and (bottom) 14 Apr 2007 over the United States.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

On occasion, the analysis compared less favorably to the NOHRSC product, in a specific region, where areas of new snow cover in the background were removed by the analysis and were not represented in the IMS data until a day or two later. This apparent time lag in the IMS data is discussed in section 4b(3) below.

2) Quantitative verification

Quantitative verification of the analyzed snow fields, in terms of snow cover, is problematic for a number of reasons. The IMS product comprises so many different data sources, including other analyzed products, that it is very difficult to find an entirely independent dataset to use for verification. Ground station observations are often used as “ground truth,” but as point-scale, spatially sparse observations they are generally nonrepresentative of the much larger footprint of a satellite-derived product. However, SYNOP “state of ground” reports have been used here to give some measure of the effectiveness of the snow analysis. Only a subset of stations regularly submits this part of the report, however, and therefore a missing data entry cannot be taken as a no-snow report. Here, verification has been carried out using only those stations that do report snow and as such is one-sided because it tells us nothing about how well nonsnowy points are represented (i.e., it does not allow us to verify snow removal). Figure 11 shows the SYNOP station data from 7 December 2006 for Europe and North America. In red are those stations that have submitted a state-of-ground report for that day. Stations submitting state-of-ground reports in North America at this time of year are so sparse that these data have not been used here for verification purposes.

Fig. 11.
Fig. 11.

Positions of SYNOP stations in (a) Europe and (b) North America for which reports were received between 0500 and 0700 UTC 7 Dec 2006. Marked in red are those stations that have reported on “state of ground.”

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

Using SYNOP station data from Europe, where a large proportion of the stations submit a state-of-ground report, snow presence has been compared between station and model data at 0600 UTC daily for a period of 14 days within each experiment. Presence of snow was diagnosed from SYNOP state-of-ground reports for observations valid from 0500 to 0700 UTC. The UM grid box into which each SYNOP report fell was calculated, and multiple reports for a single grid box were discarded. Model snow cover was diagnosed by snow amounts greater than zero. The number of control and experiment grid boxes in agreement with SYNOP reports within them, as to the presence of snow, is shown in Fig. 12 in percentage terms.

Fig. 12.
Fig. 12.

Percentage of model grid points, in Europe, with snow presence in agreement with SYNOP snow reports within them for (a) 1–14 Dec 2006 and (b) 10–23 Apr 2007. The control run, without snow analysis, is shown by the solid line, and the experiment run is shown by the dashed line.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

During the winter experiment, the snow analysis led to an improvement in the agreement on most days, but a dramatic improvement was evident in the spring experiment, with the percentage of grid boxes in agreement with the SYNOP observations doubled on several days. Given that the predominant effect of the snow analysis in the winter experiment was snow removal, especially over Europe, and that this method only allows us to verify snow addition, it is not surprising that the results were less convincing for the winter experiment. Both results do show a clear improvement in the agreement of the model snow with ground-based observations with the snow analysis in operation, however, and can be taken as evidence that the snow analysis has improved the model snow cover representation at analysis time.

3) Time lag in IMS data

In his investigations Cameron (2003) found that, for an IMS snow cover map for a certain day, the best correlation was with the UM snow cover from the previous day. In fact, he concluded that by the time the IMS data were available, they were lagging the UM by about 36 h. Romanov et al. (2000) explains how the time lag between the IMS data and the NWP model into which it is assimilated arises, using North America as an example. Since surface reports can be included from any time in the 24-h period leading up to the cutoff time of 1200 UTC, and GOES data from the eastern United States can be used up until 1700 UTC, data used in the IMS analysis can encompass a time range of up to 29 h. The IMS analysis is received by the Met Office shortly before 0000 UTC and is assimilated at 0600 UTC the following day, allowing for a potential difference of 42 h between the validity time of an IMS data point and the model snow field with which it is compared at the Met Office. Of course, this is an extreme example that is unlikely to occur, since there will nearly always be multiple sources of data and multiple satellite passes during the period that the IMS analyst draws information, but where cloud cover is present for a large period of that time the available observations are considerably reduced and large time lags can develop for some regions of the IMS analysis.

The effects of such data time lags were sometimes apparent in the experiments, especially in the winter season when changes in snow cover occurred on very short time scales (snowfall as opposed to snowmelt). One striking case occurred during 30 November–1 December 2006 when a large-scale snowfall event over the United States occurred. The model background captured the event well, verified against NOHRSC snow depth, but the snowfall was not represented by the IMS data until 2 December, and so the snow analysis removed the snow from this area and degraded the model snow cover representation for these 2 days. The comparison of analysis and background against NOHRSC snow depth for 30 November 2006 is shown in Fig. 13, with the large-scale snowfall clearly visible, extending southwest from the Great Lakes, in the NOHRSC and UM background plots (Figs. 13b,c) but absent in the analysis (Fig. 13a).

Fig. 13.
Fig. 13.

As in Fig. 10, but for North America on 30 Nov 2006.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

The effects of time lags in the data are much less evident in the spring, when the main changes to the snow field are due to seasonal melting, which happens on a longer time scale than winter snow storms. The data composing the IMS snow cover product are perhaps also less likely to be cloud covered, allowing more cloud-free data to be available to the analyst.

A method has been devised to mitigate the effects of a time delay in the IMS data, using information from the previous day’s model background snow field as a further constraint on the analysis system. For cases in which the UM has forecast a snowfall event well and there is a significant time delay in the IMS data, the IMS data may be expected to compare better to the previous day’s background snow field than to the current one. Provided that the comparison to the previous day’s background is good, it is reasonable to trust the model representation of the snow event and make no change in the analysis. Used in this way, this method would add an additional constraint to cases of snow removal (i.e., where IMS snow cover is zero but the UM denotes snow) and reduce instances of incorrect snow removal by the analysis.

The method is illustrated in Fig. 14a for cases in which observed fractional cover is zero. Figure 14b demonstrates the effectiveness of this method when applied to the case highlighted in Fig. 13. Use of the previous day’s background model snow field has allowed the model to retain the snow feature previously removed, giving an analysis that compares much better to the NOHRSC product.

Fig. 14.
Fig. 14.

(a) Schematic diagram showing how the improved snow removal method works, where fc = fractional cover, Sb = background snow amount, and Sa = analyzed snow amount. (b) Analyzed snow amounts for 30 Nov 2006, using the improved method.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

In the longer term there are developments planned to the IMS product itself to tackle the time delay, and other data sources—such as the snow products from the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Land Surface Analysis Satellite Applications Facility (see http://landsaf.meteo.pt/)—will become available for consideration. Helfrich et al. (2007) report that there are plans to release a second IMS product each day over North America to increase the timeliness of data in this region. There will also be available a file giving “time of last observation” for the whole IMS grid, which will allow the data to be quality controlled based on their timeliness. The new MODIS cloud-gap-filled product (Hall et al. 2010) is also of interest, giving snow measurement age information for each pixel.

c. Forecast verification

Overall forecast RMS error impacts, verified against observations and against corresponding analyses from the assimilation experiments, are shown for both experiments in Table 2. These are global impacts, using weighted averages over all forecast ranges and over the prognostic variable pressure at mean sea level (PMSL) and profiles of geopotential height, wind speed, temperature, and relative humidity (RH). As anticipated, the implementation of the snow analysis has a neutral impact overall in both experiments. Impacts on all individual forecast variables are small, although it is interesting to note that the largest impacts are on tropical and Southern Hemisphere components (where there is no direct action by the snow analysis). This result is probably attributable to noise rather than any genuine scientific effect, but it is worth noting that scientists have identified teleconnections between the snowpack in certain regions and subsequent meteorological conditions in other regions (Rodell and Houser 2004), although the time frame is likely to be somewhat longer than the experiment periods considered here.

Table 2.

Average forecast RMS error changes (experiment minus control), over the main prognostic variables, and main characteristics of the winter and spring snow analysis experiments.

Table 2.

Regional verification has been performed for the main NH continental regions, since the snow analysis tends to affect different regions differently, depending on factors such as existing model biases, orographic characteristics, and typical weather patterns. The most noticeable changes are evident in Europe and North America, where the snow analysis behaves very differently. In the winter experiment, although snow was predominantly removed overall, there was more snow addition than removal in North America and more snow removal than addition in Europe. In the spring experiment, there was net addition increasing throughout the period in both regions, but more so in North America.

Though small, there is evidence of consistent improvements in surface and low-level temperature and RH forecasts in situations in which snow is predominantly removed by the snow analysis. In the winter experiment, surface and lower-level temperatures were generally slightly improved, with the most consistent improvements in Europe. Of interest is that surface RH was improved for Europe but degraded for North America (Figs. 15a,b). This suggests that the repeated snow addition, seen in North America, and subsequent melting before the following daily snow analysis has degraded the surface RH forecast by upsetting the surface hydrological balance. Improvements were seen in above-surface RH, especially in North America, where the effects of the repeated snowmelt are more remote.

Fig. 15.
Fig. 15.

Impacts on surface variables during the December experiment. Forecast mean (upper panel of group) and RMS (lower panel of group) error verified against observations for surface RH over (a) Europe and (b) North America and verified against analysis for (c) snow depth over Europe. The experiment run (with snow analysis) is shown by a dashed line, and the control run (with no snow analysis) is shown by a solid line.

Citation: Journal of Applied Meteorology and Climatology 50, 5; 10.1175/2010JAMC2527.1

In the spring experiment, changes were very small and were mixed. Low-level temperatures tended to be slightly improved, but not at the surface. This is consistent with the results of the winter experiment, where surface temperatures could only be improved where snow was predominantly removed by the analysis. Rodell and Houser (2004) found similar benefits to forecast fields when removing snow. This situation, which was typical of Europe in the December experiment, also produced the only visible impact on snow-depth forecasts, with a small improvement in mean and RMS error (Fig. 15c), verified against analyses.

5. Summary

A daily NH snow analysis, using NESDIS IMS snow cover data, has been developed and implemented in the UK Met Office’s operational global NWP model. This is the first introduction of observational snow data into the global model and aims to improve the model representation of snow cover at analysis time. Global assimilation experiments have been run during the two main snow-affected seasons for the NH: during December (2006), when snow is accumulating, and from March to May (2007), encompassing the majority of the snowmelt season. No significant impacts on forecast accuracy were found, and much of the validation focused on the analyzed snow field.

Throughout the December experiment there was net snow removal by the analysis, with snow consistently being removed from central Asia and, in the latter part of the experiment, from northern and eastern Europe, where the UM has a tendency to build up snow cover prematurely. During the 3-month spring-season experiment, there was a net addition of snow by the analysis, over increasingly large areas as the spring progressed. Large areas of snow cover were reinstated over North America and eastern Europe, where the model had melted snow too early.

The changes made by the snow analysis generally verify well in qualitative terms, against other observational and analyzed snow cover products, particularly the large-scale additions of snow made toward the end of the spring experiment. In Europe, where SYNOP stations making use of snow reporting are sufficiently numerous, the number of model grid points in agreement with SYNOP reports of snow-covered ground increased in both seasons for which the snow analysis was used. This, along with qualitative comparisons of snow cover, gives clear evidence that the snow analysis has improved the analyzed snow field in terms of the presence or nonpresence of snow.

Changes to the forecast accuracy of all of the main prognostic variables were small, especially in the NH, but there is some evidence of improvements in surface and low-level temperature and relative humidity forecasts. This was noted in situations in which snow was predominantly removed by the snow analysis, with the most consistent improvement observed for Europe during the winter experiment. Many of the changes made by the analysis were not retained, particularly in cases of snow addition. It is therefore unsurprising that positive impacts on forecast skill were not found where daily additions and subsequent melting of large areas of snow occurred. Future planned developments to the UM snow scheme physics are expected to help the retention of added snow. Note also that an improved surface snow analysis will enable more satellite observations to be used over land, which may contribute to improved forecast skill.

Acknowledgments

Thanks are given to Roger Saunders, Bruce Macpherson, and Adrian Lock for their advice and guidance and to Mike Thurlow and Nigel Atkinson for technical assistance. Sean Helfrich (NESDIS) and George Gayno (NCEP) were both very helpful when arranging the IMS data transfer to the Met Office.

REFERENCES

  • Armstrong, R. L., , and M. J. Brodzik, 2001: Recent Northern Hemisphere snow extent: A comparison of data derived from visible and microwave satellite sensors. Geophys. Res. Lett., 28, 36733676.

    • Search Google Scholar
    • Export Citation
  • Bitner, D., , T. Carroll, , D. Cline, , and P. Romanov, 2002: An assessment of the differences between three satellite snow cover mapping techniques. Hydrol. Processes, 16, 37233733.

    • Search Google Scholar
    • Export Citation
  • Blyth, E., and Coauthors, 2006: JULES: A new community land surface model. Global Change Newsletter, No. 66, International Geosphere–Biosphere Programme, The Royal Swedish Academy of Sciences, Stockholm, Sweden, 9–11. [Available online at http://www.igbp.net/documents/NL_66-3.pdf.]

    • Search Google Scholar
    • Export Citation
  • Brasnett, B., 1999: A global analysis of snow depth for numerical weather prediction. J. Appl. Meteor., 38, 726740.

  • Brubaker, K. L., , R. T. Pinker, , and E. Deviatova, 2005: Evaluation and comparison of MODIS and IMS snow-cover estimates for the continental United States using station data. J. Hydrometeor., 6, 10021017.

    • Search Google Scholar
    • Export Citation
  • Cameron, J., 2003: Comparison of Unified Model snow and sea-ice fields with NESDIS and USAF data sets. Forecasting Research Tech. Rep. 416, Met Office, 40 pp.

    • Search Google Scholar
    • Export Citation
  • Davies, T., , M. J. P. Cullen, , A. J. Malcolm, , M. H. Mawson, , A. Staniforth, , A. A. White, , and N. Wood, 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 17591782.

    • Search Google Scholar
    • Export Citation
  • Drusch, M., , D. Vasiljevic, , and P. Viterbo, 2004: ECMWF’s global snow analysis: Assessment and revision based on satellite observations. J. Appl. Meteor., 43, 12821294.

    • Search Google Scholar
    • Export Citation
  • Essery, R., , E. Martin, , H. Douville, , A. Fernández, , and E. Brun, 1999: A comparison of four snow models using observations from an alpine site. Climate Dyn., 15, 583593.

    • Search Google Scholar
    • Export Citation
  • Hall, D. K., 1988: Assessment of polar climate change using satellite technology. Rev. Geophys., 26, 2639.

  • Hall, D. K., , G. A. Riggs, , J. L. Foster, , and S. V. Kumar, 2010: Development and evaluation of a cloud-gap-filled MODIS daily snow-cover product. Remote Sens. Environ., 114, 496503, doi:10.1016/j.rse.2009.10.007.

    • Search Google Scholar
    • Export Citation
  • Helfrich, S. R., , D. McNamara, , B. H. Ramsay, , T. Baldwin, , and T. Kasheta, 2007: Enhancements to, and forthcoming developments in the Interactive Multisensor Snow and Ice Mapping System (IMS). Hydrol. Processes, 21, 15761586.

    • Search Google Scholar
    • Export Citation
  • Maurer, E. P., , J. D. Rhoads, , R. O. Dubayah, , and D. P. Lettenmaier, 2003: Evaluation of the snow-covered area data product from MODIS. Hydrol. Processes, 17, 5971.

    • Search Google Scholar
    • Export Citation
  • Nolin, A. W., , and S. Liang, 2000: Progress in bidirectional reflectance modelling and applications for surface particulate media: Snow and soils. Remote Sens. Rev., 18, 307342.

    • Search Google Scholar
    • Export Citation
  • Ramsay, B. H., 1998: The Interactive Multisensor Snow and Ice Mapping System. Hydrol. Processes, 12, 15371546.

  • Rodell, M., , and P. R. Houser, 2004: Updating a land surface model with MODIS-derived snow cover. J. Hydrometeor., 5, 10641075.

  • Romanov, P., , G. Gutman, , and I. Csiszar, 2000: Automated monitoring of snow cover over North America with multispectral satellite data. J. Appl. Meteor., 39, 18661880.

    • Search Google Scholar
    • Export Citation
  • Romanov, P., , D. Tarpley, , G. Gutman, , and T. Carroll, 2003: Mapping and monitoring of the snow cover fraction over North America. J. Geophys. Res., 108, 8619, doi:10.1029/2002JD003142.

    • Search Google Scholar
    • Export Citation
1

Note that 10 kg m−2 areal density ≡ 100-mm snow depth ≡ 10-mm SWE based on snow density of 100 kg m−2 (Brasnett 1999).

Save