Abstract

Significant climatic changes over northern Eurasia during the twentieth century are revealed in numerous variables including those affecting and characterizing the state of the cryosphere. In addition to commonly used in situ observations of snow cover such as snow depth and snow courses, synoptic archives in the former Soviet Union contain regular daily and semidaily reports about the state of the ground in the area surrounding the station. Information about frozen, dry, wet, ponded, and snow-covered land, and in the case of snow-covered land, about the characteristics of snow cover, is available in these reports. A new Global Synoptic Data Network (GSDN) consisting of 2100 stations within the boundaries of the former Soviet Union created jointly by the National Climatic Data Center (NCDC) and Russian Institute for Hydrometeorological Information (RIHMI) was used to assess the climatology of snow cover, frozen and unfrozen ground reports, and their temporal variability for the period from 1936 to 2004. Comparison with satellite measurements of snow cover extent is also presented.

During the second half of the twentieth century and over many regions in northern Eurasia, an increase in unfrozen ground conditions (5 days since 1956 over the Russian Federation) was observed. The most prominent changes occurred in the spring season in Siberia and the Far East north of 55°N during April and May by 3 to 5 days, which constitute a 15%–35% change in these regions compared to long-term mean values. Since the beginning of the dataset, surface temperature changes in high latitudes have not been monotonic. As a result, linear trend analyses applied to the entire period of observations can lead to paradoxical conclusions. Specifically, changes in snow cover extent during the 1936–2004 period cannot be linked with “warming” (particularly with the Arctic warming) because in this particular period the Arctic warming was absent.

1. Introduction

In the twentieth century, northern Eurasia was the region with the largest and steadiest increase of surface air temperature, which became the most pronounced during the past 50 years (Fig. 1). This warming should manifest itself in changes of environmental characteristics affecting both terrestrial ecosystem dynamics and human activity. The length of time when the soil is unfrozen, frozen, and/or when it is covered by snow is among the most important of these characteristics in high latitudes. These characteristics are linked to both biogeophysical and biogeochemical processes by which terrestrial ecosystems feed back to climate, enhancing or damping changes on different spatial scales (Schultze et al. 2001; Kabat et al. 2004; NEESPI 2004). Recently, information about the changes of the surface conditions became widely available from various remote sensing products (Robinson et al. 1993; Robinson and Frei 2000; Fung 1997; Ramsay 1998, 2000; Myneni et al. 1997; Kidd et al. 2004). Fortunately, changes in high latitudes are large, so changes reported from both in situ observations and remote sensing are not seriously impacted by time series inhomogeneities and usually support each other.

Fig. 1.

Mean (top left) winter, (top right) spring, (bottom left) summer, and (bottom right) autumn temperature change 1956 to 2005 over the globe. The change estimates were made by fitting linear trend lines. [Data source is Jones and Moberg (2003), updated.] Processed by the NCDC Global Climate at the Glance Mapping System. (More information about the system is available online at http://www.ncdc.noaa.gov/gcag/gcag.html)

Fig. 1.

Mean (top left) winter, (top right) spring, (bottom left) summer, and (bottom right) autumn temperature change 1956 to 2005 over the globe. The change estimates were made by fitting linear trend lines. [Data source is Jones and Moberg (2003), updated.] Processed by the NCDC Global Climate at the Glance Mapping System. (More information about the system is available online at http://www.ncdc.noaa.gov/gcag/gcag.html)

An important attribute of the in situ observations is their availability throughout the entire period of instrumental observations well before the first satellite was ever launched. The primary climate variable considered in this paper has been observed over northern Eurasia during the past 70 years using the same “instrument”—observers’ reports. The “state of the ground” is a standard climatological variable reported at synoptic stations worldwide according to World Meteorological Organization (WMO) Instructions (codes 0901 and 0975; e.g., http://badc.nerc.ac.uk/data/surface/code.html). Virtually every station in the former Soviet Union made these noninstrumental observations twice a day (in the morning and in the evening) up to 1982. Thereafter, only morning observations were preserved. The observational instructions did vary slightly over time, but these changes were well documented (Shakirzyanov and Razuvaev 2000). The observer was required to report the condition of the surface in the following categories: dry, wet, ponded (extremely wet), frozen (but not snow covered), covered by snow partially (less than 50% or more than 50% of the area around the observer’s visual scan), or completely snow covered. When digitized, this information has been generalized by 10 codes (Table 1) available during the entire period of observations.1 In our previous use of the state of the ground information, we focused on the assessment of dry ground conditions versus all others (Groisman et al. 1997). Now, we investigate its characteristics during the cold season as well as transition seasons. Specifically, climatology, variability, and linear trends2 of three major characteristics of these seasons are considered: (i) the duration of the cumulative period when soil is unfrozen and is not covered by snow; (ii) various indicators of the presence of snow on the ground; and (iii) the characterization of changes between frozen/unfrozen (or snow covered) land during the day. Our interest in the last characteristic is driven by new remote sensing products related to frozen/unfrozen ground (Running et al. 1999; Kimball et al. 2004; Kidd et al. 2004). These new products have not yet been comprehensively assessed with “ground truth” (i.e., in situ) observations. The in situ data may enable us to expand these products backward into the past using the state of the ground observations. Therefore, we try to characterize better the representativeness of remote sensing products from polar orbiters and in situ observations where the observation frequency is limited to only a once-a-day observation of a given land pixel3 in relation to the diurnal cycle.

Table 1.

State of the ground codes used in this study prior to 1984.

State of the ground codes used in this study prior to 1984.
State of the ground codes used in this study prior to 1984.

2. Data and processing

a. Data availability

For the former Soviet Union, more than 2100 stations systematically reported the state of ground conditions during the twentieth century. Data holdings of the World Data Center B for Hydrometeorology in the Russian Institute for Hydrometeorological Information (RIHMI) contain this information in digital form since 1936 with the period of maximum data availability between 1956 and 1991 (cf. NCDC 2005). Thereafter, only Russian stations are being archived in RIHMI and all other republics of the former Soviet Union maintain their own meteorological archives, which were not available to the authors of this study. Approximately 1900 stations within 1209 1° × 1° grid cells have more than 80% of the data during the 1961–90 reference period (Fig. 2). Visual screening for time-dependent biases and quality control routines further reduced the number of stations used for our analyses to 1811 and the number of grid cells to 1197. Only these stations were used in subsequent analyses of this paper.

Fig. 2.

Data availability in the archive used for this study: (left) 1° × 1° grid cells that contain long-term stations with state of the ground information during the 1961–90 reference period (in total, 1209 grid cells encompassing ∼1900 meteorological stations). Albers projection that preserves areas of the regions is used here and in other figures throughout this paper with regional (i.e., with averaged over the region) data loads. Regional averaging has been performed over 11 regions outlined in this figure (rationale for their selection presented in 2b). Results for the Russian Arctic (regions poleward of the Arctic Circle), Siberia (partitioned in three regions: west, central, and southern Siberia), the Russian Far East (partitioned in two regions: north and south of 55°N), mountainous regions of Central Asia and the Caucasus, and northwestern Russia (north of 60°N and westward of the Ural Mountains) are discussed throughout the paper and/or shown in the following figures. (right) The number of 1° × 1° grid cells with data in archive.

Fig. 2.

Data availability in the archive used for this study: (left) 1° × 1° grid cells that contain long-term stations with state of the ground information during the 1961–90 reference period (in total, 1209 grid cells encompassing ∼1900 meteorological stations). Albers projection that preserves areas of the regions is used here and in other figures throughout this paper with regional (i.e., with averaged over the region) data loads. Regional averaging has been performed over 11 regions outlined in this figure (rationale for their selection presented in 2b). Results for the Russian Arctic (regions poleward of the Arctic Circle), Siberia (partitioned in three regions: west, central, and southern Siberia), the Russian Far East (partitioned in two regions: north and south of 55°N), mountainous regions of Central Asia and the Caucasus, and northwestern Russia (north of 60°N and westward of the Ural Mountains) are discussed throughout the paper and/or shown in the following figures. (right) The number of 1° × 1° grid cells with data in archive.

b. Generalized classes of state of the ground assessed in this study

Keeping in mind the documented changes in the reporting and coding practices of the state of the ground conditions, state of the ground was generalized in our analyses as follows: for the period up to July 1959, when reporting snow on the ground, only the codes for more than 50% and less than 50% of snow on the ground (and none) were reported in the former Soviet Union. Moreover, prior to 1950, we found an inconsistency in the reporting of class 5, which was mixed with class 4, and, during the 1950 to 1984 period, an inconsistency with class 8 that was virtually unused. Thus, we had to abandon attempts of using the term “stable snow cover,” attributing it to “classes 7 and 9” only, and all of our special assessment of frozen ground was confined to the period from summer 1950 (a period when the aforementioned inconsistency was first noted) to 1982 (when the last report of the afternoon state on the ground was archived). Class 5 was grouped with classes 3 and 4 when periods that include the pre-1950 years were analyzed. Table 1 has been constructed as a combination of two elements that were mostly reported in the warm season (0 through 2) and during the cold season (3 through 9). Since 1984, state of the ground coding was done using WMO codes, shown in Table 2, which were converted (with some loss of resolution) into the codes used in Table 1. Prior to 1984, in the warm season in steppe and desert zones, codes 8 and 9 were also used for reporting sand- and/or dust-covered surroundings (instead of snow-covered areas: Razuvaev et al. 1995). In our analyses these occurrences were filtered out using the temperature criteria, and all these cases were assigned code 0 (dry soil without snow cover). A few stations located in sandy deserts of Central Asia, where this precaution was found to be insufficient, were excluded from analyses.

Table 2.

WMO codes recommended for use in reporting state of the ground (available at http://badc.nerc.ac.uk/data/surface/code.html).

WMO codes recommended for use in reporting state of the ground (available at http://badc.nerc.ac.uk/data/surface/code.html).
WMO codes recommended for use in reporting state of the ground (available at http://badc.nerc.ac.uk/data/surface/code.html).

Classes 6 through 9 were used to report “snow cover” conditions over more than half of the observer line of sight. This corresponds to the commonly used practice to assign a pixel to be snow covered in remote sensing (Robinson et al. 1993; Ramsay 1998). Classes 0 through 2 were used to report “unfrozen soil” and classes 3 to 5 to report “frozen soil and/or area with remnants of snow cover” in the station neighborhood. In some of our analyses, a “frozen soil without snow” (classes 3 and 4) and “soil with some snow” (classes 5 through 9) grouping was used. Table 3 summarizes the group definitions and the instances when they were used in our analyses.

Table 3.

Definitions of various characteristics of state of the ground used throughout this paper. For climatological purposes, we used SC, UF, FSWS, and RSC. Trend analysis has been condacted for SC, UF, and FSRSC. For analyses of intradaily changes of the state of the ground as a function of daytime temperature (section 4) all characteristics, except RSC, were employed.

Definitions of various characteristics of state of the ground used throughout this paper. For climatological purposes, we used SC, UF, FSWS, and RSC. Trend analysis has been condacted for SC, UF, and FSRSC. For analyses of intradaily changes of the state of the ground as a function of daytime temperature (section 4) all characteristics, except RSC, were employed.
Definitions of various characteristics of state of the ground used throughout this paper. For climatological purposes, we used SC, UF, FSWS, and RSC. Trend analysis has been condacted for SC, UF, and FSRSC. For analyses of intradaily changes of the state of the ground as a function of daytime temperature (section 4) all characteristics, except RSC, were employed.

c. Major type of analyses

We assess climatology and linear trends of seasonal and annual duration of each of the aforementioned five groups of the state of the ground: snow cover (SC), unfrozen soil (UF), frozen soil and/or area with remnants of snow cover (FSRSC), frozen soil without snow (FSWS), and soil with some snow (SSS). The last two groups were analyzed and presented only from autumn 1950 due to the aforementioned discrepancy in reporting of class 5 conditions. Seasons were selected by partitioning the hydrological year (October–September) into the cold season (December–March), the warm season (June–September), and two transient seasons (cf. Groisman et al. 1994).

Processing includes calculating anomalies relative to 1961–90, grid cell and regional area averaging within 11 regions of the former Soviet Union, and finally the overall averaging with weights proportional to the regional areas. Most of the results of the following sections are presented in the form of a 1° × 1° grid cell or regionally averaged quantities. Climatologically motivated regionalization was selected for the former Soviet Union. Eleven regions of the former Soviet Union used throughout this paper are the same as in Groisman et al. (2005) and schematically represent climatic regions of the former Soviet Union according to Alisov (1957) and Shver (1976) with additional consideration of differences in data availability (e.g., in Siberia south and north of 55°N). Area averaging over the regions was performed as follows: First, station data anomalies from the long-term mean for the reference 1961–90 period (or statistics) were averaged within each 1° × 1° grid cell. This step allows for suppression of unduly impacts of station clusters on the area mean values. The grid cell values (those with at least one valid station value within) were then averaged further over the region with weights proportional to the grid cell latitude cosine. Area averaging over megaregions (e.g., the Russian Federation or the former Soviet Union) is then conducted by averaging the regional mean values with the weights proportional to the areas of the regions that compose the megaregion. This averaging routine was tested many times for different meteorological elements, is robust, and represents a reasonable compromise compared to the optimal area-averaging routines when we are lacking sufficient information about the spatial covariance function of the averaged meteorological field. It does not claim to cover the entire region but focuses on the data-elucidated regions where people live and have maintained meteorological observations for sufficiently long periods of time. Because people tend to settle in locations with a milder climate, we anticipated potential biases when compared to our in situ estimates using “global” satellite products (cf. section 5a). Therefore, we used an additional method of area averaging using Thiessen polygons (Thiessen 1911) to calculate regional climatology and estimate the causes of biases in comparison between the remote sensing and in situ snow cover products in section 5a. Comparing to our area-averaging routine, the Thiessen polygon averaging routine is supposed to deliver less biased estimates of the long-term mean regional values when the measurements are “accurate” and/or the micrometeorological variability is small, but for regionally averaged time series it yielded results with less accuracy.4

Section 3 presents these results for 11 regions of the former Soviet Union as well as the geographical distribution of mean seasonal values of the state of the ground and its changes during the periods 1936 to 2004 (Russia) and 1936/1956 to 1991 for the entire Soviet Union. Trend analysis has been conducted only for the period beginning with the 1956 hydrological year due to better data availability since then. Throughout this paper, changes are estimated by fitting the line to the time series and present trend estimates in unit (°C, days, or %) per the period duration (69 yr, 36 yr, or 49 yr). Before conducting linear trend analyses, we realized that linearity is not the best hypothesis when dealing with snow cover products over this period (since 1936). Snow cover extent (duration) on regional and continental scales is correlated with temperature variations (Voeikov 1949; Karl et al. 1993; Groisman et al. 1994), and our data on regional (hemispheric or Arctic) temperature variations (Groisman et al. 1986; Vinnikov et al. 1990; Lugina et al. 2005; National Climatic Data Center 2005) clearly show that the middle of the twentieth century (and the beginning of our time series) was warm in high latitudes just like the past two decades. Nevertheless, when presenting time series for a sufficiently lengthy period of 69 yr, we have no choice but to answer the question pertaining to what were the changes during this particular period. In this regard, the mean rate of these changes is provided by linear trend estimates.

d. Three modes of presentation of results

For presentation of the results of this study, the following approach is used. Climatology and trends of each characteristic are presented generalized in maps twice: first for each 1° × 1° grid cell where we have stations with sufficiently long time series of state of the ground observations (1197 grid cells) and, thereafter, area averaged within 11 preselected climatic regions. Nationwide results area averaged for the former Soviet Union territory and Russia are presented in tables. Each mode of presentation (grid cell, regional, and nationwide averages) has its purpose.

Presenting results in 1° × 1° grid cells, we demonstrate the spatial pattern of climatology and trend estimates for each characteristic of the state of the ground as well as data coverage. Gridcell values clearly show where we have station data and provide clues about the nature of changes (or their absence). This mode of presentation for linear trends does not yet deliver statistically significant estimates5 but might show a pattern of changes (if they are revealed in regionally averaged results). Presenting results area averaged within 11 preselected climatic regions and thereafter the nationwide averages, we demonstrate the mean regional climatological values and trend estimates that describe the large-scale features of the state of the ground characteristics and their mean changes during the analyzed period (if these changes exist). The last matter is being checked by testing statistical significance of the mean regional changes. Only statistically significant changes are worthy of discussion. Insignificant regional trend estimates mean, in fact, “no trend” and that the signal of the changes is too weak to be detected. It may well be that the changes over the region are insignificant in a statistical sense, but as a part of the larger region (e.g., all of Russia or the former Soviet Union) they contribute to statistically significant nationwide trends.

The reader is warned upfront about two major features of trend analyses presented in this paper: Each point (grid cell) trend estimate is usually insignificant. Regional and national change estimates could be statistically insignificant too or differ significantly from zero. In the last situation, (i) maps of trend estimates allow the reader to track down the areas where most of the signal originated (if it has a distinctive regional pattern) or to conclude that a weak but spatially well-spread signal of changes is present in the data and area averaging allows us to visualize and quantify it; and (ii) regional and nationwide trend estimates give quantitative estimates of mean changes.

e. Network of “long term” stations

During the first 20 years (1936–55), the state of the ground observations were not available for most of the meteorological stations. Thereafter, this information is practically always present in the station digital records. To incorporate these first two decades into the analyses, a frozen network assessment was performed. Only those stations were selected that were available in the decade of 1946–55, that is, prior to a dramatic increase in the data availability. This reduces the number of stations to 205 within Russia and to 350 within the former Soviet Union. This reduced network is reasonably well distributed over the country (Fig. 3) and is quite similar to (or slightly larger than) the network that has been used in several studies of snow depth changes during the past 70 years over the former Soviet Union. Usually, these studies have been based on 223 stations of international exchange or the 284 stations in the archive, the Historical Soviet Daily Snow Depth Version 2.0 (Armstrong 2001; Ye et al. 1998; Ye 2001; Heino and Kitaev 2005). Therefore, we had an opportunity to assess the representativeness of the network used in these studies versus our dataset. Figures with regional SC, UF, and FSRSC time series were expanded back to 1936. To decide if the first 20 years are reliable for trend analyses, an experiment was conducted using the regional time series based on the long-term network shown in Fig. 3. For the 1956 to 1991–2004 period, these time series were then compared with the corresponding time series that were estimated using the complete network shown in Fig. 2. When the time series are closely correlated and conclusions about changes in regional SC, UF, and FSRSC coincided, we place more confidence in the information for the first two decades in our dataset.

Fig. 3.

Map of long-term stations with state of the ground information available at least at 70% of days during the 1961–90 reference period and more than 50% of days during the first post–Word War II decade (1946–55). Sinusoidal projection centered at 60°N, 90°E is used in this and many other maps of the paper to better visualize the spatial pattern of station (grid cell) distribution.

Fig. 3.

Map of long-term stations with state of the ground information available at least at 70% of days during the 1961–90 reference period and more than 50% of days during the first post–Word War II decade (1946–55). Sinusoidal projection centered at 60°N, 90°E is used in this and many other maps of the paper to better visualize the spatial pattern of station (grid cell) distribution.

f. Understanding of the impact of the diurnal cycle in the state of the ground reports

The morning observation (usually before 0800 local time) is close to the time of the temperature minimum. The period when twice-a-day observations were available (hydrological years from 1951 to 1976) was used to assess the climatological change at the surface during the day. After 1977, the afternoon observation (conducted 12 h after the morning observations) became infrequent and was discontinued in 1981. Using detailed meteorological information for the period of the afternoon observations, we calculated the probabilities of occurrences of a switch from one state in the morning to a different state in the afternoon. Although this may sometimes catch daytime snowfall, our focus is on information about (i) the snow cover retreat during a single day and (ii) the thawing process during the day, both of which would be missed by having only the morning observation. This enables a better interpretation of remote sensing products of snow cover, which are delivered by polar-orbiting satellites with once-a-day resolution (see footnote 3), and the freeze/thaw cycle, which requires a better calibration (cf. Ramsay 1998; Kidd et al. 2004).

g. Satellite snow cover products

The time series based on the NOAA satellite snow cover product (Robinson et al. 1993; Ramsay 1998) were used to compare them with the in situ snow cover data. The gridded weekly snow cover data were downloaded (from http://www.cpc.ncep.noaa.gov/data/snow/) and processed in the manner described by Groisman et al. (1994), which resembles the Rutgers approach (Robinson et al. 1993), considered a standard for this type of data processing. From these data the fraction of weeks with snow on the ground within the hydrological year has been calculated for each grid cell and then area averaged within the country’s limits for each hydrological year from 1972 to 2002.6 Similar time series were constructed using our daily in situ data for the former Soviet Union and Russia for the 1936–91 and 1936–2004 periods, respectively. The station data were first averaged within the 1° × 1° grid cells inside the regions shown in Fig. 2 (for Russia, accounting for the Russian Federation boundaries) and thereafter averaged with the weights proportional to the regions’ areas. Owing to serial completeness of the remote sensing snow cover product, there was no need for these intermediate steps when it was area averaged over Russia and/or the former Soviet Union.

3. Seasonal climatology and trends of state of the ground characteristics

a. Climatology

In Figs. 4 –6 and Table 4 the former Soviet Union national climatology is shown for the 1961–90 reference period. The state of the ground categories were combined in three groups:

  1. no snow and/or ice on the ground and soil is not frozen (categories 0 through 2);

  2. less than a half of the surrounding area is covered by snow and/or the ground is frozen (categories 3 through 5) [This group was further partitioned to single out the frost days frozen ground without snow cover (categories 3 and 4).];

  3. more than half (or all) of the surrounding area is covered by snow, melting snow, and/or ice (categories 6 through 9).

Fig. 4.

Long-term mean annual number of days (top) with unfrozen ground (categories 0–2); (middle) frozen ground and/or the remnants of snow cover (categories 3–5); and (bottom) snow on the ground (categories 6–9).

Fig. 4.

Long-term mean annual number of days (top) with unfrozen ground (categories 0–2); (middle) frozen ground and/or the remnants of snow cover (categories 3–5); and (bottom) snow on the ground (categories 6–9).

Fig. 6.

Regionally averaged long-term mean seasonal number of days (a) with unfrozen ground (categories 0–2); (b) frozen ground (categories 3–4); (c) remnants of snow cover (category 5); and (d) snow on the ground (categories 6–9) in winter (December–March), spring (April–May), summer (June–September), and autumn (October–November).

Fig. 6.

Regionally averaged long-term mean seasonal number of days (a) with unfrozen ground (categories 0–2); (b) frozen ground (categories 3–4); (c) remnants of snow cover (category 5); and (d) snow on the ground (categories 6–9) in winter (December–March), spring (April–May), summer (June–September), and autumn (October–November).

Table 4.

Long-term mean number of days with different state of the ground conditions area averaged over the former Soviet Union. For two major groups, UF and SC, in parentheses, results of the alternative regional averaging (using the Thiessen polygons method) are presented.

Long-term mean number of days with different state of the ground conditions area averaged over the former Soviet Union. For two major groups, UF and SC, in parentheses, results of the alternative regional averaging (using the Thiessen polygons method) are presented.
Long-term mean number of days with different state of the ground conditions area averaged over the former Soviet Union. For two major groups, UF and SC, in parentheses, results of the alternative regional averaging (using the Thiessen polygons method) are presented.

Only morning observations were used for this analysis. The maps in these figures reveal the domination of cold climates of the former Soviet Union, with approximately half of the region (and most of Russia) having less than five months with unfrozen ground (Fig. 4a). This suppresses soil microbial activity and vegetation growth. Changes in the duration of this period are of critical importance for northern regional ecosystems. Results of implementation of an alternative area-averaging routine (using Thiessen polygons) are presented in Table 4 and Fig. 5 to illustrate the possible spread when a different approach to area averaging is used.

Fig. 5.

(top two rows) Regionally averaged long-term mean annual number of days (a) with unfrozen ground (UF categories 0–2); (b) frozen ground (FSWS categories 3–4); (c) remnants of snow cover (RSC category 5); and (d) snow on the ground (SC categories 6–9). (bottom) As in top but results of the alternative regional averaging (using the Thiessen polygons method) for two major groups, UF and SC.

Fig. 5.

(top two rows) Regionally averaged long-term mean annual number of days (a) with unfrozen ground (UF categories 0–2); (b) frozen ground (FSWS categories 3–4); (c) remnants of snow cover (RSC category 5); and (d) snow on the ground (SC categories 6–9). (bottom) As in top but results of the alternative regional averaging (using the Thiessen polygons method) for two major groups, UF and SC.

b. Trends for the former Soviet Union (1956–91) and Russia (1956–2004)

Figures 7 –12 present linear trends for periods when most of the station data are available (1956–91 for the former Soviet Union and 1956–2004 for Russia, respectively), for the annual and spring (April–May) number of days with SC, UF, and FSRSC. Each figure presents four maps. The upper-row maps provide trend estimates within 1° × 1° grid cells while the lower-row maps provide the same trend estimates but area averaged within 11 regions. The nationwide trend estimates (for 1956–91 for the former Soviet Union and 1956–2004 for Russia) are presented in Table 5. The regional estimates colored in purple are statistically significant at the 0.05 significance level. Left-column maps show estimates for the 1956–91 period for the entire former Soviet Union (trends are expressed in days per 36 yr), while the right-column maps show estimates for the 1956–2004 period for Russia (trends are expressed in days per 49 yr).

Fig. 7.

Linear trend in annual number of days with snow cover. The regional estimates colored in purple are statistically significant at the 0.05 significance level. The estimates for the western region of the former Soviet Union south of 60°N include all stations of the former Soviet Union in the left column but only Russian stations in the right column (where they are marked by asterisks).

Fig. 7.

Linear trend in annual number of days with snow cover. The regional estimates colored in purple are statistically significant at the 0.05 significance level. The estimates for the western region of the former Soviet Union south of 60°N include all stations of the former Soviet Union in the left column but only Russian stations in the right column (where they are marked by asterisks).

Fig. 12.

As in Fig. 7 but for spring (April–May) number of days with frozen ground, ice, and/or remnants of snow cover.

Fig. 12.

As in Fig. 7 but for spring (April–May) number of days with frozen ground, ice, and/or remnants of snow cover.

Table 5.

Nationwide changes in three major characteristics of state of the ground during the period with major data availability (1956–91 and 1956–2004 for the former Soviet Union and Russia, respectively). Estimates are based on linear trends calculated for the regionally averaged variables over 11 regions shown in Fig. 2 that were thereafter averaged with weights proportional to areas of the regions within the former Soviet Union and Russia, respectively. Estimates of change per analyzed period are presented in days and in relative values (percent of long-term mean values for the 1961–90 period). R2 is a fraction of the variance ascribed by linear trends. Statistically significant estimates at the 0.05 or higher levels are shown in bold numbers. For each season, region, and period of time the algebraic sum of changes is equal to zero (but deviates sometimes from zero due to rounding errors).

Nationwide changes in three major characteristics of state of the ground during the period with major data availability (1956–91 and 1956–2004 for the former Soviet Union and Russia, respectively). Estimates are based on linear trends calculated for the regionally averaged variables over 11 regions shown in Fig. 2 that were thereafter averaged with weights proportional to areas of the regions within the former Soviet Union and Russia, respectively. Estimates of change per analyzed period are presented in days and in relative values (percent of long-term mean values for the 1961–90 period). R2 is a fraction of the variance ascribed by linear trends. Statistically significant estimates at the 0.05 or higher levels are shown in bold numbers. For each season, region, and period of time the algebraic sum of changes is equal to zero (but deviates sometimes from zero due to rounding errors).
Nationwide changes in three major characteristics of state of the ground during the period with major data availability (1956–91 and 1956–2004 for the former Soviet Union and Russia, respectively). Estimates are based on linear trends calculated for the regionally averaged variables over 11 regions shown in Fig. 2 that were thereafter averaged with weights proportional to areas of the regions within the former Soviet Union and Russia, respectively. Estimates of change per analyzed period are presented in days and in relative values (percent of long-term mean values for the 1961–90 period). R2 is a fraction of the variance ascribed by linear trends. Statistically significant estimates at the 0.05 or higher levels are shown in bold numbers. For each season, region, and period of time the algebraic sum of changes is equal to zero (but deviates sometimes from zero due to rounding errors).

These figures and table clearly show that since 1956, a large-scale redistribution among three states of the ground—SC, UF, and FSRSC—has occurred. The annual duration of the period with unfrozen ground has increased over all of Russia, showing statistically significant increases from 4 to 9 days per 49 yr within five of eight regions (Table 5; Fig. 8) and a nationwide increase of 5 days (which is above the standard deviation of this quantity equal to 4 days). The same figure and table for the former Soviet Union indicates that most of these changes have occurred prior to the early 1990s. This increase was compensated by a reduction in two “cold land” states of the ground—SC and FSRSC (note that algebraic sums of changes in these three states is zero for each region, season, and period of time, although rounding in Figs. 7 through 12 and Table 5 sometimes contaminate the sums by one unit). The annual numbers of days with snow cover and days with frozen ground and remnants of snow on the ground are decreasing, although only nationwide over the former Soviet Union is the SC retreat statistically significant. An annual reduction of the number of days with frozen ground and remnants of snow on the ground, FSRSC, has been more pronounced than for SC. For four of eight regions in Russia, this reduction was statistically significant, and nationwide a 3-day or 10% decrease in annual FSRSC was documented for the past 49 years (Fig. 9, Table 5: this is above the standard deviation of this quantity equal to 2 days). For Russia, SC and FSRSC changes were coherent on an annual time scale in six of eight regions balanced by changes in UF. As a result, in some regions we observed insignificant changes in SC duration (e.g., in the Arctic and Far East north of 55°N) but a statistically significant redistribution of the state of the ground conditions between two other classes. One important conclusion from these results is that changes in SC are less noticeable than changes in two other state of the ground characteristics. Snow cover is exactly what is being observed from space (cf. section 5a), and the in situ data analyzed here show that it is less sensitive to changes7 characteristic of the surface conditions.

Fig. 8.

As in Fig. 7 but for annual number of days with unfrozen ground.

Fig. 8.

As in Fig. 7 but for annual number of days with unfrozen ground.

Fig. 9.

As in Fig. 7 but for annual number of days with frozen ground, ice, and/or remnants of snow cover.

Fig. 9.

As in Fig. 7 but for annual number of days with frozen ground, ice, and/or remnants of snow cover.

Spring changes (April–May, Figs. 10 –12) mirror those in annual duration but relative changes in spring are much stronger (Table 5). In particular, it appears that in southern Siberia, the entire decrease in SC (by 3.4 days per 49 yr) (i) was due to the changes in these two months; (ii) was statistically significant at the level higher than 0.01; and (iii) represented a 30% decrease in the April–May regional long-term mean SC value.

Fig. 10.

As in Fig. 7 but for spring (April–May) number of days with snow cover.

Fig. 10.

As in Fig. 7 but for spring (April–May) number of days with snow cover.

Figure 13 shows annual and summer (June–September) variations in UF and FSRSC days for two regions of Russia with the longest snow season duration, the Arctic and Far East north of 55°N (240 and 210 days per year, respectively). While the time series are shown for the 1936–2004 period, a green dashed divider line indicates the starting year of the time series for which linear trend estimates were presented in the figure. Since 1956, statistically significant changes in the Russian Arctic were an increase in the annual number of days with unfrozen ground by 4 days (or 4%) and decreases in the annual and summer number of days with frozen soil, ice, or remnants of snow on the ground by 3 days each (or 15% and 19%, respectively). Statistically significant changes in the Russian Far East north of 55°N were increases in the annual and summer number of days with unfrozen ground by 9 days (or 7%) and 3 days (or 3%), respectively, and decreases in the annual and summer number of days with frozen soil, ice, or remnants of snow on the ground by 5 days (or 17%) and 2 days (or 27%), respectively. Only statistically significant trend estimates (days and % per 49 yr) are outlined above for annual and summer time series. In these two regions, only one statistically significant negative trend in SC was detected for the Russian Far East north of 55°N in the summer time (a 1-day or 44% decrease per 49 yr).

Fig. 13.

Number of days without snow cover in two regions of northern Russia: the Arctic (north of 66.7°N) and Russian Far East north of 55°N. Annual (solid lines) and summer season (June–September, dotted lines) day counts are shown for days with unfrozen soil (black lines) and with frozen soil and/or remnants of snow on the ground (red lines).

Fig. 13.

Number of days without snow cover in two regions of northern Russia: the Arctic (north of 66.7°N) and Russian Far East north of 55°N. Annual (solid lines) and summer season (June–September, dotted lines) day counts are shown for days with unfrozen soil (black lines) and with frozen soil and/or remnants of snow on the ground (red lines).

c. Trends for 1936–2004 for the Russian Federation: Stability of the estimates

Initially, we were disinclined to present and to analyze results based on the first 20 years of data (1936–55) because of the network paucity. However, these decades are of particular interest in high latitudes because this period is associated with the first warming of the Arctic in the twentieth century (Vinnikov et al. 1990; Polyakov et al. 2003; Lugina et al. 2005). Moreover, in several earlier analyses (Ye et al. 1998; Ye 2001; Heino and Kitaev 2005) an increase in snow depth over the former Soviet Union (Russia) was reported for periods that began in 1936. The following analysis was performed to test the reliability of the sparse network available during these two decades. We performed the same calculations described in the previous section for the Russian stations included in the long-term network (Fig. 3) and compared the results of regional averaging with those based on our complete network. Particular attention was paid to the Arctic (north of the Polar Circle 66.7°N) and Russian Far East (south of 55°N) regions where for the entire 69-yr-long period statistically significant increases in the number of days with snow cover were found (12 and 8 days per 69 yr, respectively). Figure 14 shows that the restricted and complete networks produce time series that vary in close proximity to one another with R2 for the post-1956 period varying from 0.84 in the Russian Far East south of 55°N to 0.97 in West Siberia (in the Arctic north of the Arctic Circle, R2 = 0.94). The first two decades should not be considered in this comparison because the time series were constructed from practically identical sets of stations. Systematic differences in the Arctic between the two datasets are attributable to the northernmost and eastward stations in the complete network compared to the long-term network. The similarity of the last 49 years (Fig. 14) suggests that the sparse network successfully represents variability of the regionally averaged snow cover extent. One exception is the Russian Far East (south of 55°N) where, in the 1990s, an artificial positive trend in snow cover extent is found when a sparse long-term network is used. The complete network does not support this increase for the 1990s.

Fig. 14.

Annual number of days with snow on the ground in seven regions of the Russian Federation. Comparison of estimates based on complete (black lines) and long-term (red lines) datasets. Numbers of 1° × 1° grid cells with data available in each dataset are shown by dotted lines. Regional partition is shown in Figs. 1 and 5. Last plot in this figure shows linear trends of annual snow on the ground for the 1936–2004 period over the Russian Federation using the long-term dataset only.

Fig. 14.

Annual number of days with snow on the ground in seven regions of the Russian Federation. Comparison of estimates based on complete (black lines) and long-term (red lines) datasets. Numbers of 1° × 1° grid cells with data available in each dataset are shown by dotted lines. Regional partition is shown in Figs. 1 and 5. Last plot in this figure shows linear trends of annual snow on the ground for the 1936–2004 period over the Russian Federation using the long-term dataset only.

In the Arctic, the complete dataset includes 53 stations within fifty-two 1° × 1° grid cells. This is a twofold increase compared to the long-term network. Nevertheless, during the post-1956 period, snow cover extent variability and trends are well reproduced by a less populated network (R2 = 0.94). This allows more confidence in the Arctic snow cover extent variations during the first two decades. In these decades, when the Arctic was relatively warm, the duration of the period with snow cover was approximately 10 days shorter than in the 1990s. This causes a statistically significant increasing trend for the 1936–2004 period in the length of the period with snow on the ground in the Arctic (by 12 days per 69 yr, i.e., a 5% increase, which is statistically significant at the 0.01 level).

Our results for the 1936–2004 period summarized in the Fig. 14 map after nationwide averaging (Table 5) show an increase in SC over Russia by 5 days (or 3% of the long-term mean). Contrary to the 1956–2004 period, this change is mostly due to an autumn (October–November) redistribution between SC and FSRSC. These results support (or at least do not contradict) the conclusions by Heino and Kitaev (2005) relative to the increase in annual duration of days with snow on the ground during the 1936–2000 period over northern Eurasia. Furthermore, our results provide (i) reinforcement to these findings from an independent source of information and (ii) a justification for the use of a relatively sparse network as a representative source of information about the regional snow cover extent variations. We realize that our network (Fig. 2) itself is quite sparse in the Arctic and northern Siberia. To assess the representativeness of such a network, special studies of the covariance of the spatial distribution of snow cover are needed (e.g., Groisman et al. 2005) that are beyond the scope of this paper (cf. however, section 5a).

4. Intradaily changes of the state of the ground as a function of daytime temperature

During the 1951–76 period, across the former Soviet Union an average 5.3% of days during the year8 had an afternoon observation that was different from the morning observation. Most of these situations are associated with snowmelt below the 50% threshold of areal coverage (∼23%), soil thaw (∼63%), or both (Tables 6 and 7).9 The changes associated with soil thaw (conversion from the frozen state and/or from snow-covered state to unfrozen ground conditions) are separated in Fig. 15 and aggregated in Fig. 16. These figures provide the total probability of change by season and depict conditional probabilities of seasonal switches from frozen to unfrozen conditions. Without assuming that the afternoon state of the ground changed compared to the morning state, the chance of obtaining an afternoon thaw as a function of daytime temperature is shown in Fig. 17. Figure 17 shows that the chance of unfrozen afternoon state of the ground is increasingly high with warmer temperatures and reaches 30% of the sample when the daytime surface air temperatures are in the range of 2° to 10°C. For these temperatures, the highest chance for a freeze-to-thaw change occurs when the morning ground is already warm and the switch to the unfrozen state has not yet happened, making the judgment based only on the morning observation invalid. These calculations can be further regionalized, blended with snow depth information available at the same locations, and used to correct in situ freeze–refreeze data based on once-per-day measurements and known time of the observation.

Table 6.

Conditional frequencies of different switches, from one state of the ground in the morning to another state of the ground in the afternoon (%, annual).

Conditional frequencies of different switches, from one state of the ground in the morning to another state of the ground in the afternoon (%, annual).
Conditional frequencies of different switches, from one state of the ground in the morning to another state of the ground in the afternoon (%, annual).
Table 7.

Conditional frequencies of different switches, from one state of the ground in the morning to another state of the ground in the afternoon (%).

Conditional frequencies of different switches, from one state of the ground in the morning to another state of the ground in the afternoon (%).
Conditional frequencies of different switches, from one state of the ground in the morning to another state of the ground in the afternoon (%).
Fig. 15.

Conditional probability of changes to unfrozen conditions when state of the ground has changed as a function of the daytime temperature. Conversions FSWS→UF and SSS→UF are presented by season (December–March, April–May, and October–November). Total probabilities of switches within the system (UF, FSWS, and SSS) by season are also shown. Approximately 80% of all these switches represent a switch to unfrozen soil conditions (i.e., changes shown in these two graphs).

Fig. 15.

Conditional probability of changes to unfrozen conditions when state of the ground has changed as a function of the daytime temperature. Conversions FSWS→UF and SSS→UF are presented by season (December–March, April–May, and October–November). Total probabilities of switches within the system (UF, FSWS, and SSS) by season are also shown. Approximately 80% of all these switches represent a switch to unfrozen soil conditions (i.e., changes shown in these two graphs).

Fig. 16.

Conditional probability of changes to unfrozen conditions when state of the ground has changed as a function of the daytime temperature. To construct this, diagram conversions FSRSC→UF and SC→UF were assessed by season (December–March, April–May, and October–November). Total probabilities of switches within the system (UF, FSRSC, and SC) by season are also shown. Approximately 80% of all these switches represent a switch to unfrozen soil conditions.

Fig. 16.

Conditional probability of changes to unfrozen conditions when state of the ground has changed as a function of the daytime temperature. To construct this, diagram conversions FSRSC→UF and SC→UF were assessed by season (December–March, April–May, and October–November). Total probabilities of switches within the system (UF, FSRSC, and SC) by season are also shown. Approximately 80% of all these switches represent a switch to unfrozen soil conditions.

Fig. 17.

Probability of changes to unfrozen soil conditions as a fraction of cases with a given daytime temperature. To construct this diagram conversions FRSC→UF and SC→UF were assessed against all other switches and cases when no changes within these three classes occur.

Fig. 17.

Probability of changes to unfrozen soil conditions as a fraction of cases with a given daytime temperature. To construct this diagram conversions FRSC→UF and SC→UF were assessed against all other switches and cases when no changes within these three classes occur.

5. Discussion

a. Comparison with remote sensing snow product

Comparisons of in situ and satellite snow cover observations have a long history. First, the in situ data were used to verify and calibrate the remote sensing products (e.g., Wiesnet et al. 1987). Then the joint analyses of both types of data were made to better study the long-term variability of snow cover extent (e.g., Brown 2000). One of the conclusions based on such analyses for level terrain was that a sufficiently dense network of station reports containing observations of the presence of snow on the ground provides an unbiased representation of the area of snow cover extent (e.g., Fig. 17 in Groisman et al. 2004). Therefore, it was not a surprise that the estimates based on in situ and remote sensing information are reasonably well correlated (Table 8; Fig. 18). We also detected systematic differences, for example, nationwide satellites report longer periods with snow on the ground than station data. The use of polygon area averaging for the long-term mean values (cf. section 2c) reduced the biases (Table 8) but does not eliminate them. The largest differences for obvious reasons are in mountainous regions of Central Asia in summer, where satellites report from 6% (2001) to 47% (1980) of the area as snow covered while stations report less then three days (and usually only one) with snow on the ground during the June–September season.10 However, for the Russian Federation, the use of an alternative area-averaging routine (Thiessen polygon averaging) produced the snow cover duration estimates for late spring (April–May) that coincide (on average) with the remote sensing product (Table 8).

Table 8.

Results of the nationwide comparison for the former Soviet Union and the Russian Federation of information about the days with snow on the ground from this study (in situ data) and satellite products (Groisman et al. 1994, updated). Time series are shown in Fig. 18. All correlations are statistically significant at the 0.01 or higher levels. For bias estimates (satellite − in situ, in days) in the last column, results of the alternative regional averaging (using the Thiessen polygons method) are presented in parentheses.

Results of the nationwide comparison for the former Soviet Union and the Russian Federation of information about the days with snow on the ground from this study (in situ data) and satellite products (Groisman et al. 1994, updated). Time series are shown in Fig. 18. All correlations are statistically significant at the 0.01 or higher levels. For bias estimates (satellite − in situ, in days) in the last column, results of the alternative regional averaging (using the Thiessen polygons method) are presented in parentheses.
Results of the nationwide comparison for the former Soviet Union and the Russian Federation of information about the days with snow on the ground from this study (in situ data) and satellite products (Groisman et al. 1994, updated). Time series are shown in Fig. 18. All correlations are statistically significant at the 0.01 or higher levels. For bias estimates (satellite − in situ, in days) in the last column, results of the alternative regional averaging (using the Thiessen polygons method) are presented in parentheses.
Fig. 18.

(top) Annual and (bottom) April–May snow cover duration over the former Soviet Union (dashed lines) and Russia (solid lines) as derived from satellite (red lines: Groisman et al. 1994, updated) and in situ observations (black lines; this study). Satellite estimates have been shifted to account for systematic biases (Table 7) by 16 days for annual estimates and in spring by 9 and 8 days for Russia and the former Soviet Union, respectively. Separately, the Eurasian snow cover extent (106 km2) (downloaded from http://www.cpc.ncep.noaa.gov/data/snow/) is shown for comparison (blue lines, right y axes).

Fig. 18.

(top) Annual and (bottom) April–May snow cover duration over the former Soviet Union (dashed lines) and Russia (solid lines) as derived from satellite (red lines: Groisman et al. 1994, updated) and in situ observations (black lines; this study). Satellite estimates have been shifted to account for systematic biases (Table 7) by 16 days for annual estimates and in spring by 9 and 8 days for Russia and the former Soviet Union, respectively. Separately, the Eurasian snow cover extent (106 km2) (downloaded from http://www.cpc.ncep.noaa.gov/data/snow/) is shown for comparison (blue lines, right y axes).

In Fig. 18, we present the nationwide (for the former Soviet Union prior to 1991 and separately for the Russian Federation) in situ time series of snow cover duration versus the estimates of this duration derived from the satellite data. To preserve the integrity with the time series presented in this paper, we chose to bias correct the satellite data for differences shown in Table 8. For comparison, in Fig. 18 the continental snow cover extent (106 km2) is also presented. After the unit conversion (snow cover extent to days with snow on the ground), the annual (spring) variability of remote sensing product resembles11 the variability of annual (spring) duration of snow on the ground area averaged over Russia and even better describes this variability over the former Soviet Union up to 1991 with R2 = 0.52 (0.51).

Brown (2000) blended in situ snow cover observations from a sparse network in Eurasia using satellite estimates of spring snow cover extent available after 1972. Using the available in situ data, Brown was able to reconstruct snow cover extent as far back as 1915. Our assessment shows that we can do this on a continental scale back to 1936 using the presently available data (Fig. 2). Keeping in mind that observations of snow on the ground exist prior to 1936 and are simply waiting to be rescued, efforts to extend snow cover time series analyses have a good chance of being successful.

b. Interpretation of trend estimates in context to warming in high latitudes

While the climatology of snow on the ground presented in this paper is relatively stable, changes reported for several periods considered in this paper (1936–2004, 1956–2004, and 1956–91) in the form of linear trends require further scrutiny and an assessment of their usefulness for meaningful interpretation. First, this relates to the analyses that ended in 1991. Keeping in mind that there were large changes during the past 50 years in the region (Groisman and Bartalev 2004; Figs. 1 and 19), the information relevant to the 1956–91 changes has to be updated. However, when the 1956–91 changes were compared with those in the 1956–2004 period for Russia (not shown), it was found that those 36 years corresponded to the period of the most intense changes in regional temperatures (http://www.ncdc.noaa.gov/oa/climate/monitoring/gcag/gcag.html; Fig. 19) as well as in the state of the ground including the snow cover extent (Figs. 7, 8, 9 and 18).

Fig. 19.

Mean annual temperature of the Arctic (latitudinal band north of 60°N) and northern Eurasia (north of 40°N, east of 15°E). Dashed lines indicate beginning years of trend analyses in this study. Data are from archive of Lugina et al. (2005, updated to 2005). Temperature time series are shown as anomalies from the reference period 1951–75.

Fig. 19.

Mean annual temperature of the Arctic (latitudinal band north of 60°N) and northern Eurasia (north of 40°N, east of 15°E). Dashed lines indicate beginning years of trend analyses in this study. Data are from archive of Lugina et al. (2005, updated to 2005). Temperature time series are shown as anomalies from the reference period 1951–75.

The 1956–2004 period had approximately the same mean rate of temperature changes as was observed during the past 120 years in northern Eurasia and in the Arctic zone. For the entire period from 1881 to 2004 (cf. Fig. 19; Lugina et al. 2005) an increase of 1.3 K in the Arctic temperature is observed (for the 1956–2004 period, the increase is 1.2 K). Both linear trend estimates are statistically significant at the 0.01 or higher levels. For northern Eurasia, increasing trends for the 1881–2004 and 1956–2004 periods show statistically significant increases of 1.3 and 1.4 K, respectively. In contrast, for the 1936–2004 period, no systematic increases in the Arctic can be reported due to the warm first two decades of this period (the linear trend estimate for this period is 0.1 K and is not statistically significant). The mean annual temperature in northern Eurasia in this period increased (on average), but at about two-thirds of the rate compared to the entire twentieth century and/or to its second half. Generally speaking, the above means that the relationship that could be implied from “simultaneous” linear trends of the temperature in high latitudes and snow cover extent over northern Eurasia for the 1956–91 (2004) periods cannot be based on the data for the 1936–2004 period.

One might assume that the warmer Arctic is responsible for a larger atmospheric water vapor content that leads to an increase in snowfall (Groisman 1991; Kattsov and Walsh 2000), snow depth (Ye et al. 1998), and snow cover extent (Fig. 14). However, the Arctic temperature changes (Fig. 19) do not support this assumption in our data. Therefore, another hypothesis can be provided for the observed snow cover extent changes since 1936:

  • The warming in the midtwentieth century (1930s–40s) was confined mostly to high latitudes north of 60°N particularly in the Atlantic sector of the Arctic (Arctic Climate Impact Assessment 2004). The warmer Arctic in the 1930–40s reduced the meridional temperature gradient causing more winter anticyclonic conditions in northern Eurasia, thus resulting in less precipitation and generally drier winters and springs. This notion is supported by direct estimates of changes in meridional surface air temperature gradient, which is linked to changes in atmospheric circulation indices that control anticlyclonic highs in the cold season over Eurasia (Vinnikov 1976; Groisman 1983) and by precipitation changes in interior regions of northern Eurasia (Vinnikov and Groisman 1979; Groisman 1991). All of the above could be used to explain the lower snow cover extent over the former Soviet Union in the midtwentieth century.

  • The development in the second half of the twentieth century is qualitatively different (Fig. 1). Areas of maximum warming are shifted from the Atlantic sector of the Arctic to the interior of the Eurasian continent and a direct warming effect (especially in the spring season, cf. Fig. 1) might outweigh dynamic factors and cause the increase in the number of days with unfrozen soil on the expense of the cold land conditions (i.e., SC and FSRSC).

This is a descriptive explanation that gives a noncontradictory picture of observed changes in snow cover extent in the region. Speculation on the reasons for the changes in Arctic temperatures and atmospheric circulation in the high latitudes is beyond the scope of the present study.

6. Conclusions

  • During the past 50 years (1956–2004 period):

    • The annual duration of the period with unfrozen soil conditions has increased everywhere east of the Yenissei River by 4 to 9 days with maximum changes in southern Siberia and the northeastern Far East (with a nationwide annual increase of 6 and 5 days for the former Soviet Union and Russia, respectively; Fig. 8; Table 5).

    • In most cases, statistically significant increases in the number of days with unfrozen soil conditions were due to a reduction of days with frost, ice, and remnant snow on the ground conditions rather than due to snow cover retreat (Figs. 7 –12).

    • In the last decade of the twentieth century all changes diminished.

  • During the past 69 years (1936–2004 period), an increase in duration of the period with snow on the ground over Russia and the Russian polar region north of the Arctic Circle has been documented by 5 days or 3% and 12 days or 5%, respectively. That is in agreement with other findings but cannot be associated with the “Arctic warming.” This warming was not apparent in this particular time interval (opposite of the past 120 yr and/or to the past 50 yr). This nonlinearity of the surface air temperature changes in high latitudes affects the ground condition changes. Linear trend analyses applied to the entire period of observations lead to these paradoxical conclusions. Trend analyses became popular in climate change studies because we tend to link them (sometimes unintentionally) with the global change that is apparent in century-long temperatures practically everywhere over the globe. But, for time intervals when the major warming tendency is not seen (e.g., 1936–2004 in the Arctic) additional analyses (or just precaution) are needed for interpretation of observed trends.

  • During the 1956–91 period, duration of the period with snow cover has decreased nationwide over the former Soviet Union by 3% (Table 5). The most significant decreases have occurred in Siberia and in the late spring (April–May nationwide spring decrease over the former Soviet Union of 11% and in central Siberia of 20%).

  • Satellite observations available since 1972 support major features of snow cover variations over the former Soviet Union and Russia. Therefore, remote sensing products can be used to expand the in situ observations into the twenty-first century while the in situ data may be used (i) to better calibrate the remote sensing product now and (ii) to extend it into the past.

  • Changes associated with daytime soil thaw and/or snow retreat are quite frequent with positive daytime temperatures and cannot be discarded when morning observations are used to characterize the entire daily state of the ground. Therefore, their estimates can be potentially used for a postprocessing correction of snow cover extent data based on once per day measurements and/or for better calibration of freeze–refreeze information delivered with higher temporal resolution. For example, having the information about the “daily” snow on the ground (based on morning-only observations) and temperature estimates, one could use a conditional probability for a “switch” for a less biased estimation of the surface energy budget components for the day.

Fig. 11.

As in Fig. 7 but for spring (April–May) number of days with unfrozen ground.

Fig. 11.

As in Fig. 7 but for spring (April–May) number of days with unfrozen ground.

Acknowledgments

NASA Grant GWEC-0000-0052, the NOAA Climate and Global Change Program (Climate Change and Detection Element), and INTAS Grant 01-0077 (SCCONE) provided support for this study. Thoughtful recommendations of three anonymous reviewers helped us to significantly improve the manuscript.

REFERENCES

REFERENCES
Alisov
,
B. P.
,
1957
:
Principles of climatic regionalization of the USSR (in Russian).
Izv. Russ. Acad. Sci. Ser. Geogr.
,
1957
,
6
.
118
125
.
Arctic Climate Impact Assessment
,
2004
:
Arctic climate system and its global role.
Impact of a Warming Arctic, G. McBean, Ed., Cambridge University Press, 21–60
.
Armstrong
,
R.
,
2001
:
Historical Soviet Daily Snow Depth Version 2 (HSDSD). National Snow and Ice Data Center, CD-ROM
.
Brown
,
R. D.
,
2000
:
Northern Hemisphere snow cover variability and change, 1915–1997.
J. Climate
,
13
,
2339
2355
.
Chang
,
A. T. C.
,
J. L.
Foster
,
R. E. J.
Kelly
,
E. G.
Josberger
,
R. L.
Armstrong
, and
N. M.
Mognard
,
2005
:
Analysis of ground-measured and passive-microwave-derived snow depth variations in midwinter across the northern Great Plain.
J. Hydrometeor.
,
6
,
20
33
.
Folland
,
C. K.
, and
T. R.
Karl
,
2001
:
Observed climate variability and change.
Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 99–181
.
Fung
,
I.
,
1997
:
A greener north.
Nature
,
386
,
659
660
.
Groisman
,
P. Ya
,
1983
:
On the change of certain atmospheric circulation characteristics during processes of global warming and cooling.
Sov. Meteor. Hydrol.
,
1983
,
11
.
26
29
.
Groisman
,
P. Ya
,
1991
:
Data on present-day precipitation changes in the extratropical part of the Northern Hemisphere.
Greenhouse-Gas-Induced Climatic Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier, 297–310
.
Groisman
,
P. Ya
, and
S. A.
Bartalev
,
Eds.
2004
:
Science plan. Northern Eurasia Earth Science Partnership Initiative, 273 pp. [Available online at http://neespi.org/science/science.html.]
.
Groisman
,
P. Ya
,
V. V.
Kukushkina
,
K. M.
Lugina
, and
A. V.
Mihailov
,
1986
:
On obtaining of century time series of mean zonal monthly surface air temperature (in Russian).
Trans. State Hydrol. Inst.
,
320
,
118
128
.
Groisman
,
P. Ya
,
T. R.
Karl
,
R. W.
Knight
, and
G. L.
Stenchikov
,
1994
:
Changes of snow cover, temperature, and radiative heat balance over the Northern Hemisphere.
J. Climate
,
7
,
1633
1656
.
Groisman
,
P. Ya
,
E. L.
Genikhovich
,
R. S.
Bradley
, and
B. M.
Ilyin
,
1997
:
Assessing surface–atmosphere interactions using Russian standard meteorological network data. Part II: Cloud and snow cover effects.
J. Climate
,
10
,
2184
2199
.
Groisman
,
P. Ya
,
R. W.
Knight
,
T. R.
Karl
,
D. R.
Easterling
,
B.
Sun
, and
J. M.
Lawrimore
,
2004
:
Contemporary changes of the hydrological cycle over the contiguous United States: Trends derived from in situ observations.
J. Hydrometeor.
,
5
,
64
85
.
Groisman
,
P. Ya
,
R. W.
Knight
,
D. R.
Easterling
,
T. R.
Karl
, and
G. C.
Hegerl
,
2005
:
Trends in intense precipitation in the climate record.
J. Climate
,
18
,
1343
1367
.
Heino
,
R.
, and
L.
Kitaev
,
2005
:
Snow cover changes over Northern Eurasia during the last century—INTAS project SCCONE in its final stage. BALTEX Newsletter, No. 7. International BALTEX Secretariat, Geesthacht, Germany, 10–11
.
Jones
,
P. D.
, and
A.
Moberg
,
2003
:
Hemispheric and large-scale surface air temperature variations: An extensive revision and an update to 2001.
J. Climate
,
16
,
206
223
.
Kabat
,
P.
, and
Coauthors
, and
Eds.
,
2004
:
Vegetation, Water, Humans and the Climate—A New Perspective on an Interactive System.
Springer-Verlag, 600 pp
.
Kagan
,
R. L.
,
1997
:
Averaging of Meteorological Fields.
Kluwer Academic, 279 pp. (Translated from Russian by L. S. Gandin and T. M. Smith.)
.
Karl
,
T. R.
,
P. Ya
Groisman
,
R. W.
Knight
, and
R. R.
Heim
Jr.
,
1993
:
Recent variations of snow cover and snowfall in North America and their relation to precipitation and temperature variations.
J. Climate
,
6
,
1327
1344
.
Kattsov
,
V. M.
, and
J. E.
Walsh
,
2000
:
Twentieth-century trends of Arctic precipitation from observational data and a climate model simulation.
J. Climate
,
13
,
1362
1370
.
Kidd
,
R.
,
K.
Scipal
,
Z.
Bartalis
, and
W.
Wagner
,
2004
:
A diurnal difference indicator for freeze-thaw monitoring from Ku band scatterometer applied within the Siberia II project. Proc. IGARSS Symp. (IEEE Int. Geoscience and Remote Sensing Society Symp.), Anchorage, Alaska, IEEE, 1671–1674
.
Kimball
,
J. S.
,
K. C.
McDonald
,
S.
Frolking
, and
S. W.
Running
,
2004
:
Radar remote sensing of the spring thaw transition across a boreal landscape.
Remote Sens. Environ.
,
89
,
163
175
.
Livezey
,
R. E.
, and
W. Y.
Chen
,
1983
:
Statistical field significance and its determination by Monte Carlo techniques.
Mon. Wea. Rev.
,
111
,
46
59
.
Lugina
,
K. M.
,
P. Ya
Groisman
,
K. Ya
Vinnikov
,
V. V.
Koknaeva
, and
N. A.
Speranskaya
,
2005
:
Monthly surface air temperature time series area-averaged over the 30-degree latitudinal belts of the globe, 1881–2004.
Trends: A Compendium of Data on Global Change, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy. [Available online at http://cdiac.esd.ornl.gov/trends/temp/lugina/lugina.html.]
.
Myneni
,
R. B.
,
C. D.
Keeling
,
C. J.
Tucker
,
G.
Asrar
, and
R. R.
Nemani
,
1997
:
Increased plant growth in the northern high latitudes from 1981 to 1991.
Nature
,
386
,
698
702
.
NCDC
,
2005
:
Data documentation for DATASET 9290c, Global Synoptic Climatology Network. C: The former USSR. Version 1.0, 15 pp. [Available online at http://www1.ncdc.noaa.gov/pub/data/documentlibrary/tddoc/td9290c.pdf.]
.
Polyakov
,
I. V.
,
R. V.
Bekryaev
,
G. V.
Alekseev
,
U. S.
Bhatt
,
R. L.
Colony
,
M. A.
Johnson
,
A. P.
Maskshtas
, and
J. D.
Walsh
,
2003
:
Variability and trends of air temperature and pressure in the maritime Arctic, 1875–2000.
J. Climate
,
16
,
2067
2077
.
Ramsay
,
B. H.
,
1998
:
The interactive multi-sensor snow and ice mapping system.
Hydrol. Process.
,
12
,
1537
1546
.
Ramsay
,
B. H.
,
2000
:
Prospects for the Interactive Multisensor Snow and Ice Mapping System (IMS). Proc. 57th Eastern Snow Conference, Syracuse, NY, Eastern Snow Conference, 161–170. [Available online at http://www.easternsnow.org/proceedings/2000/ramsay.pdf.]
.
Razuvaev
,
V. N.
,
E. G.
Apasova
, and
R. A.
Martuganov
,
1995
:
Six- and three-hourly meteorological observation from 223 USSR stations. ORNL/CDIAC-66, NDP-048, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, Oak Ridge, TN, 68 pp
.
Robinson
,
D. A.
, and
A.
Frei
,
2000
:
Seasonal variability of northern hemisphere snow extent using visible satellite data.
Prof. Geogr.
,
51
,
307
314
.
Robinson
,
D. A.
,
K. F.
Dewey
, and
R. R.
Heim
Jr.
,
1993
:
Global snow cover monitoring: An update.
Bull. Amer. Meteor. Soc.
,
74
,
1689
1696
.
Running
,
S. W.
,
J. B.
Way
,
K. C.
McDonald
,
J. S.
Kimball
, and
S.
Frolking
,
1999
:
Radar remote sensing proposed for monitoring freeze-thaw transitions in boreal regions.
Eos, Trans. Amer. Geophys. Union
,
80
.
213, 220–221
.
Schultze
,
E. D.
,
M.
Heimann
,
S.
Harrison
,
E.
Holland
,
J.
Lloyd
,
I. C.
Prentice
, and
D.
Schimel
, and
Eds.
,
2001
:
Global Geochemical Cycles in the Climate System.
Academic Press, 350 pp
.
Shakirzyanov
,
R. I.
, and
V. N.
Razuvaev
,
2000
:
History of changes in snow cover observation methodology over the Russian territory (in Russian).
Trans. RIHMI-WDC
,
167
,
41
57
.
Shver
,
Ts A.
,
1976
:
Atmospheric Precipitation over the USSR Territory.
(in Russian). Gidrometeoizdat, 302 pp
.
Thiessen
,
A. H.
,
1911
:
Precipitation averages for large areas.
Mon. Wea. Rev.
,
39
,
1082
1084
.
Vinnikov
,
K. Ya
,
1976
:
Spectrum of temporal-spatial variations of zonally-averaged mean monthly temperature of the Northern Hemisphere during the 1881–1969 period (in Russian).
Trans. Main Geophys. Obs.
,
374
,
95
120
.
Vinnikov
,
K. Ya
, and
P. Ya
Groisman
,
1979
:
An empirical model of the present-day climatic changes.
Soviet Meteor. Hydrol.
,
1979
,
3
.
18
27
.
Vinnikov
,
K. Ya
,
P. Ya
Groisman
, and
K. M.
Lugina
,
1990
:
The empirical data on modern global climate changes (temperature and precipitation).
J. Climate
,
3
,
662
677
.
Voeikov
,
A. I.
,
1949
:
Snow cover, its effect on soil, climate and weather and methods of investigations (in Russian).
Izbrannye Proizvedeniya (Selected Papers), Vol. 2, Gidrometeoizdat, 15–157
.
Wiesnet
,
D. R.
,
C. F.
Ropelewski
,
G. J.
Kukla
, and
D. A.
Robinson
,
1987
:
A discussion of the accuracy of NOAA satellite-derived global seasonal snow cover measurements.
Large Scale Effects of Seasonal Snow Cover, Publication 166, International Association of Hydrological Sciences, 291–304
.
Ye
,
H.
,
2001
:
Quasi-biennial and quasi-decadal variations in snow accumulation over northern Eurasia and their connections to the Atlantic and Pacific Oceans.
J. Climate
,
14
,
4573
4584
.
Ye
,
H.
,
H-R.
Cho
, and
P. E.
Gustafson
,
1998
:
The changes in Russian winter snow accumulation during 1936–83 and its spatial patterns.
J. Climate
,
11
,
856
863
.

Footnotes

Corresponding author address: Pavel Ya. Groisman, National Climatic Data Center, Federal Building, 151 Patton Avenue, Asheville, NC 28801. Email: Pasha.Groisman@noaa.gov

1

In different periods of time more information was available, but we used and/or recalculated all available information into codes of Table 1 for consistency reasons (cf. section 2b). The Russian Empire and thereafter the former Soviet Union had a stable and relatively dense meteorological network over most of the country for the past 100 years but unfortunately the element that we are analyzing has not been digitized prior to 1936.

2

Any trend analysis is a simplistic way to describe what has happened on average during the period under consideration. Linear trend analysis gives a mean rate of this change. It cannot be extrapolated and does not deliver the entire picture if there is a nonlinearity. There are always nonlinearities and this is always an approximation. Twenty years ago, trend analysis was considered as a trivial exercise, keeping in mind the above (especially, the uselessness of trend extrapolation attempts without physical justification). However, since that time a kind of such justification emerged: systematic low-frequency changes in the global Earth System due to changes in the atmospheric gases and aerosols composition. Attribution of these changes was left to modelers, but from the observational point of view, scientists began looking at the trends as a hint of manifestation of these systematic changes (Folland and Karl 2001). This justifies this obsolete method as a first step to report what has happened. In this paper, we give the mean rate of changes for the period in question in one number (mean slope, i.e., linear trend) and show figures with actual time series of regionally averaged quantities.

3

Sometimes the second shot of daily observations are ignored for other reasons (e.g., afternoon snow depth measurements by passive microwave sensors; Chang et al. 2005) or are unavailable (e.g., cloudiness prevented observations in the optical and near-infrared wave bands; cf. Robinson et al. 1993).

4

We found that the micrometeorological variability for categorized state of the ground characteristics is sufficiently high on the month-to-season time scale. Therefore, the use of the methods that assign to some of the sites very high weights of area averaging (as the Thiessen polygon averaging routine does for remotely located sites) leads to a reduced accuracy of the area-averaged time series (Kagan 1997).

5

Each point (grid cell) estimate of trends for most climatic variables for the past 50–100 years is condemned to be statistically insignificant because of “weather noise”, a high level of micrometeorological variability, and sometimes, due to nonlinearities of climatic changes. Outliers in these fields could be interpreted as a manifestation of “statistically significant” changes only if we prove a “field significance” of these outliers (Livezey and Chen 1983). Furthermore, very frequently the outliers in fact are due to specifics of the field (or its measurements) in a given point due to observational inhomogeneity (e.g., station relocation) or natural inhomogeneity (e.g., strong site-specific changes) because of the local change in land use (such as reforestation, urbanization, reservoir construction, etc.). In both cases, these outliers should be studied separately in order to locate and fix inhomogeneity in the data or even be removed from the dataset outright to allow further investigation of large-scale climatic changes that are not affected by these mishaps. Fortunately, the dataset analyzed in this study has a minimal amount of the above mishaps, the most serious of which have been described in section 2b.

6

The gridded snow cover product after 2002 is “much better” (Ramsay 2000) but not exactly the same as previous time series (D. Robinson, Rutgers University, 2005, personal communication).

7

Or it is more noisy, if a 15% higher variance (on average) for SC than for UF is taken into account.

8

The seasonal numbers are 6% for the December–March cold season, and 7.3% and 11.0%, respectively, for the April–May and October–November transient seasons.

9

Causes for the a.m./p.m. observation change can be also attributed to a cold front passage and/or snowfall in the daylight hours that freeze the ground in the afternoon or cover it with snow. Table 6 shows that from the entire set of more than 600 000 cases, only 4% can be attributed to cold front passages and about 10% to daytime snowfalls.

10

It would be unfair to associate all regional biases with the in situ data only. Remote sensing snow cover products may overestimate duration of the period with snow cover for different reasons (P. Romanov, NOAA NESDIS, 2005, personal communication).

11

In this case, “resemble” means that variances of the time series are statistically indistinguishable, and the measure of similarity is described by the square of the correlation coefficient, R2, shown in Table 8. Hydrological year was selected to avoid any significant autocorrelation in time series of snow cover extent (duration).