[Index page of Masuda]
The Annual Cycle of Snow Cover Extent over the Northern Hemisphere as Revealed
by NOAA/NESDIS Satellite Data
Geographical Reports of Tokyo Metropolitan University, 28,
113 - 132 (1993).
Updated HTML version, 9 February 2001.
(Affiliation updated 25 June 2003.)
(Affiliations are as of June 2003).
Frontier Research System for Global Change, Yokohama 236-0001, Japan.
School of Commerce, Meiji University, Tokyo 168-8555, Japan.
Deceased in June 2001, formerly at Graduate School of Environmental Earth Science, Hokkaido University, Sapporo 060-0810, Japan
(also affiliated with Frontier Observational Research System for Global
Change, Yokohama 236-0001, Japan)
Center for Climate System Research, University of Tokyo, Tokyo 153-8904,
(also affiliated with Frontier Research System for Global Change,
Yokohama 236-0001, Japan.)
Abstract: The hemispheric distribution of the timing of appearance
and disappearance of snow cover is obtained. Weekly digital data of snow
cover for 25 years based on NOAA satellite observations are used. In plain
areas, we find zonal pattern in median despite of patchy pattern in individual
years. The phase lines are in general parallel to latitudinal circles,
but there is considerable east-west gradient both in Eurasia and North
America. Mountainous areas are characterized by late snowmelt, large variability
Key words: Snow cover, snowmelt, northern hemisphere, NOAA satellite
Snow cover is one of the most prominent features of the earth that vary
seasonally. It affects the heat balance of the earth in several ways, as:
Continental snow cover is also one of the major players in interannual
variation of the climate system, together with sea water temperature and
monsoon rainfall (Yasunari, 1991). Hahn and Shukla (1976) showed correlation
between snow cover area over Eurasia and monsoon rainfall over India in
the following summer. Since then, many studies have been conducted. They
are reviewed in Morinaga and Yasunari (1993).
Because of its high albedo (reflectivity), it absorbs less solar radiation
than bare soil or vegetated surface.
Melting snow acts as heat sink, and it keeps the ground temperature near
0 degree Celcius despite of diurnal variation of radiative fluxes.
In middle to high latitudes, snow cover is stock of water substance almost
immobile during wintertime, and source of soil moisture during springtime.
The wetness of soil makes the surface heat balance of summertime considerably
The seasonal variation of snow cover is much larger than the interannual
one. We should have better understanding of seasonal variation and then
we can perhaps better understand interannual anomalies too. We begin with
a primitive idea that the "front" of snow cover advances from higher latitude
to lower latitude, and that it retreats oppositely. We hope that we can
draw isopleths of the timing of advance and retreat. Such studies for one
continent or another have been published (Anonymous, 1960; Potter, 1965).
They are based on ground observations. In this paper, we try to describe
a standard seasonal cycle of snow cover in the hemispheric scale using
weekly data of satellite observations.
A preliminary report of this study was published as a researh grant
report (Masuda et al., 1989). We used 14 years' (1973 - 1986) data there.
In this paper (as published in 1993), the time period of analysis is extended
to 17 years (1973 - 1989), and the analysis procedure is improved as discussed
in Section 3. In this HTML version, the time period is 25 years (1973 -
We used the weekly values of "Weekly digital northern hemisphere snow and
ice product" compiled by NOAA (National Oceanic and Atmospheric Administration)
NESDIS (National Environmental Satellite, Data and Information Service)
of the United States. This is the data set of hemispheric-scale snow cover
with the longest history that starts in November 1966. Recently, the data
set is updated by Climate Prediction Center (CPC) which is part of National
Centers for Environmental Prediction of NOAA, rather than NESDIS.
This data set contains information of existence (1) or non-existence
(0) of snow cover at each of the 89 x 89 square grid boxes on a polar stereographic
map that covers most part of the northern hemisphere. It is a digitized
version of hard-copy charts. The hard- copy charts are created by manual
analysis of satellite imageries - mainly those of visible channels. The
satellites and sensors used for the production are listed in a table of
Matson et al. (1986) and reproduced in Wiesnet et al. (1987). The digitization
of data for 1966 - 1980 was done by Dewey and Heim (1982) and that for
the following years is done annually within NESDIS. (This digital data
set contains information of sea ice until 1980 but it does not since 1981).
The size of a grid box is roughly 200 km x 200 km.
Data from the autumn of 1966 were avaliable to this study. Until 1971,
data of some weeks were missing. Also, the sensor used until 1972 had lower
spatial resolution (3 km) than newer VHRR and AVHRR (1 km) sensors (Matson
et al., 1986). Considering these facts, we excluded these earlier years
The method of production of this data set and its problems are discussed
in the paper of Masuda and Morinaga (1990), based on Wiesnet et al. (1987)
and other sources. Here, a few points that are relevant to this study are
It just contains information about existence of snow cover, not about its
No information can be obtained if a grid box is cloudy every day during
a week. In such a case, the same condition as the previous week is assumed.
If multiple grid boxes are partially covered by snow, and if the sum of
the areas is equivalent to one box, the analysts subjectively determine
which box they should assign `1'.
3. Method of Analysis
In this paper, we calculate statistics at each grid box, and view the distribution
of the statistic values geographically. It is contrary to the more popular
approach that first calculates regional snow cover areas and that second
analyzes its time-series.
The problems of the NOAA/NESDIS data set discussed above, in particular
items (2) and (3), result in considerable statistical noise in time-series
analysis at single grid boxes. We try to overcome them by using robust
statistics as discussed below, and by paying attention to spatial pattern
of the statistic values.
Definition of `snowmelt week' and `snowfall week'
At a certain place in a certain year, there may be multiple timings of
appearance and disappearance of snow cover. We focus on the timing of the
last disappearance, because we consider it more important in controling
summer dryness of soil. For the appearance of snow cover, we pick up the
first one. This decision does not have an independent rationale. We just
wanted to treat the appearance and disappearance symmetrically. Our convention
is different from that of the figures on Page 42 of USSR Agriculture Atlas
(anonymous, 1960), where the first and last dates of `stable' (i.e. temporally
continuous) snow cover are shown.
We define the `snowmelt week' of a grid box of a year as the week when
snow cover is lastly observed between Week 8 and Week 30. (This definition
is slightly different from that adopted by the previous report of Masuda
et al., 1989).
We define the `snowfall week' as the week when snow cover is observed
at the first time between Week 36 and Week 52. It actually corresponds
to appearance of snow cover rather than just snowfall, but we use the short
name for brevity.
Use of order statistics
When a batch of numerical values is sorted, the value that comes at the
center is called the median. The value whose order is one fourth from the
smallest [largest] is called the lower [upper] quartile, and the difference
between the two quartiles is called the quartile range. It is known (e.g.
Mosteller and Tukey, 1977) that these order statistics are more robust
to outliers (extraordinary values) than the average and the standard deviation.
In this paper, medians and quartile ranges are calculated using 25 years'
data. Here, the median is the 13th largest case, and the quartile range
is the difference between the 7th largest and the 19th largest cases.
In our previous report (Masuda et al., 1989), we showed means and standard
deviations of snowmelt and snowfall weeks. Then we had to exclude such
cases from the numerator and denominator where the grid box is always snow-covered
or always snow-free during the period under examination. In this study,
if it is always snow-covered [snow-free], an arbitrary large [small] number
is assigned as flag values to preserve the order. Similarly, for `snowfall
week', an arbitrary small [large] number is assigned in such cases. With
this convention, such terminal cases are taken into account properly.
Amendment of data
After writing the previous report (Masuda et al., 1989), we made following
additional revisions to the data.
The data tape from NESDIS contained a land-sea template file used by Dewey
and Heim. Time-series plot of 0 / 1 values at each box revealed that the
template used since 1981 must be a little different from it. In this study,
only such grid boxes that are treated as land in both templates are taken
The data of Week 17 of year 1987 were damaged, where 0 and 1 are partly
(but not completely) reversed. We obtained a hard-copy chart of that week
and digitized it.
In some weeks between 1983 and 1997, there are physically inplausible line-like
patterns, such as a belt of 0's traversing Greenland. We corrected them
subjectively. It is likely to be noise of scanning hardware, and there
may be still other similar errors that cannot be subjectively corrected.
4. Preliminary Facts
As background information, the orographic height of each grid box is shown
in _Fig. 1. It is based
on the GLOBE (Global One-kilometer-Base Elevation) data set (Hastings and
Dunbar, 1999). In our 1993 paper, we used 1o x 1o
grid-box data set of Gates and Nelson (1975) with some smoothing. Names
of major mountain ranges and other places discussed later are shown in_Fig.
2 (with the smoothed orography based on Gates and Nelson 1975).
Seasonal cycle of total snow cover area
The time series of weekly values of total snow cover area is shown in _Fig.
3. The annual cycle is one decimal order-of-magnitude larger than interannual
_Fig. 4 shows the same data
differently. For each week of year, there are 25 values corresponding to
25 years (1973 - 1997). Maximum (1st), upper quartile (7th), median (13th),
lower quartile (19th) and minimum (25th) values from the largest are selected
for each week and connected. The curves do not correspond to the seasonal
march of any single year. The range between the uppermost and the lowermost
curves shows the whole range of sample values. The half of the sample values
are contained within the range covered by the second and the fourth curves
(the quartile range).
In a year, snow cover area reaches maximum at Week 1 - 8 (January to
February), and minimum at Week 30 - 36 (about August). The growth in autumn
season is about 1.5 times faster than the decay in spring. This feature
of the seasonal cycle is qualitatively the same as documented earlier by
Dewey and Heim (1982).
Frequency of snow cover
The `frequency', or empirical probability, of snow cover is the ratio of
occurrence of 1's to the total count of samples in each grid box. Maps
of geographic distribution of snow-cover frequency are presented by Matson
et al. (1986), using data for 1967 - 1981. Our results are not much different
_Fig. 5 shows the frequency
of the whole year. Cross-hatches denote the areas where there is always
snow cover. Such areas are limited to the interior of Greenland. Values
larger than 90 % are found in Greenland, Canadian Arctic Archipelago, and
in limited parts of Himalayas and northern Rockies. The areas where no
snow cover is observed at all are left white. Most of the area to the north
of 30 oN experience snow cover.
_Fig. 6 shows the snow cover
frequency of Weeks 1 - 8, that is the season where the area attains the
maximum. Most of the area in the higher latitudes has frequency larger
than 90 %. The boundary of this zone is approximately the 45oN
parallel, but it is shifted northward of this latitude in the eastern part
of each continent: i.e. in western and central Europe and in the `Great
Plains'. To the north of this boundary, it can be assumed that stable snow
cover exists continuously in winter.
5. Geographical Distribution of `Snowmelt Week' and `Snowfall Week'
_Figs. 7 (a) and (b) are examples
of the distribution of `snowmelt week' in individual years._Fig.
8 shows the median, and_Fig.
9 shows the quartile range._Figs.
10 (a) and (b) are examples of the distribution of `snowfall week'
in individual years. _Fig.
11 shows the median, and _Fig.
12 shows the quartile range.
Characteristics of maps for individual years
Both snowmelt and snowfall in a week occur in irregular patches. The size
of patches that correspond to a single week is much larger in snowfall
than in snowmelt. The typical spatial scale of snowmelt patch in one week
is about 1000 km, while that of snowfall patch is about 3000 km or more.
We consider it rational because snowfall is often caused by extra-tropical
cyclones that moves about 7000 km in a week, while snowmelt is caused essentially
by local heat balance. However, it may be partly artifact of the data set.
When snow falls it is inevitably cloudy, and it is difficult to detect
the accurate timing of snow cover set-up by observation of visible radiation.
The patterns are considerably different from year to year. We comment
on just a few examples. The advance of snow cover in Eurasia from the Arctic
coast to 45oN took only 4 weeks in 1976 (Fig.
10(a)) while it took 10 weeks in 1977 (Fig.
10(b)). Snow cover existed in a large area in the western Siberia and
eastern Europe as late as Week 17 in 1987 (Fig. 7(b))
while it melted much earlier in 1975 (Fig. 7(a))
Different nature of plain and mountainous areas
As a result of stacking 25 samples, the patterns of median snowmelt and
snowfall weeks (Figs. 8 and 11)
look much more zonal in areas with elevation generally below 500 m. We
will call these areas "plain areas". The patterns are still irregular in
mountainous areas (above 1000 m). The areas with elevation between 500
m and 1000 m can be considered as having transitional nature between plains
In plain areas, the isopleths are parallel to latitudinal circles as
the first approximation, though the east-west gradient is evident over
Europe and over the Great Plains of North America. The quartile range (Figs.
12) is generally less than 5 weeks in
In mountainous areas, large values of the quartile range are often found.
Some mountain areas, however, do not show large quartile range but are
rather characterized by late snowmelt and early snowfall (in median as
shown in Figs. 8 and 11).
The characteristics in plain areas and mountainous areas will be discussed
more in Sections 6 and 7, respectively.
6. Characteristics of Median `Snowmelt Week' and `Snowfall Week' in Plain
Speeds of north-south propagation of snowmelt and snowfall
When we compare Figs. 8 and 11,
we find some assymmetry between snowmelt and snowfall. As we have already
seen with Fig. 4, snowfall proceeds more rapidly
than snowmelt. We also see at Fig. 11 that in
median large advance of snow cover occurs in a single week: Week 45 in
North America and Weeks 40 and 44 in Eurasia.
In the following analysis, we take such grid boxes where both the median
`snowmelt week' and the median `snowfall week' are defined (i.e. not the
terminal flag values) and plot the week numbers against latitude. In_Fig.
13, we select the area of Eurasia, to the north of 40oN,
and the longitude being between 30oE and 120oE. In
this area, the isopleths can be regarded roughly parallel to the latitudinal
circles. We will first look at the grid boxes with elevation less than
500 m (+'s and o's in Fig. 13 together). `Snowmelt'
proceeds from 45oN to 70oN in 15 weeks, thus the
speed is about 1.7o latitude / week. `Snowfall' proceeds oppositely
in the same interval in 10 weeks, the speed being 2.5o latitude
/ week. Accelerated `snowfall' around Week 44 is also evident.
Dependence on elevation
The samples are sorted by the elevation of land. Difference between the
250 m - 500 m group (o's in Fig. 13) and the 0
m - 250 m group (+'s) is larger in snowmelt (about 4 weeks) than in snowfall
(about 2 weeks). The samples with elevation higher than 500 m are also
shown as small dots in Fig. 13. The distribution
is much more irregular than o's and +'s and it is likely to be affected
by regional orographic effects.
General east-west gradient
It is known that in middle and high latitudes in winter, western side of
a continent tends to be warmer than eastern side. An easy explanation of
this general trend is as follows:
Concerning snow cover here, too, the western side experiences earlier snowmelt
and later snowfall than the eastern side in both Eurasia and North America.
The slant of isopleths is evident in Figs. 8 and
in Europe and Western Siberia as well as in plain areas of North America.
The phase propagation is very roughly 1 to 2 weeks / 15o of
longitude, though it is not constant as discussed below.
In winter, the interior of a continent is cooler than oceans;
Accordingly, the surface air pressure is higher;
By near-geostropic balance with pressure, the winds tend to be northerly
in the eastern side and southerly in the western side.
Regional east-west gradient in Eurasia
The constant east-west difference in Europe and Western Siberia breaks
at the Ural mountain range at 60oE. Comparing distributions
of symbols in _Fig. 14,
we find that the belt of 45 - 60oE experience anomalously late
snowmelt. The western side is a hilly region and the eastern side is plain
area, but the difference of average elevation is less than 250 m. It is
likely that the differnce of snow depth is the controling factor. The annual
maximum snow depth is generally larger than 60 cm to the west of the Urals,
with a maximum of 80 cm over the mountain range. It is smaller to the east,
in particular, less than 40 cm to the south of 60oN. (Anonymous,
1960; Kotlyakov, 1968 quoted by Lydolph, 1971; Igarashi, 1992). It is likely
that prevailing westerly winds are forced to uplift at the mountain range
and result in large precipitation in windward and become drier in leeward.
The slanted zonal pattern appears to stop at 90oE. At 90oE,
snowmelt is later than other places with similar latitude and elevation.
There is a maximum of snow depth here to the west (windward) of the Central
Siberian Highlands. The snowmelt / snowfall pattern to the east of this
longitude is complicated, presumably due to complex orography.
The east-west gradient is again evident near the Pacific coast as shown
in_Fig. 15. Snowmelt is as
much as 8 weeks later in the grid boxes of northern Japan and Sahalin (42
- 52oN, 135 - 150oE) and Kamchatka (53 - 60oN,
150 - 165oE) than the majority of the 30oE - 120oE
zone. Grid boxes to the east of 120oE and to the south of 60oN
also have later snowmelt though the difference is smaller. In these areas,
which are near oceans, snow depth are larger than the continental interior
of eastern Siberia. It is likely that sea ice in Ohotsk and Bering Seas
has also some influence, but our knowledge is insufficient to discuss it
here. We just mention the fact that the seasonal cycle of sea ice has a
different phase from that of snow cover. The total sea ice area of the
northern hemisphere reaches maximum in March and minimum in September in
the northern hemisphere (Parkinson and Cavalieri, 1989).
Regional east-west gradient in North America
Snowmelt is rapid in the `Great Plains' region at the eastern slope of
the Rocky Mountains, while it is much slower in the longitudes of the
Great Lakes and the Hudson Bay._Fig.
16 shows the difference in snowmelt is as large as 4 weeks / 15o
longitude, while the difference in snowfall is not so large. In this case,
too, difference of snow depth is a likely cause of part of the large difference.
According to Canada Department of Transport (1967) quoted by Bryson and
Hare (1974), annual sum of snowfall is 80 to 120 cm in the `Great Plains'
region of Canada and 200 to 300 cm in the area between the Great Lakes
and the Hudson Bay. Influence of sea ice and lake ice in the eastern region
may also be suggested, though we have not examined it yet.
7. Characteristics in Mountainous Regions
An attempt of categorization
Comparing Figs. 8, 9, 11
and 12 with the orography (Fig.
1), mountainous regions are characterized by
By `large variability', we mean both large values of the quartile range
and large spatial inhomogeneity in the median. We often find these two
features together. We consider that the coincidence is probably real. But
it is possible to be an `aliasing' effect of this data set (problem (3)
mentioned in Section 2). In case when there are snow patches of the scale
smaller than a grid box, variability in judgements of the analysts who
produced the digital product can result in apparent temporal variability.
late snowmelt and early snowfall, or
large variability of snowmelt and snowfall seasons.
In this report (HTML version), we present a tentative categorization
of all areas with snow cover (not only mountains) in _Fig.
17. This figure is made by the following process. Using Fig.
8, the grid boxes where median `snowmelt week' is Week 22 or later
are marked as `late snowmelt'. Also, using Figs. 9
and 12, the grid boxes where the quartile range
of either snowmelt or snowfall week is 5 weeks or larger are marked as
`large variability'. Grid boxes are classified according to the two criteria
Thus, this figure is made by an objective process, though the values of
thresholds are determined subjectively. The areas which is always snow-covered
in median year, except permanent ice, are added to the category `4'.
1: Early snowmelt, small variability
2: Early snowmelt, large variability
3: Late snowmelt, large variability
4: Late snowmelt, small variability
Category `1' (painted orange in in Fig. 17)
corresponds to `plain areas' except Arctic region.
Category `4' (painted light blue in in Fig. 17)
includes most of the regions around the Arctic Sea irrespective of land
elevation. This category also includes part of Alaska Range, Canadian and
U.S. Rockies in North America, Pamir Highlands and part of Scandinavia,
Caucasus (Kavkaz), Himalayas, Altay and Sayan Mountains and Stanovoi Highlands
in Eurasia. Most of these are very high mountains or mountains in higher
Category `3' (painted green in Fig. 17) includes
Canadian Rockies and Sierra Nevada in North America, part of Himalayas,
Tian Shan, Caucasus and Alps in Eurasia. It is certain that pixels of this
category includes parts with higher elevation that should be included in
Category `4' if the data have higher spatial resolution. It is possible
that the remaining parts of the pixels have the same characteristics as
Category `2' and therefore Category `3' may be just an illusion due to
coarse spatial resolution. Whether this suggestion holds cannot be decided
based on available data.
Category `2' (painted yellow in Fig. 17) includes
most of U.S. Rockies in North America, and Tibetan Plateau, part of Mongolian
Plateau, and Zagros Mountains (in Iran) in Eurasia. The common factors
of these areas are mountains in relatively low latitudes and low supply
of water vapor which causes precipitation. In eastern Europe, this category
extends to plain areas near the Baltic Sea. Explanation of this feature
is not available yet.
Remark about Stanovoi Highlands
In the Stanovoi Highlands (about 55oN, 115oE, to
the east of Lake Baikal), snowmelt occurs as late as Week 23 to 25 (June)
in median. The smoothed height does not reach 1000 m, though the highest
peak is nearly 3000 m high. The late snowmelt is confirmed by ground-based
studies: anonymous (1960) also shows that in this region `stable' snowcover
exists until the third 10-day period of May, and that the total duration
of snowcover is 240 - 280 days per year. Though we cannot discuss causal
relationships, this area is also known as the southernmost part of the
continous permafrost zone of the world (e.g. Pewe, 1983).
8. Concluding Remarks
In plain areas, average elevation being less than approximately 500 m,
we find zonal pattern of the progress and retreat of seasonal snow cover
by stacking 25 years' data. The phase lines are in general parallel to
latitudinal circles, but there is east-west gradient both in Eurasia and
North America, with considerable regional anomalies.
Mountainous areas, elevation generally being more than 1000 m, are characterized
by late snowmelt, large variability or both. A tentative categorization
is shown. It should be noted, however, that temporal variability may be
aliased as spatial variability, or vice versa, because of the nature of
the data set used.
We got information of snow depth from existing literature, because the
NOAA/NESDIS data do not tell it. We consider we should examine the distribution
of snow mass more throughly by combining microwave and ground-based data.
An attempt in this direction was made by our colleague, Igarashi (1992).
We should have discussed the cause of the patterns, but we are not able
to do it. The results of this study suggest that following subjects should
be more clarified.
Discussion of the signals of interannual variations is also an interesting
matter that is left to future studies.
Surface heat balance
Development of seasonal mean atmospheric pressure system
Development and movement of cyclones
Effects of mountain ranges on the distribution of snowfall amount
Behavior of sea ice and its interactions with snow cover
Behavior of permafrost and its interactions with snow cover.
We are greateful to many persons who encouraged this work.
The original data which we used for this HTML version are from Data
Support Section, National Center for Atmospheric Research, USA (for 1966
-- 1991) and from CPC (for 1988 -- 1998). Dr. Wataru Morishima (at Tokyo
Institute of Technology) combined these two series and made them available
The data used for the version published in 1993 were from Satellite
Data Services Division (SDSD) of NOAA/NESDIS. We acquired the data for
1973 - 1987 from SDSD. Dr. Akio Kitoh (Meteorological Research Institute,
Japan Meteorological Agency) provided the data for 1988 and 1989 which
he obtained from SDSD.
(*: in Japanese; **: in Japanese with English abstract; ***: in Russian)
Anonymous (1960): Atlas Sel'skogo Hozyaistva SSSR (Atlas of Agriculture
of the USSR), Glavnoe Upravlenie Geodezii i Kartografii, Moskva, 308
Bryson, R. A. and Hare, F. K. (eds.) (1974): Climates of North America.
Survey of Climatology, 11, Elsevier, 420 p.
Canada Department of Transport, Meteorological Branch (1967): Climatic
Charts. Climatological Division, Toronto. (cited by Bryson and Hare,
Dewey, K.F. and Heim, R.H. Jr. (1982): A digital archive of northern hemisphere
snow cover, November 1966 through December 1980. Bull. Amer. Meteorol.
Soc., 63, 1132 - 1141.
Gates, W.L. and Nelson, A.B. (1975): A New (Revised) Tabulation of the
Scripps Topography on a 1o Global Grid. Part 1: Terrain Heights.
Rand Corporation, R-1276-1-ARPA, 132 p.
Hahn, D.G. and Shukla, J. (1976): An apparent relationship between Eurasian
snow cover and Indian monsoon rainfall. J. Atmos. Sci., 33,
2461 - 2462.
Hastings, D.A. and Dunbar, P.K. (1999): Global Land One-kilometer Base
Elevation (GLOBE): (Digital Elevation Model Version 1.0; Documentation
Version 1.0). NGDC Key to Geophysical Records Documentation No.
34, 138 p. National Geophysical Data Center, Boulder, Colorado, USA.
Igarashi, H. (1992): Yurasia Tairiku ni okeru Sekisetu no Eisei Kikogaku
(Satellite Climatology of Snow Cover over the Eurasian Continent),
Master's Thesis, Graduate Course of Geoscience, Univ. of Tsukuba, 76 p.
Kotlyakov, V. M. (1968): Snezhnyuy Pokrov Zemli i Ledniki (Snow cover
on the Earth and Glaciers). Gidrometeoizdat, Leningrad, 479 p. (cited
by Lydolph, 1971). ***
Lydolph, P. E. (1977): Climates of the Soviet Union. World Survey of
Climatology, 7, Elsevier, 443 p.
Masuda, K., Morinaga, Y., Numaguti, A. and Ouchi, A. (1989): Kita-hankyu
sekisetu hihuku no kisetu henka (Seasonal variation of snow cover over
the northern hemisphere). In Masuda, K. (ed.) Tairiku Kibo no Tihyomen
Zyotai no Kisetu Henka no Kaiseki (Analysis of Seasonal Variation of Land
Surface Conditions in Continental Scales), Report for Grant-in-aid
for Scientific Research FY 1988 No. 62540301, 12 - 35. *
Masuda, K. and Morinaga, Y. (1990): Mesh data sets of snow cover of the
northern hemisphere. J. Geography (Tokyo), 99, 695 - 703.
Matson, M., Ropelewski, C.F. and Varnadore, M.S. (1986): An Atlas of Satellite-derived
Northern Hemispheric Snow Cover Frequency. NOAA Atlas NOAA/NESDIS/NWS,
Morinaga, Y. and Yasunari, T. (1993): Koiki sekisetu ni okeru taiki seppyo
sogo sayo (Atmosphere-cryosphere interactions of large scale snow cover),
Kenkyu Noto (Meteorological Research Notes), No. 177, 41 - 129. *
Mosteller, F. and Tukey, J.W. (1977): Data Analysis and Regression.
Addison-Wesley, 588 p.
Parkinson, C.L. and Cavalieri, D.J., 1989: Arctic sea ice, 1973 - 1987:
Seasonal, regional and interannual variability. J. Geophys. Res.,
14499 - 14523.
Pewe, T.L. (1983): Alpine permafrost in the contiguous United States: A
review. Arctic and Alpine Research, 15, 145 - 156.
Potter, J.G. (1965): Snow cover. Canada Department of Transport, Meteorological
Branch, Toronto, Climatol. Stud., 3, 69 p. (cited by Bryson
and Hare, 1974).
Wiesnet, D.R., Ropelewski, C.F., Kukla, G.J. and Robinson, D.A. (1987):
A discussion of the accuracy of NOAA satellite-derived global snow cover
measurements. In Goodison, B.E., Barry, R.G. and Dozier, J. (eds.) "Large
Scale Effects of Seasonal Snow Cover", IAHS Publications, 166,
291 - 304.
Yasunari, T. (1991): The Monsoon year - A new concept of the climatic year
in the tropics. Bull. Amer. Meteorol. Soc., 72, 1331 - 1338.
HTML version updated 2001-02-09, 2003-06-25
Frontier Research System for Global Change, Yokohama 236-0001, Japan
_[Top of this text]
[Index page of Masuda]