1
2
3
4
Change in spring snowmelt timing in Eurasian Arctic rivers
Amanda Tan1, Jennifer C. Adam2 and *Dennis P. Lettenmaier1
5
6
1. Department of Civil & Environmental Engineering, University of Washington, Department
7
of Civil and Environmental Engineering, University of Washing ton, Wilson Ceramic Lab,
8
Box 352700, Seattle, WA 98195, USA.
9
10
2. Department of Civil & Environmental Engineering, 405 Spokane Street, Sloan 101, PO
Box 642910, Washington State University, Pullman WA 99164-2910, USA.
11
12
*Correspondence to: Dennis P. Lettenmaier, [email protected]
13
1
14
Abstract
15
Changes in the amount and timing of the discharge of major Eurasian Arctic rivers have been
16
well documented, but whether or not these changes can be attributed to climatic factors or to the
17
construction of manmade reservoirs remains unclear. Here we endeavour to identify the key
18
processes (snow cover and air temperature) that have regulated seasonal streamflow fluctuations
19
in the Eurasian Arctic over the last half century (1958-1999), and to understand the regional
20
coherence of timing trends, using a set of Eurasian Arctic rivers selected specifically because
21
they have minimal regulation (dam construction) effects. We find a shift toward earlier onset of
22
spring runoff ranging from modest (26 of 45 stations) as measured by an index of spring pulse
23
onset, to strong (39 of 45) by a centroid of timing index. Winter streamflows increased over the
24
period of record in most rivers, suggesting that observed trends in larger regulated Eurasian
25
Arctic rivers may not be entirely attributable to reservoir construction. Upward trends in air
26
temperature appeared to have had the largest impact on spring and summer flows for tributaries
27
in the coldest of the major Eurasian Arctic river basins (e.g., the Lena). While the overall
28
duration of snow cover has not significantly changed across the Eurasian Arctic, snow cover
29
disappearance has trended earlier in the year and appears to be related to the increased May and
30
snowmelt season fractional flows.
31
2
32
1. Introduction
33
Pronounced land surface process changes have occurred in the Arctic in recent decades. Surface
34
air temperatures generally have risen while precipitation has remained largely unchanged (Yang
35
et al., 2002; Shiklomanov et al., 2007; Adam and Lettenmaier, 2007); satellite data show that
36
average snow cover extent has decreased, especially in spring and summer (Comiso et al., 2002;
37
Foster et al., 2006). Modest changes in accumulated spring snowpack can lead to substantial
38
runoff changes during the late spring and early summer snowmelt period, affecting the seasonal
39
distribution of streamflow. Over approximately the same period, the discharge of Eurasian Arctic
40
rivers have increased, mostly in winter (McLelland et al., 2006; Adam et al., 2007; Shiklomanov
41
et al., 2007). Despite anecdotal relationships among these changes, they have occurred
42
contemporaneously with the construction of large dams on many of the major tributaries of the
43
largest Eurasian Arctic rivers over the last 50 years, and the extent to which observed Arctic
44
river discharge trends are attributable to climatic change, as contrasted with river management
45
effects, remains unclear (Yang et al., 2002; Adam et al., 2007; Peterson et al., 1998; McClelland
46
et al., 2002; Shiklomanov et al., 2001).
47
48
One implication of the ablation of snow cover earlier in spring is that since snow cover extent
49
has decreased, some of the energy that was once used to melt snow is now absorbed by the land
50
surface, thereby resulting in surface warming and advection of heat to surrounding snow covered
51
areas and thus leading to more snowmelt. Furthermore, some of this increased surface energy is
52
available to melt permafrost. Increased snowmelt and permafrost melt have been linked to
53
increases in Arctic river discharge that have the potential to affect the thermohaline circulation
54
and are postulated to interfere with the formation of the North Atlantic Deep Water and the
3
55
northward-flowing Gulf Stream (Peterson et al., 2002; Arnell et al., 2005; McLelland et al.,
56
2006). Dankers et al. (2004) found that a much shorter snow season, decreased sublimation and
57
increased evapotranspiration was occurring in the Scandivanian Arctic. As the snow-free season
58
is lengthened, surface albedo increases, especially during the part of the (former) snow cover
59
season when solar radiation is greatest.
60
61
Within the Eurasian Arctic domain, operation of reservoirs constructed over the last 50 years in
62
the Former Soviet Union have substantially affected the discharge of some of the major rivers.
63
In general, reservoir-related changes are evidenced primarily in streamflow seasonality
64
(increasing winter discharge and decreasing summer discharge, although significant shifts
65
towards earlier spring peak discharge have also been documented by McLelland et al., 2006;
66
Adam et al., 2007; and Shiklomanov et al., 2007). Recent satellite-based studies (e.g. Syed et al.,
67
2007) have also suggested that Arctic river discharge is increasing. They suggest that traditional
68
stream-gauge measurement data may not be capturing discharge trends accurately.
69
70
Adam and Lettenmaier (2007) suggest that observed river discharge increases in the colder
71
basins in the eastern Eurasian Arctic may be at least partially attributable to permafrost melt,
72
whereas for the relatively warmer western basins such as the Ob’, precipitation and
73
evapotranspiration (ET) changes may be responsible. Likewise, Yang et al., 2002 analyzed
74
monthly records of temperature, precipitation, streamflow and river ice thickness in the Lena
75
River basin and found increasing cold season streamflow and decline in river ice thickness with
76
shifts in seasonality of streamflow which they hypothesized to stem from changes in climate and
77
permafrost degradation. To date, however, causality for observed hydrologic trends and changes
4
78
in climatic forcings to the land surface system remains elusive, primarily because linkages
79
between hydrologic and climatic sensitivities are not well established. Uncertainty in climate
80
models, land surface models, and anthropogenic contributions are all barriers to accurate
81
diagnosis and prediction of the hydrology effects of climate change in the arctic regions.
82
83
We attempt herein to understand the degree to which snow cover and air temperature variability
84
and change influence the discharge of Arctic rivers. Snowpacks integrate the effects of climate
85
variability over the snow accumulation season; modest changes in spring snowpack can lead to
86
substantial runoff changes during snowmelt, affecting the seasonality of streamflow (Nijssen et
87
al., 2001; Yang et al., 2003). We seek to understand changes in streamflow timing and the
88
regional coherence of seasonal streamflow fluctuations in recent decades and to identify the key
89
processes that regulate streamflow timing changes. We also address the extent to which land
90
surface characteristics and hydrologic components might explain the observed streamflow
91
trends.
92
93
One obstacle to evaluating these interactions is that much of the arctic river discharge data that
94
are readily available and reasonably current, especially of smaller and intermediate sized rivers,
95
are archived as monthly accumulations, a time interval that is too long to evaluate changes
96
effectively. Smith et al., 2007 used a new R-ArcticNET (http://www.r-arcticnet.sr.unh.edu/v4.0)
97
daily observed discharge dataset to analyze minimum flows for small to medium sized,
98
unregulated stations across European Russia and the Ob’, Yenisei and Lena River basins for two
99
time periods: 1936 – 1999 (30 stations) and 1958 – 1989 (111 stations). Their study
100
demonstrated the difficulty in obtaining long-term, continuous daily discharge records.
5
101
This paper differs from previous analyses of Arctic river discharge change (e.g. McLelland et al.,
102
2004; Rawlins et al., 2004; Smith et a.l, 2007; Shiklomanov et al., 2007; Adam et al., 2007,
103
Rawlins et al., 2009) in its extensive use of land surface modeling to develop an adjusted daily
104
discharge dataset for all unregulated streamflow gauges in the Eurasian Arctic for a consistent
105
period of record 1958 – 1999. We use an approach that combines a hydrological model with
106
monthly discharge observations to infer continuous daily streamflow time series that allows us to
107
study streamflow timing changes over the Eurasian Arctic in more detail than has been possible
108
previously. Our specific objectives are: (1) link snowcover and temperature anomalies with
109
streamflow timing trends in order to determine how changes in air temperature affect spring
110
snowpack and hence the onset of the snowmelt period (2) detect timing changes attributable to
111
climate signals alone (3) analyze long term changes in seasonal streamflow of unregulated
112
Eurasian Arctic basins through the use of a continuous long-term daily adjusted streamflow
113
dataset and (4) develop relationships between snow cover extent and snow melt timing.
114
115
Because the stations we use were chosen (using methods described below) so as to reflect only
116
unregulated streamflow, we avoid the problem of confounding climate and water management
117
effects. The use of continuous, long-term daily adjusted streamflow data sets allowed us to
118
investigate changes in seasonal streamflow timing for unregulated stations, and to analyze long-
119
term changes in seasonal streamflow. Furthermore, through coincident analysis of satellite
120
records of snow cover extent, we develop relationships between snow cover extent and snowmelt
121
timing.
122
123
6
124
3. Selection of Streamflow, Temperature and Snow Cover Data
125
The runoff from the Lena, Yenisei and Ob’ Rivers combined represents 45% of the total riverine
126
flux of freshwater to the Arctic Ocean of about 3600 km³/year2. We focused on rivers within
127
these three large basins for this reason, and because they reflect a strong east to west temperature
128
gradient, with correspondingly differing permafrost extent and seasonal snow cover dynamics.
129
From a list of 1968 potential streamflow gauges on the R-ArcticNet database (http://www.r-
130
arcticnet.sr.unh.edu/) and RivDis12, 45 stations fulfilled our criteria of unregulated stations with a
131
long discharge record (1958 – 1999), and were the basis for our subsequent analysis
132
133
Since the late 1930s hydrological observations in the Siberian region, such as discharge, stream
134
water temperature, river-ice thickness, dates of river freeze-up and break-up, have been carried
135
out by the Russian Hydrometeorological Service. The observational records have been quality-
136
controlled and archived by the same agency. Some streamflow records begin in the early 19th
137
century, but many end in the early 1980s. Hydrological observations declined rapidly in the
138
1990s following the collapse of the former Soviet Union
139
140
From a list of 1968 potential streamflow gauges on the R-ArcticNet database that have recorded
141
data until at least 1999 within the three major river basins, we screened the monthly
142
uninterrupted, full-length records for the period of 1958 to 1999 and identified 45 stations. 80%
143
of the drainage areas are less than 75,000 km². The selection criteria for stations were that they
144
had to be uninfluenced by streamflow regulation, or nearly so. In order to choose stations that
145
met this restriction, it was imperative to select stations upstream of dams. This can be difficult to
146
ascertain especially in the Ob’ and Yenisei basins where many of the dams are relatively far
7
147
upstream. Dam information was collected from a combination of sources (ICOLD, 2003; Yang et
148
al., 2002; Shiklomanov et al., 2001). The gauges selected for these basins were a combination of
149
stations located in tributaries that appeared to be free of dams, and mainstream locations that
150
coincide with dam sites. In the latter case, we verified the location of stations through visual
151
inspection with the cooperation of Alexander Shiklomanov (University of New Hampshire).
152
153
Although we might have used one of several recently developed naturalized streamflow data sets
154
(Adam & Lettenmaier, 2007; Smith et al., 2006), we preferred to use directly observed flows in
155
our analyses. It should be noted that while a new daily discharge dataset is also now available on
156
R-ArcticNet Version 4.0 (Shiklomanov et al., 2007), it only includes a relatively small number of
157
stations that have long discharge records. To investigate timing changes over a larger spatial
158
domain, we developed a larger data set using the temporal disaggregation scheme described
159
below and compared the values obtained with several stations in the R-ArcticNet database.
160
161
Each monthly time series was used as the starting point for generation of daily streamflow
162
sequences as described below. We disaggregated the monthly observations to daily through use
163
of simulated streamflow output for each of the 45 gauged basins from the Variable Infiltration
164
Capacity Model (VIC), where the observed monthly flows were apportioned to daily values
165
using the ratio of the modeled daily to modeled monthly flows. Through this method we were
166
able to generate daily streamflow time series for all stations for the 42 year period 1958 – 1999.
167
The VIC model output was taken from Su et al. (2005), who performed 100 km (EASE grid)
168
simulations over the pan-Arctic domain.
Her simulations used gridded daily precipitation,
8
169
maximum and minimum temperature and wind speed, all downscaled to the model time step of
170
three-hours following methods outlined in Maurer et al. (2002).
171
172
VIC makes use of a routing model that is effectively a post processor that maps the combined
173
surface runoff and base flow to any point of interest in the basin (Bowling et al., 2002). Runoff
174
can exit a cell in only one of the possible eight flow directions and fluxes are summed using the
175
convolution integral. Using a combination of the monthly observed data and daily simulated
176
data from VIC, we developed daily adjusted streamflow data (DASD) for use in determining
177
streamflow timing changes at the 45 stations as follows. First, we obtained a ratio, Rs, of
178
observed monthly flow to simulated monthly flow at the specific VIC grid cell (with the latitude
179
and longitude specific to the streamflow gauges). We then multiplied the daily VIC-generated
180
streamflow at the same location by Rs to obtain the DASD.
181
182
In order to test this method, we compared the DASD developed with observed daily streamflow
183
for randomly selected gauges that had uninterrupted observed daily streamflow from 1958 –
184
1999 in the Ob’, Yenisei and Lena River basins (Table 1). Five observational gauges were
185
chosen over the entire domain for comparison with the disaggregation method.
186
187
Centroids of timing (CT) for these five gauges were calculated using both the simulated and
188
observed data and correlation coefficients were calculated for seasonal and annual flows. The
189
results show that there were significantly high correlations (p<0.1) for both the Lena and Yenisei
190
basins. For the Ob’ basin however, the correlation coefficients while still significant, are lower.
191
There are two explanations for this: Ob’ basin streamflow seasonality was not captured well by
9
192
the VIC model. This may be attributable to irrigation demands not being taken into account
193
(Adam et al., 2007). Large discrepancies occur even between reconstructed and observed
194
streamflows.
195
196
Surface air temperature data were obtained from the Global Historical Climatology Network
197
(GHCN) data base (Peterson et al., 1997, Serreze et al., 2000). Within the three large basins,
198
there were a total of 28 temperature stations with records that span the entire period 1958 – 1999.
199
Snow cover data from 3 October 1966 through 23 October 1978 were obtained from Northern
200
Hemisphere EASE-Grid Weekly Snow Cover and Sea Ice Extent Version 3 (Armstrong and
201
Brodzik, 1997). The snow cover extent is based on the digital NOAA-NESDIS Weekly Northern
202
Hemisphere Snow Charts and regridded to the EASE-Grid. Although the snow cover records
203
date to 1966, 1972 is the first year for which records over the Eurasian Arctic are complete.
204
205
4. Trend analyses
206
Long-term changes in streamflow characteristics were analyzed using three methods discussed
207
below. To examine trends over time in the timing, magnitude and duration of the snowmelt and
208
snow accumulation seasons, we defined two temporal windows. May and June are the months
209
with highest flows in the Eurasian rivers, with peak discharge at the mouths of the three major
210
rivers typically occurring in mid-June (Shiklomanov et al., 2007, Adam et al., 2007, Peterson et
211
al., 2000). A longer snow melt season (May 1 to August 31) was also defined for purposes of
212
analysis of snow cover ablation. A snow accumulation period was defined as September 1 –
213
March 1. Finally, a winter low flow season was defined as December- February.
214
10
215
The first streamflow timing characteristic we considered was spring pulse onset (SPO). SPO is
216
defined as the date of the beginning of snowmelt-derived streamflow when the cumulative
217
departure from the mean annual flow is at its minimum (Stewart et al., 2005; Cayan et al. 2001).
218
The day on which the mass of flow before and after is approximately equal is known as the
219
centroid of timing (CT). CT provides a time-integrated perspective of the timing of snowmelt
220
and represents the overall distribution of flow for each year. Monthly and seasonal fractional
221
flows (FF) are defined as the ratio of the streamflow that takes place in a given month or season
222
to the total streamflow in the water year. Spring season FF is the ratio of the streamflow
223
occurring in May and June, summer FF for July and August, and winter FF for December,
224
January and February.
225
226
Spring Pulse Onset (SPO) is the point when the cumulative departure from the mean annual flow
227
is minimum. In other words, it is the date of beginning of the spring or early summer snowmelt-
228
derived streamflow for snowmelt dominated rivers (hence the term “spring pulse onset”). Our
229
analysis assumes that the calculated day of minimum cumulative departure falls between 15
230
March and August 31 (i.e. between the end of winter season and the end of the summer flow
231
season).
232
233
Temporal trends were analyzed using the non-parametric Mann-Kendall test. Linear trends were
234
tested using both least squares linear regression and the non-parametric Mann-Kendall slope
235
(Helsel and Hirsch, 1992). Correlations tests between seasonal streamflows and snow cover were
236
analyzed using the Pearson’s “r” linear correlation coefficient. To test for statistical significance
11
237
of ‘r’ (being different from zero), the test statistic tr is approximately distributed as Student’s t
238
with n-2 degrees of freedom.
239
240
Surface air temperature data were tested for correlation with streamflow timing and magnitudes.
241
GHCN temperature stations were chosen to be within a 100 km radius of the drainage basin
242
centroid. This provided a representative temperature for each drainage basin. If more than one
243
GHCN station was located within the 50 km radius of the basin centroid, the observed
244
temperature data were aggregated using a distance-weighted algorithm with the nearby
245
streamflow gauges having more weight (inverse distance squared) than the farther stations
246
(Dudley and Hodgkins, 2005). The algorithm for aggregating the discharge data weighted the
247
values such that the sum of the weights for all the aggregated stations was one.
248
249
The snow cover data are given as presence-absence for 25-km EASE grids. Grid cells with more
250
than 50% snow cover were assigned a binary value of 1; otherwise the grid cell was assigned a
251
value of zero. This method of assigning snow cover extent, while ignoring subgrid variability,
252
provides the fastest and simplest method to determine snow cover disappearance and onset.
253
Limitations in the use of the NOAA-NESDIS Northern Hemisphere snow cover data have been
254
noted by several authors (Ye et al., 1998; Frei et al., 1999). However, in the absence of snow
255
water equivalent (SWE) data over the entire domain, we determined that it was best to make use
256
of snow cover extent.
257
258
The snowmelt season starts over the southern portion of the Eurasian Arctic around mid-March
259
with nearly complete snow cover depletion occurring between week 20 – week 23 (i.e. mid-
12
260
May) in all three basins. Established snow cover usually first occurs between mid-September and
261
mid-October based on NOAA satellite observations. The snow cover cycle was determined by
262
assigning a sequential series to the weekly data for each year (1 – 52, representing the number of
263
weeks in a year). Each week consists of a sequenced number of days (1 – 7, 8 – 14, etc). In that
264
way, a time series for 29 years for each grid cell was generated consisting of 1508 weekly snow
265
cover (binary) values. The leap year effect on the date of the vernal equinox was neglected. From
266
this data set, we mapped the binary snow cover extent to the basins and converted the values to
267
overall snow cover fraction and number of snow cover days. We then extracted four variables
268
described below which we used in our snow cover analysis.
269
270
The areal extent of snow cover is expressed as a fraction of basin area. It should be emphasized
271
that the binary measure of snow cover we used implies that only if a more than 50% of a grid cell
272
is covered by snow will it be considered to “contain snow”. Therefore, snow cover fraction
273
(SCF) of zero does not necessarily imply that there is no snow in the area. The SCA is then used
274
to determine the date of snow cover onset (SCO), date of start of snow cover disappearance
275
(SCD), snowmelt period (SMP) and snow-free duration (SFD) over the entire Lena, Ob’ and
276
Yenisei basins. The date assigned to each of the variables was the average day of the week in
277
which the event occurred, (the 4th day of the week) using the following definitions:
278
279
(1) SCO is defined as the week on which snow cover is first detected by the satellite as
280
determined by the computed fraction of the entire basin that is snow covered. It is
281
assumed that snow is present in the basin if the SCF is more than 0.02. SCO must also
13
282
meet a requirement that SCF is increasing for consecutive weeks, and that the date must
283
fall between weeks 31 – 46 i.e. sometime between August and November.
284
(2) SCD is determined to be the week on which the SCF for the basin falls below 0.98. This
285
date is constrained to be between weeks 13 – 26 and the SCF must be decreasing for
286
(two) consecutive weeks.
287
(3) Snowmelt period (SMP) is the time from SCD to the first week of snow-free conditions
288
i.e. SCF = 0.00. This period must fall between weeks 13 to week 28. SCD is an index to
289
the rate at which snow cover is depleting.
290
(4) Duration of snow free period (SFD) is the number of consecutive weeks for which SCF is
291
zero. If there is an occasion during which SCF is more than 0.00 during the snow free
292
period, it is disregarded if it the values are not consecutively increasing.
293
294
4.0 Results
295
This section presents the results of our analyses highlighting changes in streamflow timing where
296
we find that spring/summer flows are generally decreasing and winter flows are increasing. We
297
also present findings of correlations between streamflow with temperature and snow cover.
298
299
4.1 Streamflow timing
300
Analysis of the SPO time series for the 45 stations showed a 0 – 12 day shift earlier at 27 gauges
301
and a shift later (ranging from 2 – 4 days) at the remaining 18 stations (Fig. 1a). Due to the
302
considerable interannual variability in the SPO values, only 4 trends of the 27 shifts to earlier
303
SPO were statistically significant (p=0.10, a threshold used throughout our work), and none of
304
the 18 trends toward later SPO were. Although stations with trends toward earlier SPO were
14
305
located in all three basins, they tend to be primarily concentrated in the colder Lena and Yenisei
306
basins.
307
308
Overall annual flow timing, as quantified by the CT, generally showed shifts earlier in the water
309
year at a much larger fraction of stations than for SPO. 39 gauges had trends toward earlier CT
310
(between 1 – 24 days), 12 of which were statistically significant (Fig. 1b); 48% of those trends
311
were larger than 5 days. 6 gauges had trends toward later dates, 2 were statistically significant.
312
For 33 of the stations, CT was positively correlated with SPO (r = 0.1 – 0.6); however, the
313
correlation was statistically significant for only 10 of the 33 gauges. This suggests that SPO
314
alone may not explain the changes in CT. Among the plausible explanations for the shift in CT
315
are: (i) increased precipitation (more rain events) (McLelland et al., 2001; Adam and
316
Lettenmaier, 2007, Rawlins et al., 2009); (ii) modifications in frozen ground activity (melting of
317
permafrost especially during the winter seasons) (McLelland et al., 2001; Serreze et al., 2001)
318
and (iii) warmer spring temperatures causing more rapid snowmelt period even though SPO is
319
not occurring earlier (Yang et al., 2002; Rawlins et al., 2009). While changes in CT alone are
320
not necessarily related to timing of snowmelt, movement of both CT and SPO to earlier dates
321
suggests that snowmelt is occurring earlier in the year, and contributing to a peak annual
322
discharge that is likewise moving earlier into the year.
323
324
Trends in spring FF and in summer FF are additional indicators of possible shifts in the timing of
325
snowmelt runoff. May FF increased at 26 of the gauges (9 statistically significant) (Fig. 2a),
326
whereas June discharge decreased for 32 of the gauges (4 statistically significant). Statistically
327
significant trends were mostly at higher latitude stations, towards the eastern part of the region,
15
328
and mostly within the Lena basin (Fig. 2b). Overall, summer FF decreased at 25 of the gauges
329
with some of the stations experiencing up to 30% decreases in streamflow, particularly in Lena
330
River tributaries, where changes were most evident and more likely to be statistically significant.
331
26 of the gauges in the Lena showed a decrease in July streamflow, 6 of which were statistically
332
significant (Fig. 2d). The trend of decreasing flows in June and July appeared to be a
333
compensation for the increased May FF.
334
335
Increasing winter discharge trends were observed for 41 of the 45 gauges. The trends were
336
statistically significant for 12 of the 41 gauges and were distributed across the entire study
337
domain (Fig. 3). It is particularly noteworthy that these trends are similar in character to those
338
observed in larger rivers that have been affected by reservoir construction, suggesting that this
339
may not be the dominant cause of winter flow trends observed in previous studies (Shiklomanov
340
et al., 2007; Adam and Lettenmaier, 2007). Instead, increases in winter streamflow are
341
consistent with the hypotheses of permafrost melting, expedited by the conductive heat transfer
342
from decreasing snow cover (Barnett et al., 1989; Yang et al., 2007; Stieglitz et al., 2003; Zhang
343
et al., 2005).
344
345
4.2 Interseasonal Correlation
346
Correlations between SPO and spring FF, summer FF and winter FF, showed that SPO was often
347
correlated with spring FF, with statistically significant correlations at 21 of the 45 stations. The
348
snowmelt season centroid was mostly negatively correlated with summer flow, consistent with
349
previous studies that have found that winter and spring discharge increases are accompanied by
350
reductions in late summer and early fall streamflows (Yang et al., 2002; McClelland et al., 2006;
16
351
Adam et al., 2007; Smith et al., 2007). Spring flows were positively correlated with the winter
352
flow centroids for 11 of the gauges (7 statistically significant). This suggests that if spring FFs
353
are increasing, then it is likely that winter flows are also increasing. The link between these two
354
seasonal flows may be an increase in mean temperature as discussed below.
355
356
4.3 Correlation between Seasonal Streamflow and Temperature
357
We segregated the 45 streamflow gauge stations into their respective drainage basins to obtain a
358
clearer view of regional patterns of correlation between seasonal streamflow with monthly mean
359
observed temperature from the 28 GHCN stations. Temperature appeared to have a large impact
360
on spring and summer flows in the relatively cold Lena basin and the intermediate Yenisei basin,
361
but have little effect in the much warmer Ob’ basin (Fig. 4).
362
363
For the stations in the Lena basin, higher mean temperatures particularly in December and
364
January were associated with high winter season streamflow but the relationships were not
365
statistically significant. Higher mean spring and summer temperatures produced lower summer
366
discharge. For the Ob and Yenisei basins, the correlations were not statistically significant except
367
in June. Spring-summer temperature and fractional flows for the Lena and Yenisei basin were
368
positively correlated but for the warmer Ob’ basin, the relationship was inverse. This suggests
369
that temperature influences streamflow in the colder basins, but another mechanism (such as
370
changes in snow cover, precipitation and evapotranspiration) may play a larger role in the
371
warmer Ob’ basin. It is also possible that temperature influences the Ob’ discharge trends to
372
some degree but the effects are opposite of that for the Lena and Yenisei basins.
373
17
374
Recent studies further point to changes in precipitation which may in part explain the trends we
375
observed. Rawlins et al. (2009) found that winter precipitation was increasing particularly in
376
northern Eurasia, while during the warm season, discharge was lower. Likewise, there is also a
377
divergence between discharge trends and precipitation particularly for the warmer basins (such
378
as the Ob’) where evapotranspiration may be sensitive to both precipitation and temperature
379
changes.
380
381
4.3 Correlation between Seasonal Streamflow and Snow Cover
382
Trends in the four snow cover variables (SCD, SMP, SCO, and SFD) were also analyzed.
383
general, the overall number of snow cover days has not changed. From 1972 – 1981, SCD
384
decreased by about 1.2% per year for the Lena, 0.8% for the Yenisei and 1.5% for the Ob’, but
385
after this period, the number of snow cover days increases again. This is consistent with findings
386
of Comiso et al. (2002) who point to increasing snow cover over the Arctic in the late 80s to
387
early 2000. On average, snow cover begins to decline in the Lena basin at week 16, in the Ob’ at
388
week 13 and in the Yenisei at week 12. There is prolonged melt duration over the Yenisei of as
389
much as 14 weeks in order for the snow to completely disappear while for both the Ob’ and
390
Lena, the snow free season starts much earlier.
In
391
392
Studies point to earlier snowmelt associated with higher winter and spring temperatures (Serreze
393
et al., 2000; Zhang et al., 2001). We find that for all three basins, the week of snow cover
394
disappearance, SCD has moved earlier into the year (Table 3, Table 4). SMP, which is the
395
duration of the snow melt period, has been decreasing since the early 1980s for all three basins;
396
which may help explain the snow melt fractional flow trends. With exception of the Yenisei
18
397
basin, SCO is moving later into year, leading to longer duration of SFD except for 1992 – 1996,
398
when for all three basins, there was a decrease in the snow free period.
399
400
As shown in Table 3, SPO and SMP are significantly correlated for all three basins with the
401
highest correlation occurring in the Ob’ basin. Pronounced changes in timing of the SMP (i.e. the
402
rapidity of the snowmelt season) and changes in May fractional flow may account for the timing
403
changes in spring pulse and CT, where the bulk of annual flow is now occurring earlier in the
404
year due to the prompt melting of snow cover.
405
406
5.0 Conclusion
407
The overall implication of this study is that spring streamflows are moving earlier into the year,
408
prompted by an earlier snowmelt season. Winter flows are increasing in many parts of the
409
Eurasian Arctic. The rapidity of snow melt and a prolonged snow free duration and may be the
410
main contributors to spring and summer/peak flow timing trends. We find that while average
411
annual snow cover has decreased slightly from 1972 – 2001, this apparently does not account for
412
observed discharge trends – winter discharge increase in 75% of gauges, and June discharge
413
decrease for 82% of the gauges
414
415
Overall annual flow timing, as quantified by the center mass of flow or the centroid, has also
416
moved earlier into the water year. These measures indicate an earlier snow melt onset and
417
increased spring runoff. Likewise, trends in increased May flow, decreases in spring and summer
418
season flow and decreased June, July and August flows are compelling evidence that snowmelt
419
runoff is now occurring earlier in the water year. The most regionally coherent timing trends
19
420
occur in the eastern part of the Eurasian Arctic, particularly within the Lena basin, where
421
changes are most evident and more likely to be statistically significant. More importantly, there
422
are increasing winter streamflow trends in most of the basins studied and these trends are
423
distributed across the entire study domain (Figure 3). Runoff during the winter season in the
424
Arctic is usually small because the ground is frozen at such low temperatures. Although winter
425
flow trends for the large main stem rivers have been attributed to reservoir construction, the
426
streams we studied are unregulated, but nonetheless many show upward trends during this
427
season. An increase in winter streamflow is consistent with a hypothesis of increased
428
groundwater flow.
429
430
431
432
Acknowledgements
We thank A. Shiklomanov for comments and review on data. This work was supported by the
433
NASA Earth and Space Science Fellowship (NESSF) NNX08AU96H and The National Science
434
Foundation NSF Award #0629491.
435
436
20
437
438
439
References
440
streamflow products to streamflow trend attribution in Northern Eurasia. J. Climate 21(8), 1807-
441
1828 (2007).
Adam, J.C. & D.P. Lettenmaier, D.P. Application of new precipitation and reconstructed
442
443
Adam, J.C., Haddeland, I., Su, F. & Lettenmaier, D.P. Simulation of reservoir influences on
444
annual and seasonal streamflow changes for the Lena, Yenisei and Ob' Rivers. J. Geophys. Res.
445
112, D24114 (2007).
446
447
Armstrong, R. L., & Brodzik, M.J. Northern Hemisphere EASE-Grid weekly snow cover and sea
448
ice extent version 3. Boulder, Colorado USA: National Snow and Ice Data Center. Digital media
449
(2005).
450
451
Brown R. D. Northern Hemisphere snow cover variability and change, 1915–97. J. Climate 13,
452
2339–2355 (2000).
453
Barnett, T., Dümenil, L., Schlese, U., Roeckner,E. & Latif, M. The effect of Eurasian snow cover
454
on regional and global climate variations. J. Atmos. Sci., 46, 661–686 (1989).
455
Cayan, D. R. Interannual climate variability and snowpack in the western United States. J.
456
Climate, 9, 928–948 (1996).
457
458
Comiso, J. C. A rapidly declining perennial sea ice cover in the Arctic. Geophys. Res. Lett.
459
29(20), 1956 (2002).
460
21
461
Dudley, R. W. & Hodgkins, G. A.Trends in timing, magnitude, and duration of summer and
462
fall/winter streamflows for unregulated coastal river basins in Maine during the 20th century:
463
U.S. Geological Survey Scientific Investigations Report 2005-5021, 26 p. (2005).
464
465
Frei A. & Robinson D.A. Northern hemisphere snow extent: regional variability 1972-1994.
466
International Journal of Climatology 19, 1535-1560 (1999).
467
468
Groisman, P. Y., 2005: National Climatic Data Center Documentation for TD-9813, Daily and
469
sub-daily precipitation for the Former USSR, 16 pp.
470
471
Groisman, P. Y. and E. Y. Rankova, 2001: Precipitation trends over the Russian permafrost-free
472
zone: Removing the artifacts of pre-processing. International Journal of Climatology, 21,
473
657-678.
474
475
Helsel D. R. & Hirsch R.M. Statistical methods in water resources (Studies in environmental
476
science, Vol. 49) (Elsevier, New York, 1992).
477
478
McClelland, J. W. et al. A pan-Arctic evaluation of changes in river discharge during the latter
479
half of the 20th century. Geophys. Res. Lett. 33, L06715 (2006).
480
481
Peterson, B. J. et al. Increasing river discharge to the Arctic Ocean. Science 298, 2171-2173
482
(2002).
483
22
484
Peterson, T. C. & Vose, R.S. An overview of the Global Historical Climatology Network
485
temperature database. Bull. Amer. Meteor. Soc., 78, 2837–2848 (1997).
486
487
Rawlins, M.A., H. Ye, D. Yang, A. Shiklomanov, and K. C. McDonald. "Divergence in seasonal
488
hydrology across northern Eurasia: Emerging trends and water cycle linkages." Journal of
489
Geophysical Research 114 (2009): D18119.
490
491
Serreze, M. C. et al. Observational evidence of recent change in the northern high-latitude
492
environment. Clim. Change 46, 159–207 (2000).
493
494
Shiklomanov, A. I., Lammers, R.B., Rawlins, M.A., Smith, L.C. & Pavelsky, T.M. Temporal
495
and spatial variations in maximum river discharge from a new Russian data set. J. Geophys. Res.
496
112, G04S53 (2007).
497
498
Smith, L. C., Pavelsky, T.M, MacDonald, G.M., Shiklomanov, A.I. & Lammers, R.B. Rising
499
minimum daily flows in Eurasian suggest a growing influence of groundwater in the high-
500
latitude water cycle. J. Geophys. Res., 112, G04S47 (2006).
501
502
Stewart, I.T., Cayan, D.R. & Dettinger, M.D. Changes toward earlier streamflow timing across
503
western North America. J. Climate, 18, 1136–1155 (2005).
504
23
505
Stieglitz, M., S. J. Déry, V. E. Romanovsky, and T. E. Osterkamp (2003), The role of snow
506
cover in the warming of arctic permafrost, Geophys. Res. Lett., 30(13), 1721,
507
doi:10.1029/2003GL017337.
508
509
Su, F., Adam, J.C., Bowling, L.C. & Lettenmaier, D.P. Streamflow simulations of the terrestrial
510
Arctic domain. J. Geophys. Res., 110, D08112 (2005).
511
512
Vörösmarty, C.J., Fekete, B. & Tucker, B.A. River Discharge Database, Version 1.1 (RivDIS
513
v1.0 supplement). Available through the Institute for the Study of Earth, Oceans, and Space /
514
University of New Hampshire, Durham NH (USA) (1998)
515
516
Yang, D., Zhao, Y., Armstrong, R., Robinson, D. & Brodzik, M.J. Streamflow response to
517
seasonal snow cover mass changes over large Siberian watersheds, J. Geophys. Res. 112,
518
F02S22 (2007).
519
520
Yang, D. Q., Kane, D., Hinzman, L., Zhang, X., Zhang, T. & Ye, H. Siberian Lena river
521
hydrologic regime and recent change. J. Geophys. Res. 107(D23), 4694 (2002).
522
523
Ye, H., Cho H. & Gustafson, P.E. The changes in Russian winter snow accumulation during
524
1936– 83 and its spatial patterns, J. Clim., 11, 856–863 (1998).
525
526
Zhang, T. (2005), Influence of the seasonal snow cover on the ground thermal regime: An
527
overview, Rev. Geophys., 43, RG4002, doi:10.1029/2004RG000157.
24
Figures
(a)
(b)
Figure 1: (a) Trends in SPO (day of year marking the beginning of the snowmelt season) and (b)
trends in CT (the date when accumulated flow before and after is equal).
25
(a)
(b)
(c)
(d)
(e)
(f)
Figure 2: Monthly fractional flow linear trends for (a) April, (b) May, (c) June, (d) July, (e)
Snowmelt Season Fractional Flow (MJ), and (f) Summer Season Fractional Flow (JA).
26
(a)
(b)
(c)
(d)
Figure 3: Monthly FF (fractional flow) linear trends for (a) December, (b) January, (c) February,
and (d) Winter seasonal fractional flow (DJF).
27
Figure 4: Mean snowmelt and summer FF as a function of mean surface air temperature for the
months of April (Row 1), May (Row 2), June (Row 3), July (Row 4) and August (Row 5).
(Column 1: Lena, Column 2: Ob’, Column 3: Yenisei). Pearson’s R correlation coefficients are
indicated in each panel.
28
Tables
Table 1: Correlation coefficients for the centroid of timing (CT) of annual and seasonal flows
between the observed daily streamflow and the DASD method. Trends that are significant at
p<0.1 are highlighted in bold.
Basin
Lena
Yenisei
Ob
528
529
530
R-ArcticNET
Station
Namana at
Myakinda
(6224)
Aldan at
Tommot (6232)
Utulik
Bol’shoy Patom
Tara at
Muromtsevo
(7030)
Annual
Winter
0.795
0.673
0.703
0.908
0.895
0.816
0.822
0.681
0.849
0.913
0.781
0.851
0.721
0.828
0.421
0.801
0.885
0.873
0.985
0.899
0.643
0.602
0.852
0.510
0.529
Table 2: Pearson’s correlation coefficients between (column 2) SPO and SCD, and (column 3)
SPO and SMP. Trends that are significant at p<0.1 are highlighted in bold.
Basin
Lena
Ob’
Yenisei
531
532
533
534
Pearson Correlation Coefficients
Spring
Summer
Autumn
SCD
0.078
0.084
-0.088
SMP
0.348
0.600
0.501
Table 3: Linear trends of snow cover timing for the Lena, Yenisei and Ob’ basins. Trends in bold
are significant at p =0.10.
Basin
Lena
Yenisei
Ob’
SCO
-0.28
-1.01
-0.47
SCD
-0.88
-1.15
-0.59
SMP
0.78
1.16
1.40
SFD
-0.96
-0.86
-1.44
535
536
29