Permafrost hydrology in changing climatic conditions: seasonal variability of stable isotope composition in rivers in discontinuous permafrost

Role of changing climatic conditions on permafrost degradation and hydrology was investigated in the transition zone between the tundra and forest ecotones at the boundary of continuous and discontinuous permafrost of the lower Yenisei River. Three watersheds of various sizes were chosen to represent the characteristics of the regional landscape conditions. Samples of river flow, precipitation, snow cover, and permafrost ground ice were collected over the watersheds to determine isotopic composition of potential sources of water in a river flow over a two year period. Increases in air temperature over the last forty years have resulted in permafrost degradation and a decrease in the seasonal frost which is evident from soil temperature measurements, permafrost and active-layer monitoring, and analysis of satellite imagery. The lowering of the permafrost table has led to an increased storage capacity of permafrost affected soils and a higher contribution of ground water to river discharge during winter months. A progressive decrease in the thickness of the layer of seasonal freezing allows more water storage and pathways for water during the winter low period making winter discharge dependent on the timing and amount of late summer precipitation. There is a substantial seasonal variability of stable isotopic composition of river flow. Spring flooding corresponds to the isotopic composition of snow cover prior to the snowmelt. Isotopic composition of river flow during the summer period follows the variability of precipitation in smaller creeks, while the water flow of larger watersheds is influenced by the secondary evaporation of water temporarily stored in thermokarst lakes and bogs. Late summer precipitation determines the isotopic composition of texture ice within the active layer in tundra landscapes and the seasonal freezing layer in forested landscapes as well as the composition of the water flow during winter months.


Introduction
Numerous studies indicate that Northern Eurasia is experiencing increases in air temperature (Fedorov et al 2014), reductions in sea ice cover (Stroeve et al 2012), a lengthened ice-free period in lakes and rivers (AMAP 2011, Shiklomanov andLammers 2014), and landcover and ecosystem changes (Bhatt et al 2010, Callaghan et al 2011. Observations of river discharge from the six largest Russian rivers, all flowing into the Arctic Ocean (Northern Dvina, Pechora, Ob, Yenisei, Lena, and Kolyma), have shown a 7% increase in discharge over the 1936-1999 period (Peterson et al 2002). Recent studies showed an increase in the annual discharge of these six large Eurasian rivers to be 2.7 km 3 yr −1 or 210 km 3 over the period 1936-2010, with winter discharge contributing about 70% of this increase, i.e. 1.9 km 3 yr −1 or 146 km 3 (Shiklomanov and Lammers 2013). One peculiarity of Eurasian Arctic rivers is the presence of permafrost which underlies extensive areas of these watersheds. Over the last three decades, permafrost in the Eurasian Arctic has begun to experience warmer ground temperatures (Romanovsky et al 2010), and thickening of the active layer (Shiklomanov et al 2012) leading to decreased permafrost extent (Oberman andShesler 2009, Streletskiy et al 2014). Permafrost degradation has been associated as a potential cause of the observed increases in winter baseflow in some Arctic rivers (McClelland et al 2004), but permafrost degradation could also lead to increased groundwater storage and hence a reduction in river discharge.
One of the challenges to evaluating the effects of permafrost degradation, and the potential northward migration of discontinuous permafrost, on river flow is the large number of possible responses of the hydrological systems to changing permafrost conditions. For example, a recent study found that melting ground ice in discontinuous permafrost was not a source of observed changes in river discharge, but rather the thickening of the active layer was more significant to the changing watershed hydrology (Landerer et al 2010). Geochemical and isotope compositions of precipitation, groundwater, rivers, lakes and ground ice may help to distinguish between various sources of water in hydrological systems, assuming that the different sources have distinct stable water isotope compositions (Klaus and McDonnell 2013, Lacelle and Vasil'chuk 2013). Various mixing scenarios between different sources define the isotopic compositions of water in a river. For example, it was shown that the seasonal stable isotope composition of two small rivers in Northeastern Russia follow the isotopic composition of precipitation very closely (Huitu et al 2013). Several studies have revealed significant changes in the isotopic composition of river water during the spring snowmelt period, as snow cover has a lower isotopic composition relative to rain (Welp et al 2005, Sugimoto and Maximov 2012, Huitu et al 2013. Studies conducted in the Kolyma and Mackenzie river watersheds (Welp et al 2005, Yi et al 2010 were able to trace the effects of surface water evaporation in wetlands, bogs, and thermokarst lakes prior to entering the river flow in summer, while a study of the Lena river did not show that river flow was affected by evaporation of surface water (Sugimoto and Maximov 2012).
Although studies have investigated the effects of changing permafrost conditions on the hydrological regime at local (Quinton and Baltzer 2013), regional (Walvoord and Striegl 2007, St Jacques and Sauchyn 2009, Sjöberg et al 2013 and circumpolar scales (Yi et al 2012), studies of the stable isotopic composition of rivers from the Eurasian Arctic are rare, even though this region contains three of the four major Arctic rivers  and contributes 74% of the total terrestrial runoff to the Arctic Ocean (McClelland et al 2006). This study aims to: (i) determine the possible effects of changing climatic conditions on river discharge from 1970 to 2014 for watersheds situated near the town of Igarka in the lower Yenisei River region; and (ii) identify the sources of water contributing to 2013-14 river flow on seasonal scale. The study area is located at the transition between continuous and discontinuous permafrost on relatively shallow, ice-rich permafrost with temperatures close to the melting point. This makes the permafrost in this area sensitive to changes in climatic and environmental conditions, and offers the potential to evaluate the effects of changing permafrost conditions on hydrological regime of rivers and creeks draining to the Yenisei River.

Study area
The study area is located in the vicinity of the town of Igarka, 120 km North of the Arctic Circle, on the right bank of the Yenisei River (figure 1). The climate of the region can be classified as Continental subarctic climate (Dfc) (Peel et al 2007) with low winter temperatures and a relatively warm summer. The mean annual air temperature is −7.6°C for the 1973-2013 period. The coldest month is January (−27.6°C) and the warmest month is July (+15.0°C). The region receives 513 mm of precipitation annually, with 70 days of rain and 130 days of snow per year. Seasonal snow cover is established by early October and melts by mid-June. The regional average snow depth is 0.7-1.0 m, but can exceed 1.5 m in topographic depressions.
The region is located at the boundary between continuous and discontinuous permafrost and is characterized by Pleistocene and Holocene age permafrost reflective of the complex history of sedimentation and landscape development (Rodionov et al 2007, Flessa et al 2008. On the third Yenisei terrace at 50-60 m asl underlain by predominantly limnic fine-textured sediments, permafrost formed during the Zyryan epoch (Early Weichselian, 71 000-57 000 years BP, MIS 4 (Bassinot et al 1994)). During Kargino interstadial (57 000-24 000 years BP, MIS 3) widespread permafrost degradation occurred. Subsequent cooling during the Sartan stadial (Late Weichselian, MIS2, Last Glacial Maximum) resulted in syngenetic freezing of the coarse alluvial and fluvioglacial sediments of the second Yenisei terrace at 30-40 m asl (Astakhov and Mangerud 2007). The degradation of continuous Pleistocene permafrost during the Holocene Climatic Optimum was accompanied by the formation of numerous peat bogs and subsequent permafrost aggradation in organic-rich soils during the colder Holocene periods (Astakhov and Isaeva 1988). As a result two distinct permafrost layers are present in the area: deeper Pleistocene permafrost with its table at 3-5 m depth, and near-surface Holocene permafrost with its table at 0.4-1.5 m depth, depending on soil and vegetation conditions. The age-related depth to permafrost is relatively well identified by differences in landscape conditions (Tyrtikov 1964Rodionov et al 2007. The near-surface Holocene permafrost layer is present under moss-, grass-and sedge-dominated tundra landscapes formed on organic-rich fine lacustrine (limnic) sediments. Relatively dense taiga forests, formed on better-drained sandy and loamy soils, are dominated by larch (Larix Siberica), birch (Betula Pendula), and Siberian pine (Pinus Siberica). These landscapes, with the exception of steep Northfacing slopes, are underlain only by the deeper Pleistocene permafrost (Rodionov et al 2007).
The total permafrost thickness varies between 10 and 40 m depending on the presence of the near-surface Holocene permafrost layer (Kuznetsova et al 1985). Throughout the area permafrost temperature is around −1.0 to −0.1°C. The gravimetric ground ice content ranges from 35 to 40% in mineral soils to 60% and higher in peatlands, floodplains, and drained thermokarst lake basins (a.k.a alas depressions). Observations, conducted in a deep permafrost tunnel under the Igarka Permafrost Museum (http:// igarka-permafrostmuseum.ru/) show that between 3 and 20 m depth there is an extensive network of ice lenses formed in glacio-lacustrine sediments and reaching 0.3 m in thickness and 8 m in length (Shur and Zhestkova 2003).
Poor drainage and relatively high ground ice content of limnic tundra sediments promotes the development of thermokarst lakes that dot the landscape. The groundwater table is located at various depths depending on the local topography. The reported values range from 6 to 17 m in the floodplains to 15-30 m on fluvial terraces (Vasil'chuk 1989). Several small rivers and streams dissect the study area and drain into the Yenisei River. The major contribution to streamflow comes from snowmelt followed by summer precipitation. Groundwater base flow is present in larger streams.

Data and methods
Field investigations were conducted over the 2013-2014 period at three watersheds within the 500 km 2 study area. The three watersheds include: Graviyka, Fox Creek, and Little Graviyka that drain into the Yenisei River. Watersheds of different sizes were selected to represent a range of stream sizes and environmental conditions found within the region (figure 1): Graviyka is the largest stream out of the three studied with a watershed occupying more than 320 km 2 primarily within the second and third Yenisei River terraces. Two thirds of the watershed is covered by forest while a quarter of the basin is characterized by tundra landscapes, affected by shallow Holocene permafrost. Numerous small lakes and bogs occupy the upper portion of the Graviyka basin. A small portion of the watershed is covered by non-vegetated surfaces, predominantly sand blowouts.
Fox Creek is a medium size stream located mostly within the second Yenisei terrace with a 25 km 2 watershed. While the relative proportion of forested and permafrost-affected tundra landscapes are similar to Graviyka, a significantly smaller portion of the Fox Creek basin is occupied by lakes as the second Yenisei terrace is primarily composed of alluvial and glaciofluvial sediments formed during Kargino interstadial (MIS 3) and is well drained.
Little Graviyka is the smallest stream with a watershed of about 1 km 2 . It is located on the narrow permafrost dominated first terrace of the Yenisei River at 20-40 m asl formed primarily during Sartan stage (MIS2).

Climatic and permafrost characteristics
Daily meteorological data (air temperature, precipitation, and snow height, and soil temperature) were obtained from NCDC GSOD for the Igarka weather station (WMO ID 23274, 67.47 N, 86.57 E, 20 m a.s.l.). Daily soil temperature data were available at standard depths of 0.2, 0.4, 0.8, 1.6, and 3.2 m for 1977-2014.
Active layer thickness data are available for 2008-2014 from the Circumpolar Active Layer Monitoring (CALM) R40 site (Zepalov et al 2008, Sergievskaya et al 2012 located 5 km North-West from Igarka (http://gwu.edu/~calm). The site is representative of two main landcover classes found in the area.
Permafrost temperature at 0.5, 3.0, 5.0 and 10 m depths were monitored at two GTN-P boreholes: Graviyka1 and Graviyka2, representing forest and tundra landscapes respectively, in the vicinity of the CALM site (http://gtnpdatabase.org/). Temperatures in both boreholes were recorded every 4 h at 0.5, 3.0, 5.0 and 10 m depths using Onset U12 4-channel temperature loggers.

Landcover classification and temporal change
In order to perform the landcover classification for the three studied drainage basins, the location of the creeks were first digitized from a 1989 topographic map (Q-45-03, 04; State Geodetics and Cartography Service, 1989) (1:200,000 scale) and verified using Google Earth imagery from 2012. The drainage basins were then delineated around each of the creeks using the topographic map and an ASTER GDEM V2 digital elevation model.
Landsat TM-based 30 m resolution imagery was used to create landcover classification maps for the study area to assess 30 year changes in areal extent of shallow permafrost-affected landscapes for each watershed of this study. A modified version of the methodology known as 'Landsat dense time stacking,' was used to classify Landsat imagery for the area containing the three basins previously delineated (Schneider 2012, Nyland 2015. The method works by stacking all available scenes, including those normally rejected due to date or obscuring cloud cover, with the thought that enough scenes will compensate for data gaps. For this work, all available Landsat 5 and 8 scenes were downloaded from the USGS Global Visualization Viewer (http://glovis.usgs.gov/) and georectified as needed. Using ERDAS Imagine 2013 and ArcGIS 10.1 software, NDVI was applied to each scene in the 1985-1987 and 2013-2014 time periods. All NDVI images were stacked and the maximum value for each pixel extracted. The resulting images of maximum NDVI per pixel represent a composite of the least cloud affected pixels. Using 15 training sites per class, the maximum and minimum NDVI values were used in a conditional statement to classify the entire scene into a thematic map. Lastly, an accuracy assessment was performed for the later period using high resolution imagery and ground truth data collected using GPS units within the study area during the summer of 2013. A 2006 Quickbird scene was used for this 2013-14 classification accuracy assessment. While not from the time period this was the best high resolution scene that could be obtained in proximity to the time period of interest. The assessment yielded an overall accuracy of 83.33%.

Hydrological characteristics
Water flow was measured four times at Fox Creek and Little Graviyka gauging stations during the summer using flow probe ISP 1 M (Russia) and Flow Probe FP201-Global Water (USA) at 0.2, 0.6, and 0.8 m depths with 5% accuracy. Daily discharges were calculated based on rating curves established for each gauging station. Water levels were measured with portable loggers by Global Water (USA) at Graviyka River and Fox Creek gauging station during the summers of 2013 and 2014, however malfunctioning loggers prevented the use of these data and were therefore excluded from the paper. Therefore discharge data from the Hydrograph Hydrological Model specifically parameterized for the Graviyka watershed were used in the absence of observed discharge data during the sampling period (Lebedeva et al 2014).
Historical daily discharges available for Graviyka River gauging stations were digitized from the Hydrological Yearbooks from 1940 to 1994 and have the most complete records during the 1973-1990 period when only few days had missing observations. Electrical conductivity (EC) and pH were measured using Global Water EC500 in situ only at Graviyka River at the same time that water samples were collected for isotope analysis. The accuracy is 2 μS for EC and ±0.01 for pH.

Water samples for stable isotope analysis
Water samples for the stable isotope analysis were collected from various water sources within the study area during 1 January 2013 to 1 October 2014 period. Samples of stream water were collected at 1 to 10 day intervals, depending on the rate of change in hydrologic conditions and sites' accessibility. Samples of precipitation (rain and snow) were collected to establish the local meteoric water line (LMWL). Additional snow samples were collected from all three watersheds near the corresponding gauging stations during the winter 2013/14. To determine the stable isotope composition of the upper permafrost and seasonally frozen layer, six 1.2 m long cores were extracted from boreholes drilled. Three cores represented near-surface permafrost in tundra landscape and three represented the seasonally frozen layer within the birchlarch forest (figure 1). Additional ground ice samples were obtained in the permafrost tunnel under the Igarka Permafrost Museum at depths of 4.5, 7.5 and 10 m. To determine the influence of summer evaporation and potential ground ice contribution to the isotope composition of the water, samples were collected from thermokarst lakes and bogs, and from a ground water spring located in Graviyka watershed. The location of sampling sites is shown in figure 1. A detailed description of sampling procedures is given in appendix.
All collected water samples were analyzed at the Core Facility RC 'Geomodel' at St. Petersburg State University, Russia using cavity ring-down spectroscopy (Picarro L-2120i). Both δ 18 O and δD were determined for the same water sample. The resulting values were expressed in δ-notation relative to the Vienna Standard Mean Ocean Water (VSMOW2). The precision of the method is ±0.1‰ for δ 18 O and ±1.0‰ for δD.

Results and discussion
Changes in ground temperature and active layer thickness An increase in mean annual air temperature of 0.4°C/ decade was observed at the Igarka weather station over the last forty years. The increase is attributed primarily to spring months. Annual precipitation has not experienced significant change over the same period, however, starting in 2000, precipitation increased slightly relative to the long-term mean. The long-term increase in regional air temperature, accompanied by recent increase in snow cover can contribute to the degradation of relatively warm and thin permafrost within the study area. Daily permafrost temperature observations are available from two boreholes within the study area for 2006-2013. Mean annual ground temperature at 10 m depth is −0.25°C within forested areas and −0.20°C in tundra landscapes suggesting that permafrost is near the melting point.
Although there are no long-term direct permafrost observations within the study area, several lines of evidence suggest widespread permafrost degradation throughout the area which manifested itself through the decrease of permafrost extent, thermokarst development, thickening of the active-layer, and increases in soil temperature.
The long-term (1977-2014) soil temperatures at several depths (up to 3.2 m) are available from the weather station located in the town of Igarka. Observations are conducted at a standard meteorological leveled plot under undisturbed vegetation and snow cover conditions. Vegetation is primarily composed of mature seeded grass. While this vegetation cover is not representative of the vegetation within the study area, the data can be used to analyze the general trends of climate-induced changes in the ground thermal regime. The data shows an increase in soil temperature at 3.2 m depth from −0.5°C in 1970 s to +2.5°C in 2010 s. Depths of annual thawing and freezing, estimated by polynomial interpolation of temperature data from different depths are shown in figure 2. The lowering of the permafrost table and the decrease in the depth of seasonal freezing have together promoted development of the residual thaw layer above the permafrost in late 1980s.
Although the response of permafrost to climatic forcings (e.g. temperature, precipitation) in natural tundra landscapes are moderated by thick moss and organic layers, the observed trend can indicate an overall soil temperature change in the Igarka region. This notion is indirectly supported by comparing the landscape-specific active-layer observations collected in the study area during 1950s with those collected at the local CALM site over the period 2008-2014. Although the exact locations of historic measurements are not known, the summary data (Tyrtikov 1964) provide detailed landscape characterizations making it possible to compare the data with modern CALM measurements from similar landscapes. The comparison indicates that from the 1950s to the 2010s the average active-layer thickness has increased from 0.38 m to 0.55 m in tundra landscapes dominated by organic soils.

Changes in landcover
Extensive field work conducted within the study area in the 1950s and 1960s suggests that tundra dominated landscapes are indicative of the presence of Holocene permafrost and that sparse forest landscapes are generally corresponding to areas with a significantly deeper permafrost table associated with Pleistocene permafrost (Tyrtikov 1964). As a result, a generalized landcover classification can be used to assess the spatial distribution of shallow permafrost.
The comparison of landcover classification maps for 1985-1987 and 2013-2014 reveals a decline in the areal extent of tundra predominantly due to the expansion of forested areas (table 1). Such landscape changes can, at least partially, be indicative of permafrost degradation in the Igarka area that occurred over the 30 year period. However, these results should be interpreted with caution as they are less than the reported classification error, even though the probability of capturing noise rather than actual change was minimized by the consistent methodologies used for both time periods.

Changes in hydrological characteristics
The historic observational data were used to show the changes in the hydrological regime in changing climatic conditions; while model estimates of the 2013-14 are used further to identify the sources of water flow on a seasonal scale. Graviyka is the only stream within the study area which has historic (1973)(1974)(1975)(1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)) discharge observations. Monthly discharge values were analyzed to assess stream flow seasonality for the 1970s and 1980s. Monthly discharge statistics in m 3 s −1 for two periods are presented in table 2. Relative contributions of monthly stream discharges to annual totals for 1973-1981 (1723 m 3 s −1 ) and 1982-1990 (1804 m 3 s −1 ) and relative monthly discharge and precipitation change between the two periods are shown in figure 3. The period 1973-1990 is characterized by decreasing precipitation in all months except August. Simultaneously, a pronounced increase in discharge has occurred over cold (October-April) and snow melt (May) periods. June and July were characterized by decrease in discharge. The increase in May and decrease in June suggest a hydrologic regime shift toward the earlier snowmelt attributable to the spring warming observed during this period.
The results of this study indicate that with the exception of spring flooding, rain is the dominant source of water in these streams. To assess the relationship between monthly precipitation and discharge we have used regression analysis for two periods (1970s and 1980s). The results indicate an increase in the time-lag between precipitation events and stream discharge between 1970s and 1980s. During the warm months of the earlier decade there is a positive relationship between precipitation and discharge with a one month time-lag between July precipitation and August streamflow (R 2 =0.38), and August precipitation and September streamflow (R 2 =0.61). For the 1980s, the highest correlation was found for a twomonth time-lag between July precipitation and September discharge (R 2 =0.71) and between August precipitation and October discharge (R 2 =0.33). The effect of time-lag produced increases in discharge is most pronounced for September, the last month of consistent liquid precipitation (figure 4). All R 2 >0.45 are statistically significant at p<0.10.
In the 1970s September precipitation was strongly correlated with December discharge (R 2 =0.57), while their influence on January flow was greatly reduced (R 2 =0.23). No correlation was found between September precipitation and February and March flow. For the 1980s, September precipitation contributed to discharge throughout the cold season. The highest correlation was found for January (R 2 =0.62), followed by February (R 2 =0.49), and March (R 2 =0.44).
The estimated annual total discharge during 2013 was 1240 m 3 s −1 and 2014 was 2740 m 3 s −1 . While model estimates cannot be directly compared to the observational data, as the model does not reproduce winter base flow adequately, they give insight to the peculiarities of the hydrological conditions of the two project years. Lower than average discharge in 2013 is attributed to low winter and summer precipitation (225 mm and 50 mm respectively), followed by an extremely hot summer in 2013 (where the mean July temperature was 20.1°C). The substantially higher discharge in 2014 is attributed to extremely high snow accumulation in winter (474 mm), followed by a rainy (71 mm) and relatively cool summer (where the mean July temperature was 14.7°C).
Stable isotope composition in a hydrological system Precipitation: rain and snow From the 38 precipitation samples three have been excluded as outliers, characterized by extremely high    (table 3). D-excess in most cases is close to 10‰. Values of δ 18 O are increasing from January to April from −28 to −25‰, while D-excess is decreasing from 13 to 7‰ over the same period. The substantial differences in isotopic composition between rain and snow make it possible to differentiate these two sources of river flow.

Ground ice
Although the number of ground ice samples collected from the soil profile were limited, they provide general regional information on the isotopic composition of the ground ice from different depths. Stratigraphic similarity between cores of various lengths (1.2 m and 3 m) and the permafrost tunnel observations at 3-10 m depths allowed for the use of ground ice samples from multiple sources to reconstruct an isotopic profile up to 10 m depth.
Ground ice samples collected from the tundra cores indicate that the active layer is characterized by slightly heavier isotopes compared to the upper permafrost. The δ 18 O values are decreasing slightly from −15‰ within the active layer to around −17‰ in the upper permafrost. Heavier isotope composition of the active-layer may be attributed to late-summer precipitation (−15.5 to −17‰ δ 18 O for the second half of September) stored within the active layer prior to freezing. D-excess below 8‰ within the active layer is indicative of evaporation processes likely within the active layer. Below 1-1.5 m D-excess is above 8‰. Samples collected from the layer of seasonal freezing in forested landscapes show almost no difference in isotopic composition compared to the tundra active layer indicating the freezing of late summer precipitation.
The deep Pleistocene permafrost is represented by nine samples collected in the permafrost tunnel at 4.5-10 m depth. It showed δ 18 O values between −16.1 and −16.9‰ regardless of depth corresponding with D-excess from 10.6 to 11.9‰ (table 3). The low variability of the ground ice isotope composition may indicate a fast epigenetic freezing of limnic sediments with corresponding lowering of the groundwater table, supporting previous study of the isotope composition. Similar values of δ 18 O in late summer precipitation and in permafrost texture ice make the separation of these two sources of water flow difficult and might present the major limitation to this study.

Ground water
Ground water collected throughout the summer season is characterized by a depleted isotopic composition relative to summer precipitation and ground ice (table 3, figure 6). From late June to late September, δ 18 O is increasing from −19.7 to −18.9‰ corresponding with decreases in D-excess from 12.4 to 8.8‰ suggesting that the water originated in an aquifer likely filled with snowmelt water with increasing contributions from liquid precipitation throughout the summer. Higher δ 18 O values in the texture ice compared to the ground water suggest no contribution of melting ice over the observational period.

Thermokarst lakes and bogs
Isotopic composition of water collected from thermokarst lakes is characterized by an isotopic composition similar to late summer precipitation with δ 18 O from −17.7 to −16.4‰ and D-excess from 10.2 to 21.4‰. While δ 18 O of the texture ice in the active layer and upper permafrost is similar to the water composition in the thermokarst bogs the mean D-excess of 13.8‰ and slope of 7.2‰ further suggests that late summer precipitation, and not ground ice, is the dominant water source.

Local streams
Rivers in the study area vary from less than 1 km to more than 50 km in length with corresponding watershed areas from 1 to 323 km 2 (table 4). The largest stream (Graviyka) is characterized by a heavier isotopic composition reflecting longer storage of water in surface (thermokarst lakes, bogs) and subsurface (ground water) reservoirs within the watershed. The isotopic composition of water in smaller streams (Fox Table 3. Stable isotope (δ 18 O, δD, and D-excess) minimum, mean and maximum values, standard deviations as well as slope and intercept in a δ 18 O-δD co-isotopic plot for all sampled precipitation, watershed snow cover, water flow and river ice of the three studies rivers, thermokarst lakes and ground ice within the study area. Creek and Little Gravyika) corresponds well to precipitation indicating quicker watershed drainage after each precipitation event compared to the larger Gravyika stream (figure 6). Figure 6 shows daily weather characteristics such as air temperature, precipitation, and snow accumulation, hydrochemical characteristics of waterflow in Graviyka stream (pH and EC), and the isotopic composition of precipitation and stream water from three watersheds. It should be noted, that the regression of the isotopic composition of water from the Graviyka stream, δD=6.7 δ 18 O−14, is similar to that of water from the Yenisei as measured 225 km downstream from the study area (δD=7.04 δ 18 O −9.32, flux weighted δ 18 O is −18.4‰ and δD is −139.4‰; (Yi et al 2012)). Figure 7 provides summary frequency distributions of δ 18 O in water flow of the three streams with all other end-members shown on the single graph. The isotopic composition of stream water varies depending on major annual hydrological events such as spring flood, summer-fall low flow, and winter baseflow. There is a clear distinction between rain and snow; however other sources of water in a river flow have similar isotopic composition making further separation quite difficult. Despite of this limitation, the isotope data provide valuable insight into the understanding of the hydrological processes on permafrost.
Spring flood Streams within the study area are characterized by steep maxima in discharge during the spring flood, which occurs in late May through the beginning of June. Spring floods are dominated by snowmelt water with substantially lighter isotopic composition and higher values of D-excess (figure 6). The magnitude of the spring flood depends on the watershed characteristics (area, topography, exposure), amount of snow, rate of air temperature change, and rain intensity during the snowmelt. The onset of the flood in smaller streams precedes that of larger streams by about a week. The winter of 2014 had considerably higher snow accumulation compared to 2013 (111 and 78 cm, respectively). However, the spring flood of 2014 had a smaller magnitude as the duration of snowmelt was twice as long (20 days in 2013 and 9 in 2014). The maximum daily discharge of Graviyka River estimated for 2013 was 94 m 3 s −1 , while 160 m 3 s −1 in 2014.
The isotope composition of river flow is almost identical to the isotopic composition of the snowpack in corresponding watersheds prior to snowmelt with δ 18 O around −24‰ to −25‰ (table 3). Isotopically depleted melting snow is mixing with enriched liquid precipitation throughout the snowmelt season resulting in a gradual increase of δ 18 O in the waterflow. Gradual increases in temperature during the spring of 2013 and a short, but intense snowmelt is clearly seen in the isotopic curves of all three studied rivers ( figure 6). However, the long spring flood of 2014, with temperatures above and below 0°C and several rain-on-snow events, with particularly strong rain (8 mm d −1 ) on 3 June 2014 resulted in the temporary enrichment of snowmelt water by heavy rain precipitation dividing the δ 18 O minimum into two minima: one prior to the rain event that was produced predominantly by snowmelt, and another smaller minimum later that was produced by the combination of liquid precipitation and snowmelt from snow held in watershed depressions (e.g. gullies, thermokarst). Significant decreases in EC from 250 to 40 μS and pH change from 8.7 to 7 in Graviyka River flow confirms dominating snowmelt water's contribution to river discharge. The presence of near-surface permafrost does not seem to play an important role during the snowmelt period as both permafrost and seasonal frost create an impermeable layer close to the surface allowing effective watershed drainage for all landscapes regardless of the permafrost depth.

Summer-fall flow
The frozen ground starts to thaw by the last week of May, first week of June. In forested areas with no nearsurface permafrost, the layer of seasonal freezing thaws completely by the third week of July allowing the current year's precipitation to recharge water previously stored in the soil. The downward movement of water in tundra landscapes is limited to a shallow depth by the presence of the permafrost table. However, lateral water migration is possible on slopes, especially in organic-rich soils characterized by high hydraulic conductivity. The thaw depth in areas with near-surface permafrost increases throughout the  warm season and reaches its annual maximum in the beginning of October. The active-layer thickness depends on edaphic properties of the landscape and ranges from 0.4-0.7 m in organic-rich soils, 0.8-0.9 m in loamy soils, and up to 1.0-1.5 m in proximity to large thermokarst lakes. The propagation of thawing throughout the summer season increases soil storage capacity resulting in a longer time-lag between precipitation events and increased stream discharge. Heavier isotopic composition and lower deuterium excess of the summer streamflow suggests that liquid precipitation is the dominant source for the flow after the spring flood(s) (figures 6 and 7). Variations of EC and pH of the river flow follow the magnitude of precipitation events. On average, the discharge of Graviyka is 4.6 m 3 s −1 , while Fox Creek discharge is only 0.46 m 3 s −1 . The discharge of smaller streams responded to the precipitation events quicker and without alteration of the isotopic composition. The waterflow of the larger Graviyka River shows heavier isotope composition relative to precipitation events. This indicates that initially water is stored and subject to evaporation in numerous bogs and thermokarst lakes within the watershed prior to entering the stream. The effect of evaporation, visible in higher δ 18 O and lower D-excess values, was most pronounced during the abnormally warm and dry summer of 2013. By contrast, it was less evident during the cooler summer of 2014.

Baseflow condition
Air temperature drops below 0°C during the first week of October initiating seasonal freezing. Initial development of a downward migrating freezing front reflects in water flow as a general decrease of δ 18 O by 1-1.5‰ and of δD by about 10‰, after which δ 18 O and δD vary little during the entire winter low with exception of late March (figure 6). The depletion of water flow isotope composition occurs due to preferential freezing of heavy isotopes during ice segregation (Konishchev et al 2014) and the fractionation of water due to congelation of ice formation (Vasilchuk 2011). Within the study area the isotope composition of river is complicated further by soil/groundwater intrusion on the surface of the ice and the formation of naleds (a. k.a. river icings).
Field observations during late winter-early spring show numerous events of ejection of organicrich distinctively brown soil water flowing on the surfaces of the frozen streams. The downward propagating freezing front precludes surface water infiltration and builds up the hydrostatic pressure within the soil column. Soil water moves to the locations with smaller hydraulic head, such as streams channels, contributing to the late fall-winter increase in stream discharge. High hydraulic pressure of the stream flow under the ice was evident during several ice drilling tests conducted throughout the winter. The water was quickly ejected from the relatively small hole forming a 0.05-0.06 m thick water layer on the surface of the ice. Smaller streams of less than 1.5 m depth freeze up completely by the end of February-early March limiting water movement to taliks below the stream channel. The water commonly reaches the surface of river ice via multiple cracks. The cracks are usually found near the river banks, where excessive windblown snow cover prohibits the formation of a thick layer of river ice. By the end of the winter the ground water source is usually depleted, causing creeks to dry. The only river that is not freezing to the bottom is Graviyka. The minimum estimated discharge of 0.1-0.2 m 3 s −1 is found in late April and May, just before the spring flood. Maximums of δ 18 O observed in March of 2013 and 2014 in the Graviyka River and negative D-excess are similar to the ones observed during mid-summer suggesting that enriched summer water is stored in the watershed soil and had finally reached the water flow (figure 6). This time also corresponds to the highest level of EC of 240-260 μS and pH of 8.5-8.7. The fact that these events occurred during maximum freezing suggests that the advancing freezing front and therefore closing of the hydrological system was able to deliver the water previously stored in the soil to the river flow.

Role of permafrost in hydrology in changing climatic conditions
Observational results suggest that permafrost plays an important role in the maintenance of discharge during the cold months (Walvoord and Striegl 2007, Walvoord et al 2012, Woo 2012, Jorgenson et al 2013, Quinton and Baltzer 2013. The short observational record prevents long-term analysis of the hydrologic and permafrost systems' behavior within the study area. However, indirect evidence, described above, can be used to infer the effects of long-term climate and permafrost changes on the hydrologic regime in the Igarka region. The changes in groundwater storage resulting from thickening of the permafrost active layer in tundra landscapes and the thinning of the seasonal freezing layer in forested landscapes can, at least partially, explain the changes to the hydrologic regime evident on figure 3. Progressive lowering of the permafrost table and decreases in depth of seasonal freezing during the 1980s promoted more water storage in deeper soil layers during late summer months (July and August) and increasingly delayed release of water into the streams relative to the colder 1970s. The development of residual thaw layer above the permafrost in late 1980s allowed for soil water migration during the winter months, contributing to increases in winter flow. Our field observations of ground water ejections during the winters of 2013 and 2014, a heavy isotopic composition and high EC of winter flow in the Graviyka during the period of maximum freezing, formation of extensive river icings (naleds) in smaller streams further confirm redistribution of summer precipitation toward discharge during winter months.
Enhanced release of relatively warm soil water into streams during winter months can potentially contribute to decreases in river ice thickness (Jones et al 2013) further promoting increases of winter discharge. A recent study showed that the duration of ice conditions across Russian Arctic rivers have decreased over the last fifty years (Shiklomanov and Lammers 2014). For the Yenisei River near Igarka, the river ice formation is initiated 2 days later, while the river ice melts 5 days earlier and the maximum ice thickness decreased from about 1.3 m to 0.8 m.
Although this study did not find any direct evidence of ground ice melt in streams within the study area it is likely that melting ground ice can contribute to river discharges under continuing climate change, especially in areas with the presence of ice-wedges and tabular ground ice, such as Yenisei North. Ice wedges form during snowmelt by the immediate refreezing of melt water in thermal contraction cracks. Their isotopic composition is indicative of winter temperatures (Meyer et al 2015) and therefore differs considerably from summer precipitation. Reflecting the different climate conditions during ice wedge formation, isotopic differences can be used to distinguish Holocene from Late Pleistocene ice wedges. The closest ice wedges studied in detail (Cape Sopochnaya Carga, 71 53' N, 82 40 E, 500 km, downstream on the Yenisei River, revealed mean δ 18 O (δD) values of −20.3‰ (−154.4‰) for Holocene and −26.0‰ (−199.5‰) for Late Pleistocene ice wedges (Streletskaya et al 2011). Tabular massive ground ice of likely glacier origin on the banks of the Yenisei River (namely the 'Ice mountain' 100 km upstream) exhibit a mean δ 18 O value of −20.4‰ (Vaikmae et al 1988). As these isotope values are significantly lower than that of summer precipitation the impact of melting massive ground ice should be better detectable in river water relative to the texture ground ice.

Conclusions
Rivers and creeks in the discontinuous permafrost region have characteristic intra-annual variability evident from changes in hydrological characteristics and the stable isotope composition of river water. Heavy isotope composition during the summer is attributed to enriched summer precipitation, commonly amplified by evaporation effects for rivers with watersheds occupied by thermokarst lakes and bogs allowing for the storage of water from precipitation. Watersheds freezing in the fall results in a slight decrease in isotopic composition of river water relative to the late summer values due to the formation of segregation ice. Isotopic composition of water during winter low flow is primarily attributed to isotopically enriched ground water from the late summer precipitation events. Lighter values in spring are attributed to snow melt.
Permafrost plays an important role in the hydrology of the study area. Increases in air temperature observed in the study area resulted in permafrost degradation and shallower seasonal freezing of the ground which is evident from soil temperature measurements, permafrost and active-layer monitoring and analysis of satellite imagery. The lowering of the permafrost table and thinning seasonal freezing layer have resulted in increased soil storage capacity and allow the redistribution of late summer precipitation towards late winter-spring months when the freezing front is at its maximum depth.
While the study did not find a direct contribution of melting ground ice to river flow, the redistribution of summer precipitation to the winter low, observed ejections of ground water and the formation of naleds (river icings) provide evidence for the indirect influence of permafrost degradation to river discharge, supporting findings of changes of the terrestrial water budget of the Eurasian Pan-Arctic from GRACE satellite (Landerer et al 2010).
These results have improved our understanding of interactions between permafrost and hydrology in small arctic streams and of the relative contributions from different sources to overall river discharge. They can also be useful for improving the representation and parameterization of freeze-thaw processes in various hydrological models and as validation for isotope models in the Arctic.
water freezing and to assess the groundwater contribution to ice formation in the studied streams.
To determine the stable isotope composition of the upper permafrost and seasonally frozen layer, six 1.2 m long cores were extracted from boreholes drilled in February of 2013. Three cores represented nearsurface permafrost in tundra landscape and three represented the seasonally frozen layer within the birch-larch forest (figure 1). At each tundra site, ice samples were collected every 0.40±0.1 m along the entire core. At forested sites seven samples were collected from the seasonally frozen layer up to 0.3-0.4 m depth. Additional permafrost samples were obtained from each stratigraphic layer of a single threemeter core extracted from the tundra in July 2013. Samples from the permafrost were collected in the permafrost tunnel under the Igarka Permafrost Museum at depths of 4.5, 7.5 and 10 m (three samples at each depth).