Abstract

River runoff to the Arctic Ocean has increased over the last century, primarily during the winter and spring and primarily from the major Eurasian rivers. Some recent studies have suggested that the additional runoff is due to increased northward transport of atmospheric moisture (and associated increased precipitation), but other studies show inconsistencies in long-term runoff and precipitation trends, perhaps partly due to biases in the observational datasets. Through trend analysis of precipitation, temperature, and streamflow data, the authors investigate the extent to which Eurasian Arctic river discharge changes are attributable to precipitation and temperature changes as well as to reservoir construction and operation between the years of 1936 and 2000. Two new datasets are applied: a gridded precipitation product, in which the low-frequency variability is constrained to match that of long-term bias-corrected precipitation station data, and a reconstructed streamflow product, in which the effects of reservoirs have been minimized using a physically based reservoir model. It is found that reservoir operations have primarily affected streamflow seasonality, increasing winter discharge and decreasing summer discharge. To understand the influences of climate on streamflow changes, the authors hypothesize three cases that would cause precipitation trends to be inconsistent with streamflow trends: first, for the coldest basins in northeastern Siberia, streamflow should be sensitive to warming primarily as a result of the melting of excess ground ice, and for these basins positive streamflow trends may exceed precipitation trends in magnitude; second, evapotranspiration (ET) in the warmer regions of western Siberia and European Russia is sensitive to warming and increased precipitation, therefore observed precipitation trends may exceed streamflow trends; and third, streamflow from the central Siberian basins should respond to both effects. It is found that, in general, these hypotheses hold true. In the coldest basins, streamflow trends diverged from precipitation trends starting in the 1950s to 1960s, and this divergence accelerated thereafter. In the warmest basins, precipitation trends consistently exceeded streamflow trends, suggesting that increased precipitation contributed to increases in both ET and streamflow. In the central basins, permafrost degradation and ET effects appear to be contributing to long-term streamflow trends in varying degrees for each basin. The results herein suggest that the extent and state of the permafrost underlying a basin is a complicating factor in understanding long-term changes in Eurasian Arctic river discharge.

1. Introduction

Annual average temperatures over the Arctic have increased at nearly twice the rate observed for the rest of the globe over the last few decades (ACIA 2005), and paleoclimate records indicate that recent temperatures in the Arctic are the highest they have been in the last 400 yr (Overpeck et al. 1997). Scientists and indigenous peoples have observed many changes in the Arctic, indicating that the region is undergoing a system-wide response to a changing climate (Hinzman et al. 2005). Research interest in Arctic regions derives partly from the region’s sensitivity to global change, and partly from the fact that climate-induced changes in the Arctic have the potential to feed back to the global climate system; for example, increasing freshwater flux to the Arctic Ocean may cause a weakening of the World Ocean’s thermohaline circulation (THC) by inhibiting deep-water formation in the North Atlantic (Broecker 1997; Clark et al. 2002; Curry et al. 2003). Implications of THC changes include slowing the rate of warming (and perhaps even cooling) over the North Atlantic and northern Europe, and slowing of CO2 transport to the deep ocean, leading to increased global warming (ACIA 2005).

Combined annual streamflow volumes from the six largest Eurasian basins increased by approximately 7% (or 2.0 ± 0.4 km3 yr−1) between 1936 and 1999 (Peterson et al. 2002; Shiklomanov et al. 2006). Changes in annual runoff have been accompanied by shifts in seasonality. In general, the largest increases have been observed during the cold season, which is also the season of low flow (October–April), while the spring snowmelt peak has shifted earlier (Georgievskii et al. 1996; Yang et al. 2004a, b; Yang et al. 2002), although this is not the case for all rivers. Changes in other seasons are generally less robust; for the Lena and Ob’ River basins, summer discharges have slightly increased, whereas fall discharges have slightly decreased (Yang et al. 2004b; Yang et al. 2002; Ye et al. 2003).

Notwithstanding the potential for observed Arctic river discharge changes to affect global climate, there is little current understanding of the primary controls on Arctic river streamflow change and variability, as evidenced by conflicting explanations for observed historical changes. Knowledge of these controls would arguably facilitate better parameterization of key physical processes in land surface models that represent runoff generation processes in global and regional climate models, and in turn should improve the predictability of climate feedback processes. The potential controls on Arctic river discharge are discussed in section 2.

Trend attribution for northern Eurasian rivers is complicated by the paucity, density, and quality of observation data in Arctic areas, particularly that of precipitation. Berezovskaya et al. (2004) suggest that inconsistencies between observed precipitation and streamflow trends could be a result of data quality, primarily that of the gridded precipitation data. Known problems include sparse and nonuniform station networks (Adam and Lettenmaier 2003; Serreze et al. 2003b), temporally inhomogeneous measurement and preprocessing techniques (NCDC 2005; Groisman and Rankova 2001; Groisman et al. 1991; Mekis and Hogg 1999), orographic effects that are not well represented in station data (Adam et al. 2006), large precipitation gauge measurement biases (Adam and Lettenmaier 2003; Groisman et al. 1991; Yang et al. 2005), effects of climate on undercatch adjustments (Forland and Hanssen-Bauer 2000; Yang et al. 2005), temporally inconsistent station networks (Hamlet and Lettenmaier 2005; Rawlins et al. 2006), and a decline in observing stations since 1990 due to station closure (Serreze et al. 2003b; Shiklomanov et al. 2002). These biases are discussed in more detail in section 4b.

The goals of this study are twofold. The first is to describe a new gridded precipitation product for the Eurasian Arctic that includes adjustments that make it more suitable for evaluation of long-term trends. The second is to apply the new precipitation product as well as observed temperature, observed streamflow, and new reconstructed streamflow data to investigate the most probable contributors to observed streamflow trends over the Eurasian Arctic. More specifically, we apply the new reconstructed streamflow data to isolate the effects of reservoir operations on long-term trends in streamflow from other causes. Following this, we apply the new precipitation data to identify when and where precipitation trends are inconsistent with streamflow trends, and we develop a hypothesis to explain how temperature changes may account for these inconsistencies.

2. Background: Potential controls on observed streamflow changes

Recent research has focused on any or a combination of the following potential controls on observed Eurasian Arctic streamflow trends: 1) human effects (primarily through reservoir construction and regulations); 2) increased northward transport of moisture (as a result of an intensification of the hydrologic cycle); 3) release of stored frozen groundwater via permafrost degradation; and 4) factors affecting evapotranspiration (ET). Recent work in each of these areas is reviewed briefly below.

a. Reservoir construction and operation

Beginning in the mid-1950s, the former Soviet Union constructed several major hydroelectric dams in the large northern Eurasian basins (McClelland et al. 2004). The effects of reservoirs on long-term trends at the basin outlets have been analyzed using a variety of approaches. McClelland et al. (2004), Ye et al. (2003), and Yang et al. (2004a) used statistical techniques to develop reconstructed streamflow data, in which the effects of reservoirs were removed from basin-outlet streamflow. Using these products, they concluded that estimates of natural river discharge trends at the outlets of these river basins tend to be underestimated in the summer and overestimated in the fall and winter as a result of reservoir operations. Furthermore, McClelland et al. (2004) conclude that reservoir operations cannot be responsible for the observed increases in annual streamflow, and may have even reduced annual streamflow volume, either through groundwater storage dynamics or increases in evaporation due to increased surface water storage. Using a physically based reservoir model that determines reservoir releases by maximizing hydropower production for each operational year, Adam et al. (2007) show similar effects of reservoirs on long-term streamflow trends. Although the annual effects are small, seasonal effects are large. For instance, Adam et al. (2007) show that at the outlet of the Lena basin, reservoirs account for approximately 80% and 30% of the observed 1938–98 winter and spring streamflow trends, respectively. At the outlet of the Yenisei basin, reservoirs account for 100%, 40%, and 60%–100% of the observed winter, spring, and late summer to early fall streamflow trends, respectively, and at the outlet of the Ob’ basin, reservoirs account for 25%–100% of the observed streamflow trends in the winter and early spring.

b. Precipitation changes

Various coupled and uncoupled land–atmosphere modeling studies have predicted increased Arctic river discharge due to increased northward atmospheric moisture transport (Arnell 2005; Lawrence and Slater 2005; Manabe and Stouffer 1980; Miller and Russell 1992; Nijssen et al. 2001; Wu et al. 2005). Although precipitation exerts a primary control on streamflow, there is little observational evidence that precipitation is the sole or even primary driver of observed trends for many of the basins. Berezovskaya et al. (2004) compared basin-outlet discharge to basin-average precipitation for the largest basins (the Ob’, Yenisei, and Lena) for the period of 1950–2000. Their results show little consistency between precipitation and streamflow trends. A positive trend in Lena streamflow is accompanied by only a weak precipitation increase; Yenisei discharge increases are accompanied by mostly negative precipitation trends; and insignificant trends in both discharge and precipitation are observed for the Ob’. Pavelsky and Smith (2006) performed a similar analysis but at a finer spatial resolution, examining the compatibility of long-term streamflow and precipitation trends for 66 river basins in northern Eurasia. Of these basins, 40 exhibited statistically significant streamflow trends, and 29 of these also exhibited statistically significant precipitation trends (22 increasing and 7 decreasing). For these 29 basins, agreement between precipitation and discharge trends was between 35% and 62%. Pavelsky and Smith (2006) conclude that precipitation does play an important role for many of the basins, but not all of them, and may or may not be the sole mechanism.

c. Permafrost changes

Russian soil temperature observations indicate that permafrost active-layer depth has deepened by approximately 20 cm on average throughout Siberian permafrost regions between 1956 and 1990 (Frauenfeld et al. 2004). Zhang et al. (2003) hypothesize that climate-induced changes to permafrost could lead to changes in both streamflow seasonality and annual volumes. Thickening of the active layer leads to an increase in liquid groundwater storage, while the delaying of active-layer freeze-up leads to the potential for subsurface water to contribute to streamflow into the winter (Serreze et al. 2003a; Yang et al. 2002; Ye et al. 2004; Zhang et al. 2003). Annual streamflow is augmented by the thawing and release of excess ground ice (Frauenfeld et al. 2004; Zhang et al. 2003). Although the sensitivity of winter streamflow to permafrost thaw has been demonstrated by both models and observed data (Savelieva et al. 2000; Van der Linden et al. 2003; Zhang et al. 2003), the sensitivity of annual streamflow to permafrost thaw is unresolved. McClelland et al. (2004) argue that, even if the seasonality of streamflow is affected by permafrost thaw, melting of ground ice cannot produce enough excess water to account for the observed increase in annual streamflow. They demonstrate this by back-calculating the depth of permafrost thaw needed to produce the observed annual streamflow increase (by allowing all melted ice to drain completely from the soil and contribute to runoff). They conclude that permafrost, as the only agent of change, would have needed to thaw a minimum of four meters uniformly across the six largest Eurasian Arctic watersheds, which they argue is unrealistic. In contrast, Zhang et al. (2003) used ground-ice data to estimate that an increase in active-layer depth of 30 cm would produce sufficient water to account for the apparent imbalance in the water budget (precipitation minus streamflow minus ET). One possible explanation for the discrepancy in these two estimates is that Zhang et al. (2003) considered the potential of ground-ice melt to account for this water budget imbalance, whereas McClelland et al. (2004) evaluated the ability of ground-ice melt to account for all of the streamflow increase.

d. Changes in evapotranspiration

The role of ET changes is, perhaps, the least understood factor, largely because direct ET observations are essentially nonexistent in the region. Observations that may yield insight into ET trends, such as pan-evaporation estimates, lead to further uncertainties (e.g., see Barnett et al. 2005 for a discussion of the “pan evaporation paradox”). Although increased precipitation and temperature would result in an increase in ET, the possibility exists that, even with warming, annual ET trends could be negative; that is, the greatest warming occurs during the winter when there is less liquid water available for evaporation. McClelland et al. (2004) speculate that thawing permafrost could reduce ET, if lower water tables are a result of a thickening active layer. Also, Rawlins et al. (2006) suggest that an increase in solid precipitation with respect to liquid precipitation could result in less ET because a larger proportion of solid precipitation will contribute directly to runoff during snowmelt, especially in regions underlain by permafrost where infiltration is limited. They document large solid precipitation increases, which are accompanied by liquid precipitation decreases, from the mid-1930s to the late-1950s primarily across north-central Eurasia, areas that are predominantly underlain by permafrost.

e. New contributions of this work

This study builds upon previous work, particularly that of Berezovskaya et al. (2004) and Pavelsky and Smith (2006), in which we evaluate the extent to which precipitation changes can account for observed streamflow trends. In doing so, this paper makes the following contributions:

  1. We introduce and utilize a new gridded precipitation product for the Eurasian Arctic with adjustments to improve our estimation of long-term trends.

  2. We utilize the reconstructed streamflow data of Adam et al. (2007) to isolate the effects of artificial reservoirs on long-term streamflow trends.

  3. We analyze the consistency between precipitation and streamflow trends on a temporally varying basis, which allows us to define periods when precipitation is the likely cause of streamflow changes and periods when other factors are also likely playing a role.

3. Study domain and period

Figure 1 shows our northern Eurasian study basins, all of which discharge to the Arctic Ocean. We selected three primary study basins (outlined in black in Fig. 1) and eight secondary basins (outlined in red in Fig. 1). Basin attributes are listed in Table 1. The basins were selected according to the following criteria: 1) significant streamflow changes observed at the basin outlet, 2) the availability of a long-term stream gauge record, and 3) a basin area exceeding 105 km2. Furthermore, we sought to choose an even distribution of basins across the study domain. Note that there is a strong east to west temperature gradient across northern Eurasia that is reflected in the permafrost distribution, that is, continuous permafrost in the east, threshold (discontinuous, sporadic, and isolated) permafrost in central Siberia, and seasonally frozen soil in the west (Fig. 1 and Table 2). This causes the hydrologic regimes to be diverse across these basins, and therefore we expect that the responses to climatic change will vary regionally, dependant on the mean climate and permafrost state of each basin.

Fig. 1.

Brown et al. (1998) permafrost distribution. Permafrost category is defined by the percent of area underlain by permafrost, i.e., continuous: 90%–100%, discontinuous: 50%–90%, sporadic: 10%–50%, and isolated: less than 10%. Primary study basins are outlined in black while the secondary basin and subbasins are outlined in red. Note that basin 6 is nested inside basin 5 (which is nested inside basin 1). Basin attributes are listed in Table 1. Basins delineations are according to the HYDRO1k digital dataset.

Fig. 1.

Brown et al. (1998) permafrost distribution. Permafrost category is defined by the percent of area underlain by permafrost, i.e., continuous: 90%–100%, discontinuous: 50%–90%, sporadic: 10%–50%, and isolated: less than 10%. Primary study basins are outlined in black while the secondary basin and subbasins are outlined in red. Note that basin 6 is nested inside basin 5 (which is nested inside basin 1). Basin attributes are listed in Table 1. Basins delineations are according to the HYDRO1k digital dataset.

Table 1.

Attributes for the three primary study basins (basins 1–3) and the eight secondary basins and subbasins (basins 8–11). Note that most of the secondary basins are nested inside the primary basins, i.e., as indicated by the basin reference name in column 3. Also, note that subbasin 6 is nested inside subbasin 5. Period of gauging record, basin area, and mean annual Q (for the period of gauging record) are from the Regional, Electronic, Hydrographic Data Network for the Arctic Region, version 3.0 (R-ArcticNET v3.0), dataset (Lammers and Shiklomanov 2000); reservoir information is from McClelland et al. (2004). Basin-average 1930–2000 mean annual P and T are from the UW and UDel datasets, respectively (see Table 3).

Attributes for the three primary study basins (basins 1–3) and the eight secondary basins and subbasins (basins 8–11). Note that most of the secondary basins are nested inside the primary basins, i.e., as indicated by the basin reference name in column 3. Also, note that subbasin 6 is nested inside subbasin 5. Period of gauging record, basin area, and mean annual Q (for the period of gauging record) are from the Regional, Electronic, Hydrographic Data Network for the Arctic Region, version 3.0 (R-ArcticNET v3.0), dataset (Lammers and Shiklomanov 2000); reservoir information is from McClelland et al. (2004). Basin-average 1930–2000 mean annual P and T are from the UW and UDel datasets, respectively (see Table 3).
Attributes for the three primary study basins (basins 1–3) and the eight secondary basins and subbasins (basins 8–11). Note that most of the secondary basins are nested inside the primary basins, i.e., as indicated by the basin reference name in column 3. Also, note that subbasin 6 is nested inside subbasin 5. Period of gauging record, basin area, and mean annual Q (for the period of gauging record) are from the Regional, Electronic, Hydrographic Data Network for the Arctic Region, version 3.0 (R-ArcticNET v3.0), dataset (Lammers and Shiklomanov 2000); reservoir information is from McClelland et al. (2004). Basin-average 1930–2000 mean annual P and T are from the UW and UDel datasets, respectively (see Table 3).
Table 2.

Percent area coverage by each permafrost category (calculated using the Brown et al. 1998 dataset) for each study basin. Permafrost category definitions are listed in the Fig. 1 caption.

Percent area coverage by each permafrost category (calculated using the Brown et al. 1998 dataset) for each study basin. Permafrost category definitions are listed in the Fig. 1 caption.
Percent area coverage by each permafrost category (calculated using the Brown et al. 1998 dataset) for each study basin. Permafrost category definitions are listed in the Fig. 1 caption.

Changes across these basins have been documented and studied most often between 1936 and 2000 (Berezovskaya et al. 2004; McClelland et al. 2004; Peterson et al. 2002). To develop an understanding of how the controls on streamflow trend vary in time, we analyze the consistency between precipitation and streamflow trends for a large number of periods between these years. These periods vary with start year and period length, with a minimum length of 20 yr. Selection of these periods is discussed in section 5c(2).

4. Data sources and development of a new precipitation product

a. Data sources

Precipitation, temperature, and streamflow data used in this study are summarized in Table 3. Hereafter, the datasets will be referenced using their respective dataset identification numbers (IDs; see column 2 of Table 3). Among these are two new products: 1) our bias-adjusted gridded precipitation product (UW), the development of which is described in section 4b (and compared to existing products in section 4c); and 2) the Adam et al. (2007) reconstructed streamflow product (Recon), for which the effects of reservoirs have been removed. This product was created by running a coupled hydrology–river routed model with and without a coupled reservoir model, wherein the reservoir model calculates historical reservoir storage and release release by maximizing hydropower production (Haddeland et al. 2006a, b, 2007). At the outlet of each of the regulated basins (see column 6 of Table 1), simulated streamflow without the reservoir model was subtracted from simulated streamflow with the reservoir model. This produced a time series of reservoir influences which were then subtracted from the observed basin-outlet discharge, constraining the reconstructed flow to be identical to observed flow before reservoir construction. Adam et al. (2007) document this product and compare it to other reconstructed streamflow products (McClelland et al. 2004; Yang et al. 2004a; Ye et al. 2003).

Table 3.

Climate and streamflow data used for the trend attribution study. All gridded products are in 0.5° geographic projection.

Climate and streamflow data used for the trend attribution study. All gridded products are in 0.5° geographic projection.
Climate and streamflow data used for the trend attribution study. All gridded products are in 0.5° geographic projection.

b. Development of new long-term precipitation data over the Eurasian Arctic

The inconsistency between precipitation and streamflow trends may partly be a result of temporal biases in the gridded precipitation products. Trend biases can result from biases in the station data themselves or from the gridding procedure. Trend biases in the source station data may be a result of a change in precipitation gauge type, a change in the environment surrounding the station, or a change in the way the data were processed and archived, such as occurred in the way that the Soviet government handled wetting biases in the mid-1960s (Groisman and Rankova 2001). A change in climate can also incur a bias in the temporal homogeneity of station precipitation because the degree to which a gauge underestimates precipitation is a function of climate (Forland and Hanssen-Bauer 2000; Yang et al. 2005). Temporal biases in gridded precipitation can also be caused by changes in the station network; that is, over the Eurasian Arctic, station density increased up to about 1970 and decreased in the 1990s (Shiklomanov et al. 2002). Rawlins et al. (2006) suggest that, averaged over the Eurasian Arctic, gridded precipitation bias is well over +10 mm yr−1 in the 1930s network as compared to the 1972 network when station density was greatest and, therefore, most representative of true spatially averaged precipitation.

Using the method of Hamlet and Lettenmaier (2005), we produce the UW gridded precipitation product for the Eurasian Arctic domain (defined as Eurasian land areas that discharge freshwater to the Arctic Ocean; the shaded regions in Fig. 2). The Hamlet and Lettenmaier method combines the best features of three existing products: 1) the high-frequency spatial and temporal variability of the UDel gridded dataset that uses all available stations for each month; 2) the low-frequency spatial and temporal variability of long-term stations selected from the Groisman bias-adjusted precipitation dataset; and 3) the mean monthly climatology of the AdamClim dataset that incorporates bias adjustments for gauge undercatch and orographic effects. The steps in the bias adjustment procedure are outlined below.

Fig. 2.

High-quality, long-term Groisman precipitation stations used for the Hamlet and Lettenmaier (2005) trend adjustment procedure. The station color represents the 1931–2000 precipitation trend slope [calculated according to Hirsch et al. (1982) for the seasonal Mann–Kendall test, see section 5c(1)]. The spatial domain of our gridded P product is given by the areas shaded in gray, e.g., the regions that discharge Q to the Arctic Ocean. Of the 410 stations shown, 210 fall within this domain.

Fig. 2.

High-quality, long-term Groisman precipitation stations used for the Hamlet and Lettenmaier (2005) trend adjustment procedure. The station color represents the 1931–2000 precipitation trend slope [calculated according to Hirsch et al. (1982) for the seasonal Mann–Kendall test, see section 5c(1)]. The spatial domain of our gridded P product is given by the areas shaded in gray, e.g., the regions that discharge Q to the Arctic Ocean. Of the 410 stations shown, 210 fall within this domain.

1) Choose data sources

The dataset is a combination of three products:

  • (i) High-density gridded product for short-term variability. We selected the UDel monthly time series, a dataset that makes use of most of the available station data (and therefore is associated with a relatively high-density station network, but is also susceptible to spurious trends due to changes in the gauging network).

  • (ii) High-quality station data for long-term variability. We used the Groisman daily station data. These station data have been temporally homogenized for changes in precipitation gauge type as well as a change in the way that wetting biases were accounted for by the Soviet government using the method of Groisman and Rankova (2001). Furthermore, these data incorporate temporally varying corrections for wind-induced precipitation undercatch using the method of Bogdanova et al. (2002a, b). Because we only select stations from the Groisman dataset that cover the majority of the period, we effectively remove the spurious trends incurred by changes in station density with time. A further consequence of this procedure is that the Groisman adjustments made for temporal inhomogeneities are also incorporated into the low-frequency variability of the final gridded UW product.

  • (iii) Monthly climatology (AdamClim). We used the annual climatology of Adam et al. (2006) in which the 0.5° grids have been adjusted for orographic effects. This dataset incorporates the mean monthly adjustments of Adam and Lettenmaier (2003) for gauge undercatch biases.

2) Process high-quality station data

The trend adjustment method is dependent on the high-quality station network [dataset (ii) above] being consistent for the entire study period, therefore it would be ideal to choose stations that have 100% temporal coverage between 1930 and 2000. Unfortunately, considerable gaps exist in many of the stations, so we relaxed the coverage requirement, choosing stations that had 75% temporal coverage between 1937 and 1996. This resulted in a total of 410 Russian stations, 210 of which were within our domain. Figure 2 shows the distribution of the selected stations and the 1931–2000 precipitation trend magnitude for each station [see section 5c(1) for a description of the trend calculation]. The results of this analysis may be questionable over regions with sparse station density, such as over northeastern Siberia, although these are the regions where there are few inferred long-term changes in precipitation. Therefore, although there is a sparse distribution of Groisman stations over these regions, the observed spatial coherency in long-term trends suggests that the large-scale features of low-frequency precipitation variability will be captured by the UW dataset.

We filled record gaps, using data from nearest neighbor stations, via the maintenance of variance extension type two method (MOVE.2; Hirsch 1982). The objective of MOVE.2 is to maintain sample mean and variance, and is an alternative to linear regression. Linear regression has the objective of minimizing the square of the errors of the estimated values, but underestimates sample variance (Hirsch 1982). This approach is reasonable if the gaps are distributed throughout the record period, but is problematic if used for record extension. Following gap filling, the stations were gridded to 0.5° using the Synagraphic Mapping System (SYMAP) of Shepard (1984).

3) Combine high-quality and high-density datasets

The motivation of our approach is to maintain as much spatial information from the high-density gridded product as possible while constraining the long-term variability to match that of the less dense high-quality gridded data (Hamlet and Lettenmaier 2005). Using a Butterworth filter (Hamming 1989), the time series for both the high-quality and high-density gridded data were temporally smoothed for each month individually (e.g., all Januaries between 1930 and 2000). We rescaled the unsmoothed high-density series by the ratios of the smoothed high-quality series to the smoothed high-density series.

4) Incorporate monthly climatology

For the period overlapping with the AdamClim monthly climatology (1979–99), differences were computed between the monthly AdamClim climatology and the climatology of the product from step 3, and these differences were added to the full-length monthly time series of the product from step 3, in which resulting quantities were constrained to positive values. We found that adding the difference of the climatologies, rather than rescaling by the ratio of the climatologies, allowed for the preservation of the desired long-term variability in the final product.

The results of this approach are summarized in Fig. 3, which shows the unsmoothed and smoothed annual time series for the high-density (black line), high-quality (dotted line), and final (dashed line) precipitation products, averaged over the three primary study basins. As desired, the smoothed time series of the final product matches that of the smoothed high-quality product and the final amounts reflect adjustments for biases in gauge undercatch and orographic effects on gridding [both of which can result in considerable underestimation of precipitation in cold and/or mountainous regions; see Adam and Lettenmaier (2003) and Adam et al. (2006)].

Fig. 3.

Development of the UW dataset (dashed line) is via combination of three other gridded products: 1) the UDel product for short-term variability (solid line), 2) the gridded Groisman station data for long-term variability (dotted line), and 3) the AdamClim monthly climatology (not shown). For each dataset are shown both the smoothed (using a Butterworth Filter; Hamming 1989) and unsmoothed annual time series. Although the procedure was performed on a grid-cell basis, results shown here are basin averages over the Lena, Yenisei, and Ob’.

Fig. 3.

Development of the UW dataset (dashed line) is via combination of three other gridded products: 1) the UDel product for short-term variability (solid line), 2) the gridded Groisman station data for long-term variability (dotted line), and 3) the AdamClim monthly climatology (not shown). For each dataset are shown both the smoothed (using a Butterworth Filter; Hamming 1989) and unsmoothed annual time series. Although the procedure was performed on a grid-cell basis, results shown here are basin averages over the Lena, Yenisei, and Ob’.

c. Comparison of precipitation and streamflow datasets

Figure 4 shows the annual anomaly time series and monthly climatologies for the UDel and UW precipitation products and the observed and reconstructed streamflow products. The reconstructed streamflow is shown only for the basins with upstream reservoirs (i.e., Lena, Yenisei, Ob’, and Ob2 subbasin). There is close agreement in the interannual variability among all datasets, with the damped variability in the streamflow with respect to precipitation in the Yeni1 and Ob2 subbasins reflecting a lower runoff ratio (0.15 as compared to 0.3–0.5 for the other basins). For streamflow, the largest divergences between the observed and reconstructed products are for the Yenisei, because of the high degree of regulation in that basin. For precipitation, differences in the mean monthly data reflect the adjustments for gauge undercatch (the largest effect during the winter season) and orographic effects (the largest effect during the summer season). Differences in low-frequency variability between the precipitation products reflect the procedure used to constrain the long-term UW trends to match those of the gridded Groisman station data.

Fig. 4.

For each of the 11 study basins (see Table 1), basin-mean annual anomaly time series and monthly climatologies for the UDel P dataset (black line) and the UW P dataset (red line). Also shown are the annual anomaly time series and monthly climatologies for the observed Q product (blue line) and the Adam et al. (2007) reconstructed Q product (green line). The monthly climatologies for all products were calculated for the common period of 1943–97.

Fig. 4.

For each of the 11 study basins (see Table 1), basin-mean annual anomaly time series and monthly climatologies for the UDel P dataset (black line) and the UW P dataset (red line). Also shown are the annual anomaly time series and monthly climatologies for the observed Q product (blue line) and the Adam et al. (2007) reconstructed Q product (green line). The monthly climatologies for all products were calculated for the common period of 1943–97.

To explore this further, we used the seasonal Mann–Kendall test [see section 5c(1)] to calculate the 1931–2000 trends in UDel and UW basin-average precipitation (Table 4) as well as for each 0.5° grid cell in the Eurasian Arctic domain (Fig. 5). Generally, over the Eurasian Arctic, trend adjustment caused the precipitation trends to become less positive. These results are contrary to the findings of Rawlins et al. (2006), who suggested that adjusting for spurious trends due to a changing station network would result in precipitation trends that are more positive. An explanation is that, by constraining our low-frequency variability to match that of the Groisman station data, we are adjusting for spurious trends that are a result not only of a changing station network but also the change in the way the data were preprocessed for wetting biases, as well as a change in the undercatch of precipitation as a function of a changing climate. Note that in regions with little or no coverage of Groisman stations (e.g., Scandinavia, northern Mongolia, and the Chukchi region of Russia), the trend adjustment procedure had the least effect (cf. Fig. 2 and Fig. 5), therefore the long-term trends in these regions generally match those in UDel precipitation.

Table 4.

Seasonal Mann–Kendall 1931–2000 trend significance and slope [see section 5c(1) for description of trend test] for P before (UDel) and after (UW) trend adjustment for each of the study basins. “NA” indicates the trend did not pass statistical significance at the lowest level tested; i.e., tested levels were 99%, 98%, 95%, 90%, 80%, and 60%. See Fig. 5 for spatial distributions of trend slope.

Seasonal Mann–Kendall 1931–2000 trend significance and slope [see section 5c(1) for description of trend test] for P before (UDel) and after (UW) trend adjustment for each of the study basins. “NA” indicates the trend did not pass statistical significance at the lowest level tested; i.e., tested levels were 99%, 98%, 95%, 90%, 80%, and 60%. See Fig. 5 for spatial distributions of trend slope.
Seasonal Mann–Kendall 1931–2000 trend significance and slope [see section 5c(1) for description of trend test] for P before (UDel) and after (UW) trend adjustment for each of the study basins. “NA” indicates the trend did not pass statistical significance at the lowest level tested; i.e., tested levels were 99%, 98%, 95%, 90%, 80%, and 60%. See Fig. 5 for spatial distributions of trend slope.
Fig. 5.

1931–2000 precipitation trend slope [calculated according to Hirsch et al. (1982) for the seasonal Mann–Kendall test; see section 5c(1)] for (a) the UDel dataset (i.e., before trend adjustment) and (b) the UW dataset (i.e., after trend adjustment).

Fig. 5.

1931–2000 precipitation trend slope [calculated according to Hirsch et al. (1982) for the seasonal Mann–Kendall test; see section 5c(1)] for (a) the UDel dataset (i.e., before trend adjustment) and (b) the UW dataset (i.e., after trend adjustment).

5. Streamflow trend attribution: Methods

a. Overview of analyses

The continuity equation applied to a watershed is

 
formula

where P, ET, and Q are the basin-average precipitation, evapotranspiration, and runoff, respectively, and dS is the net change in water stored in the basin (including surface and subsurface moisture) over the time increment dt. In section 2, we discussed numerous factors that may have contributed to the observed Q trends: 1) changes in dS/dt and/or ET due to reservoir construction and operation, 1) increase in P due to intensification of the hydrologic cycle, 3) decrease in dS/dt due to warming-induced permafrost degradation, and 4) changes in ET due to changes in P and/or temperature (T). The contributions of the reservoirs to Q trends can be analyzed by examining the differences in long-term trends between observed and reconstructed Q [see section 6b; also see Adam et al. (2007) for a more exhaustive analysis]. The remaining factors are induced by changes in P and/or T as follows. Changes in P will translate, to some degree, directly into changes in Q; that is, increasing P results in increasing Q and vice versa. Furthermore, P changes will affect Q via changes in ET for the moisture-limited basins. Finally, T changes affect Q either through changes in ET (for the energy-limited basins) or in dS/dt (via permafrost degradation or other moisture storage dynamics). Our objective (as related to goal 2 in section 1) is to evaluate when and where Q trends diverge from P trends, and to develop a set of most likely warming-induced mechanisms that may explain these divergences.

Using the data products described in Table 3, we performed the following analyses:

  1. Temperature/streamflow correlation analysis. The purpose of this analysis was to develop a hypothesis to explain inconsistencies between observed precipitation and streamflow trends. This was done by examining the correlation between streamflow and UDel temperature for each of the basins on an annual basis and for the summer and winter seasons. This allowed us to comment on when and where streamflow is most sensitive to variations in temperature, and therefore to improve our understanding of how temperature-induced changes may have caused streamflow trends to diverge from precipitation trends. Reconstructed streamflow data were used for the regulated basins and observed streamflow data were used for the other basins. This method is described in section 5b and the results are discussed in section 6a.

  2. Multiperiod trend analysis of individual variables. Trend analysis was performed for each of the following products: UW precipitation, UDel temperature, and observed and reconstructed streamflow. The purpose of this analysis was twofold: 1) to document the significant trends in historical climate and streamflow, and 2) to comment on the role of reservoirs in observed streamflow changes by comparing significant trends in observed and reconstructed streamflow. The trend analysis techniques are described in section 5c(1) and the multiperiod analysis is described in section 5c(2). The results are presented and discussed in section 6b.

  3. Trend compatibility analysis between UW precipitation and observed and reconstructed streamflow. Trends in precipitation and streamflow were compared for periods in which streamflow trends were statistically significant. The purpose of this analysis was to explore the most probable causes for inconsistencies between precipitation and streamflow trends, based on the hypothesis developed according to the temperature/streamflow correlation analysis (section 6a). The results are presented and discussed in section 6c.

b. Hypothesis formulation via temperature/streamflow correlation

Formulation of the hypothesis to explain inconsistencies between precipitation and streamflow trends was aided by correlation of basin-average temperature and streamflow time series between 1943 and 1997 (the period common to all streamflow records). We investigated the relationships between temperature and streamflow by plotting the Pearson product-moment correlation coefficients, “T/Q”, against basin-average mean annual temperature (Fig. 6; left panel) and percent continuous permafrost (Fig. 6; right panel). The T/Q correlations were calculated for the following time aggregations: 1) annual, defined as April 1 to March 31 in order to include the effects of summer permafrost thaw on winter discharge; 2) summer, defined as April 1 to September 31; 3) winter, defined as October 1 to March 31; and 4) summer/winter, in which the correlation was calculated between summer T and the following winter Q. Note that a correlation coefficient of +1 (−1) indicates a perfect increasing (decreasing) linear relationship, whereas a correlation of zero indicates that the data are not linearly dependent. The dotted lines in Fig. 6 indicate statistical significance at the 99% level. Although most of the results shown in the figure are not statistically significant, the general patterns are robust. Furthermore, we use these results only as a framework to develop a hypothesis to explain observed streamflow changes, rather than as a basis to form conclusions.

Fig. 6.

For each of the 11 study basins (see Table 1) and for the Q record common period of 1943–97, the Pearson product-moment correlation coefficient between basin-average time series of Q and UDel temperature plotted against (left) basin-average mean annual UDel T and (right) Brown et al. (1998) percent of continuous permafrost. Observed Q was used for the basins with no large upstream reservoirs, whereas the Adam et al. (2007) reconstructed Q was used for basins with upstream reservoirs. The correlation coefficient is between (a), (b) annual Q and annual temperature; (c), (d) summer Q and summer T; (e), (f) winter Q and winter T; and (g), (h) summer T and winter Q (for which annual in this case is defined as Apr 1–Mar 31, summer is defined as Apr 1–Sep 30, and winter is defined as Oct 1–Mar 31). Basins are sorted into three T regimes according to basin-average mean annual T; i.e., “cold” is for temperatures less than −8°C (white circles), “threshold” is for T between −2° and −8°C (gray circles), and “warm” is for temperatures greater than −2°C (black circles). Dotted lines indicate the value for which the Pearson product-moment correlation is statistically significant at 99%.

Fig. 6.

For each of the 11 study basins (see Table 1) and for the Q record common period of 1943–97, the Pearson product-moment correlation coefficient between basin-average time series of Q and UDel temperature plotted against (left) basin-average mean annual UDel T and (right) Brown et al. (1998) percent of continuous permafrost. Observed Q was used for the basins with no large upstream reservoirs, whereas the Adam et al. (2007) reconstructed Q was used for basins with upstream reservoirs. The correlation coefficient is between (a), (b) annual Q and annual temperature; (c), (d) summer Q and summer T; (e), (f) winter Q and winter T; and (g), (h) summer T and winter Q (for which annual in this case is defined as Apr 1–Mar 31, summer is defined as Apr 1–Sep 30, and winter is defined as Oct 1–Mar 31). Basins are sorted into three T regimes according to basin-average mean annual T; i.e., “cold” is for temperatures less than −8°C (white circles), “threshold” is for T between −2° and −8°C (gray circles), and “warm” is for temperatures greater than −2°C (black circles). Dotted lines indicate the value for which the Pearson product-moment correlation is statistically significant at 99%.

c. Trend analysis

1) Selection of trend test

To examine trends in annual time series, we considered two nonparametric tests: Mann–Kendall on the annual sums (MK; Mann 1945) and the seasonal Mann–Kendall test (SMK; Hirsch et al. 1982; Lettenmaier et al. 1994). We opted to use SMK because it is sensitive to seasonal differences in that the statistic is calculated by summing over the seasonal statistics rather than the seasonal flows as in MK (Hirsch et al. 1982). This is particularly important for applications in the Arctic because the most significant changes are occurring during the cold season when the variance is lowest, particularly for river discharge. In this context, an advantage of SMK is that trends occurring during the low-flow season are not obscured by the large variability of the high-flow season (Lettenmaier et al. 1994).

We estimated the slope of the trend, B, for the data, X, according to the idea of Hirsch et al. (1982) as the median value of all Dijk, which are determined by

 
formula

in which i and j are indices for year, k is an index for season, ns is the number of seasons (i.e., 12 months, in our case), and nk is the number of observations for season k (note that D is positive for a positive trend and vice versa). Therefore B is sensitive to changes occurring in a particular season because the differences are always computed between values that are multiples of ns seasons apart (Lettenmaier et al. 1994). Throughout this manuscript, we refer to B as the “seasonal Mann–Kendall trend slope”. Note that, if ns = 1, the formulation for SMK collapses to MK [i.e., omitting k in Eq. (2)].

We demonstrate the behavior of MK versus SMK by plotting histograms of the differences, D, for an example with especially strong winter changes, 1937–98 Yenisei observed streamflow (Fig. 7). Histograms are plotted for individual month MKs, annual MK, and SMK (note that combining all of the histograms for the individual months produces the histogram shown for SMK). Annual MK is unable to capture the large December to April trend slope, whereas SMK is (i.e., the center of mass is shifted to the right).

Fig. 7.

Histograms of the differences Dij (mm month−1; MK statistics for individual months: “Oct”–”Sep”; and the MK statistic for annual streamflow: “Ann, MK”) and Dijk (for the SMK statistic: “Ann, SMK”) used to determine the Mann–Kendall slope estimator B according to Hirsch et al. (1982), i.e., B is the median of the Dij for MK and the Dijk for SMK. Results are shown for observed streamflow at the outlet of the Yenisei for the period of 1937–98.

Fig. 7.

Histograms of the differences Dij (mm month−1; MK statistics for individual months: “Oct”–”Sep”; and the MK statistic for annual streamflow: “Ann, MK”) and Dijk (for the SMK statistic: “Ann, SMK”) used to determine the Mann–Kendall slope estimator B according to Hirsch et al. (1982), i.e., B is the median of the Dij for MK and the Dijk for SMK. Results are shown for observed streamflow at the outlet of the Yenisei for the period of 1937–98.

Trend tests were performed for the hydrological “water year,” October through September. Trend slope units for the moisture flux variables are given as mm month−1 yr−1 (for monthly series) or mm yr−2 (for annual series). To be consistent with the literature, we use the short-hand unit of mm yr−1 for both cases (i.e., this should be interpreted as mm month−1 yr−1 for monthly data and mm yr−2 for annual data). Because the SMK trend slope is calculated using monthly data, we convert the final trend slope to the proper annual units by multiplying by 12 (for moisture flux variables only and unless stated otherwise, as in Fig. 7).

2) Multiperiod trend analysis of individual variables

Because the controls on streamflow variability operate at varying time-scales and in different periods, we determined trend significance and slope for a range of start and end dates. The periods were selected systematically: for every start year between 1936 and 2000, a set of end years was selected for period lengths in increments of 5 yr up until 2000 (with a minimum period length of 20 yr). Therefore, trends were tested for the following periods: 1936–56, 1936–61, 1936–66, . . . , 1937–57, 1937–62, 1937–67, etc. This results in a total of 225 potential periods, although many of the analyses utilize a subset of these periods, depending on the period of record for the dataset under consideration. Following Lettenmaier (1976) and Ziegler et al. (2003), we selected the minimum period length of 20 yr as the number of years needed to detect a 2% yr−1 streamflow trend for a significance level of 99% [cf. Eq. (3), Ziegler et al. 2003]. Trend testing was performed for precipitation, temperature, and streamflow using the SMK method [section 5c(1)] for a significance level of 99%.

6. Streamflow trend attribution: Results and discussion

a. Hypothesis formulation via temperature/streamflow correlation

Figures 6a,b suggest that the basins can be segregated into one of three temperature regimes using thresholds of approximately −8° and −2°C. First, the open circles are the “cold” basins (the Lena basin and the Lena1 subbasin), the only basins with at least 80% continuous permafrost coverage. Second, the gray circles are the “threshold” basins (the Lena2 and Lena3 subbasins and the Yenisei basin and its subbasins) that are underlain by a minimum of 30% continuous permafrost. Third, the black circles are the “warm” basins (the Ob’ basin and subbasins and the Severnaya Dvina basin) that are primarily in the region of seasonally frozen soil. The cold and threshold basins generally have positive correlation coefficients, indicating that as temperature increases, streamflow also increases, likely due to the melting of ground ice. The warm basins generally have negative T/Q correlations, indicating that as temperature increases, streamflow decreases, likely due to increased ET.

During the summer (Figs. 6c,d), T/Q correlations are negative for the warm basins (as a result of temperature effects on ET); and increase with percent continuous permafrost for the permafrost basins, such that the T/Q correlations are negligible for the coldest basins (possibly due to less moisture available for ET in the thinner active layers). Winter streamflow from the warm basins is minimally affected by winter temperature variations and inversely affected by summer temperature variations, possibly due to drier soil conditions at the beginning of the winter (Figs. 6e–h). Winter streamflow from the permafrost basins is positively affected by both winter and summer temperature, varying linearly with percent continuous permafrost (Figs. 6e–h). This suggests that, in basins with extensive permafrost, warming effects on discharge are manifested primarily during the winter, consistent with Serreze et al. (2003a) and Zhang et al. (2003), who hypothesized that a delayed active-layer freeze-up (partially due to a deepening active layer) leads to a delay in the release of soil moisture storage into the winter.

Overall, these results indicate that the way each region responds to precipitation and temperature changes is at least partially dependent on the region’s mean climate and permafrost state. Based on our results we formulated the following hypothesis. We expect the regions with extensive continuous permafrost (the “cold” regions) to be sensitive to warming via the melt and release of excess ground ice (according to the degree of ice richness of the permafrost); therefore, streamflow increase may exceed precipitation increase. Furthermore, we expect that the release of this meltwater is sensitive to both winter and summer temperature variations and should occur primarily during the winter. We expect the ET in “warm” regions to increase with warming and/or increased precipitation; therefore, precipitation increases may exceed streamflow increases. Streamflow from the “threshold” basins, especially in regions with less extensive permafrost, is most likely affected by both the melting of ground ice and ET changes. Therefore, it is difficult to predict the effects of temperature and precipitation changes on “threshold” basin streamflow.

b. Multiperiod trend analysis of individual variables

A summary of the multiperiod trend analysis is given in Table 5. The table shows the number of periods with significant trends for each basin and dataset, and qualitative descriptions of the signs of the trends. In general, there is no clear regional signature of precipitation trends, but temperature generally increased across the domain, as did streamflow. The trend results for the observed and reconstructed streamflow are plotted in Fig. 8. Each line in the figure represents a period for which the trend is statistically significant; the trend slope is given by the line color. This plot can be thought of as an alternate way to view a time series, wherein the objective is to view the significant changes in a noisy series. The results demonstrate that the longest periods with significant trends have been positive and the largest positive trends occurred during the later part of the record, starting around the 1970s, indicating that streamflow increases accelerated up until the end of the last century. Streamflow increases have not been monotonic since the 1930s; many of the basins display negative trends during the 1960s and 1970s, and a few of the basins show negative trends beginning prior to the 1960s.

Table 5.

The number of periods for which trends are significant at 99% and a qualitative description of the sign of these trends for UW P, UDel T, observed Q, and reconstructed Q for each of the study basins. The total number of tested periods for P, T, and reconstructed Q is 225. The total number of tested periods for observed Q depends on the period of record for each basin. This number is given after the number of significant trends in the “Count” column for observed Q. “NA” indicates no major reservoirs for that basin (Table 1), and “—” indicates no periods with significant trends. See Figs. 8 and 9 for visualizations of these trends.

The number of periods for which trends are significant at 99% and a qualitative description of the sign of these trends for UW P, UDel T, observed Q, and reconstructed Q for each of the study basins. The total number of tested periods for P, T, and reconstructed Q is 225. The total number of tested periods for observed Q depends on the period of record for each basin. This number is given after the number of significant trends in the “Count” column for observed Q. “NA” indicates no major reservoirs for that basin (Table 1), and “—” indicates no periods with significant trends. See Figs. 8 and 9 for visualizations of these trends.
The number of periods for which trends are significant at 99% and a qualitative description of the sign of these trends for UW P, UDel T, observed Q, and reconstructed Q for each of the study basins. The total number of tested periods for P, T, and reconstructed Q is 225. The total number of tested periods for observed Q depends on the period of record for each basin. This number is given after the number of significant trends in the “Count” column for observed Q. “NA” indicates no major reservoirs for that basin (Table 1), and “—” indicates no periods with significant trends. See Figs. 8 and 9 for visualizations of these trends.
Fig. 8.

Streamflow trend plots for all periods with trends significant at 99% using SMK. Each line represents a period for which the trend slope is given by the color of the line and the period length is given by the length of the line (starting and ending at the start and end of the period, respectively). Each panel is labeled according to basin reference name and ID (see Table 1) and whether the Q product is observed (Obs) or reconstructed (Recon).

Fig. 8.

Streamflow trend plots for all periods with trends significant at 99% using SMK. Each line represents a period for which the trend slope is given by the color of the line and the period length is given by the length of the line (starting and ending at the start and end of the period, respectively). Each panel is labeled according to basin reference name and ID (see Table 1) and whether the Q product is observed (Obs) or reconstructed (Recon).

By comparing the observed and reconstructed panels, Fig. 8 gives an indication of how the reservoirs have affected streamflow trends. For all four of the regulated basins (Lena, Yenisei, Ob’, and Ob2), observed streamflow trends were positive for all periods; furthermore, the number of periods with significant changes was larger for observed streamflow than for reconstructed streamflow. Recall that the SMK test [section 5c(1)] is sensitive to significant changes occurring during individual months. The relative abundance of positive trends for the observed flows (as compared to the reconstructed flows) in the regulated basins is due to large increases in the winter low flows as a result of reservoir operations. Adam et al. (2007) demonstrate that winter flows at the outlets of the Lena, Yenisei, and Ob’ basins increased by 20%–70% as a result of reservoir operations. For example, although there exists only one large reservoir in the Lena basin, the effects of this regulation caused a 500 m3 s−1 winter flow increase at the outlet of the Lena basin, corresponding to a 50% increase in late winter streamflow (Adam et al. 2007). The reconstructed streamflow trends for the regulated basins produced the same general features as observed streamflow for the unregulated basins, with negative trends occurring during the beginning or middle parts of the record and positive trends occurring during the later part of the record. The exception is for the Ob’ basin, in which there were no periods with significant changes in reconstructed streamflow, indicating that the significant changes in observed streamflow were entirely due to reservoirs. Adam et al. (2007) present a more thorough comparison of observed to reconstructed streamflow trends, using the traditional MK test on annual and seasonal streamflow, in contrast to the results presented here using the SMK statistic (see section 2a). Recall from section 5c(1) that the SMK statistic may capture some trends that the MK statistic does not because it is sensitive to seasonal differences in trend. This explains why Adam et al. (2007) detected no significant trends in observed annual Ob’ discharge, whereas several positive trends were detected here using the SMK statistic. These positive trends may reflect the large increase in winter discharge that occurred after the construction of reservoirs in 1956 and 1986 (Adam et al. 2007).

Figures 9a,b show the results of the trend analysis for UDel temperature and UW precipitation, respectively. The figures show that there has been a general long-term warming over the region, with exception of the Severnaya Dvina basin. Furthermore, warming appears to have accelerated with the largest warming occurring since the 1970s, although a few of the basins show a period of cooling between around 1980 and 2000. The few significant long-term precipitation increases are in the Lena and Lena1 basins prior to the 1980s. Because most of the literature suggests that precipitation must be a, if not the, major factor leading to streamflow increases (see section 2b), the lack of observed significant precipitation increases has made streamflow trend attribution difficult in these regions. An explanation for some of the study basins is that, other than the snowmelt period, there is considerably more variability in precipitation than there is in streamflow (e.g., for the winter months, the standard deviation in the Ob2 monthly streamflow time series is approximately 20% that of precipitation). Therefore, increases in precipitation may be occurring that can more than account for some of the observed changes in streamflow but because of the large variance, they are not statistically significant. This issue is examined in section 6c.

Fig. 9.

Trend plots for (a) UDel T and (b) UW P (see Fig. 8 caption for explanation).

Fig. 9.

Trend plots for (a) UDel T and (b) UW P (see Fig. 8 caption for explanation).

c. Comparison of precipitation and streamflow trends

In section 6b, we examined precipitation trends for periods in which they are statistically significant. Here, we examine the precipitation trends for the periods in which the streamflow trends were statistically significant (Fig. 10). Comparing and contrasting the line colors in Figs. 8 and 10 gives an indication of the importance of precipitation to streamflow change for each basin and for each period. As a further comparison, we created scatterplots of streamflow trends versus precipitation trends (Fig. 11). The black circles represent the periods with statistically significant changes at 99% (e.g., the same periods as in Figs. 8 and 10), while the gray circles are for changes significant at 90% but not at 99%. These scatterplots can be interpreted as follows: points along the one-to-one line indicate that precipitation may be contributing directly (without changes to ET or dS/dt, unless they are competing effects) to streamflow change (e.g., Lena2); a series of points that form a line that is rotated clockwise away from the one-to-one line indicate that precipitation is contributing to streamflow change, but that ET may also be changing either through changes in precipitation and/or temperature (e.g., Ob1); groups of points that are scattered in the upper halves of the plots, where streamflow trends are positive regardless of the sign of precipitation trends, indicate that there may be a release of moisture from storage, likely through warming-induced permafrost degradation (e.g., Lena1). Several of the plots show a combination of effects; for example, the Lena reconstructed streamflow trends appear to be caused by precipitation changes, increasing ET, as well as storage release effects.

Fig. 10.

UW P trend plots for all periods in which Q trends are significant at 99% using SMK. (See Fig. 8 caption for further explanation).

Fig. 10.

UW P trend plots for all periods in which Q trends are significant at 99% using SMK. (See Fig. 8 caption for further explanation).

Fig. 11.

Scatterplots of trend slopes between Q and UW P for all periods in which Q trends are significant at 90% (gray circles) and 99% (black circles) using SMK. Each panel is labeled according to basin reference name and ID (see Table 1) and whether the Q product is observed (Obs) or reconstructed (Recon). The R2 value for each set of pairs (for 99% significance) is shown in the lower right corner.

Fig. 11.

Scatterplots of trend slopes between Q and UW P for all periods in which Q trends are significant at 90% (gray circles) and 99% (black circles) using SMK. Each panel is labeled according to basin reference name and ID (see Table 1) and whether the Q product is observed (Obs) or reconstructed (Recon). The R2 value for each set of pairs (for 99% significance) is shown in the lower right corner.

Table 6 summarizes the most probable controls on long-term streamflow changes for each of the study basins. The table was based on information gathered from Figs. 8, 10 and 11, and the hypothesis formulated in section 6a. In the following paragraphs, each set of basins (cold, threshold, and warm) are discussed. Because we focus the remaining discussion on climate-induced changes, hereafter we will discuss only the reconstructed streamflow results for the regulated basins (Lena, Yenisei, Ob’, and Ob2), as well as the observed streamflow results for the unregulated basins.

Table 6.

Summary of section 6 results and hypothesized primary controls for each of the study basins. The columns are derived from the following sources: column 3 from Fig. 6, columns 4 and 5 from Table 2, and columns 6 and 7 from Fig. 11. “NA” indicates lack of reconstructed data for these basins (which do not contain large reservoirs, see Table 1), and “—” indicates no periods with significant trends.

Summary of section 6 results and hypothesized primary controls for each of the study basins. The columns are derived from the following sources: column 3 from Fig. 6, columns 4 and 5 from Table 2, and columns 6 and 7 from Fig. 11. “NA” indicates lack of reconstructed data for these basins (which do not contain large reservoirs, see Table 1), and “—” indicates no periods with significant trends.
Summary of section 6 results and hypothesized primary controls for each of the study basins. The columns are derived from the following sources: column 3 from Fig. 6, columns 4 and 5 from Table 2, and columns 6 and 7 from Fig. 11. “NA” indicates lack of reconstructed data for these basins (which do not contain large reservoirs, see Table 1), and “—” indicates no periods with significant trends.

Many of the periods for the cold basins (Lena and Lena1) are scattered above the one-to-one line in Fig. 11, indicating that positive streamflow trends exceeded precipitation trends. This suggests another source of water, most likely release from storage due to permafrost degradation. Comparison of Figs. 8 and 10 gives us an indication of the year that permafrost influences may have started. For the Lena basin, the positive streamflow trends beginning in the 1960s cannot be explained by the negative or slightly positive precipitation trends during those periods. For the Lena1 subbasin, the mismatch between streamflow and precipitation trend signs began in the mid-1940s but became most apparent in the mid-1950s and increasingly so up until 2000. Therefore, permafrost degradation effects may have begun in the 1960s for the Lena and in the 1950s for the Lena1. The Lena1 is a subbasin of the Lena, so it may have taken an additional 10 yr for the permafrost signature apparent in the Lena1 to be the dominant effect for the entire basin. There are also a number of points scattered along or to the right of the one-to-one line in Fig. 11, suggesting that precipitation combined with ET effects was the dominant control for these periods. Figures 8 and 10 suggest that, for the Lena, the positive streamflow trends between the 1940s and 1960s and the negative streamflow trends between the 1950s and 1970s are likely P/ET induced. For the Lena1, positive streamflow trends starting in the 1940s may have been P/ET induced, but for periods starting in the 1950s to mid-1960s streamflow trends were likely a combination of precipitation and permafrost effects, and permafrost effects became dominant for periods starting in the late-1960s. These results are in agreement with Frauenfeld et al. (2004), who show that the positive trend in active-layer thickness (averaged over Russian permafrost regions) has a greater slope for the 1956–90 period as compared to the 1930–90 period, indicating an increased thawing rate since the mid-1950s. This observed increase in permafrost thaw rate may explain the divergence of streamflow and precipitation trends beginning in the mid-1950s to 1960s in the Lena and Lena1 basins, as well as some of the threshold permafrost basins (e.g., Yenisei and Lena3; see below).

Figures 8, 10 and 11 indicate that precipitation, ET, and dS/dt changes have all, in varying degrees, influenced streamflow changes in the threshold basins (the Yenisei and its subbasins and two of the Lena’s subbasins: Lena2 and Lena3). For example, Yenisei streamflow trends were consistently more positive than precipitation trends (Fig. 11). This is in agreement with Serreze et al. (2003a), who showed divergent trends in streamflow and P/ET for the Yenisei. They suspected that permafrost thaw was the primary cause. Unlike the cold basins, there is no general picture of key mechanisms of change. For example, precipitation changes in the Lena2 subbasin appear to be the primary control for the entire period of record (i.e., the points in Fig. 11 are scattered along the one-to-one line), whereas permafrost effects may have also played a role for changes in Lena3 streamflow (i.e., the points in Fig. 11 are shifted upward from the one-to-one line). Recall that the Lena3 is a subbasin of Lena2, occupying approximately 50% of the Lena2 area. Compared to the Lena2 as a whole, the Lena3 is slightly warmer and is underlain by larger percentages of sporadic and isolated permafrost types, which may be more sensitive to warming. The Yenisei and its subbasins all appear to have been influenced by both precipitation and permafrost effects, with the largest discrepancies between streamflow and precipitation trends in the Yenisei and Yeni2 basins (Fig. 11). Compared to the Yeni1, the Yeni2 is colder and is underlain by larger percentages of continuous and discontinuous permafrost types. This is the opposite relationship of that observed between the Lena2 and Lena3 subbasins, which indicates that permafrost type, according to the definition given in Fig. 1, is not the sole permafrost characteristic needed to understand streamflow changes in permafrost basins. For example, it is possible that the Yeni2 basin is underlain with permafrost that has a greater volumetric fraction of ice as compared to the Yeni1, which may explain the greater permafrost role in observed streamflow changes in that basin.

Figure 11 suggests that the streamflow changes in the warm basins (the Ob’ subbasins and the Severnaya Dvina) were due to changes in precipitation and ET; that is, the points are scattered along a straight line that is rotated to the right from the one-to-one line. Therefore, although not significant, precipitation trends in each of these basins can more than account for the majority of the significant streamflow trends, especially for the positive trends. The remainder of the additional precipitation likely contributed toward increased basin ET. Of the three basin types (cold, threshold, and warm), the clearest consistent relationship between streamflow and precipitation trends is for the warm or nonpermafrost basins. This suggests that the presence of permafrost in a basin is likely the most important complicating factor in understanding the primary controls on observed streamflow changes for that basin.

7. Conclusions

Observed streamflow changes in the northward-flowing Eurasian rivers may be due to climatic influences, mainly through changing precipitation and temperature, or to human influences, mainly through reservoir construction and operation. We explored the mechanisms controlling observed streamflow changes using trend analysis of precipitation, temperature, and streamflow time series using the seasonal Mann–Kendall (SMK) trend test. Before exploring the role of climate in observed trends, the role of the reservoirs was identified. As shown by Adam et al. (2007), reservoirs had the greatest effect on streamflow seasonality, by increasing winter streamflow and decreasing summer streamflow. Because the SMK test is sensitive to seasonal changes, there were large differences between the observed and reconstructed streamflow trend results for the regulated basins. For the Lena, Yenisei, and Ob2 basins, significant observed streamflow trends were positive for all periods, whereas significant reconstructed streamflow trends were a mix of negative and positive trends, in which periods with negative trends ended prior to the mid-1980s. This is consistent with the trend plots for most of the unregulated basins (Fig. 8). There were no periods with significant trends for Ob’ reconstructed streamflow, indicating that the significant trends for observed streamflow were likely a result of reservoir effects on streamflow seasonality.

The attribution of observed streamflow changes to climatic influences is complicated by the lack of significant precipitation trends in northern Eurasia. As discussed in section 2b, much of the literature suggests that changes in precipitation should be a primary contributor to the observed streamflow increases, but there is an inconsistency between precipitation and streamflow changes for many of the basins. Our analysis has identified when and where precipitation and streamflow trends are divergent in sign and/or magnitude, and how these divergences may be attributed to changes in temperature. By correlating annual time series of temperature to streamflow (reconstructed data for regulated basins and observed data for unregulated basins), we advanced the hypothesis that warming could lead to differing agents of change for streamflow variability depending on the mean climate and permafrost state of the basin: melt of ground ice for basins in northeastern Siberia (leading to streamflow increases exceeding precipitation increases), ET effects for basins in European Russia and western Siberia (leading to precipitation changes exceeding streamflow changes), and competing effects for basins in the central threshold regions. Our trend analysis results indicated that the most likely controls on streamflow changes are generally in agreement with this hypothesis. For the “cold” Lena and Lena1 basins in northeastern Siberia, inconsistencies between streamflow and precipitation trends began in the 1950s and 1960s, and permafrost melt is the most probable explanation for this additional streamflow. The divergence between precipitation and streamflow trends accelerated up until the end of the last century. For the “warm” basins largely outside of the permafrost region (the Ob’ subbasins and the Severnaya Dvina basin in western Siberia), there is a clear and consistent conclusion that precipitation changes coupled to ET changes are the likely explanation for the observed streamflow trends; that is, precipitation trends nearly always exceeded streamflow trends in absolute magnitude. For the “threshold” basins in central Siberia (the Lena2 and Lena3 subbasins and the Yenisei and its subbasins), there is less agreement in the most likely controls on streamflow changes, suggesting that permafrost and ET effects were competing in causing streamflow trends to be inconsistent with precipitation trends. For example, there is a positive divergence between streamflow and precipitation trends for nearly all of the periods for the Yenisei, Lena3, and Yeni2 basins, suggesting that permafrost degradation played a primary role. Alternatively, for the Lena2 and Yeni1 subbasins, some of the streamflow trends were less than that of precipitation, suggesting that ET played a primary role. For the periods when streamflow trends matched that of precipitation trends, it is possible that ET effects canceled out the effects of permafrost degradation or that neither were having a significant contribution. One conclusion of this study is that the presence of permafrost is a complicating factor in understanding long-term streamflow changes for river basins in the Eurasian Arctic. The degree to which warming-induced permafrost degradation played a role in observed streamflow changes depends not only on the extent of permafrost in the basin, but also the temperature and ice richness of the permafrost, and likely other factors.

The improved understanding of the controls on northern Eurasian streamflow resulting from this analysis provides a motivation to enhance the modeling framework used for the prediction of Arctic streamflow. For example, we now understand that permafrost most likely plays a key role in long-term streamflow variability. Most large-scale land surface hydrology models poorly simulate winter base flow at high latitudes due to the presence of permafrost (Su et al. 2005), and these results provide motivation to focus our efforts on improving the permafrost algorithms in these models. Incorporation and validation of the key physical processes into the modeling framework is crucial for the accurate prediction of future river discharge rates into the Arctic Ocean and feedbacks to the climate system, because these processes respond differently to climatic changes. For example, increased precipitation due to an accelerated hydrologic cycle will likely continue to provide additional freshwater to the system, whereas warming-induced melting of permafrost provides freshwater only until Arctic permafrost is completely melted.

Acknowledgments

The authors thank Dr. Svetlana Berezovskaya for her careful reading of the manuscript, Dr. Daqing Yang for numerous conversations, and three anonymous reviewers for their thoughtful suggestions. This research was supported by NSF Grant OP-0230372 to the University of Washington.

REFERENCES

REFERENCES
ACIA
,
2005
:
Arctic Climate Impact Assessment.
Cambridge University Press, 1042 pp
.
Adam
,
J. C.
, and
D. P.
Lettenmaier
,
2003
:
Adjustment of global gridded precipitation for systematic bias.
J. Geophys. Res.
,
108
.
4257, doi:10.1029/2002JD002499
.
Adam
,
J. C.
,
E. A.
Clark
,
D. P.
Lettenmaier
, and
E. F.
Wood
,
2006
:
Correction of global precipitation products for orographic effects.
J. Climate
,
19
,
15
38
.
Adam
,
J. C.
,
I.
Haddeland
,
F.
Su
, and
D. P.
Lettenmaier
,
2007
:
Simulation of reservoir influences on annual and seasonal streamflow changes for the Lena, Yenisei, and Ob’ rivers.
J. Geophys. Res.
,
112
.
D24114, doi:10.1029/2007JD008525
.
Arnell
,
N. W.
,
2005
:
Implications of climate change for freshwater inflows to the Arctic Ocean.
J. Geophys. Res.
,
110
.
D07105, doi:10.1029/2004JD005348
.
Barnett
,
T. P.
,
J. C.
Adam
, and
D. P.
Lettenmaier
,
2005
:
Potential impacts of a warming climate on water availability in snow-dominated regions.
Nature
,
438
,
303
309
.
Berezovskaya
,
S.
,
D. Q.
Yang
, and
D. L.
Kane
,
2004
:
Compatibility analysis of precipitation and runoff trends over the large Siberian watersheds.
Geophys. Res. Lett.
,
31
.
L21502, doi:10.1029/2004GL021277
.
Bogdanova
,
E. G.
,
V. S.
Golubev
,
B. M.
Ilyin
, and
I. V.
Dragomilova
,
2002a
:
A new model for bias correction of precipitation measurements, and its application to polar regions of Russia.
Russ. Meteor. Hydrol.
,
10
,
68
94
.
Bogdanova
,
E. G.
,
B. M.
Ilyin
, and
I. V.
Dragomilova
,
2002b
:
Application of a comprehensive bias-correction model to precipitation measured at Russian North Pole drifting stations.
J. Hydrometeor.
,
3
,
700
713
.
Broecker
,
W. S.
,
1997
:
Thermohaline circulation, the Achilles heel of our climate system: Will man-made CO2 upset the current balance?
Science
,
278
,
1582
1588
.
Brown
,
J.
,
O. J.
Ferrians
Jr.
,
J. A.
Heginbottom
, and
E. S.
Melnikov
,
cited
.
1998
:
Circum-Arctic Map of Permafrost and Ground-Ice Conditions. National Snow and Ice Data Center/World Data Center for Glaciology, Boulder, CO, digital media. [Available online at http://nsidc.org/data/ggd318.html.]
.
Clark
,
P. U.
,
N. G.
Pisias
,
T. F.
Stocker
, and
A. J.
Weaver
,
2002
:
The role of the thermohaline circulation in abrupt climate change.
Nature
,
415
,
863
869
.
Curry
,
R.
,
B.
Dickson
, and
I.
Yashayaev
,
2003
:
A change in the freshwater balance of the Atlantic Ocean over the past four decades.
Nature
,
426
,
826
829
.
Forland
,
E. J.
, and
I.
Hanssen-Bauer
,
2000
:
Increased precipitation in the Norwegian Arctic: True or false?
Climatic Change
,
46
,
485
509
.
Frauenfeld
,
O. W.
,
T. J.
Zhang
,
R. G.
Barry
, and
D.
Gilichinsky
,
2004
:
Interdecadal changes in seasonal freeze and thaw depths in Russia.
J. Geophys. Res.
,
109
.
D05101, doi:10.1029/2003JD004245
.
Georgievskii
,
V. Y.
,
A. V.
Ezhov
,
A. L.
Shalygin
,
I. A.
Shiklomanov
, and
A. I.
Shiklomanov
,
1996
:
Assessment of the effect of possible climate changes on hydrological regime and water resources of rivers in the former USSR.
Russ. Meteor. Hydrol.
,
11
,
66
74
.
Groisman
,
P. Ya
, and
E. Y.
Rankova
,
2001
:
Precipitation trends over the Russian permafrost-free zone: Removing the artifacts of pre-processing.
Int. J. Climatol.
,
21
,
657
678
.
Groisman
,
P. Ya
,
V. V.
Koknaeva
,
T. A.
Belokrylova
, and
T. R.
Karl
,
1991
:
Overcoming biases of precipitation measurement: A history of the USSR experience.
Bull. Amer. Meteor. Soc.
,
72
,
1725
1733
.
Haddeland
,
I.
,
T.
Skaugen
, and
D. P.
Lettenmaier
,
2006a
:
Anthropogenic impacts on continental surface water fluxes.
Geophys. Res. Lett.
,
33
.
L08406, doi:10.1029/2006GL026047
.
Haddeland
,
I.
,
D. P.
Lettenmaier
, and
T.
Skaugen
,
2006b
:
Effects of irrigation on the water and energy balances of the Colorado and Mekong river basins.
J. Hydrol.
,
324
,
210
223
.
Haddeland
,
I.
,
T.
Skaugen
, and
D. P.
Lettenmaier
,
2007
:
Hydrologic effects of land and water management in North America and Asia: 1700–1992.
Hydrol. Earth Syst. Sci.
,
11
,
1035
1045
.
Hamlet
,
A. F.
, and
D. P.
Lettenmaier
,
2005
:
Production of temporally consistent gridded precipitation and temperature fields for the continental United States.
J. Hydrometeor.
,
6
,
330
336
.
Hamming
,
R. W.
,
1989
:
Digital Filters.
3rd ed. Prentice Hall, 284 pp
.
Hinzman
,
L. D.
, and
Coauthors
,
2005
:
Evidence and implications of recent climate change in northern Alaska and other arctic regions.
Climatic Change
,
72
,
251
298
.
Hirsch
,
R. M.
,
1982
:
A comparison of four streamflow record extension techniques.
Water Resour. Res.
,
18
,
1081
1088
.
Hirsch
,
R. M.
,
J. R.
Slack
, and
R. A.
Smith
,
1982
:
Techniques of trend analysis for monthly water-quality data.
Water Resour. Res.
,
18
,
107
121
.
Lammers
,
R. B.
, and
A. I.
Shiklomanov
,
cited
.
2000
:
R-ArcticNET: A Regional Hydrographic Data Network for the pan-Arctic Region. [Available online at http://www.r-arcticnet.sr.unh.edu/abstract.html.]
.
Lawrence
,
D. M.
, and
A. G.
Slater
,
2005
:
A projection of severe near-surface permafrost degradation during the 21st century.
Geophys. Res. Lett.
,
32
.
L24401, doi:10.1029/2005GL025080
.
Lettenmaier
,
D. P.
,
1976
:
Detection of trends in water quality data from records with dependent observations.
Water Resour. Res.
,
12
,
1037
1046
.
Lettenmaier
,
D. P.
,
E. F.
Wood
, and
J. R.
Wallis
,
1994
:
Hydro-climatological trends in the continental United States, 1948–88.
J. Climate
,
7
,
586
607
.
Manabe
,
S.
, and
R. J.
Stouffer
,
1980
:
Sensitivity of a global climate model to an increase in CO2 concentration in the atmosphere.
J. Geophys. Res.
,
85
,
5529
5554
.
Mann
,
H. B.
,
1945
:
Nonparametric tests against trend.
Econometrica
,
13
,
245
259
.
Matsuura
,
K.
, and
C. J.
Willmott
,
cited
.
2005
:
Arctic land-surface precipitation: 1930–2004 gridded monthly time series (V. 1.03). [Available online at http://climate.geog.udel.edu/~climate/html_pages/Arctic4_files/README.arctic.p_ts4.html.]
.
McClelland
,
J. W.
,
R. M.
Holmes
,
B. J.
Peterson
, and
M.
Stieglitz
,
2004
:
Increasing river discharge in the Eurasian Arctic: Consideration of dams, permafrost thaw, and fires as potential agents of change.
J. Geophys. Res.
,
109
.
D18102, doi:10.1029/2004JD004583
.
Mekis
,
E.
, and
W. D.
Hogg
,
1999
:
Rehabilitation and analysis of Canadian daily precipitation time series.
Atmos.–Ocean
,
37
,
53
85
.
Miller
,
J. R.
, and
G. L.
Russell
,
1992
:
The impact of global warming on river runoff.
J. Geophys. Res.
,
97
,
2757
2764
.
NCDC
,
2005
:
Daily and sub-daily precipitation for the former USSR. Dataset 9813, 16 pp
.
Nijssen
,
B.
,
G. M.
O’Donnell
,
A. F.
Hamlet
, and
D. P.
Lettenmaier
,
2001
:
Hydrologic sensitivity of global rivers to climate change.
Climatic Change
,
50
,
143
175
.
Overpeck
,
J.
, and
Coauthors
,
1997
:
Arctic environmental change of the last four centuries.
Science
,
278
,
1251
1256
.
Pavelsky
,
T. M.
, and
L. C.
Smith
,
2006
:
Intercomparison of four global precipitation data sets and their correlation with increased Eurasian river discharge to the Arctic Ocean.
J. Geophys. Res.
,
111
.
D21112, doi:10.1029/2006JD007230
.
Peterson
,
B. J.
,
R. M.
Holmes
,
J. W.
McClelland
,
C. J.
Vörösmarty
,
R. B.
Lammers
,
A. I.
Shiklomanov
,
I. A.
Shiklomanov
, and
S.
Rahmstorf
,
2002
:
Increasing river discharge to the Arctic Ocean.
Science
,
298
,
2171
2173
.
Rawlins
,
M. A.
,
C. J.
Willmott
,
A. I.
Shiklomanov
,
E.
Linder
,
S.
Frolking
,
R. B.
Lammers
, and
C. J.
Vörösmarty
,
2006
:
Evaluation of trends in derived snowfall and rainfall across Eurasia and linkages with discharge to the Arctic Ocean.
Geophys. Res. Lett.
,
33
.
L07403, doi:10.1029/2005GL025231
.
Savelieva
,
N. I.
,
I. P.
Semiletov
,
L. N.
Vasilevskaya
, and
S. P.
Pugach
,
2000
:
A climate shift in seasonal values of meteorological and hydrological parameters for northeastern Asia.
Prog. Oceanogr.
,
47
,
279
297
.
Serreze
,
M. C.
,
D. H.
Bromwich
,
M. P.
Clark
,
A. J.
Etringer
,
T.
Zhang
, and
R. B.
Lammers
,
2003a
:
Large-scale hydro-climatology of the terrestrial Arctic drainage system.
J. Geophys. Res.
,
108
.
8160, doi:10.1029/2001JD000919
.
Serreze
,
M. C.
,
M. P.
Clark
, and
D. H.
Bromwich
,
2003b
:
Monitoring precipitation over the Arctic terrestrial drainage system: Data requirements, shortcomings, and applications of atmospheric reanalysis.
J. Hydrometeor.
,
4
,
387
407
.
Shepard
,
D. S.
,
1984
:
Computer mapping: The SYMAP interpolation algorithm.
Spatial Statistics and Models, G. L. Gaile and C. J. Willmott, Eds., D. Reidel, 133–145
.
Shiklomanov
,
A. I.
,
R. B.
Lammers
, and
C. J.
Vörösmarty
,
2002
:
Widespread decline in hydrological monitoring threatens Pan-Arctic research.
Eos, Trans. Amer. Geophys. Union
,
83
,
13
.
Shiklomanov
,
A. I.
,
T. I.
Yakovleva
,
R. B.
Lammers
,
I. P.
Karasev
,
C. J.
Vörösmarty
, and
E.
Linder
,
2006
:
Cold region river discharge uncertainty—Estimates from large Russian rivers.
J. Hydrol.
,
326
,
231
256
.
Su
,
F. G.
,
J. C.
Adam
,
L. C.
Bowling
, and
D. P.
Lettenmaier
,
2005
:
Streamflow simulations of the terrestrial Arctic domain.
J. Geophys. Res.
,
110
.
D08112, doi:10.1029/2004JD005518
.
Van der Linden
,
S.
,
T.
Virtanen
,
N.
Oberman
, and
P.
Kuhry
,
2003
:
Sensitivity analysis of discharge in the Arctic Usa Basin, East-European Russia.
Climatic Change
,
57
,
139
161
.
Wu
,
P. L.
,
R.
Wood
, and
P.
Stott
,
2005
:
Human influence on increasing Arctic river discharges.
Geophys. Res. Lett.
,
32
.
L02703, doi:10.1029/2004GL021570
.
Yang
,
D. Q.
,
D. L.
Kane
,
L. D.
Hinzman
,
X.
Zhang
,
T.
Zhang
, and
H.
Ye
,
2002
:
Siberian Lena river hydrologic regime and recent change.
J. Geophys. Res.
,
107
.
4694, doi:10.1029/2002JD002542
.
Yang
,
D. Q.
,
B. S.
Ye
, and
D. L.
Kane
,
2004a
:
Streamflow changes over Siberian Yenisei River Basin.
J. Hydrol.
,
296
,
59
80
.
Yang
,
D. Q.
,
B. S.
Ye
, and
A. I.
Shiklomanov
,
2004b
:
Discharge characteristics and changes over the Ob River watershed in Siberia.
J. Hydrometeor.
,
5
,
595
610
.
Yang
,
D. Q.
,
D.
Kane
,
Z. P.
Zhang
,
D.
Legates
, and
B.
Goodison
,
2005
:
Bias corrections of long-term (1973–2004) daily precipitation data over the northern regions.
Geophys. Res. Lett.
,
32
.
L19501, doi:10.1029/2005GL024057
.
Ye
,
B. S.
,
D. Q.
Yang
, and
D. L.
Kane
,
2003
:
Changes in Lena River streamflow hydrology: Human impacts versus natural variations.
Water Resour. Res.
,
39
,
1200
1213
.
Ye
,
H.
,
D. Q.
Yang
,
T.
Zhang
,
X.
Zhang
,
S.
Ladochy
, and
M.
Ellison
,
2004
:
The impact of climatic conditions on seasonal river discharges in Siberia.
J. Hydrometeor.
,
5
,
286
295
.
Zhang
,
T.
,
M. C.
Serreze
,
D.
Yang
,
D.
Gilichinsky
,
A.
Etringer
,
H.
Ye
,
R. G.
Barry
, and
S. M.
Chudinova
,
2003
:
Permafrost degradation and hydrologic response in the Russian Arctic drainage basin.
Geophysical Research Abstracts, Vol. 5, Abstract 12990. [Available online at http://www.cosis.net/abstracts/EAE03/12990/EAE03-J-12990.pdf.]
.
Ziegler
,
A. D.
,
J.
Sheffield
,
E. P.
Maurer
,
B.
Nijssen
,
E. F.
Wood
, and
D. P.
Lettenmaier
,
2003
:
Detection of intensification in global- and continental-scale hydrological cycles: Temporal scale evaluation.
J. Climate
,
16
,
535
547
.

Footnotes

Corresponding author address: Dennis P. Lettenmaier, Department of Civil and Environmental Engineering, Box 352700, University of Washington, Seattle, WA 98195-2700. Email: dennisl@u.washington.edu