Changes of the streamflow of northern river basins of Siberia and their teleconnections to climate patterns

The Arctic rivers contribute more than one‐third of the total freshwater streamflow into the Arctic Ocean and play an essential role in the heat and mass circulation of the Arctic atmosphere/ocean system. As the Arctic is warming faster than the global average, the streamflow from Arctic basins increases. This study analyzed the streamflow of the three largest Siberian rivers: the Lena, Yenisei, and Ob', at multiple temporal scales. Results show that the annual streamflow of each river basin exhibits statistically significant increasing trends, while the seasonal streamflow of sub‐basins generally decreases in the summer but increases in the winter. Both autocorrelation and long‐term persistency are often found in the streamflow time series, which indicates significant changes in the large‐scale climatological environment. Therefore, wavelet coherence between the streamflow and large‐scale climate patterns, including the El Niño–Southern Oscillation (ENSO), North Pacific pattern (NP), Arctic Oscillation (AO), and the Pacific/North America Pattern (PNA), have been conducted. NP and ENSO are found to have positive relationships with the precipitation and the ratio of potential evapotranspiration over the precipitation. AO and the PNA are found to have positive relationships with the streamflow of the Ob' and Yenisei rivers at decadal and multidecadal scales. This study demonstrates that the existence of nonstationarities within the Siberian streamflow as the combined impact of climate change alters the hydroclimatological and terrestrial environment of Siberia. These findings provide new insights into the mechanisms underlying the hydrologic changes to warming trends and oscillations of climate patterns, which contribute to our understanding and the prediction of streamflow of these northern rivers.

decreases in the summer but increases in the winter.Both autocorrelation and long-term persistency are often found in the streamflow time series, which indicates significant changes in the large-scale climatological environment.Therefore, wavelet coherence between the streamflow and large-scale climate patterns, including the El Niño-Southern Oscillation (ENSO), North Pacific pattern (NP), Arctic Oscillation (AO), and the Pacific/North America Pattern (PNA), have been conducted.NP and ENSO are found to have positive relationships with the precipitation and the ratio of potential evapotranspiration over the precipitation.AO and the PNA are found to have positive relationships with the streamflow of the Ob' and Yenisei rivers at decadal and multidecadal scales.This study demonstrates that the existence of nonstationarities within the Siberian streamflow as the combined impact of climate change alters the hydroclimatological and terrestrial environment of Siberia.These findings provide new insights into the mechanisms underlying the hydrologic changes to warming trends and oscillations of climate patterns, which contribute to our understanding and the prediction of streamflow of these northern rivers.

K E Y W O R D S
climate patterns, Siberian river basins, statistical analysis, streamflow, teleconnections, wavelet analysis The Arctic is warming much faster than the global average rate, known as Arctic Amplification (Serreze et al., 2009;Serreze & Barry, 2011).Under a warmer climate, the Arctic hydrological processes are becoming more intense (Bring et al., 2016;Déry et al., 2009) featuring the changes in hydroclimatological variables.The rainfall was found to offset snowfall (Han et al., 2018) and the shorter duration of snow cover and lower maximum winter snow storage was found in the mountainous Siberian (Bulygina et al., 2009).Severe hydrologic extremes, such as thunderstorms and extreme droughts, were found to more frequently occur (Dirmeyer et al., 2013;Groisman et al., 2007;Iijima et al., 2016).The land surface of the Arctic also has drastic changes that have a trend of greening and browning (Myers-Smith et al., 2020;Phoenix & Bjerke, 2016), which indicates the luxuriant growth of vegetation (Elmendorf et al., 2012;Macias-Fauria et al., 2012;Tape et al., 2006) and the extensive degradation of permafrost (Reginald & Vladimir, 2009;Ye et al., 2009).These changes could affect multiple hydrological processes.The enhancement of the subsurface flow connectivity (Walvoord & Kurylyk, 2016;Watson et al., 2013), higher groundwater storage (Reginald & Vladimir, 2009), and higher active layer moisture storage (Smith et al., 2007) have been observed, and could have an influence on the streamflow from the Arctic rivers.
Three large rivers in Siberia, namely the Lena, Yenisei, and Ob' rivers, occupy as large as 7959 km 2 and contribute more than 1500 km 3 Áa −1 freshwater into the Arctic oceans, which is also the largest three river basins in the Arctic region (Shiklomanov et al., 2021).As one of the main freshwater inputs, the streamflow from Siberian rivers is crucial to the salinity stratification (McPhee et al., 1998), sea ice formation (Park et al., 2020;White et al., 2007), oceanic thermohaline circulation (Bintanja et al., 2018;Kattsov et al., 2007) and the heat and mass balance of the Arctic atmosphere (Park et al., 2020).The streamflow of these rivers has been found to have increased at the end of the 20th century, especially in winter and spring (Berezovskaya et al., 2004;Fukutomi et al., 2003;Yang et al., 2002Yang et al., , 2003Yang et al., , 2004aYang et al., , 2004b;;Ye et al., 2003).As global warming has enhanced in the 21st century, changes in streamflow from these Siberian rivers were found to intensify extensively.Besides the streamflow at multitemporal scales, the peak flood was also found to increase (Tananaev et al., 2016).The contribution from deep groundwater to the baseflow was found to increase by investigating the ratio between the maximum and the minimum streamflow (Xu et al., 2020).
In addition to the warming over the Arctic, largescale climate patterns are also important factors affecting the hydrological cycle and the streamflow of the Siberian river basins.The Siberian High is the dominant meteorological system over the Siberian region featuring a large anticyclone centre in the inner of Eurasia continent (Cohen et al., 2001;Gong & Ho, 2002).It was found to be negatively correlated with the Siberian temperature and precipitation (Cohen et al., 2001;Gong & Ho, 2002).The strength of Siberian High has a strong teleconnection with Arctic Oscillation (AO).When the winter AO is in a positive phase, Wu and Wang (2002) found both the winter Siberian High and the East Asian winter monsoonal flow tend to be weaker than normal, while Gong and Ho (2002) found the temperature tends to be higher and eastern Siberia tends to be drier.The teleconnections between the Arctic regions and climate patterns over the Pacific Ocean, such as the El Niño-Southern Oscillation (ENSO) and the Pacific North America Pattern (PNA), were also founded and affecting the climate of east Siberia (Cheung et al., 2012;Zhang et al., 2020).
Concerning the complex changes of streamflow in the large Siberian rivers and the multiple influencing factors, this study investigated the spatiotemporal changes of streamflow in the Lena, Yenisei, and Ob' rivers at multiple temporal scales based on historical observations.The slowly varying changes, under the possible effects of autocorrelation and persistence, were considered in this study by a conducting series of Mann-Kendall test, while the abrupt changes are also considered through the Pettitt test, which could identify the possible abrupt change points.Then the teleconnections between the streamflow and large-scale climate patterns, including the AO, the North Pacific Index (NP), the Pacific/North American Pattern (PNA), and ENSO, were analysed and discussed.As the climate patterns and streamflow have different periodicity or seasonality, the wavelet analysis was applied to investigate the dominant modes of variability in streamflow and its phase relationship with climate patterns.
This paper is organized as follows: the streamflow data and large-scale climate patterns are described in section 2; the technical details of the trend analysis, change point detection, and wavelet analysis are described in section 3; results are provided in section 4; the discussion of mechanisms behind the hydrological changes and the teleconnection with climate patterns are given in section 5; and conclusions are given in section 6.

| Study area and streamflow data
The study area consists of the three major river basins of Siberia: Lena River Basin (LRB), Yenisei River Basin (YRB), and Ob' River Basin (ORB).Collectively, these basins contribute more than one-third of the total streamflow into the Arctic Ocean (Shiklomanov et al., 2021).The locations and topography of these river basins are shown in Figure 1, and the summary of the land surface types is given in Table 1.The Ob' River, the westernmost of the three Siberian rivers, originates from the Altay Mountains and flows northward into the Gulf of Ob.The Yenisei River, the largest river flowing into the Arctic Ocean, originates from the Mungaragiyn-gol Ridge in the Mongol Nation and follows a northward course to the Yenisei Gulf in the Kara Sea.The Lena River, the easternmost of the three Siberian rivers, originates from the Baikal Mountains south of the Central Siberian Plateau and flows northeast into the Laptev Sea.All three basins exhibit specific northern features such as extensive permafrost and snow-dominated flooding, and they all have high flows in summer and low flows in winter.
Until the end of the twentieth century, Siberian streamflow are observed systematically and the observed data quality was controlled and archived by the Russian Hydrometeorological Services (Shiklomanov et al., 2000), which is available from the Regional Hydrographic Data Network for the Pan-Arctic Region (R-ArcticNet v. 2.0) F I G U R E 1 Geographical location of the study area, the distribution and upstream area of hydrological stations, and the permafrost extent of the Ob, Yenisei, and Lena river basins (The color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com] (Lammers et al., 2016).After 2000, the streamflow measured at the mouth of the Arctic river basins under the direction of the Arctic and Antarctic Research Institute is available from the Arctic Great River Observation (ArcticGRO) (Shiklomanov et al., 2018).A total of 160 stations from R-ArcticNet v. 2.0 and three stations at the mouth of each river basin from ArcGRO were selected in this study.The selected stations from R-ArcticNet v. 2.0 should have at least 30 years of continuous monthly records until 1999.Based on the two datasets, the total streamflow of the whole basins with durations of 1936-2018, 1935-2018, and 1930-2018 for YRB, LRB, and OBR, respectively, collected in this study.Detailed information about the selected stations is summarized in Table S1, Supporting Information.
Human activities in the three Siberian river basins are relatively weaker than in the low and mid-latitude regions (Vörösmarty & Sahagian, 2000).With low population and slow economic development, the water consumption by industries and agriculture in these areas is not significant compared with the amount of total discharge of the river basin (Durocher et al., 2019;Shiklomanov, 1997;Shiklomanov et al., 2000Shiklomanov et al., , 2021;;Yang et al., 2004b).The construction of large dams and reservoirs has been mainly concentrated in upstream of river basins.There are 11 reservoirs with capacities larger than 1 km 3 , and seven of them are in the Yenisei River basin, three of them in the Ob' River basin, and the other one in the Lena River basin.The Bukhtarminskoe reservoir, the largest reservoir in ORB, was completed in 1960 and first filled in 1967 with a maximum capacity of 49.8 km 3 , representing 13% of the annual flow volume at the outlet.The largest, Bratskoe reservoir in YRB, was first filled in 1967 with a maximum capacity of 169 km 3 that represents 29% of the annual flow volume of YRB.The largest reservoir of LRB, the Vilyuy reservoir, was completed in 1974 with a maximum capacity of 35.9 km 3 , which represents 7% of the annual flow volume at the outlet (Ye et al., 2003).The capacity and timing of filling large reservoirs (>1 km 3 ) within the study area (Adam et al., 2007;Yang et al., 2004b) are summarized in Table S2.

| Climate indices and hydroclimatological data
We investigate possible teleconnections between the streamflow of these three Siberian river basins and the Arctic Oscillation (AO), the North Pacific Oscillation (NP), the Pacific/North American Pattern (PNA), and ENSO.The correlations between these climate indices and the deseasonalized and standardized monthly streamflow.The deseasonalization was conducted on the long-term monthly streamflow time series by removing the multiyear average of each month from the monthly streamflow and then divided by the standard deviation of each month.The AO index represents the first leading mode of the mean geopotential height anomalies at 1000 hPa, with a positive index indicating a lower-than-normal pressure over the Arctic.The PNA index represents the variability of opposite geopotential height anomalies centred between the Aleutian and the Hawaiian Islands.A positive phase of PNA coincides with a lower autumn sea level pressure over Northern Eurasia (Zhang et al., 2020).ENSO is based on the Niño3.4index which indicates an El Niño (La Niña) mode if a 5-month running mean of sea surface temperature anomalies in the Niño3.4region of the Tropical Pacific exceeds (falls below) 0.4 C for 6 or more months (Trenberth, 1997).Climate indices were downloaded from the Global Climate Observing System, https://www.esrl.noaa.gov/psd/gcos_wgsp/Timeseries/.
In addition, hydro-climatological variables, including potential evapotranspiration (PET, mm), near-surface temperature (Ta, C) and precipitation (P, mm), at 0.5 × 0.5 spatial resolution from January 1901 to December 2018 were extracted from the CRU-TS 4.03 of the University of East Anglia (Harris, 2020).The monthly time series of these climate variables for each river basin are the areally-average of all grids located within the outlines of each river basin.Note: Data from Yang et al. (2002Yang et al. ( , 2004aYang et al. ( , 2004b) ) and Ye et al. (2003). a The mean annual runoff of ORB, YRB and LRB are from no. 7142 stations, no.6656 stations and no.6342 stations, respectively (Shiklomanov et al., 2021).Details of stations are provided in Table S1.

| Temporal trend analysis
The temporal changes of streamflow are detected and assessed using a series of Mann-Kendall (MK) test, which is a rank-based and nonparametric test (Kendall, 1955) that has been widely used in hydrological process analyses.By the estimation of Kendall's statistic S, the trend of a time series can be detected without distribution fitting.In the traditional MK test, the null hypothesis H 0 is that the original time series Y i , i= 1, 2, …, n f g is independent and identically distributed, while the alternative hypothesis H 1 is that a monotonic trend exists in Y i f g.The statistic S is defined as the proportion of concordant pairs minus the proportion of discordant pairs in the time series, where N is the record length and Y j and Y i are the sequential data values.When N>1, the statistic S is normally distributed with the mean and variance, where t m is the number of ties of extent m.The standardized test statistic Z is computed by The statistic Z follows the standard normal distribution with a mean of zero and a variance of one.Therefore, the probability can be estimated through the normal cumulative distribution.
von Storch (1999) found that the existence of autocorrelation within a time series will lead to the misestimation of the significance of the trend.Yue et al. (2002) proposed a "trend-free prewhitening" (TFPW) procedure to reduce the impact of autocorrelation.The first step is to remove the trend from the original time series Theil-Sen approach (Theil, 1992).Next, we remove the lag-1 correlation by is then an independent series.Then the identified trend T t f g and the residual Y 0 t È É are combined as X t =X 0 t +T t .The MK test is applied to the blended series and the significance of the trend can then be detected without being affected by the autocorrelation.
Besides the autocorrelation, the presence of long-term persistence (LTP) can also lead to the misestimating of the significance of the trend (Cohn & Lins, 2005;Su et al., 2018).LTP is a fractional Gaussian noise characterized by a long-term slowly decaying autocorrelation function.In hydrologic and climatic modelling, LTP is usually used as a "fingerprint" in reconstructing a time series for it implies the presence of long-term variability in the background climatic environment (Koutsoyiannis & Montanari, 2007;Vincent et al., 2015).The Hurst exponent H is usually used to describe the scaling properties within a time series, and ranges from 0 to 1. H is derived from the following equations: where ρ l indicates the autocorrelation function of lag l for a given H (Hamed, 2009).The value of H can be estimated by maximizing the log-likelihood function.When the H value is close to 0.5, there is the presence of LTP (Hurst, 1951).Then, whether the H value in step 1 is significantly different from 0.5 is quantified as the significance of H.It is calculated through the mean μ H and variations σ H when H = 0.5, which are estimated by empirical equations from Hamed (2008) as follows: σ H =0:77654n − 0:5 − 0:0062: This study used 10% significance level when determining significant H.The variance of the statistic S with a significant H is calculated by the following equations: where ρ ij are the correlations between equivalent normal variates.As H is calculated from a given time series, V S ð Þ H0 is a biased estimate.The unbiased estimate V S ð Þ H is obtained by multiplying with a bias correction factor B, where B is the function of H and n with empirical parameters, which could be found in Kumar et al. (2009).The significance of MK3 is computed by replacing the V(S) in Equation ( 5) with V S ð Þ H .The details of this estimation technique can be found in Hamed (2008), Hamed (2009), and Kumar et al. (2009).
Therefore, to investigate the trends of the streamflow data, three types of Mann-Kendall test are performed: the traditional Mann-Kendall test (MK1) is first performed, which reveals the sign and magnitude of the trend.Then a Mann-Kendall test with a lag-1 autocorrelation (AR) and trend-free prewhitening (MK2), and a Mann-Kendall test with LTP (MK3) are applied to the streamflow data.The results from MK2 and MK3 could indicate whether the trend detected through MK1 has autocorrelation and long-term persistence.

| Change-point analysis
A change-point analysis based on the nonparametric Pettitt test (Pettitt, 1979) (a robust method for detecting abrupt change points in a continuous time series; Villarini et al., 2012) was conducted on the mean and variance of the streamflow time series.The Pettitt test is a rank-based test that examines whether two samples come from the same population by building a Mann-Whitney statistic that allows the detection of a single change point in the mean of the variables of interest at an unknown point in time, which makes it more robust against outliers and skewed distributions than parametric tests.The p-value of the test statistic is calculated using an approximate limiting distribution of the Kolmogorov-Smirnov goodness-of-fit statistic (Muggeo, 2003;Villarini et al., 2012).

| Wavelet analysis
The possible teleconnections between streamflow and climate patterns are investigated by wavelet analysis, which includes a continuous Morlet wavelet transformation, a global wavelet spectrum (GWS), a scale-averaged wavelet power (SAWP), and the wavelet coherence (WTC).In a continuous Morlet wavelet transformation, a time series of monthly streamflow with seasonality removed is decomposed into time-frequency fields, with the variance of wavelet coefficients of various frequency bands summarized by a GWS.Based on the results of the GWS, fluctuations over certain periods are filtered out by SAWP and are correlated with climate indices in the corresponding frequency bands.Then WTC with climate indices as covariates is conducted to investigate the possible effect of climate patterns on the streamflow data over time, giving a phase relationship between the two variables with respect to time and frequency.Because the wavelet method decomposes a time series at continuous timescales, degrees were used as the measurements of correlation instead of temporal units.Zero degrees mean two time series are absolutely in-phase at the corresponding timescales, or absolute positive correlation, while 180 represents an antiphase or out-of-phase relationship, or absolute negative correlation.Degrees can be transformed into time if necessary.Details of these methods can be found in Jevrejeva et al. (2003), Mwale and Gan (2004), and Tan et al. (2016).

| Temporal trend of the streamflow
The regular Mann-Kendall (MK1) test was used to test the possible temporal trend in the streamflow of the three basins at annual, seasonal, and monthly timescales (Figure 2).The annual streamflow of stations located downstream and the main branches show statistically significant positive trends, while stations in upstream of ORB and LRB show negative trends (Figure 2a).As expected, monthly streamflow has higher spatial heterogeneity than seasonal streamflow.Monthly streamflow of small sub-basins (less than 100,000 km 2 drainage areas) at the mid-and upstream parts of ORB shows significant trends, but only a few stations in the middle of YRB have significant increasing trends (Figure 2b).
Figure 2c-f shows stations with significant trends in their seasonal streamflow time series.In ORB and LRB, the streamflow of upstream rivers has predominantly decreased in spring and summer (Figure 2c,d) but increased in winter and autumn (Figure 2e,f).In these areas, there are large dams and relatively greater water use along river valleys than in other regions, which reduce summer peak flood and release water for hydropower generation in winter (Yang et al., 2004a(Yang et al., , 2004b)).For LRB, where there are weak human activities and high permafrost coverage, the general increasing trends were found in the annual, seasonal, and monthly streamflow, which is probably affected by climate change (Figure 2).Along with climate warming, more rainfall was observed in Siberia regions, which contributes to high streamflow (Berezovskaya et al., 2004;Yang et al., 2004b;Ye et al., 2009).The spatial distribution of stations with time series showing LTP and AR1 was shown in Figure 3.The upper stream of ORB, in the mainstream of YRB, as well as in the central LRB have more stations with AR1 in the annual streamflow than other regions (Figure 3a).It seems that streamflow in these regions tends to be correlated to data of the previous or the following years.Nearly all stations with significant trends also have LTP in the monthly streamflow data, mainly located in the ORB, the west of YRB and the east of LRB (Figure 3b).The monthly streamflow in these regions may exhibit persistent wet, dry or average conditions for months.For seasonal streamflow data, the upstream of ORB and YRB and the east of LRB have more stations with AR1 in winter (Figure 3f) than other three seasons (Figure 3c,d), reflecting a stronger relationship of the streamflow with the contiguous year in winter than that in other three seasons.ORB and the east of YRB have more stations that show long-term persistence in spring and summer than other regions (Figure 3c,d) probably because of the dam regulation and the large-scale climate patterns.
The findings from the MK analysis reveal that the trend is widely detected in the streamflow data at multiple time series that mixed increase and decrease varies with seasons and locations.The short-and long-term persistence were also widely found within the streamflow data that indicate the correlation with the data in the previous and following year and the persistence of the current trend.As previously mentioned, the existence of long-term persistence is usually related to large-scale climate conditions, which was discussed in section 4.4.

| Abrupt change of the streamflow
The Pettitt test is applied to the deseasonalized and standardized monthly streamflow to detect abrupt change points and their significance.Forty out of 160 stations have statistically significant change points in the monthly streamflow time series, and most of them occurred between 1960 and 1985 (Figure 4a).As listed in Table S2, the upstream of ORB and YRB have large reservoirs that could reduce the peak flood in boreal summer and release water for hydropower generation overwinters.The increasing winter streamflow and reduced summer streamflow of tributaries are partly a result of dam regulations (Yang et al., 2002(Yang et al., , 2004a(Yang et al., , 2004b;;Ye et al., 2009).
The streamflow observed at the outlets of ORB, YRB and LRB also have significant change points in 1969, 1984and 1979, respectively (Figure 4b-d), respectively (Figure 4b-d).Furthermore, the streamflow of ORB, YRB and LRB has been higher after the detected change points.In addition to the regulation of reservoirs, the development of industries and agriculture along the river, especially the large cities located in upstream of the Ob' River, could be other contributing factors to the detected abrupt changes, in addition to the impact of climate change.However, due to the limited social-economic data, the contribution of human activities to the abrupt change of the streamflow of the large Siberian rivers cannot be estimated accurately.

| The frequency characteristics of streamflow
The monthly streamflow data from stations located at the most downstream locations were subjected to a wavelet analysis (Point ID: 7142, 6656, 6342 in Table S1).Figure 5 shows the local wavelet spectra of the deseasonalized monthly streamflow.In ORB, there is a statistically significant periodicity of 4-16 months (annual scale) over periods of 1930s-1960s and 1990s-2010s, reflected  global wavelet spectrum in the right panels as well (Figure 5a).In YRB, significant interannual oscillations at the 64-128 months timescales are detected between 1935 and 2018 (Figure 5b).In LRB, significant interannual oscillations at 64-128 months are detected (Figure 5c).Therefore, we also estimate the SAWP at annual, interannual and interdecadal (1-16, 64-128 and >256 months) timescales.
The SAWP of the deseasonalized monthly streamflow at 1-16, 64-128, and >256 months timescales are shown in Figure 6.At the 1-16 months timescales (Figure 6a), the monthly streamflow has become fluctuated more after the 1990s.At the 64-128 months timescales, all the monthly streamflow of the three basins show increasing trends after the 1990s (Figure 6b).At longer than 256-month timescales, YRB and LRB generally experience increasing trends while ORB shows decreasing trends, showing contrasting hydrologic responses between these three Siberian river basins to climate changes over the study period (Figure 6c).

| Teleconnection between the streamflow and the large-scale climate patterns
Based on the spectrum analysis of the streamflow through the wavelet method, the relationship between large-scale climate patterns, including the AO, PNA, NP and ENSO, and the streamflow from the three large Siberian river basins were evaluated.Spearman rank correlation was applied to the SAWP of monthly streamflow at multiple timescales and climate indices, including AO, PNA, NP and Niño3.4 (Table 2).Results show that NP and Niño3.4 have a significant positive correlation with the streamflow of ORB at the 1-16 month timescale, while AO and NP have significant positive correlations with that of LRB.AO and PNA have significant positive correlations with the monthly streamflow of ORB at both the 64-128 months timescale and the longer than 256 months timescale.For LRB, AO, PNA and Niño3.4 have significant negative relationships with the monthly streamflow at the higher than 256 months timescale.
The phase relationship between the climate patterns and the streamflow was further analysed through wavelet coherence (Figures 7-9).From the results, NP and Niño3.4 show a stronger relationship with streamflow than AO and PNA.NP has a statistically significant relationship with the streamflow of the three river basins at around 16 months timescale.For ORB, the relationship is significantly antiphase with NP during the 1990s and after the 2000s (Figure 7c).The streamflow of YRB was in-phase with NP around 16 months timescale before the 1970s but antiphased after that (Figure 8c).The streamflow of LRB was also in-phase with NP at around 16 months timescale before 1980s but antiphase after that.These diverse teleconnections between NP and the three Siberian basins have been similarly observed in other continents, such as in the western Canada (Gan et al., 2007;Gobena & Gan, 2006) and East Africa (Mwale & Gan, 2004).The teleconnection between streamflow and Niño3.4 were also significant at the annual timescale but with different phase relationship.Niño3.4 led YRB streamflow by 90 before the 1970s, but by 270 after the 1990s.From the sporadic significant relationship between Niño3.4 and streamflow in ORB and LRB (Figures 7d and 9d), the reverse of phase relationship could also be noticed.Compared with NP and Niño3.4,AO and PNA have relatively weaker phase relationships with the streamflow.The significant in-phase relationships between AO and the streamflow of LRB since the 1970s at about the 64 months timescale could be noticed (Figure 9a).Around the 1990s, PNA led the LRB streamflow by 90 at about 64 months timescale (Figure 9b).
To reveal the mechanism of the above teleconnections, the historical precipitation and the ratio of PET over precipitation (PET/P) were plotted with the NP and Niño3.4,shown in Figure 10.It could be found when Niño3.4 shows a positive anomaly, the precipitation in the three river basins shows a slight increase, while PET/P has a high slope (Figure 10a-c).For NP, the precipitation in three river basins has a stronger positive relationship than Niño3.4(Figure 10d-f).And with the increase of NP, the precipitation and PET/P have a similar increasing rate (Figure 10d-f).Therefore, the teleconnection between NP and ENSO and the streamflow of the three Siberian river basins may build on the precipitation and PET, while the tradeoff between precipitation and PET could be one of the reasons for the reverse phase relationship with NP and Niño3.4.The teleconnection between the streamflow and AO and ENSO was also built through the hydroclimatological variables.Previous studies have found that changes in sea level pressure over major Siberian river basins were affected by ENSO (Liess et al., 2017).Furthermore, the weaker Siberian High during positive AO winters led to higher surface air temperature over Siberia (Huang et al., 2016;Wu & Wang, 2002), and extreme heat waves in the summer of 2020 (Overland & Wang, 2021).The teleconnection can also affect the streamflow by altering the land surface.It was found by Li et al. (2017) that the southward transport of cold air enhanced under El Niño would suppress the spring vegetation growth over the river basins.
The normalized annual temperature, precipitation and streamflow data are presented in 11 which shows that streamflow from these three river basins has statistically significant positive relationships with precipitation as expected.But the relationship between the streamflow and temperature differs between basins, such that the streamflow has statistically significant positive relationships with temperature in Yenisei and Lena river basins, but not in Ob'.This may be because the steppe and forest-steppe in the Ob' River basin would lead to a significant increase in evapotranspiration under a warmer climate, which offsets the increases in streamflow resulting from a higher precipitation (Magritsky et al., 2018).For Lena and Yenisei river basins, a warmer climate would have positive effects on the streamflow through thawing permafrost (Barry & Gan, 2022).The previously blocked subsurface flow paths would be opened and transmit large groundwater fluxes when permafrost thaws, and then the subsurface fluxes, such as the soil drainage and recharge, groundwatersurface water exchange, and base flow, would be enhanced (Walvoord & Kurylyk, 2016).The overall effects would be more visible in the winter streamflow, mainly consisting of the subsurface flow (Wang et al., 2021a(Wang et al., , 2021b)).

| DISCUSSION
Both trends and abrupt change points were identified in the streamflow of large Siberian river basins through multiple statistical methods and wavelet methods.Generally, significant increasing trends were found in the annual streamflow, which is consistent with previous studies, while different trends within the seasonal and monthly streamflow are also identified and found to have a relationship with the changing climate and the largescale climate patterns.The contributions from possible factors were discussed in this section.
Series of past studies conducted at the end of the twentieth century demonstrates that the effects of human activities have significant impacts on the monthly streamflow of the three largest Siberian river basins (Yang et al., 2004a(Yang et al., , 2004b;;Ye et al., 2003).Reservoir regulations mainly for power generation would increase the winter streamflow but reduces the summer streamflow of the regulated sub-basins.Regarding the total streamflow at the outlets, the trend is fuzzed by the tradeoff between the reduction of streamflow from regulated sub-basins and the increase of streamflow from unregulated subbasins that resulted from the increasing precipitation (Yang et al., 2004a).However, climate warming of climate has intensified in the Arctic which is warming as much as four times faster than the global average, according to the latest research Rantanen et al. (2022), and the effects of climate change on the Arctic streamflow are also amplified.Researches in recent decades show that the changes in streamflow in Arctic river basins are mainly driven by climate warming (Feng et al., 2021;Frolova et al., 2022;Melnikov et al., 2019;Troy et al., 2012;Wang et al., 2021b).With a higher air temperature, the stronger moisture poleward transport was expected to contribute to a strong future increase in the Arctic precipitation (Bintanja et al., 2020) that makes the warmer Arctic wetter.Increasing precipitation should lead to increasing streamflow.But climate warming would reduce the streamflow through enhanced evaporation, greening vegetation that facilitates canopy interception and transpiration (Berner et al., 2020;Kerkhoven & Gan, 2013;Myers-Smith et al., 2020), and thawing permafrost that enhances the soil infiltration (Wang et al., 2021a(Wang et al., , 2021b)).Therefore, the variation of streamflow under a wetter and warmer climate is more than the results of human activities, which together represent the combined results of hydroclimatological and terrestrial environment changes, and human activities.However, without adequate observations, site experiments or the operation strategies of reservoirs, the influences of the above factors are difficult to quantify accurately.

| CONCLUSIONS
This study comprehensively investigated the changes within the streamflow data longer than 30 years from 160 stations within three large Siberian river basins: Ob, Yenisei and Lena river basins.Multiple statistic and spectrum analysis methods were applied to the streamflow data at multiple timescales.The results from MK analysis reveal that the general increasing trend is widely found in the streamflow data of stations downstream of the three river basins at annual, seasonal and monthly timescales.Stations in upstream of OBR and YRB have increasing winter streamflow and decreasing summer streamflow.The short-and long-term persistence were also widely detected in the streamflow data indicating the strong autocorrelation within time series and the possible relationship with the large-scale climate conditions.The teleconnection with large-scale climate patterns, including AO, PNA, NP and ENSO, was investigated by wavelet coherences.NP and ENSO were found to have a significant relationship with the annual streamflow of the three river basins.The precipitation and the ratio of PET over precipitation of all three river basins were found to increase with the increase of NP and Niño3.4.AO and PNA were found to have a phase relationship with the streamflow at decadal or interdecadal timescales.These findings support the idea that changes in streamflow from the Siberian river basins are significantly influenced by the local hydroclimatology and the large-scale climate patterns, and provide new insights for future studies concerning hydrological modelling and the attribution of the changes of streamflow in northern river basins to climate changes.Understanding these processes is essential to adapting local water resource management to mitigate the impact of global climate change.

F
I G U R E 2 Temporal trends of monthly, seasonal, and annual streamflow of the three Siberian river basins from the 1930s to 2018 based on the standard MK1 test.(The color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 3 Spatial distribution of AR1 and LTP in the annual and seasonal time series from the MK2 and MK3 tests (the significance level was 0.05).(The color version is available online).[Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 4 Significant change points detected through the Pettitt test: (a) spatial distribution of change points in the standardized deseasonal monthly streamflow, (b-d) the time series of deseasonal monthly streamflow of the whole ORB, YRB, and LRB, and their significant change points (the significance level was 0.05) (The color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com] in the F I G U R E 5 Continuous Morlet wavelet spectrum and global wavelet spectrum (solid grey line) with 95% confidence level (dashed line) of the monthly streamflow of the three largest Siberian basins are presented.The thick black contours depict the 95% confidence level of local power relative to a white noise background.The grey slash is the cone of influence beyond which the energy in contaminated by the effect of zero padding.(The color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com] at 1-16 months (annual), 64-128 months (interannual), and longer than 256 months scales (interdecadal) of the streamflow of the three river basins.(The color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 8 Same as Figure 7, but for YRB [Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 9 Same as Figure 7, but for LRB [Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 1 0 The scatterplot of historical precipitation and PET/P with climate indices NP and Niño3.4 at monthly scale.Those for AO and PNA were shown in Figure S1.(The Color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 1 1 The scatterplots of normalized precipitation, temperature, and streamflow at annual scales.(The color version is available online) [Colour figure can be viewed at wileyonlinelibrary.com] Summary of three large Siberian river basins T A B L E 1 T A B L E 2 Spearman's correlations between the SAWP of wavelet decomposed monthly streamflow and climate indices for selected scales Statistical significance correlations at the 5% significance level are in bold text. Note: