Changes of Hydrological Components in Arctic Rivers Based on Multi-Source Data during 2003–2016

The hydrological cycle of the Arctic river basin holds an important position in the Earth’s system, which has been significantly disturbed by global warming. This study analyzed recent changes in the hydrological components of two representative Arctic river basins in Siberia and North America, the Lena River Basin (LRB) and Mackenzie River Basin (MRB), respectively. The trends were diagnosed in hydrological components through a comparative analysis and estimations based on remote sensing and observational datasets during 2003–2016. The results showed that the annual precipitation decreased at rates of 1.9 mm/10a and 18.8 mm/10a in the MRB and LRB, respectively. In contrast, evapotranspiration (ET) showed increasing trends, with rates of 9.5 mm/10a and 6.3 mm/10a in the MRB and LRB, respectively. Terrestrial water storage (TWS) was obviously decreased, with rates of 30.3 mm/a and 18.9 mm/a in the MRB and LRB, respectively, which indicated that more freshwater was released. Contradictive trends of the runoffs were found in the two basins, which were increased in the LRB and decreased in the MRB, due to the contributions of the surface water and base flow. In addition, the mean annual cycles of precipitation, ET, TWS, runoff depth, surface flow and base flow behaved differently in both magnitudes and distributions in the LRB and MRB, the trends of which will likely continue with the pronounced warming climate. The current case studies can help to understand the recent changes in the Arctic hydro-climatology and the consequence of global warming in Arctic river basins.


Introduction
The global hydrological cycle has been disturbed by recent climate changes [1][2][3][4][5][6]. As in high latitudes, the Arctic region is particularly sensitive to climate changes [7]. The hydrological cycle of Arctic terrestrial ecosystems is a key element in the Earth's system [5]. The climate over the Arctic region has been significantly altered during the past few decades, which has directly influenced the hydrological cycle in different ways [8,9]. More specifically, the average annual temperature has increased at almost twice the global rate over recent decades in the Artic regions, which is pronounced to increase by an additional 4-7 • C over the next century [10,11].
The major hydrological components in terrestrial ecosystems include precipitation, evapotranspiration (ET), runoff and terrestrial water storage (TWS) changes. Estimations of the hydrological components can help us to comprehend the dynamics and potential mechanisms of the hydrological processes under perturbation, which is crucial for estimating and managing the availability of water resources and the interactions between land surfaces and climates in the Arctic [12][13][14]. It is noted that the hydrological components do not standalone but, rather, are interactive. For example, the precipitation in high latitudes plays an important role in the Arctic hydrologic system, which is likely to intensify the

Temperature and Precipitation Data
The monthly air temperature and precipitation data were extracted from the highresolution and long-term gridded dataset of the Global Precipitation Climatology Center (GPCC) (http://climate.geog.udel.edu/~climate/html_pages/download.html#wb_ts2017 (accessed on 16 November 2020)). The GPCC dataset incorporates observations from 75,000 stations worldwide from 1901 to 2017 [48]. The GPCC V7 dataset was based on an enlarged and improved database to reduce the associated uncertainties from the source

Temperature and Precipitation Data
The monthly air temperature and precipitation data were extracted from the highresolution and long-term gridded dataset of the Global Precipitation Climatology Center (GPCC) (http://climate.geog.udel.edu/~climate/html_pages/download.html#wb_ts2017 (accessed on 16 November 2020)). The GPCC dataset incorporates observations from 75,000 stations worldwide from 1901 to 2017 [48]. The GPCC V7 dataset was based on an enlarged and improved database to reduce the associated uncertainties from the source data, and the systematic gauge measuring error was corrected, which has been widely used to support model climate variability analyses, validations, regional climate monitoring and water resource assessment studies [49][50][51][52]. The V7 version reduced the uncertainty of the dataset using an enlarged and improved database and is expected to include an improved correction of the systematic gauge measuring error [50]. Therefore, the GPCC dataset with the highest spatial resolution (0.5 • × 0.5 • ) was used during 2003-2016. Four seasons, namely the spring (March-May), summer (June-August), autumn (September-November) and winter (December-February), were defined for the seasonal analysis in the current study.

TWS retrieved from GRACE
Data retrieved from the Gravity Recovery and Climate Experiment (GRACE) satellite were used for analyzing the TWS change [3,34,[53][54][55]. The GRACE data products were freely downloaded from the GRACE Tellus website (https://grace.jpl.nasa.gov/data/getdata/monthly-mass-grids-land/ (accessed on 17 November 2020)). Preprocessing was conducted to remove noise by filtering and truncation [56,57]. The monthly TWS changes from 2003 to 2016 were used with a spatial resolution of 1 • × 1 • .

ET Data
The Global Land Evaporation Amsterdam Model (GLEAM) was used to calculate the ET (https://www.gleam.eu/ (accessed on 15 November 2020)) [58]. GLEAM is driven by reanalysis or station-based forcing data (e.g., net radiation and air temperature from ERA-Interim and the Multi-Source Weighted-Ensemble Precipitation data for the ET dataset) and applies a two-step method for estimating the ET, including the traditional method proposed by Priestley and Taylor (1972), followed by a function of the root zone soil moisture and vegetation optical depth. A wide range of water balance-based and ECbased evaluations [25,59] suggested that GLEAM ET data have been demonstrated to be accurate in a variety of ecosystems. Therefore, GLEAM often serves as a reference data for assessing other sorts of large-scale ET data [60]. GLEAM has been used not only in estimating large-scale ET dynamics [61] but also land-atmosphere interactions and drought monitoring [62,63]. In this study, the GLEAM v3.2a ET product during 2003-2016 was used with a spatial resolution of 0.25 • × 0.25 • .

Runoff Data
In this study, the runoff data was collected by the Arctic Great Rivers Observatory (ArcticGRO, Falmouth, MA, USA) [64] jointly managed by national hydrological institutions in Russia (Roshydromet), the United States (the US Geological Survey; USGS) and Canada (the Water Survey of Canada; WSC) (https://arcticgreatrivers.org/data/ (accessed on 15 November 2020)). The Russian Hydrometeorological Services, which is in charge of the quality control of the discharge data, has conducted systematical observations of discharges in the Siberian region since the late 1930s [44]. The data have been rigorously validated by various hydrological institutions (Arctic and Antarctic Research Institute (AARI), St. Petersburg et al.) [65]. We used the monthly discharge data during 2003-2016 at the Arctic Red River hydrological station of the MRB and Kusur hydrological station of the LRB, respectively. For comparison purposes, we converted the discharge (m 3 /s) data to the runoff depth (mm).

Mann-Kendall Test with Trend-Free Pre-Whitening
The Mann-Kendall (MK) test with trend-free pre-whitening (TFPW-MK) has been used in studies of hydrological trend detection. The equations for the MK test method are given below: where X i is the value of year i, n is the length of the data and m is the number of groups with tied ranks, each with t i tied observations. If |Z| > Z 1−α/2 , the null hypothesis is rejected, and the alternative hypothesis is accepted at the significance level of α; otherwise, the null hypothesis of no trend is accepted at the significance level of α.
Trend-free pre-whitening includes the following steps: where x t is the value of year t of the time series, n is the length of the data and x t is the average value. The original MK test was applied to Y t to assess the significance of the trend.

Hydrograph Separation
Various methods have been proposed to separate the base flow from the measured total discharge hydrograph. The analysis of the base flow provided valuable insights into understanding the dynamic changes of the groundwater and its contribution to the stream flow under climate changes [66,67]. However, most of the existing methods are rather subjective, which cannot provide consistent and converged results, even with the same input data [68].
Digital filtering algorithms have become powerful tools in hydrograph analyses [68][69][70]. The stream flow is distinguished as a direct runoff and base flow using the digital filter method. The direct runoff indicates a stream flow component changing with a high frequency, including the surface runoff and interflow, while the base flow represents a stream flow component changing with a low frequency, including groundwater discharge and the shallow subsurface flow. A two-parameter digital filtering algorithm is expressed as follows: where b t is the filtered base flow at time step t; α is the filter parameter; q b−1 is the filtered base flow at time t−1; Q t is the total stream flow at time step t; and BFI max is the maximum value of the long-term ratio of base flow to the total discharge, defined as 0.25 for perennial streams with hard rock aquifers, 0.50 for ephemeral streams with porous aquifers and 0.80 for perennial streams with porous aquifers [69]. The two-parameter digital filter algorithm was validated by comparing the results with measured data in Illinois, Pennsylvania and Maryland [71]. In cold regions, this method also has been successfully applied in river drainage basins with permafrost [72,73]. Therefore, we used the automated digital filter techniques for hydrograph separation in the Arctic regions.

Recent Changes in Temperature and Precipitation
The seasonal and annual variability of the temperatures in the MRB and LRB during 2003-2016 are shown in Figure 2a,b and Table 1. Annually, the temperatures in the two basins showed increased trends (Table 1 and Figure 2a). The warming trends of annual temperatures during the study period were 0.79 • C/10a and 0.46 • C/10a in the MRB and LRB, respectively (Figure 2a). The Arctic amplification was much stronger than the global warming rate 0.19 • C/10a during the last decades. The annual air temperature increased by around 11.06 • C and 6.2 • C during 2003-2016 in MRB and LRB, respectively. The magnitude of the warm trends was more obvious in the MRB of North America than in the LRB of Eastern Siberia. The temperatures in the MRB showed increasing trends in all the seasons (Figure 2b and Table 1). The rates of the temperatures were stronger in the spring and summer than in the autumn and winter. The temperature in the LRB showed an increasing trend in the spring, summer and winter, while it had a decreased trend in the autumn (Figure 2b and Table 1). The results showed obvious warming during the past 14 years over the entire study area.

Changes in ET
The mean annual ET in the MRB and LRB were 358.3 mm and 334.4 mm, respectively, which were almost equal to the amount of precipitation. The annual ET series in the MRB had a significantly increased trend and that in LRB showed a nonsignificantly increased trend (Table 2 and Figure 3a). The ET in the two basins experienced increases of 9.5 and 6.3 mm/10a during 2003-2016, respectively ( Figure 3a). Seasonally, ET showed an increasing trend in the spring and summer in two basins, and the greatest increase appeared in the summer, with rates of 64 mm/a and 56 mm/a, respectively ( Figure 3b and Table 2). The decreased trend occurred in the autumn of the LRB and in the winter of the MRB ( Figure  3b and Table 2). The area average magnitudes of ET were 102.2 mm, 224.7 mm, 29.7 mm  The seasonal and annual variabilities of the precipitation in the MRB and LRB during 2003-2016 are shown in Figure 2c,d and Table 1. Annually, the precipitation in LRB had a significantly decreased trend and that in MRB showed a nonsignificantly decreased trend (Table 1 and Figure 2c). The mean annual precipitations were 359.2 mm and 363.1 mm in the MRB and LRB, respectively. The annual precipitations had decreased rates of 1.9 mm/10a and 18.8 mm/10a in the MRB and LRB, respectively (Figure 2c), which were unlike the positive trends in global surface temperatures over the past century. The changes of precipitation in the study basins were different with the trends in global precipitation, which showed no significant changes based on the GPCP dataset [74]. The annual precipitation increased during 2003-2006 but started to decline after 2000 in the LRB, while the precipitation increased after 2010 in the MRB. Except for the summer, the precipitations decreased in the spring, autumn and winter in the MRB, which were 60.1 mm,150.6 mm, 89.6 mm and 58.9 mm in the spring, summer, autumn and winter, respectively (Table 1 and Figure 2d). In the LRB, the spring and winter precipitations showed increased trends, while they showed decreased trends in the summer and autumn, and the precipitations were 57.0 mm, 177.3 mm, 94.6 mm and 34.2 mm in the spring, summer, autumn and winter, respectively. From the magnitudes of the decreases in the relevant seasons, it can be seen that the decrease of annual precipitation in the LRB was mainly due to the decreased precipitation in the autumn.

Changes in ET
The mean annual ET in the MRB and LRB were 358.3 mm and 334.4 mm, respectively, which were almost equal to the amount of precipitation. The annual ET series in the MRB had a significantly increased trend and that in LRB showed a nonsignificantly increased trend ( Table 2 and Figure 3a). The ET in the two basins experienced increases of 9.5 and 6.3 mm/10a during 2003-2016, respectively ( Figure 3a). Seasonally, ET showed an increasing trend in the spring and summer in two basins, and the greatest increase appeared in the summer, with rates of 64 mm/a and 56 mm/a, respectively (Figure 3b and Table 2). The decreased trend occurred in the autumn of the LRB and in the winter of the MRB (Figure 3b and Table 2). The area average magnitudes of ET were 102.2 mm, 224.7 mm, 29.7 mm and 1.6 mm for the spring, summer, autumn and winter in the MRB. The area average magnitudes of ET were 93.8 mm, 214.8 mm, 25.5 mm and 0.4 mm for the spring, summer, autumn and winter in the LRB.

Changes in TWS
TWS anomalies in the GRACE dataset showed significantly decreased trends during 2003-2016 in the MRB and LRB, with declines of 30.3 mm/a and 18.9 mm/a ( Figure 4a and Table 3

Changes in TWS
TWS anomalies in the GRACE dataset showed significantly decreased trends during 2003-2016 in the MRB and LRB, with declines of 30.3 mm/a and 18.9 mm/a ( Figure 4a and Table 3). The variations indicate more freshwater flowing out of two basins, which showed  Figure 4b and Table 3 show that the TWS in the LRB and MRB had negative anomalies in the spring, summer and winter and a positive trend in the autumn. The greatest decrease occurred in the spring in the MRB, while it occurred in the winter in the LRB. The state of the seasonal deficit was stronger in the MRB than in the LRB, while the state of gain was stronger in the LRB than in the MRB. The results indicated that more water was released in the MRB and LRB in the spring, summer and winter and stored in the autumn.

Changes in TWS
TWS anomalies in the GRACE dataset showed significantly decreased trends during 2003-2016 in the MRB and LRB, with declines of 30.3 mm/a and 18.9 mm/a ( Figure 4a and Table 3 Figure 4b and Table 3 show that the TWS in the LRB and MRB had negative anomalies in the spring, summer and winter and a positive trend in the autumn. The greatest decrease occurred in the spring in the MRB, while it occurred in the winter in the LRB. The state of the seasonal deficit was stronger in the MRB than in the LRB, while the state of gain was stronger in the LRB than in the MRB. The results indicated that more water was released in the MRB and LRB in the spring, summer and winter and stored in the autumn.

Changes in Runoff
The annual runoff showed obvious changes during 2003-2016 and experienced an upward trend with a rate of 2.0 mm/a in the LRB (Figure 5a and Table 4). A rapid increase in the annual runoff occurred before 2008, when the TWS also showed a significant increase. In the MRB, the runoff showed a slight decreased trend with a rate of −0.9 mm/a ( Figure 5a and Table 4). The mean annual runoff depths were 167.0 mm and 248.6 mm in the MRB and LRB, respectively. The runoffs in the spring and summer increased in the LRB, while it significantly decreased in the autumn and winter (Figure 5b and Table 4). The spring runoff increased in the Siberian regions primarily due to climate warming during snowmelt periods associated with early snowmelts [75][76][77][78]. The maximum and minimum magnitudes of the trends occurred in the spring and autumn in the LRB, respectively, with rates of 26.2 mm/10a and −10.7 mm/10a. The runoff depths in the spring, summer, autumn and winter were 27.5 mm, 149.9 mm, 57.5 mm and 13.7 mm in the LRB. In the MRB, the runoff decreased in all the seasons, and the minimum trend was in the summer with a rate of −6.3 mm/10a (Figure 5b and Table 4). The runoff depths in the spring, summer, autumn and winter were 34.7 mm, 74.6 mm, 39.5 mm and 18.2 mm.
ing snowmelt periods associated with early snowmelts [75][76][77][78]. The maximum and minimum magnitudes of the trends occurred in the spring and autumn in the LRB, respectively, with rates of 26.2 mm/10a and −10.7 mm/10a. The runoff depths in the spring, summer, autumn and winter were 27.5 mm, 149.9 mm, 57.5 mm and 13.7 mm in the LRB. In the MRB, the runoff decreased in all the seasons, and the minimum trend was in the summer with a rate of −6.3 mm/10a (Figure 5b and Table 4). The runoff depths in the spring, summer, autumn and winter were 34.7 mm, 74.6 mm, 39.5 mm and 18.2 mm.   Figures 6a,b showed the hydrological separation in two basins from 2003 to 2016 calculated by Equation (8). The annual surface flow and base flow showed increased trends in the LRB, whereas both showed a decreased trend in the MRB. The surface flow and base flow contributed 56% and 44% to the increased total runoff in the LRB, and the surface flow impacted stronger on the runoff than the base flow. In the MRB, the surface flow and base flow contributed to a 39% and 61% decrease in the total runoff, and the decreased runoff was mainly influenced by the base flow. The annual TWS anomalies were more   Figure 6a,b showed the hydrological separation in two basins from 2003 to 2016 calculated by Equation (8). The annual surface flow and base flow showed increased trends in the LRB, whereas both showed a decreased trend in the MRB. The surface flow and base flow contributed 56% and 44% to the increased total runoff in the LRB, and the surface flow impacted stronger on the runoff than the base flow. In the MRB, the surface flow and base flow contributed to a 39% and 61% decrease in the total runoff, and the decreased runoff was mainly influenced by the base flow. The annual TWS anomalies were more highly related to the surface flow and base flow than precipitation in the LRB (Figure 6c,e). The annual precipitation and TWS showed a significantly decreased trend during 2003-2016, with more TWS released the same as the runoff. Therefore, the increased annual runoff in this area was mainly attributed to the TWS increase. In the MRB, the correlation coefficient between annual precipitation and surface flow was 0.45, and changes in the annual TWS showed stronger relationships with the surface flow and base flow, which were 0.66 and 0.62, respectively. Both annual precipitation and TWS showed decreased trends, suggesting the control of the annual precipitation and TWS on the interannual variability in the runoff.

The Impacts of Precipitation, Temperature and TWS on the Stream Flow
Temperature is an important factor that affects the changes of runoffs in the Arctic region. The relationship between temperature and runoff is shown in Figure 7. The temperature influence on the runoff was comparatively less at temperatures less than 0 • C in the MRB and Lena (Figure 7a,b). The increased rate of the runoff depth was 0.23 mm/ • C in the two basins. This may be because the runoff was generated from the base flow. Previous studies have shown that the snow depth and runoff during the spring showed significant correlations in Arctic rivers [42]. At temperatures above 0 • C, the runoff increased dramatically as snowmelt (Figure 7a,b). In this case, the snowmelt water increased with the increase in temperature. The increased rate of runoff depth in Lena was higher than that in Mackenzie. 2016, with more TWS released the same as the runoff. Therefore, the increased annual runoff in this area was mainly attributed to the TWS increase. In the MRB, the correlation coefficient between annual precipitation and surface flow was 0.45, and changes in the annual TWS showed stronger relationships with the surface flow and base flow, which were 0.66 and 0.62, respectively. Both annual precipitation and TWS showed decreased trends, suggesting the control of the annual precipitation and TWS on the interannual variability in the runoff. Temperature is an important factor that affects the changes of runoffs in the Arctic region. The relationship between temperature and runoff is shown in Figure 7. The temperature influence on the runoff was comparatively less at temperatures less than 0 °C in the MRB and Lena (Figure 7a,b). The increased rate of the runoff depth was 0.23 mm/°C in the two basins. This may be because the runoff was generated from the base flow. Previous studies have shown that the snow depth and runoff during the spring showed significant correlations in Arctic rivers [42]. At temperatures above 0 °C, the runoff increased dramatically as snowmelt (Figure 7a,b). In this case, the snowmelt water increased with the increase in temperature. The increased rate of runoff depth in Lena was higher than that in Mackenzie.

Seasonal Cycles of Hydrological Components
As shown in Figure 8, the trends of the precipitation and ET agreed relatively well in the LRB and MRB, both peaking around June-August and reaching a minimum between November and February. The precipitation from November to March was higher in the MRB than in the LRB. However, it was lower from May to October. The ET was higher in the MRB than in the LRB. The TWS showed positive anomalies from February to June and negative anomalies from July to January in the two basins. As seen, the runoff remained high during May-October and then decreased to reach the annual low in April before increasing again throughout the second half of the year. The positive anomaly of the TWS was caused by the accumulation of snow, low ET and runoff. It is evident that winter

Seasonal Cycles of Hydrological Components
As shown in Figure 8, the trends of the precipitation and ET agreed relatively well in the LRB and MRB, both peaking around June-August and reaching a minimum between November and February. The precipitation from November to March was higher in the MRB than in the LRB. However, it was lower from May to October. The ET was higher in the MRB than in the LRB. The TWS showed positive anomalies from February to June and negative anomalies from July to January in the two basins. As seen, the runoff remained high during May-October and then decreased to reach the annual low in April before increasing again throughout the second half of the year. The positive anomaly of the TWS was caused by the accumulation of snow, low ET and runoff. It is evident that winter precipitation significantly influenced the hydrological cycle in the MRB and LRB. The surface flow was concentrated from May to September, and the differences between the high runoff and low runoff were obviously mainly due to the snowmelt [77,79]. The base flow from November to May was higher in the MRB than in the LRB, and it was obviously higher from June to October in the LRB. The maximum base flow occurred in August and September, which lagged 2 to 3 months compared to the surface flow. In the Arctic region, the permafrost conditions also highly influenced the hydrological regime; basins with more permafrost coverage (such as LRB) had lower subsurface storage capacity and, thus, a lower winter base flow and a higher summer peak flow [80][81][82].

Seasonal Cycles of Hydrological Components
As shown in Figure 8, the trends of the precipitation and ET agreed relatively well in the LRB and MRB, both peaking around June-August and reaching a minimum between November and February. The precipitation from November to March was higher in the MRB than in the LRB. However, it was lower from May to October. The ET was higher in the MRB than in the LRB. The TWS showed positive anomalies from February to June and negative anomalies from July to January in the two basins. As seen, the runoff remained high during May-October and then decreased to reach the annual low in April before increasing again throughout the second half of the year. The positive anomaly of the TWS was caused by the accumulation of snow, low ET and runoff. It is evident that winter precipitation significantly influenced the hydrological cycle in the MRB and LRB. The surface flow was concentrated from May to September, and the differences between the high runoff and low runoff were obviously mainly due to the snowmelt [77,79]. The base flow from November to May was higher in the MRB than in the LRB, and it was obviously higher from June to October in the LRB. The maximum base flow occurred in August and September, which lagged 2 to 3 months compared to the surface flow. In the Arctic region, the permafrost conditions also highly influenced the hydrological regime; basins with more permafrost coverage (such as LRB) had lower subsurface storage capacity and, thus, a lower winter base flow and a higher summer peak flow [80][81][82]. Precipitation had a positive correlation with the ET and runoff in general ( Figure  9a,b). The relationships between monthly precipitation and ET were better correlated in the LRB (R = 0.82, p = 0.001) than in the MRB (R = 0.79, p = 0.002) (Figure 9a). However, there were also phase differences for different months. There were higher positive relationships in March and October and high negative relationships in May and July ( Figure  10a). The relationships between the monthly precipitation and runoff were also better correlated in the LRB (R = 0.82, p = 0.0001) than in the MRB (R = 0.79, p = 0.0005) (Figure 9b). The monthly precipitation more highly positively influenced by the monthly runoff mainly occurring in January, April and November in the LRB, and it occurred only in October in the MRB, which indicated that the runoff was influenced by precipitation (Figure 10b). For the TWS, there were negative correlation with precipitation in general (Figure 9c). The relationships between the monthly precipitation and TWS were better correlated in the MRB (R = 0.58, p = 0.04) than in the LRB (R = 0.55, p = 0.07) (Figure 9c). In the MRB, the monthly precipitation showed higher negative relationships in January and Feb-Surfaceflow depth (mm) Precipitation had a positive correlation with the ET and runoff in general (Figure 9a,b). The relationships between monthly precipitation and ET were better correlated in the LRB (R = 0.82, p = 0.001) than in the MRB (R = 0.79, p = 0.002) (Figure 9a). However, there were also phase differences for different months. There were higher positive relationships in March and October and high negative relationships in May and July (Figure 10a). The relationships between the monthly precipitation and runoff were also better correlated in the LRB (R = 0.82, p = 0.0001) than in the MRB (R = 0.79, p = 0.0005) (Figure 9b). The monthly precipitation more highly positively influenced by the monthly runoff mainly occurring in January, April and November in the LRB, and it occurred only in October in the MRB, which indicated that the runoff was influenced by precipitation (Figure 10b).
For the TWS, there were negative correlation with precipitation in general (Figure 9c). The relationships between the monthly precipitation and TWS were better correlated in the MRB (R = 0.58, p = 0.04) than in the LRB (R = 0.55, p = 0.07) (Figure 9c). In the MRB, the monthly precipitation showed higher negative relationships in January and February, whereas they showed higher positive relationships in August and October (Figure 10c). The positive relationships between monthly precipitation and TWS were higher in July, August, September and November in the LRB, suggesting that the precipitation impacted on the TWS (Figure 10c). Overall, the impact of the precipitation on the ET, runoff and TWS on a monthly scale showed huge differences in the different river basins over the Arctic region. tionships in March and October and high negative relationships in May and July ( Figure  10a). The relationships between the monthly precipitation and runoff were also better correlated in the LRB (R = 0.82, p = 0.0001) than in the MRB (R = 0.79, p = 0.0005) (Figure 9b). The monthly precipitation more highly positively influenced by the monthly runoff mainly occurring in January, April and November in the LRB, and it occurred only in October in the MRB, which indicated that the runoff was influenced by precipitation (Figure 10b). For the TWS, there were negative correlation with precipitation in general (Figure 9c). The relationships between the monthly precipitation and TWS were better correlated in the MRB (R = 0.58, p = 0.04) than in the LRB (R = 0.55, p = 0.07) (Figure 9c). In the MRB, the monthly precipitation showed higher negative relationships in January and February, whereas they showed higher positive relationships in August and October ( Figure  10c). The positive relationships between monthly precipitation and TWS were higher in July, August, September and November in the LRB, suggesting that the precipitation impacted on the TWS (Figure 10c). Overall, the impact of the precipitation on the ET, runoff and TWS on a monthly scale showed huge differences in the different river basins over the Arctic region.

Telationship between Precipitation and Other Hydrological Components
As the key component and driving factor in the hydrologic cycle, precipitation has a prominent influence on the interactions among other hydrologic components [83]. There is a lag time from when water is introduced as the precipitation compared to when it becomes surface water, ET and TWS, which is typically described empirically as the total precipitation in the hydrological balance [84]. The time lags of the ET, runoff and TWS responses to precipitation are shown in Figure 11. The results indicated that the ET and runoff were most highly related to precipitation over a 1-month lag, i.e., the monthly ET and runoff were strongly influenced by the precipitation that occurred 1 month earlier in the study area. The negative relationships between precipitation and ET and runoff occurred after a lag of 4 months, and the maximum negative relationships both occurred after a 6-month time lag. The monthly precipitation and the TWS were negatively correlated without any lag and poorly negatively correlated over a 1-month lag. As time goes by, the relationships between precipitation and TWS converted from negative values to positive values in the study areas. The TWS were most highly related to precipitation over a time lag of 4 months in the two basins. These results suggested that more precipitation would lead to more TWS when precipitation preceded the TWS by 4 months; in contrast, it would result in less TWS when precipitation preceded it by 0 to 1 month. It is also found

Telationship between Precipitation and Other Hydrological Components
As the key component and driving factor in the hydrologic cycle, precipitation has a prominent influence on the interactions among other hydrologic components [83]. There is a lag time from when water is introduced as the precipitation compared to when it becomes surface water, ET and TWS, which is typically described empirically as the total precipitation in the hydrological balance [84]. The time lags of the ET, runoff and TWS responses to precipitation are shown in Figure 11. The results indicated that the ET and runoff were most highly related to precipitation over a 1-month lag, i.e., the monthly ET and runoff were strongly influenced by the precipitation that occurred 1 month earlier in the study area. The negative relationships between precipitation and ET and runoff occurred after a lag of 4 months, and the maximum negative relationships both occurred after a 6-month time lag. The monthly precipitation and the TWS were negatively correlated without any lag and poorly negatively correlated over a 1-month lag. As time goes by, the relationships between precipitation and TWS converted from negative values to positive values in the study areas. The TWS were most highly related to precipitation over a time lag of 4 months in the two basins. These results suggested that more precipitation would lead to more TWS when precipitation preceded the TWS by 4 months; in contrast, it would result in less TWS when precipitation preceded it by 0 to 1 month. It is also found that the monthly precipitation mainly formed ET and runoff over the next 3 months; after that, the ET and runoff would gradually decrease, and more precipitation was stored in the basins, which induced the TWS increase in the LRB and MRB. runoff were most highly related to precipitation over a 1-month lag, i.e., the monthly ET and runoff were strongly influenced by the precipitation that occurred 1 month earlier in the study area. The negative relationships between precipitation and ET and runoff occurred after a lag of 4 months, and the maximum negative relationships both occurred after a 6-month time lag. The monthly precipitation and the TWS were negatively correlated without any lag and poorly negatively correlated over a 1-month lag. As time goes by, the relationships between precipitation and TWS converted from negative values to positive values in the study areas. The TWS were most highly related to precipitation over a time lag of 4 months in the two basins. These results suggested that more precipitation would lead to more TWS when precipitation preceded the TWS by 4 months; in contrast, it would result in less TWS when precipitation preceded it by 0 to 1 month. It is also found that the monthly precipitation mainly formed ET and runoff over the next 3 months; after that, the ET and runoff would gradually decrease, and more precipitation was stored in the basins, which induced the TWS increase in the LRB and MRB.

Conclusions
In this study, the changes of the hydrological components in two representative river basins in the Arctic region, namely the Mackenzie River Basin (MRB) and the Lena River Basin (LRB), were investigated using multiple satellite remote sensing and observational datasets during 2003-2016. The main conclusions were as follows: (1) The changes of the hydrological components in the different basins of Arctic region showed different trends. The annual precipitation decreased at rates of 1.9 mm/10a and 18.8 mm/10a in the MRB and LRB, respectively, which was much greater in Eastern Siberia than in North America. The magnitude of the warm trends was obviously more in the MRB of North America than the LRB of Eastern Siberia, which were 0.79 °C/10a and 0.44 °C/10a, respectively. The ET increased more rapidly in the MRB than in the LRB, with

Conclusions
In this study, the changes of the hydrological components in two representative river basins in the Arctic region, namely the Mackenzie River Basin (MRB) and the Lena River Basin (LRB), were investigated using multiple satellite remote sensing and observational datasets during 2003-2016. The main conclusions were as follows: (1) The changes of the hydrological components in the different basins of Arctic region showed different trends. The annual precipitation decreased at rates of 1.9 mm/10a and 18.8 mm/10a in the MRB and LRB, respectively, which was much greater in Eastern Siberia than in North America. The magnitude of the warm trends was obviously more in the MRB of North America than the LRB of Eastern Siberia, which were 0.79 • C/10a and 0.44 • C/10a, respectively. The ET increased more rapidly in the MRB than in the LRB, with rates of 9.5 and 6.3 mm/10, respectively.
(2) The TWS showed a decreased trend during 2003-2016 in the MRB and LRB, with declines of 30.3 mm/a and 18.9 mm/a. The change trends of the runoffs were different with the TWS in the study area. The surface flow and base flow contributed 56% and 44% of the increased total runoff in the LRB. In the MRB, the surface flow and base flow contributed 39% and 61% of the decreased total runoff, so the decreased runoff was mainly influenced by the base flow. The increased annual runoff in the LRB was mainly attributed to a lower ET and less of a deficit of TWS, while a higher ET and more of a deficit of TWS in the MRB controlled the variability in the runoff.
(3) The mean annual cycles of precipitation, ET, TWS, runoff depth, surface flow and base flow in the LRB and MRB showed deviations in the magnitudes and distributions. The precipitation on a monthly scale influenced the ET, runoff and TWS and showed huge differences in the different river basins over Arctic region. There were obviously lag times from rainfall to surface water, ET and TWS in the study areas. The monthly precipitations mainly formed the ET and runoff for the next 3 months; after that, the ET and runoff would gradually decrease, and more precipitation was stored in the basins, which induced TWS increase in the LRB and MRB.