1. Introduction
Hydrological impacts of climate change have drawn worldwide concern, especially in cold regions due to the existence of the climate-sensitive cryospheric hydrogeology environment, such as frozen ground, seasonal snowpack, etc. [
1,
2,
3,
4,
5]. Climate change impacts the hydrological change of cold regions, both directly through changes in climate forcing (precipitation and evaporation) and indirectly through changes in runoff generation caused by the degrading cryosphere, which has increased the uncertainties and complexities of hydrological studies in cold regions [
6,
7].
As an important component of the cryosphere, frozen ground is a unique hydrogeological composition related to runoff generation processes in cold regions [
3,
8,
9]. Frozen ground, which consists of seasonally frozen ground and permafrost, is defined as a subsurface region with a temperature below 0 °C. The part of the frozen ground which remains in a frozen state (≤0 °C) for at least two consecutive years is further defined as permafrost [
10]. With global warming, the frozen depth of the active layer has decreased, leading taliks to form above the permafrost table, which will increase groundwater discharge to the river in winter [
11,
12]. In previous studies, increasing baseflow and the changing recession process induced by frozen ground degradation in cold-region basins have been reported worldwide [
13,
14,
15,
16]. Meanwhile, recession-flow characteristics, annual minimum flow, and winter baseflow were used to indicate and estimate frozen ground degradation rates in cold-region basins [
17,
18]. As another important component of cryosphere, seasonal snowpack is a key surface water resource of cold regions, which accumulates in winter and releases in spring and summer. Changes of snowpack under the warming climate have been altering the hydrological regime and hydrograph characteristics in cold regions [
19]. Meanwhile, frozen ground is characterized as having a low permeability, which limits the water infiltration [
20,
21], making the direct runoff ratio as well as the snowmelt runoff ratio higher in cold regions compared to temperate regions [
22]. The existence of frozen ground impedes snowmelt infiltration, resulting in spring floods [
23,
24,
25]. Snowmelt runoff is mainly feed by seasonal snowpack, meanwhile, its generation processes are also related to frozen ground. Therefore, the hydrological impacts of cryosphere degradation on snowmelt runoff under the changing climate are more complex and merit investigation. In conclusion, the cold season runoff, such as recession flow, winter baseflow, and spring runoff, is sensitive to climate change and cryosphere degradation. However, the interactions and alternations between recession flow, winter baseflow, and spring runoff under the changing climate are rarely studied.
The Tibetan Plateau is known as the third pole of the world, with an extremely cold environment and widely distributed cryosphere. The headwaters of the Yellow River (HWYR) are situated in the east of the Tibetan Plateau where there is a transitional area of permafrost and seasonally frozen soil. Under the warming climate, frozen ground has dramatically degraded, which has manifested as a shrinking permafrost area, a warming permafrost temperature, a decreasing frozen depth, a reducing frozen duration of seasonally frozen ground, and a formation of taliks [
26,
27,
28]. The runoff of the HWYR has undergone tremendous changes since the 1990s. The runoff change in the headwaters of the Yellow River is primarily caused by climate change rather than land use transition [
29]. The climatic impacts on long-term variations and water balance of annual runoff of the HWYR have been extensively discussed at an annual scale [
29,
30,
31]. The long-term variations of annual runoff of the HWYR can be generally divided into three periods: a pre-change period (1961–1990), a decrease period (the 1990s), and a recovery period (the 21st century) [
30,
31]. Compared to the pre-change period (1961–1990), the annual mean runoff of the headwater basins decreased by 20% in the 1990s, but demonstrated a recovery in the 21st century, especially in the sub-basin above Jimai station where runoff increased by 6.2% relative to the pre-change period [
32]. The frozen ground degradation effects on the hydrological change have also drawn concern in this area recently, but have only been discussed at an annual scale [
16,
32,
33,
34]. Although long-term annual runoff characteristic changes have been extensively discussed and investigated in previous studies, the long-term variations of runoff in different seasons are still unknown. Moreover, the ways in which runoff in cold seasons—such as the snowmelt runoff, recession flow, and the winter baseflow—responds to climate change, such as temperature increase, changes in precipitation types, have not been thoroughly studied. However, to understand how climate change directly alters the runoff in different seasons and how the hydrological processes respond to the changing cryosphere under the climate change, it is imperative to analyze the long-term variations of runoff in different seasons.
This study analyzed the long-term runoff changes in the HWYR, with the aims to (1) characterize the long-term runoff variations of different seasons in different change periods, (2) analyze the potential causes of runoff change in considering global warming, precipitation change, and frozen degradation, (3) analyze frozen ground degradation impacts on cold season runoff; and (4) provide a better understanding of runoff response to climate change in the HWYR.
5. Discussion
In previous studies, the decrease of annual mean runoff (AMR) and rainy season runoff (RSR) in the low-flow period (1990–2002) is generally attributed to decreases in precipitation, and increases in evaporation [
29,
31]. Meanwhile, the significantly decreased precipitation intensity is also a crucial factor that resulted in an evaporation increase in the 1990s. The increase in precipitation after 2002 offset the effects of increased evaporation, and is the main reason leading to a recovery of the runoff in the recovery period (2003–2013) [
31,
32].
This study focused on the cold season runoff changes. The cold season runoff, recession flow (RF), and winter baseflow (WB) shared one abrupt change point (1989) with rainy season runoff (RSR). It is a rational result, as RF and WB are results from the recession of RSR. In addition, RSR, RF, and WB are highly correlated with each other (
Table 2). However, the second abrupt change points of RF and WB have a three-year (1999) deviation and a one-year (2001) deviation compared to RSR (2002), indicating the runoff supplement and runoff processes changed during the 1990s (
Figure 5). Both cold season precipitation and freezing–thawing indices showed abrupt changes in the 1990s (
Table 4 and
Table 5). The potential reasons for the inconsistent variations of RSR, WB, and RF are the changes of cold season precipitation and frozen ground. Meanwhile, previous studies reported that the existence of frozen ground impedes infiltration, resulting in higher direct flow ratio [
23,
24,
25]. However, the ways in which the snowmelt runoff changes along with cold season precipitation changes and frozen ground degradation are still unknown. In this section, the precipitation change impacts on cold season runoff and the potential hydrological effects of frozen ground degradation are discussed.
5.1. Precipitation Change Impacts on Cold Season Runoff
The correlations between cold season runoff and cold season precipitation are shown in
Table 6. By comparing
Table 2 and
Table 6, we found that the winter baseflow (WB) is mainly controlled by recession flow (RF) and rainy season runoff (RSR) but is also affected by (Sleet + Rain)
(9–2). Snowmelt runoff (SR) is mainly controlled by (Snow
(9–5) + (Sleet + Rain)
(9–2)), RF, and WB. After removing the WB impacts on SR, the correlation coefficient between direct snowmelt runoff (DSR) and Snow
(9–5) increased to 0.54. The correlation coefficient between direct snowmelt runoff (DSR) and Snow
(9–5) + (Sleet + Rain)
(9–2) is 0.55. The correlation coefficient between direct snowmelt runoff (DSR) and Snow
(9–5) + (Sleet + Rain)
(3–5) is 0.29. Therefore, DSR is mainly fed by the snowmelt from the accumulative snow in cold season (ASC), therefore, the changes of precipitation types between September and May have little impacts on direct snowmelt runoff change. The recession flow (RF) has a relatively weak correlation with cold season precipitation, which is mainly controlled by rainy season runoff (
Table 2).
The increased (Sleet + Rain)
(9–2), caused by the increase of air temperature in the 1990s, is an extra supplement for RF and WB compared to the pre-change period (1961–1989). RSR and RF showed a decrease trend in the 1990s, however, WB increased significantly at a rate of 1.10 m
3/s/year (
Table 3). In addition, (Sleet + Rain)
(9–2) showed no significant change until 1999, indicating that the change of supplement could not fully explain the gradually increase phenomenon of WB in the 1990s. Therefore, another important factor, frozen ground degradation, should be involved in the explanation of WB changes. In the 2000s, both RF and WB showed a significant increase trend, partly because the extra supplement, (Sleet + Rain)
(9–2), increased by 6.35% from 2000 to 2013. The continuous effects of frozen ground degradation on WB increase in the 2000s are further discussed in the following section.
5.2. Frozen Ground Degradation Impacts on Cold Season Runoff
5.2.1. Frozen Ground Degradation Impacts on Winter Baseflow
As indicated by the obtained results in
Table 2, winter baseflow is significantly correlated to rainy season runoff (RSR) and recession flow (RF). Recession flow is a depletion process of rainy season runoff, in which surface runoff gradually decreased and was replaced by groundwater discharge. Winter baseflow is mainly fed by groundwater discharge, which is related to the water stored in the active layer and aquifer (WRSQ) in the rainy season [
45]. Due to the lack of groundwater observation data, RSR was used as an index to indicate volume of WRSQ in this study. Additionally, assume that RSR is linearly related with WRSQ when the aquifer characteristic does not evidently change. To investigate the variations of winter baseflow (WB), the related correlation analyses are plotted in
Figure 9a,b. The rainy season runoff (RSR) and winter baseflow (WB) showed similar variations, as shown in
Figure 9a. As shown in
Figure 9b, WB is significantly positively correlated with RSR (R
2 = 0.65,
p < 0.01) which supports the theory that winter baseflow is fed by water resources stored in aquifers in the rainy season. If we assume that aquifer characteristics, such as hydrological conductivity and storage capacity, have not undergone dramatic change, WB and RSR could show the same variation. As analyzed in
Section 4.1, however, the moving
t-test results indicate the abrupt change point of WB appeared three years earlier than RSR, indicating the change of the relationship between WB and RSR at the end of 1990s.
To analyze the effects of frozen ground degradation to winter baseflow (RSR) change, we defined winter groundwater discharge rate as the ratio of WB to RSR. The long-term variations of freezing–thawing indices and winter groundwater discharge rate (
Rw) are shown in
Figure 9c. The labelled years indicate the abrupt change points of freezing–thawing indices and
Rw detected by the moving
t-test. DDT showed a continuous significant increase (R
2 = 0.70,
p < 0.01) since 1986, DDF remained stable between 1985 and 1997 but showed a significantly decrease (R
2 = 0.22,
p < 0.01) in the period 1997–2013. The results indicate permafrost thawed depth significantly increased since 1985, and frozen depth significantly decreased since 1997. The enlarged thawed depth provided more room for water storage in the active layer in the rainy season. Meanwhile, the decrease of frozen depth leads to taliks formation which provided extra groundwater discharge channels. The combination effects of thawed depth and frozen depth changes lead more grounder water discharge in winter. The
Rw remains about 0.1–0.2 but gradually significantly increased (R
2 = 0.34,
p < 0.01) since 1985 and reached about 0.2–0.3 in the 21st century.
The correlation of
Rw and freezing–thawing indices after temperature significantly increased (1985) are plotted in
Figure 9d. The
Rw is significantly and negatively (R
2 = 0.14,
p < 0.01) correlated with DDF, and is significantly and positively (R
2 = 0.33,
p < 0.01) correlated with DDT. This finding indicates the increase in thawed depth and decrease in frozen depth will increase winter groundwater discharge rate, which is consistent with the understanding that under the warming climate the deeper permafrost table and the reduction frozen depth of active layer will lead more groundwater discharge in the winter [
12,
46]. The higher correlation coefficient between DDT and
Rw (R
2 = 0.33,
p < 0.01) indicated that the deeper thawed depth played a more important role in influencing groundwater discharge rate change. However, the correlation coefficient between thawing-freezing indices and
Rw (R
2 = 0.33, R
2 = 0.14) is lower than the correlation coefficient of RSR and WB (R
2 = 0.65,
p < 0.01), which indicates water resources stored in the rainy season is the main controlling factor of WB, and global warming induced frozen ground degradation altered the relationship between WB and WRSQ. Meanwhile,
Rw in the period 2000–2013 presented a more distinct fluctuation compared to period the 1985–2000 (
Figure 9c). This may be related to the increased sleet and rain from September to February since 1999, which provided an extra supplement to the winter baseflow.
5.2.2. Frozen Ground Degradation Impacts on Direct Snowmelt Runoff
The correlation analyses matrix between cold season runoff and cold season precipitation indicates that direct snowmelt runoff is mainly fed by accumulative snow in the cold season (ASC) (
Table 6). The direct snowmelt runoff (DSR) and respective runoff coefficient changes are plotted in
Figure 10a,b, respectively.
The direct snowmelt runoff coefficient (
Rs) was defined as the ratio of direct snowmelt runoff depth (mm) to accumulative snow in the cold season (ASC) (mm). As shown in
Figure 10a, the DSR slightly increased in the period 1961–1997 and decreased in the period 1998–2013, presenting a similar variations pattern with ASC plotted in
Figure 8b. The change pattern can be generally divided into two periods; a pre-change period (1961–1997) and a decrease period (1998–2013). The DSR remains at the same level (mean value 52.76 m
3/s) during the period 1986–1997 but sharply decreased to 36.30 m
3/s during the period 1998–2013. The decrease in the DSR (31.21%) is three times larger than the decrease in SR (8.64%) after removing the increasing winter baseflow (WB) impacts (
Table 3). Due to the impacts of gradually increased WB in the 1990s, SR presented a slight increase trend during the 1990s, although snow decreased.
Notably, the
Rs of DSR shown in
Figure 10b displayed a different variation pattern; the
Rs was approximately 0.16 from 1961 to 1980 but decreased with a rate of 0.0011/year from 1981 to 2013 (significant at 10% level (
p < 0.1)). The variation in the direct snowmelt runoff coefficient (
Rs) is different from direct snowmelt runoff, indicating there is another factor controlling snowmelt runoff generation besides accumulative snow in cold season (ASC). Since frozen ground is characterized as having low permeability, one possible reason for the
Rs decrease is the decreased frozen depth and shortened frozen duration, leading to more snowmelt infiltration in the ground in spring rather than directly contributing to the runoff. As indicated by previous studies, the hillslopes produce more runoff than those without permafrost, moreover, snowmelt runoff in cold regions has a higher direct runoff ratio than the storm runoff in watersheds without frozen ground [
22,
47,
48].
To analyze frozen ground degradation impacts on direct snowmelt runoff generation, DDF and DDT were used to represent the frozen depth and thawed depth of the frozen ground. The correlations between freezing–thawing indices and direct snowmelt runoff coefficient (
Rs) are shown in
Figure 11.
As indicated by the comparison of
Figure 11a,b, the direct snowmelt runoff coefficient (
Rs) is more related to DDF (0.42,
p < 0.01) rather than to DDT (0.16,
p < 0.01). This indicates that ground frozen depth in winter controlled direct snowmelt runoff generation, and with the deeper frozen depth and the longer frozen duration in the winter, more direct snowmelt runoff will be produced in the spring. The results indicate frozen ground degradation will reduce direct snowmelt runoff generation.
5.3. Frozen Ground Degradation Impacts on Water Balance
In this study, we ignored frozen ground degradation impacts on water balance in the HWYR and focused on its impacts on winter baseflow and snowmelt runoff. Nevertheless, the frozen ground degradation impacts have been considered in long-term water balance analyses lately [
7,
32,
34]. Duan [
7] used a sensitive-based method to analyze permafrost effects on runoff change of watersheds in north-eastern China and concluded that permafrost thaw increased runoff, but it is difficult to separate the permafrost degradation impacts from complex driving factors. Based on a modified decomposition method, Wu [
32] also found that the permafrost degradation presented a positive effect, which is contrary to temperature impacts on runoff change in the headwaters of the Yellow River (HWYR). Sun [
49] used a water balance simulation model to simulate and quantify the runoff change in the HWYR. The study found that total permafrost degradation will increase discharge at an annual scale. However, Wang [
34] reported a different conclusion based on the Budyko framework, indicating that frozen ground degradation decreases the runoff of the HWYR by about 34%. In addition, Zheng [
50] and Zhang [
51] used a physical based model with different freeze–thaw process scenarios to simulate runoff in cold regions, and the results showed that the simulated runoff generally reduced when the soil freeze–thaw mechanism is not considered.
As described above, the conclusions of frozen ground degradation’s contribution to water balance are still debatable. Meanwhile, as indicated by the moving
t-test and the statistical results, the change in precipitation, precipitation intensities, and temperature can explain the annual mean runoff changes. The results agree with most previous studies that the long-term annual runoff change in the HWYR is mainly controlled by climate changes (decrease in precipitation and increase in evaporation) [
29,
30,
31,
32]. Furthermore, it is difficult to separate the frozen ground degradation impacts on water balance, because the increase in temperature affects actual evaporation and the frozen ground at the same time [
32]. Considering frozen ground degradation impacts on water balance by a statistical method is out the scope of this study. However, frozen ground impacts on annual water balance is still an unsolved question which deserves further analyses in future research.
6. Conclusions
In order to investigate hydrological change and its relationship with the changing climate, and degrading frozen ground in the HWYR, we analyzed the long-term variations of runoff compositions, climate factors, and degrading frozen ground. Moreover, we further discussed the correlations between hydrological change and climate change, especially degrading frozen ground following climate change in this study. According to the obtained results, several conclusions can be drawn:
- (1)
The long-term annual runoff variations can generally be divided into three periods: I. a pre-change period (1961–1989), II. a low-flow period (1990–2002), and III. a recovery period (2003–2013). In period II, the decrease in high flow (Pe < 30%) and low flow (Pe > 70%) presented the same levels (about 30%) as period I. In period III, however, low flow increased by 40% but high flow increased by less than 20% relative to period I.
- (2)
The intra-annual variations of runoff in the HWYR can be divided into winter baseflow (January to February), snowmelt runoff (March to May), rainy season runoff (June to September), and recession flow (October to December), which account for 4%, 15%, 61%, and 20% of annual runoff, respectively.
- (3)
The winter baseflow is mainly controlled by rainy season runoff, but the frozen ground degradation altered the relationship between rainy season runoff and winter baseflow. Frozen ground degradation increased ground water discharge rate in winter.
- (4)
Direct snowmelt runoff remained at the same level during the period 1961–1997 but decreased 31.21% in the period 1997–2013 relative to the period 1961–1997. The direct snowmelt runoff coefficient (Rs) has been linearly decreasing at a rate of 0.0011/year since 1980, which is significantly controlled by DDF.
According to the obtained conclusions summarized above, we can deduce that with the decreasing accumulative snow in the cold season and the degrading frozen ground, the direct snowmelt runoff of the HWYR may decrease continuously under a warming climate in the future.