Processes of runoff in seasonally-frozen ground about a forested catchment of semiarid mountains

PengFei Lin,ZhiBin He,Jun Du,LongFei Chen,Xi Zhu,QuanYan Tian

1.Linze Inland River Basin Research Station,Chinese Ecosystem Research Network,Lanzhou,Gansu 730000,China

2. Key Laboratory of Ecohydrology of Inland River Basin, Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou,Gansu 730000,China

ABSTRACT Climate warming increases the variability in runoff of semiarid mountains where seasonally-frozen ground is widely distributed.However,what is not well understood are the processes of runoff,hydrological drivers,and freeze-thaw cycles in seasonally-frozen ground in semiarid mountains.To understand how freeze-thaw cycles affect runoff processes in seasonally-frozen ground, we monitored hydrological processes in a typical headwater catchment with seasonally-frozen ground in Qilian Mountain, China, from 2002 to 2017. We analyzed the responses of runoff to temperature, precipitation, and seasonally-frozen ground to quantify process characteristics and driving factors. The results show that annual runoff was 88.5 mm accounting for 25.6% of rainfall,mainly concentrated in May to October,with baseflow of 36.44 mm.Peak runoff occurred in June,August, and September, i.e., accounting for spring and summer floods. Runoff during the spring flood was produced by a mix of rainfall,melting snow,and melting seasonally-frozen ground,and had a significant correlation with air temperature. Runoff was mainly due to precipitation accumulation during the summer flood.Air temperature,average soil temperature at 0-50 cm depth,and frozen soil depth variable explained 59.60%of the variation of runoff in the thawing period, while precipitation variable explained 21.9%.Thawing-period runoff and soil temperature had a ＞0.6 correlation coefficient(P ＜0.05).In the rainfall-period,runoff was also affected by temperature,soil moisture,and precipitation,which explained 33.6%,34.1%and 18.1%,respectively.Our results show that increasing temperature and precipitation will have an irreversible impact on the hydrological regime in mountainous basins where seasonally-frozen ground is widely distributed.

Keywords:runoff;seasonally-frozen ground;semiarid mountains;Northeast margin of Tibetan Plateau

1 Introduction

Runoff in semiarid catchments is a major mechanism by which water,sediment,nutrients,and contaminants are moved and redistributed (Wilcoxet al.,1997; Heet al., 2012). However, runoff is also a recurrent problem which leads to competition for water resources between upland and lowland areas as population and water needs increase (Lacombeet al.,2008;Kelleyet al.,2017).

Mountain runoff as the"water tower"of an arid area,plays an important role in replenishing and regulating the water resources in arid areas (Messerliet al.,2004; Wanget al., 2011). Understanding the processes of runoff in mountain headwaters is fundamental to catchment hydrology (Buttle, 2006). In recent years,seasonal freeze-thaw and the generation and confluence processes of runoff have changed significantly due to climate change in mountain areas.In particular,changing precipitation affected the discharge and increasing temperature changed the freeze-thaw cycle,all of which increased the uncertainty of water resources in arid mountain areas (Rangecroftet al.,2013; Liet al., 2018). As a result, mountain runoff has become a hot topic in the study of the hydrological cycle in arid areas undergoing climate change(Ding and Zhang,2018).

In semiarid mountain regions,runoff can be divided into direct runoff and baseflow. Direct runoff is mainly controlled by rainfall intensity (Martı′Nez-Menaet al.,1998;Cantonet al.,2001).Baseflow,however, is considered to be the outflow of groundwater feeding rivers during dry periods(Wittenberg and Sivapalan, 1999). Typically, the proportion of direct runoff is much larger than that of baseflow, and there is a high correlation between the amounts of runoff and rainfall in most settings. However, in semiarid mountain regions, with widespread seasonally-frozen ground, runoff processes may be different. For example, seasonal freeze-thaw cycles, and the presence of forest soils with macropore pathways, may be very important to the generation of vertical and lateral flow even under unsaturated conditions. The processes of runoff in seasonally-frozen ground are unique considering low hydraulic conductivity, high soil water content, seasonal freeze-thaw processes, and the redistribution of melt water (Benseet al., 2012; Qinet al.,2016). Thus, the understanding of how rain, snowmelt, and seasonal freeze thaw cycles affect water flow in soils needs improvement. The reason is these processes affect soil moisture in early growing season and drive plant productivity in moisture-limited areas (Mulhollandet al., 1990; Rothfusset al., 2010;Yanget al.,2017).

The Qilian Mountains of northwestern China represent a semiarid mountain area of the Tibetan Plateau (TP), and form headwaters of the Heihe River Basin. Due to its ecological importance, the area has been extensively used in research, resulting in relatively abundant climatic and hydrological observations (Heet al., 2012; Liet al., 2018; Linet al.,2018). Thus, the spatio-temporal variability in permafrost and its influence on hydrology and vegetation in this area have been described (Zhanget al., 2003;Chenet al., 2007; Wang Qet al., 2013; Zhanget al.,2013; Zhouet al., 2014; Gaoet al., 2016; Qinet al.,2016; Penget al., 2017; Wanget al., 2017). Zhanget al.(2003)noted that the soil freeze-thaw cycle, which affected seasonal water storage, soil moisture, and evaporation, was the primary hydrological process during the monsoon season in the eastern TP. Wanget al. (2009) reported that the synergistic effect of changes in frozen soil and vegetation cover may be the major factor controlling surface runoff generation in permafrost in the TP. On the other hand,changes in frozen soil could also affect vegetation activity at the start of the vegetation growing season because of the soil water content change (Christensenet al., 2004; Guoet al., 2018). Previous studies have mostly focused on the role of frozen soil in thermal and hydrological regimes using distributed models for large-scale watersheds,or described longterm trends in the influence of frozen soil change on runoff. However, mechanistic research on changes in seasonally-frozen soil and the relationships between changes in frozen soil and runoff at small watershed scales in the semiarid mountains of northwestern China have received little attention. This is probably due to formidable challenges associated with measuring seasonally-frozen ground and flow in semiarid environments.

To address these unknowns,we studied runoff and seasonal freeze-thaw cycles in seasonally-frozen ground at a small-watershed scale in a semiarid region of the Qilian Mountains. The main objectives of this study were to: (i) determine the dynamics of seasonal freeze-thaw cycles, (ii) identify processes of runoff in response to rain events in a mountain basin,and (iii) analyze the effects of seasonal freezing and thawing on runoff.

2 Methods

2.1 Study site

Our study site is in the Pailugou catchment(100°17'E, 38°24'N) of the Qilian Mountains in Gansu Province, northwestern China (Figure 1). The catchment's total area is 2.93 km2, and elevations range from 2,700 to 3,860 m. Permanently and seasonally-frozen soils are widespread. The main parent material is calcareous rock, covered with a relatively thin (5 to 50 cm) soil layer. Soils have coarse texture, an intermediate organic matter content（9.50 to 31.09 kg/m2）, and pH ranging from 7 to 8 (Chenet al., 2016). Vegetation forms a mosaic of grassland,scrubland, and arbor forest. Forests are mostly found on shaded (north-facing) and semi-shaded slopes at 2,300-3,450 m above sea level, while grasslands are found on sunny slopes (south-facing). In the Pailugou catchment,Picea crassifoliais the constructive species, mainly distributed at 2,300-3,200 m above sea level.The dominant shrub species,mainly grow at an elevation of 2,800-3,650 m, areSalix oritrepha,Rhododendron przewalskii, andCaragana jubata.The dominant vegetation species arePotentilla fruticosa,P. glabra,Lonicera microphylla,Kobresis bellardii, andPolygonum viviparum, on sunny slopes the sedge Kobresis humilis and grasses of Stipa purpurea and Agropyron cirstatum.

Figure 1 Study area and observation instrument position

2.2 Measurements and analysis

2.2.1 Meteorological measurements

An environmental measurement system (ENVIS;IMKO, Ettlingen, Germany) and meteorological observation system (Campbell Scientific, Inc.) were installed at an elevation of 2,700,2,800 and 3,200 m,respectively.These sensors are used to continually monitor microclimate at 30-min intervals, such as air temperature, precipitation, wind velocity at two planes,global radiation, reflected radiation, net radiation, relative humidity, and soil temperature. Additionally,precipitation was measured and recorded with an automatic precipitation sensor(RG50,SEBA Hydrometrie, Germany) at elevations of 2,400-3,800 m at 200-m intervals for a total of 14 precipitation observation points.

2.2.2 Runoff measurements

The catchment runoff volume was measured in a weir installed at 2,650 m elevation, downstream from the study area;runoff volumes were recorded automatically using a water level gauge (SW40, China and Onset HOBO U20, USA) from 2002 to 2017. During winter (from December to March), when part of the weir froze and the meters could not normally work due to insufficient volume of runoff, we measured runoff volume using a container (volume = 0.01778 m3). Specifically, each day we measured the time (seconds)when runoff filled up the container, and then we used linear interpolation of runoff between pairs of measurements, graphed the results, and calculated total runoff during this period(the entire winter).

To understand the relationships between land surface,altitude,and runoff dynamics,eighteen micro-runoff plots (5×20 m) were established in the shrub, forest, shady-slope grassland, and sunny-slope grassland.We used concrete for each micro-plot construction.Concrete of the microplots was 10 cm thick and 40 cm high, with 20 cm embedded into the soil, leaving the remaining 20 cm above ground as the border for preventing runoff loss (Figure 2). Runoff was exported with plastic hose and recorded automatically using a tipper-type gauge(QT-50,China).

2.2.3Soil moisture, temperature, and soil freeze-thaw measurements

Volumetric soil water content and soil temperature were measured using ECH2O (Decagon, Inc. Decagon, USA) at depths of 10, 20, 40, 60, and 80 cm.All observation data were automatically collected and recorded by the EM50 series data collection system at 30-min intervals. Since 2012, six sets of ECH2O were installed in the shrub (3,400 m), forest (3,200, 2,850,and 2,750 m), shady slope grassland (2,700 m), and sunny slope grassland (2,700 m). In order to calibrate the ECH2O data, soil moisture was measured with the oven-drying method (108 °C) once every month during the growing season.

A series of permafrost apparatus (n=14) was installed to monitor soil freezing and thawing.First,soil was excavated to a depth of 250 cm.Then, the device was installed. To install, the outer tube (jacket pipe)with a length of 300 cm and a diameter of 5 cm is pushed to 250 cm soil depth.At the same time, a rubber freezing pipe with a length of 250 cm and a diameter of 1 cm is injected into water. Then, the gap between the jacket pipe and the soil is backfilled to prevent precipitation from entering.The thickness of frozen soil is read off a scale in the device that shows a frozen water column.

Figure 2 Micro-runoff plots and observation instrument

2.2.4 Baseflow separation

In this study, the recursive digital filter (RDF)was derived from the signal analysis and used for baseflow separation. The filter equation is as follows(Lyne and Hollick,1979):

whereqtis filtered quick streamflow at time stept,Qtis total streamflow, andαis the filter parameter.Then the filtered baseflow is obtained as

The filter parameterαin Equation(1)is determined from published data and results based on the baseflow index (BFI, volume of baseflow divided by the volume of total streamflow) from isotopic experiments in the Hulugou catchment, located in the same area (Chenet al., 2014; Chang, 2015; Sunet al.,2015a; Sunet al., 2015b; Qinet al., 2016). Thus, the filter parameterαwas taken as 0.98 considering the average relative error of BFI in this study. The time series was filtered three times: forward, backward,and forward again.

2.2.5 Statistical analysis

3 Results and discussions

3.1 Background conditions

Mean annual precipitation (2010-2017) at elevation of 2,700 m was 431.6 mm, with a minimum of 380.2 mm in 2011, and maximum of 471.5 mm in 2014 (Figure 3a). Most precipitation events (68.4%)did not exceed 5 mm, only 5.3% exceeded 17 mm,and ＞17 mm of precipitation events occurred less frequently, accounting for 4.2% and 1.3%, respectively (Figure 3b). Rainfall intensity ranged from 0.1 (mm/10 min) to 1.6 (mm/10 min), and more than 70% of events were drizzle-like with rainfall intensity of ＜0.3(mm/10 min).

Mean annual temperature ranged from 1.5 to 3.8°C(Figure 3c). The maximum monthly temperature was 15.6 °C (July), and the minimum was 12.7 °C (January). The annual total duration of sunshine ranged from 1,583.2 to 1,856.6 hours with an average of 1,756.5 hours(Figure 3c).

Mean daily soil temperature at different depths is presented in Figure 4a. The average temperature of surface soil(up to 10 cm depth)was 2.3°C;soil started to thaw at the end of April or early May when air temperature and sunshine hours increased. Start of thawing in the surface soil was 15-30 days ahead of that at soil depth of 40-cm and 60-cm.Surface soil temperature dropped below 0°C and soil started to freeze at the end of October, but the difference in timing of soil freeze between the surface and deep soil layers was only 15-20 days. Based on our previous studies(Yanget al., 2017), the dynamics of soil moisture in seasonally-frozen ground of a forested catchment in semiarid mountains can be separated into thaw com-pensation period (April to May), precipitation compensation period(June to September),water consumption period (October to November), and water stable period(November to April of the following year).

Figure 3 Characteristics of meteorological elements in Pailugou catchment(2010 to 2017):(a)annual precipitation,(b)percentage of precipitation events for different precipitation amounts,(c)annual precipitation and annual sunshine hours

Figure 4 Mean daily soil temperature and moisture at 5,10,20,40 and 60 cm depth in the study catchment,(a)soil temperature,(b)soil moisture

Moisture dynamics at different soil depths are presented in Figure 4b. At the beginning of the growing season,as snow melted,and seasonal permafrost started thawing, surface soil moisture (up to 10 cm in depth)increased significantly (from 0.15 to 0.20 m3/m3).During the precipitation compensation period, as temperature and soil temperature continued to increase(-10 to 0°C),plants entered a vigorous growth period and their consumption of soil moisture gradually increased. Therefore, surface soil moisture was sufficient to compensate for the increase in evapotranspiration and plant absorption and utilization, even though it was decreasing (0.22 to 0.13 m3/m3), while rainfall was increasing. At the same time, surface soil moisture was sensitive to rainfall and changed drastically(0.13 to 0.27 m3/m3),deep soil moisture remaining relatively stable with 0.17 m3/m3. At the stage of water consumption, the soil began to freeze, as air and soil temperatures decreased to 0 °C. Since the timing of freezing in surface soil was earlier than that in deep soil, surface soil moisture exhibited a slow downward trend in temperature (0.20 to 0.12 m3/m3), and the decline was greater than that in deep soil (0.20 to 0.18 m3/m3). The last period was water stable period,as temperature and soil temperature fell below 0 °C,plants entered dormancy and soil froze. Wet precipitation decreased, snowfall increased, and soil moisture remained stable with 0.12 m3/m3after soil moisture rapidly decreased during the initial period of freezing.

3.2 Process of soil freezing and thawing

Freezing and thawing of soil and water-heat exchange are the central link of hydrological processes in the permafrost of a forested catchment in a semiarid mountain region. The process of soil freezing and thawing is characterized by one-way freezing and two-way thawing. The progress of soil freezing and thawing is presented in Figure 5.It is divided into two stages: the freezing stage (section AB in the figure)and the thawing stage(section BCD in the figure).

Figure 5 The process of soil freezing and thawing in 2015 at Pailugou catchment

(1) The freezing stage (AB). Observation data show that the soil freezing time is on October 20. In the early winter season (October to November), the outside soil temperature dropped below 0 °C at night(Figure 4), and the daytime temperature rose to above 0°C.The soil surface underwent an intermittent freezing and thawing process characterized by night-freezing and daytime-thawing. With a continuous decrease in air temperature (3 to -3 °C) and an increase in surface accumulated negative temperature, the freezing depth increased rapidly and steadily (0 to 40 cm), and total moisture content in the frozen soil layer increased remarkably. The freeze-thaw process took 270 days, accounting for 73.60% of the year. The freezing period in the study area lasted from November to June of the following year.

(2) The thawing stage (BCD). Air temperature began to rise,and the number of daytime hours with temperatures above 0°C was also increasing since the middle of April; however, night temperatures were still below 0 °C. At that time, surface soil began to undergo an intermittent freezing and thawing process, again characterized by night-freezing and daytime-thawing.The freezing front continued to move deeper until it reached the maximum freezing depth (160 cm). With rising air temperature, the depth of thawing in the soil surface increased during the day, while the daily minimum temperatures were still below 0°C;this gave rise to a phenomenon of a double-frozen layer. During this period, the top-level frozen layer received solar radiation, the lower frozen layer was influenced by the effect of geothermal heat,and this process lead to a bidirectional thawing phenomenon. The top-level frozen layer was completely ablated until the lowest temperature was stable through 0 °C, and thawing of the lower frozen layer continued until it was also completely ablated. Frozen soil ablation showed an increase in volatility, with an average rate of 1.52 cm/d. Starting from May, due to the daytime temperature is higher than 0 °C, the surface frozen soil undergoes a thawing-freezing-thawing process, and the bottom frozen soil is still frozen. Until July, the deep soil tempera-ture is above 0 °C, and the bottom frozen soil begins to melt. The melting rate of the surface layer is greater than the deep layer at the beginning.

3.3 Processes of runoff generation in the catchment

The annual average amount of runoff from 2010 to 2017 was 88.5 mm; this accounted for 25.6% of rainfall. Runoff was concentrated in May-October,and accounted for 89.2% of the total annual amount of runoff (Table 1). Runoff exhibited three peak periods—the thawing period is in June 3 to June 20 with 1.01 to 1.18 mm/d, the maximum runoff during the rainfall period is 6.33 mm/d, which occurred on July 4, and the peak flow during the freezing period starting in September is 1.12 mm/d, respectively. A short dry season formed after the thawing period,soil moisture decreased from 0.22 to 0.16 m3/m3,and lasted until the beginning of the rainfall period (Figure 5 and Figure 6).After the rainfall period,the runoff gradually decreased (0.57 to 0.08 mm/d) until the river froze.The hydrological cycle in this small watershed in the Qilian Mountains was short (April-October), and the river was frozen from November to April(Figure 6).

Table 1 Baseflow characteristics at the catchment

Figure 6 Runoff generation processes in 2015 at Pailugou catchment

Runoff in the Pailugou catchment was under the comprehensive control of hydrometeorological parameters.Temperature mainly affected snowmelt and runoff due to the seasonal soil thawing in the thawing and freezing periods; precipitation mainly affected rainfall runoff in the rainfall period. These results are similar to those of previous studies in arid and cold regions, including that of Bayardet al. (2005) in the southern Swiss Alps, Wanget al. (2009) in the Fenghuoshan watershed, Semenovaet al. (2013) in the upper reaches of the Kolyma River, and Liet al. (2016)in the Shiyang River.

So the young man stood carefully on one side, and by-and-bye he heard a great rushing in the water; and a horrible monster came up to the surface and looked out for the rock where the king s daughter was chained, for it was getting late and he was hungry

3.3.1 Runoff generation processes during the thawing period

Runoff during the thawing period was mainly composed of a mix of precipitation, snow melt, and seasonal thawing of soil. The progress of the runoff process lasted a long time, and included one peak per day, the flood peak was long and smooth with an average of 0.085 mm/d.After freezing from November to April of the following year, soil moisture in the basin was at a low level 0.13 m3/m3. Starting in May,the average temperature reaches 7.3 °C, and the average sunshine hours are 16.2 hours, which causes the frozen water in the soil to melt and intermittent runoff was continuously generated in the soil surface and rivers.Temperature of the shallow soil also gradually increased with an 2-5 °C increase in air temperature.The seasonally-frozen soil began to melt rapidly, and the soil surface underwent an intermittent freezing and thawing process characterized by night-freezing and daytime-thawing. The infiltration of snowmelt and the thawing of seasonally-frozen soil caused the soil moisture in the shallow soil moisture to rise sharply from 0.11 to 0.20 m3/m3. However, surface runoff could not, or only rarely infiltrated downward due to the shallow depth of thawing; the lower layer of frozen soil did not thaw resulting in an impermeable or weakly-permeable layer, while shallow soil moisture was close to saturation. When the rate of infiltration is less than the intensity of runoff,a local super-osmotic runoff process is formed, and most of the runoff flows into the river channel in the form of slope runoff and soil flow;this then becomes the main form of runoff generation process during the thawing period.

3.3.2 Runoff generation processes during the rainfall period

Summer runoff is mainly due to rainfall accumulation. Rainfall during the summer period replenished the previous-season soil water deficit, and satisfied plant need for soil moisture, while excess water slowly converged to form soil flow. At the same time,when the intensity of the storm exceeded infiltration strength due to the influence of the southwest monsoon in summer,the super-osmotic flow formed locally. The runoff processes during the rainfall period were usually dominated by torrential rains with short duration and heavy rainfall intensity, and the runoff accounts for 60%-80%of the annual runoff(Table 2).Starting in July, runoff processes were highly consistent with rainfall because rainfall increased, and rainfall intervals decreased.

Table 2 Precipitation and runoff for the freezing,thawing and rainfall periods

During the rainfall period, runoff events were highly responsive to rainfall, as could be expected. In the initial stage, precipitation was mainly used to replenish shallow soil moisture, while deep soil moisture remained stable enough to generate runoff (Wang Set al.,2013;Yanget al.,2017).Soil evaporation and plant evapotranspiration gradually increase due to air and soil temperatures greater than 5 °C, rainfall should not only supplement soil water deficit in the early stage, but also confluence to the runoff by the form of interflow (Rothfusset al., 2010). This was the reason why early runoff was very small(0.23 mm)even though rainfall was very large(6.24 mm).

3.3.3 Runoff generation processes during the freezing period

In the study area, the temperature began to decrease significantly from an average of 10 °C to 0 °C in early September. By the end of September or early October, the surface accumulated negative temperature was less than 0 °C, surface soil began to freeze,and soil moisture was transferred to the frozen layer,forming an impermeable stratum. At higher elevations, the bottom of the frozen soil also began to freeze (Figure 5). The underground runoff and the inflow runoff were gradually reduced due to soil freezing. At the same time, precipitation in the basin appeared in a solid form due to low temperatures. Even with occasional liquid precipitation in the initial stage, surface runoff could not form because soil moisture was less than 0.13 m3/m3.During this period,the runoff components were mainly composed of groundwater and soil inflow runoff, and intermittent breaks or even freezes often occurred.

3.3.4 Baseflow processes at the catchment

Runoff, runoff coefficient, and baseflow exhibited a slow growth trend from 58.80 to 77.30 mm, 0.14 to 0.18, and 0.4986 to 0.5365, respectively, during the study period, indicating that the increase in baseflow provided the greatest contribution to the increase in total runoff (Table 1). The average baseflow in the study area was 36.44 mm, and the BFI was 0.5277 with a range of 0.4986-0.5415. The increasing trend in baseflow was mainly due to the increase in melting of the seasonally-frozen soil in the watershed, which increased soil water storage capacity, forming more interflow and enhancing the regulation of aeration zone and groundwater recharge; further, the increase in melting caused infiltration during the thawing season,which increased baseflow.

3.4 Influencing factors in runoff generation process

We selected average soil temperature for 0-50 cm depth, average soil moisture for 0-50 cm depth, air temperature, and precipitation as statistical variables for factor analysis to identify the key factors in runoff generation process in permafrost in a forested catchment. We found that, together, air temperature,the 0-50 cm average soil temperature, and depth of frozen soil accounted for 59.60% of the runoff during the thawing period, while precipitation accounted for 21.9% (Table 3 and Figure 7). Thawing-period runoff and soil temperature were highly correlated, with the correlation coefficient of ＞0.6 (P＜0.05). With the increase in the duration of sunshine in spring, soil temperature increased and seasonally-frozen soil melted.When frozen soil depth was less than 120 cm,the runoff was smaller, but when the frozen soil depth was more than 120 cm, the daily runoff increased rapidly.In the rainfall period,runoff was also affected by temperature and soil moisture, accounting for up to 33.6% and 34.1%, respectively, while the influence of precipitation increased to 18.1%.

Table 3 Component rotation load matrix and its contribution rate

Temperature controlled the melting of frozen soil and snow to influence runoff process in the thawing and freezing periods. In catchments with seasonally-frozen ground, almost all baseflow and most of the river runoff in the thawing period were attributed to the melting of frozen soil and snowmelt water, as were increases in discharge of several rivers with warming (Wanget al., 2009; Boucher and Carey, 2010; Lapp, 2015). At the beginning of the thawing period, the surface layer was highly porous and exhibited high infiltration rates, even when it was saturated by ice. Some of the infiltrated water became frozen in the cold peaty horizon, decreasing its hydraulic conductivity and creating the conditions for formation of surface flow in the thawing period (Semenovaet al., 2013). Surface soil temperature dynamics-controlled spring runoff processes by changing the source area of the saturation-excess runoff.Thus, initial thawing, including thawing of the surface-active soil layer and melting of snow, resulted in spring flood flow by generating saturation-excess runoff that enhanced the runoff coefficient (Wanget al., 2009; Wanget al., 2017). The effect of temperature on runoff during the freezing period was mainly due to the decrease in infiltration capacity and the increase in frozen water content (Kane and Stein,2010; Kalyuzhnyi and Lavrov, 2017). In regions with seasonally-frozen ground, the bottom of the frozen soil also begins to freeze. The layer formed by freezing resulted in an impermeable soil layer, and the soil moisture in a deep layer gradually tended to a stable state(Evans and Ge,2017;Yanget al.,2017).A lot of water in soil was frozen, and flow was due to the from-the-top and from-the-bottom two-way melting characteristics of seasonally-frozen soil (Penget al.,2013;Qinet al.,2016).Precipitation in our study basin appeared in solid form due to low temperatures.Even if liquid precipitation occurred in the initial stage of runoff formation, surface runoff could not form due to the small amount of available water. As a result, there was very little direct runoff and more baseflow during this period.

Figure 7 Relationships between runoff and soil temperature,air temperature,and frozen soil freezing and thawing in 2016

Climate change, including increasing temperatures and precipitation, will have an irreversible impact on the hydrological regime of mountainous basins where seasonally-frozen ground is widely distributed. This necessitates increasing of research efforts in the study of hydrological dynamics in mountainous basins.

4 Conclusions

Climate change is the direct and important factor that leads to the change of runoff processes in a seasonally-frozen catchment,and the change in precipitation and temperature regimes will directly affect runoff generation and aggregation process. In this study,we selected a typical semi-arid mountainous catch-ment to quantify the response of runoff to hydrological factors such as temperature,precipitation,and seasonally-frozen ground. The results provide basis for accurate simulation of the mountain runoff in cold regions with a hydrological model. Based on the results of this study,we conclude the following.

(1) This cold region basin was dominated by small precipitation events. From May to September,the average temperature in the Qilian Mountains rose to more than 3°C,and duration of sunshine was more than 140 hours, mainly by rainfall. From October to April of the following year, average temperature dropped below 3 °C, and duration of sunshine was less than 130 hours, while the form of precipitation was mainly snowfall.

(2) Annual total runoff from 2010 to 2017 was 88.5 mm, accounting for 25.6% of rainfall; baseflow was 36.44 mm, and the BFI was 0.5277. Runoff was mainly concentrated in May-October, accounting for 89.2% of the annual total runoff. Peak runoff occurred in June,August, and in September,i.e., spring and summer floods. The mixed flood process produced by rainfall, melting snow, and melting of the seasonally-frozen ground formed snowmelt runoff in spring season, and runoff and air temperatures were consistently significantly correlated. Runoff during the summer flood season was mainly due to precipitation accumulation. Precipitation in the summer flood period was needed for replenishment of soil water deficit to meet the soil water demands of vegetation. At the same time, excess precipitation, along with soil flow,formed the river channel confluence.

(3)Air temperature,0-50 cm average soil temperature, and frozen soil depth together accounted for 59.60% of runoff during the thawing period, while precipitation accounted for 21.9%.Runoff in the thawing period and soil temperature were highly correlated, with a correlation coefficient of ＞0.6 (P＜0.05).Runoff in the rainfall period was also affected by temperature and soil moisture, but they contributed 33.6%and 34.1%,respectively,while precipitation contributed 18.1%.

Acknowledgments:

We thank the editor and anonymous referees for their valuable comments which helped to improve our manuscript. Furthermore, we thank Dr. Kathryn Piatek for editorial suggestions and language assistance. This work was supported by the National Natural Science Foundation of China (Nos. 41901044,41621001 and 41701296), the "CAS Light of West China" Program (29Y829861), and Foundation for Excellent Youth Scholars of Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences(FEYS2019019).