Dynamics of green and blue water flows and their controlling factors in Heihe River basin of northwestern China

Abstract. Climate variation will affect hydrological cycle, as well as the availability of water resources. In spite of large progresses have been made in the dynamics of hydrological cycle variables, the dynamics and drivers of blue water flow, green water flow and total flow (three flows), as well as the proportion of green water (GWC), in the past and future at county scale, were scarcely investigated. In this study, taking the Heihe River basin in China as an example, we investigated the dynamics of green and blue water flows and their controlling factors during 1980–2009 using five statistical approaches and the Soil and Water Assessment Tool (SWAT). We found that there were large variations in the dynamics of green and blue water flows during 1980–2009 at the county scale. Three flows in all counties showed an increasing trend except Jiayuguan and Jianta county. The GWC showed a downward trend in the Qilian, Suzhou, Shandan, Linze and Gaotai counties, but an upward trend in the Mingle, Sunan, Jinta, Jiayuguan, Ganzhou and Ejilaqi counties. In all the counties, the three flows and GWC had strong persistent trends in the future, which are mainly ascribed to rainfall variation. In the Qilian and Shandan counties, rainfall was the major controlling factor for the three flows and GWC. Rainfall controlled the green water and total flows in the Mingle, Linze and Gaotai counties; green water flow and GWC in the Suzhou county; green water flow, total flow and GWC in the Jinta and Ejilaqi counties. Our results also showed that the "Heihe River basin allocation project" had significant influences on the abrupt changes of the flows above-mentioned. Our results illustrate the status of the water resources at county scale, providing a reference for water resources management of inland river basins.


Introduction
Much attention has been paid to the influence of climate variat ion on water resource all over the world (Jeu land and Whittington, 2014;Zhang et al., 2015).In the context of global warming and increasing extreme weather frequency and 25 intensify (Qiao et al., 2014;Zuo et al., 2015), studies have shown global climate change has led to reduction of water resource (Kundzewicz et al., 2008;Ravazzani et al., 2014), and exacerbated the shortage of water in the semi-arid regions.
Water scarcity can endanger the food safety and sustainable development of economy, as well as the health of the ecosystem (Cheng et al., 2007;Piao et al., 2010).
Falken mark (1995) first introduced the notions of blue water and green water and other researches afterwards progress it Hydrol.Earth Syst.Sci. Discuss., doi:10.5194/hess-2016-241, 2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 13 June 2016 c Author(s) 2016.CC-BY 3.0 License.and the proportion of the total accounted for by green water during 1980-2009.The aims are: 1) to analyze the dynamics of above-mentioned variables on county scale 2) to identify their controlling factors in each county.

Study area
The Heihe River Basin (HRB) lies within 38-42º N and 98-101º W in Northwest Ch ina and covers an area of 14.31 ×10 5 km 2 5 (Wu et al., 2013).Mean elevation is surpassed 1200 m and the range was 879-5573 m.The HRB co mprises eleven counties which are the Qilian, Sunan, Gnazhou, Shandan, Su zhou, Mingle, Linze, Gaotai, Jinta, Jiayuguan, Ejilaq i.The upper HRB includes Qilian and Sunan county and the midd le HRB includes Gnazhou, Shandan, Suzhou, Mingle, Lin ze, Gaotai county.The down HRB was consist of Jinta and Ejilaqi county.The basin mean rain fall is 193.4 mm a -1 fro m 1980 to 2009.Mean annual rainfall declined fro m the upper basin to down basin (Yin et al., 2015).10

Methods and data
The green water flow refers to actual evapotranspiration (AE), and the blue water flow co mprises surface runoff, lateral flows, and groundwater recharge (Schuol et al., 2008).Green water coefficient (GW C) -the proportion of green water in total flows (b lue and green flo wers, TW) was used to measure the importance of blue and green flo ws (Liu et al., 2009a).
The amount of flows was simulated by the Soil and Water Assessment To ol (SWAT), which has successfully been 15 applied to many countries and regions (Baker and Miller, 2013;Li et al., 2010;Schuol et al., 2008;Zende and Nagarajan, 2015).The Nash-Sutcliffe efficiency coefficient (NS), determination coefficient (R 2 ) and root mean square error (RM SE) -observations standard deviation ratio (RSR) was used to estimate the model simulat ion.Generally, model simulat ion is perfect when RSR range is between 0.00 and 0.50, and NS range is between 0.75 and 1.00 .The simulat ion effect is good when the range of RSR is fro m 0.50 to 0.60 and NS is fro m 0.65 to 0.75.The simulat ion effect is satisfactory when RSR 20 belongs to 0.60-0.70 and NS belongs to 0.50-0.65 (Krause et al., 2005;Moriasi et al., 2007).One this base, we fu rther assessed the accuracy of simulated green water by observed AE in the year of 2007. In

25
Soil data were fro m Harmonized World Soil Database version1.1 via the FAO website.

Dynamic and controlling factor analysis
The Mann-Kendall (M K) trend test was selected to describe trends of flows, and the sequential Mann-Kendall (SM K) (Ahmad et al., 2015a;Ah mad et al., 2015b;da Silva et al., 2015) was chosen to identify abrupt variations.The MK test is extensively applied to the trend analysis of long -time series of hydro-meteoro logical data because of its non-parametric Hydrol.Earth Syst.Sci. Discuss., doi:10.5194/hess-2016-241, 2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 13 June 2016 c Author(s) 2016.CC-BY 3.0 License.
assumption, simpleness and stability (Ah mad et al., 2015b;Bae et al., 2008;Chen et al., 2014;da Silva et al., 2015).We followed other literature to calculate the MK and SMK (Ah mad et al., 2015b;da Silva et al., 2015).Theil-Sen method (TS) method was applied to character the change amplitude, wh ich means the mean variat ion for each year.If the trend was displayed by linear, TS expresses the mean change rate per year.Determination coefficient of lin ear regression (DC) was used to gauge the contributions of the independent variables to the dependent variables and find the controlling factor 5 (Nagelkerke, 1991).In the unary linear regression, the independent variable can be identified as the controlling factor if DC >0.50 (Bin et al., 2003;Nagelkerke, 1991).Figure .2 listed the flowchart of our study.

Model cali bration and uncertainty analysis
The agreement between simulated and observed monthly discharge values in the periods of calibration and validation was 10 excellent (Figure .3).All NS and R 2 values were above 0.5, and absolute SRS values were below 25 (Table 1).These suggest model perform fairly well, in spite of small differences sometimes (Figure .3).The results simu lated by SWAT models can be applied to evaluate green water and blue water flows.
We used the observed annual AE in 2007 (Yang, 2009) to further estimate the green water flow due to the limitation of observed data.The range of relative errors is fro m 3.51% to 5.78% (Table 2)., 2003, , 2003, , 2001, , 2001, and 1985, respectively (Figure. 5), respectively (Figure. 5).It The GW C generally decreased fro m 1980 to 2009 at Qilian, Su zhou, Shandan, Lin ze and Gaotai counties, but insignificantly.Meanwhile, the GW C increased at Mingle, Sunan, Jinta, Jiayuguan and Ganzhou counties (Table 3).The dynamics of GW C generally are same as those of green water flo w, although not completely consistent.The reason for this is that the GWC depend on green water flow and total flo ws, not only green water flo w. 15

Dynamics of total flow
The trends in the blue water, green water, and total flows showed a considerable spatial variat ion among the counties.The MK test showed that total flows generally increased in nine counties, including the Su zhou, Shandan, Qilian, M ingle, Lin ze, Sunan, Gaotai and Ejilaqi (Tab le 4).The increase was significant in Mingle, Gaotai and Gan zhou fro m 1980 to 2009 (p < 0.05; Table 4).The largest increase was in the Ejilaq i county, where total flo w increased by 6.17 ×10 8 m 3 fro m 1980 to 2009,

20
equivalent to a rate of increase of 2.08×10 8 m 3 per decade.This was caused by an increase in green water flow (Figure. 5).
However, the total flows at Jinta, and Jiayuguan decreased insignificantly in the past three decades (Table 4).Th is is mainly attributed to decrease in blue water flo w (Figure.

Controlling factors for dynamics of bl ue water, green water and total flows
For the blue water flo w, the trends and influence factors differ fo r different counties of the river basin among eleven count ies.
One reason for this is that rainfall generally increased and temperature significantly increased during this period (Tab le 5 and Figure . A1).We detected that rainfall was the controlling factor at Qilian, Sunan and Shandan county (Table 6, DC=0.90 at 20 Qilian and DC=0.81 at Shandan), where their general trends and the fluctuations were closely related to their variat ions of rainfall (Figure . A1). Rain fall contributed more to blue water flow than the temperature in Mingle, Lin ze, Jiayuguan and Ganzhou.By contrast, the contribution of temperature was mo re than that of rain fall in Su zhou, Jinta , Gaotai, Ejinaqi.
However, neither the rainfall nor the temperature was the main controlling factor in these nine counties (Table 5).Hu man activities may be the main ly driver for land-use change.

25
For the green water, obviously, our analysis showed that the controlling factor was the rain fall with all DC>0.84 during the period of 1980-2009 (Table 6).This controlling action o f rainfall is ext remely obvious in Linze, Jinta, Gaotai and Ejilaqi counties with DC>0.90.

Discussion
The The channeling effect at this county resulted in different regional hydrological cycles and rainfall sources (Lan et al., 2001).Hence, at Shandan, rainfall was influenced by different atmospheric circulations and topography (Lan et al., 2001).Rainfall there was mainly influenced by The land use change maybe the driver of blue water in M ingle, Linze, Jiayuguan, Ganzhou, Su zhou, Jinta, Sunan, Gaotai, Ejinaqi counties.Surface roughness, albedo and other properties that affected the exchange of water and energy between the earth surface and atmosphere could be altered by conversions of land use, resulting in variability of surface energy and net 15 radiation (Kueppers and Snyder, 2012), and further influencing hydrological processes.Land use change, especially the expansion of agricultural land, had influenced on hydrological processes and water balance of Heihe River basin (Deng et al., 2015;Fu et al., 2014;Liu et al., 2010;Nian et al., 2014).The expansion of farmland had a significant influence on the shallow aquifer water sys tem through intensive irrigation.W ith the expansion of farmland, ground water recharge increased due to enhanced infiltration in these counties (Wang et al., 2014).Deforestation also reduced soil water retention capacity 20 and raise surface runoff (Zhou et al., 2002).The green water depended mainly on water availab ility and air temperature in the northwest of China (Han et al., 2012;Yang and Yang, 2012).Since 1850, the temperature on the Earth's surface has been increasing continuously ; climate warming is an obvious phenomenon (Stocker, 2014).In general, because of climate warming, the air near the earth's surface would become d rier, which further leads to an increase in AE (g reen water) (Cong et al., 2009).Nevertheless, variation of AE is of significant importance for hydrologic water cycle (Zhang et al., 2012).In the arid or semi-arid, the green water was limited by water availab ility (Sun and Wu, 2001).Green water was not strongly driven by temperature, but the rainfall, although there are significant changes in temperature in the study period.An increase in rainfall will increase surface runoff and soil moisture, leading to the increase in b lue water flow (A E).In the Ganzhou county, both rainfall and temperature have a positive influence, leading to increase in green water.The contribution of rainfall was much larger than that of temperature.

5
However, in the Ganzhou county, the contributions of rainfall and temperature were lo w, and both were not the main driver.
This suggests that human activities should be the controlling factor of green water in the Gan zhou county, which offset the positive influence induced by the increase of rainfall and temperature.Ganzhou county was the commodity grain base for the whole Gansu province and agriculture land depended on irrigation quite wild (Chen et al., 2006).Fo r Gan zhou county, the rapid expansion of irrigation in the past three decades has significantly altered the energy budget at the land surface (W iss er et al., 2009).A E will change when irrigation related change in soil mo isture, deep filt ration and underground recharge (Wang et al., 2014).Our results indicate that green water flow will keep trend in the past, all the counties will increase expect Jiayuguan county where the future trend will decrease (Figure .5o and 6h).
When green and blue water flows are summed, the trend in the total flow appears to reflect a co mbined influence of climate and human activities in the eleven counties in the HRB fro m 1980 to 2009 (Tab le 4).Because the green water flow 15 was much more than blue water flow in the HRB, the general trends in total flo w were similar to green water flow.It increased in Qilian, Gan zhou, Sunan, Suzhou, Lin ze, Gaotai, Shandan, Mingle and Ejilaqi counties, wh ereas decreased in Jinta and Jiayuguan counties.Meanwhile, our analysis showed the controlling factor of total flow at each county was different fro m that for green water flow: the total flo w in the most counties was mainly influenced by rainfall (DC＞0.5; Table 6), however, at Jinta and Jiayuguan county, the human activit ies were the major controlling factor.With increases in

20
rainfall and temperature, the increase in blue water and green water flows lead to the increase in the total flow at Qilian, Ganzhou, Sunan, Su zhou, Lin ze, Gaotai, Shandan, Mingle and Ejilaqi counties.But the total flow decreased due to decrease in blue water flo w and green water flo w, which was induced by increases in rainfall and temperature at Jiayuguan county.
Our results also showed that the decrease in b lue water flo w and green water flow at Jinta county would increase blue water more than green water, leading to a decrease in total flow.Hence, it wou ld be logical to predict that there is a persistent trend

25
for total flow in the eleven counties, which is consistent with the trend in green water d riven by rainfall expect Jinta and Jiayuguan counties (H＞0.5,Figure. 6).
There are variations in the dynamic and controlling factor of the GWC in different counties in the HRB (Table 5 and 6).
For all of the counties, the temperature was not the controlling factor and the contribution of rainfall to the GWC is relatively higher.The GW C increase induced by green water rapidly rise with the increase of rainfall and temperature in Lin ze, Suzhou, At the county level, different rainfall patterns also can lead t o different hydrological processes.For examp le, the rainfall decreased in Shandan county significantly (P<0.05),but increased in Mingle county.However, the difference caused the blue water flow to increase in both counties (significantly for Shandan cou nty, and insignificantly for M ingle county).In the future, the three flo ws will have an increasing trend due to the projected increase of rainfall.

15
The dynamic of the climate can largely exp lain the variations and abrupt changes of three flows in different periods (Table A1and Figure .A 1). Ho wever, there are so me abrupt changes in the 1990s and 2000s, which were close to 1992, 1997 and 2000 when rainfall and temperature had no abrupt changes in the corresponding periods.These abrupt changes should be attributed to the China's water transport project "Heihe River basin allocation policy" init iated in 1992,1997 and 2000 to ensure the enough water availability and in case of drying up of Dongjuyanhai Lake.The water transport project makes an 20 influence on green water flow and blue flo wn mainly though changing the underground water level and surface water.
"Heihe River basin allocation policy in 2000 had decreased ground water level by 1.3 -2.7m in the middle HRB (Haiyang et al., 2007) and raised it by 0.4-0.6m in the down HRB (Zhi et al., 2007).Yinchun et al (2014) indicated the water transport project deceased the number of water withdrawal gates by 26%, reduced the blue water withdrawal by 1.05×10 9 m 3 and increased the underground water exp loitation by 1.64×10 9 m 3 in the whole Heihe River basin (Yinchun et al., 2014).In the 25 middle-stream, "Heihe River basin allocation project" in 2000 induced the decrease the underground water level in the irrigation region caused by the reduction of exchange of surface and underground water, and expansion of the underground water exp loitation (Zhi et al., 2008).
This study has some limitations.First, more data need to be obtained and comp lemented, such as the hydrological data in the lower reaches of the basin and the meterological data for the whole basin.Second, the impacts of human activities need this study, land-use data come fro m the Ch inese Academy of Sciences (CAS) Environ mental Data Center.Monthly average discharge was obtained fro m the HRB Water Resources Administration.Daily climate data were fro m 16 weather stations administrated by the China Bureau of Meteorology.The DEM o f 30m resolution via the Ch ina Scientific Date Cloud.

5 5r)
) and Ejilaq i (Figure.5v) counties changed abruptly in 2001 changed abruptly in 1996, 2001 and 2005 at Jita county (Figure.5t), and in 1982 and 2005 at Jiayuguan county.It changed abruptly in 1983, 1986 and 1993 at Su zhou county (Figure.5n).However, there were always abrupt changes followed 2000, around 2000-2003.Meanwh ile, in the study period blue water flow has different trends.For example, at the Linze county (Figure.5l), b lue water flow increased fro m 1980 to 1982, decreased from 1982 to 1985, and then increased fro m 1985 to 10 1995, decreased fro m 1995 to 2001, and decreased from 2001 to 2004, and finally increased fro m 2004 to 2009.
4).Blue water flo w decreased at Jinta and Jiayuguan by 0.07 ×10 8 m 3 per decade and 0.33×10 8 m 3 per decade, respectively (Figure.4).3.5 Predicted trends in bl ue water, green water and total flows in the future 25 In the upstream including the Qilian county (Figure.6a), the hydrological variab les have insistent trends in the future, with H > 0.50 (Figure.6).The strength of the persistence of trends in descending order is: green water flow > total flow > GW C >blue water flo w.A ll the variables have strong persistent trends, will keep increasing in the upstream basin.In midstream basin, including Sunan (Figure.6b), Gan zhou (Figure.6c), M ingle (Figure.6d), Shandan (Figure.6e), Linze (Figure.6f), Su zhou (Figure.6g), Jiayuguan (Figure.6h) and Gaotai (Figure.6l) counties, all hydrological variables have 30 Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-241,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 13 June 2016 c Author(s) 2016.CC-BY 3.0 License.persistent future trends, with H > 0.5 (Figure.6).This indicates that the blue water flo w, green water flow, total flo w and GW C will keep increasing.At Jiayuguan, Sunan and Jinta county, the Hurst index of blue water flo wn were close to the value of 1.00, suggesting that persistent trends are obv ious, and the blue water flow in both counties should continue to decrease in future.However, the hydrological variab les at Shandan county do not have an obvious persistent trend with the Hurst index of g reen water flo w and total flow closing to 0.50, which represents quite independent trends.5 In the downstream basin, including Jinta and Ejilaqi counties, the trend for blue water flow has an obviously persistent change in the future, with H=0.99 at Jinta county (Figure.6j) and H=0.98 at Ejilaqi county (Figure.6k).Th is means that the blue water flow may quickly decrease in the future.The Hurst indexes for the Green water flow, total flo w and GW C also have persistent future trends, with H > 0.5.Th is imp lies that those hydrological variables will increase in the downstream basin.The most persistent change trends display in the downstream basin.10 All of Hurst index values of blue water flow were surpassed 0.5, wh ich indicates that the trends in the past were likely to continue (Figure.6) The counties including Qilian, Sunan, Gan zhou, Shandan, Mingle, Linze, Suzhou and Ejilaqi have persistent trends and will keep increasing in the future, whereas blue water flow at Jiayuguan and Jinta counties may decrease in the future inconsistent with the past trend.For eleven counties in the HRB, the blue water flow, green water flow, total flows, and GW C generally have continuous 15 future trends as in the past, with H > 0.5 (Figure.6), especially in the upper and middle HRB.
blue water fo r counties in the HRB generally increased fro m 1980 to 2009, although the blue water flow at Jiyuguan Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-241,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 13 June 2016 c Author(s) 2016.CC-BY 3.0 License.county and Jinta county decreased insignificantly (Figure.4).Qilian in the upstream basin was the runoff-product area for the whole basin, where the rainfall was generally high.An increase in rainfall led to a fast increase in runoff (Li et al., 2011).The barrier effect o f the Qilian Mountains in the upstream basin had caused different regional hydrological cycles and different rainfall sources (Jia et al., 2008).Rainfall in upstream was influenced by different atmospheric circulat ions that resulted from local topography, i.e. the West Pacific Subtropical High and by the Indian Ocean Southwest Monsoon (Jia et 5 al., 2008).The West Pacific Subtropical High and the Indian Ocean Southwest Monsoon began to change in the 1980s, leading to an abrupt change in rainfall in 1980s (Jia et al., 2008; Lan et al., 2001).In Shandan county, blue water flow increased significantly fro m 1980 to 2009.But they also fluctuated in different periods.

10 the
Western Pacific Subtropical High and by the Eurasian M iddle-High Latitude Circulat ion, which led to fluctuations of rainfall at Shandan county.
Climate will have a large in fluence on future blue water flo w.Climate models predicted a weak increase in annual rainfall and significant increase in temperature(Dahe et al., 2002).Climate change is likely to cause annual runoff variation(Zh i et   al., 2009)  and increase surface runoff by more than 10% during2010 -2050(Dahe et al., 2002)).An increase in rainfall will rarely produce large amounts of runoff because of the flatter topography combined with the high potential evapotranspiration 25 and low soil moisture content.
Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-241,2016   Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 13 June 2016 c Author(s) 2016.CC-BY 3.0 License.Shandan, Ejilaqi county.The increase of rain fall and temperature in the Jiayuguan and Jinta county would increase runoff (blue water) more than evapotranspiration (green water), leading to a lower GWC.For Qilian, Su zhou, Shandan, Jinta, Jianyuan and Ejilaq i counties, the rainfall determined the trends and variability of the GWC.All the Hurst index values surpassed 0.50, indicating the GW C depend on the past trend and will keep the trend in the future.Hence, the rainfall exerted the most influence on three flows in the HRB.The d ifferent dynamic of three flows formed 5 caused by different generation mechanisms of rainfall on the county scale.Therefore, the abrupt changes of temperature and rainfall could have influenced both the blue water and tot al flows.Moreover, the flows are not only influenced by atmospheric circu lation, but also by local human activities, such as the blue water flow o f Mingle, Lin ze, Jiayuguan and Ganzhou counties.The quantitative impacts of human activities in each county will also need further research.The similar abrupt changes in the water flows, rainfall, and temperature reflect the close relationship between these three parameters in 10 the eleven counties.

Figure 1 :
Figure 1: A map depicting the location of the Heihe River Basin in China (left plate) and the expanded study area (right plate) depicting the elevation, river and data stations used in the study.

Figure 2 :
Figure 2: Flowchart of this study.M ann-Kendall test is represented by M K; SM K is Sequential M ann-Kendall test represented by SM K; TS represents Theil-Sen method.DEM is a digital elevation model; GWC is the proportion of green water in the total flows (blue and 5

Figure 4 :
Figure 4: Dynamics of blue water flow at eleven counties of the Heihe River basin, including Qilian (a and b), Sunan (c and d), Ganzhou(e and f), M ingle (g and h), Shandan (i and j), Linze (k and l), Suzhou (m and n), Jiayuguan (o and p), Gaotai (q and r), Jinta (s and t) and Ejilaqi (u and v).Note: TS stands for the Theil-Sen test; *** indicates the significant level is at p ＜0. 01; + indicates the significant level is

Table 1 .
SWAT model performance during the calibration and validation period 5

Table 3 .
The results of the M ann-Kendall test (MK), sequential M ann-Kendall test (SMK), and Theil-Sen estimator test (TS) for eleven counties of GWC in the Heihe River basin.TS stands for the Theil-Sen test and the value of presents the change rate; ↑ indicates an increasing trend, whereas ↓ indicates a decreasing trend; * indicates the significant level at p＜0. 05; NS represents the test is not significant.

Table 4 .
Dynamics of total water flow in the eleven counties in Heihe River basin.Kendall test; SMK represents sequential M ann-Kendall test; TS stands for the Theil-Sen test and its values represent the change rate; ↑ declares an increase trend and ↓declare an decrease trend; *signals a significance at p ＜0. 05; NS represents the test is not significant.

Table 6 .
Coefficients of determination (DC) of rainfall for blue water flow, green water flow, total water flow in eleven counties in the Heihe River basin in the past three decades.Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-241,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 13 June 2016 c Author(s) 2016.CC-BY 3.0 License.