Where does the irrigated water in the Tarim Basin go? A hydrological analysis of water budgets and atmospheric transport

Irrigated agriculture plays a crucial role in the local economic and social development of the Tarim Basin (TB), but its sustainability is threatened by water scarcity due to the arid environment. In this study, we investigate the impact of irrigation on the atmospheric hydrological cycle in the region using the weather research and forecast model. We conduct simulations for a three month period under two scenarios: present-day and future warming. Our results show that, in the present-day scenario, 90.5% of irrigated water is transported via atmospheric hydrological processes, with precipitation and water vapor transport being the dominant components. However, in the future warming scenario, more atmospheric water (45.2%) will leave the area due to weakened wind regimes, resulting in significant water loss. Furthermore, our analysis using the HYbrid Single Particle Lagrangian Integrated Trajectory model indicates that irrigation contributes to extreme rainfall events, and the southwestern TB is a primary destination for irrigated water. Our findings highlight the urgency of addressing the sustainability of irrigated agriculture and local water resources in the face of impending global warming.


Introduction
The Tarim Basin (TB), occupying 44.2% of total agricultural land in northwest China, not only ensures local food security, but also contributes 90% of the country's cotton supply (Xinjiang Bureau of Statistics 2021).Due to the arid nature, more than 95% of the agriculture in TB is sustained by irrigation, which accounts for 92% of the water supply (Xinjiang Bureau of Statistics 2021).Since 2000, the total water resource in TB has been decreasing due to the increase of water consumption; however, the water supply has been increasing (Kadiresan and Khanal 2018, Yang et al 2021, Liu et al 2021a, 2021c, Monforte and Ragusa 2022), which has raised serious concerns about the sustainability of irrigation-based agriculture in TB (Wei et al 2013, Wu et al 2015, Zhang et al 2016, Bai et al 2018, Rosa et al 2019).
Irrigation is a key measure to ensure sustainable agriculture-20% of global farmland operated with irrigation produces 40% of global food (Schultz et al 2005) but consumes 70% of freshwater supply (Molden 2007, Grafton et al 2017, Liu et al 2017, Zohaib and Choi 2020).Meanwhile, irrigation is demonstrated to remarkably influence hydro-climate by altering surface energy balance (SEB) and hydrological cycle (McPherson 2007, Biggs et al 2008, Sridhar 2013, Yang et al 2019): enhanced evaporation can alter SEB by reducing sensible heat fluxes and increasing latent heat fluxes and lead to reduced surface temperature and diurnal temperature range, at both global and regional scales (Haddeland et al 2006, Mahmood et al 2006, Sacks et al 2009, Ozdogan et al 2010, Puma and Cook 2010, Sorooshian et al 2011, Kueppers and Snyder 2012, Han and Yang 2013, Jiang et al 2014, Chen and Dirmeyer 2019); irrigation modifies the hydrological cycle by increasing soil moisture and surface moisture fluxes, resulting in the increase of the amount of atmospheric water which may indirectly enhance cloud cover and precipitation (Zheng and Eltahir 1998, Boucher et al 2004, Gordon et al 2005, Mahmood et al 2008, Lo and Famiglietti 2013, Tuinenburg et al 2014, Lo et al 2021).And such influences in precipitation by irrigation are found at both local and regional scales (Segal et al 1998, Im et al 2013, Phillips et al 2022) but through different pathways.
Some studies reveal that the evaporative water vapor added by irrigation directly forms precipitation locally in the irrigated area, increasing the local precipitation recycling rate (Eltahir and Bras 1996, Eltahir 1998, Sorooshian et al 2011, Zhang et al 2019, Liu et al 2021b).At the same time, irrigation may also form a horizontal gradient in the area where the irrigated area and the non-irrigated area overlap, and strengthen the mesoscale circulation, leading to increased precipitation over the nonirrigated area (DeAngelis et al 2010, Niyogi et al 2010, Guimberteau et al 2012, Harding and Snyder 2012a, 2012b, Tuinenburg et al 2014, Pei et al 2016, Vrese et al 2016, Lu et al 2017, Nauert and Ancell 2019).
Despite the revealed mechanisms by which irrigation can influence precipitation at different scales around the world, the situation in TB remains elusive.Although increasing trends in greening and precipitation are found in northwest China (Song et al 2011, Qiu et al 2014, Li et al 2015, Ma et al 2015, Dong et al 2018, Lamchin et al 2018, Guan et al 2022, Wu et al 2022), the role of irrigation in moderating the regional hydrological cycle is yet to be identified.A concerning question in recent study is that if the altered precipitation due to irrigation can support sustainable agricultural development in TB (Xu and Lin 2021).As such, an in-depth study of irrigation on the atmospheric hydrological cycle warrants urgent need in TB.
Concerned about the food security, specifically irrigation sustainability, this study aims to assess the impact of irrigation on atmospheric hydrological cycle and to evaluate the sustainability of irrigation by tracking the irrigated water in different atmospheric processes.In the remainder of this paper, we first describe the modelling techniques (weather research and forecasting (WRF) and HYbrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT)) and their configurations, followed by water balance analysis to reveal the destination of irrigated water in the present-day scenario; we then conduct trajectory analysis to explicate the pathways by which irrigation can influence atmospheric water transport.In addition, we extend the above analysis to a 2050 scenario to understand the potential impacts of irrigation in conjunction with impending global warming on the sustainability of irrigation in TB.

WRF model and configuration
This study employs the WRF model v4.2 (Skamarock et al 2019), in which a new irrigation scheme (Valmassoi et al 2020) is introduced to allow more detailed operation of irrigation water, to investigate the impacts of irrigation on atmospheric hydrological cycle (Decker et al 2017).
For the present-day scenario, the year 2016 is selected as the climate conditions in 2016, specifically the mean air temperature, specific humidity and wind speed, were all located in the 30 year range interval and considered representative of the present climate normal standard (see figure 1 in supplementary material).In addition, the precipitation pattern during the summer of 2016 in the Tarim Basin provides an ideal scenario to examine the impact of irrigation (see figure 3 in supplementary material).The observed lower than multi-year average rainfall in June, and July (12.3 mm and 19.3 mm, respectively, compared to multi-year averages of 23.3 mm and 23.2 mm) could highlight the potential influence of irrigation in comparative simulations between two scenarios in this study.Larger than multi-year average rainfall in August precipitation (77.35 mm vs 25.43 mm) can broaden the scope of the investigation to explore the impact of irrigation on extreme rainfall, which is also key area of interest regarding the regional hydroclimate.
WRF is set up in a one-way nesting mode with two nested domains of 120 × 120 and 240 × 240 grids with resolutions of 15 km and 5 km, respectively: the innermost domain d02 includes the whole TB area and its surrounding mountains (figure 1).The model is configured with 47 sigma levels in the vertical direction with top level set to 50 hPa.Fixed time steps of 30 s and 10 s are set for d01 and d02, respectively.Initial and boundary conditions are provided by the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) dataset (Hersbach et al 2018) from the European Centre for Medium-Range Weather Forecasts (ECWMF) with a spatial (temporal) resolution of 0.25 • (1 h).For more accurate representation of land surface characteristics as of the simulation year (i.e.2016), global land cover dataset by European Space Agency Climate Change Initiative (ESA-CCI) (European Space Agency 2017) is used to prescribe the land use and land cover in the study area.Besides, albedo, leaf area index (LAI) and vegetation fraction (FVC) are updated using information from Global LAnd Surface Satellite (GLASS) dataset (Liang et al 2013a(Liang et al , 2013b(Liang et al , 2021)).The simulation is conducted from 00:00 1 May to 23:00 31 August 2016 UTC with the first month as the spin-up period while the remaining three months (June, July and August, or JJA) for analysis.Other physics parameterization schemes used in the simulation are summarized in table 1.To assess the effects of irrigation, two scenarios, irrigation (IRR) and default -no irrigation (DFT), are set up for irrigation and default (no irrigation), respectively.For IRR, the irrigation scheme is enabled for the whole simulation area in d02 only during the June, July and August (JJA) period.The amount of irrigation water is set to 6 mm per day according to the local irrigation daily quota (Wu et al 2022).In DFTthe benchmark scenario-all surface feature information (land cover, albedo, LAI and FVC) is modified to remove all agriculture-related features (i.e.no irrigated cropland).
For the global warming scenario, the year 2050 is selected as the future timeframe.As the coupled model intercomparison project phase 6 (CMIP6) can forecast future climate up to 2100, 2050 can represent the impending climate in the 21st century (see figure 2 in supplementary material).Meanwhile, 2050, 30 year timeframe, may provide mid-term references for current policy makers.Future meteorological data from bias-corrected CMIP6 global dataset for dynamical downscaling of the Earth's historical and future climate for 2050 under the ssp245 scenario (RCP4.5 global forcing pathway with SSP2 socioeconomic conditions) is used as initial and boundary conditions (Xu et al 2021).The dataset is based on 18 models from CMIP6 and ERA5 dataset, provides a state-of-the-art future background climate.Given no future projection about the land surface feature information is available in the study area, the present-day land cover information is kept in the 2050 scenario.So that we can give a more straightforward comparison of irrigation impacts between present day climate and global warming.Other settings (e.g.physics parameterization schemes, irrigation settings scenarios, etc) remain the same as the 2016 scenario.In following analysis, all results are based on d02 simulations unless noted otherwise.

Water balance analysis
To quantify changes in simulated water budgets due to irrigation, we conduct land surface water balance analysis following the equation below: where ∆S denotes the change in the total storage (including soil water storage ∆SOIL, surface snow storage ∆SNOW and canopy interception ∆CAN), P the precipitation, I the irrigation, E the evapotranspiration, R S surface runoff and R U underground runoff.All terms in equation ( 1) are WRF output variables except for I to be calculated using input information as follows: where Q d is the daily irrigation amount (6 mm in this study; see section 2.1), D JJA the number of the JJA days (i.e.92 days), and IF avg the spatially averaged irrigation fraction of d02 (cf figure 1).
The regional atmospheric water balance is given by: where ∆PW is the change of water vapor content (diagnosed by the WRF-Python tool) while ∆WFT the accumulative water vapor transport on d02 boundaries (note the water vapor content on the model top at 50 hPa is very close to zero).Given the minimal water vapor transport through the top level, we only calculate the vertically integrated moisture flux transport (VIMFT) on four horizontal lateral boundaries as follows (Fasullo and Webster 2003): where g is gravitational acceleration, q the specific humidity at each pressure level, u and v the wind vector.
For both land surface and atmospheric water balance analysis, the differences between IRR and DFT will be calculated later in section 3.1 to reveal the destination of irrigated water.We note that the accumulated water amount for all terms in equations ( 1) and ( 2) are averaged over the whole d02 (figure 2), whose corresponding WRF output names (or derivatives using WRF-python) can be found in table 2.  1), ( 3) and ( 5)).
Variables in equations ( 1), (3) and ( 5 (i) The 820 rainfall event: ten day backward trajectories are calculated to trace the origin of rainfall for each hour during the extreme event from 00 UTC 17 to 00 UTC 22 August, 2016 (figure 4).Initial locations are set as the falling area (i.e.accumulated precipitation exceeding 1 mm) of the 820 rainfall event in d02 (32 horizontal locations in figure 5) at 1500 m above ground level.The ten day integration time is chosen because the retention time of moisture in the atmosphere is ten days (Chen et al 2012).To identify the main paths from the large number of trajectories, we applied K-means++ cluster method (see Arthur and Vassilvitskii (2007) for details).(ii) The transport of irrigation water in July 2016: Forward trajectories are calculated to trace the evaporated water vapor from the ground surface over irrigation area for each hour from 00 UTC 1 to 00 UTC 10 July, 2016 without observed rainfall.Initial locations are set as the irrigation area in TB in d02 at 0 m above ground level (97 horizontal locations in figure 6).Integration time is set to 15 day due to minimal residual of air parcels in the atmosphere after a 15 day period.
For both backward and forward trajectory calculations, HYSPLIT uses the hourly wind regime variables from d02 of the IRR scenario as input and produces hourly track points as output.

Water balance analysis
We first look into the impacts of irrigation on land surface and atmospheric water balance (figure 2).For the surface water balance, the most remarkable change is in the evaporation: 90.5% of irrigated water enters the atmosphere by surface evaporation, leading to the increased precipitation that accounts for 53.4% of irrigated water.As a result of irrigation, the net water amount of land surface is increased: increases are observed in soil moisture (39.2%) and runoff (26.1%) but a slight decrease in the snow water content (−3.7%).While for the atmospheric water balance, in spite of the increased moisture supply from evaporation due to irrigation, the change in atmospheric moisture content is minimal (2.0%).Water vapor transport (39.0%) and precipitation (53.4%) dominate the total atmospheric water content.It is noting that 39.0% of irrigated water is transported out of the domain as water vapor flux.By tracking the hydrological process of irrigated water, the water balance analysis suggests the final destination of irrigated water as the surface water content and water vapor transport.And 90.5% of the irrigated water is transported via the atmospheric hydrological process, in which precipitation (53.4%) and water vapor transport (39.0%) are the two dominant components.

Impacts on summer precipitation
Remarkable spatiotemporal variability is observed in the precipitation change caused by irrigation (figure 3).The Basin area and the southern mountainous area have seen the largest precipitation increase (figure 3(d)), accounting for 92.8% of the increased precipitation in the domain (59.8% in the basin area and 33.0% in the southern mountain).Although minimal changes are found in the arid basin area during June and July (figures 3(a) and (b)), a large increase in accumulated precipitation is seen in August (figure 3(c)).Outside the basin area, the spatial pattern of precipitation varies during JJA period, especially in northern mountainous area and other area (cf regions N and O in figures 3(a)-(c)).However, the resultant change in total precipitation is minimal in these two areas (7.2%).So the basin is the most influenced area by irrigation with respect to precipitation, where 59.8% of the total increase is observed.
Given TB (black box shown in figure 4) seeing the largest precipitation increase, we analyze the process of the 820 rainfall event to reveal the detailed influences of irrigation on precipitation.This event occurred in the western TB-adjacent to irrigation area-between 06 UTC 17 August and 21 UTC 21 August.A surge of rainfall was observed from 12 UTC 19 August to 12 UTC 20 August with peaking rainfall rates of 1.6 mm h −1 and 0.8 mm h −1 In IRR and DFT, respectively.The ten day HYSPLIT back trajectory analysis (more details see section 2.3) suggests that origin locations of the precipitation are mainly distributed in the interior of TB, in particular the irrigation area at the foot of surrounding mountains (figure 5(a)).We further cluster the back trajectory results to two categories (East cluster and West cluster, accounting for 57.6% and 42.4% of total trajectories, respectively, figures 5(b)-(d)) using the K-means++ method (see Arthur and Vassilvitskii (2007) for details).
For the Eastern cluster (figure 5(c)), the track points are mainly distributed over the irrigation area (red dots in figure 5(c)), and the horizontal distribution is consistent with the contour line, indicating that almost no water vapor comes from outside the mountain area.It is also noting the water vapor source is mainly distributed in the bottom 2000 m of atmosphere (cf coloring of dots in figure 5(c)).Similar distribution of vapour particles is found for the Western cluster (figure 5(d)).So for both Eastern and Western cluster, the irrigated area in TB is the main origin of rainfall water in the 820 event.Compared with DFT, irrigation in IRR strengthens the evaporation process and facilitates precipitation with greater moisture supply.Furthermore, the excessive water vapour supply by irrigation is constrained in the basin due to the sheltering of mountains, which further enhanced the precipitation processes of the 820 rainfall event in IRR scenario.As such, we find that irrigation dominates the precipitation increase in the interior of TB for the 820 rainfall event.

Impacts on atmospheric water transport
The HYSPLIT forward trajectory analysis is conducted to identify the target area of the irrigated water during a 15 day transportation (configuration details see section 2.3).We analyse the results of northern, western and southern irrigated area for three transportation stages (0-120 h, 120-240 h and 240-360 h).For the northern irrigated area (figures 6(a)-(c)), water vapour particles mainly move to the southwest of the basin along with the dominant northeasterly wind, most of which end in the foothills of the southwestern basin.For the western irrigated area (figures 6(d)-(f)), affected by the downhill wind in the northern and southern mountainous areas, water vapor particles due to irrigation tend to move eastward.However, due to the prevailing north-easterly wind, most of the water vapor particles are transported southward along the basin boundary, where they land close to the mountains-only a small number of particles (0.86%) continue travelling to the eastern boundary.As for the southern irrigation area (figures 6(g)-(i)), the prevailing north-easterly wind could not transport the water vapour particles further southward-outside the basin-due to the blocking of mountains on the southern boundary.In all three areas, most water vapour particles could not travel in the atmosphere longer than ten days-only 8.0% of water vapor particles could.Besides, the height distribution (figures 6(j)-(l)) indicates that the transportation of water vapor is limited to the lower atmosphere (i.e.below 5000 m) with 50.0% and 85.0% under 2000 m and 5000 m, respectively.Putting together all the results from three irrigated areas, we find the southwest of TB is a main destination of water vapor from irrigated areas with notable increase in precipitation (cf section 3.2).
However, given 39.6% of the irrigated water lost via atmospheric water vapor transport from d02 as revealed by the water balance analysis (cf section 2.1), we look into the difference of cumulative water vapor flux on four boundaries between IRR and DFT for causes to understand the destination of the irrigated

Implications of global warming
Given the impending global warming, we further look into the future water balance in TB by conducting simulations of the 2050 scenario (details refer to section 2.1) to understand the implications for water resource sustainability.For the surface water balance (brown box in figure 8), in relation to fractions of irrigated water, the evaporation is strongly strengthened under the 2050 scenario to 108.8% (cf 90.5% under the present-day scenario), the runoff (underground and surface) and soil moisture decrease to 48.6% (cf 61.6% at under the present-day scenario).Both precipitation and water vapor transport processes are enhanced in 2050 (blue box in figure 8): with respect to fraction of irrigated water, the precipitation will increase from 53.4% to 59.6%, while water vapor transport increases from 39.0% to 45.2%, suggesting more water vapor is lost by atmospheric transport.The eastern boundary of the basin is still the dominant gateway for the loss of water vapor in 2050 with notable increase (17.9%) in the atmospheric water loss even without the expansion of agriculture and irrigation.
Putting together the water balance analyses in 2016 and 2050, we conclude that in the context of global warming, irrigation causes stronger evaporation that brings more water to the atmosphere and thus enhances the process of both precipitation and water vapor transport.More water vapor (6.2% of irrigated water) will flow out of the basin than that in 2016.As a result, more attention needs to be paid to the agriculture sustainability and freshwater shortage for the future of TB.With the decrease of both surface and underground water resources and the increase of water supply in Northwest China, the water loss-10.5 billion tonnes, or 1/3 of annual irrigation in Xinjiang (Xinjiang Bureau of Statistics 2021)-caused by irrigation seems to have a negative impact on the sustainable development of agriculture in TB.

Conclusion
In this study, we examine the impact of irrigation on atmospheric hydrological cycle in TB by conducting WRF simulations under the present-day and future climate scenarios.The calculation and analysis of surface and atmosphere water balance illustrates the atmospheric transport pathways of the irrigated water.The HYSPLIT model is used to further reveal the destination of the irrigated water and identify the origin of precipitation in the basin.The main findings are summarized as follows: (i) In the present-day scenario (2016 as the representative year), 90.5% of the irrigated water enters the atmosphere by evapotranspiration.As the result of atmospheric hydrological cycle, 53.4% of the irrigated water goes back to the ground surface by precipitation, and 39.0% leaves the area by atmospheric transport as water vapor.However, in the future scenario (2050 as the representative year), both the evaporation and atmospheric hydrological cycle are enhanced due to global warming, leading to increased loss of irrigated water via atmospheric transport as water (45.2% of 2050 vs. 39.0% of 2016).
(ii) The irrigated water is largely recycled in TB due to the blocking of surrounding mountains, which can be evidenced by both forward and backward trajectory analyses: the forward analysis shows the water vapor from the irrigated area can only be transported to the southwest of the basin by prevailing wind, causing the increase of precipitation in the western part and the southern mountain area to the basin; while the backward trajectory of the precipitation in the basin proves that the origin of the precipitation is the irrigated area.(iii) Although the irrigation may supply the atmosphere with more water vapor via enhanced evapotranspiration, given the dominant role of external water vapor in replenishing the atmospheric water, the weakening of prevailing wind over the eastern basin boundary lead to a net water loss as a whole.And this situation will be worsened under the future warming scenario: a reduction in atmospheric transport equivalent to 6.2% of irrigated water will be induced by irrigation compared to the present-day value of 39.0%.
Overall, WRF simulations for a three-month period under two scenarios-present-day and future warming-were conducted to examine the impact of irrigation in TB on atmospheric hydrological cycle.In addition, backward HYSPLIT calculation was used to reveal the contributions of irrigation to an extreme rainfall event, and forward HYSPLIT calculation was applied to identify the destination of the irrigated water.By combining WRF simulations and HYSPLIT calculations, this study draws a clear conclusion on the destination of irrigated water in TB.The analysis of current and future scenarios can help policymakers pay more attention to the sustainability of local irrigation water resources.However, we note that only one year irrigation period is simulated, thus the cumulative long-term impact of irrigation still needs to be explored.Besides, this study is focused on the land surface and atmospheric water balance; the impacts on groundwater resources of irrigation warrant future research.

Data availability statement
All datasets used in this study are publicly available from the references indicated.ERA5 data can be requested from cds.climate.copernicus.eu.More information on ESA and GLASS data can be found from www.esa-landcover-cci.organd www.glass.umd.edu, respectively.The data that support the findings of this study are available upon reasonable request from the authors.

Figure 1 .
Figure 1.Terrain map of the WRF simulation domain (d01 and d02) covering TB with fractions of irrigated area used in the IRR scenario.

Figure 2 .
Figure 2. Difference in land surface (brown box) and atmospheric (blue box) water balance budgets between IRR and DFT in 2016.Note: (1) different directions between top and bottom axis; and (2) percentages (in parentheses) for the fractions of irrigated water.

Figure 3 .
Figure 3. Differences in the accumulated precipitation between IRR and DFT (IRR-DFT) for (a) June, (b) July, (c) August and (d) JJA (cumulated value).Letters indicate different analysis regions: B-Basin area, S-Southern Mountainous area, N-Northern Mountainous area and O-Other areas.

Figure 5 .
Figure 5. HYSPLIT-based backward trajectory results for the TBP event (a) locations, (b) geographical clustering (purple and green for the east and west clusters, respectively; see section 2.3 for the clustering algorithm) and altitude above ground level (in color bar) for (c) the eastern and (d) western clusters.Red dots indicate the locations of irrigated areas.Colored lines represent topographic contour.

Figure 6 .
Figure 6.HYSPLIT-based forward trajectory results for the period from 1 July-10 July: locations (left nine subfigures) and height distribution (right three histogram) from three main irrigation areas in three different trajectory periods for 0-5 days (left), 5-10 days (middle), 10-15 days(right).(a)-(c) Horizontal results of northern irrigation area; (d)-(f) horizontal results of western irrigation area; (g)-(i) horizontal results of southern irrigation area; (j)-(l) height distribution in three periods.Blue dots represent locations of irrigation.Colored lines represent topographic contour.The arrow indicates the JJA mean wind speed at 10 m above the ground level.

Figure 7 .
Figure 7. Accumulated water vapor amount for the eastern boundary of the analysis domain (cf figure 1): (a) the IRR scenario, (b) difference between IRR and DFT (i.e.IRR-DFT) and the vertical profiles of: (c) water vapor flux, (d) water vapor and (e) wind speed.Note the coloring indicates the flow direction: red for outflow while blue inflow.

Figure 8 .
Figure 8. Same as figure 2 but for the future scenario of 2050.

Table 1 .
Overview of the WRF physics options.

Table 2 .
WRF output variables used in water balance analysis (cf equations ( Draxier and Hess 1998,  Jiang et al 2017, Shi et al 2019, Zhang et al 2021)is used in this work to compute three-dimensional paths of water vapor particles using wind regimes and topography.Specifically, we use HYSPLIT to calculate backward and forward trajectories of water vapor for the extreme rainfall event peaking on 20 August 2016 (the 820 rainfall event hereinafter) and the transport of irrigation water in July 2016, respectively, for different purposes illustrated below: