Assessing ecosystem services under water stress in the largest inland river basin in China based on hydro-ecological modeling

Ecological changes in arid lands are often determined by the available water resources. For ecological protection, 15 it is important to consider both hydrological and ecological processes. This paper presents a modeling approach to assess environmental changes and ecosystem services (ESS) in the largest inland river basin in China. Current water cycle and ecosystem protection measures were simulated, and future land use change scenarios were proposed accordingly. China is on the frontline of ecosystem protection and afforestation, but according to the simulation results, the available water resources cannot support more vegetation in its largest inland river basin. Without an additional water supply, 25.9% of the existing 20 area of natural vegetation will be degraded by 2050. A reduction in the area of cotton plantation did not substantially reduce farming incomes in the simulation. Drip irrigation and ecological flooding was effective for improving ESS, especially agricultural provisioning and riparian forest provisioning services, as well as regulating and supporting services. The assessment of ESS can achieve socio-economic and ecological benefits in a sustainable way, and also enable the impact of the complex environmental factors to be understood. The results can be used to evaluate current situations and develop 25 creative solutions.

downstream. High evapotranspiration rates and high groundwater salinity levels can intensify water shortages (Pang et al. 2010). In 2018, the total afforested area of the Tarim River Basin was 5,200 km 2 . It is not clear if the available water resources are able to support the establishment of new vegetation in the river oases.
The Tarim River Basin has developed as a vast, unique, and variable natural landscape during its long history (Wu and Cai 70 2004;Zhao et al. 2013). The world's largest natural Populus euphratica specimens are distributed in the basin oases (Thomas et al. 2017). The Taklimakan Desert is the second largest desert in the world, and the mysterious Lop Nor has long been attractive to adventure tourists (Tao et al. 2016). Due to the hyper-arid climate, the natural vegetation (other than P. euphratica) is mainly comprised of low bushes, including Achnatherum splendens, Apocynum spp., Tamarix chinensis, Halimodendron halodendron, Kalidium foliatum, and Nitraria tangutorum. These drought-tolerant plants can alleviate 75 desertification conditions to a certain extent (Yu et al. 2015). In the region's arable land, the soil types are mainly loamy sand. Because of the high evapotranspiration, irrigation water shortages, and poor drainage systems, soil salinization has been aggravated in recent years. Statistical data has revealed that 38% of the region's agricultural fields suffer from salinization (Xu et al. 2008). The most serious salt contaminated area is in the middle and lower parts of the Tarim River (Xu et al. 2014). Economic development in the Tarim River Basin is far behind that of eastern China. The basin is dominated by 80 agriculture and grazing. In recent years, petroleum exploration and the petrochemical industry have been developed (Huang et al. 2016), but most of the local populace have not been benefited from this activity. In 2000, the Chinese central government initiated the western development strategy (Lu et al. 2013), and a huge amount of investment was made to develop the western regions of China, including the Tarim River Basin (Li et al. 2011). New railways are currently under construction and the basin economy is developing rapidly (Tong et al. 2015). Under such circumstances, the utilization of 85 water resources and protection of the ecological environment are more important than ever before. Population growth and farmland reclamation have substantially changed the land use, water resources, and ecosystems in the basin (Zhao et al. 2013). In the mainstream of the Tarim River, the landscapes are very different from those of the upper to lower reaches.
Most population, farmland, and industries in the river oases are located in the upper reaches (Feike et al. 2015). Riparian forest and scattered desert plants are mainly distributed in the middle and lower reaches. How best to evaluate ESS changes 90 in this arid basin remains uncertain, but will have a large influence on decision-making and the sustainable development of the whole region. This study adopted a hydro-ecological modeling approach to assess ESS changes, while considering water availability. The results provide further guidance for regional sustainable development and ecosystem protection in arid lands.

Method 95
To simulate hydro-ecological processes and assess ESS changes, two hydrological models (MIKE HYDRO and MODFLOW) were employed and a decision support system (DSS) was established based on Qt Creator and C++ programming. Meteorological, hydrological, geographical, ecological, and socio-economic data were collected as inputs into https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. the models, and the ESS evaluations were provided by DSS outputs. The connection between MIKE HYDRO and the DSS was at the sub-catchment level with regard to the surface water consumption and allocation strategies. Groundwater 100 conditions were simulated by MODFLOW and were fed to the DSS on a monthly basis. The equations and other logical inputs in the DSS were created by reference to expert knowledge in cross-disciplinary research fields. The massive amounts of research data used in the study were based on years of sample collections from field stations and observations, supported by the Sino-German Sustainable Management of River Oases Along the Tarim River (SuMaRiO) project, which is a collaboration between 11 German and 9 Chinese research institutions. 105 Ecosystem services are usually comprised of provisioning, regulating, and supporting services. In this study, ESS were divided into three classes: services from agriculture, riparian forest, and grassland ( Fig. 1). Agriculture can provide provisioning services, including cotton production, fruit production, production of other crops, and farming incomes.
Riparian forest provides provisioning services (biomass production), regulating services (drifting dust control, sand stabilization, wind speed reduction, and carbon sequestration), and supporting services (tree species). Grassland provides 110 provisioning services (Apocynum and reed production) and regulating services (drifting dust control and sand stabilization).
In the first edition prototype of the model, only the major functions of the ESS were considered. This reduced the computational time required for each scenario. If enough knowledge and data are collected in the future, it would be possible to add other ESS functions, such as regulating the regional climate, protection of wild animals, and flood control.  MIKE HYDRO model is a versatile and physically-based modeling tool for the analysis, planning, and management of integrated water resources within a river basin (DHI 2014). The model used in the Tarim River Basin (Fig. 2) contained four sub-catchments (namely SC1, SC2, SC3, and SC4), a mainstream with Muskingum routing (Chow et al. 1988), a simplified 120 one-layer groundwater module, a Food and Agriculture Organization (FAO) dual crop coefficient module, an FAO 56 climate module using the Penman-Monteith method for evapotranspiration (Allen et al. 1998), an FAO 56 soil module with nine different soil types, more than 50 water users (for irrigation, ecosystem, livestock, domestic, and industrial uses) in the oases, and eight reservoirs to regulate seasonal surface runoff. The MIKE HYDRO model was fully calibrated and validated over a nine-year period using observed discharges from gauging stations, and a high level of agreement was attained 125 between simulated and observed values (Yu et al. 2017). The MIKE HYDRO model provided information for irrigation water demand, ecological water, infiltration, and fruit and crop production in the DSS simulation. The MODFLOW model was applied in parallel to simulate groundwater conditions over the same period. MODFLOW is a commonly used threedimensional modular groundwater model (Harbaugh et al. 2000). The MODFLOW model in the Tarim River Basin provides an updated water head, river leakage, irrigation seepage, ecological water percolation, and other simulated groundwater 130 movements using 500 × 500 m cells. wells. To build connections among hydrological processes, ecological processes, and ESS outputs, a certain logic must be established in the DSS. The logical inputs mainly include mathematical equations, computer algorithms, and fuzzy logics (Mendel, 1995). These logical inputs were formed by geoscience theorems, empirical equations, experimental outcomes, and 140 expert knowledges. Many empirical equation need parameter modifications before they can be applied in this arid region (Patrick et al. 2015;Yu et al. 2017). In addition to the scientific experts, many local farmers and stakeholders were also invited to provide their opinions on the socio-economic developments in the basin. Their advice and suggestions were collected and considered in the management scenarios and future ESS changes. In total, there were more than 100 equations, 200 data files, and 10,000 lines of C++ program in the DSS. 145 The river discharge was based on observation data and was simulated by the Water Availability in Semi-Arid Environments (WASA) model in the upstream mountainous region for future projections. River outflow was determined by inflow and water consumption along the oases (with no rainfall-runoff in the mainstream Tarim River). Farming incomes were calculated by cotton, fruit, and other crop production multiplied by the profit of each crop's production. Cotton production was aggregated from the cotton yields in each cell, and further determined in terms of water availability, soil 150 salinity, and an evapotranspiration-yield relationship model (Stewart et al. 1977;Yu et al. 2015). Drifting dust control, sand stabilization, and wind speed reduction were all determined in terms of tree height and crown area by fuzzy rules in each cell, and then aggregated to calculate a total value. Tree species were determined by the fuzzy logic between groundwater level and the flooding of natural vegetation. Apocynum and reed production were influenced by groundwater level, groundwater salinity, and grazing area. Drifting dust control and sand stabilization by grassland were determined by grassland area and 155 density. The simulation period was from 2012 to 2050, with monthly time steps.
Water-saving irrigation is a watering strategy that can be applied using different types of irrigation application methods.
The correct application of water-saving irrigation requires a thorough understanding of the yield response to water (crop sensitivity to drought stress) and the economic impact of a reduction in the harvest (English 1990). In regions where water resources are restricted it can be more profitable for a farmer to maximize crop water productivity rather than maximizing 160 the harvest per unit land (Fereres and Sorian 2007). In the study area, the implementation of drip irrigation under mulch is becoming popular to save water and guarantee productivity. Our model (Fig. 3)   The development of the DSS involved contributions from a variety of disciplines, such as computer science, hydrology, 180 geography, pedology, botany, agronomy, humanities, and economics. By including this cross-disciplinary knowledge in the DSS, the links between the environmental elements were reproduced and the resulting scenarios enabled future predictions to be made.

Water availability 185
The outputs of the hydrological models (Fig. 5) indicated that water availability was very limited in spring, autumn, and winter, but there was a substantial improvement in the summer. For surface water, summer floods (in July, August, and September) accounted for 78.8% and 75.2% of the total river discharge in 2012 and 2013, respectively. There was little surface runoff in winter and spring, especially in SC2, SC3, and SC4. There was an obvious reduction in the discharge from upstream to downstream. The discharges in SC4 were only 13.7% and 8.9% of those in SC1 in 2012 and 2013, respectively. 190 Huge amounts of water resources were consumed by agriculture and natural vegetation in the upper and middle reaches. In addition, because the water volume in 2013 was lower than in 2012, a higher percentage of the surface water would have been used in the upper and middle reaches than in a wetter year, which would in turn intensify downstream water shortages in the dry year. On the other hand, groundwater conditions were also better in summer than in winter. The water heads were raised by 2.51 and 2.74 m from February to August in 2012 and 2013, respectively. As with the surface water, the 195 groundwater availability was also much poorer in the middle and lower reaches than it was in the upper reaches, especially in winter. Groundwater levels were lower than 10 m in a parts of SC4. This result concurs with some of the previous findings reported in this region (Feng et al. 2005;Chen et al. 2015), and indicates that in some regions of the lower reaches, the groundwater table is beyond the reach of most desert vegetation (Fan et al. 2004;Hao et al. 2010).
https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. general, both surface and ground water availability were poor in the downstream oases. Better agricultural water allocation strategies should be considered for integrated water resource management in the whole catchment. 210

Natural flooding through ecological gates
Due to low precipitation levels and water heads, natural flooding is crucial for the riparian forest along the river oases. There are more than 100 flood gates along the river. The designed flow rate varies from several to hundreds of cubic meters. In addition, local farmers also occasionally open-up small breaches on the river bank to flood their fields and trees. It is therefore difficult to determine how much ecological water passes through all of these gates. We monitored several typical 215 large gates in SC1, SC2, and SC3 (Fig. 6), then simulated the flooding rates by considering their capacities and thresholds. There was a clear decreasing trend in the flooding rate from the upstream to downstream ecological flood gates. The level 220 of ecological flooding in SC3 was about 1/7 of that in SC1, which clearly indicated the shortage of ecological water in the middle reaches. In SC4 (lower reaches), due to the very low water volume and river interception, flooding from ecological flood gates was not included in the model. Spring flooding only occurred in SC1, and flooding in SC2 and SC3 only occurred in summer and early autumn. This is a critical condition for riparian forest in the downstream area considering the groundwater levels. For example, for P. euphratica, adult trees will not be largely influenced as long as their roots can reach 225 the water head, but young trees may not survive without ecological flooding. Therefore, the regeneration of Populus trees is threatened by the severe water scarcity problems in the middle and downstream oases. For grassland and scrub vegetation, https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. the living conditions are even more challenging because their roots are usually shorter than those of Populus euphratica. In SC4, plants mainly grow close to the river banks, and flooding from the ecological flood gates was not considered in the model. 230

Irrigation water consumption and water deficit
Irrigation water demand was calculated in the model by considering climate conditions, crop growing factors, and the cultivated areas. The actual irrigation water consumption was dependent on the water availability for different water users.
Whenever both surface and groundwater were unable to supply the irrigation water demand, a certain water deficit would occur. Land management decisions regarding the reclamation or abandonment of farmland are mainly determined by water 235 and land productivity, which are further constrained by the water deficit along the river oases. The simulation results for total irrigation water consumption and the water deficit of each sub-catchment are shown in Fig. 7. Most of the farms in the study area were located in SC1, and this sub-catchment therefore consumed the largest share of 240 irrigation water. All sub-catchments had a certain amount of water deficit, varying from 45 to 90 million m 3 , but the ratios of water deficit to irrigation water consumption were much higher in SC2, SC3, and SC4 than in SC1. This indicated that for a certain unit of farmland, water deficits were larger in the middle and lower reaches than in the upper reaches. A peculiar phenomenon was also found, in which as the water consumption increased from 2012 to 2013, the water deficit was not reduced. After further investigation, this was found to be due to the increase in farmland area. In the study area, crop rotation 245 and leaving some land fallow are very common practices due to water scarcity. The area of farmland to cultivate each year is https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License.
largely dependent on the upstream river discharge and government water policies. In 2013, an additional 49.15 km 2 of farmland area was cultivated compared to 2012, which led to an increase in the total irrigation water demand.

Land use changes
Under the future climate scenario CCLM RCP 2.6 (Duethmann et al. 2016), water movements and vegetation growth were 250 simulated and land use changes were predicted until 2050. In this scenario (Fig. 8), our research focused on the middle reaches where most of the natural vegetation was located. Based on the logic and rules of the model, if a number of vegetation cells vanished in future years, it would mean that the groundwater level was too low for plant growth in the corresponding areas. Furthermore, it would indicate that ecological flooding was not sufficient and groundwater recharge was too low to support natural vegetation growth in these areas. The farmland area would not change naturally in the DSS, 255 because there is a basic rule in the model that if surface water is not sufficient for the farmlands, groundwater will be pumped to guarantee the irrigation water supply. Based on the simulation results, land degradation was quite obvious along the river oases, especially in the middle and 260 lower reaches. From the comparison shown in Fig. 8, it is noticeable that many "green areas" had vanished in 2050, with this phenomenon occurring more often in locations far from the river course. By the end of simulation period, 25.9% of the area of riparian vegetation had degraded. The DSS suggested that this degeneration of natural vegetation was caused by the lack of ecological flooding and low groundwater table. Fallow land and crop rotation have been applied for many years in the study area, but the extra water resources in wet years would be better used to improve the decaying ecosystem in the 265 downstream oases rather than in new cultivated farmlands. The simulation results clearly indicated ongoing land degradation https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License.
in the middle and lower reaches until 2050. To maintain the newly planted trees and grassland in the oases, additional water sources must be guaranteed in the next few years.

Ecosystem services
There are many possible scenarios in which the potential ecological water supply could be increased and the riparian 270 ecosystem maintained (as indicated by ESS). In our model, the ESS were simulated under three different management scenarios ( Fig. 9): (1) everything kept as usual (reference scenario); (2) the cotton area reduced by 20%; and (3) the extent of drip irrigation increasing (from 50% to 90%), while ecological flooding increased by 50%. The ESS changes were mainly dependent on management alternatives, water availability, and climatic variation during the simulation period. The resultant impacts on socio-economic factors were closely associated with the well-being of human society and regional sustainable 275 development. The ecosystem provisioning, regulating, and supporting services were weighted from 0 to 1, representing how many of our management goals could be fulfilled. These goals were generated from the expectations of local stakeholders and decision-makers, who expressed their suggestions on how to improve the ecosystem.
https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. The results for the reference scenario indicated increasing trends in the total ESS values for agriculture, riparian forest, and grassland, with increases of 16.9%, 24.7%, and 12.8%, respectively from 2012 to 2050. However, grassland provisioning and regulating services were expected to decrease. A number of grassland areas are either located far from the ecological flood gates or the groundwater table is too deep for the plant roots to reach. This situation was improved in the 285 cotton reduction scenario because 20% of the cotton area could easily be replaced by natural grassland. Cotton production and farming incomes would not be substantially reduced by a reduction in the cotton area, which suggests that water supply is a bottleneck in the crop yield in the river oases. This point was proven by the drip irrigation and ecological flooding scenario, which resulted in large amounts of irrigation water being saved and more ecological water being available. Drip irrigation is likely to be very effective in increasing cotton production and farming incomes in future years. An increase in 290 ecological flooding would also improve the provisioning, regulating, and supporting services supplied by riparian forest (e.g., drifting dust control and the mean number of plant species). However, in the period to 2050, there would be a significant decline in agricultural provisioning services, grassland provisioning and regulating services, and the total ESS value in the drip irrigation and ecological flooding scenario. The reason for this was the soil salinization effect. Drip irrigation would cause the accumulation of salt in the soil profile, which is harmful for crop growth in the long term. Cotton has a strong 295 tolerance to soil salinity, but many other crops and fruits cannot resist high levels of salinization, and their yields would be substantially reduced based on our model. Overall, in comparison with the reference scenario, the cotton reduction scenario would slightly increase the total ESS, and the drip irrigation scenario would substantially increase the ESS in the near future.
However, there would then be a large decline by 2050 due to soil salinization.

Conclusion and discussion 300
China is on the frontline of combating desertification and ecosystem protection. In arid regions of China, planting trees is promoted by the central government and is undertaken by social communities and the general public. It is not clear of this is an appropriate and sustainable way to expand the vegetation area in the dryland ecosystems of China. In this study, we applied hydro-ecological models to assess the provision of water resources and predict ESS in the largest inland river basin in China. The first conclusion of our study was that the current groundwater and surface water resources are not sufficient to 305 support the sustainability of ecosystems in their current state, and therefore the further expansion of natural vegetation is not possible. The simulation results indicated that riparian forests will remain prone to deterioration if no significant changes are made to guarantee an ecological water supply along the river oases.
The Tarim River has the typical seasonal characteristics of a dryland river. Summer floods dominate the river runoff each year by supplying more than 3/4 of the total water volume. Surface runoff is quite low in spring and winter. Most of the 310 water resources are consumed in the upper and middle reaches. The discharge in the downstream is less than 1/7 of that in https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. the upstream. A higher percentage of the total water is consumed in the upper and middle reaches in dry years, which can be very harmful for the ecosystem in the lower reaches. On the other hand, groundwater levels decline substantially from summer to winter, and from the upper to lower reaches. In SC4, the groundwater level can be lower than 10 m in winter.
Overall, water stress sets variable and challenging limits on agricultural development and ecosystem rehabilitation in this 315 arid region.
Ecological flooding displayed a decreasing trend from upstream to downstream oases. Without additional ecological water, it is very difficult for riparian trees to regenerate in the middle and lower reaches. Irrigation water consumption in SC1 was much higher than in the other three downstream sub-catchments combined. In recent years, the Chinese government and social communities have made huge efforts to plant trees and undertake ecological restoration, such as "transforming 320 cropland into grassland" and "returning cultivated land to forest" in the Tarim River Basin. However, the current ecological water resources (including surface ecological water and groundwater) are not sufficient to maintain the riparian green belt in future years. Based on the simulated land use changes, more than 1/4 of the total vegetation area will be degraded by 2050 without the provision of additional ecological water resources.
The ESS scenarios evaluated in the DSS could lead to the provision of sustainable socio-economic and ecological benefits. 325 The DSS enabled the interpretation of a complex range of environmental factors and the results were used to evaluate the current situation and propose creative solutions. The reference scenario indicated a 12.8% increase in the total ESS value, but a decrease in grassland provisioning and regulating services was also noticeable. The scenario with a reduced cotton area resulted in steadily increasing ESS values. This scenario was expected to have little effect on farming incomes, due to the increased sales price of crops in future years. The drip irrigation and ecological flooding scenario has the best performance in 330 terms of increasing the total ESS value in the near future. Cotton production, farming incomes, and riparian forest provisioning, regulating, and supporting services were all significantly improved in this scenario. In the far future (i.e., 2050), the soil salinization problem should be seriously considered to avoid the deterioration of ESS in the oases.
Data availability. Data available from SuMaRiO project website: https://www.sumario.de/. The DSS is also free available on 335 request.
Competing interests. The author declares that there is no conflict of interest.
Author contribution. MD, XC, AB and RY initiated the research project. YY, PH and MH established the hydro-ecological 340 models. YY wrote the manuscript. HZ and RY performed revisions. JL, FZ and RY provided ideas and data in the research area. LS and YG contributed preparation of data sets and modeling assistance. All the co-authors provided discussion and insights into the research. https://doi.org/10.5194/hess-2020-80 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License.