*3.5. Baseflow Drought Determination*

Due to the hydrological drought with a higher accumulation period [63] and to provide insights for water planning and drought alerts for other basins facing water shortage events, the annual baseflow anomalies were implemented to determine hydrological droughts in historical and future climate conditions.

#### **4. Results**

#### *4.1. Baseflow Estimation*

The meteorological outputs in the GCMs were extracted as the inputs to the SWAT model to predict the streamflow in the future climate change, and then, a well-revised Lyne–Hollick method was used to implement the baseflow separation. Overall, the SWAT model had a good performance in both the calibration (Figure 3a) and validation (Figure 3b) stages to simulate the streamflow on a long-term scale (e.g., with R<sup>2</sup> > 0.7). In addition, the annual mean baseflow in the calibration and validation of the SWAT model also is shown in Figure 4.

**Figure 4.** Relationship between the simulated and observed annual mean baseflow in the historical period. The black and red error bars represent standard errors in simulated and observed annual total baseflow. The black line is the 1:1 line.

The baseflow time scale of predicted streamflow and recession constants are shown in Table 3. The K and α are ranged from 62 ± 6 days (SD) and 0.98 ± 0.002 (SD) 1/day in the two future climate scenarios, respectively. For the CSIRO and FGOALSg2 models, the prediction of streamflow under the two scenarios was very close. For the highest streamflow condition, the MIROCS baseflow results were much higher than those of the other two models.

**Table 3.** Recession analysis derived from three general circulation models and two scenarios for future climate conditions (2010–2054).


#### *4.2. Detection of Baseflow Changes*

All three models showed an insignificant increasing trend in both scenarios before 2020 in the future period (Figure 5). From the perspective of changing points, there was a similar and/or general pattern over a long future period; nevertheless, the numbers of changing points were different. In 2020, 2026 and 2034, changing points occurred for CSIRO and FGOALSg2 in both scenarios and for MIROC5 in the RCP8.5 scenario. After 2020, all three models showed an insignificant decreasing trend in both two scenarios. Specifically, in the FGOALSg2 model, there was a significant decreasing trend, and in the MIROC5 model, this trend occurred after 2049.

**Figure 5.** Mann Kendall test statistics for three GCMs in two scenarios (RCP4.5 and RCP8.5). UF is the sequential values of a statistic under the random hypothesis; UB is the reversed UF data statistic series. The positive and negative values indicate the increasing and decreasing trend. The intersections of UF and UB present the changing point.

The baseflow derived from the observed daily streamflow (Figure 6) showed a changing point in 1970. Before this year, baseflow showed an insignificant trend. However, after this year, there was a decreasing trend, both in 1977–1983 and after 1995.

**Figure 6.** Mann Kendall test statistics for the baseflow separated from historical observed daily streamflow data. The abbreviations are the same as Figure 5.

#### *4.3. Quantitative Baseflow Analysis Combining Historical and Future Climatic Conditions*

The baseflow exhibited a decreasing trend in the long-term periods (all *p* ≤ 0.005, see Figure 7). Herein, we first calculated the baseflow anomaly for the entire time series and then added the regression line for each GCM using local polynomial fitting. Specifically, CSIRO had a relatively more variation compared to the other two GCMs in both climate scenarios. Despite the trend with fluctuations, the three GCMs showed a similar performance in the two climate scenarios.

**Figure 7.** Baseflow anomaly plot from the entire streamflow time series in three models and two scenarios. The blue line is the linear regression line. The red vertical dashed line divides the time series into historical and future periods.

#### **5. Discussion**

#### *5.1. Baseflow Trends in Historical and Future Climate Periods*

The baseflow separation algorithm used in this study was derived from the revised version of the Lyne–Hollick algorithm. The outcomes of this method were more reproducible than the traditional methods (e.g., graphical approaches and empirical function [40]), thus this approach would greatly reduce the uncertainties of baseflow estimation. As baseflow was not directly measured under experimental conditions and was often estimated from the original total streamflow [40], in addition, the digital filter combined the recession analysis with more physical meanings containing more catchment-specific groundwater drainage characteristics [44] and provides a robust tool to decrease uncertainties in baseflow estimation [64].

The climate scenarios provided a robust tool to project the water balance of the catchment [65], and detecting trend characteristics was beneficial to understanding the hydrological variability at a long-term scale. In this study, the MK test was adopted to detect baseflow changes under future climate conditions (Figures 4 and 5). A declining baseflow trend was predicted for future climate scenarios. The baseflow change point years were 1970 and 1990. These years are not consistent with the streamflow change points reported by Zhan, et al. [33]. It was demonstrated that runoff had a decreasing trend in this basin after 1990 due to human activities, and the changing points of streamflow lagged the baseflow changes by about 20 years. However, the baseflow change points were in the range of the streamflow change points in another catchment on the Loess Plateau. Herein, the streamflow change points for different sub-catchments ranged from 1970 to 1990 [66]. As a delayed water resource, baseflow provides water to the land surface and sustains ecological health under dry spells.

Projections of baseflow and trend analysis are important to prevent and palliate drought losses on the catchment and regional scale [67]. Analyses of climate variability and baseflow improve our understanding of the effects of drought on environmental protection [4]. A drier trend has been reported for most areas of China based on PDIS (the Palmer Drought Severity Index) [67]. The degree of drought is characterized by a high frequency and has a long-term effect on hydrological connections in the WRB. Baseflow characteristics were used to evaluate hydrological droughts because baseflow is relatively steady and can represent catchment water storage [68]. Quantifying the impacts of climate variability on baseflow can provide insights for future water-resources plans [4]. Yang, et al. [41] showed that baseflow recovery had a longer lag than streamflow recovery across 130 unimpaired catchments in eastern Australia. Further, it has been reported that the hydrological cycle is intensified with changes in global mean precipitation in GCM projections [69]. This means that dry areas with limited water may become much drier.

#### *5.2. Variability of the Baseflow Index*

The BFI is an important hydrological indicator representing the water flow from groundwater/delayed resources to streamflow. It contains a lot of information on catchment characteristics [70,71], which reflects the holistic attribute of baseflow and terrestrial water balance [72]. The relationship between total baseflow, streamflow, and the baseflow index was demonstrated in Figure 8. To address the total baseflow contribution to streamflow, we also assessed the BFI for the historical observed and the simulated results for the three GCMs (Figure 9). There was an increasing trend in the BFI in the long-term climatic period. This means that the role of baseflow was remarkably strengthened in the sustenance of local water in this catchment. Compared to baseflow yield, the BFI is a relative ratio that varied from 0.42 to 0.49 and averaged 0.45 in our study. This means that the contribution of baseflow from groundwater storage or delayed sources accounted for 45% in the WRB from the perspective of future climate conditions in GCM projections. The magnitude of baseflow was very similar in the three models. Nevertheless, streamflow showed relatively greater variations. This confirmed that the baseflow is a relatively stable flow that sustains the terrestrial hydrological ecosystem [73].

**Figure 8.** Total baseflow and streamflow for each GCM. Numbers in bars are the baseflow index. Red bars are the total baseflow, and blue bars are the total streamflow.

**Figure 9.** Variation in the baseflow index derived from historical observed streamflow data (1960–2010) and the ensemble means of three general circulation models for two scenarios (simulated from six models) for the future climate changes (2010–2054). The line and equation represent a linear regression.

Additionally, to clarify baseflow and streamflow trends, the relationships of historical data and/or projected streamflow and baseflow from the three models under two scenarios were also assessed (Figures 4 and 5). The response of baseflow to streamflow had a relative laggy time interval. This may be related to the increasing degree of anthropogenic activities in this basin due to the heavy exploitation intensity of groundwater resources. Singh, et al. [72] reported that groundwater abstraction significantly influenced flow regimes, with higher baseflow under constrained pumping conditions. Further, the effects of baseflow increase vary among river reaches, and baseflow and stormflow increases have relatively greater impacts on downstream areas by increasing flow volume [16]. Estimating other anthropogenic pumping effects is a meaningful way to assess the baseflow response to local hydrological variations.

#### *5.3. Factors Influencing Baseflow Variations*

It has been reported that climate change and anthropogenic activities are drivers of groundwater storage [74,75]. The baseflow yield is associated with the interactions between climate variability and vegetation changes [66,76] and would be influenced by a variety of catchment physical factors [72]. It is characterized by seasonal precipitation variations, i.e., from June to September, which creates the summer-dominated baseflow feature in this basin [1]. Furthermore, land-use change has affected 50% of the area on the Loess Platea [66]. This directly influences streamflow and leads to changes in baseflow. The basin covers three geological classes. Land use was predominantly agricultural in the long term. However, due to the widely distributed loess-deposition areas is in this region, extensive agricultural development causes heavy soil erosion and water-conservation issues [77]. To sustain the water quality and supply of the WRB, the government has taken measures to prevent ecosystem recession (e.g., soil-conservation measures [78]). This should lead to delayed surface runoff and increase the baseflow in small catchments [60]. However, in dry seasons on a long-term scale, the baseflow should be reduced by the loss of groundwater through more plant evapotranspiration. This is associated with vegetation-type changes from grass/bare land to the forested area [79]. Additionally, this complex effect is also influenced by other potential conditions such as topography. For example, Li, et al. [80] showed that the topography plays a paramount role in low flows (flow magnitudes ≤ *Q*75%) in snow-dominated catchments.

The effects of anthropogenic activities associated with agricultural production also strongly control the water cycle in catchments [35]. It has been shown that the plantation intensity on agricultural land reduces downstream water availability [76]. Irrigation is an important factor influencing groundwater processes [29], leading to variations in baseflow [8]. The WRB is the main agricultural region, with large irrigation areas responsible for the food production for the regional population. To maintain living standards and sustain ecological health, the water demands have been increased for decades, and groundwater pumping supports much of the municipal water demand. Additionally, from the perspective of water depletion, agricultural development, and ecological recovery projects were all needing a large amount of water, including surface water and groundwater, it would create a baseflow shortage event for the Loess Plateau. For example, large-scale afforestation may exacerbate baseflow conditions as evapotranspiration increases through the amplification of leaf area and rooting depth [2,81]) for the catchment with constant precipitation input. This impact on baseflow variations would be amplified by climatic variations in this basin.

#### *5.4. Implications of Baseflow Droughts*

To provide insights for water planning and drought alerts for other basins facing water-shortage events, the annual baseflow anomalies were implemented to determine the hydrological droughts in historical and future climate conditions (Figure 7). It is noted that the baseflow has an apparent decreasing trend overall. Specifically, there was a relatively richer baseflow in approximately 2035. However, there was a lack of baseflow in 2041–2050, leading to a prolonged impact on the hydrological cycle (e.g., baseflow hydrological droughts) in the long term (~10 years).

It has been reported that, when disentangling climatic effects (e.g., precipitation) on hydrology, the uncertainties were much larger in the high-emission scenario RCP8.5 than the relatively low-emission scenario RCP4.5 [52]. In the baseflow estimation in the FCP, there was no remarkable difference between higher- and lower-emission scenarios, and the annual baseflow anomalies were very similar (Figure 7). The uncertainties of this study were likely associated with coarse temporal or spatial resolution and systematic

errors derived from GCMs [67]. Besides, the baseflow relies upon runoff estimates in the model, confined by temperature and precipitation [82]. The variations of the climate phenomenon of wet-getting-wetter and dry-getting-drier [83] also influence baseflow changes in the catchment.
