Annual, seasonal, and monthly baseflow trend in an arid area in Loss Plateau, China

With growing concerns regarding the importance of baseflow, researchers from the USA and worldwide have investigated trends in baseflow in different streams (Ahiablame et al. 2017; Singh et al. 2019; Ayers et al. 2021; Cheng et al. 2021). A baseflow index, BFI, refers to the portion of baseflow in streamflow and has been used widely (Bloomfield et al. 2009; Ahiablame et al. 2013). BFI values for different basins in Texas varied between 0.140 and 0.730, with an average value of 0.45, indicating that baseflow contributed to 45% of long-term streamflow. The long-term BFI ranges between 0.20 and 0.96, with an average of 0.53, based on baseflow and total streamflow data from 482 gauged locations across New Zealand, suggesting that baseflow accounted for 53% of New Zealand's long-term streamflow (Singh et al. 2019). Except for the Yellow River, winter baseflow variations in 14 main Eurasian rivers revealed an average growth ratio of 53.0% over the past century, from 1879 to 2015 (Qin et al. 2020). A statistical modeling approach was built to analyze how climate system changes have affected observed monthly baseflow data at 3,283 United States Geological Survey (USGS) gauges during the past 30 years (1989–2019). It demonstrated that baseflow trends and their affecting variables varied by region and month (Ayers et al. 2022). Clarifying portion and changes of baseflow variations provides valuable information on guiding the sustainability of regional water resources.

Significant changes in baseflow and BFI have been seen globally, and numerous researches have been conducted on the factors influencing baseflow and BFI globally. Climate change, particularly increases in precipitation and temperature, has been identified as an essential factor contributing to the increase in baseflow (Tan et al. 2020). Changes in evapotranspiration (ET), infiltration, land use, human activity, geography, and soil conditions could also contribute to the baseflow changes (Bloomfield et al. 2021; Ayers et al. 2022; Briggs et al. 2022). However, baseflow in different regions might show feedback to various factors, and regional analysis is necessary to identify the influencing factors.

In the Loess Plateau region of China, unique geographical, climatic, and soil conditions exert a significant influence on the behavior of baseflow. Therefore, conducting research on baseflow in these areas can unveil the impact of local characteristics on hydrological processes, thereby providing targeted recommendations for regional water resource management. However, despite the evident importance of baseflow in arid regions, research on baseflow and its trends in the Loess Plateau region remains limited. Consequently, investigating the behavior and impacts of baseflow in these areas can bridge this knowledge gap and offer practical guidance to local governments and research institutions. In the present study, we chose the Xiliu River Basin in Inner Mongolia to study its baseflow changes through a close inspection. The primary goals were to: (1) Investigate the hydrological characteristics and temporal variations of baseflow within the Xiliu River Basin; (2) Quantify annual, seasonal, and monthly fluctuations in baseflow over two distinct time periods and identify the change points; (3) Examine the potential influencing factors and their effects on baseflow dynamics in the basin. By addressing these objectives, our study aimed to shed light on the dynamics of baseflow in the Xiliu River Basin, contributing to a better understanding of this critical component of streamflow and its implications for water resource management in the region.

The Xiliu River, one of the 10 reaches that drains to the Yellow River on the right bank, is located in Inner Mongolia, shown in Figure 1. It originates from Ordos, a city in Inner Mongolia, with an altitude of 1,551m and a total length of 106.5 km, covering an area of 1,356.3 km². There are three major landscape feature classes from the southern Loess hilly and gully region to the middle Kubuqi desert and the north alluvial plain area where it drains to the Yellow River. The upstream Loess hilly and gully region covers 876.3 km², accounting for 64.6% of the total drainage area (Dang et al. 2019).

The area experiences long, cold winters and short, hot summers in a semi-arid continental environment. The yearly average temperature is approximately 6 °C, with the highest temperature recorded at 40.2 °C and the lowest at −34.5 °C. The annual rainfall ranges from 240 to 360mm, and the mean wind speed is 3.7m/s, with considerable potential evaporation of 2,200mm per year. The amount of rain and runoff varies throughout the year. The flood season runs from June to September and receives most of its heavy rain, contributing to 82 and 66% of annual precipitation and runoff, respectively (Xiao et al. 2020).

The daily streamflow data for the Longtouguai hydrological gauge, located at the outlet of the basin, were collected from the Hydrological Yearbooks of China (http://loess.geodata.cn). The data are subjected to rigorous quality control before being released by the Ministry of Water Resources of the People's Republic of China. Daily precipitation and temperature data were collected from the National Meteorological Administration of China (https://data.cma.cn/). However, there is a data gap due to historical reasons, and we selected the periods from 1981 to 1990 and 2006 to 2005.

The climate data used to describe the influencing factors are shown in Table 1. We chose PET (potential evapotranspiration), precipitation, HI (humidity index, HI = P/PET), and mean temperature as the climate factors based on the previous studies (Segura et al. 2019; Tan et al. 2020; Ayers et al. 2022). We used the Penman–Monteith equation to calculate the PET. Other topographical, land use, and basin factors may also change the baseflow significantly. However, these indicators did not change significantly on a monthly or seasonal scale; therefore, they are not considered in our study.

We used linear regression analyses and the Mann–Kendall non-parametric test (Li et al. 2017; Adib & Tavancheh 2019) to identify the annual, seasonal, and monthly baseflow and BFI trends within two periods (1981–1990 and 2006–2015). A p < 0.05 (the 95% confidence level) was considered to show significant trends in baseflow and BFI. We also used the linear regression between changes in baseflow (BFI) and PET, precipitation, HI, and temperature to identify their correlation. Additionally, we applied a series of statistical tests and analyses to investigate baseflow and BFI patterns. We employed the Pettitt test to detect change points or abrupt shifts in these series. Subsequently, as part of our trend analysis to identify the factors influencing baseflow and BFI, we utilized the Durbin–Watson (DW) Test to assess first-order linear trends and autocorrelation. Additionally, we employed the Lagrange Multiplier (LM) test to detect second-order autocorrelation in the data.

In the equation, ‘sgn’ represents the sign function, and ‘n’ is the length of the hydrological time data sequence. The values ‘x_j’ and ‘x_k’ correspond to the actual measured data values of the variables.

The sign of the Z statistic obtained from the M-K test represents the direction of the trend. At a given confidence level, if the Z value is positive, it indicates an upward trend; if the Z value is negative, it suggests a downward trend. If the absolute value of the Z statistic is greater than the threshold for the confidence level, it signifies that the time series data exhibit statistically significant trend behavior.

A change point at time t is deemed statistically significant at a 95% confidence level if p ≤ 0.05. The variable T is considered a pivotal transition point, leading to the division of the original sequence into two distinct subsequences. The identical approach is utilized to ascertain the presence of any potential boundaries. Hence, it is possible for many change points to arise, necessitating a comprehensive investigation in order to accurately identify and determine the final change points. This methodology has a high level of efficacy in identifying alterations in means at designated endpoint locations.

The DW statistic is only applicable for first-order autocorrelation testing and is not suitable for testing higher-order autocorrelation (Evans & King 1985). The LM autocorrelation test is a statistical test used to detect and assess the presence of autocorrelation in data. It evaluates whether there is a relationship or pattern of dependence between values at different time points within a time series (Dariane et al. 2018). This test involves the use of statistical models and calculations to determine the significance of such autocorrelation. In essence, the LM test helps determine if data points are not independent and exhibit patterns over time (Wang et al. 2005). The LM test is conducted through an auxiliary regression, and the specific steps are as follows (Anselin 1988).

The discrimination rule is as follows:

• If ⁠, the null hypothesis of no autocorrelation is accepted.

• If ⁠, the null hypothesis of no autocorrelation is rejected.

The BFI, the proportion of the baseflow to the streamflow, is used to study the yearly, seasonal, and monthly variable features of the baseflow in the Xiliu River Basin. Natural years and months were chosen for annual and monthly analysis. Based on the meteorological characteristics of the research area, the months of March through May were classified as spring, June through August as summer, September through November as fall, and December through February as winter.

As stated in section 2.4.2, the Mann–Kendall (M-K) method may not be robust enough in small sample situations and can be susceptible to the influence of extreme values. When dealing with a short time series or a significant amount of missing data, the reliability of the results may be compromised. Therefore, we employ the Pettitt test to identify change points for annual baseflow and BFI. The research findings indicate that there was a significant change point in the baseflow in 2007, which could potentially be attributed to climate change, alterations in precipitation patterns, or other influencing factors. Simultaneously, the BFI did not reveal any significant change points, showing a more stable BFI result.

Similarly, we applied the Pettitt test to detect change points in seasonal baseflow and BFI, and the results are presented in Table 1. Based on the information presented in Table 2, it shows that the studied area did not display any discernible and statistically significant patterns of change points in terms of seasonal baseflow and BFI throughout various seasons. Regarding baseflow, a transition point was identified during the autumn season of 2008 and the winter season of 1985. Nevertheless, no significant alterations were detected during the spring and summer seasons. Regarding the BFI, alterations were seen throughout the spring and fall periods of 1990 and 2006, respectively. No significant differences were seen in the summer and winter seasons. Furthermore, it is crucial to acknowledge that the temporal distribution of change points for baseflow and BFI did not exhibit a consistent pattern. The findings suggest that the fluctuations in seasonal baseflow and BFI within the examined region exhibit a level of intricacy and unpredictability. The intricate nature of this phenomenon can be ascribed to the wide range of hydrological systems and environmental circumstances, resulting in the emergence of change points during different years across multiple seasons and indices.

Moreover, June, July, August, and September were characterized as flood seasons, contributing 82% of the total yearly streamflow. And rainfall is concentrated in flood season, mainly in heavy rain, which provides limited infiltration and baseflow. Therefore, although the baseflow in these months is substantial, the proportion to the streamflow in these months, that is, the BFI, will vary with the difference in annual precipitation and streamflow, resulting in a significant difference each year.

Though baseflow in the three seasons of spring, summer, and fall exhibited a downward trend, all passed the M-K mutation test. The monthly baseflow was not, however, in the same situation. In January, the baseflow displayed an upward trend and passed the 95% confidence level test in 2006 but showed a downward trend in 2012. BFI also showed a significant upward trend but didn't pass the 95% confidence level test. In February, the baseflow showed an inclining trend from 1985 and then declined from 2012 without passing the 95% confidence level test. BFI increased in 2012 and then decreased. Like February, the March baseflow change trend shows a significant downward trend after 1985 and a significant upward trend for the BFI value since 2009. Baseflow in April fluctuated but finally demonstrated a statistically significant downward trend, passing the 95% confidence level in 2012.

In contrast, BFI displayed no statistically significant upward or downward trend. In May, both baseflow and BFI fluctuated. Baseflow in June exhibited a general downward trend, with a notable downward trend after 1990 and a very significant upward trend of BFI since 2008. In July, changes in baseflow were identical to those of June, indicating a significant downward and upward trend from 2011. BFI showed a non-significant upward trend. Baseflow in August demonstrated a downward trend, whereas BFI demonstrated an upward trend with no significant increase. Baseflow and BFI exhibited, respectively, downward and upward trends in September, but the trend for BFI was not significant. Baseflow and BFI in October behaved like September, with no significant trend for BFI. Baseflow demonstrated a significant downward trend in November, whereas BFI demonstrated no apparent upward trend. Baseflow exhibited a negligible downward trend in December, while BFI exhibited a downward trend.

To conclude, the baseflow in all the months showed a downward trend, and except for February, May, and December, this trend was evident. BFI, however, showed a different trend. It showed no statistically significant upward trend in February, April, May, July, August, September, and November. The remaining months exhibited a significant upward trend except for December.

Statistical analysis of monthly baseflow and BFI was conducted using the Pettitt test, and the results are presented in Table 3. When the M-K test results indicated only one change point, the findings from the Pettitt test were generally consistent with the M-K results, shown in Figure 10 and Figure 11. Table 3 shows that baseflow has fluctuating change points throughout the months. Some months exhibit no detectable change points, whereas others exhibit significant structural alterations. January and February have change points in 1987 and 1985, indicating that the baseflow was significantly altered in those years. In March, April, May, July, August, October, and December, there are no change points in the baseflow, indicating that these months are relatively stable. In 2008 and 2010, baseflow change points occurred in June and September, indicating significant alterations during these months.

Similar to baseflow, BFI exhibits change points with irregular patterns across months and years. The change points for March, February, and June are 1990, 1989, and 2006, respectively, indicating that these months experience significant variations. In April, May, July, August, October, November, and December, the BFI demonstrates no change points, indicating relative stability.

In conclusion, the monthly and yearly variations in baseflow and BFI are complex and irregular, with no distinct periodic or consistent patterns. This may be a result of the diversity of hydrological systems and environmental conditions, which causes change points to occur in various years and months. Understanding the fundamental causes of these changes and their implications for hydrological processes requires additional research.

Many studies have confirmed that the baseflow was significantly influenced by PET, precipitation, HI, and temperature since all these factors contribute to the water cycle changes (Li et al. 2013). In their study, Wilby et al. (1994) employed a hydrological model to simulate the correlation between climate change and baseflow in a watershed. The findings indicated that the impact of climate change on baseflow was most pronounced during periods of increased precipitation. The study conducted by Heydari Tasheh Kabood et al. (2020) examined the impacts of climate change on the variability of stream flow within the Urmia Lake basin. The findings indicate that precipitation, which is a significant climatic component, plays a crucial impact on the variability of streamflow. However, at the same time, many scholars have also discovered that the generation of baseflow is associated with various other factors, such as soil class, catchment properties, and so on. Consequently, despite a large body of literature, there is still no general theory capable of interpreting the relative influence of climate and catchment controls on baseflow generation.

Some scholars have conducted researches on baseflow in the Loess Plateau region. In a study conducted by Wu et al. (2019), it was determined that climate variability emerged as the primary driver of alterations in the ‘BFI’ across 11 catchments situated within the semi-arid Loess Plateau. Notably, the impact of climate on the ‘BFI’ was shown to surpass that of human activities, highlighting the substantial influence of climatic factors in this context. Yan et al. (2023) found that precipitation and hydrogeological condition (e.g., Loess thickness) were likely to dominate the spatial heterogeneity of baseflow in Loess Plateau, while PET, population, sub-surface runoff, normalized difference vegetation index (NDVI), and soil water were responsible for the decrease in baseflow with contributions of 27, 21, 17, 16, and 16%, respectively. However, we did not observe this finding in our study area. The possible reason is that our period is only 20 years and is not long enough to reflect the trend. To confirm this assumption, we conducted the linear analysis of PET, precipitation, HI, and temperature and found that none of these factors showed a statistically significant increasing or decreasing trend. This might explain why the climate change in 1981–1990 and 2006–2010 was not significant either and may not be the contributing factors to explain the changes in Baseflow and BFI.

Another possible reason might be that the study area is located upstream Yellow River basin, and many check dams have been built to mitigate water and sand loss (Wei et al. 2017). It has been reported that there were 113 check dams in the study area by the end of 2015 (Liu et al. 2020). Check dams profoundly influence the hydrological process of the basin by obviously reducing the runoff and flood peak discharge and flattening the flooding process (Guyassa et al. 2017). The check dams may have changed the baseflow by changing the runoff process.

In this study, streamflow records during 1981–1990 and 2006–2015 were used to calculate and analyze changes in annual, seasonal, and monthly baseflow and BFI. Connections between baseflow (BFI) and the climate factors of PET, precipitation, HI, and temperature were also discussed. The annual baseflow showed a statistically significant decrease, while no significant changes in annual BFI were found. The distribution pattern of the baseflow was similar, and BFI varied significantly across the four seasons. Baseflow and BFI were stable during the winter, but during May and June, baseflow was unstable while BFI remained stable. Baseflow for the spring, summer, and fall seasons exhibited a downward trend and passed the M-K mutation test. BFI showed an overall upward trend in spring and summer and a downward trend in fall and winter. During 1981–1990, January and December exhibited a slight variation in baseflow, while January, May, and June exhibited a slight variation in BFI. From 2006 to 2015, baseflow was stable, with limited fluctuations in January, February, March, April, May, November, and December.

The correlations between baseflow and PET, precipitation, HI, and temperature are neither statistically significant nor strong. The baseflow did not change with the PET, precipitation, HI, and temperature change from the annual, seasonal, and monthly time scales. The possible reason is that our period is only 20 years and is not long enough to reflect the trend. Another possible reason might be that check dams in the study area profoundly influence the baseflow by changing the runoff process.

This research was funded by Basic Public Welfare Research Program of Zhejiang Province, grant number LZJWD22E090001; Major Science and Technology Program of Zhejiang Province, grant number 2021C03019; Projects of Open Cooperation of Henan Academy of Sciences (220901008) and The Key Scientific Research Project Plan of Colleges and Universities in Henan Province, grant number 21A170014.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.