Using the Mann–Kendall test and double mass curve method to explore stream flow changes in response to climate and human activities

Climate change is an important factor in the hydrological cycle. When temperature and precipitation are altered, stream flow regimes can change significantly and lead to floods and droughts (Dai 2013; IPCC 2014). Stream flow regimes can also be modified considerably by intensive human activities such as land use change, irrigation, deforestation, and dam construction (Zhao et al. 2012; Liu et al. 2017). Dam construction is an important human activity that supplies water for various uses (Song et al. 2015). Despite their benefits, the regulatory services of dams have led to environmental damage including reduction of river connectivity (Stanford et al. 1996), alteration of hydrological processes (Ouyang et al. 2011), changes in river morphology (Wang et al. 2011), reduction of water quality (Kummu & Varis 2007), and modifications in the structure of river ecosystems (Poff et al. 2007; Hu et al. 2008). Dams have also resulted in socio-economic benefits such as reduction of water shortages, flood control, and generation of energy (Kummu & Varis 2007).

At a global scale, climate variability is one of the most important factors affecting stream flow. At a local scale, human activities such as land use change and dam construction are significant (Piao et al. 2010; IPCC 2014). For example, Zhang et al. (2015) showed that human activities were the main factor causing reduced stream flows of 4.34% and 3.17% in the north and west of Jiulong River watershed, China, respectively. Some studies found that water pumping for irrigation led to decreased downstream flow (Kustu et al. 2010; Zhang et al. 2013). Based on a previous study in the Tajan basin, Iran, Pirnia et al. (2014) reported that land use change and increased area of irrigated lands could lead to a considerable decrease in downstream flows. Yuan et al. (2015) reported that both climate variability and human activities led to stream flow changes, and dam construction resulted in decreased water levels. Wang et al. (2015) indicated that among human activities, the contribution of regulated reservoirs was more important than land use change impacts on stream flow variation in dry and wet seasons. As shown, human activities such as dam construction generally resulted in localized stream flow reduction downstream. The hydrological responses of basins to climate change and variability however, are different throughout the world. River flow rate is determined by a combination of variables such as temperature, precipitation, evaporation, snowmelt, and water use. In some regions, significant changes in temperature and precipitation have resulted in increased base flow and annual mean discharge and flooding (Ahn et al. 2016), while in other areas this has caused decreased base flow and annual mean discharge, and increased droughts (Dai 2013; Wang et al. 2015; Zhao et al. 2015).

One of the most important effects of dam construction is the natural river flow regime change and flow reduction downstream, which are influenced by reservoir performance. Song et al. (2015) referred to the regulative function of reservoirs in daily, seasonal, annual, and multi-year time scales that are different based on the size and purpose of dam construction. Ren et al. (2002) explained that reservoir construction could also lead to an alteration of the monthly runoff distribution during one year.

To analyze the impacts of climate change and human activities on stream flow alteration, most studies have applied hydrological models such as SIMHYD (Wang et al. 2006), VIC (Chang et al. 2015), or SWAT (Zhang et al. 2012, 2017). Although the models are useful and versatile instruments to analyze stream flow changes, the results of these studies have many uncertainties and require significant amounts of data (Wang et al. 2015). Furthermore, the uncertainty in the selection of hydrological models for in-depth analysis (for instance, as required for this study) significantly influences the simulation results regarding possible attribution (Onyutha 2016a). It has been shown that statistical methods are useful options to determine runoff response to some drivers such as climatic and human factors (Mu et al. 2007; Zhang et al. 2014). Long-term time series of hydrological data show temporal oscillations of stream flow caused by climate and land cover changes. Analysis of these changes can recognize the runoff variations caused by both climate change and human activities (Zhang et al. 2011). Therefore, to better evaluate runoff changes caused by different drivers, particularly human activities, investigation of runoff alterations at intra-annual time scales is useful (Ling et al. 2014; Wang et al. 2015; Zhang et al. 2016).

The Tajan River basin is considered one of the most important basins in northern Iran. It is a main source of water and plays a vital role in the irrigation of agricultural lands, societal development, and ecological and environmental protection in the basin. Shahid Rajaei dam, one of the largest dams in Iran, was constructed on the Tajan River because of the basin's importance to supply water for different uses and to satisfy increasing water use in agricultural lands. Agriculture in this area takes place primarily in the spring (April to June) and summer (July to September) where the dominant crop is rice with high water consumption requirements. Although some studies have discussed various effects of the Shahid Rajaei dam (Sharghi et al. 2011) and reported on land use changes (Kelarestaghi et al. 2006) and climate variability (Pirnia et al. 2014), none have investigated the impacts of human activities (especially dam construction) and climate variability on flow regime changes. Stream flow changes influenced by climate variability and human activities have not been studied in the Tajan basin; most studies have focused on the evaluation of human activity effects (dam construction impacts) on water quality and the environment. In this study, the following were explored: (i) annual, seasonal, and monthly changes of hydro-climatological variables during 1971–2010 and (ii) runoff changes driven by climate variability and human activities.

The Tajan River is 143.75 km in length and is one of the longest rivers in Mazandaran, Iran (Figure 1). The river basin lies between 35° 56′ to 36° 17′ N and 53° 07′ to 53° 42′ E with an elevation varying from −26 m to 3,782 m above sea level. The drainage area of the Tajan River basin is approximately 4,006 km². In general, July and August are considered low-water months, whereas March and April are high-water months. For example, at the Kord Khail station (Figure 1), discharge in a high-water month is 30.9 m³/s and 2.21 m³/s in a low-water month (Pirnia et al. 2014). Precipitation decreases in the southern regions of the Tajan River basin (with an elevation higher than 1,500 m); these phenomena occur in all of the basins of northern Iran, which are located near the Caspian Sea. In the Tajan river basin, 55–60% of the precipitation occurs in autumn and winter and less than 45% is in the spring and summer. The highest and lowest average annual temperatures are 16.8 and 9°C, and occur at the Sari and Sangdeh stations, respectively. The highest evaporation values occur in July and August, and the lowest evaporation values occur in December and January. Shahid Rajaei dam (also known as the Soleyman Tangeh dam) is an arch dam built in 1999 on the Tajan River and is located 38 km south of the city of Sari. This dam was built for hydroelectric power production, flood control, and provision of water for industrial and agricultural uses (Meskar & Fazloula 2013). The basin area is 60% agricultural land (Khosrojerdi & Fallah 2016).

In this study, hydro-meteorological data from 1971 to 2010, including monthly, seasonal and annual precipitation time series, mean air temperature, evaporation and runoff were analyzed. Climate and hydrological data were collected from the Meteorological Organization and Regional Water Office in Mazandaran, respectively. Based on precipitation characteristics, the time series were described as winter (January–March), spring (April–June), summer (July–September), and fall (October–December). As precipitation varied in intensity and duration at different locations of the river basin, precipitation time series were interpolated using the inverse distance weighted method for receiving basin average values (Tian et al. 2016). Additionally, data quality control is important to correctly analyze time series (Arbabi Sabzevari et al. 2015; Olusegun Mayowa et al. 2015). The homogeneity of the data was investigated using the double mass curve method. For this purpose, the cumulative values of a station were plotted against the average cumulative values of other stations in the area. The results showed that the data were homogeneous. The location of 11 meteorological stations and one hydrological station (i.e., Soleyman Tangeh station located downstream of Shahid Rajaei dam) are shown in Figure 1.

Different statistical methods were used to analyze the time series trends. These methods are divided into two general groups of parametric and non-parametric methods, where the latter have more extensive applications (Xu et al. 2003).

The Kendall τ test is a non-parametric method for testing correlations (Kendall 1938). In this research, it is used to detect relationships among runoff and the climatic parameters of precipitation and temperature.

UF is the forward sequence, which has a standard normal distribution, and UB is the backward sequence and is calculated similarly to UF with an inverse time series. In the two-sided test, rejecting the null hypothesis shows an increasing (UF > 0) or a decreasing (UF < 0) trend. The intersection of the two curves with a significant trend of the series indicates an abrupt change point (Bao et al. 2012; Zarenistanak et al. 2014).

In this study, the double mass curve method (Searcy & Hardison 1960) was used to detect the contribution of effective factors to runoff changes. The double mass curve is a simple, visual and practical method, widely applied to study consistency of hydro-climatic data (Gao et al. 2013). This method is based on the relationship between precipitation and runoff in which a straight line is plotted on cumulative values of these two parameters. Lack of deviation from the line shows that human activities have not made noticeable contributions to the changes of stream flow and it is only influenced by precipitation. Significant deviation from the straight line plotted on the cumulative values of precipitation and runoff indicates that, in addition to precipitation, human activities – such as reservoir construction, land use change, and direct river withdrawals – also affect the runoff change.

Runoff values during 1999–2010 decreased considerably compared to 1971–1998 and the average runoff was 340 mm and 255 mm for the two periods, respectively (Figure 2).

Wet season runoff showed abrupt changes in 1991 and 1999 (Figure 3). The decreasing trend in wet season runoff was significant at the 0.05 confidence level after 1996 (Table 1). Since 1999, the decline in runoff during the wet season has been severe with a decreasing trend at the 0.01 confidence level. Similarly, the dry season runoff had an abrupt change in 1999 (Figure 4); however, it was accompanied by a significant increasing trend after 2001 (Table 1). The main cause of these changes was attributed to the construction of the Shahid Rajaei dam, which reduced annual runoff values (Figure 2) and caused the reversion of wet and dry season runoff between the 1999–2010 (post-dam construction) and the 1971–1998 (pre-dam construction) periods (Figures 3 and 4).

Considering the results of the Mann–Kendall change-point test and the year of dam construction, the runoff data was divided into pre-dam (1971–1998) and post-dam (1999–2010) periods. A correlation analysis among seasonal runoff, precipitation, and temperature was carried out using the Kendall τ test for these two periods (Table 2). The results showed that the runoff changes were less influenced by precipitation in the post-dam construction period when compared with the pre-dam construction period, indicating that changes in runoff values were strongly caused by dam construction. As shown, the relationship among seasonal runoff, precipitation, and temperature was significant at the 0.01 confidence level in the pre-dam construction period in the dry season. A high correlation among seasonal runoff, precipitation, and temperature in the wet season in the pre-dam construction period (1971–1998) was expected; however, the results of the correlation test showed that the relationship among these variables was not significant. This is mainly attributed to snowfall in winter (January–March) and runoff from snow melt in winter and April, which led to a reduction in correlation among runoff, precipitation, and temperature (Pirnia et al. 2014).

The effects of dam construction on runoff were also investigated for two periods of 12 years pre-dam (1987–1998) and post-dam (1999–2010) (Figure 5). In general, it was evident that the highest amount of runoff before dam construction was in the wet season with the highest values occurring in April due to snow melt and precipitation. The lowest amount of runoff before dam construction was during the dry season, especially during July to September. As shown in Figure 5, the annual distribution of runoff was very different pre-dam vs. post-dam construction. The highest amounts of runoff were shifted to the dry months while the lowest values of runoff were shifted to the wet months (October to April) due to dam construction. In the periods of pre- and post-dam construction, the mean runoff in the wet season was 192.4 mm and 88.6 mm and dry season runoff was 128.5 mm and 170 mm, respectively (Table 3).

The above results were supported by the Mann–Kendall and CRD tests (Tables 4 and 5). It was evident that the absolute values of all the standardized trend statistics of both methods were greater than the standard normal variate at 5%. Therefore, the null hypothesis H₀ (no trend) was rejected at the 5% level. The runoff of dry and wet seasons decreased by −0.31 mm/y and −1.11 mm/y, respectively, during the period of pre-dam construction (1971–1998). The decrease in wet season runoff was significant at the 0.05 confidence level. When data from 12 years after dam construction (i.e., 1971–2010) were included in the analysis, the results demonstrated that wet season runoff decreased by −2.42 mm/y, over two times the decrease in wet season runoff in the pre-dam construction period (1971–1998). There was a significant decreasing trend at the 0.01 confidence level while the difference of precipitation reduction between the periods was not important. The decrease in dry season runoff in the pre-dam construction period (−0.31 mm/y during 1971–1998) was transformed into an increase in runoff of 1.41 mm/y during 1971–2010 after the dam was in place. As shown in Figure 6, the construction of the dam changed the annual distribution of monthly runoff. Ren et al. (2002), Lopez-Bustins et al. (2013), and Vicente-Serrano et al. (2016) have referred to this subject previously.

Rainwater can be divided into three forms after its occurrence: stream flow, natural losses through evaporation, and non-natural losses driven directly and indirectly by human activities. Direct non-natural loss refers to water diversion from channel or reservoir to riverside lands and indirect non-natural loss refers to evaporation increases from higher reservoir water levels (Song et al. 2015). Kelarestaghi et al. (2006) indicated that land use changes in the Tajan basin included a significant increase in cropland area, which had occurred in recent years. In addition, Pirnia et al. (2014) concluded that the increased area of irrigated lands led to an increase in water consumption and, therefore, to greater transfers to riverside lands. Thus, non-natural losses of precipitation have directly increased in the Tajan River basin. The construction of a dam can also increase indirect non-natural losses of precipitation (Song et al. 2015). As a result, intra-annual evaporation values were investigated in the 12-year periods pre-dam and post-dam construction. In general, evaporation values after dam construction, particularly in April to August, were considerably greater than in the pre-dam construction period (Figure 6). Dam construction and higher reservoir water levels led to increased indirect non-natural loss. The rise in evaporation was considerable in April to August when reservoirs were at their highest volume capacity due to increased runoff caused by snow melt, spring precipitation, and storage of stream flow in the reservoir.

In general, changes in stream flow are due to both climate variability and human activities. To better analyze the contribution of effective factors towards runoff changes, the double mass curve method was applied. Analysis of runoff variations at an intra-annual time scale can be more important than on an inter-annual scale. The effects of climate variability and anthropogenic activities were therefore investigated in the dry and wet seasons.

The values of cumulative precipitation and runoff in the wet season deviated in 1991 and 1999 (Figure 7). These deviations showed that, in addition to precipitation, human activities also contributed to runoff changes. When compared with the wet season baseline period (1971–1990), it is evident that runoff deviation in the 1999–2010 period was more significant than deviation in the 1991–1998 period, indicating that dam construction resulted in a considerable reduction in runoff during the wet season after 1999. Deviation of the cumulative curve of precipitation and runoff in the dry season was also important in 1999, as it indicated that stream flow in the dry season was considerably increased due to dam construction.

Table 5 shows the contribution effects of both climate variability and human activities on runoff changes. According to observed and predicted values, the contribution of climate variability and human activities on the runoff decrease in the wet season (1991–1998) compared to the reference period (1971–1990) was 57.76% and 42.24%, respectively. Climate variability contributed 24.68% and human activities contributed 75.32% to the decrease in runoff of 116.55 mm between the 1999–2010 period and the baseline period (1971–1990). For runoff changes in the dry season, the effects of climate variability and human activities on the runoff were reversed. Consequently, climate variability and anthropogenic activities contributed −30.68% and 130.68%, respectively, to the runoff increase of 41.45 mm in the 1999–2010 period (Table 6). Based on these results, the runoff changes of the 1999–2010 period were mainly driven by human activities, specifically dam construction, which resulted in a noticeable increase and decrease of stream flow in dry and wet seasons, respectively.

Climate variability as one of the important drivers of stream flow changes has interested many researchers (Yuan et al. 2015). Climate variability may change stream flow through variation in meteorological variables, especially precipitation and potential evapotranspiration (Green et al. 2013; Tan & Gan 2015), and these may affect the performance and biodiversity of river ecosystems (Pall et al. 2011). Our results showed that the contribution of climate variability was greater than human activities on stream flow changes before dam construction, and that these changes were considerable in the wet season. Among meteorological variables, precipitation and temperature made the greatest contributions to stream flow changes in the Tajan River basin. Annual and seasonal precipitation and temperature changes in the Tajan basin have been reported. Sheydaeian et al. (2013) showed that annual precipitation had decreased by 8.9% and average annual temperature had increased by 9.5% in recent years. They also indicated a possible future reduction in precipitation of 32.4% and a temperature increase of 9%. At a seasonal scale, Pirnia et al. (2014) reported that precipitation and temperature had decreased and increased, respectively, in all seasons. They also showed that stream flow changes and distribution coincided with climate variability in different seasons upstream of the dam. Hence, it was concluded that climate variability had an important role in stream flow changes in the Tajan basin, especially before dam construction.

Decadal variations of annual/seasonal temperature, rainfall and runoff (1971–1980, 1981–1990, 1991–2000, and 2001–2010) are shown in Tables 7 and 8. Temperature increased, and rainfall decreased, from the first decade to the third decade. The lowest and highest amounts of annual runoff in relation to the long-term mean occurred during 2001–2010 and 1971–1980, respectively.

The decadal changes in runoff rather than the long-term average were consistent with the temperature increase and precipitation reduction for the first three decades studied, but there the lack of correlation between these three parameters was considerable in the 2001–2010 decade. It was expected that stream flow would increase as precipitation increased and temperature decreased in 2001–2010 compared with 1991–2000, but results showed that runoff decreased by 19 mm in 2001–2010 and was lower than the long-term average by 56 mm (Table 8). Based on the precipitation variability index (Figure 8), the lowest precipitation value was during 1991–2000 despite lower than average precipitation in the last decade (2001–2010). The seasonal scale also showed that the decadal changes rather than the long-term average of these three parameters had a better correlation in the first three decades. During the final decade studied, runoff, precipitation, and temperature changes had less correlation than in the 1991–2000 decade. This lack of correlation is more important in the summer and spring. One main reason for these results could be related to the extension of irrigated agricultural lands on river banks in recent years and the subsequent increased transmission of Tajan River water for irrigation.

As some abrupt change points occurred before dam construction, other effective drivers such as climate variability and human activities, including land use change, increased the area of irrigated lands and water transfer from the river to agricultural lands, and this also contributed to the decrease in runoff. Some studies showed that land use changes and water transfer to riverside lands were considerable in the study area (Kelarestaghi et al. 2006; Pirnia et al. 2014).

Kelarestaghi et al. (2006) reported that the extent of agriculture lands increased between 1967 and 2002, and particularly between 1994 and 2002 (Table 9). It is likely that expansion of agricultural lands continued in the subsequent ten years, namely, the 2001–2010 decade, and this caused the significant runoff reduction in this decade despite increased precipitation and reduced temperature. Pirnia et al. (2014) concluded that changes in spring and summer runoff were caused more by land use change and increased area of agriculture lands, while autumn and winter runoff was more influenced by climate change. In this study, the lack of runoff correlation with temperature and precipitation in the last decade (compared to 1991–2000) was related to increased extension of irrigated agricultural lands.

Although the Tajan River basin is important because its high stream flow contributes to valuable environmental conditions, and also because it is very useful for agriculture, mining, and industrial activities (Fallah & Farajzadeh 2008; Namin et al. 2013), there have been no useful quantitative evaluations of stream flow. Most studies in this basin have focused on water quality and the eco-environment effects of the dam (Aazami et al. 2015a, 2015b; Yousefi et al. 2013; Aazami 2017). The Tajan basin is significantly regulated by large and small dams. The large Shahid Rajaei dam was constructed because of increasing demands in the basin downstream, and it has an important role in equitable water distribution during the year. Agriculture in spring and summer (dry seasons) is important in most Iranian basins, especially in the north (including the Tajan basin) when precipitation and temperatures are low and high, respectively. Also, due to the considerable decrease of precipitation and increase in temperature, especially in upstream areas (Pirnia et al. 2014), and the considerable increase in irrigated agriculture (Kelarestaghi et al. 2006), stream flow values have decreased significantly. Thus, to supply water for different uses (especially agriculture), regulating stream flow is important. Construction and management of the Shahid Rajaei dam has led to increased and decreased water values in low-water and high-water seasons, respectively, and thus plays an important role in stream flow distribution during the year. Although the dam resulted in a balanced distribution of water during the year, its eco-environment impacts are considerable (Yousefi et al. 2013). When one considers the recent droughts and possible further reduction in precipitation and higher temperatures in the future (Sheydaeian et al. 2013), the significance of this dam to supply water resources for different uses, and especially agriculture, is important (Khosrojerdi & Fallah 2016).

According to the results of this study, stream flow changes were influenced by a combination of anthropogenic and climatic factors. Climatic factors (especially precipitation) were the main drivers for hydrological alterations during the period of pre-dam construction (1971–1998). On the other hand, anthropogenic activities (especially dam construction) were the major drivers in runoff changes during 1999–2010.

This study carried out trend analysis of stream flow data and separated the impacts of climate and human activities on the alterations of stream flow; however, it did not consider land use change. Dam construction, especially large dams, can also result in long-term impacts such as damage to biodiversity in downstream regions (McManamay et al. 2012). Therefore, in future studies, it is important to further explore this issue.

In this study, we investigated temporal changes in runoff caused by construction of the Shahid Rajaei dam. For this purpose, hydro-climatic parameters in three periods: pre-dam construction (1971–2000), post-dam construction (2001–2010), and the entire period (1971–2010) were evaluated by the Mann–Kendall statistical method. The conclusions of this study are summarized as follows:

• 1.

  Wet and dry season runoff had abrupt change points and the runoff of the wet and dry seasons had significant decreasing and increasing trends after 1999, respectively.

• 2.

  Correlation analysis indicated that changes in runoff in the post-dam construction period compared with the pre-dam construction period were less influenced by climatic variables than human activity. Dam construction, as one of the most important human activities, led to stream flow change.

• 3.

  Mean runoff values in dry and wet seasons changed significantly from pre-dam to post-dam construction; hence, the mean value of wet season runoff in the period of post-dam construction was lower than that in the period of pre-dam construction. Also, the mean value of dry season runoff in the post-dam construction period was higher than in the pre-dam construction period.

• 4.

  The null hypothesis H₀ (no trend) tested using two methods including the Mann–Kendall and CRD tests was rejected at the 0.05 confidence level. Both tests indicated a decreasing trend in wet season runoff and precipitation during both the periods of 1971–1998 and 1971–2010. Both tests also indicated a decreasing trend in dry season runoff and precipitation during the period of 1971–1998, before 1999. They also displayed an increasing trend in dry season runoff during the period of 1999–2010. Results of the Mann–Kendall and CRD tests indicated that dam construction had led to significant changes in the runoff of wet and dry seasons, representing a significant decrease in the wet season runoff in the 1971–2010 period when compared to the 1971–1998 period. The difference in wet season precipitation levels between the two periods was negligible. Also, the increase of dry season runoff in the 1971–2010 period was significant at the 0.05 confidence level while dry season precipitation decreased in both the 1971–1998 and 1971–2010 periods.

• 5.

  Investigation of mean evaporation values showed that the increase in water level caused by dam construction and reservoir performance, had resulted in a sharp increase in mean evaporation values in the post-dam construction period, especially from April to August.

• 6.

  Based on the observed and simulated values in the double mass curve, the contribution of climate variability and human activity was 57.76% and 42.24%, respectively, on the runoff decrease of the wet season during 1991–1998. Climate variability and human activities contributed 24.68% and 75.32%, respectively, to the wet season runoff decrease of 116.55 mm for the 1999–2010 period. For runoff changes during the dry season, climate variability and human activity impacts were reversed. The contribution of climate variability and anthropogenic activities were −30.68% and 130.68%, respectively, on the runoff increase of 41.45 mm for the 1999–2010 period.

The authors are grateful to the Islamic Republic of Iran Meteorological Organization (IRIMO) and the Water Resources Management Company of the Ministry of Energy, Islamic Republic of Iran for the provision of meteorological and hydrological data.