Trend in river water quality: tracking the overall impacts of climate change and human activities on water quality in the Dez River Basin

The Dez River, as the second largest river in Iran and one of the major tributaries of the Karun River, is one of the main water resources in the Khuzestan province in order to provide fresh water needed for drinking, domestic uses, and agriculture purposes. Hence, besides water quantity, it is important to evaluate the water quality of the Dez River Basin. In arid and semi-arid climates, with considerable shortage of potable water, monitoring of river water quality and the influencing factors should receive great importance. Analysis of water quality as well as monitoring possible water quality changes are important tasks for better management of water resources. Therefore, many studies have considered these issues in different parts of the worlds (Luo et al. 2011; Ballantine & Davies-Colley 2014; Sayemuzzaman et al. 2018) and on a global scale (Jiang et al. 2014). Water quality can be affected by anthropogenic activities such as land use change (Soko & Ababio 2015; Shrestha et al. 2018; Wijesiri et al. 2018), chemical toxin discharge (Gevrey et al. 2010), as well as hydrometeorological changes of temperature (Crossman et al. 2013), evaporation (Abdel Wahed et al. 2014), rainfall and streamflow (Tabari et al. 2011; Xie et al. 2015; Murphy & Sprague 2019). Assessment of long-term trends in water quality parameters can be considered as an appropriate approach to determine any environmental changes over time, help to identify and analyze the most substantial factors affecting the water quality (Ballantine & Davies-Colley 2014; Goodrow et al. 2017), and clarify some new aspects of hydrological processes. There are various measures to determine the trend in water quality time series that can be named as time series analysis methods (Long et al. 2009) including parametric (Kundzewicz & Robson 2004; Yenilmez et al. 2011) and non-parametric tests (Bouza-Deano et al. 2008). Due to skewness, serial correlation, non-normal data, ‘less-than’ (censored) values, outliers, and missing values in hydrological data, it is recommended to use non-parametric methods to determine the trend of water quality parameters (Helsel & Hirsch 1992; Hamed & Rao 1998; Mozejko 2012). Hence, based on several studies, in different parts of the world, the non-parametric Mann–Kendall method and Sen's slope test (Yenilmez et al. 2011; Naddeo et al. 2013; Ballantine & Davies-Colley 2014; Entry & Gottlieb 2014; Niazi et al. 2014; Yevenes et al. 2018; Chen et al. 2019), seasonal Kendall (Helsel & Hirsch 2002; Ballantine & Davies-Colley 2014; Dabrowski et al. 2014; Entry & Gottlieb 2014; Hughes & Quinn 2014; Cloern 2019), and LOWESS (Helsel & Hirsch 2002; Hughes & Quinn 2014; Entry & Gottlieb 2014; Ahn & Kim 2019; Biswas & Mosley 2019) have been chosen to assess the possible trend in water quality time series.

In river basin management, analysis of the relationship between water quality parameters and streamflow rate is a crucial step to distinguish changes in water quality (Assilis et al. 2001). Therefore, this case study considers the influence of common seasonal fluctuations and flow rate changes on water quality using a set of non-parametric statistical techniques, i.e., modified Kendall, LOWESS, and seasonal Kendall on residues resulting from the LOWESS test, to identify possible climate, environmental, and anthropogenic factors controlling water quality changes in the Dez River Basin.

The Dez River Basin, located in the western part of Iran between 48° 22′ 16″ to 50° 18′ 54″ E and 32° 33′ 48″ to 34° 06′ 52″ N, covers approximately 15,990 km². Elevation of the study site ranges from 292 to 4,049 meters above sea level (masl), and the length of the river is 415 km (Figure 1). The basin receives an average annual precipitation of 100 to 460 mm, mostly in winter (70–80%), especially during February and March. The average annual temperature and actual evaporation are 16 °C and 260 mm, respectively. The dominant land-use type in the Dez River Basin is rangeland. Land-use proportions were derived from satellite images – Landsat 4–5 Thematic Mapper (TM) images acquired in 2000 and 2009 (USGS 2019). To accurately identify the farming areas, the could-free images were selected as they are representative of the cropping season of the basin (Wilken et al. 2017). Land-use maps were provided using the Semi-Automatic Classification Plugin (SCP), version 6.4.2 in QGIS environment. The hydrochemical parameters of streamflow were monitored in a monthly time interval over 44 years (1967–2010) at 12 monitoring sites by the Ministry of Energy of Iran. These sites are scattered within the study area and covered the main river channel (the Dez River) and some important tributaries (Figure 1). For water quality analysis, eight stations including Dorud Tireh (S1), Marbareh (S2), Sepid Dasht Cezar (S3), Sepid Dasht Zar (S4), Keshvar Sorkhab (S5), Tangeh Panj Cezar (S6), Tangeh panj Bakhtiar (S7), and Taleh Zang (S8), were selected based on their time series length and quality of data (Table 1). Human activities such as dam construction do not effect these monitoring sites. In addition, there are no main industrial companies in the upstream of the Dez River, where cities are located in the downstream of the Dez dam. The Dez dam is located in the downstream of the outlet of the study area (Figure 1).

Ten hydro-chemical parameters including electrical conductivity (EC), pH value (pH), bicarbonate (HCO₃⁻), sulfate (SO₄²⁻), chloride (Cl⁻), calcium (Ca²⁺), magnesium (Mg²⁺), sodium (Na⁺), total dissolved solids (TDS), and sodium adsorption ratio (SAR), were monitored and analyzed by Iran Water Resources Management Company (IWRMC) at monthly intervals. EC and pH were measured in the field by portable multi-parameter water quality meter. TDS were determined by gravimetric analysis. The dissolved water samples were shipped to the laboratories of IWRMC for analysis of Ca²⁺ and Mg²⁺ (by EDTA titration), Na⁺ (by flame photometer), SO₄²⁻ (by barium chloride titration), HCO₃⁻ (by sulfuric acid titration), and Cl⁻ (by silver nitrate titration). SAR, generally used to assess water suitability for irrigation, was calculated using Na⁺, Ca²⁺, and Mg²⁺ concentrations. It should be noted that time series were entirely checked by graphical and statistical methods for outliers and any possible errors. However, missing values in time series (no sampling dates or deleted data as results of our data quality check) were not estimated because of possible uncertainty in reconstruction of water quality data.

In order to consider the effect of seasonal patterns on water quality, temporal trend in water quality is assessed by seasonal Kendall test, after preparing water quality time series. The Mann–Kendall method is the most common non-parametric test (Kendall 1975) for assessing trends in water quality parameters and hydrological components (Cloern 2019). For the detection of trends in precipitation, temperature, evaporation, and discharge time series, the modified Mann–Kendall test (Hamed & Rao 1998) is used. Serial correlation is considered following the methodology used by Kumar et al. (2009). Further information about the modified Mann–Kendall test is available in Mahmoodi et al. (under review). In the next step, the LOWESS test is applied to distinguish the effects of discharge, as an exogenous variable, on water quality variations. Then, seasonal Kendall test is performed on the residuals from LOWESS to assess trend in water quality time series when the main exogenous effect (effects of discharge) is considered and removed. Residuals from LOWESS test were considered to search another possible controller of water quality in the Dez River Basin. The slope of the trend line is estimated by Theil–Sen method.

Positive values of S_k indicate a positive trend, and vice versa. The test statistic Z_sk gives the significance level of rejecting the null hypothesis (Kumar et al. 2009). Therefore, if the test statistic is smaller than the Z_crit statistical value of the considered significance level (such as ±1.645 for a confidence level of 90%), then the identified trend is not significant (Carmona & Poveda 2014).

Moreover, besides searching for possible correlation of different variables with the residuals’ time series from LOWESS test, seasonal Kendall test on residuals of LOWESS test is performed, to search any possible trend in controller of river water quality beyond discharge effects. Smoothing factor (f) in the LOWESS test should be selected rationally and reasonably based upon the target of smoothing factor (Helsel & Hirsch 1992), that f ≈ 0.5 in most cases of this study was found to be ideal (the same value is reported by Dabrowski et al. 2014).

The lack of internal correlation between the data is required for using non-parametric tests. It is because such correlations can have an effect on data analysis and significance of trend statistics (Kumar et al. 2009). Therefore, the effects of autocorrelation were considered in trend analysis of whole time series data.

The significance levels were considered at confidence levels of 90, 95, and 99%. Z_crit statistical values of these considered significance levels are ±1.645, ±1.96, and ±2.575, respectively.

Trend test results are categorized as being significant or non-significant as follows:

1. No significant change: the null hypothesis for the modified Mann–Kendall test or seasonal Kendall test was not rejected (If −Z_crit < Z (Z_sk) <+ Z_crit).

2. Significant increase/decrease: the null hypothesis for the modified Mann–Kendall test or seasonal Kendall test was rejected (If −Z_crit >Z (Z_sk) or Z (Z_sk) >+ Z_crit).

Due to a strong seasonal pattern in surface water quality, the elimination of the seasonal effects can help to have more detailed trend analysis. Therefore, by using the seasonal Kendall test, the trend of water quality parameter without the seasonal effects was determined (Table 2). It shows the general trend of water quality after adjustment for seasonal variations. Ca²⁺, HCO₃⁻, and Mg²⁺ at all stations show rising trends at confidence level of 95% and increasing trends in EC and TDS were significant for most sites. The slopes of the trend lines are calculated by Theil–Sen method in Table 3. Considering the seasonal variations, there are still many significant changes in water quality which corroborate the role of other factors in water quality changes. In the next section, the effects of the main controllers of water quality, i.e., streamflow, are also removed by LOWESS test and then seasonal Kendall test is performed on the residuals of LOWESS test to explore the effects of other water quality controllers in the study area. Temporal trends in streamflow discharge of the Dez River Basin for given stations is reported by Mahmoodi et al. (under review).

After eliminating seasonal variations, the second major external variable affecting water quality is discharge and its temporal changes. For elimination of discharge effect and more detailed trend analysis of water quality, the LOWESS test is executed. Therefore, water quality data (Y) against discharge were plotted, then a smooth fitted line defined by LOWESS test and residuals (the differences between observed values (Y) and the y-value on the fitted line (Y″)) were calculated. In the next step, to identify trends, seasonal Kendall test was applied on the residuals’ time series. The results of trend analyses on the residuals are presented in Table 4. As the results demonstrate, most of the water quality parameters show positive significant changes. For all monitored stations, increasing trends in Mg²⁺, HCO₃⁻ are observed at confidence level of 95% and for Ca²⁺ at confidence level of 90%. TDS and EC also carry inclining significant trends in confidence level of 95%, at all stations except S₅. pH is the only parameter showing no significant trend for all the stations except S₅, when discharge effect is eliminated. There were some declining trends of SO₄²⁻, Na⁺, and SAR in some sites. The slopes of the trend lines calculated by Theil–Sen method are presented in Table 5. After elimination of the two major exogenous variables, i.e., discharge and seasonal patterns, any possible effects on water quality can be clarified by searching the relevance between environmental and anthropogenic activities and the residuals from LOWESS. In other words, these analyses reveal the possible effects of some other environmental factors on surface water quality, such as change in surface water–groundwater interaction, precipitation, temperature, evaporation, and any other human activities (such as land-use changes). Also, it might help for better understanding the changes in river water quality and identifying any possible controller.

Due to the non-industrialized condition of the district, the most important human activity that might affect water quality is land-use change (Shrestha et al. 2018). The different pattern of sediment production in response to land-use change could result in water quality degradation (Bussi et al. 2016). Sediment loadings transported from the basin into the river results mainly from agriculture intensification (Taniwaki et al. 2017). The spatial distribution of the major land-use classes in 2000 and 2009 are shown in Figure 4. The available could-free Landsat images that cover the whole basin in the mosaic manner within the same season, to accurately represent land uses, limit our analysis of land-use change for the years 2000 and 2009. In the year 2000, the Dez River Basin catchment is dominated by rangeland (77.1%). Bare land covers 7.1% and is mostly found at the middle part of the basin (subbasins 4 and 7). Rainfed and irrigated farming areas accounted for 8.8% and 6.9% of the study area and are mainly found in the northern part of the basin (subbasins 1 and 2). Comparing the land-use map of the years 2000 and 2009, the most obvious changes occurred in the rangeland areas, about +3.9% (Table 6). From 2000 to 2009, the areal coverage of bare land and irrigated farming decreased by −2% and −1.7%, respectively, while the percentage of change for rainfed farming is almost null (0.2%). This shows that those reductions in bare land and agricultural areas could lead to decreasing sediment production in the whole basin and improve the water quality. However, despite this fact, water quality parameters mainly experienced an increasing trend (Table 4, S₈). This means, although the change in land use could have an impact on water quality of the Dez River Basin; it is not the main driver over the study period. Moreover, this could be supported by the results in the smaller scale, comparing subbasin 2, where the main reduction occurred in the area of rainfed farming (−2%), irrigated farming (−4.8%), and bare land (−9.7%), with other subbasins. A trend pattern of river water quality at this station is quite similar to others.

The relationship between surface water and groundwater is inevitable, as in most arid and semi-arid basins in Iran such as central wadis, the base flow is supplied by shallow aquifer (Mahmoodi et al. 2020a, 2020b). In a particular area, a large difference between surface water and groundwater quality is common. In the Dez Basin, EC of groundwater is greater than the EC of surface water. Therefore, the different contribution of groundwater in streamflow can alter streamflow quality where there is an important interaction. Temporal trend in groundwater levels is monitored at 11 piezometric wells in Azna-Aligudarz plain and 13 piezometers in Doroud-Borujerd from 2001 to 2010. The modified Mann–Kendall test is applied to these available short time series (Table 7). The geographical positions of piezometers are presented in Figure 1. The results of the test show mostly declining trends in groundwater level of both aquifers, but these negative trends are significant only in P₂ and P₄ in Azna-Aligudarz plain and in P₁, P₈, P₁₁, and P₁₂ in Doroud-Borujerd plain at a confidence level of 90%. Obviously, because of higher EC of groundwater rather than that of surface water, declining trends in groundwater levels can improve surface water quality in rivers which are affected by groundwater depletion and result in negative trends of EC values of surface water. While EC of surface water has positive significant trends in many cases, especially for station S₂, it can be interpreted that the minor declining trends in groundwater table cannot explain the observed positive trends in surface water quality constituents.

One of the climate factors affecting streamflow is precipitation, which can change water quality by change in river discharge. As mentioned in previous sections, streamflow discharge is the main controller of stream water quality around the world, but it is not responsible for the whole change in river water quality. The trend in precipitation of three rain gauges R₁, R₂, and R₃ (Figure 1) was analyzed. More details are available in Mahmoodi et al. (under review). They reported that the trends were not significant for any of the mentioned stations, but streamflow reductions were possibly caused by overall reductions in precipitation. Due to a general strong relation between both precipitation and discharge with water quality, any small changes in precipitation, and consequently streamflow rates, can lead to significant changes in water quality parameters. Therefore, it can be concluded that precipitation may justify parts of the observed trends in the Dez River water quality thorough changes in the Dez River water quantity.

The effects of temperature on evaporation rate, discharge, and, in addition, on dissolution can influence river water quality. Hence, the temporal changes in temperature at three stations T₁, T₂, and T₃ were analyzed by modified Mann–Kendall test. The overall results show positive significant trends in T₂ (at confidence level of 90%), T₁ and T₃ (at confidence level of 99%). Hence, variations in temperature are in accordance with trends of residuals’ time series of water quality extracted from LOWESS test. Therefore, beside discharge effects, temperature variations may have a considerable influence on water quality by changing the river water quantity in the Dez River Basin.

Evaporation, as one of the related factors under the influence of temperature, can change water quality by change in salt quantity. The variations of potential evaporation in the five stations of E₁, E₂, E₃, E₄, and E₅ (Figure 1), is also provided in more detail by Mahmoodi et al. (under review). The results highlight positive significant trends at confidence level of 95% in all monitoring stations, which is in agreement with trend in river water quality (the residuals from LOWESS test). Figures 5 and 6 show the relationships between residuals of EC (as a principal water quality parameter) at S₈ and monthly evaporation at E₅ for October and September. These graphs draw attention to changes in evaporation, resulting from trends in temperature in the Dez River Basin, which might have a key role in water equality degradation.

After an integrated temporal trends’ analysis of river water quality, the role of environmental and human factors (in agreement with Ballantine & Davies-Colley 2014) as well as seasonal changes and discharge affect are presented in a comprehensive analysis by a set of non-parametric tests (seasonal Kendall, Sen's slope, and LOWESS). Our results declared that non-parametric tests are powerful tools for trend analysis in water quality time series (previously proved by Helsel & Hirsch 2002; Yenilmez et al. 2011; Naddeo et al. 2013; Ballantine & Davies-Colley 2014; Entry & Gottlieb 2014; Hughes & Quinn 2014; Dabrowski et al. 2014). By removing the effects of seasonality and streamflow fluctuations, trends in water quality remained significant. It clarifies the roles of some other factors on river water quality in the Dez Basin. By trend analysis of modified water quality time series (the residuals from LOWESS test) besides analyzing environmental and human factors, it is revealed that changes in land use and groundwater table were not in accordance with the observed trends in water quality. Our analyses show that besides powerful seasonality pattern in river water quality and discharge-related fluctuations, positive trends in temperature and evaporation show a pattern, which may explain significant parts of observed trends in treated water quality time series (after removing the effects of seasonality and streamflow fluctuations). By our findings, it can be concluded that, besides changes in streamflow and seasonality, water quality degradation in the Dez River Basin is accelerated by increasing trend in temperature and, consequently, in evaporation (the same findings are reported by Crossman et al. (2013) and Abdel Wahed et al. (2014)). Stable isotope application in the upper Karkheh River Basin (Osati et al. 2014), located in the west of the Dez River Basin, has shown that the largest seasonal isotope variations are in correspondence with clear seasonal patterns and evaporation impacts on streamflow. Such analysis in the Dez River Basin can verify and trust our results about evaporation influences on river water quality and quantity. It should be noted that changes in groundwater–surface water relationships (as it is reported as a main cause of trends in river water quality of the Karkheh River Basin after removing the effects of seasonality and streamflow fluctuations by Osati et al. (2012)) may have some impacts on river water quality, especially in Azna-Aligudarz and Doroud-Borujerd plains, which are located in the upstream part of the Dez Basin. This is because of potential changes in groundwater contribution in streamflow.

Our findings can provide important information needed for proper management of streamflow resources in the Dez River Basin and guarantee sustainable water supplies for drinking, domestic uses, and agriculture purposes. There are diverse river water quality controllers which altogether might result in important changes in river water quality under improving/degrading conditions.

This study identified the effects of climate change, environmental, and anthropogenic factors on water quality of the Dez River Basin using statistical tests, i.e., modified Mann–Kendall test in conjunction with Theil–Sen method and LOWESS. Our findings show the following:

1. Combination of non-parametric statistical methods is a suitable approach to evaluate the temporal variability of hydrochemical components as well as to separate the impact of climate change, environmental, and anthropogenic factors on variability of these components.

2. In addition to alterations in streamflow as a main factor, climate variability present in temperature and, consequently, in evaporation plays a key role on temporal trends of hydrochemical components.

3. Impact of groundwater contribution on stream water quality is more pronounced in subbasins where the groundwater–surface interaction is higher.

4. Comparing subbasin 2, with a considerable land-use alteration, with other subbasins corroborates that although land-use change could have an impact on water quality of the Dez River Basin, it is not the main driver over the study period.

Changes in temperature and evaporation emphasize a new aspect of climate change effects on river water quality and warn of a new challenge for water resources management in semi-arid areas facing climate changes.

We are particularly grateful to Dr Jens Kiesel for helpful comments on the abstract. The authors received no specific funding for this article. The authors have declared that no competing interests exist.

All relevant data are included in the paper or its Supplementary Information.