Spatial and temporal variation of meteorological parameters in the lower Tigris–Euphrates basin, Türkiye: application of non-parametric methods and an innovative trend approach Open Access

The main driver of climate change is expected to be an increasing greenhouse gas concentration in the atmosphere (IPCC 2014). Due to global warming, climate change is on the rise and has a significant impact on today's socioeconomic development in addition to the effects of our ecological processes (Chang et al. 2019). The Intergovernmental Panel on Climate Change (IPCC 2018) asserts that the global and local hydrological cycles may be affected by the rise in average global surface temperature. The research community has focused its attention on how climate change is globally affecting hydroclimatic variables such as temperature, precipitation, relative humidity (RH), evaporation, discharge, and wind speed, which are stochastic and complex (Hırca et al. 2022). Although climate change has affected the entire world, its effects frequently differ from one region to the other (Masson-Delmotte et al. 2021; Sesana et al. 2021). Therefore, assessing the change in hydroclimatic variables is crucial for detecting climate change. Numerous studies have shown that climate change significantly affects natural ecosystems, society, and the economy by altering the hydrological cycle, which can result in a lack of water resources and an overabundance of floods and droughts (Chao & Feng 2018; Terzi et al. 2019; Esit et al. 2021; Yang et al. 2021).

One of the best methods for monitoring the effects of climate change on hydro-meteorological parameters is trend analysis (Almazroui & Şen 2020; Vishwakarma et al. 2022). Significant hydrological changes, such as decreasing or increasing trends and prolonged recurrence intervals for extreme events, are predicted to result from climate change (Eris et al. 2019; Yuce et al. 2019; Umar et al. 2022). While analyzing their trends can give valuable insight into water resources and meteorological sciences, the intrinsic characteristics of hydroclimatic variables require specific methods for trend analysis (Şen 2012; Dabanlı et al. 2016; Sa'adi et al. 2019). Numerous parametric and non-parametric methods are employed to analyze the trend. The foundation of parametric approaches is the presumption that the data follow a normal distribution. Because non-parametric approaches do not require the data to fit a normal distribution, they are typically preferred in trend analysis research (Akçay et al. 2022). Non-parametric methods include the Mann–Kendall (MK) test, Spearman's rho (SR) test, and Sen's trend slope test. Numerous studies indicate that the MK test is frequently employed in hydroclimatic variables such as precipitation (Vishwakarma et al. 2020a, 2020b; Mallick et al. 2021; Baig et al. 2022), temperature (Vishwakarma et al. 2020a, 2020b; Agbo et al. 2021; Duy et al. 2022), streamflow (Shahid et al. 2018), RH (Eymen & Köylü 2019), and evaporation (Malamos & Tegos 2022) throughout the globe. Table 1 summarizes previous studies on trend analysis.

Although the MK test is frequently employed to identify trends, there are several issues with serial correlation in time series data. Trend-free pre-whitening Mann–Kendall (TFPW-MK) tests have been developed as a solution to this issue. This test offers more accurate results than the MK test, but it also relies dependent on sample size and data distribution (Yue et al. 2002). In order to address the problems with the MK test, innovative trend analysis (ITA) was developed by Şen (2012). ITA has gained popularity despite these limitations since it can identify hidden trends while utilizing other methods, taking trend analysis a step further (Girma et al. 2020; Wang et al. 2020). The major benefits of the ITA approach, in contrast to traditional trend methods like the MK and SR tests, do not include monotonous trends and do not include restrictions such as data length, independent structure of time series, and normality (Kisi 2015). Recently, researchers have employed the ITA method and have compared it with traditional trend analysis approaches. Alifujiang et al. (2021) compared ITA results with the MK trend test at a 95% confidence level at 13 hydrological stations in the Lake Issyk-Kul Basin, Central Asia. Results of the comparison showed that the ITA approach could successfully identify the trends captured by the MK trend test. Nguyen et al. (2022) analyzed sea surface height referenced against the WGS84 ellipsoid at the Hon Dau tidal gauge station. With strong growth trends of about 3.38 mm/year with the MK test and 3.08 mm/year with the ITA method for 1961–2020, MK and ITA demonstrated complete agreement among tests.

ITA allows data to be categorized into groups such as low, medium, and high and provides information about significant climatic occurrences such as floods and droughts. The internal trend of the time series is identified by dividing the data into low, medium, and high values (Şen 2012). However, the MK test only allows for the observation of monotonic trends, and categorization, as mentioned above is not possible to assess. Additionally, the MK test does not allow for the detection of different trends, whereas the ITA allows for the detection of increasing, decreasing, and no trend scenarios (Akçay et al. 2022). The innovative trend analysis with significance test (ITA-ST) presented by Şen (2017) uses numerical computations to determine the importance of the results obtained using the ITA suggested by Şen (2012). By using the linear trend slope approach, the essential statistical equations are supported, and the trend significance is evaluated.

Although numerous attempts have been made to analyze the trends in precipitation and temperature in Türkiye over the past few decades (Ay 2020; Sezen & Partal 2020; Esit 2022), no comprehensive research has been performed to evaluate the trends of RH and evapotranspiration (EP) comparing ITA with classical trend methods. Additionally, a few investigations are employed on a basin or regional scale. Odemis et al. (2010) investigated quantifying long-term changes in water quality and quantity of the Euphrates and Tigris rivers, Türkiye by the MK test. Sezen & Partal (2020) evaluated annual and seasonal precipitation trend analysis in the Euphrates–Tigris basin, Türkiye, using the ITA method. Gumus et al. (2022) analyzed streamflow trends in the Tigris basin using ITA and MK methods. Furthermore, there is no study to detect the approximate year of the beginning of the significant hydroclimatic variables for the lower Tigris–Euphrates basin. The aim of this paper evaluates the trend and magnitudes in annual hydroclimatic variables (precipitation, temperature, RH, and EP) of the 23 stations located in the LTEB. The uniqueness of the paper is the comparison recent of the ITA-ST with the classical MK and SR trend test approaches. Free R packages (open-source) and ArcGIS programs are employed for analysis. The study emphasizes the potential for extensive statistical analysis to provide a deep insight of hydroclimatic variables distribution patterns throughout time and space.

The Euphrates–Tigris basin, which covers the LTEB, is one of Türkiye's 25 basins. The Tigris–Euphrates sub-basin, which covers approximately 7% of Türkiye's surface area, has a precipitation area of 54,695.7 km². It is bordered by the Euphrates sub-basin in the northwest and west of the Tigris sub-basin, the Lake Van Basin in the north, Syria and Iraq in the south, and Iran in the east. All or a part of the provinces of Diyarbakır, Batman, Siirt, Bitlis, Şırnak, Hakkari, Şanlıurfa, Elazığ, Bingöl, Van, Siirt, and Mardin are located in the basin. In terms of climatic conditions, it can be noted that the basin has a variety of climatic characteristics. By diminishing the cold winds coming from the Southeast Taurus Mountains in the basin's north, the effect of the Mediterranean climate seen in the southwest parts of the upper Tigris lower basin is quite moderate. In the summer months, especially in the Southeastern Anatolia Region, drought is severe. The annual average precipitation in the basin increases significantly from Southeast Anatolia to the north toward the Black Sea. Most of the precipitation falling in the Eastern Anatolia region in winter falls as snow (Gumus et al. 2022).

The GAP (Southeastern Anatolia Project) region, which is constituted of broad plains in the lower Euphrates and Tigris basins, contains 20% of Türkiye's 8.5 million ha of irrigable land. Twenty-eight percent (28%) of Türkiye's total water potential is controlled with the project's completion, irrigation of 1.7 million ha of agricultural land is accomplished, and 27 billion kWh of electricity per year with 7.485 MW of installed hydraulic power are produced (Alivi et al. 2021).

In this paper, a comparison of traditional approaches (MK and SR tests) and an ITA is used to analyze the trend in the annual total precipitation, EP, and annual mean temperature and RH time series of the LTEB in Türkiye. Before applying MK, it is determined whether the data are inherently dependent. The time series where autocorrelation is found then performed the TFPW approach to remove the autocorrelation. Afterward, trend studies are carried out annually utilizing the MK, SR, and ITA methodologies. Finally, change-point detection tests, including the Cusum and sequential Mann–Kendall (SMK) tests are performed to determine the possible beginning year of a trend.

A positive Z number shows an upward trend, while a negative value shows a downward trend. Critical test statistical values at the 90, 95, and 99% probability levels are 1.645, 1.96, and 2.57, respectively (Yuce & Esit 2021).

Rx_i (rank statistic) is obtained by sorting the data, where n is the length of the time series. Negative z values show decreasing trends, while positive z values show increasing trends. At the 10% significance level, for z >± 1.645, the null hypothesis of no trend is rejected.

A distribution-free CUSUM (trend change package in R) test is utilized to detect the abrupt change point in climatic records. The change point in the time series will be the point at which the cumulative sum reaches its maximum value. The significance of the change point is indicated if the highest value is equal to or greater than the critical value (Patakamuri & Das 2019).

The backward sequential statistic, u′(t_i), is calculated in the same way, but starting from the end of the series. Many scientists have used this approach to determine the starting point of trends (Rahman et al. 2017; Salehi et al. 2020; Alhathloul et al. 2021) rather than identifying the whole trend. In this study, the SMK method is applied to detect an abrupt change in the time series of hydro-meteorological data.

Table 3 shows the r₁ values at lag-1 for the annual meteorological variables (precipitation, temperature, EP, and RH). The critical limits of r₁ range from 0.2 to 0.4. Any value outside of this range demonstrates the presence of serial correlation in the series. Table 3 indicates that significant serial correlation at lag-1 is the annual total precipitation, annual mean temperature, annual mean RH, and annual total EP series of 4 (17.4%), 16 (69.5%), 21 (91.3%), and 17 (74%), respectively. However, negative serial correlations in the annual total precipitation are not significant. The intrinsic dependency in meteorological variables is removed in this study using the TFPW methodology. The new meteorological series, which is free of serial correlation, is then applied to the MK test.

While a decreasing trend (12 stations, 52.1%) is observed at the annual mean RH series in the eastern and southern parts of the basin, no significant trends are captured in the western part considering both two tests. Furthermore, the highest z values are seen in stations 17880 (Van), 17920 (Hakkari), and 17948 (Mardin). In the annual total EP, all stations show an increasing trend across the LTEB basin except station 17852, which shows no significant trend. The highest spatial z-value distribution is observed in the eastern and southern parts of the basin.

In the annual mean RH, the highest magnitude decreasing trend is found as −0.260 and −0.272%/year at stations 17920 (Hakkari) and 17948 (Mardin), respectively. In addition, the lowest magnitude decreasing trend is detected as −0.013%/year at station 17275 (Mardin). In the annual total EP, stations 17287 (Şırnak), 17950 (Şırnak), and 17944 (Şanlıurfa) reveal the highest magnitude increasing trend as 5.015, 4.54, and 5.66 mm/year, respectively. Furthermore, station 17280 (Diyarbakır), compared to other stations, shows the lowest magnitude increasing trend of 1.006 mm/year.

In the annual total precipitation, there is no observed significant trend at stations 17920 (Hakkari) and 17880 (Van). The same station results are similar to both MK and SR test results. Trend slopes are detected to be higher than the upper limits, which means an increasing trend at stations 17944 (Şanlıurfa), 17914 (Şanlıurfa), 17852 (Van), 17172 (Van), and 17287 (Şırnak), while trend slopes are observed to be higher than lower limits in the remaining stations (significantly decreasing trend). In addition, the highest magnitude increasing trend is observed as 4.004 mm/year at station 17944 (Şanlıurfa), and the lowest magnitude decreasing trend is evaluated as −4.74, −4.52, and −4.05 mm/year at stations 17275 (Mardin), 17874 (Diyarbakır), and 17810 (Bitlis), respectively.

In the annual mean temperature, all stations across the basin demonstrate an increasing trend at a significant level. The highest magnitude increasing trend appears at station 17944 (Şanlıurfa) as 0.08 °C/year, whereas the lowest value is detected at station 17280 (Diyarbakır) as 0.0072 °C/year. According to the annual total EP, all stations display an increasing trend over the LTEB basin. Like the annual mean temperature, station 17944 (Şanlıurfa) observed the highest magnitude trend at 6.54 mm/year. Furthermore, the lowest magnitude decreasing trend is found at station 17852 as 0.195 mm/year. In the annual mean RH series, trend slopes are demonstrated to be higher than the upper limit at stations 17950 (Şırnak), 17912 (Şanlıurfa), 17270 (Şanlıurfa), 17874 (Diyarbakır), 17847 (Diyarbakır), and 17210 (Siirt). In the remaining stations except 17966 (Şanlıurfa), the trend slope is found to be higher than the lower limit. The higher magnitude decreasing trends are noted as −0.3227 and −0.3207 at stations 17852 (Van) and 17948 (Mardin), respectively.

Table 4 shows the results of the MK, SR, and ITA-ST tests for each hydro-meteorological variable. There are significant differences between the traditional MK test, SR test, and ITA-ST test when the table is interpreted generally. The corresponding table shows that the findings obtained by MK and SR are completely parallel. According to the table, while MK and SR demonstrate a significant trend in 7 of the total annual precipitation series, ITA-ST captures the existence of a significant trend in 21 of the 23 total annual precipitations. ITA-ST captures the same trends as MK detects a significant trend in the months. However, in 14 annual total precipitation stations, MK fails to detect any trends, while ITA-ST identifies the existence of trends. The explanation for the difference in results between MK, SR classical trend methods, and the innovative method is that the innovative method is more sensitive in detecting trends.

Table 4

View Large

Three methods reveal that there is an increasing trend in both the annual mean temperature and the annual total EP except for one station across the basin. Stations 17810 and 17880 do not show any significant trend in the annual mean temperature and annual total EP, respectively. The most significant differences between the three methods are seen in the precipitation and humidity data. For example, MK, SR, and ITA-ST capture a significant decreasing trend in 10, 8, and 16 of the 23 annual mean RH series, respectively. Furthermore, while ITA-ST is detected to increase the trend of six stations of total annual mean RH, no significant trends are observed in both MK and SR tests.

According to the results presented in Table 5, there is no change in year captured in the annual total precipitation except for stations 17810 (Bitlis), 17270 (Şanlıurfa), 17968 (Şanlıurfa), 17275 (Mardin), and 17948 (Mardin). Change years are detected the same year in 1998 and 1989 by two tests for stations 17270 (Şanlıurfa) and 17968 (Şanlıurfa), whereas the beginning of trend years are observed by the Cusum test as the year 1989 for stations 17275 (Mardin) and 17948 (Mardin) and the year 1994 for station 17810 (Bitlis). In the annual mean temperature, two tests capture the same year at stations 17980 (Şanlıurfa), 17287 (Şırnak), 17950 (Şırnak), and 17847 (Diyarbakır) which start in the years 2008, 1987, 1995, and 1995, respectively. In addition, no change point detection is observed at station 17280. In the remaining station, two tests capture different change years. In the annual mean RH, stations 17880 (Van) and 17810 (Bitlis) start to change in the same direction considering two tests in 1992 and 2005, respectively. In general, change point years are not significantly detected at most stations. In the annual total EP, while the Cusum test captures changes in increasing trend direction at 22 of the 23 stations, SMK indicates to detect significantly at 16 of the 23 stations. In addition, both tests were performed to identify the same trend increasing change direction at stations 17112 (Van), 17980 (Şanlıurfa), 17275 (Mardin), 17948 (Mardin), 17285 (Hakkari), 17950 (Şırnak), and 17847 (Diyarbakır) as 1994, 2008, 1998, 1998, 1998, 1995, and 1995, respectively.

As it can be seen from Table 5, while the Cusum test can detect a break at the beginning of the time series, the SMK catches the break toward the end of the time series. Furthermore, the Cusum test is performed as more sensitive than the SMK test. For instance, Cusum and SMK are detected in the start of trend year 21.7 and 8.6% of the annual total precipitation time series, 95.65 and 69.56% of the annual mean temperature time series, 47.82 and 17.4% of the total mean RH, and 95.65 and 69.56% of the annual total EP time series.

This study examines annual meteorological variables including annual total precipitation, annual mean temperature, annual mean RH, and annual total EP over the LTEB basin using traditional non-parametric techniques (MK and SR) and the ITA-ST to identify long-term trends. LTEB is one of the important water basins of the Middle East, especially in Türkiye, Iraq, and Syria. Before applying the MK test, serial correlation is investigated in annual meteorological variables. The TFPW methodology is used in this study to remove the intrinsic dependence on meteorological variables. The MK test is then performed using the new meteorological series, which eliminates serial correlation.

MK and SR tests indicate similarities in terms of the confidence interval. According to two test results, the annual total precipitation shows a decreasing trend in the southern part of the basin when taking into account both tests, while no other significant trends are seen in the other parts of the basin, except the northern part stations 17810 (Bitlis) and 17847 (Diyarbakır). Among the 23 stations located throughout the LTEB basin, there are substantial increasing trends in the annual mean temperature for both tests except for station 17810 (Bitlis). Furthermore, a higher increasing trend is seen in the eastern part of the basin and a lower increasing trend in the western part. While the annual mean RH series in the eastern and southern parts of the basin show a decreasing trend (12 stations, 52.1%), no significant trends are seen in the western part. All stations in the LTEB basin exhibit an increasing trend in the annual total EP except station 17852. According to ITA-ST results, increasing trends are detected at stations 17944 (Şanlıurfa), 17914 (Şanlıurfa), 17852 (Van), 17172 (Van), and 17287 (Şırnak), while a decreasing trend is observed in the remaining stations, except for 17920 (Hakkari) and 17880 (Van), which showed no significant trend. In the annual mean temperature and annual total evaporation series, all stations across the basin demonstrate an increasing trend at a significant level except station 17810 (Bitlis) for the annual mean temperature and station 17852 (Van) for the annual total EP. In the annual mean RH series, an increasing trend is demonstrated at stations 17950 (Şırnak), 17912 (Şanlıurfa), 17270 (Şanlıurfa), 17874 (Diyarbakır), 17847 (Diyarbakır), and 17210 (Siirt). A decreasing trend is found in the remaining stations except 17966 (Şanlıurfa).

Based on the above results, the explanation for the difference in results between MK, SR classical trend methods, and the ITA-ST innovative method is that the innovative method is more sensitive to detecting trends (Tosunoglu & Kisi 2017; Ali et al. 2020; Ashraf et al. 2021; Hırca et al. 2022). The result obtained by MK and SR is entirely consistent. ITA-ST captures the existence of a significant trend in 21 of the 23 total annual precipitations, but MK and SR only show a trend of relevance in seven of the total annual precipitation series. The same trends that MK notices in the months are also captured by ITA-ST. However, in 14 annual total precipitation stations, MK is unable to identify any trends, although ITA-ST captures. Both the annual mean temperature and the annual total EP have an increasing trend, according to three tests. The most significant differences between the three methods are detected in the precipitation and humidity data. For instance, MK, SR, and ITA-ST capture a significant decreasing trend in 10, 8, and 16 of the 23 annual mean RH series, respectively. In addition, while ITA-ST is detected to increase the trend at six stations of the total annual mean RH, no significant trends are observed in both MK and SR tests. Mallick et al. (2021) analyzed the trend of precipitation in the Asir region of Saudi Arabia using the ITA, MK, and modified MK (MMK) tests. The MMK test appeared to be the best-performing approach within the MK test family, while ITA appeared to be the best trend detection technique among the techniques based on the results of the tests and their performance. Hajani et al. (2022) investigated the trends in the rainfall data over the Kurdistan region, Iraq using ITA and MK tests. Most of the stations have reported an increasing trend in annual rainfall, with just four stations indicating statistically significant trends. The results of this study are in agreement with previous studies in nearby regions.

Annual hydro-meteorological variables are used to detect any abrupt changes using the SMK and Cusum tests. While the Cusum test can capture a break at the beginning of the time series, the SMK detects the break toward the end of the time series. In addition, the Cusum test is revealed as more sensitive than the SMK test. For example, Cusum and SMK are detected in the start of trend year 21.7 and 8.6% of the annual total precipitation time series, 95.65 and 69.56% of the annual mean temperature time series, 47.82 and 17.4% of the total mean RH, and 95.65 and 69.56% of the annual total EP time series. According to annual precipitation results, there is no change year recorded in the annual total precipitation (Mardin) except for stations 17810 (Bitlis), 17270 (Şanlıurfa), 17968 (Şanlıurfa), 17275 (Mardin), and 17948 (Mardin). While the beginning of trend years is identified by the Cusum test as the year 1989 for stations 17275 (Mardin) and 17948 (Mardin) and the year 1994 for station 17810, change years are found in 1998 and 1989 by two tests for stations 17270 (Şanlıurfa) and 17968 (Şanlıurfa) (Bitlis). In the annual mean temperature, two tests perform the same year at stations 17980 (Şanlıurfa), 17287 (Şırnak), 17950 (Şırnak), and 17847 (Diyarbakır) which start in the years 2008, 1987, 1995, and 1995, respectively. While the Cusum test captures a change in the direction of the annual total EP at 22 of the 23 stations, SMK indicates to detect considerably at 16 of the 23 stations.

The ITA-ST approach is easy, simple to comprehend, and more sensitive in identifying trends than the classic MK and SR methods. The study thoroughly examined the changes in rainfall, temperature, RH, and EP trends throughout the LTEB between 1963 and 2021, which will assist in understanding regional climatology and hydrology as well as their effects on industries based on water resources. Furthermore, other parameters (wind speed, groundwater levels, sunshine, etc.) should be examined for more precise results. In addition, the trends of the data on a monthly and seasonal basis need to be investigated separately.

The authors acknowledge State Water Works (DSI), General Directorate of Meteorology (MGM) for providing meteorological data.

M.E.: data gathering, hydro-meteorological data trend analysis, interpretation of the findings, manuscript writing, and submission. R.Ç.: supervision and editing. E.A.: material preparation, data collection, and analysis.

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

The authors declare there is no conflict.