Reevaluating streamflow declines across the middle East and central Asia with insights from change point detection

Abstract

Change point detection (CPD) methods are widely used to identify abrupt shifts in streamflow, often linked to human activities. This study assesses the effectiveness of CPD techniques across countries in the Middle East, Central Asia, and Pakistan—regions vulnerable to water scarcity and climate variability. Analysis of annual streamflow data (1970–2018) revealed that 1998 was the most frequent breakpoint, coinciding with a major El Niño event, suggesting climatic anomalies were key drivers of the initial change. The coherence of breakpoints across the region highlights the influence of large-scale climate signals. However, comparison of pre- and post-break conditions indicates that the magnitude of hydrological shifts cannot be explained by climate alone, highlighting the limitations of CPD methods in distinguishing climatic from anthropogenic drivers. To explore these dynamics in more detail, the Karkheh River Basin (KRB) in Iran was examined. A marked change in streamflow patterns around 1998 was observed, along with shifts in temperature and precipitation. These results underscore the need for cautious use of CPD methods when attributing hydrological changes to human versus climatic factors.

Introduction

The streamflow of a catchment, which is an essential part of the hydrological cycle and has experienced nonstationary changes globally¹, is predicted to continue to alter in the future². Climate change and human activities are the main causes of these nonstationary variations³. In the last century, the influence of climate variability associated with global warming has been regarded as one of the main reasons for the decrease of per capita available water resources and shifts in their geographical distribution⁴. More intense runoff occurrences may result from climate factors including increased temperatures and precipitation^5,6,7. Furthermore, the spatiotemporal distribution of water resources is impacted both directly and indirectly by human activities such irrigation, dam construction, urbanization, and changes in land use/land cover (LULC)^8,9,10,11.

Numerous studies on the identification and attribution of runoff changes have produced an abundance of results, but there is no consistent trend in runoff globally. Most regions have shown increasing trends in runoff, while others have shown decreasing trends⁴. Some studies have concentrated on examining the effects of meteorological elements on streamflow, including temperature, potential evapotranspiration, and precipitation. For instance, Li et al.¹² determined that the Heihe River streamflow will drop by 2.0% and 2.9%, respectively, with a temperature increase of 2.0 °C and 2.9 °C. According to the findings of Shah et al.¹³, a 10% shift in air temperature and precipitation causes a 12 to 20% and 8 to 18% change in streamflow, respectively, in six main rivers in Pakistan. A number of studies have examined how human activity affects runoff. For example, studies have examined the impact of urbanization¹⁴ or changes in vegetation cover¹⁵ on streamflow variability. Researchers are more interested in accurately separating the impacts of climate and human activity on streamflow variability. Finding change points in streamflow time series is an essential initial step in determining the impacts of human activities from those of climate variability³.

Sometimes known as “breakpoints”, “change points” are abrupt changes in the distribution parameters of a dataset. In general, change points are observed at that point in the data which splits the same into two subsets of data with distinct statistical properties. For instance, in normally distributed data, such break points occur when the means of two subsets differ¹⁶. Detection of change points in time series is an active research area in statistics, but they are of particular interest to hydrologist because of their important implications for hydrologic data analysis. Change points are of value to both water-resource managers and data analysts, as they may indicate natural or anthropogenic changes in climatic, hydrologic, or landscape processes. Examples of such sudden changes might involve rainfall, runoff, reservoir storage, diversions, urbanization, and agricultural practices¹⁷. Added to these is the most common one: sudden change in streamflow and water-level of rivers due to the construction of dams and other water-retaining structures^18,19.

A few methods for change-point detection have been developed. Homogeneity tests, such as the Buishand test²⁰, the Standard Normal Homogeneity Test (SNHT)²¹, and the Pettitt test²² are often employed techniques for identifying these break-points. Ordered clustering and cumulative anomaly testing are additional methods that could be taken into account for the detection of change points. The SNHT was initially established by Alexandersson in 1984 as a means of quality control for rainfall time series data. Although a small departure from normalcy may also be allowed by the test, the Buishand test statistic is expected to be normally distributed. The Pettitt test is non-parametric change point detection (CPD) method, hence it can be used to identify breakpoints in time series and does not need fitting probability distribution functions to the data. These tests have been widely utilized to identify a single breakpoint in time series data by several researchers^{3,23,24,25,26,27,28,29,30}.

In practice, sudden changes in the hydrological time series are usually associated with human activities, and the gradual variations are considered due to the change in climate^31,32,33. The reference period is the period before the detected breakpoint, where the human impacts on the runoff can be considered minimum and even neglected, and climate change is thought to be the controlling factor of the runoff. By contrast, the period after the shift point is called the human interference period, during which the impact of human activities on runoff became obvious; the real runoff in this period is relevant not only to the climatic change but also to human interference^34,35. Many works have been carried out for different parts of the world which often pinpoint the late 1990s, and in particular 1998, as the breaking point of the streamflow^{27,30,33,34,36,37,38,39,40,41,42,43,44,45,46}. As mentioned earlier, according to common CPD methods, the year 1998 (late 1990s) is considered the breaking point after which human activities are believed to have significantly increased. However, it is important to note that the slim possibility of these activities all occurring in a single year, combined with the occurrence of a strong El Niño in 1997–1998 with global impacts, suggests that 1998 is a result of climate change rather than human activity. Thus, the most significant factor for affecting hydrological variables in the years studied was the occurrence of the most powerful El Niño in 1997/1998 ever recorded around the world⁴⁷. Therefore, the break point detection methodologies from those years may suggest that the primary cause is more likely to be attributed to a severe climate change rather than human activity. The period of 1997/1998 is detected by many researchers around the world as the breaking point using various CPD methods.

In this respect, this study investigates various effective factors to streamflow changes across selected countries in the Middle East and Central Asia, with a particular focus on the Karkheh River Basin (KRB)-one of the most important agricultural regions in southwestern Iran. Besides frequent droughts, large-scale agricultural activities and dam construction programs in the last decades have caused serious streamflow reductions in the KRB. Despite the extensive application of change-point detection (CPD) techniques in hydrological studies, there remains uncertainty about the extent to which detected breakpoints reflect climatic versus anthropogenic drivers. Therefore, the objectives of this study are: (1) to detect the abrupt change points of streamflow across the broader region, and in streamflow, precipitation, and temperature specifically within KRB; (2) to analyze the relative contribution of climate and human activity changes on streamflow; and (3) to evaluate the effectiveness of CPD methods in determining when human activity becomes the primary factor influencing streamflow.

Study area and dataset

Study area

This study broadly focuses on the arid and semi-arid regions of the Middle East and Central Asia—two climate-sensitive regions. The Middle East in this context includes countries such as Iran, Iraq, Oman, Syria, Jordan, Turkey, Egypt, the United Arab Emirates, Israel, and Lebanon. Central Asia refers to Kazakhstan, Uzbekistan, Turkmenistan, Afghanistan, Tajikistan, and Kyrgyzstan, with Pakistan also considered part of this region due to its geographical and climatic similarities. These regions share common challenges such as increasing water scarcity, high evapotranspiration rates, and growing pressure on transboundary water resources. The overall spatial extent of the study domain is illustrated in Fig. 1, where the locations of the hydrometric stations used in the regional analysis are also marked.

Fig. 1

Overview of the study area across the Middle East, Central Asia, and Pakistan.

Datasets used in the study

Historical daily streamflow data for selected countries across the Middle East and Central Asia were obtained from two major international sources. The primary source was the Global Runoff Data Centre (GRDC, https://portal.grdc.bafg.de), from which discharge records were extracted based on the availability of sufficiently long and continuous time series suitable for trend and change point detection analyses. In regions where GRDC data were sparse or unavailable, supplementary records were collected from the Copernicus Emergency Management Service (CEMS) via the Global Flood Awareness System (GloFAS) historical dataset. This dataset provides quality-controlled daily discharge observations and is openly accessible through the Climate Data Store (https://ewds.climate.copernicus.eu). The geographic distribution of all hydrometric stations used in this study is shown in Fig. 1.

In Iran, and specifically in the Karkheh River Basin (KRB), more comprehensive hydro-climatic data were utilized. Daily streamflow records from sixteen hydrometric stations with long-term observations and good spatial coverage across the basin were analyzed to calculate mean annual streamflow and detect change points. In addition, climate variability—including trends in precipitation and temperature—was examined using both ground-based meteorological observations and reanalysis data. Due to data gaps in ground-based measurements during the study period (1970–2018), ERA5 reanalysis datasets from the European Centre for Medium-Range Weather Forecasts (ECMWF) were employed to ensure data continuity. ERA5 represents the fifth generation of ECMWF reanalysis and provides comprehensive global weather and climate data from 1940 onward (https://cds.climate.copernicus.eu). Areal precipitation for upstream sub-basins was estimated using the Kriging interpolation method in ArcGIS. Hydrological and meteorological data for Iran were acquired from the Iran Water Resources Management Organization (https://www.wrm.ir).

Methods

Change point detection

Most of the traditional models used in the simulation of hydrologic time series for applications of forecasting or synthetic data generation assume stationarity. However, in reality, most hydrologic variables - for example, runoff discharges - are naturally of a nonstationary nature due to modifications imposed by climate change and human activities⁴⁸.

Abrupt changes in hydrologic time series normally are attributed to human activities, while gradual variations are usually due to climate change^31,32. It is also assumed that the reference period is the period before the change point, during which time runoff can hardly be affected by human activities and thus human activity impact can be ignored during the reference period. According to the hypothesis, the runoff in the reference period is only subjected to the impact of climate change. The subsequent period is considered as human activities interference period (impacted period), in which the impacts of human activities on runoff become remarkable, and runoff is subject to the influence of climate change and human activities in conjunction³⁴. For detection of breakpoint in the annual streamflow time series, five tests including SNHT²¹, Buishand test²⁰, Pettitt test²², Ordered Clustering (OC) and Cumulative Anomaly tests have been resorted. These tests are briefly explained in the following. In addition, Sen’s slope method⁴⁹ was employed as a non-parametric estimator of trends in annual streamflow and precipitation for the periods before and after the detected breakpoint.

Pettitt test

The Pettitt test²² has been used to identify the onset of a declining trend in streamflow, which allows for the division of the streamflow time series into two shorter periods - one before and one after the start of the decline. This division enables further hydro-climatic analysis to be conducted. For a given time series of continuous data ${\text{X(X}}_{\text{1}}{\text{,X}}_{\text{2}}\text{,}\ldots{\text{X}}_{\text{N}}\text{)}$, the Pettitt test is described stepwise as it follows:

Step (1) Initialize an empty list to store the test statistic $\:{U}_{t}$ for each year $\:t$ with $\:U\left[1\right]=0$.

Step (2) For t = 2 to N, calculate:

$$\:U\left[t\right]=U\left[t-1\right]+\sum\:(sgn\left({X}_{t}-{X}_{j}\right)\:\:\:\:for\:\:j=1\:to\:N)$$

(1)

where:

$$sgn\left( {X_{t} - X_{j} } \right) = \left\{ {\begin{array}{*{20}l} {1~} \hfill & {if\quad X_{t} > X_{j} } \hfill \\ {0~} \hfill & {if\quad X_{t} = X_{j} } \hfill \\ { - 1} \hfill & {~if\quad X_{t} < X_{j} } \hfill \\ \end{array} } \right.$$

(2)

Step 3. Find $\:{K}_{n}$:

$$\:{K}_{n}=\:{max}_{(1\le\:t\le\:N)}\left|U\left[t\right]\right|$$

(3)

Step 4. Calculate p-value:

$$\:p\stackrel{\sim}{=}2*exp\left(\frac{-6*{K}_{n}^{2}}{{N}^{3}-{N}^{2}}\right)$$

(4)

Step 5. If p < significance threshold (e.g., 0.05): reject null hypothesis ($\:{H}_{0}$: no change point exists) and identify breakpoint at $\:t={K}_{n}$.

Standard normal homogeneity test

The SNHT test developed by Alexandersson²¹, is applied to find out the location of a breakpoint in a rainfall time series. This test employs a statistic, $\:{T}_{x}$, which compares the average of the initial $\:x$ years with that of the last $\:N-x$ years and is expressed for a given time series of continuous data $\:{\text{X(X}}_{\text{1}}{\text{,X}}_{\text{2}}\text{,} \ldots {\text{X}}_{\text{N}}\text{)}$, step by step in the following:

Step (1) Calculate the mean ( $\:\overline{X}$ ) and standard deviation ( $\:S$ ) of the entire time series $\:X\left[\:\right]$.

Step (2) Initialize an empty list $\:T\left[\:\right]$ to store the test statistic $\:{T}_{x}$ for each year $\:x$.

Step (3) For each $\:x$ from 1 to $\:N-1$, do:

• 3.1. Calculate $\:\stackrel{-}{{z}_{1}}$ (mean difference for the first $\:x$ years):

  $$\:\stackrel{-}{{\text{z}}_{1}}=\frac{1}{\text{x}}\sum\limits_{\text{i}=1}^{\text{x}}\frac{({\text{X}}_{\text{i}}-\stackrel{-}{\text{X}})}{\text{S}}\:\:\:\:\:\:\:\:$$

  (5)

• 3.2. Calculate $\:\stackrel{-}{{z}_{2}}$ (mean difference for the remaining $\:N-x$ years):

  $$\:\stackrel{-}{{z}_{2}}=\frac{1}{N-x}\sum\limits_{i=x+1}^{N}\frac{({X}_{i}-\overline{X})}{S}$$

  (6)

• 3.3. Compute the test statistic $\:{T}_{x}$ and append $\:{T}_{x}$ to the list $\:T\left[\:\right]$ :

  $$\:{T}_{x}=x\stackrel{-}{{z}_{1}}+\left(N-x\right)\stackrel{-}{{z}_{2}}$$

  (7)

Step 4. Find the maximum value $\:{T}_{0}$ in $\:T\left[\:\right]$, and its corresponding year $\:{x}_{0}$.

Step 5. If $\:{T}_{0}$ exceeds the critical value (based on the sample size), reject the null hypothesis ($\:{H}_{0}$: no change point exists) and set $\:{x}_{0}$ as the breakpoint.

Buishand range test

The Buishand Range test, developed by Buishand in 1982, defines the adjusted partial sum. The steps used in finding a breakpoint by Buishand range test for a given time series of continuous data $\:{\text{X}\text{}\text{(}\text{X}}_{\text{1}}{\text{,X}}_{\text{2}}\text{,} \ldots{\text{X}}_{\text{N}}\text{)}$, are described as follows:

Step (1) Calculate the mean ($\:\overline{X}$) of the entire time series $\:X\left[\:\right]$.

Step (2) Initialize an adjusted partial sum array $\:{S}_{x}^{*}\left[\:\right]$ with $\:{S}_{0}^{*}=0$.

$$\:For\:x=\:1\:to\:N:$$

$$\:{S}_{x}^{*}=\sum\limits_{i=1}^{x}\left({X}_{i}-\overline{X}\right)$$

(8)

Step 3. Calculate the range $\:R$.

$$\:R=\frac{\left(\underset{1\le\:x\le\:N}{{max}}{S}_{x}^{*}-\underset{1\le\:x\le\:N}{{min}}{S}_{x}^{*}\right)}{S}$$

(9)

where $\:S\:$is the standard deviation of the time series $\:X\left[\:\right]$.

Step 4. Calculate the test statistic.

$$\:Test\:statistic=R/\sqrt{N}$$

(10)

Step 5. If test statistic exceeds the critical value from Buishand’s table for the sample size and significance level, reject the null hypothesis ($\:{H}_{0}$: no change point exists) and output the year with the maximum or minimum $\:{S}_{x}^{*}$ as the breakpoint.

Ordered clustering

For a given time series ${\text{X(X}}_{\text{1}}{\text{,X}}_{\text{2}}\text{,}\ldots{\text{X}}_{\text{N}}\text{)}$, the OC method⁵⁰ is explained step by step below:

Step 1. For each possible breakpoint $\:\tau\:$ from 2 to $\:N-1$:

• a. Compute the average of the first segment ( $\:{\overline{X}}_{\tau\:}$), and then calculate $\:{V}_{\tau\:}$:

  $$\:{V}_{\tau\:}=\sum\limits_{t=1}^{\tau\:}{\left({x}_{t}-{\overline{X}}_{\tau\:}\right)}^{2}$$

  (11)

• b. Compute the average of the second segment ($\:{\overline{X}}_{N-\tau\:}$), and then calculate $\:{V}_{N-\tau\:}$ :

  $$\:{V}_{N-\tau\:}=\sum\limits_{t=\tau\:+1}^{N}{\left({x}_{t}-{\overline{X}}_{N-\tau\:}\right)}^{2}$$

  (12)

• c. Calculate the total sum for this breakpoint:

  $$\:{S}_{N}\left(\tau\:\right)={V}_{\tau\:}+{V}_{N-\tau\:}$$

  (13)

Step 2. Find the breakpoint $\:{\tau\:}_{0}$ that minimizes $\:{S}_{N}\left(\tau\:\right)$:

$$\:{\tau\:}_{0}=argmin\left({S}_{N}\left(\tau\:\right)\right)\:\:\:\:\:\:for\:2\le\:\tau\:\le\:N-1$$

(14)

Cumulative anomaly

The cumulative anomaly is an index used to identify trends in discrete data, and its process for a given time series ${\text{X(X}}_{\text{1}}{\text{,X}}_{\text{2}}\text{,} \ldots {\text{X}}_{\text{N}}\text{)}$ is outlined step by step below⁴¹:

Step (1) Calculate the mean ( $\:\overline{X}$ ) of the time series $\:X\left[\:\right]$, and then initialize cumulative anomaly array $\:C\left[\:\right]$ with $\:{C}_{0}=0$.

Step (2) For each time step $\:t$ from 1 to $\:N$:

$$\:C\left[t\right]=\sum\limits_{i=1}^{t}\left({x}_{i}-\overline{x}\right)\:$$

(15)

This calculates the cumulative anomaly for each point, where positive values indicate values above the average, and negative values indicate values below the average.

Step 3. Analyze the cumulative anomaly curve by looking for the turning points in the $\:C\left[\:\right]$ array where the trend shifts from increasing to decreasing, or vice versa.

• A significant positive cumulative anomaly (upward slope) indicates a phase where values are above the mean.

• A significant negative cumulative anomaly (downward slope) indicates a phase where values are below the mean.

Step 4. If a clear turning point is identified, output the corresponding year $\:{t}_{0}$ as the breakpoint.

Results and discussions

Identification of abrupt change points in streamflow time series

To evaluate the temporal distribution of abrupt shifts in streamflow, change point detection was conducted across hydrometric stations located in the Middle East and Central Asia. The relative frequency of identified change points is presented in Fig. 2. A distinct peak is observed during the 1997–1999 period, with 1998 being the most frequently identified year of abrupt change in both regions. Notably, Pettitt’s test was the only method applied uniformly to all stations in this regional analysis, ensuring a consistent basis for cross-country comparisons. This non-parametric and widely used test was selected for its robustness to non-normal data and its suitability for detecting a single abrupt change in long-term hydroclimatic series.

The findings from this regional assessment align closely with previous studies conducted in various parts of the world, many of which reported significant shifts in streamflow during the late 1990s. For instance, Al-Hasani³⁶ identified change points between 1995 and 1998 in the Tigris River Basin in Iraq, while Saifullah et al.⁴⁴ reported a 1996 shift in the Kunhar River Basin, Pakistan. In China, several watersheds including the Minjiang³⁴, Taizi⁴⁶, and Taoer²⁷ River Basins experienced breakpoints between 1996 and 1998. Similar temporal patterns have been noted in Australia⁴², Italy³⁹, and Sri Lanka⁴³. These synchronous hydrological shifts across diverse climatic and geographic contexts reinforce the broader pattern observed in the present study, emphasizing the late 1990s as a critical turning point in global streamflow regimes.

Fig. 2

Relative frequency distribution of detected change points at hydrometric stations across (a) the Middle East and (b) Central Asia.

The results from the Karkheh River Basin (KRB) further reinforce these regional findings. Change point detection across 16 hydrometric stations—whose spatial distribution is shown in Fig. 3, and whose broader geographical context within Iran is illustrated in Fig. 1 ("Study area" section)—using five statistical techniques—Pettitt’s test, Standard Normal Homogeneity Test (SNHT), Buishand’s range test, Ordered Clustering, and the Cumulative Anomaly method—revealed remarkable agreement among methods. Thirteen of the stations exhibited a breakpoint in 1998 across all applied approaches, while three other stations showed change points in 1996 and one in 1994. This consistency underscores 1998 as a dominant turning point in the hydrological behavior of the KRB. Following this year, streamflows at nearly all stations declined significantly and failed to return to their pre-1998 levels. Consequently, the possible drivers of this sustained reduction are explored in the next section.

Fig. 3

Geographical distribution of hydrometric and meteorological stations across KRB.

Similar temporal shifts have also been documented across various Iranian watersheds. For instance, Kazemzadeh and Malekian⁵¹ determined that between 1994 and 1999, the river flow changed in the western Caspian Sea region across all hydrometric stations. Similarly, major changes were noted at hydrometric sites in the Zarrinehrood River watershed in 1995 and 1997 by Dariane and Pouryafar²⁴. Sharifi et al.⁵² discovered a dramatic shift in temperature and runoff time series of the Ghaleh-Shahrokh watershed in 1996 on central plateau of Iran. Moreover, the Minab River watershed in southern Iran had its first notable modifications in 2000, according to Mikaeili and Shourian⁵³. Furthermore, in the runoff time series for the Zayandehrud Basin, Jalali et al.⁵⁴ found a transition point in 1996, and Ansari Mahabadi and Delavar⁵⁵ determined that 1997 was the hydrological breakpoint for the upstream watershed of Boukan Reservoir. Nourani et al.³⁵ determined 1998 as the time of abrupt change point across the hydrometric stations for the Ahar Chay watershed.

The quasi-periodic El Niño–Southern Oscillation (ENSO) exerts a profound influence on global climate and hydrological patterns, with particularly strong effects during major El Niño episodes such as those in 1982/1983, 1997/1998, and 2015/2016^56,57. These events are known to disrupt atmospheric circulation and alter precipitation regimes across many parts of the world, often triggering ecological, economic, and social impacts. Numerous studies have documented hydrological shifts associated with both El Niño and La Niña phases in regions as diverse as Australia, South Asia, and the Middle East⁵⁸. In particular, the Middle East and Southwest Asia are especially vulnerable, having experienced some of the most severe droughts in recent decades during La Niña periods, such as 1999–2001 and 2007/2008⁵⁹.

Given this context, the abrupt change points identified in the present study—particularly the clustering around 1997–1998—are likely to be associated with large-scale climatic variability rather than local anthropogenic influences alone. This underscores the need for caution when interpreting statistical breakpoints as direct indicators of human-induced impacts. While change point detection methods are valuable tools for identifying shifts in hydrological regimes, their results should be contextualized within broader climatic frameworks, such as ENSO cycles, to ensure accurate interpretation and application in water resource assessments.

Change point of precipitation and temperature time series

To further investigate the drivers of streamflow shifts, the non-parametric Pettitt test was applied to the temperature and precipitation series from 1970 to 2018. The results for the KRB region are presented in Table 1. Precipitation records showed a sudden decrease in the years 1991, 1994, 1997, and 2007. However, the most pronounced and sudden changes within the ERA5 precipitation data indeed took place during the year 1994. The spatial distribution of the meteorological stations used in this analysis is shown in Fig. 3.

Table 1 Change-point of the annual precipitation and temperature at KRB stations change points showing rising temperature were observed in all stations in 1993 and 1997.

The findings indicate a broader trend towards higher temperatures in the climate regime, while also revealing that the impacts of climate change became increasingly apparent in this area during the 1990s. Numerous factors, including urbanization, increased cloud cover, concentrations of anthropogenic greenhouse gases, and global warming, are associated with the observed rise in air temperature. The temperature and precipitation change points observed in this study are in good agreement with findings from other studies conducted throughout Iran including Kazemzadeh and Malekian⁵¹, Salehi et al.⁶⁰, and Zarezadeh et al.⁶¹. As for temperature, Zarenistanak et al.⁶² discovered that for southwest Iran, the most positive shifts in mean, maximum, and minimum temperatures began in the 1990s. Their research also shows that the 1990s were the warmest decade in the 20th century in this region. Overall, Salehi et al.⁶⁰ found that a significant downward trend in annual and seasonal precipitation has taken place after the middle of the 1990s across Iran, except in spring. After 1990, there has been a noticeable increase in spring precipitation at certain stations, suggesting a shift in precipitation from winter to spring.

Although the Pettitt test identifies different breakpoint years for precipitation and temperature, these years consistently cluster within the same broader transition period. This reflects the inherent behavior of single-breakpoint detection methods, which tend to select one “best” year even when the underlying changes develop gradually over several years. As seen in Figs. 4b and 5b, temperature exhibits a progressive warming trend throughout the 1990s, with a more sustained and persistent shift toward higher values emerging in the late 1990s rather than at the earlier statistically detected year (e.g., 1993). In contrast, precipitation shows substantial interannual variability without a sharply defined transition. Consequently, differences in the detected change-point years are expected and should be interpreted as methodological sensitivity rather than as truly discrete climatic events.

The impact of climate change

This study extensively examined the climate conditions within the KRB, focusing on the principal variables of precipitation and temperature, which significantly impact the hydrologic cycle and agricultural activities. The temporal distribution of normalized annual streamflow, precipitation, and temperature data at PayePol and PolDokhtar is depicted in Figs. 4 and 5 as a representative sample, allowing a detailed comparison of hydrological and climatic patterns across the two stations. While there are differences between the observations and ERA5 data for precipitation and temperature, their trends are consistent. A clear relationship between precipitation and streamflow is observed during the period prior to the 1998 breakpoint. At both stations, increases (or decreases) in precipitation were accompanied by corresponding changes in streamflow (Figs. 4a and 5a). This is supported by the correlation coefficients, which were 0.64 at PayePol and 0.69 at PolDokhtar before 1998. The Sen’s slope estimates further confirm this consistency: before the breakpoint, both stations showed positive precipitation trends (0.97 mm/yr at PayePol and 3.10 mm/yr at PolDokhtar) and positive streamflow trends (0.51 and 0.38 m³/s per year, respectively). However, after 1998, this relationship weakened substantially. Streamflow continued to decline despite sporadic increases in precipitation. The post-1998 correlations dropped to 0.30 (PayePol) and 0.50 (PolDokhtar), indicating a reduced sensitivity of streamflow to precipitation variability. Sen’s slope results also illustrate this shift: at PayePol, precipitation showed a negative trend (–1.85 mm/yr) and streamflow a sharply negative trend (–2.49 m³/s per year). At PolDokhtar, although precipitation exhibited a modest positive slope (5.59 mm/yr), the streamflow trend was nearly flat (0.03 m³/s per year), suggesting that increased precipitation no longer translated into proportional increases in runoff. The results further showed that there was a 7% and 2.4% reduction in average annual precipitation at PayePol and PolDokhtar, respectively, after the breaking point, while the reduction in streamflow at these stations was much larger, at 59% and 40%.

Fig. 4

Comparing normalized values of observed streamflow with (a) precipitation, and (b) temperature at the PayePol station.

Figures 4b and 5b show that the annual average temperature is higher during the impacted period (after 1998) compared to the reference period (before 1998). The average annual temperature difference between the reference and impacted periods is 1.3 ℃ at PayePol and 1.4 ℃ at PolDokhtar, while in the entire KRB it is 1.7 ℃. This clearly reflects an intensified warming signal across the basin, consistent with the regional expression of global climate change, which could enhance evapotranspiration from vegetation, soil and water surfaces, resulting in greater runoff losses. Overall, temperature significantly influences streamflow changes by affecting both precipitation (including snowfall and rainfall) and water requirements of various users.

Fig. 5

Comparing normalized values of observed streamflow with (a) precipitation, and (b) temperature at the PolDokhtar station.

The average streamflow, precipitation, and temperature during the reference period (1970–1998) and the impacted period (1999–2018), as well as the relative percentage of changes between the two periods are shown in Table 2. As shown in Table 2, all stations experienced a rise in temperature and a decrease in both streamflow and precipitation during the impacted period compared to the reference period. The average decline in streamflow is approximately 53.6%, while precipitation has decreased by about 5.7%, and the temperature has increased by around 1.7 °C. Moreover, the specific yield of subbasins at the selected stations have declined in accordance with the streamflow by approximately 53.6% in average.

At first glance, such a large reduction in streamflow compared to a relatively modest decline in precipitation suggests the presence of hydrological amplification. This discrepancy can be explained by both climatic and anthropogenic factors. The values in Table 2 represent long-term averages between the pre-change (1970–1998) and post-change (1999–2018) periods. During the latter period, nearly two decades of human interventions—including dam construction, groundwater abstraction, and land-use changes—further reduced streamflow. In addition, interannual variations around the breakpoint year (1998) reveal that the hydrological shift was initially triggered by climate anomalies. For example, at the PayePol station, precipitation in 1999 was about 24% lower than in 1998 and 31% lower than in 1997, accompanied by a ~ 2 °C rise in temperature. Similarly, at the PolDokhtar station, precipitation decreased by 17% (1999 vs. 1998) and 28% (1999 vs. 1997), with a comparable temperature rise of about 2 °C. These concurrent reductions in rainfall and increases in temperature likely intensified evapotranspiration and reduced effective runoff.

Therefore, while the 1998 streamflow decline was primarily triggered by climatic changes, subsequent human activities appear to have amplified and sustained the reduction. Considering both influences provides a more comprehensive understanding of the observed magnitude of change.

Table 2 Average streamflow (Q), precipitation (Pr), and temperature (T) during the reference and streamflow changed periods and the relative percentage of changes between the two periods.

The impact of human activity

Increase of agricultural areas

In the KRB basin the increase of agricultural areas, both irrigated and rainfed, is the most important human factor that would reduce streamflow within the region. The increase of cultivation areas compared to the base year (1982–83) was analyzed during 1982–83 to 2015–16 water years (Fig. 6). The data on the cultivated areas in the KRB were obtained from the Ministry of Agriculture Jahad (https://www.maj.ir).

Seasonal precipitation has long been recognized as the primary factor influencing the annual variation in the proportion of cultivated land in different areas. Annual, spring, and autumn precipitation patterns are presented in Fig. 6a–c. The annual and seasonal correlations between the amount of precipitation and the growth of irrigated and rainfed cultivation were determined. The correlation between irrigated cultivation and precipitation is − 0.004, 0.001, and − 0.13 for annual, autumn, and spring precipitation respectively. As expected, the correlations between the amount of the autumn precipitation and the variations of irrigated cultivation are near zero and negligible. In fact, in irrigated cultivation people rely on their water rights from groundwater, Qanat, and riverflow rather than the precipitation. In contrast, there is a strong correlation between the growth of rainfed cultivation and early spring and autumn precipitation as well as Oct-Mar. precipitation. To be precise, early spring (Mar. and Apr.), autumn (Oct. and Nov.), and October to March precipitation shows relatively good correlations of 0.55, and 0.34, and 0.60, respectively.

Fig. 6

The development process of agricultural lands and the amount of (a) annual, (b) Spring, and (c) Autumn precipitation during the period in the KRB basin.

To more rigorously examine precipitation–cultivation relationships, a cross-correlogram analysis was performed (Fig. 7). Because future precipitation cannot influence past cultivation decisions, only meaningful lags (– 3 to 0 years) were evaluated, representing the effect of antecedent precipitation on current-year cultivation. The results show that rainfed cultivation responds most strongly to precipitation in the same year (lag 0), which is consistent with farmers’ reliance on rainfall conditions from autumn through early spring when determining the extent of rainfed planting. In contrast, irrigated cultivation exhibits no meaningful correlation with precipitation at any lag, reflecting farmers’ dependence on stable water rights from groundwater, qanats, and streamflow rather than rainfall variability.

It is also evident from Fig. 6 that the areas of rainfed cultivated lands varies over the years due to the precipitation as mentioned above. In years with higher precipitation, the areas of rainfed cultivation have also increased, and conversely, it has declined in drier years. Notably, in 1998–99 and 2007–08, there were significant decreases in the area dedicated to rainfed cultivation, with a particularly steep drop of 20.2% in 2007–08. Although the percentage of irrigated cultivation generally increased over the specified period, the intensive growth of it began from 2001 to 02, with a share that was of 67% higher than the base year. As discussed in "Identification of abrupt change points in streamflow time series" section, the methods used to determine the breaking point have frequently identified 1998 as the year when human activities had an impact, while the area of irrigated cultivation remained almost stable from 1991–92 to 2000–01.

Fig. 7

Cross-correlograms between precipitation and cultivation areas: (a) rainfed, and (b) irrigated.

Temperature plays an additional role by increasing crop water demand. As shown in  "The impact of climate change" section, basin-wide temperature increased by 1–2 °C after the breakpoint. Using the CROPWAT model, three temperature scenarios were evaluated—representing current conditions (Scenario 1) and + 1 °C and + 2 °C warming (Scenarios 2 and 3). Results (Table 3) indicate that crop water demand rises by approximately 5–15% under the warming scenarios. These increases, combined with the continued expansion of irrigated areas, impose additional pressure on the basin’s water resources. A comprehensive model integrating all contributing factors is required to fully disentangle their combined effects on streamflow decline in the KRB.

Table 3 Water requirement of crops under different scenarios (mm/ha).

Overall, it appears that the 1998 breakpoint in streamflow across the KRB and other basins corresponds to a major climatic anomaly. Several studies conducted in Iran, the Middle East, Central Asia, and Pakistan have similarly reported this breakpoint, attributing it to climate change, particularly the strong 1997–1998 El Niño event, which triggered a shift in the hydrological regime. However, the subsequent two decades, extensive human interventions, including dam construction, groundwater abstraction, and land-use modifications, have increased and likely amplified and prolonged the decline in streamflow. These activities have altered the seasonal flow distribution, increased reservoir evaporation losses, and reduced downstream flow availability. Therefore, it is worthwhile to mention that the present study does not suggest that human influence is negligible; rather, it highlights that the initial hydrological shift in 1998 was primarily climate-induced, whereas the continuing reduction in streamflow during 1999–2018 reflects a combined climatic and anthropogenic impact. This interpretation is consistent with the broader objective of this research, which is to evaluate and refine methods for identifying hydrological breakpoints rather than to perform a detailed attribution analysis.

Conclusions

This study investigated abrupt shifts in annual streamflow across selected countries in the Middle East, Central Asia, and Pakistan from 1970 to 2018, primarily using the non-parametric Pettitt test. The results revealed that a large number of hydrometric stations experienced significant breakpoints in the late 1990s, with 1998 being the most frequently detected change year. This timing aligns with the strong 1997–1998 El Niño event, suggesting that large-scale climatic variability acted as an initial trigger of hydrological regime shifts across the region. These findings highlight the need for caution when interpreting streamflow shifts as direct evidence of human influence, as natural climate anomalies may play a dominant role.

A more detailed assessment in the KRB revealed the same breakpoint timing, accompanied by clear post-1998 warming and rapid agricultural expansion. These findings suggest that although climate anomalies likely triggered the initial shift, human activities have increasingly contributed to the persistence and intensification of streamflow reductions in the years that followed.

Overall, the results demonstrate both the utility and the limitations of current CPD techniques. While they reliably detect the timing of hydrological regime shifts, they cannot alone differentiate climatic from anthropogenic influences. This underscores the need for interpreting CPD outcomes in conjunction with complementary hydro-climatic indicators rather than relying on single-method detection results. Future work will require integrated hydro-climatic modelling frameworks that combine climate variability, land and water use changes, and their associated uncertainties, in order to support more robust interpretations of hydrological shifts and improve water-management strategies in vulnerable regions.