Spatial and temporal hydro-climatic trends in the transboundary Jhelum River basin

The concentration of greenhouse gases (GHGs), especially carbon dioxide, has increased dramatically during the last few decades, resulting in global warming and global energy imbalance. Anthropogenic forces such as burning of fossil fuel and biomass, land-use changes, rapid industrialization, and deforestation are considered to be the main reason for the increased GHGs in the atmosphere (Chu et al. 2010; Mahmood & Babel 2013). The consequence of these gases is an increase in the global (land and ocean) average temperature by 0.85 °C (0.65–1.06 °C) over the period 1880–2012 (IPCC 2013), and an increase of 0.74 ± 0.18 °C has been detected during the last hundred years (1906–2005) (IPCC 2007). This global warming is strongly projected to continue in the future, with an increase of 1.7–4.8 °C for the period 2081–2100 under the representative concentration pathways as compared to 1986–2005.

The projected increase in temperature is likely to disturb the global hydrological cycle. As a consequence, hydrological systems are likely to experience changes in average availability of water and changes in extreme events such as increase in precipitation intensity, frequency, and amount (Khattak et al. 2011). Such disturbances in hydrological systems can pose problems for public health, industrial and municipal water demands, water energy exploitation, and the ecosystem. However, the climate change impacts on hydrological systems may vary from region to region (Chu et al. 2010; Khattak et al. 2011; IPCC 2013). Thus, it is of great importance to assess hydro-climatic changes at the regional level, which can help in proper utilization and management of water resources. Hydro-climatic systems are of great importance because of their effects on the environmental and economic development of the region and are highly complex as they comprise the atmosphere, cryosphere, hydrosphere, biosphere, and geosphere. A hydrological cycle is mainly influenced by the physical characteristics, climatic conditions, and human activities in a river basin. Most studies on climate change have focused on temperature, precipitation, and evaporation (Wang et al. 2012) as these factors are considered to be the key indicative factors of the climate change and climate variability in a basin (Dhital et al. 2013). There is an increasing consensus that changing trends in climatic variables, especially temperature and precipitation, may change the hydrological and ecological conditions of a river basin (Gleick 1987). However, it is still unclear which climatic variable is the major driving factor of discharge in a basin, and how temperature and precipitation changes in a basin affect discharge (Wang et al. 2012).

Many studies have been reported that explore changing trends in temperature, precipitation, and discharge and their relationships throughout the world, e.g. in the USA (Kumar et al. 2009; Chen & Georgakakos 2014), Europe (Longobardi & Villani 2010; Reiter et al. 2012), Africa (Tekleab et al. 2013; Oyerinde et al. 2014), Asia (Fu et al. 2004; Darand et al. 2015), and South Asia (Khattak et al. 2011; Dhital et al. 2013). Trend analysis of the historical climatic data is a central process in assessing the state of climate of a region. It provides an overall estimate about the variations in climate variables during a specified time period. The main emphasis of trend analysis is to provide the general direction (increase or decrease) and change in magnitude but not to provide details of internal dynamics of climate, e.g. how the climate is changing (Ahmad et al. 2014).

Several parametric (e.g. T-test, F-test, and linear regression) and non-parametric (e.g. Mann–Kendall (MK) test (Mann 1945; Kendall 1975), Kruskal–Wallis test (Kruskal 1952), Sen's slope estimator (Sen 1968)) methods have been developed to detect trends in hydro-meteorological data (Wang et al. 2012; Ahmad et al. 2014). The parametric tests are applied when there are clear linear increasing or decreasing trends in datasets, and when the datasets are normally distributed with minimum outliers. Since long-term meteorological datasets (especially precipitation) are mostly non-normally distributed and rarely free of outliers, the non-parametric methods are preferred by several researchers for their simplicity and robustness against outliers. However, problems of detecting precise trends arise when data series contain inhomogeneity, serial correlation, skewness, and seasonality (Gilbert 1987; Ahmad et al. 2014).

Long-term climatic series that span from decades to centuries are required for trend analysis and climate change studies, and these are rarely free of outliers and inhomogeneities due to non-climatic factors such as changes in instruments, changes in surroundings, relocation of monitoring stations, and changes in observation methods (Li-Juan & Zhong-Wei 2012). These factors may hide true signals of climate variability and climate change and thus can potentially distort the conclusions of climate and hydrological studies. Thus, it is essential to produce homogeneous and quality-controlled climate records before using them in climate analysis (Costa & Soares 2009). For this purpose, several techniques, such as Buishand range test (Buishand 1982), Kruskal–Wallis test (Kruskal & Wallis 1952), MK test, Pettit test (Pettitt 1979), and the standard normal homogeneity test (SNHT) (Alexandersson 1986), have been developed for detection of irregularities in time-series and their adjustment.

Serial correlation and seasonality effects can also distort the true signals of climate change. For example, if there is positive serial correlation in a series, the trends will be overstated. The effects of these problems can be removed with the trend free pre-whitening (TFPW) method (Yue et al. 2002) and resampling method (Kundzewicz & Robson 2004) before applying any test for trend detection.

Although the detection of hydro-climatic trends has become an important issue in the utilization and management of water resources of a region, few studies (Archer & Fowler 2004; Fowler & Archer 2006; Khattak et al. 2011; Ahmad et al. 2014) have been reported in Pakistan. Archer & Fowler (2004), Fowler & Archer (2006), and Khattak et al. (2011) conducted their studies in the upper parts of the Indus basin, above the upper Jhelum River basin (UJRB), and Ahmad et al. (2014) in the lower parts of the Indus River basin, below the upper Jhelum basin. Archer & Fowler (2004), Fowler & Archer (2006), and Ahmad et al. (2014) detected trends only in temperature and precipitation data and Khattak et al. (2011) in temperature, precipitation and discharge data. In these studies, Archer & Fowler (2004) and Fowler & Archer (2006) used linear regression for the detection of trends, and Ahmad et al. (2014) and Khattak et al. (2011) used the MK test along with Sen's slope estimator. However, none of these studies checked datasets for outliers and inhomogeneities which can distort the results. To the best of our knowledge, no studies have been reported on detection of hydro-climatic trends in the UJRB, although this basin is the biggest tributary of the Indus River. The main reason is that this river is a transboundary river located between India and Pakistan, and it is not easy to access data from the Indian side.

Thus, in the present study, hydro-metrological trends in the UJRB were detected with the MK test, and Sen's slope estimator was used to find out the magnitude of trends. Before application of trend analysis, the meteorological time-series were checked for outliers and homogenized to remove irregularities (inhomogeneities or abrupt changes) caused by non-climatic factors, which can distort the results. In addition, the TFPW approach was also used to remove the effects of serial correlation and seasonality in hydro-climatic time-series. This study can provide significant benefits for the allocation and management of water resources of the Jhelum basin, which are highly vulnerable to climate change.

The average annual precipitation in the basin is about 1,258 mm, with the highest at Murree (1,736 mm) and lowest at Srinagar (746 mm), calculated for the period 1961–2009 (Table 1). The mean maximum and minimum temperatures in the basin are 23.1 °C and 10.9 °C, respectively, calculated for the period of 1971–2009. Jhelum is the hottest climate station in the basin, with a mean maximum temperature of 30.4 °C, and Naran is the coldest, with a mean minimum temperature of 2.7 °C, (Table 1). The mean annual stream flows at different sites are given in Table 2. Azad Pattan is the main hydrometric gauge, just above the Mangla reservoir; the annual flow at this gauge is about 829 m³/s (989 mm) calculated for the period 1978–2009, which represents 80% of the flow of the entire basin.

The observed daily maximum temperature (Tx), minimum temperature (Tn), and precipitation (Pr) data of 21 climate stations were collected from the Pakistan Meteorological Department (PMD), the Water and Power Development Authority of Pakistan (WAPDA), and the Indian Meteorological Department (IMD); the mean monthly discharge data of eight hydrometric stations were obtained from WAPDA. The data for Gulmarg, Kupwara, Qazigund, and Srinagar weather stations were obtained from the IMD. The PMD provided climate data for Astore, Balakot, Garidopatta, Kotli, Muzaffarabad, Murree, and Jhelum climate stations, and the remaining data were collected from WAPDA. Most of the precipitation time-series have data periods from 1961 to 2009, and most of the temperature time-series range from 1971 to 2009 (Table 1). The daily meteorological data were checked for quality control and converted into mean monthly time-series. Then, three relative homogeneity tests, the SNHT, Maronna and Yohai bivariate test (Maronna & Yohai 1978; Potter 1981), and Easterling and Peterson test (Easterling & Peterson 1995), were applied to the monthly time-series to find inhomogeneities in the datasets, and finally the inhomogeneous times were adjusted to remove the inhomogeneities. In the present study, season and annual hydro-climatic time-series calculated from the homogeneous monthly time-series were used for trend analysis. The geographic distribution of hydro-climatic stations is shown in Figure 1. This shows that most of the stations are located in the eastern parts of the basin and on lower altitudes. Basic information about meteorological and hydrometric stations is given in Tables 1 and 2, respectively.

There are many parametric and non-parametric methods for detection of trends in time-series data (Zhang et al. 2006). The parametric tests are more powerful than the non-parametric but require normally distributed data. Since hydro-climatic times series often do not fulfil the normality requirement, the non-parametric tests are considered to be more robust than the parametric (Hess et al. 2001; Khattak et al. 2011). In the present study, the MK test, a non-parametric test, was applied for the detection of hydro-meteorological trends. This method has been commonly used to identify the trends in hydro-climatic time-series as in Burn et al. (2004), Fu et al. (2004), Wang et al. (2012), Tekleab et al. (2013), and Oyerinde et al. (2014).

The MK test requires hydro-climatic time-series to be free of serial correlation and seasonality because these can exaggerate the significance level of the trends. Since the serial correlation and seasonality can distort the result obtained from the MK test, the effects of these two factors were removed by using TFPW, developed by Yue et al. (2002). This method has frequently been used to remove the effects of these factors from the time-series (e.g. Burn et al. 2004; Kumar et al. 2009; Khattak et al. 2011). The steps for TFPW are as follows:

• 1. Calculate the lag-1 serial correlation coefficient (r) at 5% significance level, using a two-tailed test: 

  

  7

  If (−1–1.96(n−1) ≤ r ≤ (−1 + 1.96 / (n−1), then time-series are considered to be free of serial effect at 5% significance level and no pre-whitening is required. If r crosses the upper and lower limits, then the series are considered to be serially correlated and pre-whitening is required.

• 2. Calculate non-parametric slope (β) using Equation (6) and then de-trend the series by using the following equation: 

  

  8

• 3. Compute the lag-1 correlation coefficient (r) of de-trended series by using Equation (7).

• 4. Remove the lag-1 autoregressive component from the de-trended series to get residual series as follows: 

  

  9
• 5. Then, add trend (β_i) back into the residual series to get a trend free pre-whitened series (y_i) as below: 

  

  10

Finally, apply the MK test on the trend free pre-whitened time-series (y_i) to find out the trends and Sen's slope to find out the magnitude or rate of change in a variable for a specified period.

Since serial correlation in time-series can exaggerate the true signals of trends, a total of 325 seasonal and annual hydro-meteorological time-series were checked for serial correlations and corrected with the TFPW method before application of the MK test. The TFPW method not only reduced the effect of serial correlation but also seasonality in the time-series. However, the results of annual Tx, Tn, and Pr time-series are presented in this study. Table 3 describes the serial correlations of annual time-series of Tx, Tn, and Pr before and after the application of the TFPW at a significance level of 5%. The significant serial correlations ranged from 0.35 to 0.84 before the application of the TFPW. It can be seen in Table 3 that serial correlation effect is highly reduced after application of the TFPW method.

Table 4 indicates the percentages of seasonal and annual hydro-climatic time-series that were serially correlated at 5% significance level. It was seen that 33% (110 out of 325) of hydro-climatic series had significant serial correlations, and about 42%, 38%, 17%, and 33% of them were related in Tx, Tn, Pr, and streamflow, respectively. This shows that Pr time-series were minimally affected by the serial correlation. Table 4 shows that about 22% to 52% of seasonal and annual series were detected as serially correlated in Tx, Tn, Pr, and streamflow, which can greatly exaggerate the final results of trends.

Trends in seasonal and annual Tn time-series are presented in Figure 2, and percentage of seasonal and annual time-series with positive or negative trends at different significance levels (<0.1%, 0.1%, 0.05%, and 0.01%) are described in Table 5. Although a large number of increasing and decreasing trends were observed at different sites in the whole basin, increasing trends dominated. Table 5 indicates that 47% to 68% of the seasonal time-series showed positive trends (non-significant and significant), and 11% to 33% (summer‒spring) of the stations indicated significant increasing trends. In contrast, the decreasing trends were detected on 32% (autumn and summer) to 47% (spring) of the stations in the basin. However, 5% to 21% of the stations showed significant decreasing trends. On the whole, about 60% to 70% of the stations showed non-significant or no trends, and 30% to 40% of the stations showed significant trends (increasing and decreasing). However, increasing trend slightly dominated in the basin, especially in spring and autumn. In case of annual time-series, 27% of the stations exhibited increasing and the same number of the stations showed decreasing signals. However, non-significant increasing trends were detected on 32% of the station and decreasing signals only on 5%. These results do not agree with Khattak et al. (2011)'s research conducted in the Upper Indus basin. They showed slightly decreasing trends in seasonal and annual Tn time-series. It might be due to two reasons: (1) the upper Indus basin is less populated and mostly covered with glaciers and (2) they did not check data for quality control and homogenization. Nonetheless, these results well match with Ahmad et al. (2014) reported in the lower Indus basin, below the Jhelum basin towards south. They showed a steady warming trend in seasonal and annual Tn time-series.

Table 5 describes percentage of the stations having positive, negative, or non-significant trends in seasonal and annual Pr, and spatial distributions of these trends are presented in Figure 2. In case of seasonal analysis, positive trends (significant and non-significant) were observed in 33% to 86% (spring‒winter) of the stations. However, 10% to 20% (spring‒winter) of the series only indicated significant positive signals. Very strong decrease trends were detected only on 5% and 10% of the stations in winter and spring, respectively. On the other hand, 14% (winter) to 67% (spring) of the stations showed decreasing significant and non-significant signals. However, only 15% (each of spring, summer, and autumn) of the stations exhibited significant decreasing trends. In annual analysis, most of the stations (about 86%) revealed non-significant trends, with 38% and 48% having non-significant positive and negative signals, respectively. However, 10% of the annual time-series displayed the very strong decrease trends. On the whole, most of the stations disclosed non-significant changes all over the Jhelum basin. These results are also supported by Khattak et al. (2011) and Ahmad et al. (2014).

In case of Tn, all season and annual series also showed positive trends. However, no strong evidence of warming was found except in autumn, where very strong trends (α = 0.01) were found, with 0.86 °C/39 yrs (Table 6). Autumn was also identified as most affected season by the linear regression method (Figure 4). An annual increase of 0.44 °C/39 yrs in Tn has been observed in the basin.

In case Pr, no strong evidence was found about increasing and decreasing trends in the basin. However, non-significant decreasing trends dominated in the basin (Table 6) as all seasons, except winter, showed non-significant decreasing trends (Table 6 and Figure 5). An overall non-significant decrease (NSD) of 111.3 mm/49 yrs has been found in the basin. It can be concluded that the basin will most likely be warmer in the future and likely drier which can produce many problems in different fields of life, such as agriculture, health, and water management.

Table 7 shows Sen's slope in seasonal and annual flow at different gauges in the Jhelum basin. Seasonal and annual streamflows are also presented in Figure 5, showing linear regression and Sen's slope lines. Decreasing seasonal and annual trends dominated at most of the hydrometric stations in the basin except in winter where four of eight stations showed non-significant increasing trends and one station (Kotli) showed strong increasing trends. This increase could be due to an increase in precipitation and rise in temperature in winter, as mentioned above. In spring, all stations showed non-significant decreasing trends. On the other hand, in summer (the peak flow season in the basin), non-significant to very strong decreasing trend were found at different hydrometric stations, with strong significant decrease trend at four stations and very strong decreasing trend at two of eight stations. Azad Pattan, where the flow is about 80% of the total flow from the Jhelum basin, showed strong decreasing signals in summer. Khattak et al. (2011) also showed decreasing signals in summer over the Upper Indus basin due to decrease in precipitation. Autumn also showed negative trends at all the stations, with significant decreases at Muzaffarabad and Gari Habibullah. Similarly, in annual case, all the stations except Kotli experienced decreasing signals, with strong decrease trends at Azad Pattan, Domel, and Naran and very strong decrease at Muzaffarabad. At Azad Pattan, 232 mm decrease of flow has occurred in the last 30 years.

Since Azad Pattan is the most important hydrometric station in the basin located just above the Mangla reservoir, seasonal and annual flows at this gauge are presented in Figure 5 to give more information about trends and variation in flow in the basin. It was observed that, except in winter, all seasons showed significant decreasing trends at Azad Pattan. These decreasing trends in flow can most probably be due to decreasing trends in precipitation. However, other factor such as land use changes, urbanization, and industrialization may also be the reasons for this decrease. It can be concluded that a NSD in precipitation can produce a strong significant decrease in the streamflow. This is an alarming situation for the reservoir operation and management because the Mangla reservoir is used for irrigation of 6 million hectares of land and hydropower production.

To investigate the linkages of stream flow with temperature and precipitation, correlations were calculated between streamflow at the main hydrometric stations (Azad Pattan, Kotli, Gari Habibullah, and Domel) and the corresponding meteorological stations located in their drainage areas. For example, the correlation at Kotli was obtained between the streamflow at Kotli and mean temperature of all climate stations located only in Kotli drainage area. The correlation coefficients of seasonal and annual streamflow with mean temperature and precipitation are described in Table 8. Streamflow in all seasonal and annual series showed inverse relation with mean temperature at all hydrometric stations. Seasonal flows did not show significant correlation at most of the stream gauges except in spring (at Azad Pattan, Kotli, and Domel) and in summer at Domel. However, in the annual case, mean temperature exhibited significant (α = 0.05) correlations with streamflow at all the gauges. Almost all seasonal and annual flows at all hydrometric station showed strong significant correlations with Pr, except in summer at Azad Pattan and Gari Habibullah. The highest significant correlation was observed at Kotli in spring, with 0.82, and the lowest at Domel, with 0.31.

From these results, it can be concluded that both temperature and precipitation have strong linkages with streamflow in the Jhelum basin. Thus, changing trends in temperature and precipitation may have serious implications for streamflow in the Jhelum River basin. So, it could be very difficult to manage and utilize the water resource of the basin without considering the changes in climate.

In the present study, hydro-meteorological temporal and spatial trends and their magnitudes were investigated in the Jhelum River basin by using the non-parametric MK, Sen's slope, and linear regression methods. The linkages of streamflow with temperature and precipitation were also examined. Before application of trends analysis, datasets were checked for quality control, homogenization, and serial correlation, which can distort true results. The major conclusions of this study are as follows:

• 1. Both maximum and minimum temperatures showed increasing trends in the Jhelum basin. However, the increasing trends of maximum temperature in all seasons as well as in annual datasets were stronger and statistically significant than minimum temperature. An overall change in maximum temperature and minimum temperature ranged from 1.75 °C/39 yrs (summer) to 2.8 °C/39 yrs (winter) and 0.14 °C/39 yrs (spring) to 0.86 °C/39 yrs (autumn), respectively. Significant increasing trends were observed in the western parts of the basin.

• 2. In case of precipitation, both increasing and decreasing trends were almost evenly spread throughout the basin, which did not give a clear picture about spatial patterns of precipitation trends in the basin. However, decreasing trends in seasonal and annual precipitation dominated in the basin, except for winter precipitation. An overall decrease of 111 mm was observed during the last 49 years in the basin. Nonetheless, the decreasing trends were statistically non-significant in seasonal and annual precipitation over the Jhelum basin.

• 3. In case of streamflow, seasonal and annual decreasing trends dominated in the basin. Summer showed more strong and significant decreasing trends at all hydrometric stations in the basin. The highest annual decrease was observed at Muzaffarabad, with annual decrease of 10 mm, followed by Azad Pattan, with annual decrease of 8 mm. These decreasing trends are most probably due to increasing trends in temperature and decreasing trends in precipitation because very strong negative and positive correlations have been found between temperature (0.02‒0.71) and precipitation (0.11‒0.82), respectively.

This study revealed some trends and changes in hydro-metrological variables and their linkage in the Jhelum basin. However, an intensive research is essential to explore the exact reasons for these change whether these changes have links with global or regional effects or due to some local effects such orography, land use changes, industrialization, and urbanization.

The authors would like to acknowledge the PMD, the WAPDA, and the IMD for providing important and valuable data for this research. This study is supported by Natural Sciences Fund of China (41471463) and CAS (Chinese Academy of Sciences) President's International Fellowship Initiative.