Introduction

Energy is a key component in human lives and the world’s economy, although it is considered one of the major causes for greenhouse gas (GHG) emissions and, consequently, the climate change (Edenhofer et al. 2011). Hence, attention has arisen over the impacts of climate change on physical and economic environments. Thus, researches have been carried out around the world to invent novel and advanced technologies to generate carbon-free electricity (Williams et al. 2012). This has led to deployment of renewable resources for the power generation, and countries like Denmark, Ireland and Spain expect to yield more than half of their power generation requirement through renewable sources within next two decades (Smith 2018). Hydropower is used mainly as a dispatchable power generation for integrating renewable sources in electric grid and carbon reduction (Tarroja et al. 2016). Currently, 17% of the world’s electricity generation is streamed from hydroelectric power (van Vliet et al. 2016). Like other developing countries, Sri Lanka too heavily depends on hydropower generation to fulfill its energy requirement (45–50% of total electricity demand from 2010 to 2015) and highly favorable to generate hydropower in the future (Khaniya et al. 2018).

Run-of-the-river (ROR) hydropower plants worldwide are capable of generating about 200 GW of hydroelectricity with special interest identified in developing countries in Asia and Africa (International Energy Agency 2016). However, interest in these projects has diminished since past studies have shown that global warming and climate variability may have severe impacts on available water resources for hydropower generation (Schär et al. 2004; Lehner et al. 2005; Hamududu and Killingtveit 2012).

Kao et al. (2015) studied on projected climatic changes in the future annual and regional hydropower generation with developed climate models and runoff-based assessment approach. Results showed that although there is a decrease in annual power generation, it appears to be marginable. Majone et al. (2016) have predicted that there will be an 2–6% increase in water yield during 2040–2070 compared to later 20th century in the southeastern Alpine region due to lesser winter precipitation. Partial equilibrium bottom-up optimization model (TIMES_PT) was used to assess the impact of climate change on power generation in Portugal’s water resources by Teotónio et al. (2017). They derived that there will be a 41% decrement in hydropower by 2050, while electricity prices will be increased by up to 7.2%. In addition, they describe that the effect of climate change on generation of electricity based on non-thermal sources shows geographical variability due to forecasted changes in temperature and precipitation. However, Turner et al. (2017) showed that impact of climate variation and climate change would increase the hydropower generation in some parts of the world while decrease in some other parts of the world. Hamududu and Killingtveit (2012) have used simulations of regional patterns of runoff and related these to hydropower generation through geographical information systems (GIS) and showed that there will be fluctuations in power generation. Similarly, few other researchers have used different methods to predict the impact of climate change on hydropower generation. Global circulation models, which predict future climatic conditions using hypothetical scenarios, have been used by Aronica and Bonaccorso (2013), Shrestha et al. (2014) and Pilesjo and Al-Juboori (2016). Kabo-Bah et al. (2016) and Machina and Sharma (2017) have accessed the impacts of climate changes on hydropower generation with multi-year temperature and rainfall trend analysis. Khaniya et al. (2018) stated that this approach is easier and most suited given that the quality of data is at a satisfactory level. In addition, it states that this method depends on factors such as hydropower system installed, geographical coverage and availability of data for the analysis.

Sri Lanka being a tropical country is highly vulnerable to impacts of climate change and variability due to global warming and anthropogenic activities (Ministry of Mahaweli Development and Environment 2016). Over the past few decades, rainfall and temperature patterns in Sri Lanka have shown some variations and these have been resembled through the decrease in cumulative rainfall from southwest monsoon (SW) and northeast monsoon (NE), while an increment has observed in the intermediate season rainfall (Jayasundara and Shantha 2005; De Silva 2006; Iimi 2007; De Silva et al. 2007). According to Eriyagama et al. (2010), Sri Lanka’s mean air temperature has increased by 0.016 °C per year and mean annual precipitation has decreased by 144 mm per year in between the periods of 1931–1960 and during the period 1961 to 1990. In addition, they have predicted that due to the change in volume and distribution of rainfall, there would be an increment of 0.9 to 4 °C of mean temperature by 2100.

Although there is no exact quantitative justification, experts have suggested that there would be an increase in rainfall in wet zone, while a progressive decrease is expected in dry zone (Marambe et al. 2014). However, at present, no concerns have arisen regarding power deficit since most of the hydropower plants are located in wet zone (Ceylon Electricity Board 2014). Thus, there is no conclusive evidence on what will exactly happen in the future. Nevertheless, it is essential for Sri Lanka too to move on with world to study on climate change, depletion of water resources and power management for a better economy. However, studies on impact of climate change on hydropower production from river basins in Sri Lanka are rare in the literature. According to Jayasundara and Shantha (2005), there was a 39.12% reduction in rainfall from 1902 to 2002 and they have forecasted that there would be a 16.6% reduction by 2025 on Mahaweli river upper watershed area. Herath and Ratnayake (2004) showed that there is a significant decrease in rainfall over 60 rainfall stations from 1964 to 1993 period. However, they found that the intensity of rainfall was increasing. In addition, Ranatunge et al. (2003), Jayawardene et al. (2005), Wickramagamage (2016) and Karunathilaka et al. (2017) have put efforts in studying annual and seasonal rainfall trends in Sri Lanka. (Khaniya et al. 2018) presented some interesting rainfall trend analyses and linked them with the hydropower generation of a ROR hydropower plant. However, the literature does not provide any similar analyses in Sri Lanka for un-gauged catchments in some other parts of the world. Singh et al. (2001) and Kebede et al. (2011) have studied the impact of climate change on hydropower generation in un-gauged catchments using rainfall data from the surrounding rain gauges in Himalaya and Tana floodplains and Ethiopia, respectively.

Therefore, this paper presents the impact of climate change on hydropower generation in one of the un-gauged catchments in Sri Lanka, Erathna hydropower plant. This hydropower plant is an ROR hydropower plant, and it is becoming popular in Sri Lanka (Weerakoon and Rathnayake 2007). Mann–Kendall test and Sen’s slope estimator test were used to analyze the trends, while Pearson’s correlation coefficient was used to identify linear relationship between rainfall and power generation.

Homogeneity tests for rainfall data series

Trend analysis of climatic data totally depends on the quality of the data series which was used. Therefore, the reliability of the recorded climatic data is highly important. Thus many researches check the homogeneity of the climatic data series before it is used for any potential trend analysis (Alexandersson 1986; Kang and Yusof 2012; Alghazali and Alawadi 2014; Al-lami et al. 2015; Agha et al. 2017). Different tests are used to test the homogeneity of data series used for climatic studies. These tests help to validate that the same instrument was used to collect data within same time period at the same location (Alexandersson 1986; Alexandersson and Moberg 1997). Pettit’s test, Buishand’s test, Standard Normal Homogeneity Test (SNHT) and von Neumann’s test are several homogeneity tests used (Buishand 1982; Alexandersson 1986; Alexandersson and Moberg 1997; Wijngaard et al. 2003; Haylock et al. 2008; Sahin and Cigizoglu 2010).

Mann–Kendall test

Mann–Kendall test is widely used trend analysis test in climatic data series. Mann (1945) first introduced this test to analyze the trends; however, it was implemented in 1975 (Kendall 1975). The nonparametric test was further improved with seasonality by Hirsch et al. (1982).

This test is commonly known as Mann–Kendall test and, as stated above, widely used to check temporal monotonic upward and downward trends of climate (Robson et al. 1998; De Luís et al. 2000; Cannarozzo et al. 2006; Kumar et al. 2006; Longobardi and Villani 2009; Ahmad et al. 2015; Machina and Sharma 2017; Sridhar and Raviraj 2017). Equation (1) gives the Mann–Kendall statistic S.

$$S = \sum\limits_{i = 1}^{n - 1} {\sum\limits_{j = i + 1}^{n} {\text{sgn} \left( {x_{j} - x_{i} } \right)} }$$
(1)

where S, \({X_{j}}\) and \(X_{i}\) are Mann–Kendall’s statistic S, time series and \(n\) is the number of data points in the time series. The “sgn” sign function can be expressed as given in Eq. (2).

$$\text{sgn} \left( {x_{j} - x_{i} } \right) = \left\{ {\begin{array}{*{20}l} { + 1,\,\, > \left( {x_{j} - x_{i} } \right)} \hfill \\ {0,\,\, = \,\left( {x_{j} - x_{i} } \right)} \hfill \\ { - 1,\,\, < \left( {x_{j} - x_{i} } \right)} \hfill \\ \end{array} } \right.$$
(2)

The variance of the Mann–Kendall’s test is given by Eq. (3)

$${\text{Var}}\left( S \right) = \frac{{n\left( {n - 1} \right)\left( {2n + 5} \right) - \sum\nolimits_{i = 1}^{m} {t_{i} \left( i \right)\left( {i - 1} \right)\left( {2i + 5} \right)} }}{18}$$
(3)
$$Z_{\text{c}} = \left\{ {\begin{array}{*{20}c} {\frac{S - 1}{{\sqrt {{\text{Var}}\left( S \right)} }},\,\,S > 0} \\ {0,\,\,S = 0} \\ {\frac{S + 1}{{\sqrt {{\text{Var}}\left( S \right)} }},\,\,S < 0} \\ \end{array} } \right.$$
(4)

where \(t_{i}\) is the number of ties up to sample \(i\). Then, the Mann–Kendall’s statistics Z is given by Eq. (4).

The \(Z_{c}\) follows the standard normal distribution. A positive \(Z_{c}\) value shows an upward trend, whereas a negative \(Z_{c}\) gives a downward trend for the data period.

Sen’s slope estimator test

Sen’s slope is used to assess the magnitude of trend at a given time (Sen 1968) and used widely in rainfall trend analysis (Kumar et al. 2006; Ahmad et al. 2015; Sridhar and Raviraj 2017). Equation (5) is used to calculate slope for all data pairs.

$$d_{\text{k}} = \frac{{X_{j} - X_{i} }}{j - i}$$
(5)

for (\(1 \le i < j \le n),\) where \(d_{\text{k}}\) is the slope, \(X_{j}\) and \(X_{i}\) are data values at time \(i\) and \(j\), respectively, and \(n\) is the number of data. The median of \(n\) values of \(d_{k}\) is given as Sen’s slope estimator (\(Q_{i} )\) and given by Eq. (6).

$$Q_{i} = \left\{ {\begin{array}{*{20}c} {d_{{\frac{n + 1}{2}}} ,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,n\,\,is\,\,{\text{odd}}} \\ {\frac{1}{2}\left( {d_{{\frac{n}{2}}} + d_{{\frac{n + 2}{2}}} } \right),\,\,\,\,n\,\,is\,{\text{even}}} \\ \end{array} } \right.$$
(6)

Positive \(Q_{i}\) values suggest that there is an increasing (upward) trend in climatic data series, while negative values suggest the opposite.

Relationship between time series

The Pearson’s correlation coefficient is used to find the linear relationship between two data series. This is widely used for the statistical analysis of climatological and geophysical data (Puth et al. 2014; Wiedermann and Hagmann 2016; Jilin 2017; Ahmed and Kumar 2018). The linear correlation coefficient (r) is given by Eq. (7).

$$r = \frac{{n\sum {xy - \left( {\sum x } \right)\left( {\sum y } \right)} }}{{\sqrt {n\left( {\sum {x^{2} } } \right) - \left( {\sum x } \right)^{2} } \sqrt {n\left( {\sum {y^{2} } } \right) - \left( {\sum y } \right)^{2} } }}$$
(7)

where \(x = \left\{ {x_{1} , \ldots ,x_{n} } \right\}\), \(y = \left\{ {y_{1} , \ldots ,y_{n} } \right\}\) are two sample data sets and \(n\) is the number of pairs of data. The value of \(r\) varies such that \(- 1 \le r \le + 1.\) where + and − signs indicate positive and negative linear correlations between \(x\) and \(y\), respectively.

Case study application—Erathna mini-hydropower plant

Erathna mini-hydropower station is in Ratnapura district, Sri Lanka. Ratnapura district is in wet zone of the country. This area is highly vulnerable to frequent flooding due to higher rainfall volumes (Eriyagama et al. 2010; Karunathilaka et al. 2017). Rathnapura district receives a significant rainfall from SW monsoon from May to September (Punyawardena and Cherry 1999). However, district is famous for receiving rainfall throughout the year. The hydropower plant is a ROR type and utilizes the water flow from upper reaches of the Kuru Ganga. Figure 1 shows the catchment area of Erathna hydropower plant and its location within Sri Lanka.

Fig. 1
figure 1

Erathna catchment and its surrounding rain gauges

Full size image

Rainfall data for ten rain gauges were collected for 30 years (1988–2017) from the Meteorological Department, Sri Lanka, for the trend analysis. These rainfall gauges are shown in Fig. 1. They are Anhetigama estate (6.93 N, 80.37 E), Pussella S.P. (6.80 N, 80.35 E), Keragala (6.78 N, 80.35 E), Maliboda (6.88 N, 80.43 E), Galaboda estate (6.70 N, 80.47 E), Alupolla group (6.72 N, 80.58 E), Hapugastenna estate (6.72 N, 80.52 E), Laxapana (6.90 N, 80.52 E), Maskeliya Hospital (6.83 N, 80.57 E) and Maussakelle (6.85 N, 80.55 E). Missing rainfall data due to instrument issues and recording issues were filled using normal ratio method. Then, the Pettitt’s test, SNHT, Buishand’s test and von Neumann’s test were used to check the homogeneity of data series. Mann–Kendall test and Sen’s slope estimator test were applied on ten rainfall series on monthly, seasonal and annual basis to find any possible trends. The rainfall data series show that the area receives a significant rainfall throughout the year. This justifies the construction of Erathna ROR hydropower station. In addition, Pearson’s correlation coefficient was computed between each rainfall series with the hydropower generation. This helps in linking the un-gauged catchment with the rainfall data from the surrounding rain gauges. The correlation coefficients were found for the data series from January 2010 to December 2017. The power generation data were obtained from Vallibel Power Erathna (2018) (http://www.vallibel-hydro.com/projects/erathna-hydro-power.php).

Results and discussion

Figure 2 shows the annual rainfall variation over 30 years in the surrounding rain gauges of Erathna catchment. The figure clearly shows the zigzag pattern of the rainfall over the years. However, the annual rainfall in a particular rain gauge does not show significant sudden drops and rises other than in few cases (circled) for rain gauges in Keragala (years 1988 and 2004) and Maliboda (year 2010). These two gauging stations have extremely high annual rainfalls in those 2 years compared to the other corresponding rain gauging stations.

Fig. 2
figure 2

Annual rainfall variation over 30 years

Full size image

Homogeneous test results for the monthly rainfalls show that rainfall data are homogeneous. Therefore, the rainfall data can be used for the further analysis to find trends. The Mann–Kendall test results and Sen’s slope estimator test results for annual rainfall for 30 years are tabulated in Table 1.

Table 1 Mann–Kendall test and Sen’s slope estimator test results for annual rainfall
Full size table

Among the rain gauges, only four rain gauges show significant trends in annual rainfall. Hapugastenna estate, Laxapana, Maskeliya and Maussakelle are these rain gauging stations. All other gauging stations do not show any potential trends in annual rainfall. Interestingly, the all four significant trends are negative trends. Therefore, this analysis shows that there is a decrease in the annual rainfall in the above-stated rain gauging stations. Sen’s slope estimator gives these decreases in numerical values. For example, Laxapana rain gauge has a Sen’s slope of − 37.3. This means the Laxapana area has a decrease of 37 mm of rainfall annually over the 30 years. This is a significant decrease in rainfall per year. However, this trend results are spread over a year span. Therefore, the trend analysis was carried for lower spans in time.

Nevertheless, an interesting feature can be observed for these four rain gauges which have negative trends. They are in the eastern side of the Erathna catchment. In addition, the trend analysis carried out for higher resolutions for the seasons and months shows the negative trend occurs in the SW monsoon of the year. This can be further looked at the higher resolution for the months, and interestingly, it was found that the month of July is the critical month to have the negative trends. However, Samanala mountain range (including the world-famous Adam’s peak) lies in between the Erathna catchment and the concerned four rain gauges (Laxapana, Maussakelle, Maskeliya and Hapugastenna). The Samanala mountain range has an elevation around 2000–2250 m, while Laxapana, Maussakelle, Maskeliya and Hapugastenna have elevations around 1300–1400 m, 1200–1300 m, 1000–1100 m and 1150–1300 m, respectively. Therefore, these elevations clearly show that the SW monsoon is covered for the four rain gauges by Samanala mountain range. Figure 3 presents this scenario. Therefore, even if it is not a sound conclusion, it can be presented herein that the SW monsoon is getting weaker to break the Samanala mountain range due to ongoing climate variability. However, it is highly encouraged to complete a detailed study to obtain rich and sound conclusions on this matter. Nevertheless, it is very important to have such finding as the concerned area is having its maximum rainfall during the SW monsoon.

Fig. 3
figure 3

Erathna catchment and Samanala mountain range

Full size image

Table 2 shows the trend analysis results in the seasonal basis. As it was stated above, Sri Lanka experiences two major monsoon seasons and two intermediate seasons (SW and NE monsoons and 1st and 2nd intermediate seasons).

Table 2 Mann–Kendall test and Sen’s slope estimator test results for seasonal rainfall
Full size table

The table shows that there are several significant trends in some of the rain gauging stations but for SW monsoon and 1st intermediate season. Interestingly, seasonal rainfall shows one positive trend in 1st intermediate season for Pussella rain gauging station. This is around 8 mm increase per the season. (The season lasts for only 2 months, and therefore, the increase can be a considerable rainfall.) However, Laxapana and Maussakelle experienced significant rainfall decrease only in the southwestern monsoon period. Sen’s slope estimator test shows the decreases are around 30–37 mm per the season. Interestingly, the wet zone of the Sri Lanka receives its maximum rainfall during the southwestern monsoon period, and therefore, the decrease in rainfall during the same period shows a concerned climatic variation. In addition, these climatic variations can adversely impact the water resources not only in this area but also in the downstream catchments. The rain-gauged areas are the primary catchment areas for few major rivers in Sri Lanka. Therefore, the results from the seasonal trend analysis illustrate the significance of the carried research work. In addition, the rainfall decrease can be clearly seen for the Laxapana and Maussakelle rain gauges for both annual and seasonal trend analyses. Furthermore, these decreased rainfall values are almost similar in the annual to seasonal. For example, Laxapana has a decrease of 37.3 mm per year in the annual trend analysis, while it has 36.6 mm in the SW monsoon season. However, other seasons do not show any significant trends in rainfall. Therefore, the SW monsoon trend is the only trend for the year and that also represents the annual base. The similar observation can also be seen in Maussakelle rain gauge too.

However, this observation cannot be seen in the Hapugastenna and Maskeliya rain gauging stations. They show negative trends in annual rainfall trend analysis but no significant trends in the seasonal rainfall trend analysis. Therefore, the trend analysis was carried to further smaller time spans in the monthly base. The results from the analysis are given in Table 3. As it was expected and also shown in Table 1, Pussella rain gauge shows a positive trend in March. However, the rain gauge also shows a negative trend in July. The July negative trend is common for most of the other rain gauges. Therefore, the monthly rainfall trend analysis also shows the decrease in rainfall in southwestern monsoon period. In addition, Hapugastenna estate, Laxapana, Maskeliya and Maussakelle rain gauges show the usual decrease in rainfall similar to that shown in other time span trend analysis.

Table 3 Sen’s slope estimator test results for monthly rainfall
Full size table

These trend analyses clearly show the impact of climate variability over the past 30 years from 1988 to 2017. Therefore, the trend analyses prove the ongoing climate variability in the world. Identification of the impact of these climate variabilities on the available water resources would be interesting. Therefore, the following paragraphs present the impact of climate change on the hydropower generation in Erathna run-of-the-river hydropower plant.

Figure 4 demonstrates the generated hydropower from Erathna run-of-the-river hydropower plant since it was in operation (January 2010). The variations clearly reflect the available water resources in the catchment during the different months of the years. As it was expected, the power plant has lowered its power generation during the northeastern monsoon period. However, the power generation during the 1st intermediate season was gradually increased and reached the maximum during the southwestern monsoon period (May to September). Then, it has gradually decreased during the 2nd intermediate season. The variation follows the amount of rainfall received by the catchment area during the four seasons. Interestingly, drops can be observed in the power generation during the month of July in all the years compared to preceding (June) and following (August) months. Therefore, it is highly important to investigate the relationship of power generation to the rainfall trends in the month of July. As it was stated above (Table 3), there is a considerable decrease in rainfall in July over the past 30 years. However, these rain gauges are outside catchment but closely surrounded the catchment. The power generation results suggest that there is a link to the rainfalls recorded in the surrounded rain gauges.

Fig. 4
figure 4

Hydropower generation of Erathna over the years

Full size image

Figure 5 presents the annual power generation variation against the annual rainfall in the surrounding rain gauges during 2010 to 2017. Power generation variation shows couple of severe drops in years 2012 and 2016. The same observation can be seen in the rainfall variations too. Most of the rain gauges show the drops in 2012 and 2016. Similarly, power generation peaks can be seen in the years 2010, 2013, 2015 and 2017. The rainfalls during these years have significant increase compared to the preceding years. Therefore, the annual power generation variation and annual rainfall show a good agreement.

Fig. 5
figure 5

Comparison of power generation against the annual rainfall

Full size image

This relationship can further be seen in Fig. 6a, b. They present the monthly power generation variation against the monthly rainfall variation in years 2010 and 2017. They show clear monotonic relationship. Therefore, it can be concluded herein that the un-gauged catchment’s water resources availability was successfully investigated using the catchment’s surrounding rain gauges. However, to prove this scenario, a statistical analysis was carried out in order to find the Pearson’s correlation coefficient between the rainfalls and the power generations. The results are presented in Table 4.

Fig. 6
figure 6

Comparison of monthly power generation against the monthly rainfall

Full size image
Table 4 Pearson’s correlation coefficients of rainfall and power generation time series
Full size table

The coefficients are in between 0.55 and 0.8 which suggested that there is a clear link between the rainfall of the catchment’s surrounding rain gauges and the power generation. Coefficient of 1.0 suggests that there is a total positive linear correlation between the two time series, whereas − 1.0 suggests a total negative linear correlation. However, coefficient value of 0 suggests that there is no link between the time series. Therefore, the following table (Table 4) shows that there is a better correlation between the power generation with Laxapana, Maliboda, Galaboda estate, Hapugastenna estate and Maussakelle rainfalls. Out of these rain gauges, Laxapana, Hapugastenna estate and Maussakelle showed a clear reduction in rainfall for the southwestern monsoon. In addition, other rain gauges show some correlation between the power generation. Even though Anhetigama estate, Pussella and Keragala show a slightly lowered correlation coefficient (0.58, 0.55 and 0.58, respectively) compared to 0.7–0.8, these rain gauges clearly showed a negative rainfall trend for the month of July, which is in the southwestern monsoon period.

Interestingly, the trend analysis found out that the rainfalls in better correlated rain gauges have negative trends in the major rainfall season to the area. Therefore, this finding is significant. There are clear evidences of rainfall decrease in this monsoon (southwestern) period, and the impact of that has clearly affected the hydropower generation in Erathna run-of-the-river hydropower station. Therefore, the un-gauged catchment’s water availability is in a risk during its major rainfall season.

Conclusions

The trend analyses clearly show that there are decreases in annual rainfalls of Hapugastenna estate, Laxapana, Maskeliya and Maussakelle. However, seasonal trend analyses support only Laxapana and Maussakelle rainfall decreases. These decreases are observed in the major rainfall season to the area (southwestern monsoon). Therefore, results clearly show the impact of climate change/climate variability on the rainfall. In addition, results show a clear correlation between the rainfall and the power generation of the Erathna run-of-the-river hydropower plant. Therefore, the impact of climate change/climate variability on the generated hydropower can be projected. Furthermore, results validate the rainfall data usage of catchment’s surrounding rain gauges in such a study when the catchment is an un-gauged catchment. Therefore, the analysis can be handy for any future studies, where the catchment’s rainfall data are unavailable. However, the Erathna run-of-the-river hydropower plant is merely a new one which has 8 years of power generation data. Therefore, the study should be further broadened with more power generation data in the future. Nevertheless, the presented research work clearly shows the impact of the ongoing climate variability on the received rainfall and water resources.