3.1. Seasonal Cycle of Rainfall and Streamflow
shows the seasonal cycle of mean monthly streamflow and rainfall in CA, NA, PA, LA, and MA catchments of the MRB for the period 1990–2014. The close association between the annual cycle of streamflow and rainfall over catchments in the UMRB and LMRB can be found. For the UMRB, two dominant peaks of rainfall (June and October) are observed in the CA, NA, and PA catchments (Figure 2
a–c). During the SWM season (June to September), the maximum rainfall is recorded in June with relatively less rainfall observed during August and September in the UMRB. The lowest rainfall and streamflow of the CA, NA and PA catchments are recorded from December to February (Figure 2
a–c), which are responding to the cease of the NEM season.
The LA and MA catchments in the LMRB have recorded more rainfall and streamflow from November and January. According to Figure 2
d–e, the lowest streamflow of the LA and MA catchments are observed from May to September, and less rainfall can be found from February to September, except that there are small peaks in April both for rainfall and streamflow. Based on the station data analysis, we can find that the streamflow of each catchment is closely associated with rainfall variation over both the upper and lower reaches of the MRB. For the long-term average (1990–2014) of rainfall in MRB, annual total rainfall of the LMRB and UMRB are estimated to be 1823 and 2158 mm, respectively. Out of these, the LMRB received the highest seasonal rainfall during the NEM season, which is about 41.0% of the annual total. In addition, the LMRB received 30.5%, 16.9%, and 11.6% rainfall during the SIM, FIM, and SWM seasons, respectively. The UMRB recorded more than 50% of the total rainfall during the SWM (29.7%) and SIM (25.9%) seasons. On the other hand, the contributions of the NEM and FIM rainfall to annual total rainfall of the UMRB are 24.3% and 20.1%, respectively.
The total rainfall distribution in each season over the MRB during the period of 1990–2014 is displayed in Figure 3
. In addition, Table 2
provides the rainfall variation in selected catchments for different seasons. During the FIM season, the total rainfall is higher in the southwestern part of the catchment (NA, 924 mm) than rest of the MRB catchments (Figure 3
a), while most of the area received less than 350 mm rainfall in total (Table 2
). Due to the orographic rainfall on the windward side of the mountain ridge, the western and southern parts of the MRB receive more rainfall during the SWM season (Figure 3
b). The maximum and minimum rainfall during the SWM season is recorded in NA (1987 mm) and LA (271 mm) catchments, respectively.
c represents total rainfall distribution in the SIM season. In general, the whole basin receives more than 500 mm in total rainfall during the SIM season, regardless of the orographic influence of central mountains in Sri Lanka. In the SIM season, the NA (712 mm) catchment recorded maximum rainfall, while the lowest rainfall is observed in the CA (353 mm) catchment (Table 2
). According to Figure 3
d, the total amount of rainfall from the NEM season in the LMRB and eastern part of the UMRB is higher than rainfall received in the western and southern parts of the UMRB. During the NEM season, the LA catchment recorded the highest rainfall (923 mm), and the CA catchment receives the lowest rainfall (277 mm).
The seasonality of streamflow differs extensively among the basins and is influenced mostly by the local seasonal cycle of precipitation [45
]. To identify the seasonality of Mahaweli streamflow, the median, mean, minimum, and maximum, streamflow in each catchment for the four seasons are analyzed (Figure 4
). In addition, the 10th and 90th percentile of the seasonal streamflow are displayed by the lower and upper boundary of the box. For the UMRB, CA (9.73 m3
) and PA (66.43 m3
) catchments recorded the larger mean monthly streamflow during the SIM season than the other three seasons (Figure 4
a,c). The NA (28.2 m3
) catchment recorded the highest average monthly streamflow during the SWM season (Figure 4
b). The mean streamflow of the LA (17.1 m3
) and MA (231.8 m3
) catchments during the NEM season are higher than the other three seasons. Lowest monthly mean streamflow in the LA (0.85 m3
) and MA (22.08 m3
) catchments have been observed in the SWM season (Figure 4
The large variation between maximum and minimum streamflow is also observed in the SWM season for the UMRB catchments. The difference between minimum (2.6 m3 s−1) and maximum (16.2 m3 s−1) streamflow in the CA catchment during the SWM season is larger than other seasons. For the NA (PA) catchment, the minimum and maximum streamflows are 15.5 (16.4) and 38.4 (180.0) m3 s−1, respectively, which observed during the SWM season. In the same season, both catchments exhibit larger variation of streamflow as compare to other seasons. The LA catchment depicts large variation between maximum (39.1 m3 s−1) and minimum (3.7 m3 s−1) streamflow during the NEM season. Similarly, the MA catchment recorded a large difference between minimum (57.0 m3 s−1) and maximum (687.3 m3 s−1) streamflow during the NEM season compared to other seasons.
The CA catchment recorded the highest 95th percentile streamflow (15.8 m3 s−1) during the SIM season. The NA (32.4 m3 s−1) and PA (116.4 m3 s−1) catchments recorded the highest 95th percentile streamflow during the SWM season. The lowest 10th percentile streamflow for the CA (2.6 m3 s−1), NA (3.9 m3 s−1), and PA (10.3 m3 s−1) catchments are recorded in the FIM season. The highest 90th percentile streamflow in the LA and MA catchments is 33.5 and 441.2 m3 s−1, respectively, that recorded in the NEM season. The lowest 10th percentile streamflow for the LA (0.5 m3 s−1) and MA (39.1 m3 s−1) catchments are observed in the SWM season.
As extreme streamflow indicators, the average Max10-day flow and Max30-day flow for the FIM, SIM, SWM, and NEM seasons are shown in Table 3
. The largest average value of Max10-day flow in the CA (21.0 m3
), NA (76.3 m3
), and PA (134.8 m3
) catchments is observed during the SWM season. Compare to the other seasons, the average value of the Max30-day flow also recorded the largest magnitude during the SWM season for catchments in the UMRB. The lowest Max10-day flow and Max30-day flow of the CA and PA catchments are observed in the FIM season. The average Max10-day flow (14.7 m3
) and Max30-day flow (10.9 m3
) events in the NA catchment are lowest, which is recorded during the NEM season. As compared to the maximum value in a different season, the average Max10-day flow (70.0 m3
) and Max30-day flow (38.4 m3
) at LA catchment in the LMRB are observed during the NEM season. For MA catchment, the largest average value of Max10-day flow (Max30-day flow) is 759.1 m3
), which is recorded in the NEM season. The lowest Max10-day flow and Max30- day flow at both catchments in the LMRB are recorded during the SWM season (Table 2
). The results clearly show that the high streamflow extremes in the UMRB catchments are mostly found in the SWM season when heavy and very heavy rainfall occurs frequently. In the LMRB catchments, extreme high flow events can be found during the NEM season, the rainy season for the catchments. This is consistent with the result by Groisman et al. [46
], i.e., the variations of high and very high streamflow are closely associated with heavy and very heavy precipitation.
3.2. Inter-Annual Variation of Streamflow
Understanding of year-to-year variations in streamflow is important for the development and management of water resources in most regions. Considering the contribution of seasonal rainfall to the total annual rainfall, the SWM season for the UMRB and NEM season for the LMRB are important than the other two inter-monsoonal periods. Meanwhile, maximum streamflow in the UMRB (LMRB) catchments is recorded during the SWM (NEM) season, while the minimum streamflow in the UMRB (LMRB) catchments is observed during the NEM (SWM) season. Therefore, the SWM and NEM seasons are selected in this study.
The anomalous streamflow variation of each catchment during the SWM season is depicted in Figure 5
. For the UMRB, where rainfall is significantly influenced by the SWM, different features for the interannual variation of seasonally averaged streamflow can be found for different sub-basins. The streamflow of CA catchment indicates a positive anomaly from 1992 to 1996, with streamflow anomalies in 1993, 1996, and 2013 exceeding one standard deviation (σ). Generally, it is noticed that most of the years after 1997 inherited a negative anomaly of streamflow except 1998, 2006, 2010, and 2013 (Figure 5
a). The streamflow of the NA catchment shows a negative anomaly from 1991 to 2006 except 1992, 1993, 1995, 1998, and 2005. Among negative anomalies, streamflow anomaly in 1994, 2001, 2003, and 2014 are below the −1σ (5.2 m3
) level. Positive streamflow anomalies can also be found from the 2007–2013 period, except 2012. Furthermore, streamflow anomalies in 2007, 2010, and 2013 can exceed the +1σ level (Figure 5
In the PA catchment, streamflow during the SWM season shows a negative anomaly for most years from 1990 to 2008, except for 1992, 1993, 2004, and 2007 (Figure 5
c), with the magnitude of anomalies below −1σ (34.7 m3
) level. Based on all catchments in the UMRB, the year 2013 shows a remarkable positive anomaly of streamflow. The streamflow of the LA catchment shows a negative anomaly from 1999 to 2012 except 2006, 2007, and 2010, and 2009, 2011, and 2012 recorded negative anomaly below the −1σ (0.4 m3
). Considering the interannual variation of streamflow anomaly in the LA catchment during the SWM season, the positive streamflow anomalies can be found from 1993 to 1998, and for 2013 and 2014, with large than +1σ (0.4 m3
) positive streamflow anomaly found in 1995, 1998, 2013 and 2014 respectively (Figure 5
d). The MA catchment indicates a negative anomaly of streamflow below the −1σ (19.7 m3
) level in 1992, 1993, 1994, and 2014, while the positive anomalies exceeding the +1σ level are recorded only in 1995 (Figure 5
shows the anomalous streamflow at five catchments in the MRB during the NEM season. The magnitude of streamflow anomalies of the UMRB catchments during the NEM season is smaller compared to that during the SWM season (Figure 6
a–c). The contrast pattern is observed for the LA and MA catchments (Figure 6
d–e). The CA catchment showed the largest positive streamflow anomaly (5.5 m3
) in 2011, followed by 2013, while the year 2004 indicates a negative streamflow anomaly, which is below −1σ (2.7 m3
) level. For the NA catchment, larger than +1σ (3.4 m3
) positive streamflow anomalies are recorded in 1990, 2011, and 2013. In the PA catchment, the negative anomalies of streamflow are observed from 1990 to 2005 except 2004, while positive anomalies that exceed the +1σ (24.8 m3
) level are recorded in 2004, 2010, and 2013 (Figure 6
c). The positive streamflow anomaly of streamflow of the LA catchment exceeds +1σ (8.5 m3
) in year 1994, 2011 and 2013, while a strong negative anomaly (below –8.5 m3
) recorded in 1997, 2004 and 2009 (Figure 6
d). In the MA catchment, 1993, 1996, 2002, 2008, and 2014 exhibit negative anomalies of streamflow, which exceed the −1σ (140.3 m3
) level, while streamflow for the NEM season in 2011 and 2013 show large positive anomalies exceeding +1σ (140.3 m3
) (Figure 6
To identify the variability in each catchment, the variance (σ2
) has been calculated for the SWM and NEM seasons. The variance of the streamflow in CA, NA, and PA catchments is 10.24 m6
, 27.04 m6
, and 1207.4 m6
, respectively, for the SWM season. In the NEM season, the variance of the CA (7.29 m6
), NA (11.56 m6
), and PA (615.04 m6
) catchments are smaller than that in the SWM season. Based on variance, the strong interannual variation of mean streamflow at CA, NA, and PA catchments in the UMRB is observed during the SWM season. The strong interannual variation of streamflow during the NEM season is recorded in the LA (72.5 m6
) and MA (19,600 m6
) catchments (Figure 6
d–e) compared to the SWM season. As shown in the annual cycle of the rainfall over each catchment (Figure 2
), rainfall also showed peaks during the SWM and NEM seasons in the UMRB and LMRB, respectively.
Max10-day flow, Max 30-day flow, PQ5, and PQ95 are selected to analyze the interannual variation of extreme flow events in SWM and NEM season for selected catchments. The variation of Max10-day and Max30-days flows in the CA, NA, and PA catchments during the SWM season are shown in Figure 7
a–c. The lower two panels show the interannual variation of the aforementioned extreme flow indices in the LA and MA catchments during the NEM season.
According to the magnitude of an anomaly, Max10-day is larger than Max30-day in each catchment, and the interannual variation of both indices exhibit a similar pattern. In the CA, NA, and PA catchments, the recorded average Max10-day flow (Max30-day flow) are 20.9, 76.0, and 146.2 (14.5, 52.1, and 110) m3
, respectively. In terms of the Max10-day flow and Max30-day flow anomalies of catchments in the UMRB, the year 2003, 2008, 2011 and 2012 showed strong negative anomalies, while 1993, 1999 and 2013 showed positive anomalies (Figure 7
a–c). In general, the more negative anomaly of Max10-day and Max30-day flows in the CA catchment is recorded after 2000 except 2013. The same observation is recorded in the NA catchment. The positive anomalies of Max10-days flow in the NA, and PA catchments that exceed the +1σ are observed during the SWM season in 1993, 1999, and 2013 (Figure 7
b–c). According to the variation of Max10-day, and Max30-days flow over the study period, catchments in the UMRB recorded more negative anomalies in the SWM season during 1990-2014.
d shows the interannual variation of Max10-day and Max30-day flow in the LA catchment during the NEM season. The averaged Max10-day is 69.0 m3
in the LA catchment, and averaged Max30-day is 38.4 m3
. The anomaly of Max 10-days and Max 30-days flow in the LA catchment exceed the +1σ in 1993, 2005, 2010, and 2012. For 1996, 2008, and 2009, the anomaly of Max10-day and Max30-day flow is below the −1σ (Figure 7
d). In the MA catchment, the negative anomaly below the −1σ of Max10-day and Max30-day flow are observed in 1992, 1995, 1996, 2003, and 2008. However, the positive anomalies of these indices in MA catchment are observed in 2004, 2006, 2010, 2012, and 2014 (Figure 7
Although there exists strong interannual variability of Max10-day and Max30-day flows at both catchments in the LMRB, the magnitude of the variability is relatively higher after 2005 than that before 2005. This is different from the situation in UMRB, where the interannual variability is relatively larger before 2005. And this could be ascribed to the different changes of interannual variability of SWM and NEM monsoon rainfall.
a–c shows the interannual variation of the number of low flow days with streamflow below the 5th percentile and high flow days with streamflow larger than the 95th percentile during the SWM season. The lower two panels (Figure 8
d–e) are the same as above but for the NEM season. The 5th (95th) percentile flow in the CA, NA, PA, MA, and LA catchments are 2.7 (18.6), 2.1 (65.8), 8.1 (147.3), 10.1 (840.8), and 2.3 (57.3) m3
, respectively. The number of low flow days in the CA catchment is recorded more after 2000, while the number of days recorded high flow events are larger during 1990–2000 than the 2001–2014 periods. It also shows the decadal difference of the PQ5 and PQ95 in the SWM season (Figure 8
a). The Q95 flow at NA catchment recorded less than 15 days during the every SWM season. The PQ5 events in the NA catchment are higher during 1990–2005 (Figure 8
b) compared to the 2006–2014 period. In the PA catchment, PQ5 during the SWM season is higher in 1990–2005, while the maximum number of high flow days above the Q95 flow threshold in SWM season is observed in 2013 (57 days), followed by 2004 (20 days) and 2010 (17 days) (Figure 8
For the NEM season, the prevalence flow above Q95 in the LA catchment shows more events during the 1990–2000. In the year 2013, it recorded the maximum number of days (36 days). The recorded low flow events in the LA catchment during the NEM season are comparatively less in 2000–2014 except for 2008 and 2012 (Figure 8
d). Figure 8
e shows the prevalence flow of Q5 and Q95 in the MA catchment for the NEM season. In the MA catchment, 45 and 40 days in the year 1991 and 1992 are the recorded number of days, which are below the Q5 percentile flow event. However, most of the years after 2000 have zero-days, which recorded Q5 percentile flow events, except 2008 and 2009.
3.3. The Relationship Between Rainfall and Streamflow
Numerous studies in many different countries have conclusively shown that rainfall changes affect the streamflow [10
]. In Figure 9
, we take a closer look at the correlation between standardized rainfall and streamflow (Q50) in MRB catchments for the SWM and NEM seasons separately (Figure 9
and Figure 10
The statistically significant positive correlation between rainfall and Q50 in the CA (0.86), NA (0.69), and PA (0.56) catchments are observed for the SWM season (Figure 9
a–c). Considering the LMRB, LA catchment (0.72) recorded a statistically significant positive correlation between rainfall and streamflow for the SWM season (Figure 9
d). However, the corresponding correlation is not statistically significant for the MA (0.20) catchment (Figure 9
e). For the NEM season, the correlation coefficient of Q50 and rainfall at the CA, NA, and PA catchments in the UMRB are 0.82, 0.44, and 0.48, respectively, while the LA (0.78) and MA (0.62) catchments recorded statistically significant correlation between streamflow and rainfall (Figure 10
It is interesting to notice that the relationship between rainfall and streamflow in the UMRB during the SWM season is stronger compared to that in the NEM season. However, at the LA and MA catchments in LMRB, the rainfall and streamflow relationship is stronger during the NEM season than in the SWM season.
Abghari et al. [48
] found that strong relationships between river discharge and annual precipitation in Iran during the past 40 years. Abeysingha et al. [49
] revealed that the significant positive correlation between streamflow and catchment rainfall in the Gomti River basin. When considering the correlation coefficient and year to year variation, it can also be concluded that the rainfall over catchment is the major influencing factor for the streamflow variation in the MRB.
However, the correlation between seasonal rainfall and streamflow in the PA catchment is relatively smaller compared to those catchments in the upper reaches of the Mahaweli basin. Previous studies have suggested that the precipitation–runoff relationship could be modulated by human activities, such as the construction of reservoirs [50
] and the hydromechanics project [51
]. So the relatively weak rainfall-streamflow correlation could be ascribed to the effects of the construction and operation of the Kothmale (1979–1985) and Upper Kothmale reservoirs in the MRB, which can affect the streamflow in PA catchment. The weak relationship between streamflow and rainfall in the MA catchment is also observed, especially during the SWM season. The main factors, which contribute to show a weak correlation between rainfall and streamflow in the MA catchment, could be the river diversion project for irrigation purpose and reservoirs operation of Mahaweli hydropower complexes.
3.4. Streamflow Response to ENSO and Indian Ocean Dipole
It is noticed that the climate of several countries located in the Indian Ocean, as well as the entire globe, is modulated by the El Niño-Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) [52
]. According to Ward et al. [53
], ENSO has significant impacts on streamflow and flooding around the world. To understand the relationship between streamflow and the large scale circulation indices, the concurrent correlation coefficient has been evaluated using the Nino3.4 and DMI indices with streamflow in the SWM and NEM seasons separately.
shows the relationship between streamflow in each catchment and the large-scale SST anomaly signal in the tropical Pacific Ocean (Nino3.4) and the Indian Ocean (DMI) during for the SWM season. The statistically significant negative correlation can be found between the Nino3.4 and streamflow in the CA (–0.35), NA (–0.35), and MA (0.4) catchments for the SWM season. However, the correlation between Nino3.4 Index and streamflow is weak in the PA (–0.26) and LA (–0.22) catchments (Figure 9
c,d). The positive mode of the Nino3.4 (El Niño) shows a negative influence on streamflow in MRB, while the contrast response of streamflow in MRB for the negative Nino3.4 (LaNiña) is observed. Similar to our findings, the previous study suggested that the relationship between ENSO indices and Mahaweli streamflow was quite significant from January to September [33
] between 1954 and 1993, and El Niño (LaNiña) conditions are also closely associated with annual rainfall and streamflow for January to September in Kelani River basin in Sri Lanka [27
]. Furthermore, Ouyang et al. [54
] also pointed out that the streamflow during the LaNiña events in different river basins in China was higher compared to the El Niño event. In contrast, the total water flow of the Cauvery river basin in India was lower in the La Niña years than El Niño years as a result of the amount of rainfall received during the La Niña years was lower than rainfall in normal years [55
A strong negative relationship between the DMI and streamflow in the CA (−0.48), NA (−0.43), MA (−0.44), and NA (−0.48) catchments during the SWM season are observed except streamflow in the LA (−0.26) catchment (Figure 9
a–e). It revealed that a positive mode of the DMI shows a negative influence on streamflow in the MRB. For the SWM season, the strong relationship between the DMI and streamflow in the MRB has been observed as compared to the Nino3.4. In contrast, Sahu et al. [56
] found a positive correlation (0.36) between the DMI and Citarum River in Indonesia.
As shown in Figure 10
, the correlation between the Nino3.4 index and streamflow in each catchment for the NEM season indicates a weak negative correlation. Similarly, the relationship between streamflow and the DMI recorded a negative correlation. Furthermore, it’s found that the influence of the DMI and Nino3.4 on streamflow in MRB is stronger in the SWM season than in the NEM season. Moreover, the influence of the ENSO on streamflow in UMRB catchments during the SWM season is relatively dominant than the influence on streamflow in the LMRB catchments. The predominance of the observed significant correlation suggests that the phase and magnitude of the ENSO and DMI indices could be one of the factors of the streamflow variability in the Mahaweli River basin during the SWM season.
3.5. Influence of ENSO and DMI on Rainfall in MRB
The UMRB received more rainfall during the SWM season, which contributes to the enhanced streamflow in the Mahaweli River. However, the dry zone in Sri Lanka, including the Mahaweli development zones (MDZ), exhibits severe water shortage as a result of less rainfall over the area during the same monsoon period, so the Mahaweli water resource is diverted at Pollgolla barrage to overcome the water scarcity in the dry zone and MDZ. So it’s very important to understand the influence of SST anomalies on the SWM rainfall pattern in the Mahaweli river basin, in order for better water resource management. Therefore, the correlation between SWM rainfall in the UMRB and LMRB and SST is analyzed in this section. During the NEM season, the LMRB receives 41% total rainfall. Therefore, recognizing the relationship between two ocean indices is also beneficial to the NEM rainfall prediction and water resource management.
a shows the correlation of the SWM rainfall anomaly in the UMRB with SST anomalies in June, which is the starting month of the SWM over Sri Lanka. The lead-lagged correlation between the SWM rainfall in the UMRB and SST are shown in Figure 11
a1. The maximum correlation between the rainfall over the UMRB with Nino3.4 (−0.40) is observed in August. The month of June recorded the maximum correlation (−0.57) with the DMI and the SWM rainfall in the UMRB. Figure 11
b is the same as Figure 11
a but for the LMRB. The rainfall anomaly in the initial month of the SWM shows a strong correlation (−0.35) with the Nino3.4 index. Similarly, the correlation between the DMI and rainfall anomaly in the LMRB is −0.31. However, the positive correlation with SWM rainfall anomaly and the DMI is observed in lagged 3, 4, and 5 months for the LMRB (Figure 11
The spatial correlation between SST in December (Starting month of NEM) and rainfall anomaly of the NEM in theUMRB is shown in Figure 11
c, and associated lead-lagged correlation is displayed in Figure 11
c1. The correlation between SST in December with the NEM rainfall in the UMRB is −0.36, which is statistically significant, as shown in Figure 11
c1. However, the DMI showed weak correlation (–0.12) with the NEM rainfall anomaly in the UMRB for December. Figure 11
d shows the correlation between SST and NEM rainfall over the LMRB. In the LMRB, the correlation between Nino3.4 (DMI) and December rainfall is −0.16 (−0.03) while observed maximum correlation (0.45) with the DMI is recorded in the lead month of March (Figure 11
d1). Considering the line graphs, the negative relationships between rainfall anomalies in two monsoon seasons and the DMI are found. However, the DMI in lagged and lead months depict a positive correlation with monsoon rainfall anomaly except for the SWM season in the UMRB. Furthermore, the negative correlation between the Nino3.4 index and rainfall anomaly in both monsoon seasons are noted. Similar to our findings, De Silva M and Hornberger [57
] found a negative correlation between rainfall in the Kelani river basin Sri Lanka with the Nino3.4 and DMI indices. Addition, Li et al. [58
] revealed that a warm phase of the El Niño-Southern Oscillation (ENSO) and a positive Indian Ocean Dipole (IOD) diminish rainfall over the Indian subcontinent and southern Australia.