ENGINEER Evaluation of Climate Elasticity of Runoff based on Observed Rainfall, Streamflow and Simulated Future Streamflow using SWAT Model in Kelani Ganga Basin

: Kelani Ganga basin is the 7 th largest watershed in Sri Lanka, spanning over 2,292 km 2 and annually discharging 4,225 MCM flow to the sea. The basin currently hosts over 19% of the country’s population and is the primary source of drinking water to over 4 million people living in Greater Colombo. Hence, the present study was undertaken to evaluate the Climate Elasticity of Runoff based on observed rainfall, streamflow data and simulated future streamflow using the SWAT Model in the Kelani Ganga basin, targeting sustainable management of basin water resources in future. The runoff elasticity ( ε ) is assessed by two methods for the present and 2040 scenarios. The selected three hydrometric gauging stations exhibit significant downward trends for the period of 1980 to 2016. An 80% of the rain gauges in the middle and upper basin show significant decreasing trends for high to low rainfall totals for Yala season as per Innovative Trend Analysis (ITA) for the period of 1980 to 2016. Mass balance performance error ( Er ), Nash–Sutcliffe Efficiency ( NSE ) and Coefficient of determination ( R 2 ) are used as multi-objective functions and 8.90%, 0.65, 0.72 and 9.10%, 0.69, 0.69 are obtained for the above objective functions in SWAT model for the calibration and validation periods of 1970 to 1980 and 1982 to 1992, respectively. A 1 ⁰ C of temperature increase causes a 6.9% and 7.4% runoff decrease for the current scenario and it causes 0.4% increase and 1.5% decrease of runoff for Future Pessimistic Climate change Scenario as evaluated by the methods proposed by Zheng et al. [24] and Sankarasubramanian et al. [22], respectively. A 1% of rainfall increase causes a runoff increase of 0.002% and 0.370% for the current scenario and a runoff increase of 0.005% and 0.360% for 2040 as evaluated by the two methods, respectively. It is recommended to further analyse the water allocation model for better results with practical implementations by considering the identified trend after 1995 in future research for better planning and management of water resources in future.


Overview
The rapid population growth, urbanization and industrial expansion cause remarkable pressure on the available water resources in the Kelani Ganga basin. Climate change is an additional driver in the 21 st century [1]. Climate Change may affect water resources through impact on long-term water balance due to temperature changes, unusual spatio-temporal variability and sea-level rise, which lead to adverse implications for food security, water security, human livelihood and health, and ecosystems.
Surface temperature is currently increasing at 0.2 °C per decade and it is projected to increase by 1.5 °C by 2050, if the projected anthropogenic activities continue to increase at the current rate [2]. Thus, Climate change impacts will be a huge problem for developing countries because of their poor adaptation and mitigation measures to Climate change [3]. Sri Lanka also falls under this category, hence some effects may be irreversible or long-lasting, such as the loss of some ecosystems [2]. Climate elasticity of runoff defined as the proportional change in the runoff to the change in climatic variables such as Precipitation (P), Temperature (T) etc., can be used in identifying basinwide impacts due to future Climate variability.
Climate change impacts will intensify the water crisis as well as natural disasters in the Kelani River basin in future. Hence a proper Climate trend analysis based on hydro-meteorological parameters in the basin for current and future scenarios is an essential and timely requirement, due to its highest basin population and as it is ranked as third (3 rd ) in the country in terms of water resources. Therefore, it is vital to evaluate the climate trends for the past period of 1980 to 2016 considering climate elasticity of runoff based on observed rainfall and streamflow data. Not only that, but simulated future streamflow also using SWAT Model in Kelani River Basin for the planning and management of the basin water resources efficiently and sustainably in future.

Hydrological Models and SWAT Model
Hydrological models are very valuable tools to recognize the response to the issues in water resources planning and management [4]. Hydrological phenomena are highly nonlinear, highly variable and extremely complex in space and time [5]. Soil and Water Assessment Tool (SWAT) is a semi-distributed, physically-based rainfall-runoff model. It has become a powerful tool which measures the effects of Climate change on water resources planning and management in the recent past [6]. Calibrated SWAT model was used to simulate the potential effects of future Climate change on streamflow. Subsequently, the model was used to estimate rainfall-runoff elasticity in the Kelani River basin. This analysis ultimately facilitates the screening of more efficient and sustainable future water resources planning and management alternatives.

Study Area
Kelani River is the second largest river after Mahaweli Ganga by volume of discharge in Sri Lanka [7]. It is the 7 th largest river basin in the country with a watershed area of 2,292 km 2 , annually contributing 4,225 MCM flow to the sea. It is bounded by Attanagalu Oya, Maha Oya, Mahaweli Ganga and Kalu Ganga basins. The Kelani basin is entirely located in the wet zone with the highest annual rainfall in Sri Lanka, with the annual average rainfall ranging between 2,000 mm to 5,700 mm. The mean temperature varies a little over the year, between 28 o C and 30 o C in the basin. It flows along 145 km into the sea at Modara and elevation varies from 2,500 -0 m AMSL from downstream to upstream.
The basin currently has a population of approximately 2.5 million. This amounts to more than 19% of the total Sri Lankan population in less than 4% of the total land extent. The population density is over 1,000 people per km 2 in the Kelani basin. The population of the Kelani basin will rise to 3.3 million by 2040, with an increase of about 31% from 2016 [8]. Water supply from the Kelani Ganga will experience deficits by the year 2025, even corresponding to 2 year return period daily average low flow value [9]. The basin also contains parts of the administrative districts of Kegalle (44%), Colombo (19.6%), Nuwara Eliya (18.4%), Gampaha (14.3%), Ratnapura (3%), Kalutara (0.5%) and Kandy (0.2%) as shown in Figure 1. It is apparent that nearly two-thirds of the total area of the Colombo district (64%) is situated within the Kelani Ganga basin.

Figure 1 -District Boundaries and Stream Network in Kelani Ganga Basin
Hence, it is vital to evaluate the Climate Elasticity of runoff based on observed rainfall and streamflow data and simulated future streamflow using the SWAT model in Kelani Ganga basin, as the above given factors clearly illustrate the importance of the assessment of water resources in Kelani Ganga basin.

Study Approach and Setting
In order to better understand the present water resources in the basin and Climate impacts on future water availability, the study focused on the existing basin water resources, their Climate, Climate trends and trend analysis.

Climate Change
A number of research have been carried out to identify Climate change impacts regionally as well as globally. Several General Circulation Models or Global Climatic Models (GCMs) and Regional Climatic Models (RCMs) have been developed to facilitate the analysis of climatic change. However, the situation related to Sri Lanka is quite different, since there are significant research gaps with respect to the Sri Lankan context and basin level estimates.
The future variations in Climate will subsequently have an impact on regional water resources and regional hydrologic conditions in terms of both quality and quantity [10,11]. Potential effects may comprise changes in hydrological processes, hence research of global change on the hydrologic cycle plays a rising role [12]. IPCC confirms that global warming will be increased by 1.5°C by 2050 [2].

Climate Trends in Sri Lanka
Climate change has been predicted to affect the pattern of rainfall, hence would change the timing of the receipt of reservoir inflows [13]. In addition, the shifting of climatic zones would be expected due to Climate change. There will be a significant expansion of dry areas of the country by 2050 due to Climate change, thereby imposing significant pressure on water resources [4]. The regional and basin level rainfall trends, temperature trends, evaporation and evapotranspiration trends were identified during the literature review.
The World Meteorological Organization (WMO) guidelines on the standard climatological normal calculation endorse a rolling 30-year period, upgraded every 10 years for operational Climate monitoring [12]. The latest time range for the base year period is 1981 -2010 [12].

Analysis of Climate Change Impacts
Innovative Trend Analysis (ITA), Mann-Kendall (M-K) test and Sen's slope tests were carried out to check the trends in hydrometeorological parameters such as rainfall, temperature and flow. These methods were used to analyse trends in annual and seasonal variations in each parameter. The four seasons were defined as First Inter Monsoon (FIM: March -April), South West Monsoon (SWM: May -September), Second Inter Monsoon (SIM: October -November) and North East Monsoon (NEM: December -February). These tests were used to analyse Maha (October to April) and Yala (May to September) seasons as well. Mann-Kendall and Sen's slope methods were used to verify the results of Innovative Trend Analysis (ITA).

Statistical Tests for Climate Impacts
Three (3)

Mann-Kendall Test
The Mann-Kendall test, proposed by Kendall [16,17], is a non-parametric test, which is a widely used most popular method to detect trends in hydro-meteorological time series [14,15]. The test statistic S is given by: ...(1) where n is the number of observations, x i and x j are the i th and j th (j >i) observations in the time series, and: where m is the number of tied groups and t k is the number of ties of extent k. The standard normal test statistic Z used for detecting a significant trend is expressed as, A positive value of Z indicates an upward trend, while a negative value of Z indicates a downward trend.

Sen's Slope Estimator
Sen [16] developed a method to estimate the slope of the trend using a non-parametric procedure in the sample of N pairs of data. The N values of Qi are ranked from smallest to largest and the median of slope or Sen's slope estimator is computed as:  The governing equation in the SWAT model is given below [26].

Materials and Methods
The research methods followed and the data used in the present study are explained below.

Methodology
The methodology adopted for this research is briefly described as follows.
 During the literature review, research gaps, the extent of analysis and prevailing issues were identified. The research objectives and specific objectives were originated.
 The study area was selected based on the research gaps and other identified issues during the literature survey.
 Data collection was initiated and data checking was carried out for all meteorological and hydrological data series and the missing data threshold is taken as less than 10% for all time series data.
 Gap-filling was carried out for all five (5) alpha parameters for rainfall and temperature data using Inverse Distance Weighting (IDW).
 Three (3) streamflow (hydrometric) stations were selected among six (6) hydrometric stations based on the data quality. Gap filling of streamflow was carried out using linear interpolation and nearby station's records.
 Root-mean-square error (RMSE) was calculated for each month and each percentile to determine the most suitable combination of power value (α) for both rainfall and temperature time series in the analysis.  The applicability of the same hydrological parameters, which were used for Glencourse, was also evaluated for Hanwella and Kitulgala gauging stations in the basin.
 Future rainfall and temperature series already derived for pessimistic Climate change scenarios and landuse for 2040 [8] were used to simulate the future flow series in the Kelani Ganga basin using the SWAT model for 2040.
 The runoff elasticity (ε) was assessed by two methods based on the assessment of impacts of Climate change only and impacts of Climate and land surface change on the streamflow, as evaluated by Sankarasubramanian et al. [22] and Zheng et al. [24], respectively for the current and the future pessimistic Climate change scenarios for 2040.

Data and Data Checking
Several tests were performed during data checking such as visual data checking, outlier checking, graphical checking, and consistency and homogeneity checking using: Accordingly, rainfall, evaporation, temperature, streamflow data and spatial information for all hydrometric stations were collected from different organizations in Sri Lanka. Table 3 summarises the details of data availability for the analysis. All the collected data were preprocessed to restructure the raw data into time series during the initial stage of data collection.
Data inconsistency, missing data, and outliers were assessed visually for the collected hydrometeorological data including streamflow, rainfall, maximum temperature (T max ) and minimum temperature (T min ) data.
Using the land use maps developed by the Land Use Policy Planning Department (LUPPD) and the growth and constraint factors developed as part of the population growth analysis, the existing land use maps were updated to represent the projected land use change in 2040 [8] for SWAT modelling.
Furthermore, the double mass curves and annual water balance were also used to identify the data inconsistency and/or homogeneity. Key components of the SWAT model are weather, surface runoff, return flow, percolation, evapotranspiration, transmission losses, pond and reservoir storage, crop growth and irrigation, groundwater flow, reach routing, nutrient and pesticide loading, and water transfer [19].  Even though there are many publications on rainfall trends in the basin, the selected data durations for the trend analysis are different. Therefore, the comparison of rainfall durations for the selected duration of 1980-2010 was based on [28] and Table 5 shows that all given trends clearly tally with the published trend analysis.

Climate Trend Analysis based on
Minimum Temperature (Tmin) Katunayaka station exhibits a significant increasing trend in M-K test for annual averages at a 10% confidence level, while it shows a significant increasing trend for Maha season at a 5% confidence level for the period of 1980 to 2016 ( Table 7). The Sen's slope trend magnitude varies between 0.0 to 0.3 for all annual, Maha and Yala averages for the period of 1980-2016. Katunayaka exhibits the highest upward trend for annual, Maha and Yala seasons.

Climate Trend Analysis based on
Streamflow data All selected three streamflow gauging stations exhibit significant downward trends for the period of 1980 to 2016, as per the results of M-K test. Glencourse hydrometric station exhibits a significant decreasing trend in M-K test for Maha averages at 5% confidence level, while it shows a significant decreasing trend for annual averages and for Yala season at 1% confidence level. Kitulgala station exhibits a significant decreasing trend in M-K test for annual averages at a 10% confidence level, while it shows a significant decreasing trend for Yala season at a 5% confidence level (Table 8)   Hargreaves method [29] was used to calculate evapotranspiration in SWAT. It is evident that there is a good agreement between the two methods between January and May in Colombo. After May, the actual evaporation data is higher than the calculated values, while the situation is reversed after mid-October. There is generally a good agreement between the two methods at Seetha Eliya with values generally higher than the measured evaporation data, especially after the month of August.

Selection of Model Parameters and
Objective Function Glencourse gauging station is selected as the key hydrometric station in the basin, as it is located in the narrow gorge section and it is the best gauging station in terms of data quality compared with the others in the Kelani Ganga basin. The durations for calibration and validation were selected based on the best quality observed data available periods for the Kelani Ganga basin.
By visualizing the daily simulated flow vs. observed flow, it is identified that peak flow and baseflow are the most sensitive parameters for optimization, hence rules for parameter regionalization were used to select the parameters [30]. During the sensitivity analysis, about seven (7) parameters were initially chosen for optimization [30] and finalized with four (4) parameters. Optimized four parameters are Threshold depth of water in the shallow aquifer required for return flow to occur (CN2.mgt), Threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN.gw), Available water capacity of the soil layer (SOL_AWC().sol) and Soil evaporation compensation factor (ESCO.bsn) for the selected, which indicated the highest model sensitivity for the three selected objective functions. The Nash-Sutcliffe Efficiency (NSE), Relative Error (Er) and Coefficient of determination (R 2 ) were used to assess the SWAT model performance as suggested in the literature [23,31]. Repeated sensitivity model runs were performed to optimize the three objective functions for Glencourse hydrometric station by changing four parameters using SUFI-2 parallel processing using SWAT-CUP. Due to the discrepancies in the data resolution, the actual resolution of the data series is not recorded, as the daily time step is used in the observed data series. Hence, the one-day time lag is adjusted in the observed time series to match the modelled flow series, to optimise objective functions.
Generally, model performance is termed very good if NSE >= 0.75, satisfactory if 0.36 =< NSE < 0.75, and unsatisfactory if NSE < 0.36 [31,32,33], while R 2 should be greater than 0.5 [33] and Er values are lower than 20% [32]. The objective functions of NSE, R 2 and Er obtained are 0.65, 0.72 and 8.9% for calibration and 0.69, 0.69 and 9.1%, for validation, respectively. Hence, the overall performance of the model in terms of R 2 , NSE and Er is quite satisfactory for Glencourse hydrometric station. Though the obtained R 2 and Er for Hanwella gauging station are satisfactory for calibration period with 0.6, 7.3%, the obtained NSE is very low such as 0.23 for the calibration period. None of the objective functions performed satisfactorily for Kitulgala station, presumably due to data inconsistencies.

Model Performance and Reliability 4.2.2.1 Flow Threshold Selection
The threshold for low flows and high flows were taken as 80% and 15%, respectively, for Glencourse gauging station, by visual observation of the deflection change of the Flow Duration Curve (FDC).

Model Performance
Model performance was checked by mainly NSE and R 2 for daily time-step. Low flow and high flow regions were identified for Glencourse. Though the satisfactory model performance is achieved for the objective functions for overall and high flow regions, the medium and low flow regions were unable to achieve the satisfactory model performance for NSE for calibration and validation periods. The NSE was very low as -0.63 and 0.00 for validation and calibration periods, respectively, for the medium flow region. Though the low flow region shows satisfactory model performance for objective functions for the validation period, it was unable to achieve satisfactory performance for the calibration period for R 2 , but the NSE was good for both calibration and validation of the model for the low flow region.

Runoff Elasticity 4.3.1 Current Scenario
Climate change is always linked to a lot of uncertainties, hence Climate change predictions are also very sensitive especially for a small island like Sri Lanka when downscaling of GCM results to small grid sizes.
It clearly shows that a 1⁰ C of temperature increase will cause a 6.9% to 7.5% runoff decrease, while a 1% of rainfall increase will cause 0.002% and 0.400% runoff increase at Glencourse gauging station as [25] and [23], respectively, hence temperature increase causes a higher impact on runoff than rainfall does for the current scenario for both methods (Table 9). A 1⁰ C increase in temperature causes a 0.4% rise and 1.5% reduction of runoff as evaluated by [24] and [22], respectively, for the Future Pessimistic Climate change Scenario with projected landuse for 2040. A 1% of rainfall increase causes a 0.005% and 0.360% increase of streamflow as evaluated by [24] and [22], respectively, for Glencourse gauging station for the aforementioned scenario. This implies a positive effect on runoff for the future pessimistic scenario with projected landuse for 2040 as of [24]. 7. Mass balance performance error (Er), coefficient of determination (R 2 ) and Nash-Sutcliffe efficiency (NSE) were used as multi-objective functions and 8.90%, 0.72, 0.65 and 9.10%, 0.69, 0.69 are obtained, respectively, for the calibration and validation periods for the key hydrometric station at Glencourse. 8. A 1⁰ C temperature increase in the basin causes a 0.4% rise and 1.5% reduction of runoff as evaluated by the methods proposed by Zheng et al. [24] and Sankarasubramanian et al. [22], respectively, for the Future Pessimistic Climate change Scenario with projected landuse for 2040. A 1% of rainfall increase causes a 0.005% and 0.360% increase of flow as evaluated by Zheng et al. [24] and Sankarasubramanian et al. [22], respectively, for Glencourse gauging station for the aforementioned scenario. This implies a positive effect on runoff for future pessimistic scenario with projected landuse for 2040 following the method proposed by Zheng et al. [24].

Recommendations
As the SWAT model has been calibrated and validated for the duration of 1970 to 1992, the results show a high degree of uncertainty of flow simulation for recent years. Hence, it is recommended to perform a water allocation model to obtain better calibration and validation results for the Kelani Ganga basin in future with consideration of identified trends after 1995, as water allocation models present the water users' contribution in the basin.Thus, the runoff will be reduced than that of the hydrological models at each node. Hence,the use of water allocation models could result in better calibration and validation results, which will reduce the degree of uncertainties on flow simulations for recent years after the 1990s as well as increase the degree of confidence to predict runoff elasticity coefficients for future scenarios in future researches for planning and management of water resources.