Modeling future changes in the North-Estonian hydropower production by using SWAT

Small hydropower (SHP) has been a source of electricity generation in Europe since the beginning of 20th century. SHP represents about 9% of renewable and 1.2% of the total electricity generation in the European Union (Kougias et al. 2014). In 2010, Estonia had 47 SHP plants with a total installed capacity of 8 MW, the aim is to have 55 plants with a total installed capacity of 9 MW by 2020 (Liu et al. 2013). A renewable energy support scheme is interested in increasing the number of small or micro hydropower plants as a feed-in tariff or a fixed premium is legally meant to be paid by the utility (Liu et al. 2013). However, there exists a contradictory standpoint by public organizations referring to the very low hydropower share of total electricity production (less than 0.5%) in Estonia (Punys & Pelican 2007) and to the environmental considerations, i.e. the migration of fish being blocked or adversely affected by the hydraulic structures. Thus, an evaluation of the impacts of climate change on river discharge will provide valuable information to policy makers.

Estonian topography is relatively flat and rivers have small average slopes. There are over 7,000 rivers and streams in Estonia, however discharge of less than 50 rivers exceeds 2 m³/s, and only 14 rivers discharge over 10 m³/s (Raesaar 2006). Thus, as the hydroenergetic potential is proportional to the head and the rate of discharge of the water, large hydropower plants with a high dam and large reservoir for flow regulation cannot be utilized in Estonia. Nevertheless, there are many rivers suitable for small and micro hydro power plants. For example, in northern Estonia where the steep escarpment of the Baltic Klint is cut through by river valleys.

Thus, due to the conditions described, hydro power plants in Estonia are mainly small, except the Narva Hydropower station which is owned by the Russian Federation and is operating in the border river. Small scale hydro power plants (SHP) are defined as ‘run-of-river’ plants which do not require a large impoundment of water, with little or no control of discharge, sometimes implementing diversion schemes to utilize the natural topographic gradient (Kumar et al. 2011). The energy comes mainly from the water volume and not from the head of water (Gaudard & Romeiro 2014).

SHPs are believed to be ‘clean’ and ‘green’ compared to large hydropower plants (Kumar et al. 2011), having conditionally low environmental impact (Gaudard & Romeiro 2014). However, lately this perception has been questioned (Premalatha et al. 2014). Estonian SHPs are facing many administrative barriers (ESHA 2012). To build and operate a hydropower plant, one needs a Permit for the special use of water (concession) which defines the permit owner's rights and obligations. The licensing procedure for the sector is time consuming. The authorization procedure takes from 4 months to 4 years. Concessions have a duration of only 5 years. Minimal residual flow is prescribed in the water use licensing procedure and is defined as a fraction of flow duration curve (95%). Mitigation measures, i.e. constructing fish passes, are often requested (ESHA 2012).

The ‘run-of-river’ hydropower plants are particularly dependent on river discharge. Thus, changes in pattern and amount of available water have a profound effect on hydropower generation. Possible changes in future water storage will differ from region to region around the globe. Li et al. (2015) used regional climate model (RCM) RegCM4 as a driving force to investigate the potential impact of climate change on hydrology over continental Southern Africa. By using the results from an ensemble of 16 or more CMIP3-CGCMs (coupled global climate models), Zhang et al. (2015) investigated how future changes in temperature and precipitation might influence total runoff in the headwaters of the Yellow River basin.

There have been several studies generalizing the impacts of climate change on hydropower by using the delta change approach (e.g. Lehner et al. 2005; Carless & Whitehead 2013; Gaudard & Romeiro 2014). Also the spatial resolution has been coarse (Lehner et al. 2005). The main objective of this study is to use daily generated climate data from the EURO-CORDEX (Coordinated Downscaling Experiment – European Domain) project and detailed spatial information to study future changes in the North-Estonian hydropower production.

Keila has the largest watershed among the three study areas (Table 1). According to the land use and land cover database Estonian National Topographic Database, most common land use types in the three study basins being arable land, forested land and wetland. The dominant land use for all study areas is arable land, covering around 45–60% of the area. Kunda basin is only covered 5% by wetland, while the corresponding land use coverage in Keila and Valgejõe are 12% and 16%, respectively. Forested land accounts for 28%, 40% and 44% of the land surface in Keila, Kunda and Valgejõe, respectively. All case-study regions have relatively flat slopes, with an average slope of around 1%.

The average annual measured precipitation in Keila, Kunda and Valgejõe is 725 mm, 644 mm and 701 mm, respectively. Around 20% of precipitation is snowfall. Long-term average actual evapotranspiration rate is around 400 mm per annum. The average annual temperature during the baseline period 1971–2000 was 4.5°C. The coldest month is February, with an average temperature of –6.6°C, followed by January with an average temperature of –6.2°C. The warmest month is July, with an average temperature of 15.8°C.

Kunda river is the most challenging for simulation, since its flows are significantly affected by karstic aquifer. It is estimated that groundwater contributes over 50% of flow to the river discharge (HELCOM 2011). In Estonia, the typical river hydrograph has peak discharges in spring and fall, from which the spring is more abundant in water due to snow melt. Observed daily discharge data for the period 1970–2010 are available in all three study basins, which are used for model calibration and validation.

The first fully automatic hydropower plant was built in 1893 on Kunda River, where a Francis turbine with capacity of 210 kW was installed. Today Kaplan reaction turbines have been installed in all three plants. The SHP characteristics of the three basins are provided in Table 1.

Climate impact assessment requires scenarios of future climate to be translated into potential changes in the quantity and timing of river runoff. Modeling results from the EURO-CORDEX project (Jacob et al. 2013) are used as future climate data input. EURO-CORDEX is part of CORDEX initiative (Giorgi et al. 2009), which coordinates regional downscaling of CMIP5 project to all terrestrial regions of Earth. Different institutions run their climate models on a similar grid within the EURO-CORDEX project, and make data available on the same grid. Therefore, it is possible to conduct a common model evaluation (Kotlarski et al. 2014) and provide common climate projections for climate change impact, adaptation and mitigation studies (Giorgi et al. 2009). Within the EURO-CORDEX project, models are being run with two resolutions: 50 and 12 km, and both cover all European land areas, the Black Sea, the Mediterranean Sea and Iceland. For the current project, two modeling studies are used from EURO-CORDEX high-resolution simulations: RACMO model with the boundaries from EC-EARTH r1i1p1 ensemble member from the Netherlands (KNMI) (Van Meijgaard et al. 2008) and HIRHAM5 model with the boundaries from EC-EARTH r3i1p1 ensemble member from Denmark (DMI) (Christensen et al. 2007).

Climate models in general are mathematical representations based on physical principles which estimate higher for some climate variables (e.g. temperature) than for others (e.g. precipitation) (IPCC 2001). RCMs are known to be biased, causing even more uncertainties in the future hydropower potential change prediction. Despite biases, RCM's still produce variables which are physically coherent. Muerth et al. (2013) found that bias correction of regional climate simulations provide a closer to reality representation of the climate in the use of hydrological models.

For the whole study region, both initial climate projections have cold biases year round. A larger bias is found during the spring and summer period, where it is around −2 °C according to RACMO-DMI and −3 °C according to HIRLAM5-KNMI. Both projections have cold biases of less than –1 °C for the rest of the period.

The precipitation amounts in RACMO-DMI are overestimated in winter and spring, while HIRLAM5-KNMI overestimates in winter and underestimates during summer months. In terms of average annual precipitation, there is a clear overestimation of around 100 mm by RACMO-DMI model. There is a tendency of both climate model projections to simulate too many low-intensity rain events.

KNMI tends to predict lower changes in temperature, compared to DMI. The DMI model predicts an overall 30% increase in precipitation, which means potentially higher flows throughout the year. The KNMI model predicts an overall 10% increase in precipitation, except for the summer months, where a slight reduction in monthly precipitation is expected. This suggests a reduction in summer runoff.

The Soil and Water Assessment Tool (SWAT) (Arnold et al. 1998; Neitsch et al. 2005) was applied to simulate hydrologic processes in the three study basins. SWAT is a physically-based, semi-distributed hydrological model, which uses process-based equations to simulate different hydrologic responses. Although the model time step is daily, SWAT was designed as a long-term yield model and is not designed to simulate single-event flood accurately.

One of the reasons for selecting the SWAT model was that it has been widely and successfully used in snowmelt regions to simulate the hydrologic response (e.g. Abbaspour et al. 2007; Ahl et al. 2008). Furthermore it has also been used around the world for the estimation of climate impacts (e.g. Marshall & Randhir 2008; Ficklin et al. 2009; Franczky & Change 2009). The SWAT model has not been widely used as a tool for evaluating hydropower potential change (Haguma et al. 2014; Song et al. 2014).

SWAT requires a significant amount of data and parameters for development and calibration. These include a digital elevation map (DEM), land use map, soil map and weather data. Climate inputs consist of precipitation, solar radiation, maximum and minimum temperature, wind speed and relative humidity. In a SWAT model, a watershed is subdivided into a number of sub-basins, which are then further subdivided into hydrologic response units that consist of homogeneous land use, slope and soil characteristics.

In this study the daily maximum and minimum temperature, wind speed, humidity and solar radiation data are available for all study basins. Daily precipitation data were used from six meteorological stations. Daily discharge data are available for Kunda, Keila and Valgejõe river hydrological stations. A high-resolution DEM with a 10 m grid size derived from light detection and ranging was used. Due to the absence of realistic land use change scenarios in study basins, the same land use map was implemented in current study. This uncertainty has to be considered, while analyzing the results.

The SWAT model was calibrated and validated with the SWAT CUP software. For this the SUFI2 (Sequential Uncertainty Fitting) algorithm was used. In SUFI2, parameter uncertainty takes into account the various uncertainty sources like rainfall, measured data, etc.

The NSE (Equation (2)) is a normalized statistic that determines the relative magnitude of the residual variance (‘noise’) compared to the measured data variance (‘information’) (Nash & Sutcliffe 1970). As stated by Sevat & Dezetter (1991) NSE is the best objective function for reflecting the overall fit of a hydrograph. NSE is very sensitive to peak flows, at the expense of better performance during low flow conditions. Since the objective is to model hydropower potential change, where low flow does not play a key role, NSE and R² are appropriate to be used as qualitative model evaluation criteria.

PBIAS (Equation (3)) measures the average tendency of the simulated data to be smaller or larger than their observed data counterparts (Gupta et al. 1999). Simulation results are ‘unsatisfactory’ when PBIAS is larger than 25%. Values below 10% are considered to be ‘very good’. Values ranging from 10% to 25% are rated from ‘good’ to ‘satisfactory’ during calibration and validation (Van Liew et al. 2007).

In Estonia most of the hydropower plants are classified as low head, with a head below 10 m. The energy comes mainly from the water volume and not from the head of water. In all study basins, Kaplan reaction turbines are installed. These are most suitable for low-head sites because of their higher specific speeds and flat efficiency curve where part-load performance is an important factor.

According to Equation (4) the changing flow rate and the hydraulic efficiency are functions from flow rate. These are the main parameters affecting the energy outcome from run-of-river hydropower plants. Thus a change in river flow rate, means a change in hydropower. In hydropower calculations, some limitations must be taken into account.

Hydropower plants in Estonia must guarantee a minimal residual flow (e.g. through the spillway), i.e. water cannot be extracted while river runoff is lower than minimal residual flow. The upper threshold for energy generation is limited by a maximum flowrate of the turbine. Thus the available water for energy generation is the range between the maximum flow rate of the turbine and available water for consumption. For efficiency reasons of the Kaplan turbines, only higher than 30% of the first turbine's maximum flow rate is extracted for energy generation. Technical parameters of hydropower plants and limitations by water permits have been implemented in the present study.

Validation results show that the SWAT model can effectively represent the hydrological processes in the study basins. PBIAS reveals almost no bias for Keila river basin. PBIAS values for Kunda and Valgejõe indicate a slight model bias towards overestimation, however model performance is still considered ‘very good’ (Van Liew et al. 2007). Thus, the water balance for study basins is physically representative by the SWAT model.

Tolson & Shoemaker (2004) stated that SWAT is not designed to simulate a single extreme event and the model usually underestimates largest flow events. A similar tendency is observed in this study. However, this underestimation does not significantly affect hydropower harvesting in the current study, because peak flows exceed maximum flowrate of the turbines.

The simulation results of the SWAT hydrological model indicate a change in river runoff for both climate scenarios. The mean increase in annual discharge of 15% and 55% was predicted for North Estonia by climate projections KNMI and DMI, respectively (Table 3). The spring peak in the study basins tends to occur earlier and be smaller compared to the baseline period. This pattern was more pronounced in the DMI climate projection, resulting from an increase in precipitation (Figure 7) and lesser winter snow accumulation. A significant increase in autumn discharge was projected by the same model.

Coherence between changes in annual discharge and hydropower potential was evident with some exception. Flowrates exceeding installed hydro plant capacity cannot be used for hydropower harvesting. Keila River SHP is a good example of this behavior: hydropower potential is predicted to rise 17% and 34%, according to the KNMI and DMI climate projections, respectively. The projected changes in the hydropower potential were higher in Kunda and Valgejõe. Generally, significant increase in winter hydropower potential is modeled in SHP, where installed capacity is available (Table 3). Kunda SHP revealed a lower increase in hydropower potential for summer months, compared to Keila and Valgejõe SHP.

The amount of uncertainty involved in modeling of Kunda is higher because the physical groundwater system of the karst aquifer may behave unexpectedly.

Climate-induced changes in temperature and precipitation are the driving factors in modeling river runoff. According to the climate projections used, the impact on future river discharge in North Estonia can be remarkable. In the Baltic states, Bolle et al. (2008) found that annual river runoff is forecasted to increase in North Latvia, which agrees with the results of the current study. However another study by Apsite et al. (2010) predicts a decrease in annual river runoff. Contradicting results in Latvian river runoff can be explained by the use of different climate models, scenarios and assumptions. Kriaučiūnienė et al. (2008) predicted a remarkable increase in Lithuanian winter river runoff due to the shortening of the snow period, while spring runoff peak decreases and shifts to an earlier period by the end of the century.

A comprehensive study of the climate change impact on the hydropower potential in Europe by Lehner et al. (2005) suggests an increase in the hydropower potential in North Europe by the end of 21st century. In Finland and Sweden, the increased runoff will give potentially 19% higher hydropower in both countries. Our results are in agreement with the findings of Lehner et al. (2005) who found an increase of 29% in Estonia whereas slightly higher hydropower potential was found in the current study with a mean value of 35% for North-Estonian study basins (Table 3).

High flow periods could be better exploited by increasing the number of turbines at SHP, i.e. increasing the installed capacity.

Future simulations do not take into account changes caused by climate change itself, e.g. land use and vegetation. El-Khoury et al. (2015) and Song et al. (2014) studies showed that land use change can influence future river discharge considerably. Combining the impacts of climate change, land use, technology and policy will help to improve prediction quality. In spite of the uncertainties in hydropower potential modeling, the results provide a sound basis to energy policy makers for river management in North Estonia.

In this study the impact of climate change on water resources in North Estonia was assessed. Keila, Kunda and Valgejõe study basins were selected, and basin scale hydrology was modeled with the hydrological model SWAT. Two different climate projections (RACMO-KNMI and HIRHAM5-DMI) were bias corrected and used as input for calibrated hydrological models. Change in hydropower potential for run-of-river hydropower plants was calculated from modeled river discharges.

Modeling results indicate an annual increase in water discharge and thus hydropower potential increases in all study basins due to the changes in climate. High flow periods could be better exploited by increasing the number of turbines at SHP. It can be concluded that the climate change impact on hydropower potential in North Estonia is likely to be positive.

The results provide a sound basis for energy policy makers towards river management in North Estonia. For future work, it is recommended to implement different land use change scenarios, taking into account developments in technology and policy.

Current research was funded by the Estonian Research Council within the project ESTKLIIMA and partly by the project ‘The impact of land-related decisions on the use of land’ funded by the Estonian University of Life Sciences. Authors would like to thank Estonian Weather Service for their weather data contribution. Many thanks to Ain Kull, Arno Kanal and Jaak Jaagus from Tartu University for input data and expertise regarding application of the SWAT model.