Quantifying the potential impacts of land-use and climate change on hydropower reliability of Muzizi hydropower plant, Uganda

Ugandan rivers are being tapped as a resource for the generation of hydropower in addition to other uses. Studies on the reliability of these hydropower plants due to climate and land-use/land cover changes on the hydrology of these rivers are scanty. Therefore, this study aimed to model the impact of the changing climate and land-use/cover on hydropower reliability to aid proper planning and management. The hydropower reliability of Muzizi River catchment was determined from its past (1998–2010) and midcentury (2041–2060) discharge at 30 and 95% exceedance probability under Representative Concentration Pathways (RCPs) of 4.5 and 8.5, respectively. The past and projected hydropower were compared to determine how future climate and land-use changes will impact the discharge and hydropower reliability of Muzizi River catchment. Six LULC scenarios (deforestation, 31–20%; grassland, 19–3%; cropland, 50–77%; water bodies, 0.02–0.01%; settlement, 0.23–0.37%, and Barren land 0.055–0.046% between 2014 and 2060) and three downscaled Regional Climate Model (REMO and RCA4 for precipitation and RACMO22T for temperature from a pool of four CORDEX (Coordinated Regional Climate Downscaling Experiment) Africa RCMs) were examined. A calibrated SWAT simulation model was applied for the midcentury (2041–2060) period, and a potential change in hydropower energy in reference to mean daily flow (designflow 30% exceedance probability), firm flow (flow 95% exceedance probability), and mean annual flow was evaluated under the condition of altered runoff under RCP4.5 and RCP8.5 climate change scenarios for an average of REMO and RCA4 RCM. The future land use (2060) was projected using the MOLUSCE (Module for Land Use Change Evaluation) plugin in QGIS using CA-ANN. Three scenarios have been described in this study, including LULC change, climate change, and combined (climate and LULC change). The results suggest that there will be a significant increase in annual hydropower generation capacity (from 386.27 and 488.1 GWh to 867.82 and 862.53 GWh under RCP4.5 and RCP8.5, respectively) for the combined future effect of climate and land-use/cover changes. Energy utilities need to put in place mechanisms to effectively manage, operate, and maintain the hydropower plant amidst climate and land-use change impacts, to ensure reliability at all times.


INTRODUCTION
Hydropower is a key renewable energy source widely used as a driving force to power economic development and technological, scientific, anthropogenic, and industrial transformation in many countries across the world (Hwang & Yoo ). Currently, hydropower accounts for 86% of renewable energy technology that represents 16% (3,551 TWh/a) of global electricity generation which is projected to increase by 1% by 2050 (Hamududu & Killingtveit ). Global installed and electricity generated from hydro- where there are fewer negative environmental impacts as compared to the fuels, or economical where the hydropower plant is economically viable, cost-effective, competitive, and sustainable without government subsidization. The physical reliability metrics concerning land-use and climate changes include the discharge potential, efficiency, and consistency to meet the demand. The environmental metrics include the hydropower plant's ability to not only work without fuels that emit dangerous gases but also reduce the sediment accumulation, and economic reliability metrics include the hydropower plants' self-sustainability, leading to its viability and compatibility irrespective of the effects of land-use and climate changes.
This study has determined the land-use/land cover trends in the Muzizi River Catchment for the past 30 years   with a mean annual average of 700 mm at the mouth and 1300 mm at the source of the river. The temperature in the catchment ranges from 15.8 to 33 C within the year.
Soil in the Muzizi River catchment is mapped according to three major districts that it borders. Kibaale district has granitic soils which are classified as shallow loams with moderated acidity, red clay loams, and brown gravelly clay loams.

Data sources
The SWAT model uses readily available input data such as digital elevation model (DEM), land-use data, soil data, and climatic data, and the functions are summarized in Table 1.

Digital elevation model
The DEM allows ease of identification and measurement of the surface drainage area of catchment, which is among the first steps in conducting catchment delineation. The DEM  Figure 1 shows the locations of the climatological stations overlaid.
Reanalysis precipitation data ERA5, CFSR, CHIRPS, MERRA2, TRMM 3B42, and NASA Agro climatology, which are available to the public, were chosen, corrected for bias using the method proposed by Berhanu et al. (), and evaluated for suitability. Table 2 shows the six reanalysis precipitation data used in the evaluation of the observed data. River discharge for years 1957-1977 and years 1998-2010 are considered reliable (with no missing gaps) in assessing the catchment current surface water resources and hydropower reliability. Flow Gauge Station 85200 just 20 km flow shall be considered in assessing the hydropower reliability of Muzizi River.

Methodology
The main objective of this paper is to quantify the potential impacts of land-use and climate change on hydropower reliability. Details of the methodology are highlighted in the below sections.

Performance evaluation of reanalysis data
To fill in gaps for missing data, reanalysis of precipitation data which was widely applied by scholars in hydrological modeling was used. The reanalysis data in the section 'Reanalysis precipitation data' were first assessed to identify the most accurate reanalysis dataset that can better mimic observed data within the catchment. The performances of the reanalysis data were evaluated by comparing the gridded reanalysis data with the observed point/gauged climatic data. cing between x and y coordinates between the grid points of MERRA2 reanalysis data, resampling will introduce errors.
Therefore, rain gauge data were directly compared against the nearest grid points of MERRA2 in the original resolution without resampling.
The quantitative assessment of the performance of the six reanalysis climate data in simulating observed mean monthly precipitation was undertaken using the following: A t-test was carried out at a 5% (α) level of significance to estimate the P-value to assess the reliability of the null hypothesis (H 0 ) which was formulated as follows: observed and reanalysis precipitation are not significantly different.
Reanalysis precipitation performance was judged by the magnitude of the statistical results of PBIAS, RMSE, Rs, R 2 , and N and the ability of reanalysis precipitation in reproducing mean monthly, mean daily, and sum of monthly observed precipitation.

Bias correction of selected reanalysis precipitation
The best performing/selected reanalysis is a product containing historical  precipitation data, typically containing biases when compared with observations (Mehrotra & Sharma ). Bias correction was carried out to correct the historical precipitation using the differences in the mean and variability between reanalysis and observed datasets. In this study, the biases in the daily time series of the precipitation from the selected reanalysis output were corrected using the easiest and the most common method, which was the multiplicative method for Here, the multiplicative correction factor for each month was used, and the modified daily rainfall was expressed as in the equation below: where P is the precipitation (mm/day), P is the long-term average precipitation and i, j, k are the day, month, and year counters, respectively.
Projection of future land use and climate of Muzizi River catchment Accuracy assessment of the remotely sensed/classified land-use/cover data This was carried out to compare the classified image to/with another data source (in this case Google image) that is considered to be accurate or ground truth data.
In this study, the accuracy of classified images for 1984, 2000, and 2014 was performed by creating a set of random points (also known as the ground truth point) and compared them with the classified data in a confusion matrix. The random points were termed as users' points which represent classified image pixels, while producers' points were the equivalent of users' point land use in google images. Users' and producers' accuracy were calculated as shown in Equations (2) and (3), respectively. The overall accuracy was obtained as the sum of the correctly classified pixel divided by the total number of samples expressed as a percentage in Equation (4). While the statistical test of the classification accuracy for individual pixels was determined using the Kappa statistic (Equation (5)).
where TS is the total sample and TCS is the total corrected classified sample (diagonal).
The statistics have values ranging from 0 to 1 although negative values are possible but rare. K values closest to 1 indicate almost perfect agreement (Othow et al. ).

SWAT model and model setup
The SWAT is a physically-based, semi-distributed hydrological model that predicts the impact of land management practices on water, sediment, and agricultural chemical yields in large complex watersheds of varying soils, land-use/cover, and management conditions over long periods (Neitsch et al. ). The model simulates the hydrological cycle based on the water balance (Equation (6)).
where SW t is the final soil water content and SW o is the initial soil water content of the day i, t is time in days, and R, Q, ET, P, and QR are the daily amounts of precipitation, surface runoff, evapotranspiration, percolation, and return flow, respectively, all measured in mm. The model uses readily available input data such as DEM, land-use data, soil data, and climatic data as described in the section 'Data sources'. In this study, modeling of the hydrological process was carried out using the extension of SWAT for Three statistical measures were employed. They are the coefficient of determination, R 2 (Equation (7)), the NSE (Equation (8)), and the PBIAS (Equation (9)). Other details of these measures such as their utility and satisfactory range of values are explained by Moriasi et al. ().
where n is the number of observations in the period under consideration, O i is the ith observed flow, O is the mean observed value, P i is the ith simulated flow, and P is the mean of simulated flow.
Evaluation of the performance of GCMs in simulating current climate conditions of Muzizi River catchment

Brief introduction
Information obtained from global climate models (GCMs) supports a better understanding of the climate at a global scale. The output from GCMs is too coarse (>100 km) to be used in impact assessment studies, adaptation planning, and decision-making processes at a local or regional scale (Treesa et al. ). In addition to the coarse resolution, biases and uncertainties associated with GCMs increase from global to regional and local scales, which limit the suit-    (10)) was performed for each pair of datasets after the estimation of their Spearman rank correlation coefficient (Equation (9)). The null hypothesis was rejected when the p-value obtained was greater than α-critical (0.05). That is, H 0 was rejected when p-value was >0.05; otherwise, it was not rejected. Model performance was judged by the magnitude of the statistical results of PBIAS, RMSE, Rs, R 2 , and NSE.
where IRsI is the absolute value of Spearman rank correlation coefficient Rs and n is the number of data (samples). P-value was estimated using the TDIST command in Microsoft excel after the calculation of the degree of freedom (DF ¼ n À 2).

Climate change data bias correction/downscaling
Climate model data for hydrologic modeling (CMhyd) was used to extract and bias correct the best-selected RCM model outputs and provide climate data for the SWAT model (Rathjens et al. ). Precipitation and temperature data were bias corrected using the Delta adjustment correction techniques method available in the CMhyd software.
Best-selected reanalysis precipitation data and CFRS temperature of the catchment were used for the bias correction of the RCM climate data.

Assessment of Muzizi current water resources and hydropower reliability
Flow duration curve The current/reference flow (water availability) of the Muzizi River catchment was determined from the dependable discharges that correspond to 95% exceedance probability (Equation (6) FDC is estimated by sorting the daily mean flows for the period of record from the largest value to the smallest value and assigning flow value a rank from 1 to the largest value.
The frequencies of exceedance are then computed using the Weibull formula for computing plotting position.
where p is the probability that a given flow will be equaled or where P is power (MW), η is the average total turbine efficiency, ρ is water density (kg/m 3 ), Q is discharged (m 3 /s), g is gravity (9.81 m/s 2 ), and H is the hydraulic head (m).
In this study, the average total turbine efficiency was deter-

Performance of reanalysis precipitation
The results of the performance assessment of the six different reanalysis datasets with respect to simulating mean monthly, mean daily, and mean annual precipitation for the period of 1981-2010 as compared to their respective observed precipitation are shown in Tables 4-6, respectively.
The results show that overall CHIRPS data outperformed other datasets in simulating mean monthly, mean daily, and mean annual precipitation for the period of 1981-2010 as compared to their respective observed precipitation. CHIRPS reanalysis precipitation was adopted throughout the study in assessing the impact of climate and land-use change onto Muzizi HPP reliability.

Muzizi catchment land-use/cover classification
Land-use accuracy assessment

Accuracy assessed land use
The results show that forest land area coverage increases to 41.48% in 2000 from 29.15% in 1984 (Table 8)  land-use/cover of farmland and settlement from 1984-2014 was because of the need to produce more food and built houses for the ever-increasing population in the catchment.
The increase in the area coverage of forest land between 1984 and 2000 was attributed to the good environmental policies that had been set by the new government which had been ushered in, within that period, which was restricting Best evaluated reanalysis dataset selected to drive hydrological modelling. Best evaluated reanalysis dataset selected to drive hydrological modelling.     flow is 11.14 m³/s.

SWAT calibration and validation
The semi-distributed SWAT hydrologic model was calibrated and validated for Muzizi streamflow gauges. As shown in Figure 4(a) and 4(b), respectively, the model was able to simulate daily streamflows with the goodness-of-fit values of NSE 64.5%, PBIAS 4.5, and R 2 0.59 for the calibration period

Precipitation
The daily precipitation simulations of the climate models from the CORDEX-Africa RCM datasets were averaged over the basin area, and their performances were evaluated using statistical parameters. The mean monthly, mean daily, and mean annual precipitation for the historical period 1981-2005 were compared with the CHIRPS reanalysis dataset for the same period. Summary statistics used to assess GCMs' performances in simulating observed rainfall are shown in Appendix 4.

Maximum and minimum temperature
Comparison of daily average maximum/minimum temperature for the period of 1981-2005 was compared to that of CFSR reanalysis over the catchment for the same period.
The null hypothesis was not rejected for all models as evident by their respective p-values being less than 5%. The  RCA4 RCM performed best in reproducing mean monthly, daily, and annual precipitation.

Potential change in mean monthly and annual precipitation
Future/midcentury rainfall has been differentiated at the scale of the catchment when possible (depending on the resolution used for the different climate modeling). Table 13 shows the evolution of mean monthly rainfall for the differ-     Potential change in mean daily flow and mean power generation of Muzizi River  Potential change in mean monthly and annual energy generation all the months is projected to be higher than the reference power generation.
Mean annual hydropower output is presented in

Recommendations
The study aimed at contributing knowledge to the hydropower and engineering professionals on the risks of land use and climate change on the hydropower reliability in the development of hydropower plants on small, medium, and large rivers. While this has been demonstrated, it should be noted that the analysis is limited to the hydrological dimension and has not considered aspects such as sedimentation. Given that the predicted changes are due to changes in flows caused by land use and climate changes, the risk of sedimentation on hydropower plants such as this one cannot be ruled out. It is therefore recommended that authorities pursue an environmental protection agenda through reafforestation and enforcing buffer zones alongside the Muzizi River, and a policy that governs the operation of these actions on catchments is most befitting. This study did not take into consideration of sedimentation impacts on the dam and its components, and it is therefore inferred that future studies be carried out to establish this.
This study has been undertaken under limited data in terms of period and spatial distribution. While this study has demonstrated that it is possible to utilize bias-corrected reanalysis data and historical discharge data to build a climate model in a data-scarce scenario, to improve the accuracy of the results, there is a need to invest in hydrological and climate infrastructure for improved data collection. These investments can be recouped through savings from improved operation and maintenance of the hydropower plant system and reduced unplanned downtime due to hydrological catastrophes.
It has been noted that the future discharges will be more than the designed discharges creating operation and maintenance challenges. To mitigate this impact, the spillway should be re-optimized to accommodate future overflows.
It is further recommended that further studies are undertaken on how to utilize this increased flow and how to optimize the performance of the plant. It has been found that climate change and land use impact river systems differently for different areas, and therefore, it is recommended that such studies are carried out on different river catchments to understand their responsibilities in the upcoming climate and land-use changes.