ABSTRACT
This study integrated climate projections from five global climate models (GCMs) into the soil and water assessment tool to evaluate the potential impact of climate alterations on the Rietspruit River sub-basin under two representative concentration pathways (RCP4.5 and 8.5). The model's performance was evaluated based on the coefficient of determination (R2), percent bias (PBIAS), Nash–Sutcliffe efficiency (NSE), probability (P)-factor and correlation coefficient (R)-factor. Calibration results showed an R2 of 0.62, NSE of 0.60, PBIAS of 20, P-factor of 0.86 and R-factor of 0.91, while validation produced an R2 of 0.64, NSE of 0.61, PBIAS of 40, P-factor of 0.85 and R-factor of 1.22. Precipitation is predicted to increase under both RCPs. Maximum temperature is projected to increase under both RCPs, with a major increase in the winter months. Minimum temperatures are projected to decrease under RCP4.5 in the near (−0.99 °C) and mid (−0.23 °C) futures, while the far future is projected to experience an increase of 0.14 °C. Precipitation and temperature changes correspond to increases in streamflow by an average of 53% (RCP4.5) and 47% (RCP8.5). These results indicate a need for an integrated approach in catchment water resource management amid potential climate and land use variations.
HIGHLIGHTS
An ensemble of five best-performing GCMs (MiroC5, CanESM2, SHMI-ESM, CSIRO and NorESM2) was employed.
P-factor and R-factor were used to assess model uncertainty particularly due to anthropogenic land use changes.
During winter, maximum and minimum temperatures are projected to increase and decrease, respectively.
The projected increase in streamflow seems to be aggravated by continued land use changes.
INTRODUCTION
Agriculture, energy, transport and industrial sectors are the largest sources of carbon emissions, thus contributing to climate change (IPCC 2021; Abbas et al. 2022a, 2022b; Elahi et al. 2024). As a result of anthropogenic and natural influences, there is an ongoing change in the global climate that is affecting different areas of human life and nature including agricultural production, catchment hydrology, forestry and biodiversity (Grusson et al. 2021; Alam & Rukhsana 2023; Filho et al. 2023). One of the sectors most impacted by climate change is water resources (Sang et al. 2023). In fact, water is the primary medium through which the impacts of climate change on different sectors of the economy are felt (IPCC 2014a, 2014b); hence, water security is key to creating climate-resilient communities and achieving sustainable development goals (Caretta et al. 2022). The continued rise in global air temperature adds more water vapour into the atmosphere, causing variations in weather circulations (Petpongpan et al. 2021; Das et al. 2022a). This results in the recurrent occurrence of catastrophic hydrological events including floods and droughts. Overall, this means there would be a decline or increase in the availability of both surface water and groundwater resources depending on the region under study. Thus, given the immeasurable prominence of water to mankind, it is only sensible to understand how future water resources will be affected by global climate change. This is critical for the future water resource planning, development, use and protection of the environment.
The earth's climate system and the extent of global warming are largely influenced by a nexus of factors such as increasing population, energy use, technological advancements, changes in economic activities and land use characteristics (IPCC 2014a, 2014b). As the greenhouse gas concentration increases, the risks of climate change impacts on water resources considerably escalate. For instance, under the low and high climate change scenarios, global mean surface temperature is projected to rise by 1 and >4 °C, respectively (Nazarenko et al. 2015; Li & Fang 2021). This consequently has the potential to increase evapotranspiration rates and change the precipitation patterns. Due to the unpredictability of the future climate, scenarios describing the future climate change circumstances were developed based on various scientific assumptions. In view of shifting information needs for policymakers and to generate future climate scenarios, the Intergovernmental Panel on Climate Change (IPCC) fifth assessment report (IPCC 2014a, 2014b) published a standardized set of scenarios called representative concentration pathways (RCPs) that are used for climate research and as input for running climate models. RCPs are a collection of scenarios that encompass the impact of emissions, concentrations, land use/land cover and various climate policies (van Vuuren et al. 2011).
These RCPs demonstrate the potential different socio-economic and technological advancement scenarios in the future. Climate scenarios are noticeable tools that are useful to decision-makers with respect to characterizing the future climate. RCP scenarios include a high greenhouse emission scenario (RCP 8.5) due to no policy changes to reduce emissions, two medium-range emission scenarios (RCP 4.5 and 6.0) representing low and intermediate energy intensities and a low greenhouse gas emissions scenario mostly characterized by the declining use of oil and low energy intensity (RCP 2.6) (IPCC 2014a, 2014b).
The hydrological modelling approach is recognized as one of the effective methods for assessing the impact of climate change on streamflow characteristics (Bastola et al. 2011; Shrestha et al. 2018; Bhatta et al. 2019; Ji et al. 2023). To project future hydrologic conditions and the effects of potential climate change on water resources, future climate projections from global climate models (GCMs) are integrated into hydrologic models (Teutschbein & Seibert 2012; Ye & Grimm 2013). The use of GCMs to depict future climate information has been widely acknowledged and used in various climate-related applications (Miao et al. 2016; Yoo & Cho 2018; Rhymee et al. 2022). GCMs describe the general circulation of the land surface, ocean and atmospheric processes; hence, they are able to provide climatic information on past, present and future periods (Flato et al. 2013; Ackerley & Dommenget 2016). However, due to coarse spatial and temporal resolution in the GCMs, the model outputs are downscaled to enhance and match the resolution at a local scale and minimize the systematic biases (Soriano et al. 2019). Further, each GCM is developed using its own set of underlying assumptions and distinctive mathematical representations of the climatic processes (Yoo & Cho 2018), thereby resulting in different climate projections from each GCM.
The multi-model ensemble approach, where projections from several individual climate models are combined, is widely used in different hydrologic research with an aim to reduce the uncertainties in climate models (Yimer et al. 2022; Ahsan et al. 2023; Mounir et al. 2023; Yalcin 2023).The GCM ensembles have been used to evaluate future streamflow in numerous climate change impact studies across the world. For example, Ndhlovu & Woyessa (2021) evaluated the climate change impact on the Kabombo sub-basin of the Zambezi River Basin in southern Africa using six GCMs, where it was found that annual streamflow will increase under RCP8.5 and slightly decrease under RCP4.5. Likewise, Faramarzi et al. (2013) integrated future projections from five GCMs to project the impact of climate change on water resources across Africa. The research findings showed that there is a likelihood that the total water quantity in Africa as a whole will increase in the future. Using three GCMs (MIROC5, CNRM-CM5 and MPI-ESM-MR), Petpongpan et al. (2021) projected that the Yom and Nan River basins (Thailand) will experience an increase in annual rainfall, temperature and runoff particularly in the middle of the 21st century. To assess the potential impact of future climate change on the hydrology of the Yarra River catchment, Australia, Das et al. (2022a) also used five GCMs (ACCESS1-0, CanESM2, CNRM-CM5, GFDL-ESM2M and MIROC5) under the scenarios of RCP 4.5 and RCP 8.5. The study projected a decrease and increase in precipitation and temperature, respectively, and a corresponding increase and decrease in annual evapotranspiration and surface runoff. In Pakistan, Abbas et al. (2022a, 2022b) evaluated the performance of GCMs from the Coupled Model Intercomparison Project Phase 6, where it was observed the HadGEM3-GC31-MM model performed better in simulating precipitation distribution. Precipitation was projected to increase at an annual scale with mixed increasing and decreasing trends at seasonal timescales.
In South Africa, researchers have presented findings on the potential hydrological impact of climate change using different hydrological models. Some of the commonly used hydrological models in South Africa include the Pitman hydrological model (Mvandaba et al. 2018; Du Plessis & Kalima 2021), the water evaluation and planning (WEAP) model (Remilekun et al. 2020; Dlamini et al. 2023) and the SWAT model. Notwithstanding some challenges including the need for accurate and sufficient data because of the numerous processes it models (Akoko et al. 2021; Janjić & Tadić 2023), the SWAT hydrologic model continues to be widely used for assessing the hydrological impact of climate change, and its performance has proved to be reasonable in different watersheds across the worlds (Bhatta et al. 2019; Kibii et al. 2021; Saade et al. 2021). The model's popularity largely lies on the fact that it is well documented, open access, computationally efficient and its capability to be used in different applications including water availability, water quality and different hydrological processes (Gassman et al. 2007; Tan et al. 2021; Sánchez-Gómez et al. 2022).
Likewise, in South Africa, the SWAT model continues to be applied as a tool for assessing different hydrological processes and river basin management options. Gyamfi et al. (2016) employed SWAT to assess the impact of land use changes on the hydrology of the Olifants Basin, where it was revealed that, between 2000 and 2013, the generation of surface runoff increased largely due to urbanization. Nkhonjera et al. (2021) used the SWAT model in the same Olifants River Basin to assess seasonal precipitation variability in the near- and long-term time series. The study projected a significant increase in future precipitation particularly within the 2020s–2050s. Using the SWAT model, Mararakanye et al. (2022) assessed the impact of land use changes on water quality of the lower Vaal River Catchment, whereby it was shown that nutrient loads (nitrate and orthophosphate) increased in irrigated and built-up areas, while dryland areas showed a decrease in the studied nutrients. In an attempt to model the potential impact of climate change on the sediment yield of the Tsitsa River catchment, Eastern South Africa, Theron et al. (2021) employed the SWAT model and six GCMs. The study found that sediment yield is likely to increase by about 10% on an annual basis, thereby increasing the risk of soil erosion and infrastructure longevity (Theron et al. 2021). Other recent studies that have used SWAT in South Africa include those on water resource availability (Thavhana et al. 2018; Woyessa & Welderufael 2021), on sediment yield (Scott-Shaw et al. 2020) and on water quality (Ngubane et al. 2022). These studies have shown that the SWAT model can be successfully applied in different regions and water management areas of South Africa, thus providing an additional rationale to be adopted in the Rietspruit sub-basin.
In general, the SWAT model was developed to predict how different land management practices within a watershed influence streamflow, water quality and sediment yield, with varying soils, land use and management conditions (Gassman et al. 2007). Physical properties of soil play a significant role in directing the flow of water within the hydrological system by dividing precipitation into various elements of the water cycle, including infiltration and overland flow (van Tol et al. 2020). For instance, soils with slow infiltration rates like clay, when thoroughly wet, restrict the downward movement of water, have a slow rate of water transmission and usually exhibit high runoff potential (Neitsch et al. 2005). However, land use activities have increasingly been impacting the natural soil characteristics, particularly in urbanized catchments and eventually influencing the different hydrological processes. The basin modifications usually affect magnitude and distribution of flow; hence, it is vital to understand such changes for achieving a better hydrological model.
Due to increasing land use activities including residential, agriculture and mining practices, the majority of past research works in the Rietspruit river catchment have been focused on the assessment of river water quality and distribution of pollutant loads (for instance, Steynberg et al. 1995; Showalter et al. 2000; Ijoma 2010; Dzwairo & Otieno 2014; Nyamukamba et al. 2019; Raji et al. 2022), while specific basin scale climate change related studies are limited. In contrast, Banda et al. (2021) performed the trend analysis of precipitation and temperature in the area, which was based on historical data. However, to devise proper climate adaptive measures in different sectors, it is necessary to predict and quantify the hydrological impacts and the magnitude of future climate change using different climate change scenarios. In all angles, effective climate adaptation planning should preferably be done at a local level or, to a certain degree, at the regional level (Grusson et al. 2021).
Hence, this study aims to evaluate how potential future climate change will affect the streamflow characteristics in the Rietspruit sub-basin. The Rietspruit sub-basin is one of the important river catchments supplying water to the Vaal River Basin, which is an economically important region despite it being affected by high developments (du Plessis 2021). The sustainability of the Vaal River is highly dependent on sustainable management of its sub-basins, one of which is the Rietspruit sub-basin. The Rietspruit River drains mine and domestic waste effluents into its lower elevations, which are directly downstream along the Vaal River (Raji et al. 2022; Sindane & Modley 2022). Knowledge of the water quantity changes due to potential climate change is not only significant for understanding catchment water balance but also water quality issues, for instance, release of contaminants due to extreme floods (Caretta et al. 2022). Therefore, findings from this study are useful to water resource managers, environmentalists and ecologists for developing a comprehensive and integrated water resource plan in the upper Vaal region under different climate change scenarios. Further, these results will also contribute to the growing body of literature and scientific research on the implication of global warming and climate change scenarios on streamflow.
MATERIALS AND METHODS
Study area description
The study approach
Description of the SWAT
The SWAT model was applied in the current study to assess the relationship between potential precipitation and temperature changes on the streamflow of the Rietspruit sub-basin. Developed by the United States Department of Agricultural Research Service (USDA–ARS), the SWAT is one of the universally used hydrological models for simulating streamflow. The SWAT is a physically based semi-distributed model used to evaluate the different catchment processes including hydrology, nutrient cycling, sediment yield and land use management practices in river basins with varying soils, land use and management practices (Gassman et al. 2007).
Under the SWAT model, streamflow is calculated at the hydrologic response unit (HRU) level and accumulated to the river channel within the sub-basin. HRUs are the fundamental SWAT building blocks upon which all landscape processes are generated. HRUs are determined by an overlay of a distinct combination of land use, soil and slope in each sub-basin (Srinivasan et al. 2010). Slope classes of 0–1, 1–3, 3–5% and finally greater than 5% were applied in this study, bringing about a total of 16,534 HRUs. Meanwhile, thresholds of 10, 10 and 10% were used each for land use, soil and slope, respectively, which resulted in the reduction of the number of HRUs to 78. With a threshold area of 150 km2, the SWAT model in the Rietspruit watershed produced a total of seven sub-basins.
Data collection
Daily observed precipitation and minimum/maximum temperature data were obtained from the South African Weather Services and the Agricultural Research Council of South Africa. The data were tested for homogeneity (Banda et al. 2021), and only homogenous data and stations were adopted in this study. Monthly streamflow data from the sub-basin outlet at gauging station C2H005 were used for SWAT model calibration and validation. The streamflow data were obtained from South Africa's Department of Water and Sanitation.
For the topographic data, the digital elevation model (DEM) was produced by the Shuttle Radar Topography Mission (SRTM) from the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov/), at a 30-m resolution. Soil data obtained from the Africa Soil Information Service (AfSIS), which has a spatial resolution of 250 × 250 m, were used as bases for SWAT input soil data.
The elevation of the study area ranges from a high of 1,806 m to a low of 1,422 m above sea level. Generally, the topography in the area is gentle, with a slope of less than 1% along the Rietspruit River stretch from the source to the outlet. Meanwhile, land use is dominated by cropland-grassland (CRGR), dryland grass (RYER), range shrubland (RNGB), followed by urban/residential land (URBN), particularly in the eastern part of the sub-basin where the Rietspruit river passes. There are high development activities occurring in the catchment including high and low density urban, mining, industrial and agriculture. The three most dominant soil types in the study area are the haplic lixisols (70%), calcic vertisols (21%) and haplic acrisols (9%) (Figure 4(c). Litisols, which largely cover the upper part of the Rietspruit River, are highly weathered sandy textured well-drained soil and with low clay content and hence have good natural drainage properties. Lixisols usually occur in depressed areas and are characterized by having a presence of shallow groundwater, perched water table and period water saturation (Driessen & Dudal 1991). In contrast, vertisols are cracking clay soils that exhibit a swell/shrink property, thereby causing a discontinuous and non-permanent pore system (Driessen & Dudal 1991). Thus, flooding and high runoff in vertisols are likely to occur easily due to slow infiltration of water.
Parameter sensitivity, model calibration and validation
Parameter sensitivity analysis, model calibration and validation processes are key to successfully applying a hydrologic model. In the present study, sensitivity analysis, calibration and validation were performed using the Sequential Uncertainty Fitting version 2 (SUFI2) algorithm in the SWAT Calibration and Uncertainty Program (SWAT-CUP) (Abbaspour et al. 2015). SUFI2 was used for performing uncertainty analysis as it is more efficient, particularly for complex hydrological models having different dimensions (Zhao et al. 2018).The most sensitive parameters for a particular watershed are used to determine the model calibration and validation processes. Based on SWAT default values and recommendation from previous studies on streamflow analysis, 17 model parameters were initially adopted in this study. However, it should be noted that sensitivity of a parameter varies from one river basin to another, and it reflects well for that one particular basin. Thus, model parameters from other river basins need to be systematically examined before being adopted and used for further analysis and/or calibration.
Model calibration entails estimating and adjusting values for the different model parameters that allow the model to accurately mimic the behaviour of the real system it represents, while validation involves quantitatively and qualitatively evaluating the model using new observed data not utilized in the calibration process to affirm the robustness of the model and transferability of its parameter (Shen et al. 2022). In this study, the SWAT model calibration and validation were performed for the period 2005–2009 and 2010–2012, respectively, using data from the C2H005 gauging station. Data from 2003 to 2004 were used for the model warm period. Model warm-up initializes the various biophysical processes in the model for the purpose of running the entire hydrological cycle process.
Evaluation of model performance
The NSE is a normalized statistic that determines the relative magnitude of the residual variance compared to the measured (observed) data variance (Nash & Sutcliffe 1970). NSE is a measure of how closely the plot of simulated values versus observed values fits the 1:1 line. The NSE ranges from –∞ to 1, such that larger NSE values signify better model performance; hence, NSE equal to 1 is the optimal value (Akhtar et al. 2022). PBIAS calculates the average tendency of simulated values to be greater or lower than the observed values (Mohseni et al. 2023). The PBIAS's ideal value is 0, and low magnitude values suggest accurate model simulation. Positive values indicate underestimation bias in the model, whereas negative values suggest overestimation bias. Contrarily, the coefficient of determination, which ranges from 0 to 1, measures the degree of collinearity between simulated and measured data; hence, larger values indicate a better model performance. Incorporating a combination of different metrics is considered a good practice to get a comprehensive understanding of the model's performance.
Model uncertainty prediction
In general, uncertainties in hydrological models originate from three key sources: parameter uncertainty, the structure of the model and the quality of input and calibration data (Moges et al. 2020). SUFI-2 treats all the different sources of uncertainties in terms of parameter uncertainty and quantifies such uncertainty in terms of the P-factor and R-factor. The model output uncertainty is quantified by the 95% prediction uncertainty (95PPU) calculated at the 2.5 and 97.5% levels of the cumulative distribution of an output variable (Abbaspour et al. 2007).
Hence, further to the above three model performance evaluation approaches (Section 2.3.3), this study also employed the P-factor and R-factor statistics, which are deducted from the 95% Prediction Uncertainty (95PPU), to determine the fit between the simulated and observed data. The P-factor and R-factor statistics evaluate the model output on the basis of its uncertainty, which is ignored in the majority of model performance criteria. It is worth considering that model calibration depends on a number of factors including structure of the model, quality and quantity of data for running model and calibration and modeller's assumptions and expertise; thus, uncertainty analysis is critical when assessing the overall model performance (Abbaspour et al. 2015).
The assessment of model performance based on model uncertainty is necessitated by the fact that, presently, most river basins are no longer in their pristine state and there are various human activities affecting the land and hydrological processes, which makes the catchment scale calibration an increasingly difficult task (Abbaspour 2022). Further, such basins have no to little such management data and/or information coupled with limited and poorly distributed observed climatic variables, thereby making calibration and validation a challenge. Therefore, the use of model uncertainty through statistical measures such as the P-factor and R-factor gives a better indication of model performance, particularly under a continued changing environment. There is a strong correlation between the P- and R-factors, suggesting that a higher P-factor is only a result of a higher R-factor (Akhtar et al. 2022).
Climate change impact analysis and bias correction
Data collection
The current study considered two RCPs of the CMIP5, namely, RCP4.5 and RCP8.5 (Taylor et al. 2012). Historical and future climate data from the CMIP5 for both RCP4.5 and 8.5 were obtained from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) dataset (https://cds.nccs.nasa.gov/nex-gddp/). Each of the climate projections contains precipitation, maximum temperature and maximum temperature at a daily timescale from 1950 to 2100. The dataset has a spatial resolution of 0.25° (∼25 × 25 km). Future climate scenarios were categorized into three different time frames, which were categorized as the near future (NF) (2024–2048), mid future (MF) (2049–2073) and far future (FF) (2074–2098). Nine GCMs from the CMIP5 were initially considered, having future climate projection data for precipitation and minimum and maximum temperatures. The nine preliminary GCMs were CanESM2, CM5A-MR, CSIRO, EC-EARTH, GFDL-ESM2M, MIROC5, MPI-ESM, NorESM1 and SHMI-ESM.
Bias correction
The statistical characteristics of temperature and precipitation projections from climate models and those from observed time series do not perfectly align due to the systematic errors (biases), which can be as a result of imperfect conceptualization, discretization and spatial averaging at very coarse resolutions (Soriano et al. 2019; Wörner et al. 2019).
Due to such limitations of climate models, direct use of GCM outputs to determine hydrological impact of climate change on streamflow is not recommended (Muerth et al. 2013; Vaittinada Ayar et al. 2021). Rather, projections from climate models need to be adjusted to account for their systematic errors and match the spatial resolution and distribution of observed climatic data, through what is termed bias correction. It is the bias-corrected data that can then be adopted to perform impact analysis in the hydrological models. In this study, GCM outputs were bias-corrected based on observed precipitation and temperature data using the climate model data for the hydrologic modelling (CMhyd) tool. CMhyd has been successfully used in bias correcting precipitation and temperature in different studies (Yeboah et al. 2022; Alehu & Bitana 2023; Brighenti et al. 2023).
Different bias correction techniques are available and are applicable for either precipitation or temperature, while other methods are suitable for both datasets. The techniques include the linear scaling of precipitation and temperature, local intensity scaling of precipitation, variance scaling of temperature, power transformation of precipitation, distribution mapping of precipitation and temperature and quantile mapping of precipitation (Teutschbein & Seibert 2012; Fang et al. 2015; Brighenti et al. 2023). In this study, the distribution mapping of the precipitation and temperature method was applied to bias correct the data. The method was chosen because of its ability to efficiently correct biases in the mean, standard deviation (SD), inter-annual variability and extreme events (Wörner et al. 2019; Kim et al. 2020). The distribution mapping technique involves the use of the cumulative distribution functions of the observed and projected climatic parameters to generate a bias correction function (Teutschbein & Seibert 2012). The technique's key assumption is that, over a long time series, the probability distribution function of hydro-meteorological data would be relatively constant and that the simulated probability distribution should closely match the observed probability distribution (Xue et al. 2022). However, such an assumption that the raw modelled and observed data follow the same pattern may potentially introduce new biases and hence may limit its application (Fang et al. 2015), thereby confirming the need to assess the performance of the chosen bias correction method using different statistical measures.
Comparing the various bias correction techniques based on the mean absolute error (MAE) (Teutschbein & Seibert 2012) observed that the distribution mapping was the best performing method. Worku et al. (2020) also observed that, despite all the methods being able to adjust mean monthly and annual regional climate model (RCM) simulations, distribution mapping was more effective in adjusting the 90th percentile of precipitation and temperature and wet day probability of precipitation. Detailed descriptions of the bias correction methods, their advantages and limitations are found in the literature (Gudmundsson et al. 2012; Teutschbein & Seibert 2012; Fang et al. 2015; das et al. 2022b).
In the present study, the performance of the quantile mapping technique was evaluated using five statistical coefficients, namely, SD, root mean square error (RMSE), MAE, PBIAS and the NSE. Closely matching SD between observed and bias-corrected data suggests the appropriateness of the bias correction technique. Both the RMSE and MAE indicate the difference between observed and simulated data of the variables under interest. Lower values of RMSE and MAE indicate a better fit between the modelled and observed data, with 0 being an ideal value.
RESULTS AND DISCUSSION
Performance of the SWAT model
Parameter sensitivity
One important problem in distributed hydrological models is over-parameterization, where the model has more parameters than necessary, thereby resulting in high model uncertainty and/or lower model prediction accuracy (Whittaker et al. 2010). Hence, sensitivity analysis is performed for each individual catchment to identify the parameters that can have a significant influence on the final hydrological model simulation.
The p-value and t-stat from the SUFI-2 sensitivity analysis of the originally adopted 17 parameters were used to estimate the parameter sensitivities. The more sensitive parameter is indicated by higher absolute t-stat values and a smaller p-value (Abbaspour et al. 2015). Global sensitivity analysis was employed to select the most sensitive parameters. Results for the ten most sensitive parameters for the Rietspruit sub-basin, ranked based on their t-stat and p-value, are presented in Table 1.
Parameter . | Description . | Range . | Fitted value . | t-stat . | p-value . |
---|---|---|---|---|---|
R_SOL_AWC(..).sol | Available water capacity of the soil (mm H2O/mm soil) | 0–1 | 0.9499 | −31.36 | 0.00 |
R_SOL_K(..).sol | Saturated hydraulic conductivity (mm/h) | 0–2000 | −0.0026 | −26.23 | 0.00 |
V_EPCO.bsn | Plant uptake compensation factor | 0–1 | 0.7446 | −1.73 | 0.08 |
R_HRU_SLP.hru | Average slope steepness (m/m) | 0.1413 | −1.33 | 0.18 | |
R_SURLAG.hru | Surface runoff lag coefficient (days) | 1–24 | 2.0546 | −1.24 | 0.21 |
R_SLSUBBSN.hru | Average slope length (m) | 10–150 | 39.3958 | 1.17 | 0.24 |
V_GW_DELAY.gw | Groundwater delay (days) | 0–500 | 318.3055 | 0.97 | 0.32 |
V_GW_REVAP.gw | Groundwater ‘revap’ coefficient | 0.02–0.2 | 0.3540 | −0.97 | 0.32 |
R_CH_N2.rte | Mannings n value for the main channel | 0.01–0.3 | 0.1281 | 0.66 | 0.50 |
R_RCHRG_DP.gw | Deep aquifer percolation fraction | 0–1 | 1.1423 | −0.65 | 0.51 |
Parameter . | Description . | Range . | Fitted value . | t-stat . | p-value . |
---|---|---|---|---|---|
R_SOL_AWC(..).sol | Available water capacity of the soil (mm H2O/mm soil) | 0–1 | 0.9499 | −31.36 | 0.00 |
R_SOL_K(..).sol | Saturated hydraulic conductivity (mm/h) | 0–2000 | −0.0026 | −26.23 | 0.00 |
V_EPCO.bsn | Plant uptake compensation factor | 0–1 | 0.7446 | −1.73 | 0.08 |
R_HRU_SLP.hru | Average slope steepness (m/m) | 0.1413 | −1.33 | 0.18 | |
R_SURLAG.hru | Surface runoff lag coefficient (days) | 1–24 | 2.0546 | −1.24 | 0.21 |
R_SLSUBBSN.hru | Average slope length (m) | 10–150 | 39.3958 | 1.17 | 0.24 |
V_GW_DELAY.gw | Groundwater delay (days) | 0–500 | 318.3055 | 0.97 | 0.32 |
V_GW_REVAP.gw | Groundwater ‘revap’ coefficient | 0.02–0.2 | 0.3540 | −0.97 | 0.32 |
R_CH_N2.rte | Mannings n value for the main channel | 0.01–0.3 | 0.1281 | 0.66 | 0.50 |
R_RCHRG_DP.gw | Deep aquifer percolation fraction | 0–1 | 1.1423 | −0.65 | 0.51 |
V, replacement of a parameter value with a new value of a new parameter; R, existing parameter value is multiplied by (1 + a given value).
The listed sensitive parameters had sensitivity values (p-values) ranging from 0.00 to 0.51. The most sensitive parameters were those related to soil properties, groundwater processes, channel flow and those pertaining to water uptake by the plant. The available water capacity of the soil (SOL_AWC), which controls the ability of the soil to store water, is categorized as the top ranked parameter. Higher values of SOL_AWC indicate that the soil has a greater capacity to retain water, which results in less water being accessible for surface runoff and percolation (Jha 2011). Eventually, this will influence other different water balance components including infiltration, plant uptake, evapotranspiration, groundwater recharge and discharge. These characteristics correspond with the prevailing soil types of the study area and particularly the vertisols that cover the middle and lower parts of the Rietspruit River. Vertisols are cracking clay soils that exhibit a swell/shrink property causing a discontinuous and non-permanent pore system (Driessen & Dudal 1991) and have a large surface area and high water holding capacity.
Model calibration and validation
In the field of hydrologic modelling, metrics assessing the performance system usually rely on comparing the simulated and observed streamflow at the outlet of a catchment. Along with the use of graphical techniques, this study employed two forms of statistics to evaluate model performance, namely, evaluation of the model with performance indicators (quantitative statistics) and evaluation with respect to model prediction uncertainty. The model's performance was evaluated based on the R2, PBIAS and the NSE, while the P-factor and R-factor were used to show model uncertainty. Results of the model performance assessment for monthly simulated streamflow are presented in Table 2.
Variable . | R2 . | NSE . | PBIAS . | p-factor . | R-factor . |
---|---|---|---|---|---|
Calibration | 0.62 | 0.60 | 20 | 0.86 | 0.91 |
Validation | 0.64 | 0.61 | 40 | 0.85 | 1.22 |
Variable . | R2 . | NSE . | PBIAS . | p-factor . | R-factor . |
---|---|---|---|---|---|
Calibration | 0.62 | 0.60 | 20 | 0.86 | 0.91 |
Validation | 0.64 | 0.61 | 40 | 0.85 | 1.22 |
Despite the relatively higher value in PBIAS during the validation process, the statistical indicators for the model performance, based on the R2 and NSE, generally indicated satisfactory performance of the model (Moriasi et al. 2007). R2 and NSE were almost similar for both the calibration and validation periods, while the PBIAS increased during the validation period, making the model result unsatisfactory in the validation period (Table 2). As indicated earlier, higher values of R2 indicate less error variance, while higher values of NSE indicate a good fit of observed and simulated data. Hence, the R2 of 0.62 and 0.64 and NSE of 0.60 and 0.61 during calibration and validation are deemed acceptable for monthly streamflow data. The difference in the PBIAS values during the calibration and validation period could be due to non-separation of data for the wet and dry periods during both the calibration and validation periods. Ideal calibration and validation data should typically constitute both average, dry and wet years so that the different hydrological events are incorporated (Gan et al. 1997). Observed streamflow data (Figure 5) clearly show the small peaks and lows including during the dry season; hence, the streamflow cannot entirely be attributed to precipitation. This further indicates that the hydrological processes being modelled exhibit non-stationarity, such that the model did not completely capture the dynamics and temporal variability in the processes during the validation period, hence, the need for evaluating the model performance using different statistical measures.
Although different statistical metrics have been previously used to assess the hydrological model performance based on the goodness-of-fit approach, their sole use has recently been argued not to be sufficient to determine whether the model truly represents the real world hydrologic system. This is true when, among others, the model structure, parameters and its boundary conditions are not adequately understood, thus advancing model equifinality and eventually prediction uncertainty (Acero Triana et al. 2019; Khatami et al. 2019). Addressing equifinality in hydrological modelling involves recognizing the uncertainty associated with model structure and parameterization (Akhtar et al. 2022).
Accordingly, further to the above statistical performance indicators, the P-factor and R-factor were adopted in this study to check the model's performance based on its uncertainty. The P-factor was determined as 0.86 and 0.85 while the R-factor was found to be 0.91 and 1.22 for calibration and validation, respectively. A greater P-factor number (approaching 100%) and a lower R-factor value (approaching 0) are indicators of a good model's uncertainty. Meanwhile, Abbaspour (2022) recommended that for streamflow, a model with P-factor ≥ 0.7 and R-factor ≤ 1.5 can be considered a good model. Based on that, the established P-factor and R-factor during both the calibration and validation periods provide that the model captured the uncertainties well. In this study, the P-factor and R-factor have been mainly used to explain the model performance as these statistical measures incorporate model uncertainty, which is crucial because presently, the majority of the river basins experience continued changing climate and dynamic land use, thus increasing the uncertainties.
Uncertainties in modelling the hydrological impacts of climate change originate from both the climate models (e.g., downscaling uncertainty, emission scenario uncertainty and uncertainties caused by the internal variability of the climate system) and also from hydrological modelling processes (e.g., incorrect structure of the model and use of poor input and observed or calibration data) (Akhtar et al. 2022; Yimer et al. 2022). In the current study, model uncertainty was comparatively minimized by using an ensemble of five best performing model GCM data (Ahsan et al. 2023). Therefore, knowing the level of model uncertainties and minimizing uncertainty is vital to obtaining reliable climate estimates and watershed characteristics information, thereby providing better information for assessing future outcomes and devising adaptation measures. In this study, input data uncertainty can be sourced from different land use and river basin management scenarios, such as water abstraction, effluent discharge and canalization (Aboelnour et al. 2019), and these data are not easily generated or accessible. Incorporating all the different sources of uncertainties is essential in improving the accuracy and credibility of the simulated data (Zhao et al. 2018). Thus, considering the good values of the P-factor and R-factor obtained in this study, such results provide high confidence for performing further analyses. In general, the findings from the model performance on calibration and validation show that the SWAT model uncertainties and performance statistics were falling within the acceptable ranges. This calibrated model can therefore be utilized for a variety of purposes, including planning for and managing water resources and understanding the influence of climate change on streamflow in the Rietspruit River sub-basin.
Impact of climate change assessment
Performance analysis of the bias correction method
All the preliminary obtained nine GCMs data were subjected to bias correction before they were used for further analysis. Each GCM's performance was evaluated for both precipitation and maximum and minimum temperatures. Two representative sites, namely, Westonaria in the higher altitude (1,693 m amsl) and Barrage representing low altitude (1,435 m amsl), were adopted to check the performance, particularly for precipitation. In contrast, temperature data were evaluated from Zuuberkom and Vereeniging due to their long-term observed data availability. To evaluate the performance of the bias correction method, five statistical coefficients including SD, RMSE, NSE, MAE and PBIAS were employed before and after bias correction.
Based on the values of the chosen statistical parameters, five models, namely, CanESM2, CSIRO, MiroC5, NorESM2 and SHMI, were considered to perform better while CM5A-5R, EC-EARTH, GFDL-ESM2M and MPI-ESM performed poorly. Table 3 presents the results of the statistical performance before and after bias correction for precipitation and temperature. Only results from the MiroC5 are shown in the table as a representative. In general, the NSE for both precipitation and temperature significantly increased after bias correction, with the bias corrected data having an NSE between 0.93 and 0.99. Similarly, better and lower values of RMSE, MAE and PBIAS were also observed in all the climatic variables after bias correction. Again, the bias correction resulted in SD values of climate model data being closer to the observed (measured) data, unlike with the raw climate model values. RMSE and MAE values less than half the SD of the observed data can be considered as low, Singh et al. (2005), hence indicating good model performance.
Station . | SD . | RMSE . | NSE . | MAE . | PBIAS . |
---|---|---|---|---|---|
Precipitation | |||||
Westonaria | 48.34* | ||||
Raw | 97.65 | 95.46 | −2.89 | 81.03 | −141.3 |
Bias corrected | 43.43 | 12.45 | 0.93 | 8.65 | 5.38 |
Barrage | 43.95* | ||||
Raw | 60.01 | 32.73 | 0.45 | 24.22 | −38.64 |
Bias corrected | 45.27 | 3.22 | 0.99 | 2.54 | 0.91 |
Tmax | |||||
Zuurbekom | 3.21* | ||||
Raw | 3.04 | 4.36 | −0.84 | 3.87 | 16.04 |
Bias corrected | 3.21 | 0.37 | 0.98 | 0.28 | 1.19 |
Tmin | |||||
Zuurbekom | 5.17* | ||||
Raw | 4.41 | 1.62 | 0.90 | 1.23 | −11.55 |
Bias corrected | 5.28 | 0.15 | 0.99 | 0.12 | −0.08 |
Tmax | |||||
Vereeniging | 3.38* | ||||
Raw | 2.96 | 5.32 | −1.48 | 4.89 | 19.82 |
Bias corrected | 3.31 | 0.17 | 0.99 | 0.09 | 0.38 |
Tmin | |||||
Vereeniging | 5.31* | ||||
Raw | 4.38 | 1.25 | 0.94 | 1.12 | −2.69 |
Bias corrected | 5.35 | 0.05 | 0.99 | 0.03 | −0.20 |
Station . | SD . | RMSE . | NSE . | MAE . | PBIAS . |
---|---|---|---|---|---|
Precipitation | |||||
Westonaria | 48.34* | ||||
Raw | 97.65 | 95.46 | −2.89 | 81.03 | −141.3 |
Bias corrected | 43.43 | 12.45 | 0.93 | 8.65 | 5.38 |
Barrage | 43.95* | ||||
Raw | 60.01 | 32.73 | 0.45 | 24.22 | −38.64 |
Bias corrected | 45.27 | 3.22 | 0.99 | 2.54 | 0.91 |
Tmax | |||||
Zuurbekom | 3.21* | ||||
Raw | 3.04 | 4.36 | −0.84 | 3.87 | 16.04 |
Bias corrected | 3.21 | 0.37 | 0.98 | 0.28 | 1.19 |
Tmin | |||||
Zuurbekom | 5.17* | ||||
Raw | 4.41 | 1.62 | 0.90 | 1.23 | −11.55 |
Bias corrected | 5.28 | 0.15 | 0.99 | 0.12 | −0.08 |
Tmax | |||||
Vereeniging | 3.38* | ||||
Raw | 2.96 | 5.32 | −1.48 | 4.89 | 19.82 |
Bias corrected | 3.31 | 0.17 | 0.99 | 0.09 | 0.38 |
Tmin | |||||
Vereeniging | 5.31* | ||||
Raw | 4.38 | 1.25 | 0.94 | 1.12 | −2.69 |
Bias corrected | 5.35 | 0.05 | 0.99 | 0.03 | −0.20 |
* indicates the standard deviation for the observed (historical) data.
After bias correction, all the variables produced low RMSE and MAE less than half of the observed SD, hence fitting good model evaluation. Such a thorough and accurate model evaluation is crucial to ensure the reliability and applicability of the model's predictions, thereby enhancing the quality of decision-making.
Projection of future climate change around the Rietspruit sub-basin
Projected changes in precipitation
Changes in precipitation and temperature are the main sensitive and naturally contributing factors to streamflow variability. In the present study, the future precipitation and temperature for the Rietspruit sub-basin were projected and quantified under two emission scenarios in reference to the baseline period (1994–2018). The multi-model ensemble mean was generated by averaging the projected precipitation and temperature from each individual GCM (i.e., CanESM2, CSIRO, MiroC5, NorESM2 and SHMI) (Tramblay et al. 2017).
Ensembles of the relative changes in monthly precipitation and temperature are presented in Tables 4 and 5, respectively.
Month . | Jan . | Feb . | Mar . | Apr . | May . | Jun . | Jul . | Aug . | Sep . | Oct . | Nov . | Dec . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Baseline precipitation (mm) | 120.2 | 92.4 | 85.6 | 40.9 | 21.0 | 6.6 | 4.3 | 5.4 | 18.9 | 68.5 | 108.7 | 136.5 |
RCP4.5 | ||||||||||||
% change NF | 31.7 | 47.8 | 3.9 | 21.9 | 36.5 | −0.1 | 44.4 | 171.2 | −5.6 | 24.4 | −10.4 | 3.3 |
% change MF | 18.4 | 36.3 | −7.9 | 2.3 | −1.3 | −38.6 | 14.1 | 99.7 | 14.8 | 52.8 | −15.7 | −5.6 |
% change FF | 31.5 | 21.4 | −16.8 | −1.6 | 32.2 | −24.4 | 113.1 | 40.4 | 75.9 | 64.8 | −11.8 | 13.6 |
RCP8.5 | ||||||||||||
% change NF | 20.5 | 31.5 | 4.3 | −12.0 | 58.2 | −44.6 | −35.6 | 130.6 | 5.4 | 18.3 | −27.3 | −4.9 |
% change MF | 6.8 | 28.3 | −15.3 | −13.8 | −11.6 | −34.7 | −39.3 | 32.5 | −32.9 | 62.6 | −23.1 | −11.5 |
% change FF | 1.9 | 13.8 | −25.7 | −36.1 | −42.1 | −65.5 | −50.2 | 12.1 | 51.0 | 100.8 | −19.9 | 3.7 |
Month . | Jan . | Feb . | Mar . | Apr . | May . | Jun . | Jul . | Aug . | Sep . | Oct . | Nov . | Dec . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Baseline precipitation (mm) | 120.2 | 92.4 | 85.6 | 40.9 | 21.0 | 6.6 | 4.3 | 5.4 | 18.9 | 68.5 | 108.7 | 136.5 |
RCP4.5 | ||||||||||||
% change NF | 31.7 | 47.8 | 3.9 | 21.9 | 36.5 | −0.1 | 44.4 | 171.2 | −5.6 | 24.4 | −10.4 | 3.3 |
% change MF | 18.4 | 36.3 | −7.9 | 2.3 | −1.3 | −38.6 | 14.1 | 99.7 | 14.8 | 52.8 | −15.7 | −5.6 |
% change FF | 31.5 | 21.4 | −16.8 | −1.6 | 32.2 | −24.4 | 113.1 | 40.4 | 75.9 | 64.8 | −11.8 | 13.6 |
RCP8.5 | ||||||||||||
% change NF | 20.5 | 31.5 | 4.3 | −12.0 | 58.2 | −44.6 | −35.6 | 130.6 | 5.4 | 18.3 | −27.3 | −4.9 |
% change MF | 6.8 | 28.3 | −15.3 | −13.8 | −11.6 | −34.7 | −39.3 | 32.5 | −32.9 | 62.6 | −23.1 | −11.5 |
% change FF | 1.9 | 13.8 | −25.7 | −36.1 | −42.1 | −65.5 | −50.2 | 12.1 | 51.0 | 100.8 | −19.9 | 3.7 |
Month . | Jan . | Feb . | Mar . | Apr . | May . | Jun . | Jul . | Aug . | Sep . | Oct . | Nov . | Dec . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Baseline Tmax (°C) | 26.80 | 26.76 | 25.43 | 23.28 | 20.88 | 18.50 | 18.19 | 21.38 | 25.02 | 26.27 | 26.40 | 26.93 |
RCP4.5 | ||||||||||||
% change NF | 1.96 | 1.31 | 2.06 | 2.08 | 2.23 | 1.91 | 2.39 | 1.93 | 1.77 | 2.17 | 1.90 | 1.71 |
% change MF | 2.92 | 2.29 | 2.91 | 2.50 | 2.78 | 2.94 | 4.02 | 3.81 | 3.97 | 2.49 | 3.15 | 2.85 |
% change FF | 3.06 | 2.47 | 2.95 | 2.39 | 2.90 | 3.28 | 4.97 | 4.97 | 4.86 | 2.65 | 3.19 | 3.03 |
RCP8.5 | ||||||||||||
% change NF | 2.17 | 1.57 | 2.32 | 2.44 | 2.40 | 2.44 | 2.67 | 1.92 | 1.93 | 2.24 | 2.23 | 1.96 |
% change MF | 3.89 | 3.39 | 3.92 | 3.56 | 4.05 | 4.05 | 5.27 | 4.91 | 5.80 | 4.03 | 3.70 | 3.53 |
% change FF | 5.66 | 4.88 | 5.41 | 5.40 | 5.99 | 6.49 | 8.17 | 7.63 | 9.17 | 5.27 | 5.61 | 4.85 |
Baseline Tmin (°C) | 17.30 | 16.83 | 14.80 | 10.98 | 6.41 | 2.61 | 2.50 | 6.30 | 11.27 | 14.17 | 15.58 | 16.99 |
RCP4.5 | ||||||||||||
% change NF | −0.67 | −1.02 | −0.82 | −1.00 | −0.99 | −0.52 | −0.97 | −0.84 | −1.99 | −1.38 | −0.78 | −0.94 |
% change MF | 0.28 | −0.40 | −0.42 | −1.19 | −1.24 | −0.38 | 0.11 | 0.78 | −0.10 | −0.51 | 0.28 | 0.03 |
% change FF | 0.51 | −0.29 | −0.57 | −1.49 | −1.15 | −0.20 | 0.87 | 1.91 | 0.80 | −0.06 | 0.83 | 0.52 |
RCP8.5 | ||||||||||||
% change NF | −0.52 | −0.76 | −0.57 | −0.78 | −0.77 | −0.31 | −0.91 | −0.89 | −1.69 | −1.25 | −0.54 | −0.73 |
% change MF | 1.53 | 0.63 | 0.65 | −0.10 | −0.22 | 0.90 | 1.17 | 1.87 | 1.00 | 0.93 | 1.41 | 1.14 |
% change FF | 3.58 | 2.16 | 1.77 | 0.59 | 0.62 | 2.50 | 3.47 | 4.74 | 3.45 | 2.55 | 3.45 | 3.25 |
Month . | Jan . | Feb . | Mar . | Apr . | May . | Jun . | Jul . | Aug . | Sep . | Oct . | Nov . | Dec . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Baseline Tmax (°C) | 26.80 | 26.76 | 25.43 | 23.28 | 20.88 | 18.50 | 18.19 | 21.38 | 25.02 | 26.27 | 26.40 | 26.93 |
RCP4.5 | ||||||||||||
% change NF | 1.96 | 1.31 | 2.06 | 2.08 | 2.23 | 1.91 | 2.39 | 1.93 | 1.77 | 2.17 | 1.90 | 1.71 |
% change MF | 2.92 | 2.29 | 2.91 | 2.50 | 2.78 | 2.94 | 4.02 | 3.81 | 3.97 | 2.49 | 3.15 | 2.85 |
% change FF | 3.06 | 2.47 | 2.95 | 2.39 | 2.90 | 3.28 | 4.97 | 4.97 | 4.86 | 2.65 | 3.19 | 3.03 |
RCP8.5 | ||||||||||||
% change NF | 2.17 | 1.57 | 2.32 | 2.44 | 2.40 | 2.44 | 2.67 | 1.92 | 1.93 | 2.24 | 2.23 | 1.96 |
% change MF | 3.89 | 3.39 | 3.92 | 3.56 | 4.05 | 4.05 | 5.27 | 4.91 | 5.80 | 4.03 | 3.70 | 3.53 |
% change FF | 5.66 | 4.88 | 5.41 | 5.40 | 5.99 | 6.49 | 8.17 | 7.63 | 9.17 | 5.27 | 5.61 | 4.85 |
Baseline Tmin (°C) | 17.30 | 16.83 | 14.80 | 10.98 | 6.41 | 2.61 | 2.50 | 6.30 | 11.27 | 14.17 | 15.58 | 16.99 |
RCP4.5 | ||||||||||||
% change NF | −0.67 | −1.02 | −0.82 | −1.00 | −0.99 | −0.52 | −0.97 | −0.84 | −1.99 | −1.38 | −0.78 | −0.94 |
% change MF | 0.28 | −0.40 | −0.42 | −1.19 | −1.24 | −0.38 | 0.11 | 0.78 | −0.10 | −0.51 | 0.28 | 0.03 |
% change FF | 0.51 | −0.29 | −0.57 | −1.49 | −1.15 | −0.20 | 0.87 | 1.91 | 0.80 | −0.06 | 0.83 | 0.52 |
RCP8.5 | ||||||||||||
% change NF | −0.52 | −0.76 | −0.57 | −0.78 | −0.77 | −0.31 | −0.91 | −0.89 | −1.69 | −1.25 | −0.54 | −0.73 |
% change MF | 1.53 | 0.63 | 0.65 | −0.10 | −0.22 | 0.90 | 1.17 | 1.87 | 1.00 | 0.93 | 1.41 | 1.14 |
% change FF | 3.58 | 2.16 | 1.77 | 0.59 | 0.62 | 2.50 | 3.47 | 4.74 | 3.45 | 2.55 | 3.45 | 3.25 |
In general, there have been varying precipitation projections in the study area under RCP4.5 and RCP8.5, whereby models show precipitation to be higher under RCP4.5 than under RCP8.5 scenarios. Although global climate model projections suggest that mid-latitude and subtropical dry regions are likely to experience a decrease in mean precipitation under RCP8.5 (IPCC 2014a, 2014b), it is clear that precipitation changes will not be uniformly distributed across all the regions. Likewise, studies on future precipitation as a result of climate change have produced contrasting results over the southern Africa region. Using multi-model ensemble projections, Pinto et al. (2016) observed that the total annual mean precipitation will decrease over the southern Africa region, although the magnitude of extreme precipitation events are likely to increase. Bichet et al. (2020) also observed that by the end of the 21st Century, the ensemble mean annual precipitation will decrease in the southern African region. In an attempt to assess the changes in surface water availability as a result of climate change in the Buffalo River Catchment (South Africa), Dlamini et al. (2023) employed an ensemble of six GCMs from the CMIP5 and found that in the 2070–2099 period, there will be a 5% increase in the mean annual precipitation under RCP8.5. Eventually, surface runoff and surface water availability also were predicted to increase in the basin. In a study in Wartburg, an inland area in southeast South Africa, Ncoyini-Manciya & Savage (2022) used projections from the CanESM2 model where they observed an increased trend in precipitation, particularly in the 2050s. Likewise, in Tsitsa catchment (Eastern South Africa), Theron et al. (2021) projected an increase in extreme rainfall events between 2015 throughout and 2100.
Meanwhile, in the Rietspruit sub-basin on a monthly basis, there is no definitive trend in future precipitation, under both RCP4.5 and RCP8.5, as precipitation shows both increase and decrease in different periods. This suggests that there is an increased prospect of intra-annual precipitation variability due to climate change. However, projections from January and February, which are peak precipitation months in the Rietspruit sub-basin, show an increase in precipitation in all the future periods under both scenarios (Table 4). This could be indicative of broader climate change impacts (Caretta et al. 2022); therefore, it is important to understand if the increase is more pronounced in the form of extreme events that can increase the risk of flooding. It is also important to incorporate increased precipitation projections into long-term climate adaptation plans; thus, floodplain management and infrastructure improvements might be necessary to mitigate the increased flood risk.
Projected changes in temperature
It is expected that under all emission scenarios, due to global warming, surface temperatures will increase over the 21st century caused by natural climate variability and anthropogenic emissions. An increase of above 6 °C in the FF as projected in the current study would mean that temperatures in the study area would surpass the maximum projected global average temperature rise of 4.8 °C by 2100 (IPCC 2014a, 2014b). In fact, the study area has already experienced an increase in temperatures in the past two to three decades, with a temperature rise of up to 2.2 °C (Banda et al. 2021). Temperature increases of up to 3.48 °C have also been projected in southeast South Africa (Ncoyini-Manciya & Savage 2022). Rising temperatures have significant implications on water resources not only through enhanced evapotranspiration rates but also by influencing water quality and the magnitude, intensity and timing of precipitation (Caretta et al. 2022; Soren et al. 2023; Zou et al. 2023). Hence, there is a need to enhance adaptation to climate change in different sectors, for instance, by systematically incorporating all the variables and strategies involved in the sustainable development of the resource including investment in resilient infrastructure, improved water-use efficiency and water conservation measures.
From the results, it was also observed that the highest increase in maximum temperature was in July for both time periods and emission scenarios. Meanwhile, observed (historical) temperature time series in the Rietspruit sub-basin indicate that the coldest month is usually observed in July. These findings are similar to a study by Choruma et al. (2022) in eastern South Africa, whose projections showed a more than 50% increase in winter temperatures. When these historically cold months (winter periods) are projected to experience increasing maximum temperatures, such trends can be associated with the broader phenomenon of global warming and ongoing climate change (IPCC 2021), which can disrupt ecosystems and the overall water resource management (Caretta et al. 2022). Hence, adaptation strategies and policies might need to be revised to address such local changing conditions.
The observed decreasing minimum temperatures in the NF explain that climate change is a complex phenomenon and has different implications, considering that the finding is counterintuitive because climate change is often associated with global warming and rising temperatures. Changes in maximum and minimum temperatures are known to affect precipitation patterns; for instance, it has been established that global precipitation increases with warming temperatures and decreases in cold climates (Li et al. 2013). Apart from affecting precipitation patterns, temperature changes impact biodiversity by affecting the distribution and behaviour of various species (Filho et al. 2023). In general, lowering the minimum temperature can also complicate climate adaptation efforts. This is because, presently, the majority of scientists and policy makers are planning for climate change adaptation measures based on increasing temperatures because, generally, the overarching global temperature trend is still towards a warming climate (IPCC 2021). This entails the need that climate change adaptation strategies should be tailored based on the river basin's local characteristics. This is so because while global average temperatures are generally rising due to increased greenhouse gas emissions, local and regional variations can result in seemingly contradictory effects (Sutton et al. 2015), such as lower minimum temperatures in certain areas.
All in all, as expected, significant temperature increases are projected in the MF and FF for minimum temperature, while maximum temperature shows an upward trend for near, mid and far periods under both climate change scenarios. Despite the decreasing projections in minimum temperature under RCP4.5 in the NF, average long-term temperature projections in the Rietspruit under all scenarios show an increase, similar to the historical increasing trend observed in Banda et al. (2021).
Impact of climate change on streamflow
The projected GCM outputs were used to determine the future climate scenarios in the study area. Based on their performance after bias correction (Section 3.2.1), an ensemble of five GCMs (CanESM2, CSIRO, MiroC5, NorESM2 and SHMI) was used because these GCMs were more consistent with the measured precipitation and temperature data. Consequently, the bias corrected outputs from the GCM ensemble were incorporated into the calibrated and validated SWAT model to simulate future streamflow scenarios as a result of potential climate change. To quantify the impact of climate change on streamflow, observed streamflow data from 1994 to 2018 were used as a baseline period and then compared with the projected future streamflow under different timescales (NF, MF and FF).
As indicated in Figure 9, the greatest increase in streamflow is projected to be in the NF under emission scenario RCP4.5. It is projected that streamflow will increase by almost 84% whereby the average annual streamflow will be 4.2 m3/s. This is partly in response to the increase in average annual precipitation (Figure 6), which is projected to increase by 17.2% (831 mm) in comparison to the baseline period. In contrast, the smallest increase in streamflow is also projected in the NF under the RCP8.5 emission scenario from an average of 2.26 m3/s in the baseline period to 3.09 m3/s, representing an increase of 36%.
Month . | Jan . | Feb . | Mar . | Apr . | May . | Jun . | Jul . | Aug . | Sep . | Oct . | Nov . | Dec . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Baseline flow (m3/s) | 2.52 | 2.59 | 2.60 | 2.72 | 2.50 | 2.26 | 2.19 | 2.12 | 1.98 | 1.78 | 1.80 | 2.12 |
RCP4.5 | ||||||||||||
% change NF | 194.2 | 103.5 | 48.4 | 42.9 | 56.1 | 58.9 | 51.5 | 45.5 | 43.5 | 52.7 | 82.2 | 125.9 |
% change MF | 29.8 | 52.4 | 50.5 | 39.9 | 42.2 | 46.3 | 39.0 | 32.9 | 32.6 | 46.9 | 53.3 | 40.5 |
% change FF | 29.7 | 37.9 | 39.9 | 35.5 | 42.1 | 45.5 | 38.1 | 32.1 | 31.5 | 55.1 | 67.5 | 58.6 |
RCP8.5 | ||||||||||||
% change NF | 29.3 | 32.5 | 37.7 | 28.8 | 34.9 | 39.6 | 32.4 | 27.8 | 26.9 | 40.8 | 44.2 | 41.5 |
% change MF | 40.9 | 58.5 | 55.9 | 44.5 | 46.9 | 49.9 | 42.0 | 36.0 | 35.3 | 103.1 | 64.7 | 95.8 |
% change FF | 39.6 | 48.4 | 45.2 | 37.8 | 42.9 | 47.5 | 39.7 | 33.1 | 36.9 | 80.1 | 77.9 | 70.1 |
Month . | Jan . | Feb . | Mar . | Apr . | May . | Jun . | Jul . | Aug . | Sep . | Oct . | Nov . | Dec . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Baseline flow (m3/s) | 2.52 | 2.59 | 2.60 | 2.72 | 2.50 | 2.26 | 2.19 | 2.12 | 1.98 | 1.78 | 1.80 | 2.12 |
RCP4.5 | ||||||||||||
% change NF | 194.2 | 103.5 | 48.4 | 42.9 | 56.1 | 58.9 | 51.5 | 45.5 | 43.5 | 52.7 | 82.2 | 125.9 |
% change MF | 29.8 | 52.4 | 50.5 | 39.9 | 42.2 | 46.3 | 39.0 | 32.9 | 32.6 | 46.9 | 53.3 | 40.5 |
% change FF | 29.7 | 37.9 | 39.9 | 35.5 | 42.1 | 45.5 | 38.1 | 32.1 | 31.5 | 55.1 | 67.5 | 58.6 |
RCP8.5 | ||||||||||||
% change NF | 29.3 | 32.5 | 37.7 | 28.8 | 34.9 | 39.6 | 32.4 | 27.8 | 26.9 | 40.8 | 44.2 | 41.5 |
% change MF | 40.9 | 58.5 | 55.9 | 44.5 | 46.9 | 49.9 | 42.0 | 36.0 | 35.3 | 103.1 | 64.7 | 95.8 |
% change FF | 39.6 | 48.4 | 45.2 | 37.8 | 42.9 | 47.5 | 39.7 | 33.1 | 36.9 | 80.1 | 77.9 | 70.1 |
The Rietspruit sub-basin is characterized by four main climatic seasons, namely, summer (DJF), autumn (MAM), winter (JJA) and spring (SON), whereby a great percentage of rainfall is received in summer. Thus, normally it is expected that the greatest streamflow will be observed in the summer period. However, as depicted in Table 6, the greatest increase in streamflow in summer was only observed in the NF under RCP4.5, corresponding to the increase in precipitation in this period. Meanwhile, there was almost a similar increase in future periods in all the seasons, suggesting no significant precipitation contribution to the overall streamflow change.
Although streamflow was projected to increase under all emission scenarios and future decades, there was no observed correlation between streamflow increase and precipitation or temperature changes. For instance, streamflow in the mid decades under RCP8.5 also substantially increased (by 74%), but there was no noticeable change in annual precipitation between the baseline (708 mm) and the MF period (715 mm), while during the same period, minimum and maximum temperatures were projected to increase between 0.9 and 4.1 °C. Typically, under natural circumstances, streamflow increase will mean that the impact of precipitation exceeds that of temperature. Again, theoretically, variations in streamflow over time reflect changes in either precipitation or land use/land cover (Gupta et al. 2015). A river basin that experiences an increase in streamflow without a corresponding increase in precipitation clearly suggests that besides climatic variations, land use changes and human influences such as water transfers into the river and withdrawals play a major role in influencing the overall streamflow (Aboelnour et al. 2019; Serrão et al. 2022). This entails that in addition to precipitation and temperature, changes in streamflow in the Rietspruit sub-basin are influenced by external factors including anthropogenic activities. Additionally, the projected streamflow increases in the dry season aligned with the historical streamflow during the baseline period, which also showed an increasing trend even during the dry season (Banda et al. 2022). Therefore, the projected increase in streamflow including during the dry season indicates that the SWAT model was able to capture external processes like additional input of water into the hydrological system from other sources apart from precipitation.
It is widely recognized that the variations in long-term river flows are a function of changes in water availability, including precipitation, evapotranspiration and abstractions. However, while the influence of climate change on streamflow is being assessed, it is essential to consider the contribution of land use changes on the overall flow changes. Land use changes, which are largely triggered by rapid socio-economic developments, have a direct influence on the streamflow as they affect peak runoff, evapotranspiration, groundwater recharge, baseflow, sediment yield and flood frequency (Aragaw et al. 2021; Serrão et al. 2022). For example, Idrissou et al. (2022) projected that surface runoff in the Inland Valley Catchment (Burkina Faso) would increase due to the combined impact of climate change and land use changes, whereby in comparison to the reference period, surface runoff was 158% higher when the combined impacts of climate and land use changes were assessed, while the runoff increase was 52% when only climate change was considered. Haleem et al. (2022) also observed that river runoff in the upper Indus basin will increase, whereby the largest increase will be due more to climate change than land use change. Gyamfi et al. (2016) also noted that the increase in urbanization in the Olifants Basin in South Africa contributed to the increase in surface runoff up to 47%. Land use plays a substantial role in the overall streamflow as it determines the amount of surface runoff, infiltration, groundwater recharge and the evapotranspiration rate. Thus, it is important that the future basin land use/land cover need to be projected, such that the combined and individual impacts of climate change and land use changes need to be evaluated.
The Rietspruit sub-basin holds significance as a vital river catchment contributing water to the Vaal river basin, an economically crucial catchment for South Africa's economy for many years (DWAF 2004; du Plessis 2017). Modelling the Rietspruit sub-basin processes is useful for understanding its hydrological processes including the water balance and nutrient load, which have a major impact on the sustainability of the water resource and environment of the region. As already noted, different regions and sub-basins exhibit varying climate change characteristics; hence, climate-resilient approaches and adaptive management need to be identified at a local scale, similar to the Rietspruit sub-basin. By modelling to predict future climate scenarios, findings from this study will provide the local government with necessary futuristic data in order to put in place bio-economic strategies to mitigate the negative impacts.
Overall, assessment of the potential impact of climate change on streamflow has different implications for the Rietspruit sub-basin, ranging over water resource planning, ecosystem health, urban planning and agricultural development. As precipitation is predicted to increase, net water supply may not increase due to increasing temperatures and demand from different users. On the other side, among the daunting environmental challenges facing the Rietspruit sub-basin is the deteriorating water quality, largely from non-point source pollution from urban, residential, mining and agricultural activities (Dzwairo & Otieno 2014; Raji et al. 2022). Increasing precipitation and runoff can result in overloaded storm and sewer systems causing the entry of untreated pollutants into the natural water courses. The deteriorating wastewater infrastructure in the Rietspruit has caused big water quality challenges (du Plessis 2021; Sindane & Modley 2022). Hence, the current findings are vital for urban planners and water authorities to design climate-resilient and sustainable infrastructure, which can be achieved by having an integrated knowledge and extent of climate change in the area. Over the coming decades, the demand for water resources from competition users will presumably increase in the upper Vaal catchment, owing to increasing economic activities and population growth in the area (du Plessis 2021). Despite the projected increase in streamflow, efficient water resource management would also require a holistic assessment of the competing users ranging from social and economic aspects as well as maintaining ecosystem services. Hence, effective planning and management of water resources in the Rietspruit sub-basin should incorporate issues like population growth and urban development in addition to climate variability.
CONCLUSIONS
This study aimed to evaluate the potential impact of climate change on the hydrology of the Rietspruit sub-basin using five best-fit GCMs (ACCESS1-0, CanESM2, CNRM-CM5, GFDL-ESM2M and MIROC5). The sub-basin's historical and climate model output data were integrated into the calibrated SWAT, which evaluated the future streamflow under RCP4.5 and RCP8.5 greenhouse gas forcing. The SWAT model performance was evaluated using various statistical matrices. Using the SUFI-2 algorithm in SWAT-CUP, sensitivity analysis showed that soil moisture (SOL_AWC) and saturated hydraulic conductivity (SOL_K) are the two most sensitive parameters in the study area.
Findings indicate that under both RCP4.5 and RCP8.5, precipitation will increase, although there will be a greater increase under RCP4.5 scenarios (17.2, 9.5 and 16.8%) than in the RCP8.5 (6.3, 0.9 and 3.6%) under NF, MF and FF periods, respectively. Temperatures are also projected to rise, with a maximum rise of 3.35 and 6.21 °C under RCP4.5 and RCP8.5, respectively. However, projections show minimum temperatures getting lower in the NF and MF under RCP4.5, particularly during winter months. Compared to the baseline period (1994–2018), streamflow in the Rietspruit sub-basin is projected to increase by 42–75% and 34–50% under RCP4.5 and RCP8.5, respectively.
Despite acknowledging the influence of land use/land cover changes on surface runoff and streamflow, the present study did not incorporate in the SWAT model the potential changes in future land use/cover. Hence, similar future studies need to include the combined impact of climate and land use changes amid continued development and management changes in different river basins. On this note, use of shared socio-economic pathways from the CMIP6 is also recommended since these emission scenarios incorporate potential societal changes including industrial developments, population growth, land use and climate policy changes. Although findings indicate an increase in streamflow, it is imperative to understand current and anticipated future water demands from competing users to better understand the surface water availability and water quality situation in the Rietspruit. Knowledge of potential climate change should be integrated in the design of water infrastructure in the study area. Further, climate change adaptation strategies should be tailored based on the river basin's local characteristics Overall, the present findings provide a significant platform for working out long-term plans for managing water resources in the Rietspruit sub-basin and developing climate change adaptation measures.
FUNDING
The financial assistance of the South Africa National Research Foundation (NRF) is hereby acknowledged. Opinions expressed and conclusions arrived at are those of the authors and are not necessarily to be attributed to the NRF. The research is on the Rietspruit sub-basin under the grant BRICS multilateral R&D project (BRICS2017-144), the NRF UID number 116021 and the Durban University of Technology UCDG Water Research Focus Area grant. The research was also financially supported by the University of the Western Cape under the Research Incentive Funds.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.