Assessment of climate change impacts on hydrological processes in the Usangu catchment of Tanzania under CMIP6 scenarios

Changes in temperature and precipitation patterns serve as clear indicators of climate change, which primarily arises from the increased concentration of greenhouse gases in the atmosphere. The IPCC's 2022 report highlights its widespread impact on vulnerable communities, particularly in Africa. Multiple studies confirm altered temperature and precipitation patterns, with projections showing rising temperatures (0.8–2 °C) and changes in precipitation patterns in southern Tanzania. Urgent climate action and adaptation measures are essential to protect these vulnerable regions (Mamuye 2013; Mwakalila 2014; Paul 2016; FCFA 2017; Hyandye et al. 2018; Näschen et al. 2019a; Africa 2020; Hersi et al. 2022; Mwabumba et al. 2022; Nobert 2022; Tibangayuka et al. 2022).

Climate changes directly affect catchment hydrological processes, influencing water availability (Mamuye 2013; Paul 2016; Hyandye et al. 2018; Näschen et al. 2019b; Shagega et al. 2020; Gao et al. 2021; Wen et al. 2021; Tibangayuka et al. 2022). As precipitation levels increase, evaporative loss, water output, surface runoff, and streamflow also rise, as evidenced by several studies (Mehan et al. 2016; Paul 2016; Liu et al. 2022). Rising temperatures lead to increased evaporation rates which leads to increased water consumption by vegetation, resulting in significant reductions in river flows and aquifer infiltration. This can result in decreased water availability and reduced streamflow, which impacts ecosystems, agriculture, and human water supply.

The Usangu catchment is significant for wildlife ecosystems and agriculture, contributing notably to the country's rice production and acting as a vital water source for Ruaha National Park. However, population growth has led to forest and wetland conversions into irrigation farms, altering river patterns and streamflow, causing water management challenges (Kashaigili & Majaliwa 2010; Mutayoba et al. 2018). Climate change worsens water balance issues, making the catchment vulnerable to water scarcity, evidenced by the drying of the once perennial Great Ruaha River (GRR) since 1993 (Mccartney et al. 2008; Kashaigili et al. 2009). Therefore, it is important to assess the hydrological condition of the Usangu catchment in relation to climate change as the major factor affecting water balance components.

Hydrological models, such as the Soil Water Assessment Tool (SWAT) model, are essential for understanding how climate change affects water balance. Previous studies have shown the effectiveness of SWAT in predicting climate and land management impacts, particularly in data-scarce regions (Neitsch et al. 2002; Meaurio et al. 2015; Hyandye et al. 2018; Sáez et al. 2018; Kishiwa et al. 2018; Mutayoba et al. 2018; Msovu et al. 2019; Näschen et al. 2019b; Pandi et al. 2023; Shinhu et al. 2023). Additionally, previous studies used Global Circulation Models (GCMs) from Coupled Model Intercomparison Project 5 (CMIP5) scenarios along with SWAT models to analyze climate change's impact on hydrology (Hyandye et al. 2018; Näschen et al. 2019b; Adib et al. 2020; Gurara et al. 2021; Wen et al. 2021). However, the introduction of CMIP6 GCMs has brought notable improvements in GCMs performance in terms of spatial resolutions, earth system processes, and future outcome ranges (Kumar et al. 2022; Sun et al. 2022; Hersi et al. 2023). Also, GCMs' coarse resolutions still struggle to capture local climate variations, necessitating downscaling to finer resolutions. Weather Generators (WGs), like the Long Ashton Research Station-Weather Generator (LARS-WG), offer a valuable solution by generating daily weather variables that match observed data, enabling faster computation and downscaling of GCM predictions (Chisanga et al. 2017; Mehan et al. 2017; Kishiwa et al. 2018; Mwabumba et al. 2022; Tibangayuka et al. 2022). This approach provides a promising way to assess climate change's impact on hydrological processes at finer scales.

Therefore, this research combines the SWAT model with the newer CMIP6 GCMs, given that only a limited number of studies from the region and none from the case study have utilized CMIP6 model projections to quantitatively evaluate the effects of climate change on hydrology. Additionally, the study addresses the challenges posed by limited ground data availability by integrating readily available satellite data from Climate Hazards Group Infrared Rainfall with Station data (CHIRPS) and Climate Forecast System Reanalysis (CFSR) datasets. Furthermore, the study utilizes the LARS-WG, a notable stochastic weather generator, which effectively represents climate variables with empirical distributions and offers greater flexibility compared to standard distributions found in other generators like WeaGEN, CLIGEN, and WGEN, as observed in recent studies (Mehan et al. 2017; Hersi et al. 2022; Munawar et al. 2022; Tibangayuka et al. 2022).

Data used for analysis include DEM, land use/cover (LULC) map, soil map, weather data (temperature and precipitation data), and future climate data (GCMs) as described in Table 1. However, the ground-measured data in the Usangu catchment are sparsely distributed due to the uneven weather station network mainly caused by the presence of protected areas. Moreover, for temperature data, only two stations (Mafinga and Lupatinga) had data but with large gaps missing. Since it is important for hydrological modeling to account for the contribution of all the streams that join GRR, alternative data are important. Nowadays, open-access satellite-based measurements have evolved as an alternative to sparsely distributed ground-based observations. Rainfall data from CHIRPS and CFSR have been widely used and proven to be good rainfall and temperature estimates (Hyandye et al. 2018; Dhanesh et al. 2020; Mulungu & Mukama 2022; Mwabumba et al. 2022). Therefore, these datasets were used in this study. Future projections for temperature and rainfall obtained from CMIP6 GCMs under SSP 2–4.5 and SSP 5–8.5 were used in the simulation of future streamflow and other water balance components.

The weather and streamflow data were checked for quality before being used for hydrological analysis. The process involved filling of missing data, removal of outliers, ground validation of the satellite data and checking continuity of the data and gap filling using simple linear regression methods in the R software to generate time series data for the stations with missing data as in Cherie (2018) and Kamwaga et al. (2018). Change point analysis was performed in the Khronostat software to select a baseline period with no breaks in the time series. CFSR and CHIRPS rainfall datasets were used for infilling of temperature and rainfall data, respectively.

Global dataset bias correction was carried out over the timeframe of 1990–2014, employing the linear scaling approach as outlined in Hersi et al. (2022) and Koffi et al. (2023). This methodology is widely adopted internationally, involving applying a correction factor to the global datasets with the aim of enhancing their alignment with the observed climate, as evidenced by previous studies (Chen et al. 2015; Hersi et al. 2022; Koffi et al. 2023).

The GCMs were statistically downscaled using the Long Ashton Research Station Weather Generator (LARS-WG) under SSP 2–4.5 and SSP 5–8.5. This weather generator was initially calibrated and validated using the observed station data. The calibration process involved comparing the statistical properties of the generated data with independent observed data to ensure an improved agreement between them. The performance of LARS-WG was evaluated using Kolmogorov–Smirnov (K–S) test, for comparisons of probability distributions, means and standard deviation (Chisanga et al. 2017). p-values in the K–S test were used to determine whether the generated data had the same probability distribution as the station data (Chanter 1990; Chisanga et al. 2017; Tibangayuka et al. 2022). The minimum p-value used was 0.01, whereby values below it indicate poor performance of the model (Hersi et al. 2022; Tibangayuka et al. 2022).

After calibration and validation of LARS-WG, the model was used for future predictions of climate variables. LARS-WG simulation contains a number of biases when simulating present-day climate (Kishiwa et al. 2018). Therefore, an ensemble means of three best-performing GCMs in temperature and precipitation were compared with the baseline data for climate change analysis.

The Usangu watershed was delineated and further subdivided into 31 sub-catchments. The next step was on HRU creation, which involved overlaying the baseline classified 1990 Land use map, soil map and the slope map generated from the DEM. Weather generated and station data files were fed into the model and the files were written. The surface runoff and the evapotranspiration were simulated using the curve number and Hargreaves equations selected in the model interface, respectively. The Hargreave potential evapotranspiration equation was selected due to insufficient weather data. Topographic and Hydrologic Response Unit (HRU) reports from the SWAT model setup were used to describe clearly the present hydrological processes in the catchment with their respective governing parameters. These parameters were optimized in SWAT-CUP software for a better physical representation of the catchment during calibration and validation.

SWAT calibration and validation were done using the SWAT-CUP model, which is a standalone optimization program. Streamflow data from the Msembe gauging station (1KA59) were selected based on data availability and position of the station to capture flow from the upstream areas and in Ruaha National Park located at the downstream of the catchment (Figure 1). Since the Usangu catchment has a significant number of water abstractions since the 1990s, therefore, calibration and validation periods selected were 1981–1985 and 1986–1990 respectively based on years with fewer streamflow data gaps and minimum water abstractions in the catchment. The Nash–Sutcliffe (NS) coefficient was set as an objective function for calibration in SWAT-CUP. Further, manual calibration on the parameters was performed for goodness of fit. The calibration process was preceded by global sensitivity analysis, which assisted in parameter selection for calibration. The t-test and p-value test were used to measure the sensitivity of the parameters.

The study evaluated the model's performance in simulating catchment hydrology using statistical and graphical analyses, including discharge hydrographs. Three metrics, namely Nash–Sutcliffe Efficiency (NSE), Coefficient of Determination and Percent Bias (PBIAS), were employed to assess the model's performance. Nash–Sutcliffe Efficiency (NSE) ranges from −∞ to 1 and measures the similarity between simulated and observed data, with 1 indicating a perfect fit and ≤0 indicating that the observed data mean is a better predictor. The Coefficient of Determination (R²) describes the degree of collinearity between simulated and measured data, assessing their alignment. The Percent Bias (PBIAS) measures the average tendency of the simulated data to deviate from observations, with low values indicating better simulation, positive values suggesting model underestimation, and negative values indicating overestimation. The statistical guidelines proposed by Moriasi et al. (2015) were used to determine the model's acceptability for use, considering values of NSE > 0.65, R² > 0.5, and |PBIAS| ≤ 25% as indicators of satisfactory and efficient model performance (Table 2).

After calibrating the SWAT model, it was utilized to simulate hydrological processes for two distinct time frames, i.e., the baseline period from 1990 to 2014 and the future period spanning from 2030 to 2060. Throughout these simulations, the land use map for 1990 was retained. The output values of the SWAT model for these two periods (baseline and future) were presented as relative changes in annual, seasonal, and monthly percentages.

From the topographic statistics presented in Table 5, about 56.26% of the catchment is below the mean elevation of 1,389.5 m and covers about 0.08% of the watershed. The topographic variation was evaluated using the coefficient of variation (CV). The CV for the Usangu catchment was 26.93% and the proportional of the small slope is 64.67%. From the land use HRU analysis report, woodland, bushland, grassland and wetland covered 47.17, 26.48, 15.51, and 9.46% of the total watershed area, respectively. The Usangu catchment is highly dominated by loam and sandy loam soil, which facilitates more infiltration and thus subsurface runoff processes. Following the nature of topography, vegetation cover and soil characteristics, the runoff generation mechanism may be through infiltration excess (Horton flow) and overland flow during higher rainfall events. Higher rates of evapotranspiration and relatively low deep recharge are expected to occur. The nature of the soil will facilitate more groundwater recharge. Therefore, from the topography, land use and soil type, a set of 20 parameters were selected and used in the calibration and validation of the SWAT model.

In the future periods, the annual actual evapotranspiration was projected to increase by 26 and 22% in SSP 2–4.5 and SSP 5–8.5, respectively. On a monthly scale, there was a significant ET increase of more than 30% in January, September, October, and November (Figure 14(b)) compared to the rest of the months. Increased evapotranspiration is subsequently caused by higher temperature changes in these months during future periods. Higher ET would lead to a decrease in soil moisture and therefore adversely affect groundwater supplies, water yield, and plant growth in the catchment. More losses were expected to occur from the agricultural land and the wetlands located at the central and northeast (NE) parts of the catchment.

Results revealed that changes in ET are higher during dry seasons, therefore with decreased precipitation during those months, an increase in irrigation requirements is expected to occur in the catchment. Consequently, river flows in the catchment will also decrease and further cause additional water stress in the Water–Energy–Food nexus sectors in the catchment. Spatially, the wetlands and swamps situated in the central area will experience more rise in ET than other parts of the catchment. Hyandye et al. (2018) argued that increased ET in the Usangu catchment would mean the cultivation of water-resistant crops such as sugarcane as an adaptation measure.

The study reveals that annual groundwater recharge to the shallow aquifer will decrease by 26% for SSP 2–4.5 and 18% for SSP 5–8.5. Notably, a significant 40% reduction occurs during the dry season (MJJASO), mainly in August and September for both scenarios, attributed to increased evapotranspiration (ET) and surface runoff within the catchment. The decrease in the groundwater recharge is evidence that river flows would further decrease as a result of lower baseflows. Shallow aquifers in the catchment exhibit substantial water-holding capacity, potentially serving as accessible water resources with low input costs during the dry season. However, increased ET's impact on decreasing groundwater recharge surpasses the rise in the surface runoff, particularly evident in SSP 2–4.5. Groundwater is recommended as an alternative water source for extensive irrigation in the Usangu catchment (URT 2015). Reduced recharge may exacerbate water disputes due to excessive reliance on river flows, emphasizing the need for sustainable water management strategies and equitable groundwater utilization.

The total annual water yield in the catchment is predicted to decrease by 7% for SSP 2–4.5 and increase by 5% for SSP 5–8.5 in the future. The water yield increase in SSP 5–8.5 is attributed to a notable precipitation increase in this scenario, which contributes more to surface runoffs and further to water yield. Notably, a nearly 30% decline occurs during the dry season (MJJASO), with the peak reduction in August and September resulting from rivers drying in the catchment. In the wet season (NDJFMA), water yield is projected to notably increase for SSP 5–8.5, particularly in December, January, and February. However, November demonstrates a drop in water yield for both scenarios by 32% for SSP 2–4.5 and 25% for SSP 5–8.5 (Figure 14(c)). Moreover, when viewed on an annual basis, water availability in the catchment is not seen as a constraining factor for production in the catchment. Previous studies argued that an increase in the water yield during months with increased precipitation is caused by a temporal shift in the plant growth pattern with minimum water loss through transpiration and less irrigation water demand in the catchment (Arranz & Mccartney 2007; McCartney et al. 2007; Kashaigili et al. 2009; Mwakalila & Masolwa 2012; URT 2015; Hyandye et al. 2018). Therefore, during months with increased water yield, water storage reservoirs should be constructed in the catchment to serve water demands during dry seasons.

The main objective of this study was to assess the impacts of climate change on hydrological processes in the Usangu catchment. The hydrological response under baseline conditions was simulated using the calibrated SWAT model. However, the model was unable to capture the peaks well, but generally, the data showed goodness of fit, making it suitable for use. Under the baseline period, total annual precipitation was 763 mm, whereby 39% of the precipitation was withdrawn by actual evapotranspiration, 19% returned as the streamflow, and groundwater recharge to shallow aquifers was 34%. The study revealed that groundwater contributed about 70% of the total water yield in the catchment while 30% was contributed by surface runoff.

Climate change projections were made using an ensemble mean of five GCMs downscaled using the LARS-WG statistical tool under SSP2–4.5 and SSP5–8.5 emission scenarios. This weather generator was found to be a reliable tool for downscaling with an average correlation of 0.99 between the generated and observed rainfall and temperature data and a large p-value of greater than 0.01 for all the simulated months. Annual precipitation was projected to increase by 7 and 17% for SSP2–4.5 and SSP5–8.5 emission scenarios, respectively. In both scenarios, the average temperature was projected to rise by 0.6–2 °C. Surface runoff and streamflow will increase significantly during wetter seasons in the future as precipitation increases. This will further aggravate flooding events in the catchment, thus calling for suitable adaptation measures such as reservoir construction for water storage during the wet season. However, during the dry season, these components were projected to decrease even more, thus increasing the possibility of water conflicts in the catchment. Temperature increase in the catchment is shown to escalate water loss through evapotranspiration, which in turn will lead to a decrease in water yield in the catchment.

Therefore, these findings collectively emphasize the sensitivity of water balance components to climate change and bring the need for integrated and sustainable water management strategies. Implementation of adaptive measures such as improved irrigation practices, groundwater recharge enhancement, and flood control mechanisms will be pivotal in navigating the challenges posed by changing hydrological conditions. Collaborative efforts among stakeholders, supported by informed decision-making, will be crucial in safeguarding water resources and building resilience against the impacts of climate change.

• The utilization of SWAT modeling revealed an overestimation of high flows and a concurrent underestimation of low flows, which may be attributed to the inadequate representation of rainfall distribution within the catchment. Hence, it is recommended that upcoming investigations enhance the depiction of rainfall distribution in the catchment to rectify this issue.

• Also, due to the fact that most areas are protected and ungauged, there was a need to use global data in some areas. These datasets tend to underestimate and overestimate the station data, henceforth they come with lots of uncertainties. Therefore, it is recommended to improve and modernize the hydro-meteorological monitoring system within the catchment.

• The study predicted shallow aquifer storage holds more water than deep aquifer, which could easily be extracted for use with low input costs during the dry season.

• The results showed a higher increase in the surface runoff, which could potentially be captured in a well-designed water harvesting structure such as ponds and dams.

• Future studies should consider the impacts of climate change on wetland's water availability and ecological dynamics.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.