Modeling future streamflow under climate and land use scenarios in the lower Okavango River basin

Climate and land use/land cover (LULC) change are primary drivers of hydrological changes, particularly in water-scarce transboundary river basins. This study addresses the need to understand these interactions for sustainable landscape management in the semi-arid Okavango Basin. The objective was to assess the projected streamflow response of the lower Okavango River Basin under combined climate and LULC change scenarios. We quantified future changes in streamflow and water yield across three periods, extending to 2100. In the transboundary lower Okavango River Basin, we applied the SWAT+ hydrological model. The model was forced with downscaled climate projections and land use change scenarios under moderate (SSP2-4.5) and high (SSP5-8.5) emission pathways. Results project a 5–20% decrease in precipitation by 2100, leading to a severe decline in streamflow of up to 60%. Dry season flows are projected to nearly cease. Concurrently, projected land use change indicates that cropland will expand to over 65% of the basin, replacing natural wetlands. The annual water yield becomes concentrated along the river channel, with most of the basin generating less than 200 mm of water. The basin's hydrology is highly sensitive to these combined changes. The combined climate and LULC pressures threaten the basin's hydrological integrity and water security. This underscores an urgent need for integrated land and water management. We recommend strengthening transboundary governance and implementing nature-based solutions, such as wetland conservation.

Climate and land use/land cover (LULC) changes are both interrelated global environmental challenges. The beginning of industrial revolution in the mid-1800s is a major contributor to climate change, through rapidly increasing greenhouse gas emissions (IPCC, 2022). Climate change has adversely affected food security and terrestrial ecosystems through increasing frequency of extreme events globally. Temperature warming and precipitation patterns are prominent features of climate change, with direct

impact on hydrological responses (Kaini et al., 2010). Recurrence of extreme rainfall and drought events are signs of climate change which in-turn influence hydrological responses. Similarly, LULC changes have impacts on streamflow, affecting evapotranspiration and infiltration, causing erosion and influencing the water balance of hydrological cycles (Guo et al., 2020). Therefore, the determination of complex dynamics of climate and LULC change, through analysis of time series of various indicators and better understanding of trends, patterns, and variability of climate-related phenomena, is crucial to support sustainable water resource management. This complexity arises from multi-dimensional characteristics of climate change, lack of long-term observational datasets, and the intrinsic uncertainties in predictive models.

Land degradation and desertification are global environmental problems which are attributed to climate variability and human actions e.g., loss of biodiversity due to farming activities. As such, it is vital to consider biophysical as well as socio-economic drivers to combat desertification and land degradation (IPCC, 2023). The Okavango River basin is regarded as one of the most important freshwater landscapes located in the dryland of northeastern Namibia, Northwest Botswana, and South Angola. Nevertheless, the basin is under constant pressure due to climate variability, fast population growth and varying land use practices (Mendelsohn and El Obeid, 2004). The semi-arid characteristics of the lower Okavango River basin further enhance environmental sensitivity toward human induced land degradation and climate change. Pressure on natural resources, driven by increasing population and intensified farming in close proximity to the basin, strongly impact the river's hydrological dynamics such as streamflow conditions, sediment transportation and infiltration. Approximately ninety-five percent of streamflow of the Okavango basin originates from Angola and passes through Namibia before dissipating into the Okavango Delta in Botswana (Andersson et al., 2003). Climate and LULC changes upstream will have a vital consequence downstream in the already water scarce environment.

Therefore, investigation of catchment behavior along with LULC and climate change predictions have the potential to provide insight into the feasibility of managing water and land demand and supply, while also supporting the agenda of Sustainable Development Goals (SDG) to whom these countries are signatories. A comprehensive approach to study human influence on natural resources for sustainable management of land and water resources involves assessment of watershed behavior under different land use and climate scenarios. The proposed research analyses streamflow responses of the lower Okavango River basin to climate and land use changes.

Climate and LULC changes play a vital role in predicting future hydrological responses. Researchers use projected climate data and LULC projections in conjunction with hydrological modeling to estimate future streamflow (Acharki, 2023; Ahiablame et al., 2017; Chawanda et al., 2024; Marhaento et al., 2018). Climate projection is typically produced using General Circulation Models (GCMs) and Regional Climate Models (RCMs). GCMs are attributed with reliable climate simulation at a large scale with global coverage with RCMs additionally used to generate higher resolution data with smaller spatial coverage (Chen et al., 2013). Model Outputs from GCMs and RCMs are used to improve scientific understanding of climate systems providing necessary knowledge to support informed political decision-making. Although, GCMs are generally in coarse resolution, statistical and dynamic downscaling methods have lately been used to refine outputs and bridge the gap with RCMs. Several studies using deep learning and machine learning techniques have shown downscaled GCMs to be reliable for studies at regional as well aslocal scale (Polasky et al., 2023; Soares et al., 2024). The sixth Coupled Model Intercomparison Project (CMIP6) is the latest phase of the GCM dataset used by the Intergovernmental Panel on Climate Change (IPCC) for its sixth assessment report. Unlike its predecessors CMIP3 and CMIP5, a new set of SSP scenarios are included as well as modeling groups, advanced climate simulations and increased spatial resolution (Eyring et al., 2016). Research on assessing future climate change using CMIP6 datasets show consistency of climate predictions at regional scale (Chaturvedi et al., 2024; Gbode et al., 2025).

Land use predictions recommended by CMIP6 for the different SSP and RCP scenarios are produced using different Integrated Assessment Models (IAMs). IAMs aim to integrate multi-disciplinary knowledge between socioeconomic and biophysical processes to construct representative sectors from energy to economics to climate and ecology (Parson and Fisher-Vanden, 1997). Each model can predict several SSP-RCP scenarios. There are several IAMs such as GCAM, IMAGE, AIM, REMIND-MAgPIE that generate LULC products based on climate change impacts, energy and food demand, policy choices and socioeconomic trends using Shared Socio-economic Pathways (SSPs) and Representative Concentration Pathways (RCPs). The resolutions of most available global land use projection products are usually coarse ranging between 10 and 50 km (Meyers, 2012). However, scenario-based simulation using SSPs and RCPs of future LULC changes has become a powerful tool for change assessment, providing references for sustainable land management. One such product is the 1km resolution Global land projection dataset spanning 85 years (2015–2100) for 20 land cover types (Chen et al., 2022). This datasets adopted SSP and RCP scenarios and is based on a Land-Use Harmonization (LUH2) project which reconstructed historical land use for future scenario land use patterns in a resolution of approximately 25 km (Hurtt et al., 2020). LUH2 land use projection datasets are used as input in CMIP6 Earth system models (ESMs) ensuring consistency across climate models. Therefore, making the 1 km projected global LULC datasets reliable and globally applicable to support research and applications in fields of climate change, land use planning and environmental monitoring (Chen et al., 2022).

Hydrologic models can simulate and better understand and manage hydrologic processes through simplification of the real-world water resource system (Refsgaard et al., 2007). Accurate data to represent respective variables remains a significant challenge, even though these models integrate complex spatial and temporal variables to account for catchment heterogeneity (Mengistu et al., 2019). Specific parameters define models to represent the real-world conditions of hydrologic processes. Therefore, models that explain hydrological dynamics are critical for sustainable water resources management, especially in areas subject to extreme drought and flooding (Hrachowitz et al., 2013). Various models have been used to study streamflow responses, including the Soil and Water Assessment Tool (SWAT), Variable Infiltration Capacity Model (VIC), and Hydrologic Engineering Center—Hydrologic Modeling System (HEC-HMS). Among these models, the SWAT model is widely used to simulate the water budget of a watershed. SWAT is a semi-distributed hydrologic model that is continuous over time and includes various parameters such as land use/land cover, soil properties, topography, and climate variables (Arnold et al., 2012). SWAT helps to predict system behavior and analyze key processes, including streamflow dynamics, sediment and nutrient loads in catchments with varying LULC, climate, topography, and soil conditions (Vallam et al., 2014).

In our previous research (Negussie et al., 2022), the ability of the SWAT model to simulate streamflow dynamics in the semi-arid Okavango catchment was demonstrated, implying it as a reliable tool for hydrological modeling in this data-scarce region. Other studies done in southern Africa have shown that variations in land cover, together with climate change majorly affect streamflow variability, particularly in semi-arid basins where the hydrological process is very sensitive to vegetation cover and rainfall intensity (Funk et al., 2015; Kusangaya et al., 2014). Recent global assessments, including that of the Intergovernmental Panel on Climate Change (IPCC), show that future climate scenarios will render arid and semi-arid regions much drier hence making it more pressing to implement sustainable land and water management strategies (FAO and UNEP, 2020; IPCC, 2023). Based on this, another study (Negussie et al., 2024), determined the suitability of land for sustainable legume-based agriculture in the Okavango River basin under the assumptions that land use and climate changes would have an effect on streamflow. Since this assumption has never been validated empirically in our study area, the present study is set to fill in that gap by applying SWAT+ as an upgraded version of SWAT to simulate future streamflow responses under combined climate and LULC scenarios. In doing so, it will also verify our earlier assumptions made in assessing land suitability. More importantly, it brings to the forefront the significance of integrated hydrological modeling with land use planning and climate adaptation strategies in semi-arid transboundary river basins.

This study investigates the impact on streamflow response within the lower Okavango River basin considering future climate change predictions and LULC change factors. Results provide information on projected streamflow under combined climate and LULC change scenarios. The main aim of this research is to develop reference materials for water and land resource managers, policymakers and concerned professionals to make informed decision and improve resource management and planning contributing in development of future adaptation strategies. The specific objectives are to: (i) to evaluate SWAT+ model performance at monthly time steps of the lower Okavango River basin using historical data to expose catchment behavior; (ii) to use SWAT+ with global daily downscaled climate projections derived from GCM of CMIP6 and the global land projection dataset to simulate future streamflow projections under climate and LULC scenarios for SSPs 4.5 and 8.5; and (iii) assess the changes of future streamflow in three periods (2025–2040), (2040–2070), and (2070–2100) using respective climate and LULC scenarios in comparison to historical periods.

Study area

The study area as seen in Figure 1 below, focuses on the transboundary lower Okavango River Basin, which includes the Rundu townlands in northeastern Namibia and neighboring areas of southern Angola (Mendelsohn and El Obeid, 2004). The area around Rundu is situated along the Okavango River, forming a natural border between the two countries. The region shows a semi-arid climate receiving an average annual rainfall of 450–550 mm, mainly during the rainy season (October to April) with rainfall peaks during January and March (Atlas of Namibia Team, 2022). Major contributions to local livelihoods include smallholder farming, predominantly rain-fed subsistence agriculture being highly dependent on rainfall seasonality, complimented by livestock rearing (FAO, 2014).

Recently, a rapid increase in population size along the river has been observed. Rundu, the second most populated town in Namibia, is located at the heart of the study area. Since the last census, it has almost doubled its population to 120,000 inhabitants (Namibia Statistics Agency, 2011). Predominantcrops grown in the area are pearl millet (in the local language, mahangu), maize, and sorghum, which are typically grown subsistently by local farmers on relatively small land between 2 and 5 hectares in size (Kamwi et al., 2015). The soils in the area differ substantially, with soils further from the river corridor mostly classed as low-nutrient arenosols, while river floodplains have nutrient-rich fluvisols and gleysols, supporting a higher degree of agricultural activity (Coetzee, 2020).

The study site has a very diverse and ecologically rich landscape including floodplain grasslands, woodlands, and wetlands (Mendelsohn and El Obeid, 2004). Nevertheless, there has been a significant shift in land use with more cultivation and expansion of urban sprawl, primarily in and around Rundu (Kamwi et al., 2015). Upstream precipitation from the Angolan highlands and LULC patterns are critical to regulate downstream flow regimes on the Angolan side, (Andersson et al., 2006). As a result, this localized watershed is a focal point for hydrological impact assessment within the broader Okavango system and highly susceptible to both climate variability and land use change (Wolski and Murray-Hudson, 2005).

Digital elevation model

Due to the uniform topography of northern Namibia and southern Angola, where our study area is located, we obtained a high-resolution TanDEM-X (DEM) from the German Aerospace Center (DLR), a product of the TerraSAR-X DLR mission. This product has a relative vertical accuracy of 2 meters and a spatial resolution of 12 meters (Wessel et al., 2018). The absence of high-resolution information such as TanDEM-X in the study site could potentially hinder or distort hydrological models, impacting the results of simulations. This high-resolution product proved vital in precisely delineating watershed boundaries, essential for creating HRUs and representing the smallest operational unit for streamflow simulation in the model. As demonstrated in Figure 2, the elevation map of the river basin derived by TanDEM-X is presented together with the generated subbasins of the catchment.

Soil mapping

The role of soil information in streamflow modeling is well documented. The incorporation of input variables such as soil information and resolution of soil data have significant impact on modeling streamflow in the SWAT+ model. A soil map of 1:250 000 scale was generated using soil sample profiles collected within the study area (Coetzee, 2020). Subsequently, using the acquired soil profile data, we extracted vital information for the model, including but not limited to texture classes, soil types, profile depth, and other soil attributes essential for model development. The study catchment identified seven soil classes: Cambisol, Fluvisol, Arenosol, Technosol, Fluvic Cambisol, Eutric Fluvisol, and Dystric Arenosol. The resulting soil map demonstrating the distribution of soil classes across the catchment is presented in Figure 3.

Model setup and calibration

A hydrological SWAT+ model was used to simulate the water balance components of the different sub-basins in the study area. The modeling process consisted of delineating the lower Okavango River basin, dividing it into sub-basins, and further subdividing into Hydrological Response Units HRUs to account for its spatial heterogeneity (see Figure 4). HRUs are the smallest spatial units within a catchment, having similar slope, soil, and LULC features. These units allow the simulation of water balance components, assuming that hydrologic characteristics vary between HRUs. Climate, discharge, slope, LULC, and soil variables derive different SWAT+ scenarios (see Figure 4 for the overall workflow). Subsequently, these scenarios simulated the amount of water in each sub-basin, resulting in the total water departing and arriving at each time step in and out of the channel. The water balance was determined using the following equation.

where SW t denotes the soil water content at time t, the initial soil water content is expressed by SW 0, R day refers to the amount of precipitation on day i, Q surf signifies surface runoff on day i, E a is referring to the amount of evapotranspiration on day i, W seep denotes the amount of percolation on day i and Q gw indicates groundwater return flow on day i.

In a previous research, we evaluated SWAT model and found it to be suitable for water balance calculation in the Okavango catchment (Negussie et al., 2022). Therefore, in the current study similar model set-up was applied using related attributes, to compute water balance in the catchment and to avoid duplication of efforts. Due to the flat terrain characteristics of the lower Okavango catchment, TanDEM X high-resolution digital elevation product was used to delineate the river basin in SWAT+ model setup and to divide the catchment into sub-basins. Preliminary steps include filling DEM depressions, computing flow direction and accumulation, and setting stream definition thresholds to determine stream networks and sub-basins. Subsequently, unique pairings of LULC, soil, and slope information within each sub-basin created HRUs. The combined streamflow from each HRU make-up the sub-basin water flow enabling the calculation of channel's water balance for the catchment area. Six weather stations, optimally distributed throughout the study area, were identified for rainfall, temperature (Tmin and Tmax), relative humidity, wind speed, and solar radiation input data into the SWAT+ model. Finally, a 2-year warm-up period (1981 and 1982) was used to run the simulation over a 35-year period (1981–2015). Warm-up is vital for model adjustment and reaching stable conditions (for example, flow values), as SWAT+ simulations typically underestimate basin initial conditions (Kim et al., 2018). Therefore, in this study, the hydrological parameters of the catchment were simulated for 33 years (i.e., excluding the warm-up period). This served as the baseline period through retrospective simulation to evaluate changes of prospective simulation under combined LULC and climate change.

Six sub-basins, ranging from 49 to 335 km 2 were delineated using TanDEM X which were further subdivided into HRUs. A slope map with four classes (0–5.5, 5.5–15.5, 15.5–34.5, >34.5) was generated using the same high-resolution DEM for the study area. LULC were identified from SWAT database resembling land types in the study site, such as AGRL, RNGB_SUDRF, WATR, URBN_WARM, SHRB, WETL, GRAS and BSVG representing dryland farming, tree cover, water bodies, built-up area, shrubland, wetland, grassland, and bare land and sparse vegetation, respectively. Seven soil classes were recorded for the study area namely Cambisol, Fluvisol, Arenosol, Technosol, Fluvic Cambisol, Eutric Fluvisol, and Dystric Arenosol. Thereafter, a unique combination of LULC, Soil type and slope class generated 1857 HRUs. Surface runoff occurs when precipitation that accumulates on the ground surface exceeds the infiltration rate, resulting in water flow from the surface onto the ground. During a rain event in a particular catchment location, the Soil Conservation Service Curve Number, or SCS CN, determines the surface runoff. This runoff simulation considers soil moisture conditions, soil hydrologic group, and LULC types. Potential evapotranspiration (PET) in the catchment was estimated using Solar energy, wind speed, humidity, and temperature as input through the Penman-Monteith method (Allen et al., 1998). The water flow dynamics within the basin's stream network were simulated using the variable storage routing technique pioneered by Williams (1969).

Following our previous study in the catchment, eight parameters were selected that best represent the semi-arid characteristics of the catchment area. The SWAT+ parameters which were chosen based on their sensitivity to the study site are CN2, ALPHA_BF, GW_DELAY, OV_N, ESCO, SOL_AWC, CH_K2, and SURLAG which represent the potential for surface runoff, baseflow response to recharge, groundwater delay time, speed of overland flow, soil evaporation, water storage capacity of soil, water movement through the streambed, and delay between rainfall and the generation of surface runoff, respectively. Upon completion of SWAT+ simulation, calibration and uncertainty analysis were conducted in Sequential Uncertainty Fitting version 2 (SUFI 2) in SWAT-CUP package. A split sample technique for calibration and validation process was employed using Rundu gauge station monthly streamflow data with calibration period from 1983–1996 and validation from 1997–2001. During the calibration, the model parameters are refined by running four iterations of at least 500 simulated runs each, where each round of iteration recorded smaller values of parameter ranges. Results from simulation were compared using P-factor (recommended > 0.7) and R-factor (recommended < 1.5) indices to demonstrate a good fit. Furthermore, model performance was evaluated using objective functions such as Nash Sutcliffe coefficient (NSE), Coefficient of determination (R 2), Percent bias (PBIAS) and Ratio of RMSE (RSR) to demonstrate agreements between observation and simulation of streamflow. A performance rating criteria from (Moriasi et al., 2007) as shown on Table 1 was used to evaluate model performance.

Table 2 summarizes the performance of the SWAT+ model during calibration and validation periods at a monthly time step. The model achieved satisfactory to good performance according to the criteria proposed by Moriasi et al. (2007). NSE and R 2 values indicate strong agreement between simulated and observed streamflow, while PBIAS values demonstrate acceptable bias levels. The uncertainty metrics (P-factor and R-factor) fall within recommended ranges, indicating adequate parameter uncertainty representation. The relatively short validation period reflects discharge data availability constraints typical for transboundary semi-arid African basins and may limit representation of interannual extreme events.

LULC scenario setup

Global land projections from (Chen et al., 2022) at 1 km resolution (2015–2100) was used to run scenario-based analysis in our model. The initial future land use classifications of the global land projections were based on the 2015 European Space Agency's Climate Change Initiative (ESA CCI) land cover datasets. The process of creating future LULC data followed three steps, (i) forcasting various land use using different SSP and RCP scenarios, (ii) generating the highest resolution of global land use products to date at 1-km resolution with seven land use classes (i.e. forest, grassland, barren land, cropland, urban, water and permanent snow and ice) using future land use simulation model (FLUS), and (iii) dividing the seven land classes into 20 plant functional types. The global land projection 1-km dataset based its simulation on CMIP6 recommended LULC product from LUH2 project. The SSP and RCP scenarios for LUH2 future land use dataset were reconstructed from various integrated IAM results. Our study focused on two LULC scenarios: SSP2, which represents a moderate socioeconomic pathway with balanced economic and environmental policies, and moderate greenhouse gas emissions (RCP 4.5); and SSP5, which represents a rapid economic growth pathway driven by fossil fuels, with high greenhouse gas emissions (RCP 8.5). The trajectory from 2020 to 2100 for global cropland and urban land use under SSP2-4.5 scenario showed a gentle and steep increase. Without policies for sustainable development and continuation of business-as-usual economic and environmental policies, agricultural activities will further increase, and urban centers will expand to meet demands of a rapidly growing population. Similarly, under SSP5-8.5 with rapid economic growth based on fossil fuels, grasslands are observed to decline, while urban areas are expected to double in size by 2100 compared to the initial prediction, presenting worst-case scenario. Furthermore, analysis performed in sub-Saharan Africa showed extensive forest degradation in scenarios other than SSP1 assuming lower levels of economic growth and greenhouse gas emissions indicating a more sustainable development due to biodiversity protection policies. Details of approach and analysis of land simulation processes and their different SSP scenarios are documented by Chen et al. (2022).

A combination of 24 runs for climate change and LULC scenarios of SSP2-4.5 and SSP5-8.5 were performed in SWAT+ for three time periods between 2025–2040, 2041–2070, and 2071–2100, respectively. When observing the projected LULC data in our study site for the above periods, we noticed an extreme decline in water bodies/wetlands. At the same time, a sharp increase in cropland was followed by a more gradual expansion of urban areas. The land types considered in our SWAT+ model were shrubland, tree cover, cropland, grassland, bareland, sparse vegetation, urban, and water bodies. Future streamflow processes in the catchment were simulated in SWAT+ using climate change and LULC projections assuming soil and slope data to be constant.

Climate scenario setup

Climate projections required for this study were sourced from three CMIP6 GCMs, namely GFDL-ESM4, IPSL-CM6A-LR, and NorESM2-MM of the NEX-GDDP-CMIP6 global daily downscaled climate projections. These datasets went through a process of bias correction and spatial disaggregation to refine coarse-resolution outputs for regional and local scale climate change impacts. Making it possible to understand potential climate patterns at a regional or local scale (for example watersheds, urban areas, and regional administrative boundaries). In this study we compiled a dataset from climate projections of the aforementioned GCMs across two SSPs (SSP2-4.5 and SSP5-8.5) for the period 2015–2100. These climate projections have been downscaled to a spatial resolution of approximately 25 km. Historical climate records from 1950–2014 were used to adjust future climate projections. These three GCMs were selected for three reasons: (i) to use downscaled higher spatial resolution needed for applications in regional watersheds rather than using original coarser resolution, (ii) availability of all required variables showing patterns at local and regional scale (iii) documented use of these models near and similar areas of the catchment under investigation (Engelbrecht et al., 2011; Joubert and Hewitson, 1997; Karypidou et al., 2022; Ringler, 2010; Samuel et al., 2023). The selection of three CMIP6 GCMs represents a compromise between ensemble diversity and computational feasibility in a data-scarce basin, reflecting a balanced approach to model selection that considers both model capacity and practical constraints. Three criteria drove the selection of GFDL-ESM4, IPSL-CM6A-LR, and NorESM2-MM. Firstly, these models span a range of climate sensitivities and precipitation responses over southern Africa. Secondly, they are available within the bias-corrected NEX-GDDP-CMIP6 framework. Thirdly, and perhaps most importantly, these models have been widely applied in hydrological impact studies in semi-arid African basins. While a broader ensemble might more accurately reflect the full range of uncertainty, as discussed above earlier research suggests that multi-model means derived from a carefully chosen subset can effectively represent regional trends. Therefore, the results should be interpreted as indicative only, not probabilistic.

In order to address the systematic biases that are frequently encountered in raw GCM outputs, we employed the NEX-GDDP-CMIP6 dataset. This dataset employs a statistically bias-corrected and spatially disaggregated (BCSD) framework. The application of quantile mapping in aligning GCM data with observed historical distributions facilitates the acquisition of finer spatial resolutions, a prerequisite for basin-scale analysis, while preserving the integrity of long-term climate signals. This approach is well-supported by prior studies in semi-arid African contexts, where BCSD-based products have consistently outperformed raw model data in capturing hydrological processes. However, we remain cautious regarding its limitations in representing local convective processes and extreme events, where some residual uncertainty is expected.

Daily projected climate variables (precipitation, minimum and maximum temperature, wind speed, relative humidity, and radiation) were extracted from the three GCMs in six locations spatially distributed over the catchment. These variables were prepared in a format fit for SWAT+ and were used to simulate scenarios in a total of 24 simulations for the selected GCMs across SSP2-4.5 and SSP5-8.5. The approach combined the projected climate and LULC scenarios to observe effects on annual streamflow for three periods between 2025–2040, 2041–2070, and 2071–2100. Furthermore, comparative evaluations were performed between the baseline scenario and the three future scenarios.

The analysis evaluates projected changes in precipitation, streamflow, and water yield using SWAT+ model output pre-conditioned by three CMIP6 GCMs. The analysis focuses on estimating changes in precipitation, streamflow, and water yield across three future time periods. Results are discussed in relation to land use and land cover projections in order to illustrate combined effects of climate change, effects of human-induced land transformation on water availability, spatial hydrologic patterns, and ecological sustainability in a transboundary watershed.

Projected precipitation and streamflow changes

Precipitation and streamflow projections based on SWAT+ simulations forced with three CMIP6 GCMs (GFDL-ESM4, IPSL-CM6A-LR, and NorESM2_MM) show a consistent drying trend over three future periods (2025–2040, 2041–2070, and 2071–2100) for the localized section of the Okavango River Basin near Rundu, Namibia. All three models predict a continuous decrease in rainfall relative to the historical baseline, especially during the peak rainfall period (January to March), with the decrease becoming more apparent toward end of the century. Similarly, runoff shows an even more significant and non-linear decrease compared to precipitation, with historically high runoff months (February through April) showing noteworthy decreases. These dynamics are illustrated in Figures 5a–f, where future monthly rainfall and streamflow changes are shown for the SSP2-4.5 and SSP5-8.5 scenarios. The level of streamflow decline exceeds the decline of precipitation. This highlights a higher hydrological sensitivity of the basin to climatic variability. Our findings align with observations made by Ndhlovu and Woyessa (2021) for the neighboring Zambezi catchment. Their results indicated significant streamflow reductions despite low to moderate declines in precipitation, especially during dry months of the year. Similar nonlinear streamflow responses have been documented in the Limpopo and Orange basins, where moderate rainfall declines resulted in disproportionate runoff reductions (Aich et al., 2014; Goulden et al., 2009). These parallels reinforce the broader vulnerability of semi-arid transboundary systems in southern Africa to compounded climatic and land-use pressures.

The observed decline in streamflow, when considered in relation to precipitation, underscores the compounding effects of increased evapotranspiration demand, reduced soil moisture storage, and diminished groundwater recharge. Rising temperatures have been shown to have two main consequences for water demand in the atmosphere. Firstly, they have been found to enhance the demand for water, while secondly, they have also been found to reduce the infiltration capacity and the soil water retention of the land, due to the conversion of that land to a different use. Consequently, the contributions from baseflow exhibit a precipitous decrease, particularly during the dry season, resulting in a state of near-cessation of flow despite the continued presence of seasonal rainfall. This nonlinear response is distinctive feature of semi-arid systems nearing a hydrological “tipping point,” where even minor climatic shifts can trigger a disproportionately large systemic collapse.

Furthermore, this pattern is in line with the “dry gets drier” paradigm, as examined and presented by Xiong et al. (2022) for terrestrial water storage projections. Their global evaluation based on terrestrial water storage anomalies led to the conclusion that semi-arid and transitional regions are strongly sensitive to climate-induced decreases in water availability even for modest precipitation variability. The Okavango Basin has already been stressed due to seasonal scarcity of water and high evapotranspiration; hence, it gets directly placed within this vulnerable category. Aich et al. (2014) and Goulden et al. (2009), in their respective studies of the Limpopo and Orange basins, reported similar findings. This further implies that semi-arid southern African river systems are highly sensitive to variability in rainfall patterns.

In the context of our study area, the persistent reduction in both precipitation and streamflow across scenarios and GCMs coupled with a clear inter-model consensus, provides strong evidence that the hydrological regime of the Okavango Basin will likely experience intensified stress in the future, particularly under high-emission scenarios. This reinforces global and regional assessments which suggest that semi-arid systems in sub-Saharan Africa will continue to exhibit increased sensitivity to climate change, further validating the “dry gets drier” paradigm projected for water-stressed environments.

The above-mentioned patterns become more apparent when analyzed by period. By 2040, precipitation will decline slightly by about 5–10%, depending on the model, while runoff will decline significantly, particularly during the months of March and April. For example, January precipitation is projected to decline from a historical baseline of approximately 124 mm to around 100 mm under SSP2-4.5, while March decreases from 65 mm to approximately 50 mm. However, the corresponding streamflow drops more sharply from 695 m3/s to approximately 113 m3/s in January, and from 557 m3/s to 46 m3/s in March highlighting the basin's nonlinear runoff response to even modest rainfall reductions. The variability between GCMs remains relatively high during this period, reflecting a higher uncertainty.

For the period between 2041 and 2070, precipitation decreases by a further 10-20%, while the decrease in runoff becomes much more noticeable, with extremely low runoff during the dry season (May-October). By mid-century, the predictions of all models begin to converge, indicating an increase of drying trend confidence. Precipitation in January drops from the baseline of 124 mm to approximately 79 mm (SSP2-4.5) and 86 mm (SSP5-8.5), while streamflow falls to 110 m3/s and 108 m3/s, respectively. Similarly, March values decline to 54 mm and 51 mm in precipitation, with flow reducing to 90 m3/s and 31 m3/s. Dry season months such as August show near-zero rainfall and streamflow ranging from 0–5 m3/s, suggesting seasonal cessation of flow in most models.

The most significant decreases in precipitation, often exceeding 20% relative to baseline values, are observed between 2071 and 2100. During this time, dry season river flow will nearly cease, and wet season peak flows will be both lower and shorter in duration. By 2100, January precipitation drops to 100 mm under SSP2-4.5 and 85 mm under SSP5-8.5, with streamflow reducing to 107 m3/s and 101 m3/s, respectively. March rainfall declines further to 42 mm and 38 mm, while flow decreases to 56 m3/s and 48 m3/s. During August, several models simulate complete cessation of flow. These results confirm the transition toward shorter, weaker wet seasons and more extended, intensified dry periods.

Differences between different climate models' projections have narrowed, suggesting a growing consensus that a severe drying trend will persist by 2100. These results confirm projections made for neighboring river basins, the Limpopo and Orange rivers (Aich et al., 2014; Goulden et al., 2009). They also reflect the global pattern as outlined by the “dry gets drier, wet gets wetter” paradigm. The agreement among models by 2100, suggests that the Okavango River Basin is likely to experience a fundamental shift in its hydrological regime, with substantial implications for water availability as well as ecological sustainability and transboundary management.

A closer examination reveals that streamflow responds disproportionately to changes in precipitation, showing that even modest rainfall amounts can lead to a significant decrease in river discharge. This highlights a high dependency of the Okavango River Basin hydrology on upstream rainfall from the Angolan highlands, where small changes in precipitation have large impacts on downstream water availability. Any reduction in upstream rainfall is likely to have an increased impact in the basin near Rundu, given that local rainfall contributes little to the total streamflow in this region. Similarly, susceptibility of semi-arid catchments to climatic variability were highlighted in a study on Rietspruit river in South Africa where climate projections in SWAT using multiple GCMs were investigated and found considerable decline in streamflow under different future climate scenarios (Banda et al., 2024). As a result, the projected decline of river flows poses a serious risk to water security and may affect domestic water supplies, irrigated agriculture, and the ecological sustainability of the Okavango. Traditional crop growing seasons could be disrupted, drought stress worsened, and reliance on already scarce groundwater resources increased by the loss of distinct seasonal flow peaks.

While there is still uncertainty in the early projection periods (2025–2040), the GCM outputs tend to converge by 2100, allowing for higher confidence in the direction of the projected changes. While there are still uncertainties concerning the extent of change, all models agree on the general direction of drying. The intensity of these hydrologic impacts is directly associated with the emission scenarios. Specifically, the drying trend exhibits greater severity under higher emission scenarios, such as SSP585. In contrast, the most significant hydrological impacts may be mitigated to some extent through strategic reduction of global emissions, thereby offering a potential counterbalance. The predictions indicate an essential need for integrated water management and climate adaptation efforts to ensure the conservation of water resources and livelihoods dependent on the Okavango River in the surrounding region.

Projected water yield and land use land cover change

Using the SWAT+ model, our analysis of projected water yield and LULC changes indicates a shift toward and expansion of drying conditions in the Okavango Basin near Rundu between the baseline period and 2100. During the historical baseline, the study region's northern parts, coupled with some portions in the center, demonstrated water yield potential above 200 mm/yr. Figures 6a–d illustrates the spatial distribution of annual water yield across different periods featuring prominent areas between 250 and 350 mm/yr and local peaks surpassing 350 mm/yr. These high-yield areas were associated with dense wetlands and tree cover-indicating a system that has been supported by natural vegetation, rainfall-driven infiltration, and base flow storage in the past.

The years between 2025 and 2040 show a significant decrease in the number of high-yield zones, with the highest yield class (>300 mm/yr) completely disappearing. Areas with a water yield of < 200 mm/year expand north and southwards, while areas with higher water yields become increasingly concentrated along river corridors. This shift is consistent with LULC projections illustrated in Figure 8, which show an increase in cropland and urban development near Rundu, coupled with a considerable reduction of wetland extent over the study area. Cropland cover increases dramatically between 2021 and 2040, as shown in Figures 8a, b, going from about 30% to over 45% of the total basin area. Urban land also grows very fast, nearly doubling in size around Rundu during this period. A notable increase in water bodies is seen in the 2040 projection. This temporary increase in modeled water bodies likely reflects episodic floodplain inundation captured by the land-use classification rather than a sustained increase in surface water availability, as subsequent periods show a consistent decline. For the most part, it is seen along the flood-prone areas next to the Okavango River, which might be attributed to modeled inundation resulting from high runoff concentration or possible overrepresentation of ephemeral flood storage areas captured under surface water classes. However, this increase is not carried forward into subsequent projections and therefore probably represents a short-term hydrological occurrence rather than a structural enhancement of water availability.

These results are consistent with our previous findings of SWAT model simulations of the Okavango-Omatako catchment and indicate significant water yield and baseflow reductions due to land use conversion from vegetation into cropland (Negussie et al., 2022). A previous study by Vushe et al. (2014) found that land use and land cover changes, such as the conversion of forests and grasslands to urban and agricultural areas, have significantly impacted the streamflow of the Okavango River, underscoring the substantial influence of human activities on river flow dynamics. These changes suggest that human-induced alterations to the land, particularly the conversion of wetlands to impermeable or low-infiltration surfaces, are exacerbating the already projected moderate declines in rainfall, as modeled by the GCMs.

A phase of increased aridity is projected to occur between 2041 and 2070, where only limited parts of the landscape remain within water yield classes of 250–300 mm/yr. The majority of the basin, particularly the southern and northern regions, will be characterized by the lowest water yield class. Simultaneously, LULC predictions show extensive urbanization, substantial loss of wetland cover, and the predominance of arable and shrubland across the landscape. By 2070, cropland accounts for almost 60% of the total study area, as urban areas grow more in the west and central parts of the basin. Wetlands shrink and are mostly taken over by cropland and shrubland. These results highlight the combined effects of climate and land use changes on streamflow reductions. This trend is also supported by Figures 7a, b, which shows declining surface runoff across all GCMs and future periods under both SSP2-4.5 and SSP5-8.5.

In particular, under SSP2-4.5 (Figure 7a), models such as GFDL-ESM4 and NorESM2-MM project a consistent decline in mean surface runoff from baseline values of 74 mm to approximately 29 mm and 21 mm by 2040, and further to around 13 mm and 18 mm by 2070. Similarly, the IPSL-CM6A-LR model show very low runoff projections throughout, declining to below 10 mm by the end of the century. This suggests that these models are more sensitive to precipitation declines and possibly more responsive to land degradation assumptions embedded in the SWAT+ outputs.

Under SSP5-8.5 (Figure 7b), the decreasing trend remains evident, though some inter-model variations are observed. Notably, IPSL-CM6A-LR projects comparatively lower runoff values around 7–15 mm, representing approximately 80% decline by 2100. While GFDL-ESM4 and NorESM2-MM under this high-emission scenario also simulate sustained runoff decreases with runoff values ranging between 12 and 27 mm, aligning with predictions of reduced infiltration and enhanced surface runoff due to land cover changes.

By 2100, water yields across the basin reach critically low values, with almost all areas falling below 200 mm/yr. Moderate yields are maintained only in a few isolated areas near the river corridor. LULC projections for this period indicate that cropland will become the dominant land cover in the basin with a total area coverage of approximately 65%, with urban areas, primarily around Rundu, also expanding, while notable decline in wetlands from the catchment. Most natural vegetation buffers and wetland areas disappear in the basin. These gradual shifts demonstrate a definite move toward significant LULC changes, harming ecosystem benefits and worsening surface water losses. This is clear from Figures 8a–d, with projections of land use change from 2021 to 2100.

Combined effects of climate and land use change on water sustainability/basin hydrology

The simultaneous occurrence of severe precipitation and streamflow reductions alongside extensive land degradation suggests that the region is approaching or may have already crossed ecological thresholds beyond recovery, prompting irreversible damage, even under best-case future climate scenarios. While both climate and land-use change contribute to projected streamflow decline, climate-induced precipitation reductions appear to be the dominant driver of annual flow reduction, particularly after mid-century. However, LULC conversion intensifies spatial redistribution of water yield by reducing infiltration capacity and weakening baseflow contributions, thereby amplifying hydrological stress at sub-basin scale. This interaction suggests that climate change sets the overall drying trajectory, while land-use change accelerates and spatially redistributes its hydrological impacts. These trends reveal three critical dynamics: (1) the significant impact of climate-induced drying on water yield patterns; (2) the growing significance of land-use change, particularly the loss of wetlands and the expansion of urban areas, in exacerbating hydrologic loss; and (3) the emergence of spatial inequalities in water availability, with only limited areas near rivers retaining significant water yields by end of the century.

These trends are consistent with the broader regional context, as highlighted by Kusangaya et al. (2014), who observed the effects of climate change on water resources in Southern Africa. Their research underscores the compounding effect of LULC changes and climate variability on reduced water yield and increased vulnerability to water scarcity. A report by the Food and Agriculture Organization (FAO, 2014) on the Okavango River basin notes that population growth and land use changes, such as the expansion of cropland and urban areas, have led to changes in streamflow patterns and water quality. This has serious consequences for local water security, agriculture and ecological health which will be increasingly endangered. Vast parts of the basin could become hydrologically unsustainable by 2100, if climate change coupled with land use transformations run unabated along current emissions pathways. Findings motivate an urgent need for targeted policy interventions to conserve wetland ecosystems, to manage land development, and to provide buffers against further hydrological collapse.

Our projections show a significant decrease in wetland area under both SSPs, a result that reflects how global IAM-based land-use models handle, and sometimes simplify, wetland dynamics. Because these models frequently overlook local hydrological constraints and existing conservation policies, they may overestimate the pace of land conversion in systems like the Okavango. Consequently, the findings are interpreted as an indicator of potential vulnerability under sustained pressure rather than a definitive prediction of future extent. This discrepancy underscores the imperative for future scenario frameworks to incorporate enhanced integration of regional land-use governance. Such integration is essential to ensure that these frameworks accurately reflect the resilience of protected African wetlands.

The uncertainty in the projections is attributable to three primary sources: structural differences in climate models, emission pathway assumptions, and land-use projection frameworks. Inter-model variability is more pronounced in the near-term period (2025–2040), reflecting internal climate variability and differences in regional precipitation sensitivity among GCMs. As projection periods extend toward the end of the century, there is an increase in model convergence, thereby strengthening confidence in the drying trend direction. Furthermore, the NEX-GDDP bias-correction approach reduces systematic errors but may not fully capture localized convective rainfall processes. Land-use projections derived from IAM-based LUH2 datasets are characterized by their coarse spatial scale and reliance on global socioeconomic trajectories. These projections may not fully reflect local governance or conservation policies, potentially leading to misinterpretations of environmental change. Therefore, it is imperative to interpret the results as plausible, scenario-based trajectories rather than as deterministic forecasts.

This study presents a projected assessment of hydro-climatic changes in the lower Okavango Basin near Rundu. Using SWAT+ simulations driven by CMIP6 GCMs and LULC projections, we examined precipitation, streamflow, and water yield responses under SSP2-4.5 and SSP5-8.5 scenarios, relative to a historical baseline dataset spanning 1981–2015. The following key findings are summarized below:

Nonlinear streamflow response to precipitation decline: While projected rainfall decreases are modest (approx. 5–20%), streamflow declines are far more severe (up to 60% by 2100), with several models simulating the end of dry season flow. This confirms the “dry gets drier” paradigm and highlights the vulnerability of semi-arid, downstream-dependent basins like the Okavango.

Accelerated hydrological degradation due to land use change, particularly the conversion of wetlands to cropland and urban land is projected to reach over 65% of the basin area by 2100. This transition reduces infiltration capacity, increases runoff, and eliminates ecosystem buffering functions enhancing the climate impacts on streamflow.

Spatial hydrological inequalities are expected to intensify, with water yield becoming increasingly concentrated along river corridors, while most southern and northern parts of the basin fall to extremely low yield conditions below 200 mm/year by end-century. The decline in spatially distributed water yield across the basin raises equity concerns for water access and ecosystem support.

Model convergence after mid-century especially under SSP5-8.5 enhances confidence in long-term drying trends, particularly under high-emission pathways. The agreement across models strengthens the case for urgent adaptation and mitigation responses focused on land use regulation and water resource planning.

These findings not only confirm regional trends observed in the Zambezi, Limpopo, and Orange catchments (Aich et al., 2014; Goulden et al., 2009), but also emphasize the combined role of climate and human-induced LULC changes in undermining hydrological resilience.

The findings of this investigation have the potential to provide direct insight into the realization of the following Sustainable Development Goals: SDG 6 (Clean Water and Sanitation), SDG 13 (Climate Action), and SDG 15 (Life on Land). In practice, transboundary institutions such as OKACOM could operationalize these insights through a multifaceted approach, including prioritizing wetland conservation areas, incorporating land-use scenario assessments into water allocation planning, and promoting nature-based solutions, such as riparian buffer restoration. At the national level, there is a need to align basin-scale hydrological risk assessments with regulations governing land use, zoning, and related matters. Such an action would reduce future water security risks.

Namibia has laid important policy foundations through the Integrated Water Resources Management Plan (IWRM, 2010) and the Water Resources Management Act (Act No. 11 of 2013). However, basin-scale integration with land use planning and climate change adaptation remains weak, especially for transboundary coordination with Angola and Botswana. To respond to the combined challenge of climate and LULC changes, sustainable strategies are proposed; (i) implementation of nature-based solutions such as wetland protection and riparian buffer restoration in high wateryield areas, particularly upstream and near Rundu, to retain water and enhance recharge. (ii) adopting conservation agriculture by implementing drought resistant farming practices like legume-based cropping systems, which are not widely practiced locally. These approaches allow to reduce evapotranspiration demand, soil retention, groundwater recharge and support food security in drier conditions. (iii) Strengthen the mandate and implementation capacity of OKACOM to embed climate and land-use scenario planning in transboundary agreements and water allocation protocols. (iv) Enhance monitoring and scenario-based planning for transboundary basins, with integrated datasets on land use, groundwater, and evapotranspiration to reduce modeling uncertainty and guide spatial adaptation strategies.

Without these coordinated actions, the lower Okavango is likely to cross irreversible ecological thresholds by the end of the century. Therefore, the study serves as a decision-support tool to inform water resource managers and policy makers, for integrated land-water-climate strategies that are both locally informed and regionally coordinated to prevent a systemic collapse in water security.