Soil and Water Assessment Tool reveals dominant baseflow in semi-arid catchment with sand dams in Kenya

Increasing water scarcity and climate variability threaten rural livelihoods in semi-arid regions. Sand dams are widely used as Nature-based Solutions (NbS) to retain water within the sandy matrix of ephemeral riverbeds, yet their catchment-scale hydrological effects remain insufficiently quantified. We applied the Soil and Water Assessment Tool (SWAT+) in a 22.35 km 2 ungauged ephemeral catchment in Makueni County, Kenya, containing 22 sand dams. The model combined open Earth observation and geospatial datasets to characterize baseline hydrology for 1990–2024. Future hydrological responses were explored under an RCP4.5 climate scenario through 2099, while a third scenario tested the feasibility of representing sand dams as in-channel SWAT+ reservoir objects. The baseline simulation indicates an evapotranspiration-dominated catchment with strong subsurface flow contributions, with modeled baseflow accounting for approximately 70% of total streamflow. Under RCP4.5, increased precipitation is partitioned predominantly into enhanced percolation and subsurface flow, strengthening the modeled baseflow contribution while also increasing localized surface-runoff peaks. Representing sand dams as standard reservoir objects enabled the analysis of spatial and seasonal differences among individual structures and their cascade behavior, but produced highly episodic storage dynamics and substantial quantitative uncertainty. The results demonstrate the potential of open-data-supported hydrological modeling for characterizing ungauged semi-arid catchments and exploring sand-dam networks at the catchment scale. At the same time, the SWAT+ reservoir formulation cannot adequately reproduce key sand-dam processes, particularly porous-media storage and sediment-related subsurface dynamics. The findings should therefore be interpreted primarily in terms of hydrological patterns and model behavior and not as as precise estimates of sand-dam performance, highlighting the need for improved model representation and field-based validation.

Increasing water scarcity is among the most pressing consequences of climate change, particularly in arid and semi-arid regions. In sub-Saharan Africa, where rainfed agriculture sustains rural livelihoods and freshwater access is structurally constrained, the intensification of drought cycles and rainfall variability threatens both food security and ecosystem health (IPCC, 2023). Kenya exemplifies this vulnerability: approximately 82% of its land area falls within arid or semi-arid zones (Mutimba et al., 2010), and its two short rainy seasons, from March to May and from October to December, leave communities exposed to prolonged dry periods with minimal water retention. Projected temperature increases of up to 0.5 °C per decade under high-emission scenarios (Mango et al., 2011), combined with accelerating population growth, land degradation through overgrazing and deforestation, and high levels of soil erosion, further intensify pressure on already stressed water resources (Ng’anga and Smit, 2023; Opiyo, 2025).

Land cover change alters hydrological processes by increasing surface runoff, reducing infiltration, and diminishing groundwater recharge, compounding the challenge of securing sustainable water access for rural communities and ecosystems alike (Baker and Miller, 2013; Mwangi et al., 2017).

Nature-based Solutions (NbS), defined as the sustainable management and use of natural systems to provide ecosystem services, are increasingly recognized as a transformative complement, or alternative, to conventional grey infrastructure in water resource management (IUCN, 2020). Sand dams represent one such solution: small concrete or masonry barriers constructed across sandy ephemeral riverbeds that allow alluvial sediment to accumulate behind the wall, creating a subsurface reservoir within the sand matrix. During the rainy season, precipitation infiltrates the sand body and is retained in the pore spaces, remaining available throughout the dry season while being protected from evaporation (Maddrell and Neal, 2018; Ritchie et al., 2021). Sand dams can thereby raise local groundwater tables, support riparian vegetation, reduce surface runoff peaks, and extend agricultural productivity into the dry season. With lifespans of up to 60–100 years and comparatively low maintenance costs, ranging from approximately 6,000 to 16,000 EUR per structure, sand dams are considered a cost-effective and community-centered NbS for climate adaptation in dryland environments (Maddrell and Neal, 2018; Mulwa et al., 2025; Ritchie et al., 2021). In Makueni County, southern Kenya, over 900 sand dams have been constructed to date, supplying approximately 48% of the population with water and constituting a central pillar of local water governance (Mulwa et al., 2025).

Despite their widespread adoption, the scientific evidence base for sand dam performance remains incomplete. Existing research has focused predominantly on individual structures, examining inflow dynamics, subsurface flow, water abstraction, and specific yield at the dam site. Studies such as Borst and de Haas (2006) and Quinn et al. (2019) reported storage coefficients of typically 10–16% of the total sand volume, while Eisma and Merwade (2020) highlighted the difficulty of field-measuring evapotranspiration and seepage losses. Groundwater modelling has shown that sand dams significantly raise upstream water tables, with evaporation accounting for up to 57% of total water loss (Hoogmoed, 2017), and that inter-dam spacing critically determines cascade performance in multi-dam networks (Quilis et al., 2009). Earth observation-based studies using vegetation as proxy indicators for benefits at the ecosystem level indicate greening trends in the surroundings of sand dams, yet no systematic evidence has been delivered (Eisma and Merwade, 2020; Ryan and Elsner, 2016; Walper et al., 2022). Accordingly, as Ritchie et al. (2021) conclude in a systematic review, fundamental questions regarding catchment-scale water balance, seasonal groundwater recharge dynamics, and the interactions between dam networks, land cover, and hydrological process remain unresolved. Equally important, the role of sand dam cascades in moderating flood peaks and buffering low-flow periods under intensifying climate variability has yet to be quantified. These are precisely the dimensions that policy-relevant NbS evaluation requires (Castelli et al., 2022; IUCN, 2020).

Process-based hydrological modelling offers a pathway to close this gap. The Soil and Water Assessment Tool (SWAT+) is a physically-based, semi-distributed model widely applied to assess water balance components, land use impacts, and climate change scenarios across Africa (Mwangi et al., 2017; Samimi et al., 2020). SWAT+ has been successfully deployed in Kenyan catchments with contrasting land cover and climate regimes: Githui et al. (2009a, 2009b) applied SWAT in the Nzoia basin, Mwangi et al. (2016) assessed land use change effects on streamflow in the Upper Mara, and Le and Pricope (2017) demonstrated its suitability for semi-arid East African conditions using remote sensing inputs. Nevertheless, SWAT applications in Kenya face well-documented challenges: data scarcity limits classical calibration approaches, ungauged basins, of which Makueni County is a prime example, require alternative validation strategies, and the model’s representation of onset-of-rainfall dynamics and rapid surface runoff on degraded soils remains a recognised limitation (Akoko et al., 2021; Mwangi et al., 2017). No published SWAT or SWAT+ study exists for Makueni County to date, leaving a regional gap in the model’s applicability for semi-arid NbS contexts.

The integration of sand dam structures into SWAT+ represents an additional and largely uncharted methodological frontier. Existing work that combines SWAT with sand dam analysis has used the model exclusively to generate upstream inflow time series, while the dam structures themselves were represented externally in groundwater frameworks such as MODFLOW (Choi et al., 2022; Chung et al., 2022a, 2022b). A direct, process-coupled representation of sand dams as reservoir objects within SWAT+, and a systematic assessment of its potential and limitations at the catchment scale, has not been attempted. This constitutes a critical gap in the NbS modelling literature, particularly given the growing policy interest in scalable hydrological frameworks that can support sand dam planning across semi-arid regions (Castelli et al., 2022; Chawanda et al., 2025).

A further methodological challenge concerns data availability. Stream gauging infrastructure in semi-arid Kenya has declined substantially in recent decades (Akoko et al., 2021), rendering classical discharge-based calibration infeasible in many catchments. Remote sensing and global reanalysis products, including satellite-derived evapotranspiration (GLEAM), gridded precipitation (CHIRPS), and land surface temperature retrievals, increasingly provide the observational basis for hydrological model parameterization and validation in ungauged basins (Deus et al., 2013; Le and Pricope, 2017). Leveraging such data-driven approaches for NbS evaluation represents both a practical necessity and a methodological contribution aligned with the growing field of remote sensing-informed catchment hydrology (Weerasinghe et al., 2020).

This study presents a catchment-scale hydrological analysis of a 22.35 km 2 ephemeral catchment in Makueni County containing 22 sand dams. Using SWAT+, calibrated against satellite-derived evapotranspiration (GLEAM) and literature-based hydrological ratios in the absence of observed discharge data, we pursue three objectives: (1) to establish a baseline characterisation of the catchment water balance and dominant hydrological processes under current conditions; (2) to assess projected hydrological changes under the RCP 4.5 scenario through 2099 using statistically downscaled climate projections; and (3) to evaluate the feasibility of representing sand dams as reservoir objects within SWAT+ and to quantify their modelled contribution to the catchment water balance. By integrating NbS evaluation into a process-based, remote sensing-driven hydrological modelling framework applied to an ungauged semi-arid catchment, this study contributes methodological and empirical knowledge to support the design, upscaling, and climate-proofing of sand dam systems across semi-arid East Africa.

Study area

The ephemeral catchment (Area of Interest, AOI) in southern Kenya lies within the Lower Midland Zone 5 (LM5) in Makueni County (CCU, 2023) (Figure 1). As part of Kenya’s Arid and Semi-Arid Lands (ASALs), this area experiences frequent droughts, extreme rainfall events, and dry spells (MoALF, 2016; Mulwa et al., 2025). With a strong dependency on rainfed agriculture and livestock farming, combined with increasing climate change impacts, sand dams are a key local adaptation strategy for water storage, addressing agricultural limitations and thus addressing poverty (Makueni County Government, 2026; MoALF, 2016). The ephemeral river with 22 sand dams drains into the Athi River, the second-longest river in Kenya and one of the three permanent rivers in Makueni (CCU, 2023; Munyao et al., 2017), that lacks in water quantity and quality, contributing to the issue of water scarcity (MoALF, 2016; Mulwa et al., 2025; Munyao et al., 2017; Walper et al., 2022). The sand dams in the study area vary considerably in their geometric characteristics (Supplementary Table 1). Dam lengths range from 13.6 to 82.4 m, with a median of 36.5 m, while throwback lengths range from 52 to 579 m (median: 185.5 m). Average river width upstream of the dams varies between 9.0 and 35.0 m, with a median of 24.5 m.

The AOI experiences a bimodal rainfall regime, with long rains from March to May (MAM) and short rains from October to December (OND) (Figure 2). A hot dry season in January and February is followed by a cooler dry season from June to September, both of which may extend by 1 month (Nyawira, 2022). Makueni’s lower midland zone receives 750–1,000 mm of annual, yet unreliable, precipitation with strong seasonal variability and declining trends (NDMA, 2025). CHIRPS data for the AOI (1990–2024) indicates a mean annual precipitation of 723.67 mm, with November as the wettest and July as the driest month. Temperatures in LM5 range from 12 °C to 28 °C, with recent annual means of 22.7 °C to 24 °C (CCU, 2023). ERA5 and CHIRPS data (1984–2020) show highest temperatures in February and March, lowest in July, and a steady warming trend projected to continue (CCU, 2023; Field et al., 2014; Gebrechorkos et al., 2019). Evapotranspiration rates are significant due to high temperatures. From January to February and June to October, potential evapotranspiration exceeds precipitation, leading to moisture deficits (Wanjogu et al., 2006). Humidity and wind are both subject to seasonal fluctuations, with November to May being the most humid and July to October the windiest (CCU, 2023).

The AOI catchment area ranges from 821 to 1,093 m above sea level (mean: 969.24 m, SD: 72.12 m, TanDEM-X 30 m) with elevation increasing toward the west and terrain sloping eastward toward the Athi River. Sandy soils, typical of ASAL regions, dominate the area (Nyawira, 2022). According to the Harmonized World Soil Database (HWSD v2.0), the Soil and Terrain database for Kenya (KENSOTER v2.0), and earlier mapping (Batjes, 2013; FAO and IIASA, 2023; Muchena and Njoroge, 1978), Rhodic Ferralsols occur in the higher western areas, Haplic Lixisols in the mid-elevations, and Chromic Cambisols in the lower eastern areas (Figure 3). The underlying bedrock primarily consists of acidic metamorphic rocks (gneiss and migmatite) belonging to the Mozambique Belt, transitioning to intermediate igneous rocks in the east near the Athi River outlet, as mapped in KENSOTER v2.0 and by the National Geodata Centre for Kenya. The lower midland zone is characterized by cattle and goat farming and the cultivation of drought-tolerant crops such as maize, millet, sorghum, pigeon peas, beans, and green grams (CCU, 2023; Karanja, 2006; KNBS, 2022).

The Soil and Water Assessment Tool

The Soil and Water Assessment Tool (SWAT) is one of the most applied hydrological models worldwide (Cai et al., 2023; Castelli et al., 2022). As a process-based, semi-distributed hydrological model, it operates on a daily time step and comprises eight major components: hydrology, weather, sedimentation, soil temperature, crop growth, nutrients, pesticides, and agricultural management (Arnold et al., 1998; Rajat, 2021). SWAT+, released in 2017 (Tran et al., 2023), is a revised version of the model (Bieger et al., 2017; Chawanda et al., 2025) that introduces Landscape Units (LSU), new spatial objects (Bieger et al., 2017; Wagner et al., 2022), and enhanced reservoir functions (Wu et al., 2020).

A central concept of SWAT is the spatial discretization of the catchment (referred to as watershed in SWAT). The watershed is first divided into subbasins and further into hydrological response units (HRUs), separately for landscape and floodplain areas, if selected (Arnold et al., 1998; Githui et al., 2009a). HRUs are homogeneous areas with a unique combination of land use, soil type, and slope characteristics (Odira et al., 2010). These HRUs are created through a GIS-based overlay of digital maps (DEM, land use, soil) and the definition of thresholds (Arnold et al., 1998; Hunink et al., 2013).

The typical model application process involves six steps. First, data is prepared, then the model is set up (e.g., in QSWAT+), which includes watershed delineation and defining subbasins. The third step is to define HRUs. Fourth is the configuration in SWAT+ Editor, where climate data is imported, management plans are defined, and detailed parameterization occurs. The fifth step involves sensitivity analysis, calibration, and validation. Sixth, the final simulation runs are conducted and the outputs evaluated (Arnold et al., 1998; Chawanda et al., 2025; Giuliani et al., 2013; Houteta et al., 2022; Mango et al., 2011).

A broad set of data is required for model creation: a Digital Elevation Model (DEM), maps for land use (LULC) and soil types, weather and climate data (on a daily basis: precipitation, minimum and maximum temperatures, solar radiation, wind speed, and relative humidity), with hydrological data recommended for model calibration and validation and further data depending on the scope of the study (Giuliani et al., 2013; Houteta et al., 2022). The analysis is primarily based on the official documents of the Texas Water Resources Institute, ‘SWAT Input/Output File Documentation, Version 2012’ (Arnold et al., 1998) and ‘Soil and Water Assessment Tool: Theoretical Documentation’ (Neitsch et al., 2011), which serve as the fundamental and authoritative sources. For this study, QGIS version 3.40.6, QSWAT+ plugin 3.0.3, SWAT+ Editor 3.0.8, and SWAT+ Toolbox 3.0.6 (all 64-bit) were used.

SWAT+ input data

The modelling framework integrates several categories of spatial and Earth observation data, including topography, soils, land cover, meteorological forcing, evapotranspiration, climate projections, and sand dam location data. To improve transparency, Table 1 summarizes all datasets, their principal variables, spatial and temporal characteristics, and their specific role in the SWAT+ workflow.

The methodological workflow comprised four main stages (Figure 4). First, spatial and climatic input data were prepared and used to delineate the catchment and generate HRUs in QSWAT+. Second, the reference SWAT+ model was parameterized, subjected to sensitivity analysis, and calibrated and validated using literature-based hydrological constraints and GLEAM evapotranspiration. The calibrated reference model was subsequently used to characterize baseline hydrology for 1990–2024. Two modifications of this framework were then evaluated: a climate-change scenario using statistically downscaled RCP4.5 forcing for 2025–2099, and a sand-dam scenario in which the 22 structures were represented as SWAT+ reservoir objects. Model outputs included annual and monthly water-balance components (e.g., evapotranspiration, surface runoff, percolation, lateral flow, groundwater contribution, and water yield), spatially resolved HRU-level outputs, and reservoir-specific inflow, outflow, storage, evaporation, seepage, and sediment variables. These outputs were analysed at annual, seasonal, and spatial scales, with additional reservoir-level analyses conducted for the sand-dam scenario.

Digital elevation model

The DEM is essential for the delineation of the watershed and further creation of HRUs. The location of the stream network, slope gradients, flow accumulation paths, and flow direction within the basin are defined through the topographic input (Odira et al., 2010; Opiyo, 2025). For the AOI, the TanDEM-X 30 m Edited DEM dataset (Gonzalez et al., 2020) with an original spatial resolution of 12 m at the equator (corresponding to 0.4 arcseconds) was selected, due to its consistent altitude representation within a concluded DEM comparison and the absence of abrupt elevation spikes.

Soil map

SWAT+ requires a soil map with corresponding look up table and soil database, containing properties such as soil bulk density (SOL_BD), saturated hydraulic conductivity (SOL_K), available water capacity (SOL_AWC), USDA Hydrologic Soil Groups, and the textural fractions (sand, silt, clay) (Odira et al., 2010; López-Ballesteros et al., 2023). The model uses these data to simulate the complex interactions between water, soil properties, and land use, with infiltration, percolation or runoff generation being influenced by texture, structure, and porosity (Bouslihim et al., 2025). SOTWIS (Batjes, 2013) was selected due to its inclusion of historical soil profiles across Kenya and was complemented with information from the Harmonized World Soil Database (HWSD v2.0, 1 km resolution; FAO and IIASA, 2023) and KENSOTER v2.0 (based on HWSD, exploratory maps and soil profiles from the country, e.g., Kenyan Soil Survey; Batjes, 2013), following established methods to meet SWAT+ input requirements (Cordeiro et al., 2018; USDA, 2021) (Figure 2).

Land use map

Land use and land cover (LULC) directly influences hydrological processes (Githui et al., 2009b; Kibii et al., 2021; Odira et al., 2010). Considering land use information from World Bank and CIAT (World Bank, 2015), KENSOTER dataset, the Regional Centre for Mapping and Resource Development, the Government of Makueni County (2024), the International Soil Reference and Information Centre (ISRIC), and GlobCover, the AOI has a relatively uniform land cover, which was verified with reports from Makueni authorities as well as Landsat and Sentinel-2 imagery. The focus was placed on differentiating between existing tree or shrub groups and only seasonally vegetated land because of its high influence on surface runoff (Githui et al., 2009b).

A dataset from OpenLandMap from the University of Chinese Academy of Sciences, (OLM_glc_fcs30d) was chosen as it best distinguishes between tree groups and crop areas (Liu et al., 2023). Due to the large number of LULC classes (over 30 in the AOI), similar classes were grouped together, and the raster was generalized in QGIS using the ‘Sieve’ tool to avoid excessive number of HRUs consisting of only one or two pixels. The resulting classes were ‘forest_steppe_subtropical_mixed’ (mixed deciduous/ evergreen) and ‘crgr’ (cropland/grassland mosaic). For modelling sand dams in the third scenario, reservoir areas were assigned water-land use classes, unlike in Scenarios 1 and 2 (Figure 2).

Climate data

SWAT requires inputs for daily precipitation (pcp, total accumulation in mm), maximum and minimum temperature (tmp, at 2 m in °C), solar radiation (slr, in MJ/m 2), relative humidity (hmd, as a fraction), and wind speed (wnd, in m/s) (Neitsch et al., 2011). For precipitation, CHIRPS was utilized (1983 onwards, Climate Hazards Group InfraRed Precipitation with Station data) (Funk et al., 2014), and supplemented by ERA5 and ERA5-Land (2001–2024) (Copernicus Climate Change Service, 2022). CHIRTS (1983–2020) (Verdin et al., 2020) and ERA5 (2000–2025) were used for temperature. AgERA5 was used for solar radiation, relative humidity, and wind speed. The suitability of CHIRPS was assessed against regional precipitation and drought patterns (Government of Makueni County, 2023; MoALF, 2016; Mutua et al., 2016). CHIRPS is also the operational dataset used in Makueni’s Drought Early Warning Bulletin (NDMA, 2025). It outperformed NOAA CFSR in Eastern Africa (Le and Pricope, 2017) and showed higher accuracy than satellite-only products, especially in Kenyan catchments (Dinku et al., 2018), where ERA5 also performed well (Wanzala et al., 2022).

Reference scenario model setup

As illustrated by the model setup workflow (Figure 4), the reference scenario is created with first the delineation of the watershed in QGIS with QSWAT, burning in existing stream network with the shapefile study river (channel threshold 0.3 km 2, stream threshold at 3 km 2) and an ‘inlet/outlet’ shapefile for the outlet to Athi River. Floodplain is created with DEM inversion, lowering the slope position threshold to 0.07 and leaving the ridge threshold at the default of 3 km 2. Second, HRUs are created without HRU filtering and with merging short channels (for <5% of subbasin area), resulting in 93 HRUs. In the SWAT+ Editor, the loaded climate data are then assigned to the simulated weather stations.

For the LULC class ‘crgr’, the plant communities ‘mung’ (for green grams) and ‘grsg’ (sorghum) are added, both with the status of not being a landcover. For the tree-shrub-groups with the LULC ‘frst_sust’, the plant community ‘frst_sust’ is already implemented with default values, where the initial biomass is reduced to 3,000. The plant ‘shrb’ was added, with the landcover status set to active (yes). Based on literature, official information from Makueni (Waswa et al., 2025) and testing (soft calibration) (FEWS NET, 2024; KALRO, 2024; Mulwa et al., 2025; NDMA, 2025; Nyawira, 2022), the land use management schedule ‘crgr_rot’ as crop season were set as follows: for sorghum (‘grsg’), from 1 April to 15 July and 1 November to 28 February; for mung beans (‘mung’, representing green grams), from 15 March to 1 July and 15 October to 31 January – each planted at the beginning and harvested and killed at the end of season. The simulation was run from 1 January 1983 to 31 December 2024 with a warm-up period of 7 years and output files were generated for the period from 1 January 1990 to 31 December 2024 (SWAT+ rev.61.0.1).

Since the AOI is ungauged, conventional calibration against streamflow measurements was unfeasible (Rajat, 2021). Instead, a strategy combining soft calibration and remote sensing-based calibration was adopted. For soft calibration, overall model plausibility of the hydrological mass balance is aimed at by adjusting land-use management practices and plant community compositions, with literature reference values for semi-arid regions as calibration benchmarks (see Table 2).

In addition to the hydrological mass balance soft calibration (Chawanda et al., 2025), remote sensing product GLEAM (Global Land Evaporation Amsterdam Model) is used for model calibration against evapotranspiration (Miralles et al., 2025). Measured evapotranspiration provides a suitable calibration target, as it accounts for the largest share of precipitation in semi-arid catchments and constrains the dominant component of the catchment water balance. This approach has been demonstrated in other studies including in Kenya and Africa (Bennour et al., 2022; Ningthoujam and Romeji, 2025; Odusanya et al., 2019; Parajuli et al., 2018; Rocha et al., 2023; Yang et al., 2017; Zhang et al., 2020), while taking methodological considerations and practices into account (Cakir et al., 2019; Jayakrishnan et al., 2005; Moriasi et al., 2015; Samimi et al., 2020). The parameters calibrated against were identified during the sensitivity analysis (LH-OAT and Sobol) in SWAT+ Toolbox. Modelled evapotranspiration was calibrated against GLEAM evapotranspiration for the period 2021–2022 using the DREAM algorithm, while 2023–2024 served as an independent validation period for evapotranspiration as it represents the most recent complete GLEAM observations available for the study. This calibration and validation procedure therefore evaluates the model’s ability to reproduce the dominant evapotranspiration component of the catchment water balance, but do not constitute a direct validation of streamflow, surface runoff, lateral flow, or groundwater discharge, for which no in-situ observations were available. During the calibration period (2021–2022), simulated evapotranspiration achieved a Kling-Gupta Efficiency (KGE) of 0.53 (r = 0.73, α = 0.99, β = 1.38) and a Nash–Sutcliffe Efficiency (NSE) of 0.27, while during validation (2023–2024), KGE was 0.65 (r = 0.73, α = 0.88, β = 1.18) and NSE was 0.46. For the combined 2021–2024 period, KGE was 0.61 (r = 0.73, α = 0.92, β = 1.27) and NSE 0.40 relative to GLEAM (Figure 5).

Climate change scenario

Climate change is expected to substantially alter hydrological processes, including runoff ratios and the frequency of high-flow events (Okello et al., 2020; Sarmiento et al., 2025). In Makueni County, projections indicate rising temperatures, longer and more intense heat periods, and increasingly erratic rainfall patterns (CCU, 2023; Kiprotich et al., 2021). Although total annual precipitation may increase, rainfall distribution is expected to become more irregular. The short rains (OND) are projected to intensify, while the long rains (MAM) are expected to decline (Field et al., 2014; Gebrechorkos et al., 2019; Kiprotich et al., 2021).

To assess climate-change impacts on catchment hydrology, this study employs the statistically downscaled and bias-corrected climate dataset developed by Gebrechorkos et al. (2019), covering 1961–2,100 for East Africa. The dataset was generated using the Statistical DownScaling Model (SDSM), with large-scale predictors from the second-generation Canadian Earth System Model (CanESM2), a CMIP5 global climate model (Chylek et al., 2011). It provides daily station-based projections of minimum and maximum temperature and precipitation for East Africa under different RCP scenarios. This local downscaling approach offers localized daily minimum and maximum temperature and precipitation data for the AOI (Station Box 30 until 8 Dec 2100, LAT –1.88, LON 39.15) for RCP 4.5. The baseline period 1990–2024 results from the temporal configuration of the reference SWAT+ model: Climate inputs were available from 1983 onward, with the first 7 years (1983–1989) used as a model warm-up period, leaving 1990–2024 as a 35-year multi-decadal contemporary reference period for hydrological evaluation. RCP4.5 was selected as an intermediate pathway that avoids both the strong mitigation assumptions associated with RCP2.6 and the very high emissions represented by RCP8.5. This choice is also consistent with more recent assessments of plausible future emission trajectories. While RCP8.5 was still widely used as a high-emission scenario when the underlying climate projections were generated, very high emission pathways such as RCP8.5 and SSP5-8.5 are now considered increasingly implausible in light of observed emission trends, climate policies, and the declining costs of renewable energy (Van Vuuren et al., 2026). Conversely, stronger mitigation pathways would be expected to result in less pronounced climatic changes than those derived from RCP4.5. Accordingly, the results presented here should be interpreted as representing an intermediate climate-change trajectory rather than the full range of possible future conditions.

The model was run for the period 2018–2099, incorporating a seven-year warm-up phase to ensure model stability and alignment with preceding climatic conditions under RCP 4.5, reducing the risk of biased outputs caused by abrupt transitions between reference and projected climate data. During long-term runs, program crashes occurred due to the maximum age limits of the LULC plant communities. To maintain model continuity, both communities were temporarily harvested and replanted: frst_sust in 2066 and shrb in 2077.

Because no observed streamflow data were available for independent hydrological validation, the future simulations should not be interpreted as deterministic predictions of catchment discharge or groundwater fluxes. Yet, they represent scenario-based estimates of how the calibrated model structure responds to an intermediate climate forcing, with particular emphasis on relative changes in water-balance components and seasonal patterns.

Sand dam modeling

As SWAT+ lacks a dedicated sand-dam module and previous studies have modeled them only externally, existing model components must be adapted to simulate their hydrological dynamics. This can be achieved either through modifications of the input datasets to create dedicated HRUs or through the adaptation of existing SWAT+ objects and their parameterization. The first approach modifies HRUs by assigning dedicated land use or soil classes along the channel or floodplain with adjusted properties (e.g., higher sand content, greater hydraulic conductivity, enhanced deep percolation) in the input datasets or in SWAT+ Editor. Despite this flexibility, it cannot reproduce in-channel water storage or sediment retention, allowing water and sediment to pass through the channel unobstructed. The second approach represents sand dams as ponds or reservoirs within the SWAT+ stream network. Ponds are small, off-channel features receiving inflow only from their own subbasin. As isolated pools, they cannot reproduce the flow retention behavior of sand dams. Reservoirs, in contrast, are on-channel structures with an impervious bottom that intercept upstream inflow and simulate sediment trapping and accumulation, more closely resembling sand-dam functioning. Their operation can be refined through decision tables and release rules, although reservoirs generally consist of water and maintain a surface water layer.

Ponds and reservoirs can be incorporated into SWAT+ in three ways (George, 2023). First, the ‘Add Lakes’ option can be used during watershed delineation (QGIS step 1) by importing polygon shapefiles of reservoirs. Second, an ‘Inlet/Outlet’ shapefile containing georeferenced sand dams classified as reservoirs can be imported (QGIS step 1). In this case, the corresponding areas in the land use map (QGIS step 2) must additionally be classified as ‘WATR’ to ensure recognition as water bodies during HRU creation. Third, reservoirs can be identified automatically by lowering the reservoir-threshold parameter during HRU creation so that ‘WATR’ areas exceeding this threshold are classified as reservoirs. Parameters on the dams’ locations, elevations, and reservoir classifications are given in Supplementary Table 1.

The inlet/outlet shapefile combined with the ‘WATR’ land use class proved most reliable for reservoir implementation. Reservoir parameters were subsequently corrected in SWAT+ Editor by importing a modified ‘hydrology.res’ file (Supplementary Table 2). Reservoir area and volume were calculated individually for each sand dam using throwback length, average channel width, derived from Piemontese et al. (2024), an assumed depth of 2.5 m, and a specific yield of 0.13 derived from Ritchie et al. (2021) (principal spillway equals emergency spillway). Seepage was limited (k = 0.01) to represent the nearly impervious dam base, while evaporation was set at 68% (Ritchie et al., 2021), decrease of evaporation in sand dams by 19%, initial reference scenario 84.49%). Sediment parameters were adjusted (se_amt = 0.1; D50 = 300 micrometers; carbon = 0, bd = 1.4 t/m 3). Water release downstream occurred once the reduced storage capacity was exceeded, approximating fully saturated sand storage (Supplementary Table 2).

A new watershed was delineated with the same settings as in the reference scenario. Catchment area was again 22.3479 km 2 and consists of 3 subbasins, 242 HRUs, and 122 LSUs. The modeled reservoir geometry, including length, width, depth, slope, elevation range, and its assignment to the respective subbasin and stream order, is provided in Supplementary Table 3. The model was run twice in SWAT+ Editor, once from 1983 to 2024 (as Scenario 1) and second from 1983 to 2099 with a seven-year warm-up period.

Hydrology of the reference scenario

Catchment analysis

The simulated catchment (Figure 6) for the reference scenario covers a total area of 22.3479 km 2 and is spatially discretized into three subbasins, 48 LSUs, and 93 HRUs. The model structure incorporates 24 channels, 31 point sources (inlets), seven aquifers, and 48 routing units. Topographically, the domain consists of a general landscape area (82.82%) and a floodplain (17.18%). Subbasin 1, which contains the outlet to the Athi River, occupies 60.62% of the catchment with elevations ranging from 821 to 1,075 m. Subbasin 2 covers 23.18% (985–1,093 m), and Subbasin 3 accounts for 16.19% (985–1,089 m). Land use is heavily dominated by cropland-grassland at 84.94%, while mixed forest accounts for the remaining 15.06%. Ferralsols are the most prevalent soil type (57.85%), followed by Lixisols (39.60%) and Cambisols (2.55%). The model’s performance for this reference scenario yielded a KGE of 0.61 and an NSE of 0.4 (period 2021–2024).

Annual water balance

Based on the long-term annual mean from 1990 to 2024, the water input into the system is characterized by a mean annual precipitation of 720.79 mm and an average curve number of 65.61 (Figure 7). Evapotranspiration represents the dominant hydrological output, consuming 84.49% of the total precipitation (608.97 mm). It consists primarily of soil evaporation (55.12%) and plant transpiration (42.44%), with canopy transpiration playing a minor role (2.45%). After evapotranspiration losses, 10.78% of the precipitation percolates into the shallow aquifer. Most of this percolated water becomes return flow or re-evaporates, while a very small portion, equivalent to 0.76% of total precipitation, recharges the deep aquifer. Surface runoff is limited, accounting for only 5.04% of precipitation. Overall, 16.66% of the precipitation generates streamflow. This streamflow is strongly dominated by baseflow, which includes lateral and return flow, constituting 69.74% of the total flow, whereas surface runoff contributes the remaining 30.26%.

Differences in the values or terms for return flow were noted between SWAT+ Editor (Output Check) and SWAT+ Toolbox (Model Check). According to the ‘basin_wb_aa’ output file, the annual mean of flow into the channel is 44.5 mm (similar to GW_Q), and flow is 76.7 mm (corresponds with return flow of SWAT+ Editor). In the ‘basin_aqu_aa’ file, ‘GW_Q’ is 45.16 mm per year. Lateral and return flow are also referred to as baseflow while groundwater flow (GW_Q) is defined as the groundwater contribution to the stream in the watershed for the simulation. Further, the value for recharge to the deep aquifer in SWAT+ Editor and SWAT+ Toolbox was assigned to the parameter ‘seepage’ in the output files. There, the parameter ‘rchg’ (recharge) is 115 mm, and seepage is 5.5 mm (corresponds with recharge to deep aquifer in SWAT+ Editor).

Seasonal catchment fluxes

Mean monthly precipitation is 73.34 mm, peaking in November, followed by December and April, with consistently dry months occurring from June to September (Figure 8). Evapotranspiration follows a similar seasonal pattern, reaching its highest values during the long rain season in April, followed by the short rain season in November and December. Evapotranspiration is lowest during the dry, cool season from June to September. While soil evaporation generally dominates throughout the year, plant transpiration assumes a higher share during the peak rainfall months of April and November. The curve number exhibits a delayed temporal response, increasing 1 month after the start of the rainy seasons – peaking in December and January – and declining 1 month after the onset of the dry season. Surface runoff is only generated during the rain seasons and is significantly higher during the short rains, reaching a maximum in November and a secondary peak in December. Subsurface percolation peaks after the short rains, with a maximum in January, and declines steadily from June to an annual minimum in October.

Spatial distribution

Spatially, evapotranspiration is generally lower in mixed forest areas compared to cropland-grassland, and it is slightly higher in the floodplain than in the surrounding landscape. Specifically, higher overall evapotranspiration and soil evaporation rates are observed in Lixisol regions, while plant transpiration predominates in the western areas characterized by rhodic Ferralsols. Curve number values range from 44.2 to 72.4 across the HRUs, being generally lower in forested areas than in cropland-grassland, and lower in Lixisol regions compared to Ferralsol and Cambisol areas. Consequently, surface runoff is lower in Lixisol-dominated areas than in those with Ferralsols and Cambisols. It is elevated in the floodplain compared to the general landscape and partially higher in mixed forests than in cropland areas. The generated surface runoff predominantly flows into the ephemeral river channels, except in the area near the catchment outlet, which borders the Athi River. Percolation is generally lower in Ferralsol and Cambisol areas than in Lixisols and is partially higher in mixed forests compared to cropland-grassland.

Overall, the catchment is water-stressed and evapotranspiration-dominated with percolation as the main recharge mechanism. The total water reaching the channel, referred to as water yield (sum of lateral flow, surface runoff and groundwater flow) is only 6.02% (43.4 mm per year) of annual precipitation. Monthly dynamics follow a clear seasonal pattern driven by the rainy season, whereas hydrological responses are spatially diverse.

Hydrological changes under RCP 4.5

Annual water balance

Under the RCP 4.5 scenario, the catchment exhibits alterations in its annual water balance (Table 3). Because a 7 year warm-up period was applied, precipitation and temperature inputs integrate more historical data, resulting in slight deviations from the originally downscaled regional data from Gebrechorkos et al. (2019). Annual precipitation increases by 47.87% compared to the baseline (1,065.81 mm), providing more water to the system. Evapotranspiration increases in absolute terms to 828.49 mm but decreases relative to precipitation to 77.73% (down from 84.49% in the reference scenario). The curve number increases slightly by 0.34 to 65.95. Surface runoff now constitutes a smaller fraction of the total flow (11.65%, down from 30.26%). At the same time, percolation increases to 18.54% of precipitation (compared to 10.78% previously). Return flow increases by over 200%; groundwater flow into the channel (GW_Q) increases similarly. Recharge to the deep aquifer increases to 1.5% (compared to 0.76%). Streamflow as a proportion of precipitation increases to 32.01%, up from 16.6% in the reference scenario. Baseflow as a proportion of total flow also increases, now making up 88.34% (before 69.74%). When considering groundwater flow into the channel specifically, this represents 13.98% of precipitation (before 6.27%). Ultimately, the water yield is now 60.7 mm, representing a smaller increase compared to the reference scenario than the increase of precipitation.

Again, values for return flow differed between the SWAT+ Editor Output Check and the SWAT+ Toolbox Model Check, and the presented value for deep recharge equals the parameter ‘seepage’ in the output files. There, annual percolation aligned with the value of the SWAT+ Editor (198 mm), but GW_Q differed, being 148.96 mm. The annual mean of flow into the channel is 149 mm (which corresponds with GW_Q), flow is 281 mm (which corresponds with the return flow of the SWAT+ Editor), recharge is 336 mm, and seepage is 16 mm (which corresponds with recharge to the deep aquifer in the SWAT+ Editor).

Seasonal input patterns

Precipitation and temperature values change, but the historical seasonal pattern remains visible. Modeled precipitation for 2025–2099 under RCP 4.5 falls between the 1990–2024 historical values (CHIRPS/ERA5) and the projected original data (which had an average monthly increase of 207.15%), except for higher values observed in April and May (Figure 9). On average, the monthly mean minimum temperature (T min) increases by 8.84%, with the highest increase in December (+12.60%) and the lowest in February (+6.45%). The monthly mean maximum temperature (T max) increases by 7.67%, showing the strongest increase in August (+22.23%) and the smallest in December (+0.50%). Seasonally, the long rain season (MAM) follows a similar pattern but generally features more rain, higher T max, and slightly higher T min. The following dry cool season (JJAS) is still dry but experiences more general rainfall, along with a significantly higher T min than in the reference scenario and a higher T max. The short rain season (OND) is characterized by higher precipitation, higher T max, and a similar T min. The dry hot season (JF) shows a higher T max, but similar T min and precipitation compared to the baseline.

Seasonal catchment fluxes

Monthly mean evapotranspiration persists with a temporal delay following declines in precipitation (Figure 10). Similarly, percolation also persists with a temporal delay after a decline in precipitation. As an indicator of extreme events, surface runoff increases sharply during periods of high precipitation. Groundwater flow is most pronounced in December, November, and March, whereas lateral flow dominates during the drier months from June to August.

Spatial changes for 2025, 2050, and 2099

Overall, evapotranspiration increases by 2050 compared to 2025 and stays relatively stable through 2099, with decreases observed only in some areas. This pattern is particularly visible in the central-eastern HRUs east of FRr within LXh and CMx. In the northern and central areas of FRr, evapotranspiration ranges between 800 and 900 mm per year in a few HRUs on average in 2025 but rises above 900 mm by 2050 and remains at that level in 2099. In the south-western HRUs, evapotranspiration consistently remains below 450 mm per year. The majority of subbasin 3 (completely within soil class FRr) remains at a lower level of evapotranspiration (<= 450 mm).

Regarding the mean annual curve number, no difference is observed between the years 2025 and 2050. In 2099, however, a slight increase is visible in a few HRUs, particularly near the outlet in the channel area and within the floodplain area of a channel in the central part of the AOI, where values rise from the 57–64 range to 64–71. The curve number (mean 65.95) shows higher values in cropland areas (68.68) compared to the shrubland–forest mix (50.54). Surface runoff generally increases from 2025 to 2050 and continues to rise by 2099, although some areas consistently show low values (below 30 mm per year), particularly in the central-eastern and south-western HRUs. The increase in surface runoff is most pronounced in the north-western HRUs and around the study river, especially near the north-western inlet and the outlet.

Percolation increases from 2025 to 2050 and continues to rise by 2099 across most HRUs. While values in 2025 generally range between 70 and 210 mm, they increase to between 140 and 280 mm by 2099. In the south-western area, the mean annual percolation consistently remains below 70 mm throughout the simulation period. Again, the majority of subbasin 3 maintains its value across all modeled years. Lateral flow increases from 2025 to 2050 and continues to rise by 2099. These changes are mainly observed in the north-western HRUs, near the outlet, and in the central part of the AOI. In the floodplain areas, values are below 18 mm in 2025 and increase to between 18 and 32 mm by 2050 and 2099, with most floodplain HRUs showing changes within this range. In the south-western area, the mean annual lateral flow consistently remains below 18 mm. The annual change in soil water content in 2025 is mostly positive, with a decrease observed near the outlet to the Athi River. By 2050, the outlet area experiences an even stronger decline, while the central part of the AOI shifts from a gain to a loss in soil water. Most of the western AOI still shows an increase in soil water in 2050. In 2099, this western region experiences an annual loss of soil water, with only parts of the floodplain showing a gain. In contrast, the area around the eastern outlet and the central AOI exhibit increases in soil water.

In summary, evapotranspiration remains the dominant water loss but increases at a slightly slower rate than precipitation. The catchment partitions this excess water primarily into enhanced percolation, further reducing the surface runoff fraction. Water yield grows slower than precipitation, reaching 60.7 mm. Hence, only about 5.7% of the total annual rainfall translates into actual water yield arriving in the channel. The system stays subsurface driven, with baseflow ultimately generating 88.34% of the total flow. The historical bimodal climate persists, but the catchment’s response shifts toward more extreme behavior. While baseline surface runoff is decreasing, intense rainfall events trigger runoff spikes.

Sand dam catchment hydrology

Reservoir representation of sand dams

For the period of 1990 to 2024 in the third scenario including sand dams represented by reservoirs, simulated evapotranspiration against GLEAM data achieved a KGE of 0.56 and an NSE of 0.28. Averaged for 1990–2024 for the reservoirs, SWAT+ Output Check reports an average sediment trapping efficiency of −25.26% and as average water loss, total removed and losses amount to −0.25%. The average water loss by evaporation is 47.61% and by seepage 3.88%. The 22 reservoirs were empty 83% of the time. Overall, a consistent downstream pattern is observed along the ephemeral river. Reservoirs at lower elevations (further downstream) show higher flow rates (towards 10 L/s, upstream towards 1 L/s) and slightly shorter filling times (towards 25 days, upstream towards 34 days). In contrast, precipitation inputs and incoming sediment loads tend to be higher at higher-elevation reservoirs. Seepage and evaporation are relatively similar across all reservoirs.

When examining the areas assigned by SWAT+ to the reservoirs, a significant difference compared to the actual sand dam areas becomes apparent: On average, values increased by 84%, with changes in area ranging from −67.8% (sand dam 455) to +708.8% (sand dam 460). The real areas were entered into SWAT+ Editor within the file ‘hydrology.res’ (Supplementary Table 1, 2). However, during the subsequent simulation, the reservoirs were assigned new area values (Supplementary Table 3), which also occurred when trying other methods of adding reservoirs to the model (as lakes, with thresholds, or as ponds). The average of the new measures of the modelled reservoirs is 413.96 m in length, 5.53 m in width, and 0.34 m in depth, with a mean side slope of 1.82 m/m. Assigned sand dam depth increased with decreasing elevation (all reservoirs were uniformly assigned a depth of 2.5 m in the input data). The quantitative validity of the reservoir-related results is therefore discussed later.

Annual water balance

The annual water balance is comparable to that of the scenarios without reservoirs (Table 4). The most notable difference is observed in return flow, groundwater discharge (GW_Q) and percolation. In the sand dam scenario, return flow reaches only 47.4% of the value in the reference scenario without reservoirs (GW_Q: 45.12%) and 63% of the value in the climate scenario without reservoirs. Surface runoff remains at a similar absolute level in scenarios with and without reservoirs. However, because percolation is lower in the reservoir scenario, surface runoff now represents a larger share of total flow. It accounts for 45% of total flow under current conditions and 18% under RCP 4.5, while baseflow contributes 55% instead of nearly 70% in the scenario without reservoirs. Within the climate scenario, baseflow accounts for 82% of total flow with reservoirs compared to 88% without reservoirs. Water yield corresponds to the scenarios without reservoirs, with 5.84% of annual precipitation from 1990 to 2024 and 6.72% under the RCP 4.5 scenario with reservoirs (see Table 4).

Differences in return flow and deep recharge between SWAT+ Editor, SWAT+ Toolbox and the output files occurred as in the former scenarios. Output files for 1990–2024 reported ‘flo’ (return flow) being 36.4 mm, recharge 66.6 mm, seepage 3.17 mm and flow into channel (GW_Q) 20.4 mm. Finally, 16 mm of flow was lost.

Seasonal catchment fluxes

The mean daily inflow and outflow of a reservoir, and sediment input and storage all increase during both rain seasons, with the strongest response occurring during the short rains and extending into January (inflow peaking at over 6,000 m 3 /day in late November). Consequently, sediment movement and storage hit their absolute maximum during the extreme inflow peaks of the short rains (mean storage increasing by over 240% compared to the long rains). The long rains (MAM) also lead to elevated inflow and sediment transport, although less pronounced (sediment inflow is approx. 72% lower than in OND). Overall, during the rainy seasons, inflow exceeds outflow, indicating that the reservoir is recharging (net gain ranging from ~6 m 3 /day in MAM to ~148 m 3 /day in OND).

During the cool dry season (JJAS), inflow, sediment input, and sediment storage decrease continuously and reach their annual minima towards September (inflow drops by almost 90% compared to OND averages). This leads to a period of stagnation, where sediment movement and storage drop to near-zero levels (inputs falling to ~0.3 kg/day). Reservoir outflow generally follows the inflow pattern but is less variable. During both dry seasons, outflow exceeds inflow, meaning the reservoir is depleting (net deficit of ~216 m 3 /day in JF and ~36 m 3 /day in JJAS). Evaporation and seepage as further reservoir water losses are highest during the hot dry season (JF) (evaporation reaching ~114 m 3 /day) and lowest during the cool dry season (dropping by approx. 80% to ~23 m 3 /day). Seepage remains comparatively small throughout the year (peaking at <9 m 3 /day). Overall, OND is the hydrologically most active season, whereas JJAS represents the period of lowest reservoir activity.

SWAT provides monthly inflow dynamics for individual reservoirs. Sand dam 406 (upstream, subbasin 2, located at the highest elevation of 1012.25 m) shows inflow peaks that closely coincide with the rainfall seasons (April and November), whereas sand dam 97 (near catchment outlet, located at the lowest elevation of 828.12 m) exhibits a delayed response, with peak inflows occurring one month later (May and January) (Figure 11). Under the RCP 4.5 scenario, inflows increase during the rain seasons while the overall seasonal pattern remains largely unchanged. Compared to the reference period (2019–2024), sand dam 406 reaches its maximum inflow in December rather than November, while sand dam 97 shows a relatively stronger inflow in March than in May, indicating shifts in the timing of seasonal peak flows. Spatially, mean monthly inflow shows differences across the system, with generally higher values, for example, between sand dams 408 and 456 as well as between 458 and 459, which are further amplified under RCP 4.5 (Figure 12).

Hydrological findings and catchment response

The ephemeral river (AOI) in semi-arid Makueni County with 22 sand dams lies within a 22.35 km 2 water-stressed catchment divided into three subbasins, comprising 93 HRUs. Compared to Walper et al. (2022), the SWAT model delineated additional channels in the west-southern part of the catchment, which are based on the DEM watershed delineation and were aggregated to form Subbasin 3.

Serving as the sole water input to the catchment, the annual mean precipitation of 720.79 mm is substantially exceeded by a potential evapotranspiration of 1,854.3 mm – a high evaporative demand that is consistent with the Kenya Soil Survey advisory site evaluation report (Wanjogu et al., 2006). Evapotranspiration represents the overwhelmingly dominant pathway for water loss within the catchment, consuming 84.49% (608.97 mm) of the total annual precipitation and emphasizing the environmental pressure on local water resources (MoALF, 2016). This high evaporative loss ratio is consistent with the regional hydrological literature for semi-arid East African basins, which regularly reports ET values ranging between 70 and 90% of total rainfall (Deus et al., 2013; Juma et al., 2022; Kiptala, 2016; Mutunga et al., 2022; Nyanchoka Ochengo, 2018; Oduor et al., 2014; Recha et al., 2016). Soil evaporation constitutes the largest share of ET at 55.12% (335.64 mm/year) as in further literature (Jackson and Wallace, 1999; Nyanchoka Ochengo, 2018), with higher rates observed in the central mid-elevation Lixisols, and in the cropland-grassland compared to mixed forest areas (Koohafkan and Stewart, 2008).

While the curve number value is plausible for a semi-arid catchment (65.61) (Nyanchoka Ochengo, 2018; Yoshe, 2025), simulated surface runoff – which is largely controlled by the curve number – remains comparatively low relative to values reported in most comparable studies. Surface runoff accounts for only 5.04% of annual precipitation, whereas previous studies typically report proportions of 10–20% (Juma et al., 2022; Koohafkan and Stewart, 2008; Nyanchoka Ochengo, 2018; Oduor et al., 2014; Recha et al., 2016). This discrepancy is unlikely to be explained by land cover effects (Kibii et al., 2021; Owuor et al., 2016), as mixed forest covers only 15.06% of the catchment area, while the dominant cropland-grassland mosaic provides substantial vegetation cover only during the cropping seasons from 15 March to 15 July and from 15 October to 28 February. Seasonally, surface runoff is highly reactive, being almost absent between July and October but responding rapidly to rainfall during the OND rains, peaking at 16.6 mm in November. During the MAM long rains, surface runoff rarely exceeds 6 mm. This reflects the system’s rapid hydrological response to intense rainfall and the associated potential for flash flooding, particularly as more heavy precipitation events are projected to increase under RCP 4.5. The highest runoff generation occurs on agricultural Ferralsols (seasonal cropland), while the lowest runoff is generated on forested Lixisols, reinforcing the critical role of vegetation in runoff moderation.

A central, finding of this study is the dominance of subsurface flow mechanisms. In contrast to previous Kenyan SWAT applications, which have primarily emphasized increases in surface runoff following land-use changes (Githui et al., 2009b; Opiyo, 2025; Sarmiento et al., 2025), this model identifies baseflow (lateral and return flow) as the dominant contributor, accounting for nearly 70% of total annual flow. Although this proportion is high, evidence from other semi-arid African catchments shows that it is not unprecedented. Reported baseflow contributions range from approximately 33–62% in several Ethiopian catchments (Bayou et al., 2021; Molla and Tegaye, 2019), while values of 40–70% and, depending on the separation method, approximately 69–75% have been observed in semi-arid South African catchments (Saraiva Okello et al., 2018; Welderufael and Woyessa, 2009). The simulated value for the present catchment therefore lies near the upper end of reported ranges but remains hydrologically plausible. Comparisons should nevertheless be treated cautiously because published estimates differ in temporal scale, hydrological setting, and separation method.

The dominance of subsurface flow is also consistent with the simulated recharge dynamics. Percolation, equivalent to 10.78% of annual precipitation, shows a pronounced seasonal lag and peaks in January (9.26 mm), well after the OND short rains. This delayed response suggests that substantial wetting of the upper soil profile is required before deeper infiltration occurs, thereby sustaining shallow groundwater storage and baseflow into the early hot-dry season. Spatially, recharge is highest in the central and eastern parts of the catchment, where Lixisols and Cambisols dominate, suggesting a greater contribution of these soils to groundwater recharge than the Ferralsols in the west. Percolation is also higher under vegetated mixed forest than under cropland. These patterns are consistent with evidence from East African catchments showing that high subsurface contributions can be promoted by vegetation cover, geology, climatic water availability, and catchment morphology (Abebe and Foerch, 2006; Molla and Tegaye, 2019).

The integration of statistically downscaled climate data until 2099 (Gebrechorkos et al., 2019) under the RCP 4.5 scenario projects an overall warming trend for the catchment, particularly pronounced during the cool dry season (JJAS), with August maximum temperatures increasing by 22.23%. The magnitude of these projected changes is sensitive to the selected emissions pathway. A stronger mitigation scenario would likely reduce the projected warming and associated hydroclimatic changes, whereas a higher-forcing pathway would be expected to amplify them, although the direction and magnitude of precipitation responses may be less consistent across scenarios. Annually, the data indicates a steeper rise in monthly mean minimum (+8.84%) compared to maximum temperatures (+7.67%), aligning with CORDEX simulations by Anne et al. (2017), who similarly found that minimum temperatures increase more than maxima in the Upper Ewaso Ngiro catchment. Furthermore, these trends align with regional models projecting general temperature rises (Kiprotich et al., 2021) and local assessments anticipating up to a 75% increase in ‘hot nights’ (CCU, 2023). Regarding precipitation, the dataset projects a 47.87% annual increase, exceeding the 0.1–23.2% precipitation increases modelled for other Kenyan catchments (Githui et al., 2009a; Anne et al., 2017) and aligning more closely with local RCP 8.5 projections (45.78% increase by 2050, CCU, 2023). This coincides with heightened seasonal variability and intensified short rains (Kiprotich et al., 2021; CCU, 2023); yet unexpectedly, modelled long rains (MAM) also increased.

While absolute evapotranspiration increases due to higher temperatures, its relative proportion of precipitation declines from 84.49 to 77.73%. This dynamic – where long-term precipitation increases outpace evapotranspiration losses, resulting in higher overall water availability – aligns with the trends observed by Sarmiento et al. (2025) using SSP scenarios. The catchment channels this surplus rainfall predominantly into subsurface processes, elevating percolation by 154.32%. Because re-evaporation from shallow aquifers only increases by 20.51%, most of this water feeds return flow and groundwater flow into the channel, which both increase by over 200% compared to the baseline. Consequently, the system becomes even more subsurface driven, with baseflow projected to constitute 88.34% of the total flow. Unlike Okello et al. (2020), who predicted higher runoff-to-precipitation ratios, and Sarmiento et al. (2025), who projected a broader conversion of future rainfall into surface runoff during the JJAS season, increased precipitation in our model does not translate into surface runoff. Yet, the strong subsurface buffering we found aligns with Anne et al. (2017), who reported a seasonal lag where precipitation changes only reflect in the streamflow of the subsequent season. Collectively, the diverse dynamics within the future water balance of the catchment validate the non-linear hydrological responses described by Mango et al. (2011).

Simultaneously, the projection identifies localized risks. Although the catchment overall shows lower values of surface runoff, specific HRUs in the western headwaters and the eastern outlet are projected to experience surface runoff exceeding 300 mm per year. This signifies a substantially increased localized risk for soil erosion, which corroborates the broader warnings by Sarmiento et al. (2025) regarding increased high flows and future flood risks. Particularly noteworthy is the low water yield: in the reference scenario, only 6.02% (43.4 mm/year) of precipitation remains annually (5.84% with reservoirs), and in the climate scenario, it is only about 5.7% (6.72% with reservoirs), with higher values occurring during the OND rains.

Sand dam modeling in semi-arid Kenya

This study provides a methodological contribution by integrating open-source Earth observation data (including the TanDEM-X 30 m DEM, KENSOTER and SOTWIS soil datasets, OpenLandMap land use data, GLEAM satellite-derived evapotranspiration, climate data from ERA5, CHIRPS, CHIRTS) with process-based hydrological modeling (SWAT+) to evaluate Nature-based Solutions (NbS) at the catchment scale. This approach establishes a framework for spatiotemporal hydrological assessments in a completely ungauged, semi-arid basin (Chawanda et al., 2020; Odusanya et al., 2019; Parajuli et al., 2018). Such a framework provides essential baselines for NbS planning and resilient water management in data-scarce Sub-Saharan regions, particularly in the face of climate change.

Prior literature addressing sand dams via SWAT differs fundamentally in its methodology. Studies by Chung et al. (2018) explicitly utilized the SWAT-K model as a precursor tool to generate upstream watershed runoff and inflow time-series data for a specific valley. In those subsequent analyses, the physical sand dam structure, the alluvial sediment matrix, and the specific yield storage processes were not modeled inside SWAT itself; instead, the SWAT outputs were manually routed into specialized, external 3D groundwater models like MODFLOW to simulate the dam’s behavior (Chung et al., 2022a, 2022b). This present study attempted to couple 22 sand dams directly within the SWAT+ environment utilizing the standard ‘reservoir’ object. This approach aims to achieve a representation of dam networks interacting with land use, soil properties and varying climate within a single software environment. While this integration proved challenging and resulted in quantitatively uncertain outputs, the resulting qualitative trends provide valuable insights into seasonal patterns at sand dams. Further, parameterizing a network of 22 interacting dams would allow for the observation of catchment-wide cascading effects, building on studies such as Quilis et al. (2009) who found that inter-dam distance was a critical parameter, and overlapping areas lead to decreasing stored water volume. SWAT+ spatial outputs can show distinct, quantifiable variations in daily inflow and sediment trapping efficiencies for each individual sand dam, which here were visible between higher-elevation upstream and downstream dams. It must be noted that the absolute quantitative values in this scenario – such as total incoming flow and sediment volumes – are highly uncertain. Within the SWAT+ setup, the reservoirs were assigned disproportionately large surface areas, and seepage rates are configured manually by setting hydraulic conductivity. Accordingly, the simulated hydrological response should be interpreted in relation to the modelled reservoir geometry and its position within the drainage network (Supplementary Table 3). While the exact parameter outputs varied noticeably between model runs, the qualitative trends – specifically the seasonal inflow patterns, the dams’ highly episodic nature, and their vulnerability to evaporation – remains physically plausible.

The model indicates that the sand dams act as local evaporation hotspots, increasing the catchment’s overall evapotranspiration-to-precipitation ratio from 84.5 to 89%. Of all the water lost from the reservoirs, 47.6% evaporates directly into the atmosphere, whereas only 3.8% is lost through groundwater seepage. This aligns with field-based groundwater models from the neighboring Kitui District, which identified evaporation as the largest loss component, accounting for 57% of total water depletion from sand dams (Hoogmoed, 2017; Lasage et al., 2008). As maximum reservoir inflows coincide with the hot-dry season in January, the largest stored water volumes are exposed to the year’s maximum evaporative demand. Assigned with a low hydraulic conductivity (k = 0.01) to represent partial sealing, the large reservoir areas appear to hinder natural riverbed percolation, as percolation and baseflow contribution to total flow decrease with the implementation of the dams. Consequently, while these dams retain water locally, they likely limit downstream flow. Simulated water availability at dam sites is highly episodic. The reservoirs remain entirely empty for 83% of the simulation period, acting primarily as short-term storage rather than permanent water bodies. Inflow and sediment-trapping occur mostly as fast responses during the November and December short rains, which supports recent surveys, that found that most existing dams fail to hold water year-round due to seasonal flow limitations and rapid depletion (Ngugi et al., 2020).

Methodological constraints

Data scarcity and model operation

The main constraint of this hydrological analysis is the availability and quality of local data (Jayakrishnan et al., 2005; Mango et al., 2011; Samimi et al., 2020). The number of gauging stations in Africa has declined drastically (Akoko et al., 2021), and many catchments are poorly gauged, making calibration difficult (Le and Pricope, 2017; Mwangi et al., 2017). No in-situ precipitation nor streamflow records could be obtained for the catchment of this study, which is in line with the broader data scarcity in semi-arid Kenya and led to calibration and validation constraints (Chawanda et al., 2025; Mwangi et al., 2017). Consequently, standard discharge-based model calibration was rendered impossible, and the model’s performance relied on parameter assumptions derived from literature rather than verification against measured local responses (Arnold et al., 1998). While SWAT can provide satisfactory results in general (Mwangi et al., 2017) through the combination of multiple data sources (Samimi et al., 2020) – such as utilizing remote sensing and global datasets as alternatives to local measurements (Dessu and Melesse, 2012; Deus et al., 2013; Kiprotich et al., 2021; Le and Pricope, 2017; Nugroho Christanto et al., 2020) – this workaround introduces uncertainty. In addition to that, the absence of discharge-based validation is particularly relevant for the climate-change experiment, because uncertainties in the partitioning between surface runoff, lateral flow, and groundwater contributions propagate into the projected future response. Consequently, absolute values of future streamflow and recharge should be interpreted with caution, whereas relative changes and broad seasonal tendencies provide the more robust level of interpretation supported by the present model setup.

With an NSE ranging from 0.27 to 0.46 (where a value of 1.0 represents a perfect match between simulated and observed data), model performance only reaches a marginally satisfactory level during the validation period (Moriasi et al., 2015). Yet, because errors are squared, the NSE is highly sensitive to deviations in peak flows. In contrast, the KGE (where 1.0 represents complete agreement) simultaneously assesses correlation, bias, and variability and therefore allows for a more balanced evaluation of how well the model reproduces both the temporal dynamics and the overall magnitude of observed flows, making it better suited for ungauged catchments (Chawanda et al., 2025; Cinkus et al., 2023; Gupta et al., 2009). For the combined calibration and validation period (2021–2024), the model correlation (r = 0.73) indicates a relatively strong linear relationship, suggesting that the seasonal and event-based dynamics of ET are well captured. The variability component (α = 0.92), expressed as the ratio of standard deviations, indicates that extremes are represented realistically, although the simulated values are slightly more dampened than those in the GLEAM dataset. Finally, the bias component (β = 1.27) indicates that the model simulated higher evapotranspiration over the 4-year period than estimated by GLEAM. The model KGE of 0.61 can therefore be considered adequate for this application. The model was calibrated using plausible data from literature alongside a calibration of actual evapotranspiration against satellite-derived evapotranspiration (GLEAM). While this is a recognized technique in data-poor regions (Mango et al., 2011), optimizing a complex hydrological model solely for evapotranspiration introduces a high risk of achieving the right results (observed evapotranspiration) for the wrong underlying physical reasons (Samimi et al., 2020; Sohoulande, 2025). A satisfactory evapotranspiration agreement (KGE of 0.61) does not guarantee an accurate representation of other crucial hydrological pathways, notably surface runoff and deep percolation. Direct comparison with published baseflow indices is additionally constrained by differences in terminology and methodology. In this study, baseflow was calculated from the SWAT+ components lateral flow and return flow, whereas many observational studies derive baseflow from digital filters, hydrograph separation, or chemical and isotopic tracers. These approaches do not necessarily partition streamflow into equivalent components, particularly in ephemeral catchments where event water, shallow subsurface flow, and groundwater discharge may overlap. The simulated value of 69.74% should therefore be interpreted as the modelled contribution of slow and subsurface flow pathways and not as a directly observed groundwater baseflow index.

Moreover, the remote sensing-derived input datasets introduce inherent uncertainties. While the chosen TanDEM-X 30 m in its edited version (Gonzalez et al., 2020) reduces anomalies over water bodies and fills residual gaps, primary uncertainties – such as X-band radar penetration limits in dense vegetation and layover effects – remain. In such small catchment, even minor DEM artefacts can disproportionately affect flow routing and the simulated performance of individual sand dams. In addition, the spatial precision of the model is constrained by the exceedingly coarse resolution of the chosen soil maps. Their polygon-based structure fails to capture the complex spatial heterogeneity of soil properties, which can significantly influence hydrological processes and model outputs, including the partitioning of water between percolation and surface runoff. Although the selected soil datasets incorporate historical soil surveys (e.g., Kenyan Soil Survey; Kinyanjui, 1996) and are broadly consistent with more recent records from the Africa Soil Profiles Database, only three profiles are located in the vicinity of the study area, confirming Rhodic Ferralsols in the west and Lixisols in the central part of the catchment. Locally collected soil data would likely result in a more accurate representation of site-specific conditions (Krpec et al., 2020) and may provide the greatest potential for improving the representation of hydrological processes within the catchment. For the land-use data, classification errors between natural vegetation and agriculture, together with the limited representation of fragmented small-scale farming systems, may affect evapotranspiration estimates. However, although evapotranspiration represents the largest water loss component in the catchment, simulated values are consistent with those reported for other semi-arid catchments in the region, suggesting that local land-use data would likely lead to only limited changes in model results. Given spatial resolutions of 0.05° to 0.1°, the catchment falls within a single pixel across all climate inputs, with CHIRPS and CHIRTS adding further uncertainty due to their reliance on infrared satellite proxies calibrated against a sparse network of roughly 500 stations over Africa (Funk et al., 2014). Without observed streamflow data to calibrate the model, the uncertainties of these remote sensing products cannot be reduced. The results provide a theoretical baseline using the best available data, but the quantitative values should be interpreted with caution.

Further uncertainties arise from operator influence and parameterization, which depend on the modeler’s understanding of SWAT, the hydrological processes, and the specific local context (Mango et al., 2011; Mwangi et al., 2017). The absence of field experience and in situ knowledge presented a significant barrier, as effective model calibration often relies on expert judgment and a practical understanding of regional hydrology (Samimi et al., 2020). This became evident when the automatic calibration algorithm (DREAM) produced mathematically optimized but hydrologically implausible parameter sets, necessitating manual iterative adjustments that inherently introduce subjective bias. Inherent SWAT Architectural Limits.

Another methodological limitation lies in the model selection itself. SWAT was historically optimized for temperate climates and perennial river systems. Its application to highly variable, semi-arid regions is a recognized conceptual challenge, especially in catchments with large terrain variations or non-uniform rainfall (Cai et al., 2023). As with all models, specific hydrological processes are represented through simplified mechanisms. Particularly in SWAT, groundwater simulation lacks the specialized complexity required for arid and semi-arid applications (Samimi et al., 2020). Consequently, the explicit spatial modeling of lateral groundwater movements into riverbanks, as well as a detailed spatial analysis of groundwater recharge based on underlying bedrock, remain constrained. SWAT also has difficulties representing specific land uses, repeatedly misinterpreting ponding water in paddy fields as immediate surface runoff (Kakarndee and Kositsakulchai, 2020). Therefore, SWAT is frequently combined with other tools tailored to specific research objectives (Aloui et al., 2023; Chung et al., 2022a), which was not done in this study. A significant limitation identified in related research is the model’s inability to simulate the dynamic co-evolution of land use and land cover with climate; it treats them as separate drivers rather than interconnected systems (Githui et al., 2009b).

The assessment of future hydrological impacts under RCP 4.5 is constrained by its reliance on a single climate model and one emission pathway. This singular approach does not encompass the broad spectrum of uncertainty in climate projections, bypassing the standard ensemble modelling typically used to capture a more robust range of potential futures. Global Climate Model outputs contain biases, and their coarse resolution introduces additional uncertainty during the downscaling process, particularly regarding variable precipitation patterns (Musau et al., 2015). Moreover, the significant precipitation increase (+47.87%) in the downscaled (Gebrechorkos et al., 2019) does not perfectly align with other regional projections (e.g., CCU, 2023). These regional projections suggest a decrease in the long rains (MAM) and link the modeled large rainfall increases in OND to the more extreme RCP 8.5 scenario.

Regarding the representation of sand dams, the most significant limitation lies in the absence of a dedicated SWAT module and the subsequent workaround using the ‘reservoir’ object. Although well-tested, the reservoir object was originally designed to simulate the thermodynamic and physical properties of open, free-water surfaces (N’guessan et al., 2024; Sohoulande, 2025). While a physical sand dam stores water within a porous sand matrix, the reservoir models evaporation and seepage based on the physics of an open lake. Adapted reservoir parameterization cannot fully capture the distinct physical processes of the actual sand dam structures, such as sediment retention or storage dynamics. This conceptual flaw was exacerbated by a technical error: manually entered reservoir surface areas, required as inputs (Sohoulande, 2025), were overwritten by the software during simulation. This led to area discrepancies of up to 84% and compromised the quantitative outputs. The modelled reservoir dimensions (length 413.96 m, width 5.53 m, depth 0.34 m) differ markedly from the original sand dam geometries. Furthermore, the simulation operated under the idealized assumption that all implemented sand dams were functional. This contrasts with field evidence from the region, which suggests that a high percentage of existing dams may be non-functional (Ngugi et al., 2020). Additionally, the locations of and information on the sand dams rely on the Global Sand Dam Database (Piemontese et al., 2024) and remote sensing verification and does not guarantee physical dam dimensions with absolute certainty.

Finally, the intensive demands of data preparation and the time-consuming nature of calibration are recognized practical limitations of the SWAT methodology. The inconsistency between terms and values among the SWAT+ Editor, the SWAT+ Toolbox, and the simulation output files is a notable limitation. This discrepancy, likely related to inconsistent terminology, may have led to errors in the interpretation and analysis of subprocesses, especially baseflow.

Given these methodological constraints – particularly the coarse input data, the lack of streamflow calibration, and the conceptual errors in sand dam representation – the quantitative results of Scenarios 2 and 3 must be interpreted as qualitative tendencies rather than absolute values. The manual calibration process may have introduced compensation effects, and the findings remain context-specific to the unique topography and soils of the study area. Still, the qualitative results provide valuable indications of seasonal patterns and may be transferable to other semi-arid catchments in the lower midlands or in Makueni County. The identification of peak periods at the sand dams and the timing of maxima and minima can provide essential context for local water management and upcoming field studies.

Implications for practice and research

Strategies for nature-based solution upscaling

Sand dams across sub-Saharan Africa are increasingly recognized as a Nature-based Solution (NbS) to protect rural communities from rainfall variability. Placing this infrastructure relies on invaluable local knowledge, often ensured by local authorities and organizations such as the Makueni Sand Dam Authority or the Africa Sand Dam Foundation. Hydrological modeling can serve as a complementary tool to further enhance this established process. By providing a spatially explicit understanding of the catchment, models can help move planning beyond topological suitability to include quantitative values and seasonal flow patterns, thereby supporting robust, adaptive water management frameworks.

First, our model results demonstrated that flow volumes and temporal peaks change across the sand dams along the ephemeral channel. For instance, the model captured a one-month delay in peak inflows from upstream to downstream dams and differences in flow rates, filling times, or incoming flow and sediment. Upscaling programs can utilize these specific temporal and volumetric insights to optimize the physical distance between structures, particularly as sand dams are often constructed in a series or cascades (Quilis et al., 2009; Ritchie et al., 2021). Second, our model attributes roughly 70% of the catchment’s current streamflow to delayed baseflow, while simultaneously showing that mixed forest areas generate lower surface runoff and enhanced deep percolation compared to seasonal cropland-grassland mosaics. To sustain the critical baseflow identified by the model, dam construction could be combined with catchment-wide measures – such as agroforestry, and vegetation restoration – to enhance slow-response percolation pathways. While HRU definitions and model calculations explicitly incorporate slope, our analysis did not specifically map out its spatial distribution across the catchment. Third, the integration of climate change projections (RCP 4.5) transforms the model into a proactive planning tool. The model identifies specific HRUs – particularly in the western headwaters and near the Athi River outlet – that are projected to experience surface runoff increases exceeding 300 mm/year. Local authorities can utilize spatial outputs to pre-emptively target vulnerable zones for land-use management, erosion mitigation, and the construction of protective check-dams before the projected climate impacts manifest and degrade the landscape.

With respect to policy implications, given the exploratory nature of the modelling results, the findings should not be interpreted as prescriptive recommendations for individual sand dam sites. They nevertheless indicate several principles that may support climate-resilient planning in semi-arid catchments. First, site selection should consider not only local channel geometry but also the wider hydrological setting, including recharge potential, soil properties, land cover, and the position of a site within the drainage network (Maddrell and Neal, 2018). Second, implementation should be accompanied by long-term monitoring of reservoir sedimentation, groundwater levels, downstream flow, and seasonal water availability to evaluate whether expected benefits persist under changing climatic conditions. Such monitoring would also provide the empirical basis required to refine model-based assessments and adapt management over time.

Sand dams should furthermore be regarded as one component of integrated catchment management rather than as stand-alone infrastructure (Ebrahim et al., 2024, 2026). Their implementation can be combined with soil and water conservation measures that reduce rapid surface runoff, erosion, and sediment loss while enhancing infiltration and soil-water retention. Depending on local conditions, these may include vegetation restoration, riparian protection, contour-based measures, infiltration structures, or improved agricultural soil management. Such combinations could strengthen the capacity of the wider catchment to retain water and may complement the local storage and recharge functions of sand dams. The specific effectiveness of these measures, however, requires site-specific assessment and should be evaluated through monitoring and further field-based research.

Dedicated NbS modules in hydrological modeling

A key modelling challenge is the conceptual limitation of representing sand dams using standard open-water reservoir objects. The coupled approach with utilizing SWAT+ outputs to dynamically feed them into dedicated, high-resolution 3D groundwater models like MODFLOW or specialized MATLAB scripts is currently one of the few ways to simulate the porous-media physics, specific yield, and subsurface sediment retention dynamics unique to sand dams (Ritchie et al., 2021; Chung et al., 2022a, 2022b). Future research within the SWAT+ community must prioritize the development of a dedicated ‘Sand Dam’ or ‘Alluvial Storage’ module. This updated tool should mathematically account for specific water yield within porous media, delayed subsurface water abstraction, the dynamic lateral groundwater recharge into adjacent riverbanks, and the protective effect of the sand layer against direct atmospheric evaporation. Crucially, it must also simulate sediment trapping efficiency accurately, distinguishing between the accumulation of useful coarse sands and detrimental fine silts.

To overcome the limitations of ungauged catchments, future studies could make use of the next generation remote sensing. The global soil and elevation datasets used in this study, while standard, are often too coarse to capture the complexities of a small 22.35 km 2 catchment. Future research could integrate high-resolution topography, drones, multispectral satellite imagery or data from newly released or future missions to refine HRU parameterization and include dynamic land-use changes.

Empirical validation and socio-economic implications

Theoretical model assumptions must be grounded in physical reality. The catchment in this study was calibrated and validated using remote sensing-derived climate inputs and further operated under the assumption that all 22 dams were functional as intended. However, this does not account for siltation or maintenance failures in the field. Additionally, information from the Global Sand Dam Database (Piemontese et al., 2024) 4 can be enriched by field-verified data, which should be flagged as such within the database. Future research must prioritize empirical, field-based validation where possible to understand and verify physical landscape dynamics. In addition, laboratory activities could provide further empirical data on sediment accumulation rates and infiltration mechanics. Calibrating hydrological models directly against measured evidence is a necessary next step for testing their accuracy and improving predictive capabilities for future climate scenarios.

Once models are robustly validated through such physical evidence, their outputs must be translated into tangible human benefits. Future research must therefore bridge the divide between hydrology and socio-economic impacts, integrating these domains at both the analytical and operational levels. Studies that couple process-based models like SWAT+ with socio-economic assessments are necessary to quantify the broader societal benefits of NbS. Measuring how the augmented water yield from a sand dam network translates into stabilized agricultural production, improved public health, and enhanced socio-economic resilience will provide the comprehensive evidence base required by policymakers. By translating hydrological data into socio-economic impacts, the scientific community can provide decision-makers with the comprehensive, compelling, and evidence-based arguments needed to support the Sustainable Development Goals and build resilient water systems as adaption measures against the growing impacts of climate change. While conducting such comprehensive, multi-disciplinary analyses must always be pragmatically weighed against the required time and resource investments, this combination of ecological data and quantified Socio-Economic Impact (SEI) is becoming increasingly critical for actual project funding and implementation. Particularly within European funding programs and reporting frameworks, demonstrating these combined benefits through specific SEI Key Performance Indicators is often a fundamental prerequisite to secure financing and political backing for adaptation projects.