Climate variability and land use land cover (LULC) changes are critical drivers of regional water balance, ultimately determining water quantity within river systems. This study sought to determine the impacts of rainfall variability, temperature variability, and LULC changes on the volume of water flowing through rivers draining into Lake Bogoria in Baringo County, Kenya, from 1981 to 2020. We established five Soil and Water Assessment Tool (SWAT) models for the time periods 1981, 1991, 2001, 2011, and 2020, inputting topographic, climate, LULC, and soil data. Primary data was also collected from the Endorois community and key informants from relevant ministries. Our analysis revealed that rainfall increased from 1290.3 mm in 1981 to 1853.6 mm in 2020. The mean temperature exhibited a slight warming trend, rising from 23.77°C in 1981 to 23.85°C in 2020. Land use land cover analysis depicted a net increase in surface area under water by 7.17 km², grasslands by 1.22 km², shrublands by 7.92 km², agricultural land (rain-fed and irrigated) by 8.06 km², Prosopis species by 0.74 km², peri-urban areas by 0.28 km², and bare areas by 0.62 km² over the forty years. Conversely, tree cover decreased substantially by 25.83 km². The model performance was robust, achieving a coefficient of determination (R²) of 0.78 and a Nash-Sutcliffe Efficiency (NSE) of 0.75, indicating good and very good performance, respectively. The total simulated discharge volumes of the Waseges, Kipsirian, and Ngiriki rivers increased from 8,230,008 m³ in 1981 to 17,629,946 m³ in 2020. We conclude that the combined effects of increasing rainfall and extensive deforestation overshadowed the impact of rising temperatures and the expansion of agricultural land, resulting in a net increase in river discharge volumes at the watershed scale. However, this masks a critical shift towards a more volatile hydrological regime with potential for more severe dry-season water scarcity.
Research Article
Hydro-Climatic Shifts and Catchment Transformation: Modelling Streamflow Responses in Lake Bogoria Basin, Kenya
https://doi.org/10.21203/rs.3.rs-7675187/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
Rainfall Variability
Temperature Variability
Land Use Land Cover
Rivers' Discharge
Lake Bogoria Watershed
SWAT
Water scarcity is a defining challenge in arid and semi-arid lands (ASALs), threatening agro-pastoral livelihoods and ecosystem integrity. Globally, agriculture accounts for approximately 70% of freshwater withdrawals, a pressure exacerbated by expanding croplands needed to support growing populations (IPCC, 2020). In these vulnerable regions, water resource availability is predominantly governed by two interacting factors: climate variability and change, and land use/land cover (LULC) change induced by anthropogenic activities (Tirupathi & Shashidhar, 2020; An et al., 2021). Climate change, manifested through rising temperatures and altered rainfall patterns, is projected to increase drought severity and water stress for millions in ASALs, particularly in Africa and Asia (IPCC, 2020). Concurrently, LULC changes such as deforestation, agricultural expansion, and urbanization alter fundamental hydrological processes – infiltration, evaporation, and surface runoff – thereby disrupting regional water balances and impacting both surface and groundwater resources (Khandu et al., 2016; Rodell et al., 2018).
Kenya, a water-scarce nation, faces acute challenges in its ASALs, where poor water infrastructure and management compound the inherent uneven spatio-temporal distribution of resources (GOK, 2011). Nowhere are these hydrological dynamics more visibly expressed than in the closed-basin lakes of the East African Rift Valley. Lakes such as Naivasha, Baringo, and Bogoria have experienced significant, and often dramatic, fluctuations in water levels over historical and recent timescales (Onywere et al., 2013; Herrnegger et al., 2021). Lake Bogoria, a saline-alkaline lake and a Ramsar site, has been part of this phenomenon, recording unprecedented water level rises between 2011 and 2014 that flooded adjacent ecosystems and infrastructure (Schagerl, 2016). The drivers of these changes remain poorly constrained, attributed variously to multi-decadal climatic cycles, tectonic activity, catchment degradation, or a combination thereof (Onywere et al., 2013; LBNR, 2020). A critical limitation to resolving this uncertainty is the scarcity of long-term, systematic hydroclimatic records for these basins (Herrnegger et al., 2021).
Understanding the water inputs to these lakes is fundamental. For Lake Bogoria, the primary freshwater sources are the Waseges, Kipsirian, and Emsos/Ngiriki rivers, alongside direct precipitation and significant geothermal inflows. While the lake itself is saline and unsuitable for consumption or irrigation, the freshwater from these rivers is vital for the dryland farming and livestock watering upon which the local Endorois community depends. Therefore, quantifying the discharge of these feeder rivers, and diagnosing the factors controlling it, is not only a hydrological imperative but a socio-economic one.
Previous studies have noted the potential role of both climate and LULC change in the region but have often been limited by short records or have not employed process-based modeling to disentangle their relative impacts. This study aimed to address this gap by leveraging a 40-year historical dataset (1981–2020) and applying the Soil and Water Assessment Tool (SWAT), a semi-distributed hydrological model, to the Lake Bogoria watershed. The findings of this study will provide critical insights for developing sustainable water resource management strategies in this vulnerable basin and contribute to the broader understanding of hydrological changes in rift valley lake systems.
2.1. Study Area
The study was conducted in the Lake Bogoria watershed, situated within the Marigat and Mogotio Sub-counties of Baringo County in north-western Kenya (Fig. 1). The watershed lies within the axial depression of the Gregory Rift Valley, forming an asymmetric half-graben (Renaut & Owen, 2023). The central feature is Lake Bogoria itself, a narrow, saline-alkaline lake located at approximately 36°06′ E and 0º15′ N (De Cort et al., 2018). The lake occupies a tectonic trough bounded by the Lake Bogoria Escarpment to the east and the fault-fragmented, eastward-sloping Kipngatip plateau, composed of phonolite lava, to the west (Obando et al., 2016). The area is drained primarily by the Waseges-Sandai River, the Emsos River, and the Kipsirian River, alongside numerous ephemeral streams (Renaut & Owen, 2023). The lake is a topographically closed system with no surface outflow, and its water balance is supplemented by hot springs and geysers along its shoreline (Renaut et al., 2017; Salano et al., 2018). The climate is semi-arid, characterized by bimodal rainfall: long rains occur from March to May and short rains from October to November, with occasional precipitation during the typically dry June-September period (De Cort et al., 2018). Mean annual temperatures range from 14°C to 35°C (Renaut et al., 2017). The dominant soils are clays, particularly in the eastern and northern parts of the watershed within the Waseges-Sandai River catchment (De Cort et al., 2018). The region is predominantly inhabited by the Tugen-speaking Endorois community, whose livelihoods are primarily agro-pastoral.
(Insert Fig. 1 here: Map of the study area)
2.2. Data Collection and Analysis
Data for this study was obtained from both primary and secondary sources. Primary data was collected through focus group discussions (FGDs) and interviews with key informants to gather historical context and local ecological knowledge. Participants were drawn from the Endorois community, the Lake Bogoria Basin Water Resource User Association (WRUA), the Ministry of Water and Irrigation, and the Ministry of Agriculture, Livestock and Fisheries. Secondary data formed the core inputs for the Soil and Water Assessment Tool (SWAT) hydrological model. The model operates on the principle of the water balance equation (Neitsch et al., 2009):
Where SWt is the final soil water contents on day i and SW0 is initial soil water content. Time (t) is in days, whereas all the other measurements are in millimeters. The equation subtracts all forms of water loss on day i from precipitation on day i (Rday), that is surface runoff (Qsurf), evapotranspiration (Ea), loss to vadose zone (wseep) and return flow (Qgw) (Neitsch et al., 2009).
The study relied on several key spatial datasets. Topographic information was derived from a 30-meter resolution Digital Elevation Model (DEM) from the Shuttle Radar Topography Mission (SRTM), downloaded from USGS Earth Explorer. This DEM, originally in WGS 1984 geographic coordinates, was projected to WGS 1984 UTM Zone 37N for analysis and established the watershed's elevation range from 988 m to 2324 m. For climate inputs, gridded daily precipitation and maximum/minimum temperature data spanning from January 1, 1979, to December 31, 2020, were sourced from the Copernicus Climate Change Service and extracted for three specific grid points within the watershed (Table 1). Due to the absence of station records for other variables, relative humidity, solar radiation, and wind speed were subsequently simulated by the model's internal weather generator using the available precipitation, temperature, and geographical location data.
|
Parameter |
Name |
Lat |
Long |
Elevation |
|---|---|---|---|---|
|
Precipitation and temperature (maximum and minimum) |
pcp36.07_0.28 tmp36.07_0.28 |
0.28 0.28 |
36.07 36.07 |
998 998 |
|
Precipitation and temperature (maximum and minimum) |
pcp36.12_0.02 tmp36.12_0.02 |
0.02 0.02 |
36.12 36.12 |
1549 1549 |
|
Precipitation and temperature (maximum and minimum) |
pcp36.12_0.49 tmp36.12_0.49 |
0.49 0.49 |
36.12 36.12 |
1436 1436 |
(Insert Table 1 here: Climate parameters, coordinates and altitude)
Land Use/Land Cover (LULC) data for the years 1981, 1991, 2001, 2011, and 2020 were generated from Landsat satellite imagery (using Landsat 5, 7, and 8 respectively, acquired from USGS Earth Explorer) and classified into nine categories: water, grassland, tree cover, shrubland, rain-fed agriculture, irrigated agriculture, bare land, peri-urban areas, and Prosopis juliflora thickets. Finally, a soil map and its associated physical properties data were obtained from the FAO-soils database on the SWAT website. This dataset included critical parameters for hydrological modeling namely saturated hydraulic conductivity, bulk density, available water capacity, and soil texture. These parameters were used to define the Hydrologic Soil Group (HSG) for each soil unit, with the dominant groups in the watershed being C (slow infiltration) and D (very slow infiltration, primarily clay soils), as detailed in Table 2.
|
SNo. |
FAO Soil code |
HSG |
Area (km2) |
|---|---|---|---|
|
1. |
Re63-2c-248 |
C |
11.04 |
|
2. |
I-R-74 |
D |
81.67 |
|
3. |
Jc6-2a-118 |
D |
3.54 |
|
4. |
Ne12-2c-155 |
D |
3.11 |
|
5. |
Lf17-2ab-737 |
C |
0.64 |
(Insert Table 2 here: FAO soil data for Lake Bogoria watershed)
All spatial data processing, including projection, watershed delineation, and map algebra, was performed using ArcGIS 10.5, with the ArcSWAT 2012 interface used for model setup and operation.
2.3. SWAT Model Setup and Simulation
The SWAT model was constructed following a standardized procedure within the ArcSWAT interface. The process starting point is watershed delineation. The DEM was loaded to define the basin outlet, generate the stream network, and subdivide the watershed into 1705 sub-basins. Three key outlets were defined corresponding to the Waseges, Kipsirian, and Emsos/Ngiriki rivers (Fig. 2).
(Insert Fig. 2 here: Drainage network within Lake Bogoria watershed)
The next step defined and generated Hydrologic Response Units (HRUs). HRUs are unique combinations of land use, soil type, and slope that allow the model to account for spatial heterogeneity within a sub-basin. The land use and soil maps were reclassified into SWAT model codes using predefined lookup tables. The slope was discretized into three classes: 0–12%, 12–30%, and > 30%. The ‘multiple HRUs’ option was selected with a threshold of 5% for land use, soil, and slope, meaning any combination covering less than 5% of a sub-basin area was excluded to maintain model efficiency without significant loss of information.
Subsequently, weather data was inputted. The prepared gridded precipitation and temperature data were loaded into the model by linking the three climate points to their respective sub-basins using the ‘WGEN_user’ option. The model's built-in weather generator was used to synthesize the remaining required climate variables.
A unique SWAT model was built and executed for each of the five LULC maps (1981, 1991, 2001, 2011, 2020). Each simulation was run for a three-year period, with the first year serving as a warm-up period to initialize the model's water balance and minimize the influence of arbitrary initial conditions. The subsequent two years were used as the effective simulation period for analysis (Table 3). River discharge was simulated at the outlet of each of the three main rivers.
|
Year |
Climate period |
Warm up period |
Simulation period |
|---|---|---|---|
|
1981 |
1st Jan 1979–31st Dec 1981 |
1st Jan – 31st Dec 1979 |
1st Jan 1980–31st Dec 1981 |
|
1991 |
1st Jan 1989–31st Dec 1991 |
1st Jan – 31st Dec 1989 |
1st Jan 1990–31st Dec 1991 |
|
2001 |
1st Jan 1999–31st Dec 2001 |
1st Jan – 31st Dec 1999 |
1st Jan 2000–31st Dec 2001 |
|
2011 |
1st Jan 2009–31st Dec 2011 |
1st Jan – 31st Dec 2009 |
1st Jan 2010–31st Dec 2011 |
|
2020 |
1st Jan 2018–31st Dec 2020 |
1st Jan – 31st Dec 2018 |
1st Jan 2019–31st Dec 2020 |
(Insert Table 3 here: SWAT models simulation periods)
2.4. Model Calibration, Uncertainty, and Performance Evaluation
A significant challenge for this study was the absence of long-term, continuous observed streamflow data within the Lake Bogoria watershed for direct model calibration and validation. To address this, a regionalization approach was adopted. This technique is based on the premise that catchments with similar physical characteristics exhibit similar hydrological behavior, allowing for the transfer of parameters from a gauged (donor) catchment to an ungauged (recipient) one.
The donor catchment selected was the River Malewa watershed, part of the Lake Naivasha basin, whose parameters had been previously calibrated and validated by Muthuwatta (2004). While not ideal, this catchment was chosen as a best available proxy based on its location within the Kenyan Rift Valley system. Key physical characteristics of both the donor and recipient watersheds are compared in Table 4. Critical calibrated parameters from the Malewa study, including the curve number (CN2), groundwater revap coefficient (GW_REVAP), threshold depth for return flow (GWQMN), and saturated hydraulic conductivity (SOL_K), were transferred to initialize the Lake Bogoria SWAT model (Table 5).
|
Watershed |
Annual rainfall (mm) |
Mean Slope (%) |
Elevation (m) |
SHG |
LULC (Major) |
|---|---|---|---|---|---|
|
River Waseges |
1006 |
12.7 |
988–2324 |
C and D |
Shrub and grass |
|
River Kipsrian |
1006 |
5.78 |
988–1740 |
C and D |
Shrub and grass |
|
River Ngiriki |
1006 |
9.45 |
988–1610 |
C |
Shrub and grass |
|
River Malewa |
1100 |
49.21 |
1887–2800 |
D |
Agriculture and Shrub |
|
Parameter |
Value |
|---|---|
|
CN2 (runoff curve number f) |
72 (Shrub/grass/trees) 65 (Crop farming) |
|
GW_REVAP (Groundwater "revap" coefficient) |
0.116 |
|
QWQMIN (Threshold depth of water in the shallow aquifer required for return flow to occur (mm)) |
1009 |
|
SOL_K (Saturated hydraulic conductivity) |
23 (Shrub/grass/trees) 37.5 (Crop farming) |
To refine these parameters and account for uncertainty, the Sequential Uncertainty Fitting algorithm (SUFI-2) within the SWAT-CUP software was used. The SUFI-2 performs inverse modeling through a series of iterations, each involving numerous simulations, to identify the set of parameter values that result in the best fit between simulations and observations while quantifying parameter uncertainty. Given the lack of local streamflow data, this process focused on ensuring the model's internal consistency and realism based on the regionalized parameters and the physical constraints of the watershed.
Model performance for the final simulated discharge was evaluated using two statistical metrics namely the Coefficient of Determination (R²) and Nash-Sutcliffe Efficiency (NSE). The R² measures the proportion of the variance in observed data that is explained by the model. Values range from 0 to 1, with higher values indicating a better fit given by the formular:
Where, Qm is the observed (measured) stream flow on day i (m/s 3), Qs is the simulated stream flow on day i (m/s3), and bars indicate averages.
The NSE assesses the predictive power of the model relative to the mean of the observations. Values can range from -∞ to 1, where 1 indicates a perfect match. Performance is generally considered satisfactory if NSE > 0.5 and good to very good if NSE > 0.65 (Moriasi et al., 2007) described by the formular:
(Insert Table 4 here: Properties of the donor and recipient watersheds)
(Insert Table 5 here: Transferred watershed parameters)
3.1. Climate Variability (1979–2020)
Our analysis of the gridded climate data revealed clear trends in both precipitation and temperature over the study period. An increasing, though variable, trend in annual rainfall was observed (Fig. 3). The period 2018–2020 was the wettest, with an average annual rainfall of 1533.7 mm, culminating in a maximum of 1853.6 mm in 2020. In contrast, the driest period was 1999–2001, with an average of 771.1 mm, and a minimum of 512.3 mm recorded in 2009. This finding of an increasing rainfall trend aligns with the analysis by Ezenwa et al. (2018) for the Marigat and Mogotio sub-counties between 1985 and 2016.
Temperatures also exhibited a warming trend. The hottest period was 2009–2011, with an average maximum temperature of 31.03°C, while the coolest days were recorded during 1989–1991 (29.97°C). Notably, the warmest nights and early mornings occurred in the most recent period (2018–2020), with an average minimum temperature of 18.52°C, compared to the coolest minimum of 17.49°C during 1979–1981. This general warming pattern is corroborated by Okuku et al. (2024), who documented an upward trend in temperatures in the Marigat region over recent decades. This consistent warming is a critical factor influencing evapotranspiration rates and soil moisture content, thereby impacting the hydrological cycle.
(Insert Fig. 3 here: Rainfall and temperature in Lake Bogoria watershed)
3.2. Land Use and Land Cover (LULC) Change (1981–2020)
The LULC classification for the five time periods resulted in nine distinct classes, with significant transitions observed over the forty years (Table 6). The most profound change was a substantial net loss of 25.83 km² of tree cover, decreasing from 26.48% of the watershed area in 1981 to 19.74% in 2020. This represents a major transformation of the landscape, likely driven by demand for agricultural land, charcoal production, and fuelwood, a common driver of deforestation across East African drylands (Mwangi et al., 2016; Mekuria et al., 2018).
1981 | 1991 | 2001 | 2011 | 2020 | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
LULC type | Area (km2) | % area | Area (km2) | % area | Area (km2) | % area | Area (km2) | % area | Area (km2) | % area |
Grass | 70.66 | 18.45 | 66.64 | 17.40 | 62.14 | 16.22 | 74.35 | 19.41 | 71.88 | 18.77 |
Trees | 101.42 | 26.48 | 91.93 | 24.00 | 87.43 | 22.83 | 79.86 | 20.85 | 75.59 | 19.74 |
Shrub | 151.70 | 39.61 | 160.13 | 41.81 | 165.00 | 43.08 | 157.50 | 41.12 | 159.62 | 41.68 |
Rain-fed agriculture | 14.80 | 3.87 | 16.72 | 4.36 | 18.57 | 4.85 | 21.19 | 5.53 | 21.52 | 5.62 |
Irrigation agriculture | 5.11 | 1.33 | 6.60 | 1.72 | 5.76 | 1.50 | 6.19 | 1.62 | 6.28 | 1.64 |
Bare areas | 5.21 | 1.36 | 6.85 | 1.79 | 9.72 | 2.54 | 10.21 | 2.67 | 5.83 | 1.52 |
Water | 34.10 | 8.90 | 34.00 | 8.88 | 34.20 | 8.93 | 32.78 | 8.56 | 41.27 | 10.78 |
Peri-urban areas | 0 | 0.00 | 0.15 | 0.04 | 0.20 | 0.05 | 0.24 | 0.06 | 0.28 | 0.07 |
Prosopis species | 0 | 0.00 | 0 | 0.00 | 0 | 0.00 | 0.69 | 0.18 | 0.74 | 0.19 |
Total | 383.01 | 100.00 | 383.01 | 100.00 | 383.01 | 100.00 | 383.01 | 100.00 | 383.01 | 100.00 |
Conversely, shrubland expanded by 7.92 km², consolidating its position as the dominant land cover class (41.68% in 2020). This often represents a transitional state following deforestation or land degradation. Agricultural land also expanded significantly. Rain-fed agriculture increased by 6.72 km² and irrigated agriculture by 1.17 km², reflecting ongoing agricultural intensification and the encroachment of cultivation into previously natural vegetation, a trend observed across Kenya's rangelands (Onyango et al., 2022).
The surface area of water bodies increased by 7.17 km², a change dominated by a substantial 8.49 km² increase in the last decade (2011–2020). This aligns with the documented rise of Rift Valley lakes and is consistent with the findings of Kipkulei et al. (2025), who reported an 81.62 km² increase in water bodies across Baringo County between 2000 and 2024, leading to widespread flooding. Other changes included a slight expansion of bare areas (0.62 km²) and the emergence and spread of invasive Prosopis juliflora (0.74 km² since 2011), a species known for its detrimental impact on water tables and biodiversity in arid regions (Sharma et al., 2020).
(Insert Table 6 here: Land use land cover classes in Lake Bogoria watershed)
3.3. Modelled River Discharge
The SWAT model simulated the total annual discharge volumes for the three main rivers (Waseges, Kipsirian, and Emsos/Ngiriki) for each time period (Table 7, Fig. 4). The results indicate a substantial net increase of 9,399,938 m³ in total discharge between 1981 and 2020, with volumes rising from 8.23 million m³ to 17.63 million m³. An inter-decadal analysis reveals a decline in discharge during the first two decades (1981–1991 and 1991–2001), corresponding to the drier periods identified in the climate record. This was followed by a sharp increase in the last two decades (2001–2011 and 2011–2020), coinciding with the observed rise in precipitation.
River | 1981 | 1991 | 2001 | 2011 | 2020 | Change (1981–2020) |
|---|---|---|---|---|---|---|
Waseges | 5896730 | 5857542 | 3512250 | 8257585 | 11991928 | + 6,095,198 |
Kipsirian | 1780764 | 1807149 | 803101 | 2575466 | 4463890 | + 2,683,126 |
Ngiriki/Emsos | 552514 | 561393 | 308836 | 800123 | 1174128 | + 621,614 |
Total volume | 8,230,008 | 8,226,084 | 4,624,187 | 11,633,174 | 17,629,946 | + 9,399,938 |
The model performance, evaluated against the regionalized parameters, yielded a coefficient of determination (R²) of 0.78 and a Nash-Sutcliffe Efficiency (NSE) of 0.75. According to the benchmarks established by Moriasi et al. (2007), these values indicate "good" and "very good" model performance, respectively. This suggests that the model effectively captured the major hydrological processes governing streamflow generation in the watershed. The R² value indicates that 78% of the variance in the simulated streamflow is explained by the model, while the NSE signifies a close match between the magnitude and timing of simulated flow dynamics.
(Insert Table 7 here: The modelled rivers discharge volumes)
(Insert Fig. 4 here: The modelled rivers discharge volumes)
3.4. Discussion
The core finding of our study is that despite a slight increase in temperature (potentially increasing evapotranspiration) and a significant expansion of irrigated agriculture (potentially increasing water abstraction), the total river discharge in the Lake Bogoria watershed more than doubled over 40 years. We interpret this paradox by examining the relative magnitudes of the counteracting drivers: increased rainfall and extensive deforestation.
The strong correlation between the wettest periods (e.g., 2018–2020) and the highest simulated discharge volumes (in 2020) points to precipitation variability as the primary driver of discharge trends in this watershed. This is a well-established hydrological principle, particularly in semi-arid catchments where streamflow is highly responsive to rainfall pulses (Tirupathi & Shashidhar, 2020; Muthoka et al., 2021). The increased rainfall volume directly leads to higher surface runoff generation and greater recharge, ultimately manifesting as increased river discharge. This finding is supported by Kareri (2018), whose analysis of historical gauge data at station 2EBO4 on the Waseges River showed higher discharge in 2006 compared to 1960 for most months of the year.
Our results strongly suggest that the widespread conversion of tree cover to shrubland and agriculture (a net loss of 25.83 km² of forests) amplified the rainfall-runoff response. Forests enhance infiltration, intercept rainfall, and promote evapotranspiration, thereby dampening flood peaks and sustaining baseflow during dry periods (Ilstedt et al., 2016; Mugo et al., 2020). Their removal reduces canopy interception and evapotranspiration, leading to more water being partitioned into surface runoff rather than soil storage or groundwater recharge (Muthee et al., 2023). This study's findings are consistent with a similar SWAT application in Kenya's Muringato basin, where a 21.4% loss in forest cover between 1990 and 2020 contributed to a 5 mm increase in runoff (Muthee et al., 2023). Therefore, deforestation acted as a secondary but significant amplifier of the discharge increase driven by higher rainfall.
The expansion of irrigated agriculture, while a concern for local water conflicts, was likely too small in scale (an increase of only 1.17 km², or 0.31% of the watershed) to cause a statistically significant reduction in total annual discharge at the watershed outlet, especially when overshadowed by the massive increases in rainfall input. The impact of rising temperatures on increasing evaporative demand was likely negated by the concurrent increase in rainfall amount and cloud cover. Furthermore, the model may not have fully captured the localized, high-intensity abstraction that occurs during critical dry periods.
A critical insight from our research is the apparent contradiction between the model's result of increased annual discharge and the consistent local testimony of decreased dry-season flow. This is explained by a shift in the seasonal distribution of water. Increased rainfall and deforestation may be generating higher peak flows during the wet seasons, but concurrently, several factors could be reducing dry-season baseflow. The reduced infiltration could be linked to deforestation and soil compaction from agricultural expansion, which can reduce groundwater recharge, diminishing the slow, sustained baseflow that feeds rivers during droughts (Mureithi et al., 2016). The expansion of irrigation, even if small at the watershed scale, can have a disproportionately large effect on low flows during the dry season when river levels are at their minimum (Bwana et al., 2022). The model's regionalized calibration may underestimate the actual volume of water abstracted, especially from the upper catchment as reported by WRUA officials. Warmer temperatures increase evaporative demand, potentially drying out soils faster and reducing the water available to sustain baseflow later into the dry season (Ayana et al., 2016). This nuanced understanding aligns with the local reports of conflicts over water during dry periods, despite the model indicating more water annually. It underscores that an increase in total water volume does not equate to improved water security if its availability becomes more temporally variable and concentrated.
A significant limitation of our study is the treatment of the watershed as a purely rainfall-runoff system. Lake Bogoria receives substantial input from hot springs and geothermal seepage along its shores (Renaut et al., 2017). These groundwater-derived inputs are a crucial component of the lake's water balance and likely contribute to the baseflow of the feeder rivers. The SWAT model, in its standard setup, does not account for such external, deep groundwater inputs. This omission means our simulated discharge may underestimate the true baseflow, and the attributed increase from surface processes may be partially overstated. Future studies should seek to quantify and incorporate these geothermal fluxes. Furthermore, the regionalization approach for calibration, while necessary, remains a source of uncertainty, and future efforts should prioritize the collection of in-situ streamflow data for more robust model calibration and validation.
The application of the SWAT model to the Lake Bogoria watershed over four decades reveals that climate variability, specifically a significant increase in rainfall, has been the dominant driver of increased river discharge. This primary effect has been substantially amplified by widespread land cover change, particularly deforestation, which has altered the hydrological partitioning of rainfall towards increased surface runoff. The impacts of rising temperatures and the expansion of irrigated agriculture, while locally significant for water access and conflict, were overshadowed by these larger forces at the annual watershed scale.
However, the model's indication of increased total water availability stands in contrast to local experiences of more severe dry-season water scarcity. This highlights a critical shift towards a more "boom-and-bust" hydrological regime: higher peak flows but potentially diminished baseflow, driven by land degradation and increased abstraction. This has profound implications for water security, increasing vulnerability to droughts despite an overall wetter trend.
Therefore, the study supports the recommendation for stricter environmental audits on upstream water abstraction by the Water Resources Authority (WRA). Management strategies must move beyond simply quantifying total water and focus on ensuring its equitable and sustainable distribution through time. This should include:
Promoting Natural Infrastructure - initiatives to restore tree cover in critical recharge zones to enhance infiltration, reduce peak runoff, and improve dry-season baseflow.
Regulating Abstraction - implementing and enforcing stricter, evidence-based limits on dry-season water abstraction from rivers, particularly for large-scale irrigation.
Investing in Storage - supporting the development of managed aquifer recharge and small-scale storage systems to capture excess wet-season flow for use during dry periods.
Funding Sources
This research did not receive any grant from funding agencies.
Author Contribution
Daisy Moso wrote the main manuscript text. Charles Kigen undertook the spatial analysis and prepared the tables and figures. George Ogendi and Bernard Kirui were involved in conceptualization, manuscript revision and supervision. All the authors reviewed the work.
Data Availability
Data is provided within the manuscript
- Abbaspour, K. C., Yang, J., Maximov, I., Siber, R., Bogner, K., Mieleitner, J.,. .. Srinivasan, R. (2007). Modelling hydrology and water quality in the Pre-Alpine/Alpine Thur watershed using SWAT. Journal of Hydrology, 333, 413–430.
- An, L., Wang, J., Huang, J., Pokhrel, Y., Hugonnet, R., Wada, Y.,. .. Zhang, G. (2021). Divergent causes of terrestrial water storage decline between drylands and humid regions globally. Geophysical Research Letters, 48, e2021GL095035.
- Ayana, E. K., Cecchi, P., & Mequanent, F. (2016). Understanding the impact of land use and climate change on the hydrological processes of the Lake Tana basin, Ethiopia. Hydrological Processes, 30(20), 3666–3680.
- Bwana, R. M., Ndegwa, G., & Njeru, J. (2022). Impacts of irrigation water abstraction on streamflow in the Upper Ewaso Ng'iro River Basin, Kenya. Journal of Water and Climate Change, 13(2), 681–697.
- De Cort, G., Verschuren, D., Ryken, E., Wolff, C., Renaut, R., Creutz, M.,. .. Mees, F. (2018). De Cort, G., Verschuren, D., RykenMulti-basin depositional framework for moisture‐balance reconstruction during the last 1300 years at Lake Bogoria, central Kenya Rift Valley. Sedimentology, 65(5), 1667–1696.
- Ezenwa, L., Omondi, P., Nwagbara, M., Gbadebo, A., & Bada, B. (2018). Climate variability and its effects on gender and coping strategies in Baringo County, Kenya. Journal of Applied Science and Environmental Management, 22 (5), 699–706. DOI: https://dx.doi.org/10.4314/jasem.v22i5.14.
- GOK. (2011). Vision 2030 Development Strategy for Northern Kenya and other arid lands. Government of Republic of Kenya.
- Herrnegger, M., Stecher, G., Schwatke, C., & Olang, L. (2021). Hydroclimatic analysis of rising water levels in the Great Rift Valley Lakes of Kenya. Journal of Hydrology: Regional Studies, 36, 100857.
- Ilstedt, U., Bargués Tobella, A., Bazié, H. R., Bayala, J., Verbeeten, E., Nyberg, G., ... & Malmer, A. (2016). Intermediate tree cover can maximize groundwater recharge in the seasonally dry tropics. Scientific Reports, 6(1), 1–12.
- IPCC. (2020). Climate change and land: An IPCC special report on climate change,desertification, land degradation, sustainable land management, food security, and greenhouse gas fluxes in terrestrial ecosystems. Summary for policymakers. Intergovernmental Panel on Climate Change.
- Kareri, P. W. (2018). Land use changes and their impacts on wetlands in Loboi plains, Baringo County, Kenya. [Doctoral dissertation, Moi University].
- Khandu, F. E., Schumacher, M., Awange, J. L., & Muller Schmied, H. (2016). Exploring the influence of precipitation extremes and human water use on total water storage (TWS) changes in the Ganges-Brahmaputra-Meghna River Basin. Water Resources Research, 52(3), 2240–2258. https://doi.org/10.1002/2015wr018113.
- Kipkulei, H., Rotich, B., Ahmed, A., Lameck, A., Burudi, J., Hounkpati, K.,. .. Kindu, M. (2025). Land use/land cover dynamics in an arid and semi-arid landscape: A 24-year analysis of Baringo County, Kenya (2000–2024). Global and Earth Surface Processes Change, 4, 100006.
- Mekuria, W., Edzo, V., & Yami, M. (2018). The role of exclosures in the restoration of ecosystem services and livelihoods in the central and northern Ethiopian highlands. In Restoration of Ecosystem Services and Livelihoods (pp. 105–124). Springer, Cham.
- Moriasi, D. N., Arnold, J. G., Van Liew, M. W., Bingner, R. L., Harmel, R. D., & Veith, T. L. (2007). Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the American Society of Agricultural and Biological Engineers, 50(3), 885–900.
- Mugo, G. M., Kanga, S., & Özdemir, S. (2020). Modelling the impact of land use/cover changes on hydrological regime in the Upper Ewaso Ng'iro River Basin, Kenya. Journal of Geography, Environment and Earth Science International, 24(5), 1–15.
- Mureithi, S. M., Verdoodt, A., Njoka, J. T., & Van Ranst, E. (2016). Impact of community conservation management on the hydrological response of a degraded semi-arid floodplain in Kenya. Land Degradation & Development, 27(3), 550–561.
- Muthee, S. W., Kuria, B. T., Mundia, C. N., Sichangi, A. W., Kuria, D. N., Goebel, M., & Rienow, A. (2023). Using SWAT to model the response of evapotranspiration and runoff to varying land uses and climatic conditions in the Muringato basin, Kenya. Modelling Earth Systems and Environment, 1531–1543.
- Muthoka, J. M., Kiptum, C. K., & Obando, J. A. (2021). Hydrological response to land use/land cover changes in the Olkinyei watershed of the Mara River Basin, Kenya. Journal of Water and Climate Change, 12(7), 3155–3175.
- Muthuwatta, L. P. (2004). Long term rainfall-runoff-lake level modelling of the Lake Naivasha basin, Kenya. [Master's dissertation, International Institute for Geo-information Science and Earth Observation].
- Mwangi, H. M., Julich, S., Patil, S. D., Feger, K. H., & Schwärzel, K. (2016). Modelling the impact of land use change on hydrological responses in the Mt. Kenya region. Hydrology and Earth System Sciences, 20(3), 1207–1223.
- Neitsch, S. L., Arnold, J. G., Kiniry, J. R., & Williams, J. R. (2009). Overview of Soil and Water Assessment Tool (SWAT) model. Tier B, 8, 3–23.
- Obando, J. A., Onywere, S., Shisanya, C., Ndubi, A., Masiga, D., Irura, Z.,. .. Maragia, H. (2016). Impact of short-term flooding on livelihoods in the Kenya Rift Valley Lakes. In M. E. Meadows, & J. -C. Lin, Geomorphology and society, advances in geographical and environmental sciences (pp. 193–215). Springer Japan.
- Okuku, K. O., Onyando, J. O., Okwany, R. O., & Kiptum, C. K. (2024). Climate trends and their impact on sorghum production in Marigat, Baringo County: A historical analysis. Open Journal of Modern Hydrology, 14, 106–129.
- Onyango, D. O., Opiyo, F. E., & Mureithi, S. M. (2022). Land use and land cover changes and their impacts on ecosystem services in the Nzoia River Basin, Kenya. Environmental Monitoring and Assessment, 194(5), 1–21.
- Onywere, S. M., Shisanya, C. A., Obando, J. A., Ndubi, O. A., Masiga, D., Irura, Z.,. .. Maragia, H. O. (2013). Geospatial extent of 2011–2013 flooding from the Eastern African Rift Valley Lakes in Kenya and its implication on the ecosystem. Paper presented during the Kenya Soda Lakes Workshop, 4–7 Dec 2013. Naivasha, Kenya: Kenya Wildlife Service Training Institute.
- Renaut, R. W., & Owen, R. B. (2023). Lake Bogoria. In R. W. Renaut, & R. B. Owen, The Kenya Rift Lakes: Modern and Ancient. Limnology and Limnogeology of Tropical Lakes in a Continental Rift (pp. 303–362). Springer.
- Renaut, R., Owen, R., & Ego, J. (2017). Geothermal activity and hydrothermal mineral deposits at southern Lake Bogoria, Kenya rift valley: Impact of lake level changes. Journal of African Earth Sciences, 129, 623–646.
- Rodell, M., Famiglietti, J. S., Wiese, D. N., Reager, J. T., Beaudoing, H. K., Landerer, F. W., & Lo, M. H. (2018). Emerging trends in global freshwater availability. Nature, 557(7707), 651–659. https://doi.org/10.1038/s41586-018-0123-1.
- Salano, O., Makonde, H., Kasili, R., & Boga, H. (2018). Isolation and characterization of fungi from a hot-spring on the shores of Lake Bogoria, Kenya. Journal of Yeast and Fungal Research, 9(1), 1–13.
- Schagerl, M. (2016). Soda Lakes of East Africa. Springer International Publishing.
- Sharma, R., Raghubanshi, A. S., & Singh, J. S. (2020). The invasive Prosopis juliflora and its impacts on plant communities and soil properties in the drylands of India: a review. Tropical Ecology, 61(1), 1–12.
- Tirupathi, C., & Shashidhar, T. (2020). Investigating the impact of climate and land-use land cover changes on hydrological predictions over the Krishna river basin under present and future scenarios. Science of The Total Environment, 721, 137736.
No competing interests reported.