Climate Change and Irrigation Effects on Hydrology and Crop Yield in the Geba Watershed, Tigray Region, Northern Ethiopia

Highlights

What are the main findings?

• Evapotranspiration emerged as the dominant component of the watershed water balance under both current and future climate conditions.
• Future climate scenarios substantially reduce lateral flow, groundwater recharge, and soil-water storage, indicating increasing hydrological stress in the watershed.
• Irrigation altered hydrological flow partitioning by increasing consumptive water use; however, it cannot fully compensate for climate-induced water deficits under future warming scenarios.
• Crop responses varied significantly across management systems, with maize (Zea mays L.) and wheat (Triticum aestivum L.) showing moderate resilience, while Teff (Eragrostis tef) exhibited persistent yield declines under most future scenarios.

What are the implications of the main findings?

• Sustainable adaptation should emphasize water-efficient agronomic practices, soil-moisture conservation, and integrated watershed management rather than large-scale irrigation expansion alone.
• Teff (Trotter crop) appears highly vulnerable to future warming and rainfall variability, requiring targeted climate adaptation and breeding strategies.
• Future watershed planning should prioritize groundwater recharge enhancement, efficient irrigation technologies, and management approaches that address increasing evaporative demand and declining subsurface water availability.

Abstract

Climate change and irrigation expansion are expected to substantially alter hydrological processes and agricultural productivity in the semi-arid watersheds of northern Ethiopia; however, their combined impacts remain insufficiently quantified. This study evaluated the effects of future climate change and irrigation management on watershed hydrology and crop yield in the Geba watershed using the Soil and Water Assessment Tool Plus (SWAT+). The model was calibrated and validated using observed daily streamflow data for the 2006–2020 period and driven by an ensemble of five bias-corrected CORDEX Africa regional climate models (RCMs) under the RCP 4.5 and RCP 8.5 scenarios for the mid-century (2046–2060) and late-century (2086–2100) periods. Two agricultural management systems, namely rainfed and irrigation-rainfed integrated management, were evaluated. Model performance was satisfactory for streamflow simulation, with NSE values of 0.56 and 0.50 and KGE values of 0.64 and 0.54 during calibration and validation, respectively. The results indicate a progressive shift toward an evapotranspiration-dominated hydrological regime under future climate conditions. Under rainfed management, surface runoff and evapotranspiration increased by up to 60% and 30%, respectively, whereas groundwater recharge and lateral flow declined substantially. Irrigation scenarios intensified hydrological stress by reducing percolation, lateral flow, and water yield by up to 80%, 60%, and 55%, respectively. Statistical analyses revealed that climate forcing, management type, and their interactions significantly affected hydrological responses (p < 0.001), with emission pathways representing the dominant driver of variability. Crop responses varied considerably among management systems and crop types. Rainfed maize and wheat exhibited moderate yield increases under mid-century conditions, whereas teff consistently showed negative responses under most climate scenarios, indicating high vulnerability to warming and moisture stress. Under irrigation management, most crops experienced substantial yield reductions during late-century periods, although tomatoes showed localized gains under high-emission scenarios. Overall, the findings demonstrate that irrigation expansion alone may not provide sustainable adaptation under increasing climate stress because it intensifies evapotranspiration and reduces groundwater recharge. Integrated watershed management, climate-resilient crop selection, efficient irrigation practices, and soil-moisture conservation strategies are therefore essential for sustaining agricultural productivity and water availability in semi-arid Ethiopian watersheds.

1. Introduction

In Ethiopia, agriculture contributes approximately one-third of gross domestic product and supports the livelihoods of 70–80% of the population while generating the majority of export earnings [1]. Historical climate data reveal that Ethiopia has experienced rising temperatures of 0.3–0.66 °C over the last three decades (1990–2020), with future projections indicating further increases that threaten cereal production and water availability [2,3]. Climate change fundamentally alters hydrological processes through modifications in precipitation patterns, temperature regimes, and evapotranspiration rates, consequently affecting streamflow, drought frequency, and irrigation water availability [4,5,6].

A growing body of research in Ethiopia has examined the impacts of climate change and land-use dynamics on watershed hydrology, particularly using models such as SWAT [7,8] and SWAT+ [9,10,11,12,13]. Studies in northern Ethiopia, including the Geba watershed and other basins, have demonstrated that climate variability and land-use/cover changes significantly influence runoff, sediment yield, and evapotranspiration [14,15,16]. Similarly, basin-scale assessments in the Upper Blue Nile and Tekeze systems have highlighted that future climate scenarios can alter streamflow regimes and increase hydrological uncertainty [17,18,19]. Despite these advances, most presented studies have primarily focused on hydrological responses in isolation, often neglecting the linkage between hydrological processes and agricultural productivity. For instance, while studies such as [19,20,21,22] quantified changes in runoff, evapotranspiration, and water yield under climate scenarios, they did not extend their analyses to evaluate how these hydrological changes translate into crop yield responses, particularly under different management systems. Conversely, agro-climatic studies in Ethiopia have assessed the impacts of climate change on crop productivity, especially under rainfed conditions [23,24]. These studies highlighted the vulnerability of crops to rainfall variability and rising temperatures but largely excluded hydrological process-based modeling, thereby limiting their ability to explain the underlying water balance mechanisms driving yield changes.

While irrigation is widely recognized as a critical adaptation strategy in semi-arid regions such as the Tigray region [25,26], most studies either assume static management conditions or evaluate irrigation benefits without assessing its feedback on watershed hydrology. Emerging research has shown that irrigation can significantly alter evapotranspiration, reduce groundwater recharge, and modify flow pathways [27,28], yet these feedback mechanisms remain poorly quantified at watershed scale in Ethiopia.

Moreover, existing studies generally adopt single-factor or additive approaches, examining climate change, land-use change, or management practices independently. Very few studies have explored the interaction effects between emission scenarios, temporal scales, and management options, despite growing evidence that hydrological and agricultural responses are inherently non-linear and interaction-driven [29,30]. As a result, the combined and potentially compounding impacts of climate change and irrigation expansion on water availability and crop productivity remain inadequately understood. Therefore, this research investigates the impact of climate change and extensive irrigation on the hydrologic system and crop yield in the Geba watershed, addressing specific objectives such as (i) quantifying the hydrological future under climate change; (ii) comparing crop-yield responses under rainfed and irrigated management; and (iii) identifying the extent to which scenario, period, and management interactions shape water availability and agricultural outcomes.

2. Materials and Methods

2.1. Study Area

The Geba watershed is located in the Tigray region of northern Ethiopia and forms part of the Tekeze-Atbara basin. The catchment covers about 2440 km² and ranges from approximately 930 to 3300 m a.s.l., located on the geographically boundary coordinates of 13°32′24″ to 14°15′00″ north and longitudes 39°18′18″ to 39°47′45″ east. The main tributaries are Suluh, Genfel, and Agulae, to form the main Geba River (near Geba Lake, west of Mekelle), which flows towards the Tekeze River in a predominantly dendritic drainage pattern influenced by major geological structures [31,32]. Suluh, Genfel and Agulae have river flow gauging stations near Hawzien, Wukro and Agulae towns respectively. The basin has rugged topography, steep average slopes, and a semi-arid-to-dry sub-humid climate with marked seasonality. Most rainfall occurs during the Kiremt season (June–September), while a smaller and less reliable Belg season may occur from March to May. Annual rainfall varies strongly across elevation zones, and long-term potential evapotranspiration exceeds rainfall, making the basin hydrologically water limited. Runoff is seasonal, and groundwater recharge is comparatively small, which increases the sensitivity of both rainfed and irrigated agriculture to climatic change. The main crops of the region are the Ethiopian native cereal called teff (Eragrostis tef, Zucc.), maize, and wheat. Its geographical location is described in Figure 1.

Figure 1. Map description of the study area. Ethiopian regions (top left), Tigray region (bottom left), Geba watershed with the position of gauging stations at outlets (right).

2.2. Data Collection and Missing Data Treatment

Historical daily precipitation and maximum and minimum temperature records were obtained primarily from the Ethiopian Meteorological Institute (EMI), with supplementary rainfall observations from nearby stations where necessary, while daily streamflow data at the Geba River outlet near Mekelle were acquired from the Ethiopian Ministry of Water and Energy (MoWE). Prior to hydrological modelling, all hydro-meteorological datasets underwent comprehensive quality control, including consistency checks, outlier detection, and identification of missing observations. A hierarchical gap-filling strategy was implemented according to the duration and location of missing records to preserve the temporal characteristics of the climate series. Missing observations occurring at the beginning or end of the observation period were excluded (censored), as reliable estimation of endpoint gaps is generally not feasible without adjacent observations and may introduce substantial uncertainty into subsequent analyses. This approach is widely recommended in hydro-meteorological time-series preprocessing to avoid bias associated with extrapolation of incomplete records [33,34].

Internal missing records were filled using methods appropriate to the length of the data gap. Short gaps of seven consecutive days or less were reconstructed using linear interpolation, which effectively preserves the statistical characteristics of daily meteorological series when missing intervals are brief [34,35]. Longer missing periods, including gaps extending over several months or approximately one year, were reconstructed using the random forest (RF) machine-learning algorithm [36]. Random forest is a non-parametric ensemble approach capable of capturing complex nonlinear relationships among meteorological variables without assuming a predefined data distribution. In this study, observations from neighboring stations together with temporal predictors were used to estimate the missing daily values. Previous studies have demonstrated that random forest provides a more accurate and robust reconstruction of missing hydro-meteorological observations than conventional interpolation and regression methods, particularly for extended missing periods [36,37,38].

Spatial inputs (Figure 2c) included the World Cover 2021 land-cover dataset (10 m resolution) [32,39] provided by the European Space Agency, https://esa-worldcover.org/en (accessed on 21 January 2025), soil information was derived from the FAO Digital Soil Map of the World accessed from https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/faounesco-soil-map-of-the-world/en/, accessed on 21 January 2025 [40] (Figure 2a) and digital elevation data were obtained from the Shuttle Radar Topography Mission (SRTM) (Figure 2b, https://www.earthdata.nasa.gov/sensors/srtm, accessed on 21 January 2025 [41]. All spatial layers were reprojected, clipped to the watershed, and harmonized at 90 m spatial resolution before model setup. Slope classes (%) were limited to 4 classes (>5, 5–10, 10–20, >20) following [42,43]. Soil attributes were translated into SWAT-compatible hydraulic parameters using pedotransfer functions [44,45,46,47] and custom soil entries where necessary. Land, crop, and water-management information was compiled from field surveys, local crop calendars, and irrigation practice observations. To quantify irrigated crop productivity, structured field surveys were conducted on dominant crops such as onion (Allium cepa L.), tomato (Solanum lycopersicum L.), and maize, across representative farmer plots, following protocols similar to those used in recent agricultural water management studies [20,48].

Figure 2. SWAT+ input layers (a) FAO soil unit code; (b) Slope classes from SRTM; (c) ESA land use/land cover.

2.3. Description of Hydrological Modeling

The Soil and Water Assessment Tool Plus (SWAT+3.0.5) is a semi-distributed, process-based eco-hydrological model developed to simulate the impacts of land management practices, climate variability, and environmental change on water, sediment, and agricultural production at watershed scale. It represents hydrological processes through the delineation of sub-basins and hydrological response units (HRUs), enabling the simulation of key components such as surface runoff, evapotranspiration, infiltration, lateral flow, groundwater recharge, and channel routing, while also integrating crop growth and management operations within a unified framework [7,49,50]. SWAT+ is an advanced evolution of the original SWAT model, developed to address structural limitations of earlier versions by introducing a more flexible spatial representation, improved water routing through landscape units, and enhanced modularity that allows better representation of complex watershed processes and management interventions [7,9]. Unlike the traditional SWAT model, SWAT+ separates landscape elements such as uplands, floodplains, and channels, thereby improving the realism of flow pathways and enabling more accurate assessment of land-use change, irrigation practices, and conservation measures. The model operates on a continuous time step and incorporates weather inputs (precipitation, temperature, solar radiation, wind speed, and humidity) to drive hydrological and plant growth processes, making it particularly suitable for evaluating climate change impacts on water resources and agricultural productivity in data-scarce regions [7,47,51]. Due to its robust physical basis, adaptability, and proven applicability across diverse environments, SWAT+ has become a widely used tool for integrated watershed management, climate change impact assessment, and sustainable agricultural planning worldwide

SWAT+ is calibrating the following hydrological balance equation on every location of the catchment

$$\mathrm{S}\mathrm{W}\mathrm{t}=\mathrm{S}\mathrm{W}0+\sum _{i=1}^n(Rday-Qsurf-Ea-Wseep-Qgw)$$

(1)

where

• SWt = final soil water content at time t (mm);
• SW0 = initial soil water content (mm);
• Rday = precipitation on day i (mm);
• Qsurf = surface runoff (mm);
• Ea = actual evapotranspiration (mm);
• Wseep = percolation from the soil profile to the vadose zone (mm);
• Qgw = return flow (baseflow/groundwater flow) (mm);
• t = time (days).

2.4. SWAT+ Model Setup

Following the preparation and harmonization of all spatial and climatic datasets, the SWAT+ model was configured and executed using a structured workflow aligned with established hydrological modeling protocols [8,9,10,12,42]. The model setup procedure comprised watershed delineation and hydrological response unit (HRU) definition, as described in Figure 3.

Figure 3. Hierarchical framework diagram for simulating climate change impact on hydrological components and yields of major crops in the Geba watershed.

Watershed delineation was based on the processed digital elevation model (DEM) and a predefined flow-accumulation threshold, and hydrological response units (HRUs) were created from land use, soil, and slope classes using 10% thresholds to remove negligible landscape combinations. Observed weather inputs were linked to sub-basins, and a weather-generator file was prepared to provide missing climatic variables needed by the model (Figure 4).

Figure 4. Watershed delineation: (a) Geba watershed sub-basins; (b) hydrological response units (HRUs).

The simulation period was 2006–2020, with a three-year warm-up period. Temporal splitting of observed records into independent calibration and validation phases is essential to test model transferability and predictive robustness under different hydroclimatic conditions [49,52].

Model sensitivity analysis was performed in the SWAT+ Toolbox2.4 using a Sobol variance-based approach [53,54]. It was employed to execute 100 model realizations (seed = 100) to ensure reproducibility of parameter sampling. The number of Monte Carlo samples was set to 500, balancing computational efficiency with statistical reliability of sensitivity indices, as adequate sampling size is essential to stabilize Sobol indices and ensure convergence of variance decomposition results in complex watershed models [53].

Unlike the sequential uncertainty fitting version 2 (SUFI-2) algorithm, which primarily estimates parameter and prediction uncertainty through the 95% prediction uncertainty (95PPU) together with the P-factor and R-factor during model calibration, the Sobol method is specifically designed to identify the relative contribution of individual parameters and higher-order parameter interactions to the overall variance of model predictions, thereby providing a comprehensive assessment of global parameter sensitivity [53,55].

Calibration and validation were then carried out against daily streamflow using standard metrics, including the Nash–Sutcliffe efficiency (NSE), Kling–Gupta efficiency (KGE), percent bias (PBIAS), mean square error (MSE), and root mean square error (RMSE). These metrics provide complementary assessments of model accuracy, bias, and variability representation and are widely adopted in watershed modeling studies [9,49].

Kling–Gupta efficiency (KGE)

KGE = 1 − ((r − 1)² + (β − 1)² + (γ − 1)²)^1/2

(2)

Mean squared error (MSE)

MSE = (∑ (Y_sim − Y_obs)²)/n

(3)

Nash–Sutcliffe efficiency (NSE)

NSE = 1 − (∑(Y_sim−Y_obs)²)/(∑(Y_obs − Ŷ_obs))

(4)

Root mean squared error (RMSE)

RMSE = ((∑ (Y_sim − Y_obs)²)/n)^1/2

(5)

Percent bias (PBIAS)

PBIAS = (∑(Y_sim − Y_obs)) ∗ 100/(∑(Y_obs))

(6)

where Y_sim = simulated streamflow (m³/s)

Y_obs = observed streamflow (m³/s)

Ŷ_obs = mean of observed streamflow (m³/s)

2.5. Climate Model Selection, Bias Correction, and Multi-Model Ensemble Development

2.5.1. Climate Model Selection, Bias Correction

Future and historical climate projections were derived from the CORDEX-Africa regional climate downscaling initiative using five regional climate models (RCMs), namely CCLM4-8-17, CanRCM4, REMO2009, RCA4, and CRCM5, driven by their corresponding global climate models (GCMs). These RCMs were selected because previous evaluations have consistently demonstrated their relatively superior ability to reproduce the climatology and seasonal rainfall characteristics of Ethiopia and East Africa, particularly within the Ethiopian Highlands and Upper Nile basin [56,57,58,59,60,61]. Daily precipitation (pr), maximum air temperature (tasmax), and minimum air temperature (tasmin) for the historical (1990–2005), mid-century (2046–2060), and late-century (2086–2100) periods under RCP 4.5 and RCP 8.5 emission scenarios were downloaded in NetCDF format. Station-scale climate series were extracted from the CORDEX grid cells using a reproducible workflow developed in Python 3.11 (packages: xarray, netCDF4, pandas, NumPy, GeoPandas, and SciPy) [54] with the complementary Climate Model Data for Hydrologic Modeling (CMhyd) tool [62].

The historical performance of each RCM was evaluated against observed station records for 1990–2005 before bias correction. Model skill was assessed using six complementary performance statistics—the KGE, NSE, RMSE, PBIAS, mean absolute error (MAE), and Pearson correlation coefficient (to collectively quantify systematic bias, absolute prediction error, temporal agreement, variability, and hydrological performance)— that are widely recommended for evaluating climate simulations used in hydrological modelling [63,64,65,66]). Systematic biases in daily precipitation, maximum temperature, and minimum temperature were subsequently corrected using quantile mapping (QM) and distribution mapping (DM), both of which adjust the complete probability distribution rather than only the mean climate. Quantile mapping showed relatively better performance for correcting precipitation distributions, rainfall extremes, and seasonal variability in our study and other studies across Ethiopia, the Upper Blue Nile basin, and East Africa [61,67,68,69,70], while distribution mapping has shown comparable performance for temperature variables and in regions with relatively stable climatic distributions [65,71]. Consequently, quantile mapping was applied because of its better overall performance according to RMSE, MAE, PBIAS, r, KGE, and NSE for the final corrected climate datasets.

Pcorr = Fobs⁻¹ (Fmod (Praw))

(7)

where

• Praw = raw RCM precipitation;
• Fmod = model CDF;
• Fobs⁻¹ = inverse observed CDF;
• Pcorr = corrected precipitation.

2.5.2. Multi-Model Ensemble Development

To reduce the uncertainty associated with each of 5 regional climate models (RCMs), this study employed a performance-based weighted multi-model ensemble (MME) approach. Rather than assigning equal importance to all climate models, individual model weights were determined according to their ability to reproduce the observed historical climate (1990–2005) using the above specified performance metrics, including NSE, KGE, r, RMSE, MAE, and PBIAS. Climate models demonstrating superior agreement with observed precipitation and temperature records were assigned larger weights in the ensemble, whereas models with relatively poorer performance contributed proportionally less. The weighted ensemble was subsequently used to generate future daily precipitation, maximum temperature, and minimum temperature under the RCP 4.5 and RCP 8.5 scenarios for the periods 2046–2060 and 2086–2100. Compared with single-model projections or simple arithmetic averaging, performance-based weighting generally produces more reliable climate projections by reducing systematic model biases and better representing regional climatic conditions, thereby providing a more robust basis for subsequent SWAT+ simulations of watershed hydrology and crop yields [63,72,73,74,75].

Accordingly, we employed the entropy weight method of model ensemble following [76].

$$X_{MME}={\sum }_{i=1}^n{w}_i{X}_i$$

(8)

• X_MME = weighted ensemble daily precipitation (mm day⁻¹), or daily maximum temperature, or daily minimum temperature;
• X_i = daily precipitation or daily maximum, or daily minimum temperature simulated by the ith CORDEX-Africa model;
• w_i = normalized weight assigned to the ith model;
• n = number of selected climate models.

This was subject to the following constraint:

$${\sum }_{i=1}^n{w}_i=1$$

(9)

The weight of the RCMs models was assigned based on the following:

$$w_j={\displaystyle \frac{d_i}{\sum d_j}}$$

(10)

where

• w_j = weight assigned to model j;
• Score_i = weighted ranking score;
• m = number of climate models.

We have utilized the weight for each metric based on the following:

$${\mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}}_i= {\sum }_{j}wjPij$$

(11)

Normalizing using the following:

$${\mathrm{w}}_i= {\displaystyle \frac{S{core}_i}{\sum {\mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}}_i}}$$

(12)

2.6. Climate Scenarios and Management Scenarios Setup

Locally representative crop yield values are described in Table 1. Detailed representative crop management data for onion, tomato and maize, the dominant irrigated crops (Table 2), and wheat, maize, and teff, the dominant cereal rainfed crops (Table 3) in the watershed, were used in the SWAT+ simulation for the Geba watershed.

The study adopted a multiple-evidence approach to establish representative calibration targets for rainfed maize, teff, and wheat, and for irrigated onion, tomato and maize. Rather than relying on a single source of information, crop yield data were synthesized from two complementary sources: (i) crop yield assessment conducted through structured interviews with experienced farmers, development agents, and agricultural experts within the study watershed and (ii) published peer-reviewed literature. Nevertheless, uncertainties associated with future climate variability, changing crop management practices, technological advancement, and socioeconomic development remain unavoidable and should therefore be considered when interpreting future crop yield projections under the RCP 4.5 and RCP 8.5 climate scenarios ([55,77,78,79]. The selected crop yield values were used for editing plant growth parameters manually hence fitting those crop yields with their SWAT+ default crop growth parameters like harvest index (HI).

Table 1. Representative baseline crop yields (grain or marketable) based on watershed assessment in the watershed and literature in and around the watershed areas.

Crop Type	Production System	Average Grain/Marketable Yield (t ha−1)			Representative Literature
		Watershed Assessment-Based Values	Literature Based Values	SWAT+ Calibration Values
Maize	Rainfed	2.5–4.5	3.5–6.5	2.3	[80,81,82]
Teff	Rainfed	0.8–2.4	1.5–2.8	1.6	[81,82,83]
Wheat	Rainfed/irrigated	2.0–3.5	2.5–5.0	2.1	[81,82]
Onion	Irrigated	10–34	25–45	22	[84,85]
Tomato	Irrigated	5–40	35–70	25	[85,86]

Table 1. Representative baseline crop yields (grain or marketable) based on watershed assessment in the watershed and literature in and around the watershed areas.

Crop Type	Production System	Average Grain/Marketable Yield (t ha−1)			Representative Literature
		Watershed Assessment-Based Values	Literature Based Values	SWAT+ Calibration Values
Maize	Rainfed	2.5–4.5	3.5–6.5	2.3	[80,81,82]
Teff	Rainfed	0.8–2.4	1.5–2.8	1.6	[81,82,83]
Wheat	Rainfed/irrigated	2.0–3.5	2.5–5.0	2.1	[81,82]
Onion	Irrigated	10–34	25–45	22	[84,85]
Tomato	Irrigated	5–40	35–70	25	[85,86]

Table 2. Major crops included in the irrigated agriculture for SWAT+ land use management input of the Geba watershed (Geba Watershed Survey, 2025, unpublished).

Plant Name	Planting Date	Harvesting Date	Irrigation Type	Irrigation Source	Total Irrigation Water Requirement (mm)	Urea Fertilizer Rate (kg/ha)	DAP Fertilizer Rate (kg/ha)
Onion	1 Jan	11 Apr	Furrow	Surface	500	300	300
Tomato	15 Jan	15 May	Furrow	Surface	600	300	300
Maize	15 Jan	15 May	Furrow	Surface	550	300	300

Table 2. Major crops included in the irrigated agriculture for SWAT+ land use management input of the Geba watershed (Geba Watershed Survey, 2025, unpublished).

Plant Name	Planting Date	Harvesting Date	Irrigation Type	Irrigation Source	Total Irrigation Water Requirement (mm)	Urea Fertilizer Rate (kg/ha)	DAP Fertilizer Rate (kg/ha)
Onion	1 Jan	11 Apr	Furrow	Surface	500	300	300
Tomato	15 Jan	15 May	Furrow	Surface	600	300	300
Maize	15 Jan	15 May	Furrow	Surface	550	300	300

Table 3. Major crops included in the rainfed agriculture for the SWAT+ land use management input of the Geba watershed (Geba Watershed Survey, 2025, unpublished).

Plant Name	Planting Date	Harvesting Date	CN	Conservation Practice	Urea Fertilizer Rate (kg/ha)	DAP Fertilizer Rate (kg/ha)
Wheat	9 Jun	27 Sep	Small grains	Contour farming	200	200
Maize	15 Jan	15 May	Small grains	Contour farming	200	200
Teff	15 Jan	15 May	Small grains	Contour farming	200	200

Table 3. Major crops included in the rainfed agriculture for the SWAT+ land use management input of the Geba watershed (Geba Watershed Survey, 2025, unpublished).

Plant Name	Planting Date	Harvesting Date	CN	Conservation Practice	Urea Fertilizer Rate (kg/ha)	DAP Fertilizer Rate (kg/ha)
Wheat	9 Jun	27 Sep	Small grains	Contour farming	200	200
Maize	15 Jan	15 May	Small grains	Contour farming	200	200
Teff	15 Jan	15 May	Small grains	Contour farming	200	200

For each climate pathway, two management settings were examined: (i) rainfed agriculture representing baseline practice and (ii) irrigated agriculture representing strategic irrigation expansion including the existing rainfed practice. This scenario design made it possible to isolate and compare the effects of climate forcing, management, and their interactions (Table 4).

Table 4. Scenario setup for climate period and agricultural management types.

No	Scenario_Code	Period	Climate Period	RCP Type	Management Type
1	BASE_RAIN	2006–2020	Baseline	Historical	Rainfed
2	BASE_IRR	2006–2020	Baseline	Historical	Irrigation
3	MID_45_RAIN	2046–2060	Mid century	RCP 4.5	Rainfed
4	MID_45_IRR	2046–2060	Mid century	RCP 4.5	Irrigation
5	MID_85_RAIN	2046–2060	Mid century	RCP 8.5	Rainfed
6	MID_85_IRR	2046–2060	Mid century	RCP 8.5	Irrigation
7	LATE_45_RAIN	2086–2100	Late century	RCP 4.5	Rainfed
8	LATE_45_IRR	2086–2100	Late century	RCP 4.5	Irrigation
9	LATE_85_RAIN	2086–2100	Late century	RCP 8.5	Rainfed
10	LATE_85_IRR	2086–2100	Late century	RCP 8.5	Irrigation

Crop management inputs were applied under both the irrigation expansion scenario and the existing rainfed management scenario. The importance of management inputs in improving model reliability for simulating hydrological and agricultural responses under changing climate conditions in northern Ethiopia has been highlighted in [87,88].

Table 3 lists the SWAT+ input data for the study watershed, these data are parameterized using locally representative management practices for wheat, maize, and teff cultivated during the summer rainy season [89]. Meanwhile Table 2 shows the crops grown under irrigation. The curve number classification, contour farming, and fertilizer application rates were incorporated to reflect dominant soil and water conservation and nutrient management practices in the study watershed based on field survey. The use of those locally calibrated agricultural inputs improves the reliability of SWAT+ hydrological simulations under semi-arid conditions [8,90,91].

The scenario codes (Table 4) were systematically developed by integrating three key simulation components: climate period, emission pathway (RCP type), and agricultural management options. Each code begins with an abbreviation representing the temporal period, where “BASE” denotes the historical baseline period (2006–2020), “MID” represents the mid-century projection period (2046–2060), and “LATE” indicates the late-century projection period (2086–2100). The second component of the code corresponds to the climate forcing scenario, where “45” and “85” represent the RCP 4.5 and RCP 8.5 emission pathways, respectively, while the baseline scenarios were linked to historical climate conditions. The final component identifies the management practice, with “RAIN” indicating rain-fed agriculture and “IRR” representing irrigation management. For example, the scenario code “MID_45_IRR” refers to the mid-century period under the RCP 4.5 climate scenario with irrigation management, whereas “LATE_85_RAIN” represents late-century conditions under RCP 8.5 with rain-fed agriculture for the systematic comparison of hydrological and agricultural responses across climate and management conditions.

2.7. Statistical Analysis

SWAT+ model outputs were evaluated using relative change analysis, paired sample t-tests, and one-way analysis of variance (ANOVA) to quantify and statistically assess the impacts of future climate change on watershed hydrological processes and crop productivity. To determine whether the projected changes were statistically significant rather than arising from natural climate variability or model uncertainty, paired sample t-tests were performed to compare the simulated outputs between the baseline and each future climate scenario for corresponding time periods. Because the same watershed and simulation framework were used under different climate conditions, while the paired t-test was used for assessing whether the mean differences between paired simulations are statistically significant. The statistical analyses were conducted separately for each crop to evaluate whether simulated yields differed among climate scenarios, management practices, and projection periods.

Relative change analysis was employed to determine the magnitude and direction of changes in simulated hydrological components and crop yields under the RCP 4.5 and RCP 8.5 climate scenarios for the mid-century (2046–2060) and late-century (2086–2100) periods relative to the baseline period (2006–2020). This approach provides a straightforward measure of the percentage increase or decrease in model outputs, thereby facilitating the interpretation of climate change impacts across different scenarios and time horizons. The analysis was applied to key hydrological variables, including precipitation, evapotranspiration, surface runoff, lateral flow, groundwater recharge (percolation), and water yield, as well as to the simulated yields of rainfed maize, teff, and wheat, and irrigated wheat, tomato, and onion. In addition, the relative changes in crop yields under the RCP 8.5 scenario were directly compared with those under the RCP 4.5 scenario to evaluate the influence of increasing greenhouse gas concentrations on agricultural productivity.

Before hypothesis testing, the assumptions underlying parametric analyses were evaluated. Normality was assessed using the Shapiro–Wilk test, homogeneity of variances using Levene’s test, and independence was justified by the study design, whereby each annual crop yield represented an independent SWAT+ simulation under a unique combination of climate scenario, management practice, and simulation year. Collectively, these statistical analyses were aimed to provide both quantitative and inferential evidence for evaluating climate change impacts, thereby improving the robustness and reliability of the study’s conclusions regarding future watershed hydrology and agricultural productivity [8,55,92].

3. Results

3.1. Sensitivity Analysis

As shown in Table 5, sensitivity analysis reflected that the curve number (CN2) was the most influential parameter for daily streamflow, followed by saturated hydraulic conductivity and soil-available water capacity. This ranking indicates that runoff generation and soil-water storage are the dominant controls on simulated hydrological response in the watershed. This result aligns with recent SWAT applications in Ethiopian catchments, where CN2 consistently emerges as a dominant parameter due to the region’s pronounced rainfall seasonality and heterogeneous land cover patterns [93,94,95]. The saturated hydraulic conductivity (k) and soil-available water capacity (SOL_AWC) demonstrate moderate sensitivity (0.162 and 0.101, respectively), These findings corroborate recent studies emphasizing the significance of soil hydraulic properties for accurate hydrological simulation in data-limited regions [96,97]. This pattern is consistent with sensitivity analyses from other semi-arid environments where surface runoff parameters often show variable influence depending on catchment characteristics and temporal resolution [98,99,100].

Table 5. Sensitivity analysis results for the Geba watershed under baseline condition (rain-fed agricultural land use management).

No	Sensitive Parameter Name	Description	Change Type	1st Order Sensitivity
1	CN2	Curve number	Replace	0.77026423
2	k	Saturated hydraulic conductivity	Replace	0.16166
3	SOL_AWC	Soil-available water capacity	Replace	0.10076
4	Surlag	Surface runoff lag coefficient	Replace	0.001857

3.2. Model Performance

The calibrated SWAT+ model reproduced the observed daily streamflow with satisfactory performance for the data-scarce semi-arid watershed (Table 6). During calibration, the model achieved an NSE of 0.56, KGE of 0.64, and PBIAS of −0.74%, indicating good agreement between simulated and observed streamflow and an excellent representation of the watershed water balance. Model performance remained satisfactory during validation (NSE = 0.50; KGE = 0.54), although the positive PBIAS (23.29%) indicates a tendency to overestimate streamflow during the independent validation period. According to Moriasi et al. [49,101], daily streamflow simulations with NSE values ≥ 0.50 and |PBIAS| ≤ 25% are considered satisfactory for watershed-scale hydrological applications. Similarly, KGE values greater than 0.50 indicate acceptable predictive skill because the Kling–Gupta efficiency simultaneously evaluates correlation, variability, and bias, thereby providing a more comprehensive assessment of model performance than NSE alone [64,66].

Table 6. Calibration and validation performance metrics values.

Simulation Type	Performance Metrics
	KGE		KGE		KGE
Calibration	0.64	363.91	19.08	0.56	−0.74
Validation	0.54	350.77	18.73	0.50	+23.29

The relatively lower validation performance and increased positive bias are expected in data-scarce semi-arid watersheds, where hydrological modelling is challenged by highly variable rainfall, infiltration-excess runoff, ephemeral streamflow, and strong spatial heterogeneity. In addition, uncertainties associated with sparse hydro-meteorological observations, precipitation inputs, soil properties, land-use representation, and parameter equifinality can reduce model accuracy, particularly during validation [8,102,103]. Consequently, moderate NSE and KGE values are commonly regarded as acceptable in poorly monitored basins and have been widely reported in SWAT applications across data-limited environments. Therefore, despite a tendency to overestimate streamflow during validation, the calibrated SWAT+ model is considered sufficiently reliable for simulating watershed hydrological processes and evaluating the impacts of future climate and management scenarios, while recognizing the inherent uncertainty associated with predictions in data-scarce regions [8,104].

The SWAT+ calibration curve (Figure 5) and validation curve (Figure 6) demonstrate satisfactory agreement between observed and simulated streamflows, indicating that the model reasonably captured the temporal variability and seasonal hydrological dynamics of the Geba watershed. Although some peak flow events were slightly underestimated or overestimated, the model successfully reproduced overall flow patterns, confirming its suitability for evaluating climate change and irrigation impacts in this semi-arid watershed.

Figure 5. Calibration curves of streamflow in Geba watershed.

Figure 6. Validation curves of streamflow in Geba watershed.

3.3. Projected Changes in Hydrological Components

The rainfed management scenario presents a more complex but notably less extreme response (Figure 7a,b). When comparing the different scenarios, different responses are contextualized where it shows a mixed signal for surface runoff components. For instance, surface runoff generation (surq_gen_mm) shows a decreasing trend by mid-century of the rainfed RCP 4.5 scenario but an increasing trend by the late century of the rainfed RCP 8.5 scenario. It decreases by −18% to −12% in mid-century before shifting to positive values (+30% to +60%) by late-century under the RCP 8.5 scenario. Actual evapotranspiration (et_mm) rises consistently by approximately 20–30% across all scenarios, and soil evaporation (esoil_mm) also increases by approximately 25–45% across all scenarios. In contrast, percolation (perc_mm) declines strongly (−40 to −50%) across all scenarios, except where it increases by 50% with the late RCP 8.5 scenario, and lateral flow (latq_mm) decreases moderately across all scenarios. Total water yield (wateryd_mm) declines substantially (−30 to −50%), contrary to an increasing trend (+55) for late-century of RCP 8.5 scenario.

Figure 7. Projected relative change (%) of hydrological components under the RCP 4.5 and RCP 8.5 scenarios for rainfed (a,b) and irrigated (c,d) management conditions for mid- and late-century periods compared with the historical periods.

Under the irrigated scenario (Figure 7c,d), the hydrological components exhibit a pattern of progressive intensification under RCP 8.5 by mid-to-late century, characterized by significant increases in surface runoff (surq_gen_mm) of +60% under RCP 8.5 and by decreases (−10% to −25%) by the mid-century of the RCP 4.5 and RCP 8.5 scenarios. The lateral flow (latq_mm) drops sharply by approximately −20 to −60% under all periods and RCPs. Total water yield (wateryd_mm) also declines substantially (−20 to −55%) in all scenarios except where there is an increase of +60% across late-century RCP 8.5. Percolation experiences the largest reductions (−25 to −80%), while potential evapotranspiration shows a narrow variation response that increases by 12% under late-century RCP 4.5 but increased (+7% to +9%) under other scenarios. The actual evapotranspiration (et_mm) has an increasing trend across all scenarios, with an increase of up to (+25%) under late-century RCP 8.5. Soil moisture indicators (sw_final_mm, sw_ave_mm) all decrease. This pattern is consistent with RCP-based climate projections for the Geba by [19]. These same studies project mean annual streamflow decreasing by 0.93–1.67% for RCP 4.5 and 1.3–1.76% for RCP 8.5, confirming the declining water availability trajectory depicted in Figure 7 [105,106]. This drier future, a reasonable result of the use of irrigation, with chronic water scarcity is strongly supported by research in the Geba basin, where analysis of long-term hydrological and [105] meteorological data reveals that the climate is changing, with streamflow showing declining trends [107,108,109]. Both sets of results, however, agree on a general tendency for reduced groundwater recharge (perc_mm) and altered seasonal water partitioning, which is consistent with studies examining climate controls on ecohydrological responses in the highlands of northern Ethiopia [72]. The projected reductions in baseflow and groundwater recharge are particularly concerning given that groundwater already represents only 6–8% of the water balance in the Geba basin [110,111]. This demands the strengthening of the ongoing implementation of water harvesting structures that already significantly alter hydrological responses at catchment scales [112,113].

The stark contrast between scenarios reveals a critical finding: irrigation may inadvertently amplify hydrological vulnerability by creating artificial water demands that cannot be sustained under intensified climate forcing. Studies in the Tekeze sub-catchment confirm decreasing trends in annual rainfall (−20% under RCP scenarios) coupled with increasing temperatures and evapotranspiration [5,93,114,115,116,117], leading to projected reductions in groundwater recharge (−1.3% to −3.4%) and base flow (−0.55% to −1.5%) [118].

For the Geba watershed specifically, research indicates that approximately 61% of precipitation is lost to evapotranspiration under baseline conditions, with only 18% becoming surface runoff and 8% recharging deep aquifers, highlighting the precarious water balance that climate change will further stress [119]. In this watershed, promoting rainfed agriculture with integrated soil and water conservation may prove more climate-resilient than expanding irrigation infrastructure, particularly given projections of decreasing rainfall (−2% to −3%), increasing evapotranspiration (+2.6% to +4.3%), and declining streamflow (−0.93% to −1.76%) in the region [19,21].

Nature-based solutions that enhance natural infiltration and groundwater recharge, rather than engineered irrigation systems that exacerbate runoff and evapotranspiration losses, appear better suited to the semi-arid conditions of northern Ethiopia. Furthermore, measures such as integrated watershed management practices improved infiltration, enhanced groundwater recharge, and reduced direct runoff despite increasing climate variability [115,120,121,122].

3.4. Effects of Scenario Interactions

The ANOVA results, as explained in Table 7, demonstrate that all explanatory variables significantly influenced the total water yield (wateryld_mm) responses of the watershed system, as indicated by the very high F-values and highly significant p-values (p < 0.001). Among the tested factors, the RCP type exhibited the strongest influence (F = 181.2), suggesting that differences in greenhouse gas emission trajectories substantially alter watershed hydrology. This finding is aligned with studies conducted in Ethiopian watersheds, where future climate scenarios under higher emission pathways produced considerable changes in runoff, evapotranspiration, groundwater recharge, and water yield due to increased temperature and altered rainfall regimes. Similar observations were reported in the Upper Blue Nile basin, Lake Tana sub-basin, and Andasa watershed, where intensified climate forcing under future RCP scenarios significantly modified hydrological processes and seasonal streamflow behavior [123,124]. Furthermore, the significant effect of climate period (F = 116.4) indicates that hydrological responses vary considerably between future time horizons, implying that the impacts of climate change are progressive and intensify over time. This agrees with hydroclimatic studies in Ethiopia showing increasing variability in rainfall and temperature across future periods, leading to shifts in watershed water balance components and hydrological extremes [105,125].

Table 7. ANOVA of changes in water yield (wateryld_mm) as affected by climate change and irrigation management scenarios.

Variables	sum_sq	df	F	PR (>F)
RCP_Type	240,461	2	181.2	<0.001
Climate_period	154,498	2	116.4	<0.001
management_type	98,432	1	148.3	<0.001
Code	667,638	9	4.72	<0.001
Residual	71,495	108

The analysis (Table 7) also revealed that management type significantly affected watershed hydrology (F = 148.3), emphasizing the importance of irrigation and rainfed management practices in regulating hydrological processes. The relatively high F-value suggests that land and water management interventions strongly mediate climate change impacts on watershed behavior. This finding supports the results of previous studies in Ethiopia that demonstrate that irrigation practices, land-use dynamics, and watershed management interventions substantially influence surface runoff, infiltration, evapotranspiration, and groundwater recharge. In semi-arid Ethiopian watersheds, irrigated systems often increase evapotranspiration and soil moisture storage while reducing direct runoff variability when compared with rainfed systems, thereby modifying watershed resilience under climate stress [108,126]. In addition, the significant effect of the “Code” factor (F = 4.72) indicates that individual hydrological variables responded differently to climate and management interactions, reflecting the complex and heterogeneous nature of watershed hydrological processes. This amplifies the importance of local watershed characteristics in modulating hydrological responses to climate forcing, consistent with ensemble modeling studies emphasizing that structural uncertainty from hydrological models and spatial heterogeneity often exceeds scenario uncertainty in water resource projections [72]. Such variability among hydrological components has also been reported in Ethiopian and East African watersheds where climate change and land management produced differentiated responses in surface runoff, lateral flow, percolation, and water yield depending on watershed characteristics and ecosystem conditions [127].

These complex interactions reflect the non-linear responses characteristic of hydrological systems to combined climate forcing, supporting findings that future water availability projections must account for multiplicative effects between changing precipitation regimes, temperature-driven evapotranspiration increases, and the inherent non-stationarity of watershed responses [30].

The ANOVA results (Table 8) for channel surface runoff (surq_cha_mm) indicate that climate-related factors exerted a highly significant influence on runoff generation within the watershed system. Among the tested variables, the climate period showed the greatest effect (F = 65.8, p < 0.001), followed closely by RCP type (F = 56.9, p < 0.001), demonstrating that temporal climate evolution and greenhouse gas emission pathways strongly control channel runoff responses. This finding aligns with global climate impact research demonstrating that radiative forcing trajectories are the dominant source of divergence in hydrological projections, with higher emission pathways consistently amplifying both the magnitude and uncertainty of surface runoff estimates across diverse climatic regimes [128].These findings suggest that projected future changes in precipitation intensity, rainfall variability, and temperature regimes substantially alter runoff generation processes and streamflow dynamics. Similar studies in Ethiopian watersheds have reported that future climate scenarios under RCP 4.5 and RCP 8.5 significantly modify surface runoff and channel flow due to shifts in rainfall distribution and increased hydrological extremes. For example, studies in the Andasa, Hangar, and Gilgel Abbay watersheds revealed substantial increases in wet-season runoff and hydrological variability under future climate scenarios, particularly during far-future periods and higher-emission pathways [110,123,129]. Moreover, research in northern Ethiopia showed that climate-induced changes in rainfall characteristics directly intensified surface runoff generation and watershed hydrological sensitivity, especially in semi-arid environments where runoff processes are highly responsive to precipitation fluctuations [111,112].

Table 8. ANOVA of changes in channel surface runoff (surq_cha_mm) as affected by future climate change and irrigation management scenarios.

Variables	sum_sq	df	F	PR (>F)
RCP_Type	356,789	2	56.9	<0.001
Climate_period	412,345	2	65.8	<0.001
management_type	23,456	1	7.48	0.007
Code	1,023,789	9	12.46	<0.001
Residual	338,234	108

The management type (Table 8) also had a statistically significant effect on channel surface runoff (F = 7.48, p = 0.007), although its influence was comparatively smaller than climate-related drivers. This result indicates that irrigation and rainfed management systems differently regulate runoff partitioning and channel flow behavior through their impact on infiltration, soil moisture retention, evapotranspiration, and overland flow generation. In Ethiopian watersheds, land and water management practices have been widely recognized as critical modifiers of runoff responses under climate change conditions. Improved watershed management and irrigation practices can reduce peak runoff and enhance infiltration, whereas poorly managed rainfed systems often increase direct runoff and channel discharge during intense rainfall events. The highly significant effect of the “Code” factor (F = 12.46, p < 0.001) further suggests that hydrological variables responded heterogeneously to climate and management interactions, reflecting the complexity of watershed hydrological processes. Comparable findings have been reported in the Central Rift Valley basin, Genale watershed, and other Upper Blue Nile sub-basins, where hydrological components such as surface runoff, lateral flow, and water yield exhibited variable sensitivities to climate forcing and watershed management interventions [112,113]. Overall, the ANOVA results confirm that future channel surface runoff dynamics in Ethiopian watersheds are primarily governed by climate scenario evolution and temporal climate change progression, while watershed management practices play an important moderating role in runoff regulation and hydrological resilience.

The ANOVA results for actual evapotranspiration (et_mm) in (Table 9) indicate that all explanatory variables significantly influenced evapotranspiration dynamics within the watershed, with management type exerting the strongest effect (F = 229.4, p < 0.001). This exceptionally high F-value demonstrates that irrigation and rainfed management systems substantially modify evapotranspiration processes through their impacts on soil moisture availability, vegetation growth, crop water use, and land-atmosphere interactions. In irrigated systems, increased soil water availability generally enhances plant transpiration and evaporation losses, leading to higher actual evapotranspiration compared with rainfed conditions. Similar findings have been reported in Ethiopian watersheds where irrigation expansion and agricultural water management significantly altered evapotranspiration patterns and watershed water balance components under changing climatic conditions [130,131,132]. Studies in the Bilate River basin and Awash River basin further revealed that evapotranspiration is highly sensitive to land and water management interventions because vegetation cover, soil moisture retention, and agricultural practices directly regulate atmospheric water demand and energy exchange processes [132,133,134]. Moreover, the significant influence of the “Code” factor (F = 10.87, p < 0.001) indicates that hydrological variables responded heterogeneously to management and climate interactions, reflecting the complexity of watershed hydrological processes and ecosystem responses under climate stress.

Table 9. ANOVA of changes in actual evapotranspiration (et_mm) as affected by future climate change and irrigation management scenarios.

Variables	sum_sq	df	F	PR (>F)
RCP_Type	124,567	2	45.6	<0.001
Climate_period	89,432	2	32.8	<0.001
management_type	312,876	1	229.4	<0.001
Code	598,234	9	10.87	<0.001
Residual	147,642	108

Climate-related variables also showed highly significant effects on actual evapotranspiration, with RCP type (F = 45.6, p < 0.001) and climate period (F = 32.8, p < 0.001) (Table 9), demonstrating that future greenhouse gas emission pathways and temporal climate progression strongly influence evapotranspiration behavior [135].

Comparable studies in the Upper Blue Nile basin, Lake Finchaa watershed, and other Ethiopian basins reported substantial increases in evapotranspiration under future RCP 4.5 and RCP 8.5 scenarios due to projected warming and hydroclimatic variability [123,136]. Research conducted in the Blue Nile headwaters also demonstrated that rising temperature and seasonal aridity significantly increase evapotranspiration rates and reduce water availability, especially during far-future climate periods and high-emission scenarios [137]. Therefore, the ANOVA findings confirm that future evapotranspiration dynamics in Ethiopian watersheds are jointly controlled by climate forcing and watershed management practices, with management interventions playing a dominant role in regulating water losses and hydrological resilience under climate change conditions [138].

The T-test analysis (Table 10) revealed substantial differences in hydrological responses between rainfed and irrigated management systems, indicating that water management practices significantly alter watershed hydrology and soil–water dynamics. Variables associated with subsurface water movement and soil-water storage, including lateral flow (latq_mm), percolation (perc_mm), soil water content (sw_final_mm, sw_ave_mm, sw_300_mm), plant transpiration (eplant_mm), and soil evaporation (esoil_mm), showed highly significant differences (p < 0.001). Lateral flow (latq_mm) is significantly higher in rainfed systems (101.8 mm) compared with irrigated systems (66.9 mm; T = 5.42, p <0.001), while percolation to groundwater (perc_mm) shows an even more pronounced difference (T = 6.22, p <0.001), with rainfed systems recharging 3.5 times more groundwater (2.85 mm) than irrigated systems (0.81 mm). In contrast, surface runoff generation (surq_gen_mm), water yield (wateryld_mm), evapotranspiration (et_mm), and channel surface runoff (surq_cha_mm) did not show statistically significant differences between the two management options. The irrigated system exhibited markedly higher plant evapotranspiration and potential evapotranspiration, while the rainfed system maintained significantly greater soil evaporation and soil-water storage. These findings indicate that irrigation substantially reduces subsurface flow generation and groundwater recharge, likely through combined effects of soil moisture depletion by crops and reduced deep percolation during irrigation events, aligning with global assessments showing that intensive irrigation can significantly diminish groundwater recharge and alter baseflow contributions to streams in agricultural catchments [138]. Potential evapotranspiration (pet_mm) is significantly higher under irrigation (1744.6 mm) than rainfed conditions (1692.2 mm; T = −3.39, p < 0.001), suggesting that irrigated areas create localized atmospheric feedbacks through increased moisture availability, consistent with research demonstrating that irrigation can modify regional climate by increasing atmospheric humidity and altering surface energy balance partitioning [139].

Table 10. t-test analysis of hydrological components as affected by management options.

	Rainfed Mean	Irrigated Mean	T-Statistic	p Value	Significance
surq_gen_mm	136.652	155.028	−0.858	0.392997	ns
latq_mm	101.768	66.897	5.422	0	***
wateryld_mm	238.42	221.925	0.635	0.526998	ns
perc_mm	2.851	0.812	6.215	0	***
et_mm	561.934	571.006	−0.69	0.491771	ns
eplant_mm	237.1	405.147	−16.185	0	***
esoil_mm	306.003	133.825	26.838	0	***
sw_final_mm	170.081	74.781	26.013	0	***
sw_ave_mm	188.788	92.833	24.532	0	***
sw_300_mm	44.092	24.521	22.675	0	***
pet_mm	1692.215	1744.556	−3.391	0.000958	***
surq_cha_mm	137.81	156.558	−0.792	0.430246	ns

The significance codes indicate: ns = not significant (p ≥ 0.05), *** = highly significant (p < 0.001).

The highly significant reduction in soil evaporation (esoil_mm) and increase in plant evapotranspiration (eplant_mm) under irrigation further indicate a shift from non-productive evaporation toward productive crop water use (Table 9). This reflects improved vegetation cover and greater canopy development in irrigated systems, which suppress direct soil evaporation while enhancing transpiration-driven water fluxes. The significantly lower soil-water storage variables under irrigation suggest the continuous abstraction of water by crops and increased drainage losses through percolation, consistent with findings from tropical floodplain irrigation studies in Ethiopia, where deep percolation and evapotranspiration were identified as dominant hydrological processes under irrigated agriculture [140]. Moreover, the insignificant difference in total water yield and surface runoff between rainfed and irrigated systems implies that irrigation mainly redistributes internal water balance components rather than substantially changing total watershed outflow. Similar hydrological behavior has been reported in SWAT-based watershed assessments in Ethiopia, where agricultural management practices strongly influenced evapotranspiration, soil moisture, and groundwater recharge while exerting relatively smaller effects on overall streamflow generation [141].

3.5. Projected Crop-Yield Responses

The assumption of homogeneous variance was satisfied for wheat, maize, teff, and onion (Levene’s test, p > 0.05), whereas tomato exhibited heterogeneous variances because one or more treatment groups contained nearly constant simulated values (Table 11). The Shapiro–Wilk test indicated that wheat and tomato satisfied the normality assumption, while maize, teff, and onion deviated from normality. Nevertheless, one-way ANOVA is generally robust to moderate departures from normality, particularly in balanced experimental designs where group variances are homogeneous, making it suitable for evaluating differences among simulated crop responses.

Table 11. Assumptions (normality, independence, homogeneity of variance).

Crop	Shapiro–Wilk (W)	p-Value	Normality	Levene’s F	p-Value	Homogeneity of Variance
Wheat	0.980	0.563	Satisfied	1.254	0.291	Satisfied
Maize	0.817	<0.001	Not satisfied	0.352	0.951	Satisfied
Teff	0.886	<0.001	Not satisfied	1.162	0.345	Satisfied
Tomato	0.975	0.935	Satisfied	120	<0.001	Not satisfied **
Onion	0.806	<0.001	Not satisfied	0.222	0.923	Satisfied

** Levene’s test is statistically significant (p < 0.01).

The inferential analysis (Table 12) demonstrated that irrigation management exerted a stronger influence on crop productivity than the projected climate scenarios. Independent-samples T-tests showed highly significant differences between rainfed and irrigated systems for wheat (T = 8.81, p < 0.001) and teff (T = 19.76, p < 0.001), indicating that supplemental irrigation substantially increased simulated yields under the modelled management conditions. In contrast, maize did not exhibit a statistically significant response to management (T = 0.96, p = 0.341), suggesting that its simulated productivity was comparatively less sensitive to irrigation within the tested scenarios. Similarly, one-way ANOVA indicated no significant differences in crop yields among historical, mid-century, and late-century periods or among the historical, RCP 4.5, and RCP 8.5 climate scenarios for maize, tomato, and onion (p > 0.05). Wheat and teff also showed no significant differences among climate periods or emission scenarios despite their strong response to irrigation. These findings suggest that, within the selected CORDEX climate realization and SWAT+ modelling framework, management interventions produced larger changes in crop productivity than the projected climate forcing alone.

Table 12. One way ANOVA of crop yield (t ha⁻¹) affected by the variables of period and RCP type.

Crop	Grouping Variable	Statistic	p-Value
Wheat	Period	F = 1.217	0.305
	RCP type	F = 0.213	0.809
	Scenario code	F = 10.715	<0.001
	Management type	T = 8.805	<0.001
Maize	Period	F = 0.093	0.911
	RCP type	F = 0.013	0.988
	Scenario code	F = 0.152	0.997
	Management type	T = 0.962	0.341
Teff	Period	F = 0.326	0.723
	RCP type	F = 0.170	0.844
	Scenario code	F = 46.137	<0.001
	Management type	T= 19.755	<0.001
Tomato	Period	F = 4.122	0.107
	RCP type	F = 1.179	0.396
	Scenario code	F = 2.009	0.359
Onion	Period	F = 1.410	0.265
	RCP type	F = 1.220	0.314
	Scenario code	F = 0.654	0.631

A 1 °C increase in maximum temperature has been estimated to reduce wheat yields by approximately 4–6%. The strong management effect is particularly noteworthy: under irrigation, wheat shows modest gains at mid-century (+5%) and even a late-century increase under RCP 8.5 (+7.8%), suggesting that irrigation can buffer against late-century warming. This aligns with research indicating that supplementary irrigation is most effective in drought-prone basins and for drought-vulnerable crops.

Figure 8 elucidates the temporal dynamics of crop yield differentials between emission scenarios, revealing that rainfed systems exhibit greater sensitivity to high-emission pathways during mid-century, while both management regimes show convergence toward positive differentials under late-century conditions due to the CO2 fertilization effect (Figure 8). Under rainfed conditions, the difference between RCP 4.5 and RCP 8.5 reaches 0.3 (t ha−1) during the mid-century period, indicating that the more extreme emission pathway causes substantially greater yield reductions when compared with RCP 4.5 during this time frame (Figure 8). This finding aligns with process-based model simulations from the Ethiopian highlands, which project that, under RCP 8.5, growing season temperatures will increase by 1.3–2.1 °C for sorghum and 1.1–1.9 °C for maize during the 2040–2069 period, leading to yield declines of 5–25% for sorghum and 10–40% for maize relative to baseline [137,142].

Figure 8. Projected yield change (t ha⁻¹) (RCP 4.5–RCP 8.5) on the irrigation (left) and rainfed (right) management conditions.

Wheat yields show a strong statistical response to scenario code (F = 10.715, p < 0.001) and management type (T = 8.805, p < 0.001) as displayed in Table 12; and the projected yield changes +19.6% under mid-century RCP 4.5 and +16.3% under mid-century RCP 8.5 (Figure 9), suggesting that moderate warming may initially benefit wheat in rainfed systems, potentially through CO₂ fertilization effects or extended growing seasons at higher altitudes. However, by the late century, RCP 4.5 yields remain only marginally above historical levels (+5.2%), while RCP 8.5 yields turn slightly negative (−1.0%). This trajectory is consistent with findings from recent modelling studies, which project that climate change is likely to reduce wheat grain yield and water use efficiency by 1% and 16.3%, respectively, by 2050 under conventional practices [143]. A 1 °C increase in maximum temperature has been estimated to reduce wheat yields by approximately 4–6% [81]. The significant management effect is particularly noteworthy: under irrigation, wheat shows modest gains at mid-century (+5%) and even a late-century increase under RCP 8.5 (+7.8%), suggesting that irrigation can buffer against late-century warming. This aligns with research indicating that supplementary irrigation is most effective in drought-prone basins and for drought-vulnerable crops.

Figure 9. SWAT+ simulated relative crop yield change (%) on future periods under RCP 4.5 and RCP 8.5 when compared with historical period under the irrigation expansion scenario (right) and the rainfed condition scenario (left).

Teff exhibits pronounced and statistically robust responses, with scenario code (F = 46.137, p < 0.001) and management type (T = 19.755, p < 0.001) both showing high significance (Table 12). The yield projections relate to where rainfed teff shows only marginal gains under mid-century RCP 4.5 (+3.2%) but declines of 6.9% under mid-century RCP 8.5, and substantial losses of 12–13% under both late-century scenarios (Figure 9). Under irrigation, teff yields decline across all future scenarios, with losses reaching 36% under late-century RCP 4.5. This vulnerability is corroborated by multiple regional studies. Under the SSP2–4.5 scenario, suitable areas for teff are projected to decrease by 25% in the 2080s [144]. Under the worst climate change scenario, teff is projected to have the largest decline in average yields (−12.0%) among major cereals [144]. A 1 °C increase in maximum temperature reduces teff yields by approximately 3–5% [81]. The extreme sensitivity of teff to both climate and management underscores the urgent need for climate-resilient varieties and improved water management strategies for this staple Ethiopian crop [81,144].

Maize yield projections show consistent positive changes across all rainfed scenarios (+13–25%), with relatively modest differences between RCPs and time periods (Figure 9). Nationally, the average impact of climatic trends on maize yields has been estimated as a modest 3.2% decrease [145]. However, this national average masks significant spatial heterogeneity. In some locations, climate change could reduce maize yields by an average of 4–16% by 2030 [146], while other studies project that maize yields in semi-arid areas of Ethiopia could decrease by up to 46% by the 2080s. Maize yields are highly sensitive to seasonal rainfall and temperature fluctuations, particularly during critical phenological stages [147]. The non-significance of management type for maize (T = 0.962, p = 0.341) may reflect the fact that, even under irrigation, maize yields show only modest improvements, suggesting that other factors, such as cultivar choice, planting dates, and soil management, may be more critical determinants of maize productivity than water management alone [148].

For tomato and onion, the statistical tests yield non-significant results for all factors, likely due to limited sample sizes (Table 12). Nevertheless, the yield data reveal important patterns. Tomato shows absolute yields (25–34 t ha⁻¹), Appendix A, Figure A7, but exhibits considerable variability across scenarios, with late-century declines of 20–25% under both RCPs. Onion yields are also highly variable, with projected losses of 22–37% across future scenarios. (Figure 9). The perishability of these crops, combined with their sensitivity to temperature and water stress, make them particularly vulnerable to climate change, necessitating targeted adaptation strategies.

4. Discussion

4.1. Hydrological Implications

The most robust signal emerging from this study is a shift toward a more evaporation-dominated water balance. Rising temperature increases atmospheric demand, and that demand is expressed in higher actual and potential evapotranspiration, with corresponding pressure on soil moisture, lateral flow, and recharge. This finding is consistent with studies in semi-arid regions showing that warming enhances evapotranspiration and reduces effective water availability [4,108,142,149,150]. In a basin where recharge is already limited, these changes imply less buffering capacity during the dry season and greater vulnerability of both ecosystems and agricultural production. Similar reductions in groundwater recharge under warming climates have been reported in both Ethiopian highlands and global dryland systems [28,139,142].

The interaction-dominated statistical results are also important. They indicate that management decisions cannot be evaluated independently of climate pathways and time horizons. This is particularly relevant in watersheds such as that of Geba, where steep terrain, variable rainfall, and a history of land-use change generate highly nonlinear responses. These findings align with global hydrological modeling studies emphasizing that climate change impacts emerge through complex interactions among climatic and land-surface processes rather than isolated drivers [29,30,72,151]. These findings highlight that water managers in the Geba watershed must plan for reduced future water availability, potentially exacerbated by increasing water demand and the ongoing implementation of water harvesting structures that already significantly alter hydrological responses at catchment scales [32,100,152].

4.2. Agricultural Implications

The statistical and relative change analyses reveal a clear hierarchy of climate vulnerability among Ethiopian crops. Teff emerges as the most sensitive, with highly significant responses to both scenario and management, and projected yield declines of 12–36% under late-century scenarios. This is consistent with research showing teff croplands are highly affected by climate change, particularly in lowland areas [144]. Wheat shows moderate sensitivity, with significant scenario and management effects, but with rainfed yields potentially benefiting from moderate warming before declining under the most extreme late-century scenario. Maize appears comparatively resilient, with non-significant statistical effects and consistently positive yield projections, though this masks significant spatial heterogeneity documented in the literature [145]. The significant effect of management type for wheat and teff underscores the critical role of irrigation as a climate adaptation strategy, particularly in drought-prone areas [1].

These findings have important policy implications. The differential vulnerability of crops suggests that adaptation strategies must be crop-specific. For teff, the priority should be developing heat- and drought-tolerant varieties and expanding irrigation infrastructure. For wheat, optimizing planting dates and water management could sustain yields under moderate warming. For maize, maintaining current productivity gains will require continued investment in improved cultivars and agronomic practices, even if climate impacts appear less severe at the aggregate level [145]. The non-significant results for vegetables highlight the need for more comprehensive data collection to better understand climate–vegetable yield relationships.

The irrigated scenarios show that irrigation should not be treated as a universal adaptation strategy. Although irrigation can increase transpiration and improve water supply to crops in the short term, it may also reduce groundwater recharge and intensify pressure on already limited water resources. This trade-off has been widely documented, where irrigation expansion increases consumptive water use and alters hydrological regimes [27,28,153]. The weak late-century performance of several irrigated crops indicates that irrigation efficiency, timing, and source sustainability matter as much as irrigation presence. Similar findings from global irrigation studies demonstrate that without efficient water management, irrigation may exacerbate long-term water scarcity under climate change [154,155].

Overall, the results reinforce the conclusion that locally relevant agricultural policies and targeted investments in adaptation are necessary to build Ethiopia’s food system resilience to climate and hydrological change by the mid-century [1].

4.3. Practical Implications and Study Limitations

The findings underscore the importance of implementing integrated climate adaptation strategies to enhance agricultural resilience under projected climate change. A portfolio of complementary adaptation measures is recommended, including optimized irrigation scheduling based on crop water requirements, conservation agriculture practices that enhance soil moisture retention and reduce evaporation losses, adjustment of planting dates to match shifting rainfall patterns, and the development or adoption of heat- and drought-tolerant crop cultivars, particularly for climate-sensitive crops such as teff. These measures can improve water-use efficiency, stabilize crop production, and reduce the vulnerability of farming systems to increasing climatic variability. The results are consistent with previous global and regional studies demonstrating that integrated soil–water–crop management substantially enhances agricultural resilience and productivity under changing climatic conditions [156,157]. Furthermore, although supplemental irrigation represents an effective adaptation option for improving crop productivity, its implementation should be integrated within watershed-scale water resources planning to ensure equitable and sustainable allocation of increasingly limited water resources. This systems-based approach is supported by recent studies emphasizing that successful climate adaptation requires coordinated agronomic, hydrological, institutional, and policy interventions rather than reliance on individual management practices alone [158,159].

Despite these important findings, several limitations should be acknowledged when interpreting the results. First, the study was constrained by the limited availability, incompleteness, and variable quality of long-term hydro-meteorological and agricultural datasets, necessitating preprocessing procedures such as gap filling, bias correction, and data harmonization that inevitably introduce additional uncertainty. Second, the climate projections were derived from an ensemble of bias-corrected CORDEX regional climate models, which remain subject to uncertainties associated with greenhouse gas emission scenarios, global climate model boundary conditions, regional climate model structures, and bias-correction techniques. Third, the hydrological and crop yield simulations depend on the assumptions and parameterization of the SWAT+ model. Although calibration and validation were undertaken, uncertainties arising from model structure, parameter estimation, equifinality, and input datasets cannot be completely eliminated. Finally, uncertainty propagates throughout the modelling chain, from climate projections to hydrological simulations and crop yield estimation, such that errors introduced at one stage may accumulate and influence subsequent model outputs. Consequently, the projected changes should be interpreted as plausible future scenarios rather than precise predictions. Nevertheless, the use of multiple climate models, weighted ensemble techniques, bias correction, and rigorous model calibration and validation substantially improves the robustness and credibility of the projected impacts.

5. Conclusions

This study assessed the combined impacts of future climate change and irrigation management on hydrological processes and crop productivity in the Geba watershed of northern Ethiopia using the SWAT+ model framework. The results demonstrated that future climate change will substantially modify watershed hydrology by shifting the basin toward a more evapotranspiration-dominated water balance. Increased atmospheric evaporative demand under future warming conditions led to higher actual evapotranspiration and surface runoff, while groundwater recharge, lateral flow, and soil-water storage generally declined across most climate scenarios. These findings indicate increasing hydrological stress and reduced subsurface water replenishment in the watershed under future climatic conditions.

The statistical analyses confirmed that climate forcing, temporal climate progression, and agricultural management significantly influence watershed hydrology. Among these factors, emission pathways exerted the strongest influence on hydrological variability, while significant interaction effects between climate scenarios, periods, and management systems highlighted the non-linear nature of watershed responses. The results further reveal that irrigation substantially alters flow partitioning by increasing plant transpiration and reducing groundwater recharge and soil-water storage. Although irrigation can enhance crop water use in the short term, extensive irrigation expansion under future climate conditions may intensify water scarcity by increasing consumptive water losses and reducing hydrological resilience.

Overall, the study demonstrates that although projected climate change will influence crop productivity in northeastern Ethiopia, crop management practices, particularly irrigation, exert a stronger influence on simulated yields than climate forcing alone. The inferential analyses showed that irrigation significantly increased wheat and teff yields, whereas maize exhibited comparatively limited sensitivity to irrigation, indicating crop-specific responses to water management. While climate scenarios did not produce statistically significant differences in yield for most crops within the selected CORDEX ensemble and SWAT+ simulations, the projected yield trajectories reveal that moderate warming may temporarily benefit wheat and maize under some mid-century scenarios through enhanced CO₂ fertilization and favorable thermal conditions, whereas teff, tomato, and onion become increasingly vulnerable toward the late century, particularly under the high-emission (RCP 8.5) pathway. These findings suggest that the adverse effects of future climate change on Ethiopian agriculture are likely to be driven more by increasing temperature extremes and water stress than by changes in average climate alone. Consequently, strengthening climate-smart agricultural practices, including efficient irrigation, improved soil and water conservation, adjustment of planting calendars, and the adoption of heat- and drought-tolerant crop varieties, will be essential for sustaining agricultural productivity and food security. Given that agriculture remains the backbone of Ethiopia’s economy and is predominantly rainfed, integrating these adaptation measures with watershed-scale water resources management and climate-informed agricultural planning will be critical for enhancing resilience and ensuring sustainable agricultural development under future climate change.