Description of the study area

The Gelana catchment is situated in the Abaya Chamo sub-basin of the Rift Valley Lake Basin in southern Ethiopia, roughly 450 km south of Addis Ababa, between the Southern Ethiopian and Oromia regional states (Fig 1). The Gelana rivers in the Gelana basin originate from the Yirgachefe district of the Gedeo zone, accounting for around 10% of the total inflow into Lake Abaya [44]. The current study area covered 624.15 km² of the catchment, with elevation ranges from 3056 to 1630 meters above sea level (masl) with rises to a maximum of 3056 and a minimum of 1630 masl. The catchment experiences bimodal rainfall, with a mean annual precipitation exceeding 920 mm. The catchment is primarily hilly and undulating, with mild to flat slopes on the valley floor in the low-lying areas [45, 46]. According to Ethiopian climate classifications, the catchment mainly falls in the Woina dega zone, with average monthly maximum and minimum temperatures spanning from 35.5°C to 32.5°C and 12.5°C to 17.3°C, respectively. The catchment change includes widespread deforestation for agriculture and urbanization, and the soil parent material in the catchment is an alluvial and colluvial deposit that reaches a considerable depth.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study area location map.

Illustrates the basins of Ethiopia, Rift Valley Basin, and Gelana Watershed, compiled using DEM data and watershed boundaries. The map also highlights river gauging stations, meteorological stations, streamlines, and the main river line. DEM data was obtained from ASTER GDEM which is Available at: https://asterweb.jpl.nasa.gov/ and the Watershed Boundary Dataset was obtained from: https://www.usgs.gov/.

https://doi.org/10.1371/journal.pclm.0000483.g001

Data sources and acquisition

The primary data needed for the present study included meteorological, hydrological, digital elevation model (DEM), soil, and land use/land cover (LULC) data. The Ethiopian National Meteorological Agency provided daily data for four stations in and near the Gelana catchment, including precipitation, maximum and minimum air temperature, relative humidity, wind speed, and sunlight hours from 1987 to 2016. The Ethiopian Ministry of Water, Irrigation, and Electricity (MoWIE) for two river-gauged stations (Yirgachafe and Tore), supplied hydrological data (daily observed stream flow) between early 1990 and 2013. Soil data, one of the crucial inputs, was collected from the FAO-UNESCO soil map database and the Rift Valley Basin master plan document, which by available from MoWIE. Remotely sensed land cover satellite data were gathered using 168 paths and 56 rows for dry and cloud-free seasons from the Earth Explorer portal of the United States Geological Survey ((https://earthexplorer.usgs.gov/).

Methods of data analysis

The utilization of partial data for analysis and modeling may lead to contradictory and inaccurate results due to gaps in records, the absence of records, or the relocation of meteorological and hydrological stations [47]. The meteorological data gaps were inferred using arithmetic and normal ratio methods based on missing percentages of data. Whereas, missing hydrological data was evaluated by a linear regression between the Yirgachefe and Tore gauge stations and precipitation fields. The standard normal homogeneity test (SNHT) was utilized to confirm the data homogeneity. This technique compares the mean of the first m years of a time series with the mean of the subsequent n-m years, where n represents the total length of the dataset, and m is the number of initial years selected for comparison [48]. By dividing the series in this way, the SNHT identifies any significant changes or shifts in the mean over time, indicating potential inhomogeneity. In our analysis using XLSTAT, the SNHT showed that data from four stations were homogeneous.

Image categorization and precision evaluation

The process of classifying an image involves dividing up a continuous raster image’s pixels for pre-determined land cover classes [49]. Prior knowledge and a clear Google image visualization of the catchment area were used to collect the set of pixels that correspond to certain LULCs for this study area. The Landsat imagery was accessible from the Earth Explorer portal and was acquired from the USGS EROS Center from the United States Geological Survey. These cloud-free images included multispectral information from the Operational Land Imager/Thermal Infrared Sensor (OLI/TIRS), Enhanced Thematic Mapper Plus (ETM+), and Landsat Thematic Mapper (TM). Due to factors including open accessibility, the presence of mid-infrared bands, and an extensive track record of Earth observation data at a worldwide scale, Landsat imagery was selected for these studies over other multispectral sources. Using supervised image classifications in Erdas Imagine 2014, four land cover categories—forest, pasture, built-up, and agricultural land classes—were identified. After that, reference points were used to evaluate the accuracy of classified images were categorized as well. Ground reference points (326) from Google Earth images were utilized to assess the accuracy of the 1987 and 2003 classification images. Whereas, 387 ground reference points (GRPs) from field surveys were used to evaluate the 2020 LULC image accuracy.

The change in the LULC series was found by analyzing the percentage of the classified image (final year area minus initial year area per initial area) and the annual rate (final year area minus initial year area per number of years between two-year periods) for each LULC. Following the evaluation of images based on accuracy classified pixels, error matrices, and a total number of pixels, an assessment of overall accuracy, producer accuracy, and user accuracy was assessed [50, 51]. The classifier’s percentage gains over a class containing off-diagonal items are determined by multiplying the row and column totals, or kappa values. The kappa values of accuracy are categorized into three main groups: a value greater than 80% denotes strong agreement, a value between 40 and 80% indicates moderate agreement, and a value less than 40% indicates poor agreement [52]. The power of predictive abilities is determined by utilizing the following equation [53].

(1)

Where:—Mc is the forecasted proportion correction by chance between the simulated and observed map, and Mo is the proportion of correctly simulated cells.

Land use/land cover projection and model validation

The dynamics of land cover change and watershed modeling are frequently modeled by using cellular automata (CA) and Markov chain models [25, 26]. The 2020 classified image for this study is the baseline image. Whereas, the assembly transition probability matrix of the 2020 LULC image was simulated by using the 1987 and 2003 LULC images. The kappa index was calculated by comparing both images (classified and simulated) using the Terrest cross-tab module. Land classes for this study were reclassified in ARC GIS 10.4 and then converted into an ASCII text file format to facilitate additional analysis in TerrSet.

A common technique for assessing a model’s predictive power is the kappa index, which compares the projected data with reference data using the validated module [26]. Kappa for location (K loc), Kappa for no information (Kno), and Kappa for standard (K standard) are the three kappa indices that distinguish between errors of quantity and errors of location between two qualitative maps. K location evaluates the degree of agreement between the simulated and classified maps according to a given location, while Kno location reflects the proportion accurately classified relative to the expected proportion correctly classified by simulation, with no ability to specify the quantity of location precisely, and the K standard compares the proportion assigned properly to the fraction corrected by chance using the validate module. For each of the aforementioned computations, the following formula was used [26, 54].

(2)

(3)

(4)

Whereas;—M(m), N(m), and P(m) denotes medium grid cell-level information, N(n) denotes no grid information, and P(p) denotes perfect grid cell-level information across the landscape.

LULC images for 2040 and 2060 for this study were projected after the evaluation of the CA-Markov model’s performance. The suitability of variables was evaluated using multi-criteria evaluation (MCE) to predict LULC changes in addition to the baseline image which reflects the variables and limitations that constrain the spatial distribution of LULC classes. The current analysis identified the most significant elements influencing changes in LULC as distance from roads, slopes, rivers, towns, hill shadows, and elevation because those factors are major elements for this study.

After the preparation of the variables above, the future LULC images were predicted using the following formula:

(5)

Where: -S(t+1) is the system status at the time of t+1; pij (transition probability), which is the likelihood that a given time state (i) will change into a different time state (j) in the future; and S(t) is the LULC status.

The transition probability matrix (pij), which is between 0 and 1, can be calculated as follows [55].

(6)

Climatic scenario models and bias correction

In the current study, the relative change of climatic variabilities between the current and future scenarios was assessed using daily climate data (precipitation, maximum, and minimum air temperatures) obtained from the coordinated regional climate downscaling experiment (CORDEX) under the Africa domain. The data was collected from historical (1987–2005) and future modeled (2021–2080) periods. The following regional climate models were utilized for this study: RCA4 (Ross by Center Regional Atmospheric Model) from the Swedish Meteorological and Hydrological Institute (SMHI), RACMO22T (Regional Atmospheric Climate Model, version 2.2) from Koninklijk Nederland’s Meteorologisch Institute (KNMI), and CANRCM4 (Canadian Centre for Climate Modeling and Analysis regional climate model) from the Canadian Centre for Climate Modeling and Analysis (CCCMA). The climate projections were based on two Representative Concentration Pathways (RCPs): RCP 4.5 and RCP 8.5, over the entire sub-catchment after bias correction.

Bias correction is typically required because climate models frequently provide biased illustrations of observed time series data due to systematic model flaws brought on by faulty conceptualization, discretization, and spatial averaging within grid cells [56]. Several statistical bias correction techniques were developed to minimize systematic model errors [53]. The bias correction processes in this study were conducted using CMhyd (the climate model data for hydrological modeling) software, which was obtained from the SWAT website (https://swat.tamu.edu/software/) and was used to correct biases in the precipitation and temperature data.

A study by [57] compared five bias correction methods using CMhyd and found distribution mapping performed excellently for a climate change impact study on streamflow dynamics of two rivers in the Northern Lake Erie basin, Canada. The research carried out by [58] employed CMhyd to forecast climate change based on temperature fluctuations in the Fincha watershed, extract the Cordex NetCDF, and remove the bias of minimum and maximum temperatures, with positive results. Based on the listed evidence above on different areas, distribution mapping fitted well, so for this study, distribution mapping in CMhyd was ranked for both temperature and precipitation. To lessen the uncertainty for each specific model gap, it is possible to ensemble bias-correct RCMs before using the model, which was carried out.

Overview of the SWAT model

The Soil and Water Assessment Tool (SWAT) is a physically based, semi-distributed simulation model designed primarily for strategic planning that forecasts the effects of land use/land cover change and management techniques on hydrological regimes over extended periods in diverse soils, land use, and management conditions [59]. The land phases of the simulation handle the loading of water, sediment, nutrients, and pesticides into the primary channel in each sub-basin, whereas the routing phases consider the flow of water, sediment, and agricultural chemicals over the channel network to the catchment outlet. Based on the water balance equation, SWAT simulates the hydrological cycle in the land phase as follows: -

(7)

Where: Swt = final soil water content, Swo = beginning soil water content, t = time (days), Rday = precipitation total, Q surf = amount of surface runoff, Ea = amount of evapotranspiration, Wseep = quantity of water entering the vadose zone, and Qgw = quantity of return flow.

Model setup and sensitivity analysis

The model uses easily available input data, including soil, climate, land use/landcover, and DEM data. Watershed delineation was carried out first by setting up a new SWAT project, and it was then divided into hydraulic response units (HRUs) by superimposing data on the watershed’s slope, soil, and LULC data. A basic SWAT simulation was executed using weather data such as precipitation, maximum and minimum temperatures, wind speed, solar radiation, and relative humidity. The simulation in the project used three-step processes. First, scenario simulations were conducted utilizing constant baseline (1987–2016) climatic data with five separate LULC scenarios,1987, 2003, 2020, 2040, and 2060, to examine the independent impact of LULC on hydrological response.

For the second scenario simulation, a separate climate effect on hydrological response was assessed by running the model twice for near and far periods (twice for RCP 4.5 and twice for RCP 8.5) in the RCP series data for (2021–2050) and (2051–2080), with a constant 2020 LULC data. The third simulation was completed to determine the combined impact on the hydrological response, either from LULC or climate, and it was done by comparing the LULC model simulated from 2040 and 2060 with the baseline climatic simulation. Evaluating sources of uncertainty by calculating, the sequential uncertainty fitting (SUFI-2) parameter. The SUFI-2 found in SWAT-CUP was used to calibrate and validate the SWAT model accounting for possible sources of uncertainties, such as in variables, model, parameters, and measured data. SWAT-CUP was performed to assess the sensitivity parameter based on the t-statistic (larger absolute values) and p-value (closer to zero) [60].

Model evaluation criteria

Calibration is the process of evaluating model parameters by comparing results to observations to ensure a similar response over time. Whereas validation is the process of determining whether the model parameters can predict flow for periods, other than those calibrated for the model [61, 62]. For model evaluation the dataset of the period 1990–2013 of observed streamflow was divided into the calibration period (1992–2005) and the validation period (2006–2013) with two-year data saved as a warm-up period to initialize distinct catchment processes. The SWAT was calibrated and validated by using the SUFI-2 method with the best resolved simulated stream flow, calibration parameter, and performance of the model. For this study coefficients of determination (R²), Nash-Sutcliffe efficiency (NSE), and percent bias (PBIAS) were used unitized to assess model performance. The coefficient of determination indicates the proportion of the variance in the measured data explained by the model; Nash-Sutcliffe efficiency compares observed versus simulated data; and the percent bias indicates an average tendency of the simulated data to deviate from the observed counterpart [63, 64]. The model performance for monthly streamflow simulation can be identified as satisfactory if 0.7<NSE<0.8, 0.75< R² <0.85, and ±5<PBIAS<±10. Mathematically, the terms above are expressed as follows:

(8)

(9)

(10)

Where: Qobs and Qsim denote actual and simulated data, and and are the mean of actual and simulated data, and n indicates the number of data points.

Land use/land cover change detection and accuracy assessment

Based on a satellite image from the USGS Earth Explorer, four LULC categories (forest, pasture land, built-up, and agriculture) were classified using supervision classification algorithms in Erdas Imagine 2014. According to LULC conversions, significant gains and losses were examined throughout the study period. The 1987 LULC image classification shows that forest cover was the highest proportion at 254.63 km² (40.79%), followed by pasture land at 229.39 km² (36.75%), agricultural land at 124.69 km² (19.98%), and built-up area at 15.44 km² (2.47%).

The classification of the thermal infrared image of the 1987 LULC yielded an accuracy assessment of 0.92 for Kappa statistics and an overall accuracy of 93.75%. The results obtained revealed that the classification was highly significant and within the suggested range, according to [50]. The LULC image of 2003 was done in the same fashion as the previous land class classification, and the following proportion of coverage areas was obtained: built-up of 31.32 km² (5.02%), pasture land of 127.9 km² (20.5%), forest land of 184.62 km² (29.57%), and agricultural land of 280.31 km² (44.91%), as shown in Table 1. Similar to the 1987 LULC classification, the 2003 LULC image classification prioritized agricultural and built-up land cover relative to pasture and forest land classes. Despite a minor increase in built-up land cover classes, the results showed that the agricultural class continuously outperformed the other land cover classes. The accuracy assessment of classifications revealed 95.39% overall accuracy and a 0.94 Kappa value, implying that the results have high significance and are within an acceptable range for further investigation [50].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The coverage area and land use/land cover percentage changed from 1987 to 2020.

https://doi.org/10.1371/journal.pclm.0000483.t001

As in the previous two series of LULC images, the 2020 LULC categories showed an increase in agricultural and built-up areas, with a compensatory reduction in other land cover percentages. The classified LULC for 2020 encompassed built-up areas of 56.87 km² (9.11%), pasture land covered 65.52 km² (10.5%), agriculture land covered 344.38 km² (55.18%), and forest land covered 157.38 km² (25.22%). The computed kappa value and overall accuracy for these land classifications were 0.93 and 94.57%, respectively. According to [50], the calculated accuracy assessment results were within the required range. When rating accuracy, not only the Kappa coefficient and total accuracy were taken into consideration, but also the accuracy of the producers and users. Thus, Table 2 shows the accuracy rates for users and producers per LULC class for each land class, which ranged from 89.71% (pasture land in 1987) to 96.265% (forest land in 2020) and 89.8% (agricultural land in 1987) to 97.46% (pasture land in 2020). Generally, the prepared LULC image and calculated results are shown below in Table 2, and the results are highly significant for further investigation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Land use/land cover accuracy assessment from 1987 to 2020.

https://doi.org/10.1371/journal.pclm.0000483.t002

Analysis of land use/land covers (LULC) change

The study year image was compared pixel by pixel using coverage area proportion, and the results are summarized in Table 1 above. The land cover change during the first period (1987–2003) showed considerable changes in agricultural (124.8%) and built-up land use (102.85%), whereas forest (27.49%) and pasture (44.24%) land use lost a significant amount relative to other classes. Similar to the first period of change, built-up (81.58%) and agricultural land (22.86%) categories increased coverage area in the second period (2003–2020), with significant losses in pasture (48.77%) and forest (14.75%) land classes due to a variety of reasons as outlined in Fig 2. Various studies in different catchments reflect current findings. One study conducted in the Gojeb watershed of the Omo Gibe basin revealed that expanding populations and crop production resulted in an increase in cultivated area at the expense of forest land cover [65].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. 1987,2002 and 2020 LULC map of Gelana catchment.

(accessed via US geological survey (http://www.usgs.gov/) data processed and analysed by using ArcGIS, ERDAS, and survey data). The maps revealed spatio-temporal trends of Land Use/land cover (LULC) change classes as built-up, forest, agriculture, and pasture land.

https://doi.org/10.1371/journal.pclm.0000483.g002

The equivalent outcome was also recorded in the Didessa basin, Anger sub-basin, where the area of land covered by forests and shrubs dropped by 39.5% between 1986 and 2010 [66]. A more in-depth investigation of the Adei watershed in Ethiopia’s central highlands identified both temporal and spatial changes in land cover between 1985 and 2015. According to one study, urban development and agricultural deforestation were the most significant drivers of the progressive fall in the coverage area of wood and forest land [67]. Similar findings are also observed in south-central Ethiopia, where agricultural land development peaked in terms of suitable land cover between 1972 and 2013 and then continued to occupy marginal areas due to the impact of forest biodiversity [68]. Research conducted in Ethiopia’s Nech-Sar and Awash National Parks on changes in land use/ land cover revealed that deforestation, changes in the land management system, and increased pressure from locals and livestock were the main factors seriously harming wildlife habitat [69]. Evaluations of LULC trends in the current watershed have shown that, as a result of ongoing population growth and agricultural development, there has been a clear increase in built-up and agricultural land classes at the expense of pasture and forest land classes.

Prediction and validation of future LULCC

This study utilized cellular automata and Markov chain (CA-Markov) models in IDRISI image processing tools in TerrSet to predict future LULC changes. First, the model efficiency was assessed by simulating the LULC in 2020 using transition area information and probabilities between the LULC series of 1987 and 2003 (Table 3). The errors in the agreement/disagreement components of the simulated 2020 LULC image and the real classified image were 0.136 (allocation/disagree grid cell) and 0.0233 (quantity/disagree quantity). The primary reason for the variations between the two images was marked by allocation error rather than quantity error. Based on the kappa indices, the model demonstrated a strong relationship in simulating future LULCs; further results show that built-up (44.54%) and pasture (-29.64%) land classes were overestimated while agricultural (1%) and forest (-5.9%) land classes were slightly underestimated. The simulation-estimated Kappa variation for Kno of 0.810, Klocation of 0.78, and Kstandard of 0.76 revealed a high level of agreement. Based on the kappa indices, the model confirmed significant agreement in forecasting future LULC.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Proportional area of actual and simulated LULC image of 2020.

https://doi.org/10.1371/journal.pclm.0000483.t003

The LULC images for 2040 and 2060 were predicted with the use of the 2020 LULC as the base map and other bio-physical variables. The coverage areas are shown in Table 4 and Fig 3 below. The coverage area for the first prediction revealed a high tendency to increase agricultural (358.49km²), built-up (65.35km²), and forest (162.26km²) land classes while decreasing pasture (38.05km²) land categories. The primary reason for the expansion of agriculture (57.44%) and built-up (10.47%) land classes, and the slight increment of forest (26%) and decrement of pasture (6.09%) land covers in the first projection, could be the current continuous growth of population, which is largely driven by rural-urban migration within the Gelana basin, as adjacent cities and towns tend to absorb the city growth. The prediction also showed a positive increase in land cover from forests (3.1%), which may be related to Ethiopia’s 2019 Green Legacy initiative effort to plant 500 million trees nationwide in just one year. Improved education, medical care, and infrastructure in the catchment were primary contributors to an increase in built-up land classes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. LULC for 2040 and 2060.

(predicted by using a Multi-Layer Perceptron Neural Network (MLP-NN) with a Cellular Automata-Markov Chain Model), which is found in the software of the TerrSet 18.32 framework. https://clarklabs.org/product/terrset-2020-academic-license-upgrade/. The maps illustrate the spatial distribution of LULC classes (Built-up, Pasture land, Forest, and Agriculture) and the projected trends over the specified periods.

https://doi.org/10.1371/journal.pclm.0000483.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. LULC proportion from 2040 to 2060.

https://doi.org/10.1371/journal.pclm.0000483.t004

The second LULC projection was done for 2060, which revealed a constant expansion by agricultural (370.83km²) and built-up (76.36 km²) land classes while forest (143.34km²) and pasture land (33.62km²) were contracted relative to earlier scenarios. Compared to the previous detection, the rate of rise in agricultural land was higher, while it was lower for forest (22.97%) and pasture (5.39%) land classes. This may reflect a lack of available land for further farming and built-up expansion. A simulation of LULC between 2040 and 2060 revealed that the various LULC categories were constantly expanding and contracting, which could have an influence on the environment, including water supplies, and have a direct impact on the catchment’s ecological integrity and water resources. In the current research, we examined several factors that contributed to LULC changes, including population growth, agricultural expansion, shifting cultivation, and fuel wood extraction, which are the primary causes of forest and pasture land loss.

Similar studies have been carried out in the Majang Forest Biosphere Reserves in Gambella, Southwest Ethiopia, using the CA-Markov model in IDRISI software. According to the findings, there will be a drop in forest and grassland and an increase in agricultural and settlement areas between 2032 and 2047 [54]. Comparable research conducted in Malawi’s Decza district using the CA-Markov Chain model embedded in IDRISI software proved that, by the years 2025 and 2035, built-up areas, water bodies, and barren land will all rise, while agricultural and forest land will substantially decrease due to increased population interference [27]. Other study attempts support the findings in several catchment areas [54, 70, 71].

Climatic change projections of the RCMs for rainfall and temperature

Three models exhibit distinct behaviors at different time series with varied magnitudes based on climatic scenarios, as shown in Tables 5 and 6. Except for CANRCM4, which estimates a 10.55% increase in the far future (2051–2080), evaluating precipitation revealed that three RCMs exhibit different rates of intensity with decreasing precipitation in each scenario. More specifically, a RCP 4.5 model scenario for precipitation may decline by 26.4% and 15.46% in the near and far future, respectively. However, for RCP 8.5 precipitation is expected to decline by 14.34% 2021–2050 scenario. In RACMO22T, precipitation decreased by 12.1% and 4.6% for near -and far-future periods when RCP 8.5 was applied; similarly, precipitation declined by 15.5% and 5.56% for near- and far-future periods when RCP 4.5 was used.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Precipitation changes (%) between the three RCMs under the RCP 4.5 and RCP 8.5 scenarios.

https://doi.org/10.1371/journal.pclm.0000483.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Temperature change (°C) for three different RCMs under the RCP 4.5 and RCP 8.5 scenarios.

https://doi.org/10.1371/journal.pclm.0000483.t006

According to Table 5, the RCA4 for both scenarios (RCP4.5 and RCP8.5) showed a larger drop in precipitation (48 to 48.64% and 41.3 to 43.16%, respectively) for the near and far future time frames. This is consistent with the RACMO22T and CANRCM4 models, which revealed a substantial reduction in rainfall. As shown in Table 5, the ensemble mean precipitation decreased by 22.58% and 12.4% under RCP 8.5 and 29.96% and 23.22% under RCP 4.5, respectively. From the results, the ensemble mean yielded relatively significant results as compared to individual models, which were used in the present study.

Several climatological studies show that various catchments in Ethiopia a decreases in rainfall between the 20^th and 21^st century [72–74]. In 2020, the output from seventeen GCMs was analyzed, and it was observed that 10 out of the 17 GCMs indicated a decrease in precipitation. However, when examining the ensemble mean of the seventeen GCMs, there was negligible change in precipitation over the specified time domain (please specify the time domain, e.g., 2021–2050 or 2050–2100) [75]. Another study, which analyzed rainfall data from the last 38 years, verified that precipitation intensity was declining in eastern, western, and southern Ethiopia [76]. As an illustration, [77] rainfall estimated in the Gilgel Abay watershed reduced by over 30% between 2010 and 2040 using statistically downscaled HadCM3 data from the model. The available model output for the RCP 8.5 scenario results in heightened variability in rainfall patterns which may lead to increased severity of drought and floods. Food security and agricultural livelihoods can be seriously by extreme climate variability beyond historic norms, especially for smallholder farmers [78].

As per the prediction results, as shown in Table 6, for model CANRCM4 under the medium emission scenario (RCP4.5), the temperature rises by 1.20°C and 1.870°C for both near and far future terms; under the higher emissions scenario (RCP8.5), the minimum temperature increases by 1.23°C and 3.1°C for both near and far future periods. Under RCP 4.5, maximum temperatures may rise by 1.44°C and 2.20°C, respectively, whereas under RCP 8.5, they may rise by 1.39°C to 2.93°C for both near and far futures. For RCA 4.5, minimum and maximum temperatures increase by 1.3°C and 2.17°C, respectively, under the near and far future terms of RCP 4.5. For both minimum and maximum temperatures, however, there may be slight increases of 0.6°C and decreases of 1.9°C, respectively, for the two near and far periods.

The minimum temperature for RACMO22T decreases by 0.4°C and increases by 0.58°C under RCP 4.5 for the near and far future, respectively, but under RCP 8.5, it increases by 1.36°C and 3.35°C for the same periods. Aside from the results above, the maximum temperature for this RCM showed that, for both the near and far periods, there would be increases of 1.2°C and 2.59°C under RCP8.5 and drops of 0.2°C and a rise of 0.5°C under RCP4.5, respectively. Under the medium emission scenario (RCP4.5), the ensemble mean of the maximum temperature will increase by an average of 0.98°C and 1.8°C over the near term, in contrast to three other regional models, whereas the minimum temperature would rise by an average of 0.7°C and 1.54°C for the near- and far-term, respectively. For high emission scenarios (RCP8.5), the maximum temperature will rise by 0.53°C and 2.04°C, and the minimum temperature will rise by 0.56°C and 2.35°C, respectively, for both the near and far futures. According to the current analysis, under RCP 8.5, the mean maximum temperature rises more frequently between 2051 and 2080 than under RCP 4.5, the medium level of emissions.

Numerous studies carried out in Ethiopia have confirmed that climate variations throughout the country are the primary driver of climate change [20, 58, 79, 80]. Based on a study conducted in the Blue Nile basin, the average annual temperature in the upper basin is projected to rise by 1.5˚C, 2.6˚C, and 4.5˚C between 2011 and 2040, 2041 and 2070, and 2071 and 2100, respectively [81]. According to [82], the average monthly maximum and minimum temperatures for the two scenarios (RCP 4.5 and RCP 8.5) for the near and far future periods indicate a growing trend in almost all of the months. According to the current findings, the Gelana watershed has warmed on average over the study period for both scenarios (RCP 4.5 and RCP 8.5), which has severe consequences for watershed management and planning.

Hydrological model sensitivity analysis, calibration, and validation

The sequential uncertainty fitting-2 (SUFI-2) under SWAT CUP was used to conduct a sensitivity analysis of SWAT model outputs to identify key parameters influencing model performance. Parameters with low sensitivity values were found to have minimal effect on model outputs. In contrast, those parameters with medium to very high sensitivity were preferred for hydrological model calibration. Table 7 presents the results of the sensitivity analysis, including the sensitivity ranks and fitted values for the Tore gauging station. The analysis identified eleven parameters as the most sensitive for calibration in the catchment outflow based on the t-statistics and the p-value. The highest t-stat value implies a high parameter coefficient to standard error ratio, whereas a lower p-value excludes the hypothesis that adding a parameter raises the variable response substantially [83]. The parameter rankings were obtained from the last iteration of SUFI-2 and are shown in Table 7.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Global sensitive flow parameters [84].

https://doi.org/10.1371/journal.pclm.0000483.t007

After identifying the sensitive analytical variables, the model parameters were calibrated and validated to ensure compatibility with the observed data. Validation was completed for a further period after calibration to ensure consistent calibrated parameters over time. Simulated and measured stream flows in the Tore Gauging Station were consistent with statistical performance (R² = 0.84, NSE = 0.77, PBIAS = 15.9,) with calibration. The SWAT model was also successfully validated for another independent period with R² = 0.88, NSE = 0.8, and PBIAS = -14.1. Even if the model results resembled the monthly observed flow trends, it overestimated and underestimated peak flows over some months and years, as shown in Fig 4. Overall, the performance evaluation results were highly significant and within the acceptable range, according to [63, 64].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Observed and simulated monthly average streamflow for calibration and validation.

(source: MoWIE; Tore Gauging Station data (1990–2013), calibrated and validated within SWAT-CUP). This figure presents the comparison between observed and simulated monthly average streamflow for the calibration (1990–2005) and validation (2006–2013) at the Tore Gauging Station.

https://doi.org/10.1371/journal.pclm.0000483.g004

Hydrological responses for land use/ land cover change alone

A calibrated and validated SWAT model was used to evaluate the impact of the LULC change on hydrological components considering five distinct LULC scenarios, three from the past (1987, 2003, and 2020), and two from the future (2040 and 2060) for invariant climatic conditions under both annual and seasonal scenarios. The long-term hydrologic simulation was performed to analyze the impact of LULC change on hydrological components in the Gelana River basin. The response of streamflow to LULC change alone indicated that the annual streamflow was increased from 103.88 to 128.62 6 m3/s during the study period 1987 to 2060, respectively. Similarly, the mean monthly streamflow was increased in the wet season (3.69 to 28.2%) while decreased in the dry season (8.04 to 0.16%). The principal drivers of this change were annual increases in agriculture and built-up land cover and extreme losses in pasture and forest land cover classes across the study period. The streamflow during the dry season, which mostly comes from baseflow decreases, whereas streamflow during the wet season increases. Thus, streamflow responses to LULC are more sensitive in dry seasons than in comparison with wet seasons.

The expansion of agricultural land and built-up area over forest land and pasture land results in the reduction of lateral flow (downward infiltration) that flows into the shallow aquifer. The streamflow change was higher during the period 1987 to 2003 as compared with the period 2003 to 2020, mostly due to LULC change reflecting less expansion of agricultural land and forest land between 2003–2020 and 1987–2020. Owing to the large expected changes in land use from 1987 to 2020 and 2020 to 2060 by agriculture and built-up areas the hydrological responses vary highly with different magnitudes (Table 8). Potential Evapotranspiration (PET) is the quantity of water that could evaporate and transpire if there is sufficient water available. The PET fell from 1987 to 2003, followed by a moderate increase (2020 to2060).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. Variation of the annual hydrological components under land use/ land cover change.

https://doi.org/10.1371/journal.pclm.0000483.t008

These trends may be associated with changes in land use and monthly and annual variations of climate at specified periods. The decrease in PET in 2003 could be linked to changes in vegetation cover (for example, deforestation or urbanization), which lower transpiration. The PET has varied over the years, with a large increase projected by 2060. Forests commonly have a higher PET because of their large leaf area and deeper root systems, which allows increased transpiration. When forest cover is replaced by agricultural land, the PET frequently decreases because of lower canopy cover and soil moisture retention. Also, the rise of urbanization creates more impermeable surfaces, which can disturb the natural hydrological process. Whereas, built-up regions may have lower PET because of less vegetation and can also yield localized microclimates that influence temperature and humidity.

Evaporation has increased slightly particularly between 1987 and 2003 reflecting an increase in urbanization and exposed water bodies, which contribute to more direct water loss. Water yield, which is the total amount of water available after accounting for evaporation and transpiration, rises from 1987 to 2040, but declines somewhat in 2060 possibly linked to interactions of land use changes such as diminished vegetation, water management practices, urbanization, and soil moisture and vegetation. Also, surface runoff increases gradually after 2020 and peaks in 2060. These limited response of evaporation and runoff for later in the 21 ^st century is most likely due to the rise of impervious surfaces, such as roads and buildings, which limit water infiltration and increase runoff. Groundwater achieves a peak in 2003 but drops dramatically by 2060 as shown in Table 8. The drop is most likely caused by lower infiltration rates as a result of increased urbanization and surface runoff, which reduces groundwater recharging.

Various studies in different watersheds in Ethiopia are consistent with this study. One study in China revealed that urbanization lowers PET due to a loss in vegetative cover, which has a direct effect on transpiration rates [85]. The importance of vegetation cover in determining PET, demonstrates that changes in land use (such as deforestation or urbanization) can reduce PET via changing surface features [85, 86]. Land use changes from forest to agricultural land in the middle reaches of the Yangtze River had a major impact on PET patterns, with a noticeable drop in places where forest cover had been lost. The study stressed the importance of vegetation type in determining local climate conditions [87]. The study conducted in the Blue Nile Basin found that deforestation initially led to increased water yield due to less water uptake by trees. However, over time, soil erosion and decreased infiltration reduced overall water yield, contrasting with the continuous increase observed in global data up to 2040 [19].

Similar results in the Awash River Basin, Ethiopia, show that water yield initially rises after deforestation but decreases due to erosion and land degradation, leading to long-term water shortages [88]. Furthermore,[89] pointed out that urbanization significantly increases surface runoff by decreasing infiltration. The change in rainfall patterns due to climate change can reduce groundwater recharge rates, further exacerbated by urban expansion [90]. A study by [91] found that groundwater recharge in the Ethiopian Rift Valley decreased due to land degradation and increased runoff, consistent with the decline in groundwater. The upper Blue Nile River’s stream flows are predicted to increase by 26% between 2010 and 2039 and then decrease by 10% between 2070 and 2099 [92]. The change in LULC for this study revealed that stream flow increased annually and throughout the wet season but dropped in the dry seasons over the study period when LULCC alone was considered. Another study explains that as agricultural land expands, the soil structure and organic contents are influenced by a plow during long-term tillage, which decreases infiltration as well as reduces forest and pastureland cover, which alarmingly reduces evapotranspiration and increases streamflow [1].

Hydrological response for climate change alone

Using 2020 LULC data, the SWAT model simulation was run four times over two near and far periods using RCP 4.5 and 8.5 scenarios. For base period data, the average yearly flow of streamflow declines in two successive scenarios (RCP 4.5 and RCP 8.5). For the RCP 4.5 and RCP 8.5 scenarios from 2021 to 2050, the mean annual model results decreased by 23.27% and 24.61%, respectively. However, the model indicated a large reduction for both the RCP 4.5 and RCP 8.5 scenarios for the far future (2051 to 2080). For the two scenarios, the mean streamflow throughout rainfall was lower than the baseline, and for the dry seasons, the magnitude was substantially smaller.

PET increases slightly from RCP 4.5 to RCP 8.5 in the period 2021–2050 besides it shows a much larger increase by 2051–2080 under RCP 8.5 (Table 9). This aligns with climate model projections indicating that with higher greenhouse gas concentrations (RCP 8.5), temperatures are expected to rise significantly, leading to more water lost to the atmosphere through evaporation and transpiration. Between 2021–2050 and 2051–2080, evaporation rises, with higher rates under the RCP 8.5 scenario. Future rising ET rates are in line with projections of rising atmospheric moisture demand driven by higher temperatures, especially under the more extreme RCP 8.5 scenario. Water yield rises significantly between 2051 and 2080 under RCP 4.5, but falls slightly under RCP 8.5 from 2021 to 2080. Water yield increases under a more moderate warming scenario (RCP 4.5) as precipitation and temperature changes are balanced. Surface runoff rises dramatically under RCP 4.5 (2051–2080) but reduces under RCP 8.5 over the same period. The high rise in surface runoff under RCP 4.5 could be ascribed to increasing rainfall intensity, which causes more water to flow off rather than infiltrate into the ground. Under RCP 8.5, however, significant warming may reduce rainfall or cause prolonged dry spells between intense storms, resulting in lower runoff. However, under RCP 8.5, significant warming can result in increased evaporation, decreasing the amount of available water. While groundwater recharge drops under RCP 8.5 over time, it rises under RCP 4.5 by 2051–2080. Balanced temperature increases and precipitation may encourage higher groundwater recharge through infiltration under moderate climate change (RCP 4.5). However, under extreme warming scenarios (RCP 8.5), groundwater levels are lowered by decreased infiltration and increased surface runoff.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 9. Variation of the annual hydrological components under Climate change.

https://doi.org/10.1371/journal.pclm.0000483.t009

A related study by [93] employed CCLM downscaling for the periods 2041–2070 and 2071–2100 and indicated 6.6% and 6.4% decreases in precipitation over the upper Blue Nile basin, respectively, which have an impact on streamflow by lowering its volume. According to [94], a study that used HadCM GCM to examine the effects of climate change on the hydro-meteorological variables of the Finchaa sub-basin identified a substantial drop in streamflow due to decreasing precipitation. Climate fluctuations, including temperature and precipitation, have a higher effect on stream flow, based on studies investigated in the Rifty Valley Lake Basin of the Meki River Catchment. For example, a 10% decrease in rainfall resulted in a 30% reduction in the catchment’s predicted hydrological response. A 1.5°C increase in air temperature, on the other hand, would result in a 13% drop in stream flow [79].

One study in Ethiopian highlands identified warming trends elevate evaporation rates [95]. Also, another study in Ethiopia shows that increasing temperatures due to climate change directly affect PET, especially in semi-arid regions [96]. The intense warming and drying periods, especially in areas with less adaptive capacity, may cause water yield to decline [97]. A similar investigation from the Awash River basin noted that rising temperatures and reduced infiltration from surface runoff decrease groundwater recharge [91]. In the current study, the hydrological response to climate scenarios (RCP 4.5 and RCP 8.5) shows that moderate warming (RCP 4.5) could increase water yield and groundwater recharge, whereas extreme warming (RCP 8.5) leads to decreases in water availability due to increased evapotranspiration and reduced groundwater recharge.

Hydrological responses for combined LULC and climate change

This study indicates that the watershed stream flow is negatively impacted by climate change with a rise in temperature and a concomitant fall in precipitation. In contrast, the LULC changes are associated with increased streamflow when there is expansion of agriculture and built-up land whereas there is a decline in land covered by pasture and forest. To evaluate the effects of LULC and climate change on hydrological response, combined scenarios of three land use/land cover data sets (2020, 2040, and 2060 LULC) and two RCPs (RCP 4.5 and RCP 8.5) with a baseline simulation were undertaken. Both RCP 4.5 and RCP 8.5 resulted in significantly lower mean annual stream flow (42.45 to 55.38%), wet season flow (44.28 to 55.2%), and dry season flow (36.1 to 56.8%) under RCP 4.5, and further drops to (59.36 to 62.15%) mean annual, (58.33 to 60.49%) for wet season, and (67.6 to 75.44%) for dry seasons under RCP 8.5 respectively. The RCP 4.5 and RCP 8.5 scenarios showed much lower percentages of mean annual, rainy season, and dry season streamflow than the baseline time frame based on the model simulation’s combined scenario.

The other hydrological responses for this case (combined LULC and climate change) were observed and shown in Table 10 and Fig 5. The PET decreases over time under both RCP scenarios. PET in the earlier period (2021–2050) shows a decreasing trend by 2051–2080, especially under RCP 8.5, suggesting less potential for evaporation as conditions become drier. The actual amount of water that evaporates or is used by plants is represented by evaporation which shows significant variability, especially between RCP 4.5 and RCP 8.5 in the 2021–2050 period, where it jumps from 2.79% to 30.60%.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Hydrological response to LULC and climate change.

This figure depicts the hydrological response to changes in land use/land cover (LULC) and climate, with climate data sourced from historical records and RCM, LULC data from USGS, and hydrological variables simulated using the SWAT model.

https://doi.org/10.1371/journal.pclm.0000483.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 10. Variation of hydrological components under combined climate and LULCC.

https://doi.org/10.1371/journal.pclm.0000483.t010

PET drops to 19.6% under RCP4.5 by 2051–2080, reflecting potential future drying conditions, while under RCP8.5, it decreases to 17.79%. In the RCP4.5 scenario, water yield increases significantly in the long term (2051–2080), reaching 63.49%. However, under RCP8.5, it declines drastically to 4.38%, indicating a potential severe reduction in water resources in a high-emission future. Under both scenarios, surface runoff is projected to increase, particularly in RCP8.5 (2051–2080), where it reaches 43.31%, suggesting more flooding risk in this scenario because more pasture land and forest would be converted by agriculture with a lower precipitation percentage. For these coupled conditions, groundwater storage increases substantially under RCP4.5 (2051–2080) to 27.06%, indicating the possibility for improved groundwater recharge, although it remains low under other scenarios, especially in the high-emission future.

Increasing temperatures (associated with higher PET) will lead to increased evaporation, drying of soils, and reduced water availability (lower water yield under RCP8.5). Climate variability may also lead to an increase in extreme weather events, such as more frequent floods (higher surface runoff in RCP8.5), with less predictability in groundwater recharge (low groundwater under RCP8.5). Changes in land use, such as deforestation or urbanization, can reduce vegetation cover, raising surface runoff and reducing ET, resulting in lower groundwater recharge. The drop in PET and ET under RCP4.5 indicates better land management techniques could help to reduce future water losses.

Table 10 reveals how climate change and LULC variations may cause significant shifts in Ethiopia’s hydrological cycle. Under RCP8.5, the country could face reduced water availability (water yield and groundwater), increasing flood risks (surface runoff), and an anticipated drying trend (evapotranspiration) in the long run. These findings highlight the necessity of climate mitigation and sustainable land use planning in Ethiopia to guarantee future water resources. Several studies confirmed the above results at different times and catchments with different magnitudes of increment and decrease by different parameters. Rising temperatures and falling precipitation may reduce streamflow, increasing the possibility of evapotranspiration and lowering farmer production [80, 98]. Similar findings have been supported by climate change studies in Africa, particularly in the Sub-Saharan region. Based on this research, the land subtropics may face various kinds of warming trends, frequent extreme heat events, increased drought, and changes in rainfall, all of which will result in a drop in streamflow [99].

A study by [100] also noted potential reductions in river flows and groundwater recharge under high-emission scenarios, particularly in water-scarce regions. This aligns with the decline in water yield and groundwater under RCP 8.5. Another study by [101] found that rising temperatures and changing rainfall patterns are expected to increase evapotranspiration rates and reduce water availability in Ethiopia, particularly in the Rift Valley and highland regions, which aligns with the reduction in water yield under RCP8.5. Studies pointed out by [102] have projected increased surface runoff and flood risks under future climate scenarios in parts of Ethiopia, particularly under high-emission scenarios (RCP8.5), which is reflected in the increase in surface runoff. In contrast with LULC and climate change alone, the combined impact of climate and LULC change had a more severe overall influence on hydrological response in both near- and far-future scenarios study periods.

Conclusions

This study investigated the impact of climate and LULC change on the hydrologic response in the GELNA catchment, Rift Valley Basin of Ethiopia. LULC images prepared from ERDAS Imagine revealed four major land use classes for the current study: agricultural, forest, pasture, and built-up area, using the maximum likelihood supervised classification method. LULC scenario prediction maps of 2040 and 2060 were generated using the CA–Markov model. The results of the study indicated an increase of agricultural and built-up land coverage by 19.98 to 55.18% and 2.47% to 9.11% over a three-decade period; however, forest and pasture land coverage decreased by 40.79 to 25.22% and 36.75% to 10.49%. Agriculture and built-up areas are likely to increase (55.18% to 59.41% and 9.11 to 12.23%), whereas forest and pasture land are anticipated to fall (25.22 to 22.97% and 10.49 to 5.39%, respectively) for two projected LULCs. The primary drivers of these LULC changes include fast population expansion inside and around the watersheds, and increased farmland and urbanization in the vicinity.

Climate projection analysis shows that mean annual precipitation declines for both individual and ensemble mean, while mean maximum and minimum temperatures rise significantly under two future climatic scenarios of from the IPCC, RCP 4.5 and RCP 8.5. The agreement between the measured and observed river flows indicated that the SWAT model could be used to simulate the hydrologic responses to climate and LULC variations in the catchment. The performance evaluation of calibration and validation of streamflow revealed a satisfactory agreement between observed and simulated values with statical parameters. The effect of LULC change alone increased streamflow, evaporation, surface runoff, and water yield but decreased groundwater and potential evapotranspiration. While climatic change scenarios for the current (2021–2050) and far (2051–2080) were projected using the ensemble mean of regional climate models under RCP4.5 and RCP8.5 scenarios with an increase in potential evapotranspiration, evaporation, surface runoff, water yield and decrease in groundwater at varying scales.

Throughout the study period, streamflow declined in mean annual, wet, and dry seasons due to increases in temperatures and precipitation decreases, deforestation, and a climate-driven increase of evapotranspiration. The consequences of altering land use/ land cover in conjunction with temperature rise and precipitation decrease have a substantial impact on hydrological response in combined situations. The combined impacts of climate and LULC variations on hydrologic responses in the coming decades are greater than the variation trends of climate or LULC change alone. This scenario results in substantial increases in surface runoff, potential evapotranspiration, evaporation, and groundwater resources.

The trends in the data reveal the impacts of land-use changes and climate variability on hydrological processes like urbanization, deforestation, rising temperature, and lowering precipitation patterns are key drivers behind the observed trends in evapotranspiration, evaporation, surface runoff, water yield, and groundwater recharge. Managing these impacts will require an integrated approach to land-use planning and water resource management. The increased rainfall, warmer temperature, and significant increment in the hydrologic component, particularly the surface runoff and associated extreme peak flow over the coming decades, are likely to put tremendous pressure on the catchment. This calls for a sustainable and effective adaptive measure for future water resource management in the Gelana watershed. The findings illustrated how the hydrological processes in the Gelna catchment respond to variations in climate and LULC. This could help with water resource and land management planning. If degraded sloping areas are rehabilitated, tree plantations, and groundwater recharge increase. As a result, the surface runoff that washes topsoil and nutrients into water bodies is minimized. Overall, the findings underline the importance of concerned bodies promoting robust climate-resilient management measures to effectively adapt and mitigate combined anthropogenic and climate changes in the Gelna catchment.