Scenario-Based LULC Dynamics Projection Using the CA–Markov Model on Upper Awash Basin (UAB), Ethiopia

Abstract

Understanding the spatiotemporal changes in land use and land cover (LULC) in the watershed is crucial for maintaining the sustainability of land resources. This study intents to understand the historical (1972–2015) and future (2030–2060) spatiotemporal distribution of LULC changes in the Upper Awash Basin (UAB). The supervised Maximum Likelihood Classifier technique (MLC) was implemented for historical LULC classification. The Cellular Automata-Markov (CA–Markov) model was employed to project two scenarios of LULC, ‘business-as-usual’ (BAU) and ‘governance’ (Gov). Results from the historical LULC of the study area show that urban and cropland areas increased from 52.53 km² (0.45%) to 354.14 km² (3.01%) and 6040.75 km² (51.25%) to 8472.45 km² (71.97%), respectively. Whereas grassland, shrubland, and water bodies shrunk from 2052.08 km² (17.41%) to 447.63 km² (3.80%), 2462.99 km² (20.89%) to 1399.49 km² (11.89%) and 204.87 km² (1.74%) to 152.44 km² (1.29%), respectively, from 1972 to 2015. The historical LULC results indicated that the forest area was highly vulnerable and occupied by urban and cropland areas. The projected LULC under the BAU scenario shows substantial cropland and urban area expansion, increasing from 8472.45 km² (71.97%) in 2015 to 9159.21 km² (77.71%) in 2060 and 354.14 km² (3.1%) in 2015, 1196.78 km² (10.15%) in 2060, respectively, at the expense of vegetation cover. These results provide insight intothe LULC changes in the area, thus requiring urgent attention by watershed managers, policymakers, and stakeholders to provide sustainable practices for the UAB. Meanwhile, the Gov scenario indicates an increase in vegetable covers and a decrease in cropland, encouraging sustainable development compared to the BAU scenario.

1. Introduction

Land use and land cover (LULC) changes profoundly affect many fields including hydrology, ecology, global economy, agriculture, and climate [1,2,3,4]. LULC change has an adverse effect on water resources [3] by altering the hydrological cycle. It also leads to disruptions of local and regional water balances and biological cycles, leading to increased natural hazards, such as flood and drought [5]. LULC changes play a vital role in all related sustainable development issues. Therefore, simulation and analysis of LULC change are essential to understand watershed hydrology and adequately managing water resources. LULC of the watershed around the world has undergone significant changes from one class to another due to extensive anthropogenic activities [6,7,8,9,10] such as population growth, urbanization [11,12], agriculture intensification, and deforestation. Urbanization and forest destruction may reduce evapotranspiration and hydrological recycling, thus reducing rainfall. The rapid population growth and the consequent demands on land resources played a major role in driving the LULC change [12,13]. Anthropogenic activities have recently altered more than 40% of the globes’ land cover [14]. However, natural phenomena can also trigger LULC dynamics [15].

For predicting LULC changes, several models are used, which are categorized as Quantity Prediction Models (QPM), Spatial Prediction Models (SPM), and hybrid models of QPM and SPM [16]. QPM, including Markov chain [17], regression, System Dynamics (SD), and neural network models (NNM) [18]. In Xi’an city, the central part of Guanzhong China, Zhao, et.al. [19] used SD to establish a dynamic model of the citys’ Cultural ecosystem services (CES) of each administrative division. Cellar Automata (CA) [10] and the Conversion of Land Use and its Effects (CLUE) models are examples of SPM [20]. More recently, Sun, et al. [21] simulated and predicted the LULC based on CLUE and CA-Markov Models in typical pastoral areas in Mongolia. The QPM models detect the amount of land class transfers from one land cover type to another; however, it is impossible to predict the spatial structure of different classes of LULC. The spatial structure of LULC, which is the composition and configuration of various LULC classes, indicates the location-based distribution of each land category or the geographical direction of LULC change under different natural and human activities [22]. The SPM can predict the spatial allocation for identified land classes [23]. Nevertheless, the land cover type conversion cannot be determined. The LULC simulated by a single model described above could not enhance the accuracy of LULC simulation. The QPM and SPM models have been identified as having several limitations. Therefore, models that can reveal spatial projection and land class conversion quantity should be employed to overcome the limitation of quantitative and spatial models [24]. A hybrid of the QMP/SMP model enhances LULC prediction results from both a statistical and spatial perspective [25]. CA-Markov, a CA and Markov hybrid, attracted many researchers interested in LULC prediction [26,27,28,29,30,31,32]. CA-Markov models are easy to apply, integrate with GIS and remote sensing data, and have a high capacity to simulate complex LULC patterns. Besides, this model enables a more wide-ranging simulation than other LULC models. Factors and constraints can be incorporated in CA-Markov to improve simulation accuracy. In general, the combination of the CA-Markov model addresses the limitations of the Markov and CA models by considering spatial and temporal LULC dynamics [33]. These qualities make it a better model for adoption in the UAB for LULC change prediction. Refs. [28,34,35] simulated the future dynamics of land use conversion using the CA-Markov models. A Machine Learning (ML) algorithm consists of several processing elements that receive input and deliver output according to predefined activation functions. There has been a rapid increase in the use of ML approaches for analyzing remote sensing data. ML techniques have provided excellent parameter estimation, image classification, and anomaly detection results [12,36]. CA-Markov model has the advantage of simplicity compared to ML. Random forest (RF), support vector machine (SVM), and artificial neural network (ANN) models are examples of ML algorithm methods [1].

Prediction and simulation of future LULC change based on historical trends alone are highly prone to uncertainty. The uncertainty is mostly due to the lack of adequate understanding of the anthropogenic impact and political circumstances, which could be improved in the future and change local development patterns [31]. Therefore, a scenario-based methodology should be used to accommodate such uncertainties [37]. To fully represent and simulate the dynamics of LULC change, it is imperative to identify the factors responsible for those changes [38,39,40]. Knowing the potential outcomes of alternative scenarios can be a powerful tool when making and implementing difficult policy decisions [41,42]. Here, we used two scenario-based prediction approaches, namely GOV and BAU. LULC is to be managed and planned in an environment-friendly and sustainable way under the GOV scenario, while the BAU scenario assumes rapid augmentation of socio-economic trends (rapid urbanization, population growth, intensive agricultural practices, and infrastructure development) would continue [37,43]. Scenario-based prediction can bring more accurate environmental outcomes to the government and decision-makers [44]. These scenarios were developed after identifying regional land use characteristics and reviewing the literature on Ethiopias’ future plans [6,45,46,47]. The GOV and BAU scenarios were derived from future economic trends, policies, and past trends. The GOV scenario was designed based on the ‘Green Legacy Initiative’ [48], a government plan for conserving forest cover. Also, BAU was developed based on a national development policy named ‘Path to Prosperity’ [49] that promoted new cities, industrial parks, and massive infrastructure construction. The scenario selection included expert judgments from discussions and consultation with experts from different organizations. From a multidisciplinary perspective, LULC dynamics can be modeled under different scenarios to gain insight into how current decisions may affect the future of the resource [41]. It also offers a chance to assess how LULC dynamics affect the future ecosystem. Using this approach, we can gain a deeper understanding of anthropogenic disturbance and conservation. Scenarios-based simulations have proved useful for testing the effect of conservation policies, changes in socio-economic conditions, land-use policies, and changes in population growth on future land-use changes [50]. A good policy decision can be made and implemented more effectively by knowing the potential outcomes of alternative scenarios. Thus, it greatly enhances early decision-making and helps to identify limitations and opportunities associated with certain courses of action. Scenario-based modeling has been applied depending on the goals of a particular study by different researchers [31,37,43,51,52,53,54,55]. Scenario-based modelling is widely accepted in the research community and attracts government interest to consider and evaluate future policies to support sustainable development goals. Planners and policymakers can better grasp upcoming trends with scenario-based future LULC change predictions.

This study aims to (i) assess the spatial and temporal changes in LULC over the past 43 years (1972–2015) and (ii) predict the possible future dynamics of LULC under two distinct scenarios (i.e., GOV and BAU) for the years 2030, and 2060. This study is the first of its kind to predict LULC change under two scenarios, i.e., (BAU and GOV) in the UAB using the cellular automata (CA)–Markov model. Compared to other studies in UAB, the scenarios presented in this study are different. The scenarios are developed based on the Ethiopian governments’ future economic and land protection policies. Based on the ‘Green Legacy Initiative’ [48], a government program to conserve forest cover, the GOV scenario was designed. Additionally, BAU was developed based on a national development policy known as ‘Path to Prosperity’ [49] that promoted new cities and industrial parks and the construction of massive infrastructure. This study differs from previous studies, which developed two scenarios based on the governments’ policies, which will be implemented over the coming decades.

2. Materials and Methods

2.1. Study Area

The UAB is located in the central part of Ethiopia between latitudes 8°16′ and 9°18′ N and longitudes 37°57′ and 39°17′ E, with an area coverage of 11,575 km². The geography of the watershed is mountainous and hilly, with elevations ranging from 1500 to 3500 m.a.s.l. (Figure 1). This hilly and mountainous basin is a highly conserved forest area due to its inaccessibility to human activity. The basin has the most population (18.9 million in 2008) of all Ethiopian cities and is forecasted to reach 25 million by 2025. The annual precipitation and average basin temperatures from 1981 to 2017 are about 1055 mm and 17.8 °C, respectively. Cropland is the dominant land use in the basin.

2.2. Data Sources

For analyzing the dynamics and trend of LULC over the last 43-year trajectory, satellite images were acquired from the United States Geological Survey (USGS), Earth explorer (https://earthexplorer.usgs.gov/, accessed on 27 December 2022) for the year of 1972, 1985, 2000, and 2015 (

Table 1. Landsat satellite data used for LULC classification.

Satellite/Sensor	Acquisition date	Path	Row	Spatial Resolution (m)
Landsat 1 MSS	31 January 1972	181	54	60
	30 January 1972	180	54	60
Landsat 5 TM	22 November 1985	169	54	30
	21 January 1985	168	54	30
Landsat 7 ETM+	26 November 2000	169	54	15
	5 December 2000	168	54	15
Landsat 8 OLI TIRS	20 December 2015	168	54	30
	28 January 2015	169	54	30

). Due to difficulties getting a cloud-free image for a tropical country [56] like Ethiopia, the cloudless dry season was selected for the image accusation date, starting from November to February in Ethiopia [57]. Cloudy images can lead to great uncertainties during the LULC classification. Geographical data collected from different sources, roads, and railways were downloaded from OpenStreetMap (https://www.openstreetmap.org/, accessed on 27 December 2022). Digital Elevation Model (DEM) of 30-m resolution was obtained from the USGS Earth Explorer. The slope gradients and elevation were calculated from the DEM output. A detailed summary of the satellite imagery used to evaluate the historical LULC change for UAB is presented in Table 1.

2.3. Satellite Imagery Pre-Processing

In monitoring and predicting changes in the LULC, satellite images need to be pre-processed before image classification [58]. Geometric corrections, radiometric calibrations, atmospheric corrections, and image enhancements were performed to get high-quality imagery data using Earth Resources Data Analysis System (ERDAS) IMAGINE 2015 software. Geometric Corrections involve correcting geometric distortions caused by sensor-Earth geometry variations and converting the data to real-world coordinates or georeferencing [59]. The radiometric calibrations improve the image quality by adjusting digital number values; using this process, remote sensing data can be interpreted and analyzed more effectively [60]. Atmospheric corrections are implemented to eliminate atmospheric distortions caused by differences in orientation sensor parameters and noise from the acquiring platform. The atmospheric correction is necessary to get the correct and actual reflectance by removing the solar effect [61]. In order to eliminate the stripping lines on the satellite images, image enhancement processes were usually carried out to improve the image visual [11].

2.4. LULC Classification

Image classification is the most efficient technique for processing satellite images. The remote sensing image classification methods can be categorized as supervised or unsupervised. Following image pre-processing, unsupervised and supervised image classification was done to develop LULC maps for 1972, 1985, 2000, and 2015 using ERDAS IMAGINE 2015 and ArcGIS 10.5. For supervised image classification, the training stage is essential, [1] meaning that first, we must select some training pixels from each class. However, a training stage is not required in unsupervised image classification, but different algorithms are used for clustering [62]. Applying unsupervised image classification is important in developing a preliminary output of the LULC classification of the study area prior to the supervised image classification [63]. Under unsupervised image classification, no prior knowledge of the study area is required. However, unsupervised classification is much less accurate than supervised classification. Supervised image classification can achieve accurate image classification by selecting training pixels from each land class identified during the unsupervised classification process. During the supervised classification, 200 plus signature files were collected for each LULC class by creating a false-color composite to identify the actual ground representative consecutively; the same pixels are merged to create a representative ground land cover type. The supervised classification method is simple, less time-consuming, and more accurate than other image classification techniques. The supervised image classicization was produced using a maximum likelihood classifier (MLC) method [64,65]. Its availability and simplicity make MLC one of the most widely used algorithms for supervised classification [66]. This method classifies pixels in satellite images based on their probability of belonging to a specific LULC class. The method entails that all input bands have a normal distribution and all classes have equal probabilities. False-color composite maps were created using 4, 3, 2 bands for Landsat 4 TM and 7 ETM+ and 5, 4, 3 for Landsat 8 OLI (Figure A1). The false composite bands in ERDAS Imagine 2015 could be easily visualized and interpreted in Google Earth Pro since it shows the actual images from different years. According to their spectral signatures and knowledge of the area, satellite images were classified into seven broad LULC classes, and the training samples of the study area were matched with the satellite images. The land class types are urban, waterbody, cropland, shrubland, forest, grassland, and unused land, as depicted in

Table 2. LULC classes utilized in this study.

LULC Classes	Description
Urban	Urbanized areas and rural settlements
Water	A stream or river, a lake, a pond or a reservoir
Cropland	A plot of land used to grow a variety of crops
Shrubland	Chaparrals, woodlands, and savannas
Forest	Dense trees
Grassland	Dense grass, moderate grass, and sparse grass
Unused land	Terrains with loose, eroded, or bare soils

. The classified LULC map was employed for future scenario prediction.

2.5. Accuracy Assessment

Accuracy assessment is essential to understand the degree of agreement between the actual ground and the classified image [67]. The confusion (error) matrix technique and the receiver operating characteristic curve (ROC) are two of the most common ways to determine classified image accuracy. As recommended in previous studies, we utilized the error matrix method to determine an excellent and reliable LULC classification and considered an accuracy level of 85% or higher to be excellent [68,69]. A confusion matrix comprises a series of numbers arranged in rows and columns that indicate the number of pixels or polygons attributed to a given land cover class relative to what exists [70]. This technique is used to gain descriptive and analytical statistics when assessing classification accuracy [70]. As Tung, et al. [71] suggested, the average accuracy was calculated based on producer and user accuracy. Producer accuracy refers to the probability of correctly classifying a reference pixel. Meanwhile, user accuracy represents the probability of each pixel on a map reflecting the real-world class on the ground [72]. As its key evaluation indices, the confusion matrix includes the overall accuracy and the kappa coefficient [73]. A measure of overall accuracy (OA) is the ratio of correctly classified pixels to the total amount of pixels in the confusion matrix (i.e., how many major diagonal entries are included). A kappa coefficient represents the percentage of correctly classified pixels in relation to the actual percentage expected by chance. In order to assess the degree of agreement or accuracy, it uses a discrete multivariate method [74]. Kappa coefficients range between 0 and 1. Kappa coefficients between 0.00 and 0.20 indicate a slight correlation, the value of 0.21 and 0.40 represents a fair agreement, a correlation coefficient between 0.41 and 0.60 indicates a moderate correlation, an agreement of 0.61 and 0.80 is considered substantial, a correlation of 0.81 to 1 is considered almost perfect [75].

In this paper, the accuracy assessment was evaluated by crosschecking the level of agreement between the signature selected from each land-use class of the classified image and reference ground data which creates several ground truth points in ArcGIS 10.5 [68,70,76,77,78]. Different approaches, such as the Kappa coefficient, producer accuracy, and user accuracy, are widely used for the accuracy evaluation in the image classification process.

2.6. Determined Driver Factors for LUCC Prediction in the CA–Markov Model

Multiple factors affect land-use change, such as socio-economic development, land-use policy, climate variability, and other physical factors. Factors could also be global or regional. Climate variability is a global issue [5,79,80] and can be considered as a global factor. However, socio-economic factors, like roads, railways, and population growth, are regional factors that vary depending on where the study is conducted. It is essential to consider natural and human driving factors to make the LULC projection more accurate and reasonable [81]. The selection of driver variables is mainly based on knowledge of the study area. This study considers multiple drivers responsible for the future land use dynamics change [82]; distance to the railway and roads, elevation, and slope [82,83] (Figure A2). The selected factors are processed in ArcGIS 10.5 software for the LUCC prediction in the Idris selva software.

In UAB, forests are mainly protected in a mountainous area at an elevation greater than 3000 m.a.s.l.; this is due to the high elevation being inaccessible to any anthropogenic disturbance. The mountainous area plays a significant role in regenerating and conserving forest land, places with a high elevation not suitable for urban development and cultivation. Similarly, climate variables such as temperature and precipitation vary with elevation. The elevation is a good predictor that is suitable for agriculture, and the surrounding area is expected to change to agricultural land. UAB elevations are between 3565 to 1560 m above sea level; areas with more than 3000 m.a.s.l are considered less suitable for cropland in the multi-criteria evaluation map. Thus, low elevations are susceptible to converting to agricultural land. The slope is another essential factor to consider, as it can reveal whether the land is suitable for humans. For example, agriculture and urban requires a fairly gentle slope; hence, areas with mild and gentle slope might be more likely to experience land cover change. Therefore, physical attributes, mainly elevation, and slope are important factors to be considered in future LULC dynamics prediction. Infrastructure development such as roads and railways can provide access to previously remote areas and promote anthropogenic disturbance near those accesses. Urban centers tend to grow and expand as the human population increases, so the area near the current settlement quarter is frequently suspect to land change. In Ethiopia, economic factors are a significant factor driving land transformation, and our factor selection was based solely on these factors [28]. There is no reliable and timely census data available in Ethiopia, so we did not consider population growth as a factor.

2.7. Simulation of Future LULC Dynamics

Future LULC prediction requires understanding LULC transformation ‘from-to’ change of quantity and location between each prescribed period [84]. The Markov model is a stochastic model that estimates change probabilities between one class and another, considering LULC changes over time [24,85,86,87,88]. In other words, Markov chains are simply random values whose probabilities change depending on the previous number value [89]. The following equations describe Markov chains:

$$F_X\left(X\left(t_{n+1}\right)\le X_{n+1}|X\left(t_n\right)=x_n, X\left(t_{n-1}\right)=x_{n-1,\dots ,}X(t_1\right)=x_1)=(X\left(t_{n+1}\right)\le X_{n+1}|X\left(t_n\right)=X_n)$$

(1)

Markov chain process for a particular time (t), represented by X(t), tn represents the present moment, whereas tn+1 describes the time for changes in the future. Furthermore, the previous changes are denoted by t_n−1.

Equation (2) is the relation that describes the probability of state transition from i to j, as described in the preceding study [90], and the state k is represented by X[k] (k = 1,2,3,…).

$$P_{i,j}=P_r(X\left[K+1\right]=j|X\left[K\right]=i)$$

(2)

A transition probability matrix can be defined as [90] for a Markov chain that follows a finite number of states (i.e., n).

$$\left[\begin{array}{ccc}{P}_{1,1}& P_{1,2 }\dots & P_{1,n}\\ P_{2,1 }& P_{2,2 }\dots & P_{2,n}\\ P_{n,1 }& P_{n,2 }\dots & P_{n,n}\end{array}\right].$$

(3)

However, the Markov chain model omits spatial explicitness [91]. There is no scientific explanation for the change processes, and the spatial distribution of LULC is not included, which is extremely important when simulating land cover patterns. In this study, the Markov transition matrix was developed between 1972 and 1985, 1985 to 2000, and 2000 to 2015, and the probability change of 1985 and 2000 were used to project a map of 2015.

The CA-Markov chain combines CA and Markov chains for spatial dimensioning of the model, thus solving Markov chain limitations. The CA-Markov model effectively simulates land-use changes due to its combination of the contiguity of a Markov model and the advantages of cellular automata. The CA-Markov predicts the trend and spatial structure of various LULC classes [10]. It identifies the likely spatial distribution transitions [30,92].

This study predicted LULC using the CA–Markov chain model for the years 2030 and 2060. IDRISI Selva was used to monitor and simulate the potential LULC change dynamics of the basin. The data for running the CA-Markov in IDRISI Selva are historical land cover maps, transition area files, and transition suitability maps. Transition probabilities were explored using multi-temporal LULC images from 1972, 1985, 2000, and 2015. Each land cover class is given a probability of changing over a specified period in this probability matrix. In the following step, the matrices are used to predict future changes to LULC. The LULC transition suitability maps are produced by compiling a collection of maps which are factors and constraints. A multi-criteria evaluation method (MCE) is used to produce the suitability maps.

Before using the CA, a suitability map for each class is required. Transition suitability maps were created for each land cover class utilized in the study. The MCE generates suitability maps by combining information from several criteria to form a single evaluation index. The MCE method evaluates and collects weighted maps based on the experts’ knowledge of how the factors affect and interact with LULC. The UAB future suitability map for LULC was mapped, considering the factors driving LULC change and distribution. The factors considered were slope, DEM, distance to the main road, and distance to the railway. As constraints, water bodies and built-up areas are vital in constructing suitability maps. Factor and constraint maps were processed with ArcGIS 10.5 and IDRISI Selva. Standardization involves converting the factor’s values into a number between 0 and 255 [93]. Those values near 255 in the particular land class are considered highly suitable for change from one class to another; on the contrary, 0 represents unsuitable for expansion or change of the given land class.

It is believed that constraints rarely change into other types of LULC. A (Boolean) 0/1 was used to divide the images into two groups. Boolean images were assigned 1 and 0 values based on the non-constraint and constraint criteria (Figure A3). 0 implied a very low probability of change, and 1 indicated a very high probability of change. IDRISI Selva has a fussy membership function for standardizing and ensuring that they are logically consistent. The three membership functions employed in this research are: sigmoidal, J-shaped, and linear (

Table A1. Factors, membership functions, control points, and constraints of the different land use classes. MD: Monotonically Decreasing function.

LULC Class	Factors	Membership Function	Control Points	Constraints
Cropland	Slope	MD-J shape	c = 4; d = 10	Water Urban
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from road *	MD-Sigmoidal	c = 0; d = max
	Distance from railway *	MD-Sigmoidal	c = 0; d = max
Grassland	Slope	MD-J shape	c= 4; d = 10	Water Urban
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from railway *	MD-Sigmoidal	c = 0; d = max
	Distance from road *	MD-Sigmoidal	c = 0; d = max
Water	Slope	MD-J shape	c= 4; d = 10	Urban
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from road *	MD-Sigmoidal	c = 0; d = max
	Distance from railway *	MD-Sigmoidal	c = 0; d = max
Urban	Slope	MD-J shape	c = 4; d = 10	Water
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from road *	MD-Sigmoidal	c = 0; d = max
	Distance from railway *	MD-Sigmoidal	c = 0; d = max
Unused land	Slope	MD-J shape	c = 4; d = 10	Water Urban
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from road *	MD-Sigmoidal	c = 0; d = max
	Distance from railway *	MD-Sigmoidal	c = 0; d = max
Forest	Slope	MD-J shape	c = 4; d = 10	Water Urban
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from road *	MD-Sigmoidal	c = 0; d = max
	Distance from railway *	MD-Sigmoidal	c = 0; d = max
Shrubland	Slope	MD-J shape	c= 4; d = 10	Water Urban
	Elevation	MD-Sigmoidal	c = 2000; d =3100
	Distance from road *	MD-Sigmoidal	c = 0; d = max
	Distance from railway *	MD-Sigmoidal	c = 0; d = max

(*) Factors not taken into account in the ‘Governance’ scenario.

). For each fuzzy membership function, the suitability of a given landscape for a specific LULC exhibits certain shapes as factor values increase. These are the following: when the factor value becomes higher, the suitability of a land surface for a particular LULC class increases (monotonically increasing), or when the factor value becomes higher, the suitability of a land surface for a particular LULC class decreases (monotonically decreasing). Three approaches can be used to determine the weighted overlay of factors: user-defined weights, equal weights, and the Analytic Hierarchy Process (AHP) [94,95]. The AHP uses the weighted linear combination algorithm (WLC) to aggregate factors and constraints [95]. The AHP approach was employed to identify the relative potential weight of each driving factor to change the land class category. In LULC change prediction, an AHP approach has become popularly used [95,96]. Therefore, the AHP approach was used in this study. In Figure 2, the overall study framework is illustrated.

2.8. Model Validation

Performing model validation of the future changes of LULC is a fundamental component. VALIDATE, and CROSSTAB modules in IDRISI have been used to compute the correlation of Kappa indexes between the simulated and base maps of 2015. Suppose the correlation between the two maps is perfect. In that case, the correlation value will approach 1. If the simulated map shows a significant deviation from the base map, the correlation value will be near zero. The correlation value indicates the relative similarity or deviation between the two maps. When Kappa ≤ 1, it showed a strong agreement level; if Kappa ≤ 0.75, it showed a moderate level of agreement; and if Kappa ≤ 0.5, it showed a weak agreement. Therefore, before implementing the prediction map for 2030 and 2060, the performance of the CA Mark model was examined by comparing the actual and simulated (BAU) 2015 maps. After achieving good validation results of Kappa for no information (K_no), Kappa for location (K_location), and Kappa for standard (K_standard), CA-MARKOV was adopted for future simulation of UAB LULC change maps of 2030 and 2060 under both scenarios [97,98]. These indices offer information about the actual and simulated map, i.e., disagreements regarding the different allocations of categories on the two maps and the different proportions in which the categories appear on each map [99].

$$K_{no}=\frac{M_m{N}_n}{P_p-N_n}$$

(4)

$$K_{location}=\frac{M_m{N}_n}{P_m-N_m}$$

(5)

$$K_{standard}=\frac{M_m{N}_n}{P_p-N_m}$$

(6)

where N(n) is no information, M_m, N_m, and P_m are medium grid cell-level information; P_p is perfect grid cell-level information across the landscape.

2.9. Scenario-Based Projection

Scenario-based future LULC predictions are robust tools for identifying and understanding the future LULC dynamic under different policies and control majors. The potential outcomes of alternative scenarios can be a powerful tool when making and implementing difficult policy decisions [41]. This study framed GOV and BAU scenarios based on future economic tendencies, policies, and past trends. For example, the GOV scenario was designed based on the ‘green legacy initiative’ planning, which is forest cover conservation released by the government. BAU was also designed based on a national development policy named ‘Path to Prosperity,’ which marked newly developed cities, industrial parks, and massive infrastructure construction. It will tend to change or abandon vegetable land for urbanization. At the same time, under GOV, urban and crop spatial expansion is limited and protected to decline the very speedy expansion of these LULC classes. The BAU followed the historical trends of the main drivers of LULC changes, widely used in LULC change prediction. In general, the BAU considering the historical trend of urban expansion, population growth, and ecological process, will continue in the future with a significant increase in related anthropogenic activities. This study will provide a beneficial scenario for achieving UAB basin sustainability under natural and human disturbances. Scenario-based modeling is widely accepted in the research community and attracts government interest to consider and evaluate future policies to support sustainable development goals. Planners and policymakers can better grasp upcoming trends with scenario-based future LULC change predictions.

2.9.1. Governance (GOV)

The GOV scenario involves implementing effective land management strategies and policies to preserve forests. In this scenario, factors, such as the distance to the main road and railway, are not included in the modeling due to the implemented controlled and governed infrastructure development in the basin. Elevation and slope were the only physical factors that remained conserved. This scenario predicts increased forest, grassland, shrubland, and water due to serious infrastructure development and anthropogenic disturbance rules compared to the BAU scenario. Therefore, this is the scenario policymakers and stakeholders that should follow to encourage sustainable development through elaborate eco-friendly policies. Constraints, including waterbody and urban, are considered protective and reserved classes to achieve scenario-based modeling under the GOV scenario. For example, waterbody can not be altered by settlement and cropland since this scenario is favorable for the natural resource. Other land class categories were not limited to transformed to the waterbody. The employed constraints under this scenario were extracted from the 2000 base map using the ‘RECLASS’ model available in IDRSI SELVA software. Constraints are mapped as a Boolean image assigned 1 representing the non-constraint class and zero for the constraint class. Policymakers should consider the GOV scenario to reduce undesirable effects on water resources, livestock, ecological systems, human health, and economic development. If the policymakers had not applied the proper policies, future LULC would show an unprecedented loss of natural resources.

2.9.2. Business As Usual (BAU)

BAU scenarios consider the continuation of current trends of socio-economic development. Factors such as distance to the road and railway, elevation, and slope are employed to generate a suitability map under the BAU scenario. Urbanization and cropland expansion is expected to be highly accelerated in this scenario as the historical pattern. This scenario aims to determine what will happen if the current trends continue [100].

3. Result

3.1. LULC Change from 1972–2015 and Accuracy Assessment

The six identified land classes in the study area are urban, waterbody, cropland, shrubland, forest, and unused land, as depicted in Figure 3. In all four-time horizons, the predominant LULC class in UAB is cropland which covers more than half of the total area, for instance, 6040.75 km² (51.25%), 7634.33 km² (64.94%), 7937.15 km² (67.22%), and 8472.45 km² (71.97%) of the area covered by cropland for 1972, 1985, 2000, and 2015 respectively, while shrubland and forest cover 2462.99 km² (20.89%), 1939.41 km² (16.50%), 2350.42 km² (19.99%), 1399.49 km² (11.89%) and 834.67 km² (7.08%), 801.93 km² (6.82%), 500.00 km² (4.25%) 875.46 km² (7.44%) during respective periods. However, the smallest portion of the basin is water bodies, unused land, and urban cover. During the past four decades, cropland and urban have expanded at the expense of forest, shrubland, grassland, and unused land. From 1972 to 2015, urban and cropland dramatically increased from 52.53 km² (0.45) to 354.14 km² (3.01%) and 6040.75 km² (51.25%) to 8472.45 km² (71.97%). In the same study area, research carried out by Shawul, et al. [73] and Daba, et al. [24] is consistent with this finding, implying the expansion of cropland and urban at the expense of vegetable covers. A study [101,102,103] shows a significant cropland expansion. The major increment rate of cropland was revealed between 1972, 6040.75 km² (51.25%), and 1985, 7634.33 km² (64.94%). The land reform proclamation of 1975 has undergone substantial land-use policy changes, and significant changes have taken place in agricultural lands due to these new policies. The most considerable cropland expansion was found in 1985, which is a 12.75% increment compared to 1972.

Grassland was highly depleted, from 2052.08 km² (17.41%) in 1972 to 447.63 km² (3.80%) in 2015, and shrubland declined from 2462.99 km² (20.89%) in 1972 to 1399.49 km² (11.89%) in 2015. Waterbody shrinkage from 204.87 km² (1.74%) in 1972 to 152.44 km² (1.29%) in 2015. The Forest area was reduced by 2.83% between 1972 and 2000. Cutting woody species for fuelwood and charcoal consumption, construction, and commercial purpose, and moving from rural to urban is a major reason in UAB for the decline of forests and shrublands [7]. Deforestation increases the area of bear land, which causes landslides, erosion, sedimentation, increased surface runoff, river flow, and subsequent flooding, including health issues. Human health is profoundly influenced by forest biodiversity and forest ecosystems. Forest systems can be radically altered through widespread deforestation, allowing disease-causing pathogens such as parasites, bacteria, and viruses, previously unaffected, to spread.

Ethiopia is a fast-growing economy among the sub-Saharan African countries, attracting foreign and local investors, thereby causing an expansion in residential and industrial areas. Urban expansion has been accelerating since the passage of economic reforms in the last three decades, caused by the growth of the economy and the influx of migrants to urban areas. The urban area expanded from 52.53 km² (0.45%) to 354.14 km² (3.01%) between 1972 and 2015. Moreover, since 2000, the UAB has experienced extreme urban expansion due to high population growth, which made it necessary for the government to implement more housing schemes at the expense of agricultural land bordering the city [63]. Moreover, housing without a city administration permit was another cause of urbanization in the study area. It has resulted in urban sprawl due to settlement expansion. Violence and unrest in the country have recently led to the influx of migrants and encroaching on farmland to build more houses. Rapid urbanization has converted natural ecosystems, forests, and shrubland into urban areas in the UAB.

Agriculture is the livelihood and backbone for more than 85% of the population in Ethiopia. Since urban alters the fertile agricultural land, there will be an effect on the crop productivity and economy of the country. Similarly, as farmers are displaced from their former agricultural land, migration of farmers to forestry and unused land will result from wiped out and demolished vegetable cover lands because of the high demand for new agricultural land, village, and fuel consumption. Displaced farmers may not get new favorable land for agriculture; therefore, they will be required to access hilly, sloppy fragile areas. As a result, crop productivity will decrease due to the loss of topsoil, soil erosion, and land degradation. Therefore, it is imperative to prioritize the protection of fertile agricultural land to ensure the countrys’ food supply. Stringent agricultural land management strategies should be implemented, which could play a vital role in reducing fertile agricultural land use.

There was an improvement in forest coverage between 2000–2015 (3.19%) in UAB due to the community afforestation program implemented by the government. The nationwide tree planting program substantially promotes land afforestation and sustained forest growth [104,105]. Therefore community-centered policy should be widely practiced to promote successful land afforestation and sustained forest growth. Ethiopia planted 350 million trees in a day in 2019 [106] to combat the effects of deforestation and climate change in the drought-prone country. Those plated trees are expected to show their outcomes in the near future. Such tree-planting campaigns should constantly be implemented to balance urban expansion and forest protection. Furthermore, more effective preservation policies and mitigation strategies to reduce deforestation should be supported. Distributions of LULC classes in UAB from 1972 to 2015 are listed in

Table 3. Distributions of LULC classes in UAB from 1972 to 2015.

LULC	1972		1985		2000		2015
	Area (km2)	Area (%)	Area (km2)	Area (%)	Area (km2)	Area (%)	Area (km2)	Area (%)
Urban	52.53	0.45	90.62	0.77	173.86	1.48	354.14	3.01
Water	204.87	1.74	192.81	1.64	188.05	1.60	152.44	1.29
Cropland	6040.75	51.25	7634.33	64.94	7937.15	67.52	8472.45	71.97
Shrubland	2462.99	20.89	1939.41	16.50	2350.42	19.99	1399.49	11.89
Forest	834.67	7.08	801.93	6.82	500.00	4.25	875.46	7.44
Grassland	2052.08	17.41	986.89	8.39	441.63	3.76	447.63	3.80
Unused land	139.95	1.19	109.75	0.93	164.64	1.40	70.28	0.60

and

Table 4. Percentage changes in LULC classes that occurred in UAB.

LULC	1972–1985	1985–2000	2000–2015	1972–2015
Urban	+72.52	+91.86	+103.69	+574.17
Water	−5.88	−2.47	−18.94	−25.59
Cropland	+26.38	+3.97	+6.74	+40.25
Shrubland	−21.26	+21.19	−40.46	−43.18
Forest	−3.92	−37.65	+75.09	+4.89
Grassland	−51.91	−55.25	+1.36	−78.19
Unused land	−21.58	+50.01	−57.31	−49.78

.

We evaluated the accuracy and usability of the classified images based on the four kinds of accuracies user, producer, overall and kappa statistics (

Table 5. Summary of LULC classification accuracy assessment in percent (%) user’s accuracy (UA), producer’s accuracy (PA), overall accuracy, and Kappa coefficient of classified LULC of the year 1972, 1984, 2000, and 2014.

	LULC 1972		LULC 1985		LULC 2000		LULC 2015
LULC Class	UA	PA	UA	PA	UA	PA	UA	PA
Urban	75.2	74.7	90.0	87.9	96.5	85	93.4	92.7
Water	80.5	80.5	95	91.0	100.0	92.13	100.0	100.0
Cropland	76.5	86.6	78.5	88.7	65.67	88	82.3	97.6
Shrubland	75.0	75.0	75.6	87.7	79.09	87.88	97.1	77
Forest	90.4	77.7	90.1	87.4	98.9	90.88	88.0	86.8
Grassland	70.9	85.6	82.0	91.8	98.9	93.07	82.1	78.6
Unused land	76.2	72.7	96.7	88.9	96.9	90	94.3	91.9
Overall accuracy	80.6		89.04		89.41		89.2
Kappa coefficient	0.76		0.87		0.87		0.87

). The overall accuracy of 1972, 1985, 200, and 2015 LULC maps recorded 80.6%, 89.04%, 89.41, and 89.2%, respectively. To measure the accuracy and confidence in the results of the LULC map, we calculated Kappa coefficients for each category and the results were 76%, 87%, 87%, and 87% for the respective years. It generally showed acceptable overall accuracy and was used for further changes and detections.

3.2. CA-Markov Model Validation

Validation of the CA-Markov model is needed to simulate the future LULC maps of UAB. The CA-Markov model projected the 2015 LULC using an existing LULC map of 1985 and 2000, including the translation probability and suitability map configuration. The 2015 LULC map was used for validation as a reference map for the projected map of 2015. Model performance was examined by comparing the simulated and actual map of 2015 before implementing further projection [107,108]; all performance statistics k_no, K_standard, and K_locality illustrate the acceptable range. The acceptable result of performance statistics indicates the capability of the model in the particular area to hold the future projection. The model achieved kappa values, k_no 0.90, K_standard 0.87, and K_locality 0.92 (

Table 6. Level of agreement between the simulated and actual map of 2015 using the validation model.

Kappa Index	Kappa Index of Agreement (%)
Kno	90
K-standard	87
K-locality	92

). The analysis assessed with CROSSTAB module showed a good performance with KIA 0.87 (

Table 7. Level of agreement between the simulated and actual map of 2015 using the CROSSTAB model.

LULC Category	KIA 1	LULC Category	KIA
Urban	0.83	Forest	0.82
Water	0.93	Grassland	0.80
Cropland	0.81	Unused land	0.71
Shrubland	0.74
Overall KIA		0.87

¹ KIA: Kappa Index of Agreement.

). This result proves that the CA–Markov model is reliable and capable of simulating future LULC changes in the basin. The projected and actual maps of 2015 are similar and close regarding the spatial distribution and quantity of land change (Figure 4). Table 6 and Table 7 display the degree of agreement between the base and simulated maps of 2015. The projected land class, such as water, urban, and forest areas, coincide highly with the actual map. However, cropland and shrubland are respectively overestimated and underestimated, as mentioned in Figure 5 and

Table 8. LULC classes at UAB in 2015: a comparison between actual and simulated areas.

LULC Category	Actual Map of 2015		Simulated Map of 2015
	km2	%	km2	%
Urban	354.14	3.01	511.14	4.34
Water	152.44	1.29	145.12	1.23
Cropland	8472.45	71.97	7835.64	66.50
Shrubland	1399.49	11.89	1628.41	13.82
Forest	875.46	7.44	1080.85	9.17
Grassland	447.63	3.80	498.30	4.23
Unused land	70.28	0.60	83.01	0.70

. Finally, future LULC change has been modeled under two scenarios of BAU and GOV for 2030 and 2060.

3.3. LULC Transition Probabilities Matrices

There is a frequent conversion between LULC categories in LULC studies. The n-by-n matrix of LULC classes displays the changes. The columns represent the newer or projected categories of LULC, while the row classes represent previous LULC types. The nondiagonal values indicate the likelihood of land use cover changing between times 1 and 2. While the diagonal value represents the probability that a land cover class will persist between times 1 and 2. The tables below show the changes in LULC categories between 1972–1985, 1985–2000, and 2000–2015.

3.3.1. Conversion between 1985 and 2000

Table 9. Transition probabilities of LULC classes from 1985 to 2000.

1985	2000
	Urban	Water	Cropland	Shrubland	Forest	Grassland	Unused Land
Urban	0.6992	0.0081	0.1117	0.0846	0.0835	0.0083	0.0047
Water	0.0125	0.7691	0.0879	0.0910	0.0322	0.0000	0.0072
Cropland	0.0097	0.0006	0.6693	0.2034	0.0572	0.0337	0.0260
Shrubland	0.0197	0.0041	0. 5736	0.2924	0.0382	0.0640	0.0079
Forest	0.0073	0.0033	0.2859	0.4671	0.1430	0.0847	0.0089
Grassland	0.0081	0.0004	0.5887	0.3067	0.0086	0.0814	0.0061
Unused land	0.0164	0.0005	0.8003	0.1088	0.0033	0.0021	0.0686

The bold diagonals represent the probability of a given LULC class to remain stable.

illustrates the transition probabilities between 1985–2000. Land-use categories that remain unchanged from time 0 to time 1 are represented by values on the diagonal. The LULC categories with the highest probability of remaining unchanged are water urban and cropland, with 76.91%, 69.92%, and 66.93%, respectively. While unused land and grassland have a low resistance to change to other categories of land. The transition probabilities of LULC classes from 1985 to 2000 showed the conversion of Shrubland (57.36%), Grassland (58.87%), forest (28.59%), and Unused land (80.03%) to cropland occurred, but cropland was reverted to none of the three LULC categories. This result highlights the significant expansion of cropland over vegetation covers and unused land. Crop expansion in the UAB may have caused deforestation.

3.3.2. Conversion between 2000 and 2015

There was a 54.62%, 62.69%, 52.43%, and 77.90% conversion rate of shrubland, forest, grassland, and unused land, respectively, to cropland area (

Table 10. Transition probabilities of LULC classes from 2000 to 2015.

2000	2015
	Urban	Water	Cropland	Shrubland	Forest	Grassland	Unused Land
Urban	0.5991	0.0082	0.1409	0.1196	0.0922	0.0387	0.0013
Water	0.0022	0.6387	0.1421	0.1180	0.0946	0.0043	0.0002
Cropland	0.0394	0.0011	0.7021	0.1477	0.0664	0.0338	0.0096
Shrubland	0.0130	0.0016	0. 5462	0.1998	0.1619	0.0759	0.0016
Forest	0.0312	0.0016	0.6269	0.0859	0.2412	0.0118	0.0014
Grassland	0.0083	0.0001	0.5243	0.1895	0.0754	0.2020	0.0005
Unused land	0.0083	0.0051	0.7790	0.0523	0.0308	0.0175	0.1069

). The fastest population growth increased the demand for food, which accelerated crop expansion in the region resulted in significant changes to the landscape. There was a substantial amount of transformation of water, with 14.21% being transformed to cropland and 11.8% to shrubland. The conversion of a water body to cropland showed that a substantial amount of water is consumed by agriculture in GRB.

3.4. Future LULC Dynamics

The future dynamics of UAB were simulated based on the historical pattern of LULC change from 1972 to 2015 under two socio-economic scenarios. Figure 6 displays the LULC of UAB in 2030, and 2060 under the BAU and Gov scenarios. Under the BAU scenario, urban 717.67 km² (6.10%), 1196.78 km² (10.15%), and cropland 8833.65 km² (75.04%), 9159.21 km² (77.71%) showed consistent expansion in 2030 and 2060.The urban expansion will continue near and around Addis Ababa to the towns connected with roads and railways. The vegetable area will be converted into urban and agricultural areas when cities and crop expansion areas continue at the current rate. This trend suggests that urbanization and agricultural planning must be efficient and environmentally friendly. Similarly, cropland expansion will occur in the upper and middle sections of the region due to the favorable slope of the area. Forest and shrubland will remain at high elevations, not accessible for human urbanization and cropland. The BAU scenario predicts forest, shrubland, and grassland will decline sharply from 2015 to 2060, reaching 875.46 km² (7.44%), 1399.49 km² (11.89%), 447.63 km² (3.8%), and 439.95 km² (3.73%), 629.61 km² (5.34%), and 239.58 km² (2.03%), respectively (Figure 7). Compared to the GOV, under the BAU scenario, vegetable covers are expected to decline; policymakers and stakeholders should be greatly concerned. The government should implement rehabilitation and afforestation programs with the participation of the community. A serious nationwide rehabilitation program can make a real difference in having a sustainable resource [104,105]. A total of 350 million trees were planted in Ethiopia in 2019 to combat the effects of deforestation and climate change [109]. Such tree-planting campaigns should be constantly implemented to balance urban expansion and forest protection. Furthermore, it is necessary to support conservation policies and mitigation strategies that reduce deforestation.

If settlements alter cropland in the same way as in the historical pattern, crop productivity may decrease and be exposed to food insecurity. The UAB has been one of the main cultivators of vegetables and fruit in the past decades. Nevertheless, suppose the current trend management under BAU continues. In that case, most agricultural land will be abandoned due to inherent and extensive agriculture and losing fertile soil due to degradation and soil erosion. Under this scenario, the basin will face great environmental concern, mainly due to the intensive agricultural practices and urban expansion; as a result, surface water, groundwater dynamics, and soil properties will be affected. Such conditions are specifically dangerous for local farmers; hence, alleviating poverty will be difficult for food security in the UAB. It is crucial in this situation to protect the land and consider the needs of farmers whose livelihoods will be mostly affected by these changes. The region will have a high demand for food as the population increases over the coming decades. Agroforestry, crop rotation, and polyculture could be applied to overcome these challenges. Sustainable farming could be implemented with renewable energy for pumping and irrigation systems.

In the Gov scenario, a water management plan is adopted, which increases water bodies across the basin by 152.44 km² (1.29%) to 179.78 km² (1.56%) (Table 5) from 2015 to 2060. As a result of environmental protection, water resources are allocated for agricultural practices and industrial uses accordingly. In this scenario, unused land is predicted to increase in 2030, and the area could be converted into a recreation park, garden, or wildlife preserve later. There has been a steady increment in cropland, with 7134.88 km² (60.54%) and 7500.90 km² (63.6%) between 2030 and 2060. The most increase in the forest, 1537.19 km² (13.03%), and grassland, 568.35 km² (4.82%) showed. Cropland declined from 8472.45 km² (71.97%) in 2015 to 7500.90 km² (63.60%) in 2060. Urban expansion has been well controlled in the GOV 595.78 km² (5.06%) and 665.80 km² (5.65%) compared to the BAU scenario 717.67 km² (6.10%) and 1196.78 km² (10.15%) in the respective years. Figure 8 illustrates the % change in LULC under the GOV scenario. It has been noted that cropland continues to decline by 700.77 km² (15.79%) between 2015 and 2030. Meanwhile, forest (57.42%), unused land (33.77%), and grassland (63.01%) show significant growth in the LULC area. Increasing vegetable cover throughout the basin will reduce the exploitation of natural resources, soil erosion, and degradation of the environment, which will help improve food security, livelihoods, and the sustainability of the environment. Compared to the high increment of the past decade, urban expansion shows a slight increase between 2030 and 2060 (

Table 11. Projected area and percentage changes in LULC classes in UAB from 2015 to 2060 under the BAU and Gov scenarios.

LULC Category	Reference		BAU				Gov
	2015		2030		2060		2030		2060
	km2	%	km2	%	km2	%	km2	%	km2	%
Urban	354.14	3.01	717.67	6.1	1196.78	10.15	595.78	5.06	665.80	5.65
Water	152.44	1.29	144.38	1.23	114.83	0.97	149.88	1.27	149.78	1.56
Cropland	8472.45	71.97	8833.65	75.04	9159.21	77.71	7134.88	60.54	7500.90	63.6
Shrubland	1399.49	11.89	976.42	8.29	629.61	5.34	1703.06	14.45	1312.03	11.12
Forest	875.46	7.44	692.21	5.88	439.95	3.73	1378.15	11.69	1537.19	13.03
Grassland	447.63	3.8	370.24	3.15	239.58	2.03	729.66	6.19	568.35	4.82
Unused land	70.28	0.6	37.13	0.32	6.73	0.06	94.01	0.8	59.72	0.51

). In order to ensure the sustainability of the region, sensitive areas (vegetable cover) and appropriate programs must be protected as effectively as possible (Table 11).

Providing incentives for afforestation and reducing deforestation should be a key component of forest incentive measures. It can limit urban expansion by implementing a national ‘smart-growth’ zoning or ‘limit urban growth’ regulation, prohibiting land conversion into urban in non-metropolitan areas. Implementing urban containment will prevent further urban expansion around many unused land and forests while allowing natural vegetation to grow in the basin.

4. Discussion

Sustainable resource management requires a scientific understanding of future LULC patterns. A simulation of future LULC change dynamics in the UAB using a historical LULC map and relevant driver factors has been successfully performed in this study. In this study, two kinds of scenarios, namely BAU and GOV, were developed. Based on the BAU scenarios, we found that croplands will likely remain an essential and influential type of land use in the basin. Because croplands have expanded, future LULC changes will be more pronounced around vegetable covers. The expansion of croplands will therefore take over first suitable (gentle slopes favorable for farming) and then unsuitable (steep slopes) areas of the basin, making vegetable covers most susceptible to spatial changes. This will have a negative impact on the area’s potential for use, leading to reduced productivity in the long run. Those whose livelihoods depend on land might suffer as a result. Moreover, a runoff will increase if crops are expanded on steep slopes, soil erosion will worsen, and water supplies may decline in the absence of vegetable cover. The land might be degraded severely to the point that any crop production in the future will not be possible. Also, the expansion of cropland at the expense of vegetable covers led to a decline in biodiversity. In such conditions, food security in UAB will be difficult; as a result, eradicating poverty will be challenging. These changes will negatively impact farmers’ livelihoods, so protecting the land and considering their needs is crucial. The region will have a high demand for food as the population increases over the coming decades. Agroforestry, crop rotation, and polyculture could be applied to overcome these challenges. A sustainable farming system could be implemented using renewable energy for pumping and irrigation. Under the GOV scenario, cropland expansion remained stable compared to BAU due to intensification through improved farming and better land protection planning. Due to this, the basin will be more beneficial and multifunctional than it currently is.

In this study, future LULC is projected based on current trends of socio-economic pathways distributions. However, future patterns are uncertain as the human population grows, socio-economic development shifts, and natural hazards change. There is a lack of socio-economic and census data in the projection of LULC development pathways that allows for a comprehensive analysis and understanding of the process of LULC change. The last time Ethiopia conducted a census was in 2007, 15 years ago [110]. A lack of adequate and timely census data hinders the study from considering population density as a factor. In order to decrease concerns related to such lack of data consideration, we intend to conduct further research once the government releases reliable census data.

5. Conclusions

Based on historical maps from 1985 to 2015, the CA-Markov model was utilized to predict two scenarios of the future LULC system: BAU and Gov in UAB. In the first step, we compare the actual 2015 LULC map with a simulated version based on 1985 and 2000 LULC maps to test the model’s accuracy. As a result, the CA-Markov model appears suitable to project future LULC in UAB based on the validation results. There were significant LULC changes in UAB during the years 1972 to 2015. We have seen increased urban and cropland areas in the past four decades. Increases in urban areas are likely due to population growth, transportation improvements, and nearby villages’ development. Meanwhile, shrubland, grassland, and unused land have decreased, while water bodies have remained relatively stable. Making effective decisions in light of such LULC changes requires predicting future changes with various assumptions. Therefore, our study considered two LULC scenarios (BAU and Gov). According to the BAU scenario, which assumes the present socio-economic advancement trends will persist, urbanization and cropland are projected to rise substantially in 2030 and 2060. However, vegetable covers (forest, grassland, and shrubland) and water bodies are predicted to decline dramatically. As long as this scenario continues, it will negatively impact the entire basins’ ecosystem, agriculture, and water resources. In the BAU scenario, cropland expansions were met at the expense of a vegetable cover. However, this will negatively affect the potential use of the land and may eventually result in a loss of productivity. It is also likely that when croplands are expanded at the expansions of vegetable cover, more runoff will occur, soil erosion will increase, and water supplies may be lowered. Land protection is essential in this scenario, and farmers who the changes would most impact need to be considered. During the coming decades, the region is expected to experience an increase in population, leading to a high demand for food. Increasing crop productivity with modern farm inputs (fertilizer and improved seeds) and better family planning is essential to meet this demand. Water bodies within the catchment need to be buffered through appropriate management. Furthermore, it could be possible to manage water in the basin efficiently if reservoirs are constructed across the basin with enough capacity to store rainwater. Such reservoirs should be well-regulated to ensure underground water pumping. And more area will be covered by water if agricultural land can be irrigated. The Gov scenario experiences modest and gradual land cover changes; vegetable covers areas and water bodies will increase and be considerably larger in the future. In this scenario, water resources are allocated in a well-planned manner for agricultural and industrial applications. Protecting vulnerable areas (vegetable covers and water bodies) and implementing programs to protect them is very important to the thriving of the region. This scenario assumes that urban centers are limited in their growth and expansion. Hence, we strongly suggest following the Gov scenario, both economically and environmentally feasible for the UAB over the long run.

For sustainable management of resources in countries like Ethiopia and those with similar conditions, the approach we framed in our study provides a useful tool. Its flexibility, spatial explicitness, and scenario-based features can be used in various educational and research applications to study future LULC changes. Additionally, it can be used for forming informed decisions. This method may also be employed to determine which types of LULC need prior attention. In addition, it may be helpful to understand better the consequences of implementing socio-economic policies without considering the impact on future LULC dynamics. Local governments can better understand the complex LULC system through our simulation results. In addition to providing insight into the future impacts of LULC dynamics, the projected LULC maps could also provide an early warning system in the face of declining vegetables and cropland expansion. The simulation results can also be considered as a strategic guide in land use planning to achieve a more effective balance between agricultural productivity and environmental protection.

This paper provides valuable information to Awash Basin authorities and stakeholders regarding LULC planning and policy management to advance forthcoming strategies and guidance within the current state-of-the-art framework of sustainable development. Besides, the results of this study will be helpful for future assessments of climate change and anthropogenic impacts on the hydrology of the basin.