Detecting, Analyzing, and Predicting Land Use/Land Cover (LULC) Changes in Arid Regions Using Landsat Images, CA-Markov Hybrid Model, and GIS Techniques

: Understanding the change dynamics of land use and land cover (LULC) is critical for efﬁcient ecological management modiﬁcation and sustainable land-use planning. This work aimed to identify, simulate, and predict historical and future LULC changes in the Sohag Governorate, Egypt, as an arid region. In the present study, the detection of historical LULC change dynamics for time series 1984–2002, 2002–2013, and 2013–2022 was performed, as well as CA-Markov hybrid model was employed to project the future LULC trends for 2030, 2040, and 2050. Four Landsat images acquired by different sensors were used as spatial–temporal data sources for the study region, including TM for 1984, ETM+ for 2002, and OLI for 2013 and 2022. Furthermore, a supervised classiﬁcation technique was implemented in the image classiﬁcation process. All remote sensing data was processed and modeled using IDRISI 7.02 software. Four main LULC categories were recognized in the study region: urban areas, cultivated lands, desert lands, and water bodies.


Introduction
Land degradation is a global issue caused by demographic growth, inappropriate land management, deforestation, climate change, and others [1].Recently, Egypt has experienced a very complicated dynamic built-up sprawl process due to the fast population increase, which is the key driver of urbanization and LULC changes in arid lands, converting productive agricultural lands to urbanized areas, prompting concern among researchers [2].Furthermore, in Egypt, growing urbanization is linked with population increases, resulting in intense farming land use to fulfill food demand.Accordingly, some Nile Valley and Delta fertile soils have been vulnerable to degradation and have lost their fertility and productivity.Hence, the sustainable development of current agricultural lands with an expansion of reclaimed soils is necessary to achieve food security, which has become the primary concern [3].Land reclamation programs for arable land outside the Nile Valley have accelerated rapidly, aiming to increase farmed areas while relieving the pressure on fertile agricultural soils [4].Thus, for Egypt, as an arid region, the detection and projecting of LULC changes are essential for long-term management and land-use planning of natural resources.
Many factors influence LULC changes, including time, scale, politics, economics, and social and cultural factors [5].At all spatial-temporal scales, LULC changes have been recognized as significant drivers of environmental changes [6].Such changes, along with other negative changes (e.g., climate change, biodiversity loss, and water, soil, and air pollution), should be the top priorities for humans [7].Therefore, studying and monitoring the changes in LULC over time is both urgent and necessary to obtain an accurate future vision for sustainable development.Monitoring and relieving the unfavorable consequences of LULC dynamics while sustaining essential resource generation has, thus, become a top emphasis for researchers and policy-makers worldwide.Land use and land cover (LULC) describe the physical land types such as forests, wetlands, impervious surfaces, agriculture, water types, and others, as well as how humans use these land types in a region.Furthermore, LULCs are spatially dispersed due to the dynamic interactions between human activities and natural factors of ecosystems [8][9][10][11].Natural, social, and economic variables all influence the complex dynamics of LULC systems.Changes in LULC availability and distribution have a substantial influence on climate, environmental challenges, and natural ecosystem conditions [12][13][14].Furthermore, changes in global LULC are a major cause of significant concern for future LULC trends [15][16][17][18][19].Moreover, LULC changes are a critical aspect of sustainability and management of natural resources [20][21][22][23].
In many cases, human activities have unfavorable impacts on LULC in diverse territories of the world.Human-induced LULC changes are driven mainly by the requirements of communities for food and economic development due to the rapidly increasing population growth globally [24][25][26][27][28].These LULC shifts, however, have severe consequences for climate change, environmental conservation, environmental pollution, soil erosion, agricultural land production, biodiversity, and water resources [29][30][31][32][33][34][35][36][37].In the investigation region, as arid land, the main driver of LULC changes is human activities due to population increase.This is similar to the rest of Egypt, which requires supporting economic development to meet food and housing requirements.
Above all, understanding and identifying detailed historical and future LULC dynamics knowledge over a lengthy series of periods and analyzing their patterns and processes is critical for land development decision-makers and planners.LULC changes are rapidly dynamic processes worldwide.Thus, it is widely recognized that LULC trends documenting and modeling improve knowledge of historical and future LULC patterns and their repercussions and guide future land-use planning [27,38,39].
Geographic information systems (GIS) and remote sensing (RS) techniques provide helpful approaches for understanding, analyzing, and monitoring LULC over time in landscapes [40,41].Many investigations on LULCs have been conducted using these tools [42][43][44][45].Additionally, Landsat periodic imagery data for a specific area is a reliable data source that can be utilized to forecast LULC trends in that area.GIS and RS techniques are effective for displaying spatial modeling methods [46].Thus, modeling, simulation, and predicting the LULC changes using temporal-spatial data are crucial for improving LULC planning and management.Although the complexity of developing models and simulations is great, they are necessary to detect the LULC changes and analyze the causes and consequences of this phenomenon [47].Furthermore, various approaches for determining historical and prospective LULC have been established to assist stakeholders in economic improvement, land conservation, and land-use planning.In addition, several models for simulating future scenarios of LULC have been created.These models deliver appropriate approaches for detecting the spatial variability patterns in LULC.Moreover, model validation is required for an accurate assessment of LULC prediction in a particular area by comparing predicted and observed LULC changes [48].
The combined Cellular Automata (CA) and Markov Chain model (CA-Markov Model) is one widespread model used with high accuracy for analyzing LULU dynamics [27,36,49].Furthermore, the CA-Markov model's robustness provides an opportune manner for modeling LULC change dynamics spatially and temporally in complex landscapes [50][51][52].The flexibility of the CA-Markov hybrid model to combine spatial and remote sensing data, as well as biophysical and socio-economic data, promotes a more extensive, detailed, and accurate projection of LULC change transitions [53].Several studies have employed this model successfully in LULC prediction, e.g., [36,[54][55][56][57][58].CA-Markov hybrid model, which was employed in the present investigation, is an effective and widely used model among several models by researchers to detect, predict, monitor, and simulate spatiotemporal change in LULC [59][60][61][62].
The most significant stage in the CA-Markov model is the rules of transition, which depend on the training data [63].Furthermore, the applied model is affected by neighborhood class and cell size, which are also considered in obtaining optimum simulation or prediction outcomes [13].CA-Markov hybrid model can effectively incorporate remotely sensed data and GIS.This model can convert the results into spatially explicit results necessary for LULC development [64].For simulating and predicting land-use changes, several modeling approaches are used [65].Some of them are applied based on the statistical matching approach of the spatio-temporal data using the main variables of prediction.The others utilize algorithms specifically designed to assess human-environment interactions [66].However, the Markov model is the earliest statistical model that can be generated with minimal input data [67,68].In the current study, assessing and analyzing historical LULC change patterns from 1984 to 2022 and simulating and predicting future LULC trends of 2030, 2040, and 2050 in the study area using the CA-Markov model is beneficial for implementing sustainable land management [69].Furthermore, there is a scarcity of research on detecting LULC and forecasting future scenarios in LULC in arid lands such as the study region.Therefore, the main goal of this work was to detect, analyze, and evaluate the historical change dynamics in LULC from 1984 to 2022 and to simulate and predict LULC trends for 2030, 2040, and 2050 using CA-Markov spatial modeling.

Study Area
In the current study, Sohag Governorate represents the arid land study area.Sohag Governorate is located in Upper Egypt and encompasses a portion of the Nile Valley beside the eastern and western desert belts.It has an area of around 11,120 km 2 and a population of 5.45 million in 2021.The Nile River runs for about 125 km through the Sohag Governorate.The study region is situated at latitudes of 26 • 07 and 26 • 57 N, as well as longitudes of 32 • 14 and 31 • 20 E, and it is bounded on the south by Qena Governorate and on the north by Assiut Governorate (Figure 1).Climatically, the research region is located in North Africa's arid region, which is characterized by hot summers and mild winters with extremely low to nil rainfall.The monthly averages of temperatures, evaporation, and rainfall data for the study region are depicted in Figure 2. Temperature varies from 15.5 • C to 36.5 • C, and the relative humidity ranges between 35% and 61%.Following Soil Survey Staff [70], Hyperthermic is the soil temperature regime, and Toric is the moisture regime.The study area includes four LULC categories: urban areas; cultivated lands; deserts; and water bodies, as shown in Table 1.
beside the eastern and western desert belts.It has an area of around 11,120 km 2 and a population of 5.45 million in 2021.The Nile River runs for about 125 km through the Sohag Governorate.The study region is situated at latitudes of 26°07′ and 26°57′N, as well as longitudes of 32°14′ and 31°20′E, and it is bounded on the south by Qena Governorate and on the north by Assiut Governorate (Figure 1).Climatically, the research region is located in North Africa's arid region, which is characterized by hot summers and mild winters with extremely low to nil rainfall.The monthly averages of temperatures, evaporation, and rainfall data for the study region are depicted in Figure 2. Temperature varies from 15.5 °C to 36.5 °C, and the relative humidity ranges between 35% and 61%.Following Soil Survey Staff [70], Hyperthermic is the soil temperature regime, and Toric is the moisture regime.The study area includes four LULC categories: urban areas; cultivated lands; deserts; and water bodies, as shown in Table 1.beside the eastern and western desert belts.It has an area of around 11,120 km 2 and a population of 5.45 million in 2021.The Nile River runs for about 125 km through the Sohag Governorate.The study region is situated at latitudes of 26°07′ and 26°57′N, as well as longitudes of 32°14′ and 31°20′E, and it is bounded on the south by Qena Governorate and on the north by Assiut Governorate (Figure 1).Climatically, the research region is located in North Africa's arid region, which is characterized by hot summers and mild winters with extremely low to nil rainfall.The monthly averages of temperatures, evaporation, and rainfall data for the study region are depicted in Figure 2. Temperature varies from 15.5 °C to 36.5 °C, and the relative humidity ranges between 35% and 61%.Following Soil Survey Staff [70], Hyperthermic is the soil temperature regime, and Toric is the moisture regime.The study area includes four LULC categories: urban areas; cultivated lands; deserts; and water bodies, as shown in Table 1.

Water bodies
The Nile River, canals, drainage patterns, and wastewater treatment plants.

Desert lands
Bare lands, sand sheets, and rocky lands in the eastern and western parts outside the Nile Valley.

Cultivated lands
Old cultivated lands in the Nile Valley and newly reclaimed lands outside the Nile Valley.

Urban areas
Residential, commercial, industrial, and road constructions in cities and rural areas.

Data Acquisition
Remotely sensed and spatial data are trustworthy data for detecting and comprehending the dynamic causes of LULC in any terrain [71].The primary remotely sensed and ancillary data utilized in this work were four multispectral images of Landsat.These images were acquired by the thematic mapper (TM), the enhanced thematic mapper (ETM+), and the operational land imager (OLI) sensors for the years 1984, 2002, 2013, and 2022, respectively.Each of the four images used was created by mosaicking three adjacent captured images to cover the investigation region.The four images' geo-coded cloud-free digital data were obtained from the Landsat archive of the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov,accessed on 5 January 2022).The principal specifications of the images used are shown in Table 2.The study site boundary was subsetted using topographic maps at 1:50,000 scales obtained from the Egyptian Survey Authority (ESA) as a reference.For creating the digital elevation model (DEM), Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data were used after radiometric and geometric corrections, as described in Table 3. Figure 3 displays the created DEM.

Pre-Processing of Remote Sensing Data
Pre-processing remotely sensed data is required to reduce noise, prepare the data, and increase its suitability for further analysis.As previously stated, three images were employed to cover the study area, totaling twelve images covering the investigated time series from 1984 to 2022.Using the ENVI platform, all Landsat images underwent preprocessing, including layer stacking, geometric correction, mosaicking, and band extraction.First, the multi-temporal Landsat images were imported into the ENVI platform, and all spectral bands were layer-stacked.Concerning the geometric correction process, the entire datasets employed were resampled and projected to the World Geodetic System 1984 (WGS84) into the Universal Transverse Mercator (UTM), with a root mean square error (RMSE) of less than 0.5 pixels [43,72].For all images of interest, a standard geographic coordinate system was used throughout the geometric rectification process.Furthermore, choosing ground control points (GCPs) is critical for geometric rectification.The digital data for 1984 were registered using 50 GCPs that were easily identifiable on both satellite images and on the ground (map-to-image registration).In addition, the first polynomial order and nearest neighborhood resampling approaches were used to set the pixel size at 30 m for image registration.The remaining images were rectified (image-toimage registration) using the corrected image of 1984 as a reference, following the same resampling method.Furthermore, the nearest neighborhood algorithm was applied over images for the resampling process without changing the original brightness magnitudes

Pre-Processing of Remote Sensing Data
Pre-processing remotely sensed data is required to reduce noise, prepare the data, and increase its suitability for further analysis.As previously stated, three images were employed to cover the study area, totaling twelve images covering the investigated time series from 1984 to 2022.Using the ENVI platform, all Landsat images underwent preprocessing, including layer stacking, geometric correction, mosaicking, and band extraction.First, the multi-temporal Landsat images were imported into the ENVI platform, and all spectral bands were layer-stacked.Concerning the geometric correction process, the entire datasets employed were resampled and projected to the World Geodetic System 1984 (WGS84) into the Universal Transverse Mercator (UTM), with a root mean square error (RMSE) of less than 0.5 pixels [43,72].For all images of interest, a standard geographic coordinate system was used throughout the geometric rectification process.Furthermore, choosing ground control points (GCPs) is critical for geometric rectification.The digital data for 1984 were registered using 50 GCPs that were easily identifiable on both satellite images and on the ground (map-to-image registration).In addition, the first polynomial order and nearest neighborhood resampling approaches were used to set the pixel size at 30 m for image registration.The remaining images were rectified (image-to-image registration) using the corrected image of 1984 as a reference, following the same resampling method.Furthermore, the nearest neighborhood algorithm was applied over images for the resampling process without changing the original brightness magnitudes of pixels.
Before image classification, the temporal images must be radiometrically adjusted to normalize brightness variations induced by changing atmospheric conditions.Radiometric normalization was performed on the images using pseudo-invariant features (PIFs), which are spatially well-defined, spectrally stable, and radiometrically stable features.The image with the highest dynamic brightness value range, i.e., 2013, was chosen as a reference or base image for radiometric correction of other images from 1984, 2002, and 2022.

Image Classification and Accuracy Assessment
The main targeted LULC categories in the study site were built-up areas, cultivated lands (old farmlands and newly reclaimed farmlands), deserts, and water bodies.In terms of image classification, each image was classified individually before extracting primary LULC maps.Furthermore, image classification was performed employing supervised classification.For image classification, a decision tree classifier (DTC) approach was applied to integrate different remote sensing-derived indices.The DTC is defined as a machinelearning-based analysis technique comprising several classes of modeling algorithms using a tree structure, in which each node shows a test on attributes; the branch represents the test results, and the leaf node shows the target classes.In addition, the DTC is a multistage classifier that can be applied to a single image or a stack of images.It comprises a series of binary decisions to place pixels into classes.Each decision divides the pixels of the image into two categories based on the expression.Furthermore, the DTC is a hierarchical model composed of decision rules that recursively split independent variables into homogeneous zones.Therefore, we used it to determine the correct category for each pixel, with certain classes being separated during each step in the simplest manner possible (Figure 4).The DCT has been widely used for land cover classification, particularly with remote sensing technology approaches, and is often used to integrate multi-source data.In practice, the DTC concept is often translated using classification rules.Three stages are used to build the classification rules, including the following: (1) generating and perfecting knowledge and rules from experts; (2) extracting variables and rules using cognitive methods; and (3) automatically generating rules from observed data.As a result, the LULC in the study area is categorized into four LULC categories: urban areas; cultivated land; deserts; and water bodies, as defined in Table 1.The Normalized Difference Vegetation Index (NDVI), Modified Normalized Difference Water Index (MNDWI), Normalized Difference Built-Up Index (NDBI), and Dry Bare Soil Index (DBSI) were among the indices utilized.These indices were chosen according to the diverse land uses in the study area and were employed to improve the precision of the image classification process.The expressions of all indices utilized are shown in Table 4.
of pixels.Before image classification, the temporal images must be radiometrically adjusted to normalize brightness variations induced by changing atmospheric conditions.Radiometric normalization was performed on the images using pseudo-invariant features (PIFs), which are spatially well-defined, spectrally stable, and radiometrically stable features.The image with the highest dynamic brightness value range, i.e., 2013, was chosen as a reference or base image for radiometric correction of other images from 1984, 2002, and 2022.

Image Classification and Accuracy Assessment
The main targeted LULC categories in the study site were built-up areas, cultivated lands (old farmlands and newly reclaimed farmlands), deserts, and water bodies.In terms of image classification, each image was classified individually before extracting primary LULC maps.Furthermore, image classification was performed employing supervised classification.For image classification, a decision tree classifier (DTC) approach was applied to integrate different remote sensing-derived indices.The DTC is defined as a machine-learning-based analysis technique comprising several classes of modeling algorithms using a tree structure, in which each node shows a test on attributes; the branch represents the test results, and the leaf node shows the target classes.In addition, the DTC is a multistage classifier that can be applied to a single image or a stack of images.It comprises a series of binary decisions to place pixels into classes.Each decision divides the pixels of the image into two categories based on the expression.Furthermore, the DTC is a hierarchical model composed of decision rules that recursively split independent variables into homogeneous zones.Therefore, we used it to determine the correct category for each pixel, with certain classes being separated during each step in the simplest manner possible (Figure 4).The DCT has been widely used for land cover classification, particularly with remote sensing technology approaches, and is often used to integrate multisource data.In practice, the DTC concept is often translated using classification rules.Three stages are used to build the classification rules, including the following: (1) generating and perfecting knowledge and rules from experts; (2) extracting variables and rules using cognitive methods; and (3) automatically generating rules from observed data.As a result, the LULC in the study area is categorized into four LULC categories: urban areas; cultivated land; deserts; and water bodies, as defined in Table 1.The Normalized Difference Vegetation Index (NDVI), Modified Normalized Difference Water Index (MNDWI), Normalized Difference Built-Up Index (NDBI), and Dry Bare Soil Index (DBSI) were among the indices utilized.These indices were chosen according to the diverse land uses in the study area and were employed to improve the precision of the image classification process.The expressions of all indices utilized are shown in Table 4.   Accuracy evaluation processes decide the reliability of the spatial information generated from remote sensing images for accurate image classification [77,78].Remotely sensed spatial information is both reliable and accurate when used in integration with ground control points that serve as a reference [79].The preliminary geomorphologic units of the Sohag Governorate area were identified and verified through field observations after visual interpretation of satellite Landsat images.In this work, the accuracy of images from 1984, 2002, 2013, and 2022 was assessed using the visual interpretation of Landsat images [80].A semi-detailed survey was completed in 2022 to obtain more detailed spatial information about the soil patterns, landforms, and landscape characteristics as ground truth data throughout the investigated area.Moreover, various statistical methods of the error matrix can be used for accuracy assessment [81].Therefore, the accuracy of each categorized image was determined and evaluated by calculating the producer's accuracy, user's accuracy, overall accuracy, and Kappa coefficient values, as described in Equations ( 1)-( 4) [79,82,83].Furthermore, due to the nature of GIS and its integration with remote sensing (RS), the combination of GIS, RS data, and the Markov model was supported [53,84].Therefore, thematic maps of different LULC for studied periods were created using ArcGIS 10.8 software.
where A p is the accuracy of the producer; A u is the accuracy user; A o is the overall accuracy; K is the kappa coefficient; p ii is the percentage of zones of categorized and reference categories, and P +i = P i+ = ∑ m i=1 p ii is the percentage of zones of categorized categories and ground truth categories.

LULC Change Analysis Using CA-Markov Modeling
CA-Markov hybrid model was employed to simulate future LULC trends in this study.Using transition probability, the CA-Markov hybrid model, as a separate random process, forecasts the next scenario of LULC and all future scenarios depending on the current situation [85].Furthermore, the CA-Markov analysis technique is a simple statistical tool that employs a transition probability matrix depending on the influences of the neighborhoods via a spatially influenced algorithm [86,87].The CA-Markov hybrid model has been extensively and increasingly utilized in LULC prediction in recent years because of its ability to fit the complicated spatial nature [47,[88][89][90][91].However, the matrix of transition probability of each LULC class may be accurate, but the spatial distribution of the occurrences is unknown [46].This is because the Markov chain model has a deficiency in projecting the spatial assignment and distribution of LULC categories [92].Therefore, to fill this gap, the CA-Markov hybrid model was developed to enhance the precision of the change detection of different land-use categories by providing spatial dimension via cellular automata (CA) filter to the used model [93,94].
The cellular automaton (CA) model assumes that a grid cell's states are determined by the changes (dynamics) of the cell itself and its neighbors' grid cells [95,96].Thus, a hybrid Markov-Cellular Automata model can be applied to detect the spatial and temporal changes in the Sohag Governorate's different land uses with a dynamic degree of estimation by setting a dataset of land uses at a specific time, then projecting the changing probabilities of these data for a future time [43,90].Change detection in LULC changes, on the other hand, is accomplished by predicting changes in satellite image pixels from one land-use category at a time (t1) into the other at a time (t2).Equations ( 5)-( 7) express the employed models [97].
where P ij is the transitional probability matrix of changing from a specific land use studied (n) to another, and its value ranges from 0 to Markov Chain model, when combined with Cellular Automata, offers an excellent chance to forecast and simulates the spatial-temporal LULC changes.Furthermore, this integrated approach is also effective in modeling and forecasting the complex LULC categories [50,52,98].Therefore, the CA-Markov hybrid model with two variables was employed in the process of LULC simulation and prediction.The two variables used are the discrete variable in time and space and the local variable, which they usually allocate to interactions [99].Grid sizes, cell neighborhoods, cell spaces, time phases, and transition rules are the core components of the CA-Markov hybrid model [100,101].Cells are objects in any dimensional space that are adjacent or close to one another.Each cell can only be in one of the states that define the system's attributes at a time.The state of any cell is determined by the states of other cells in its neighborhood, defined as the immediately adjacent set of cells that are "next" to the cell in question.Finally, transition rules drive changes in the state of each cell as a function of what exists or is happening in the cell's vicinity.Employing the inter-period change-transition probability matrix, the CA-Markov hybrid model integrates the idea of projecting the next time state depending on the previous period states [102,103].Furthermore, this model can determine neighbors, and the greater the weight factor, the closer the distance between cells and their neighbors [43,90].Therefore, the weight factor and transition probability can forecast the states of neighbor grid cells [44].The land-use changes were calculated using Equation (8).
where Si is land-use change; LU (i,t1) is land-use change at an earlier period; LU (i,t2) is land-use changed area at a later period, and LUA i is the area with no change.Future LULC trends for future dates studied were projected based on the historical patterns (from 1984 to 2022) in the study area.CA-Markov was employed in IDRISI 7.02 software to forecast the future LULC trends in 2030, 2040, and 2050.In the present study, LULC images of the Sohag Governorate area over two-time series in 1984 as earlier land-use (t1) and 2022 as later land-use (t2) were utilized, as described by Hyandye and Martz [50].A standard contiguity filter (5 × 5 pixels) was applied to the images to identify the neighbor-hood cells of each land-use class.To achieve a good influence, each cell was surrounded by a matrix scope of 5 × 5 cells.This spatial filter occurs near an existing category and rules out changes in land use randomly [104].Furthermore, the pixels close to the existent LULC class are more suitable than those far away [105].In the current study, neighboring pixels were utilized to generate spatially neighboring weights to project LULC categories for 2030, 2040, and 2050.The contiguity filter used for LULC change analysis is shown below (Equation ( 9)).
The methodology of analyzing, simulating, and projecting LULCs employing the CA-Markov hybrid model is summarized into four main steps, which are as follows: 1.
The Predicted LULC maps for future dates were produced by overlapping the results obtained in the previous steps.

CA-Markov Model Validation
According to Eastman [106], the utility of any predictive change model is dependent on the outcome of the validation process.The CA-Markov model outputs had to be validated before modeling the future LULC trends for 2030, 2040, and 2050.Therefore, the CA-Markov model was validated using the validation module in IDRISI 7.02 software to compare the degree of agreement between the predicted and the classified maps.The validation process was performed by comparing the LULC predicted results (projected maps) for 2013 and 2022 to their corresponding observed datasets (classified maps), which were used as LULC reference data.Furthermore, Kappa coefficients were calculated and utilized to evaluate the performance precision of the applied model in the LULC projection maps.The current study applied the statistic Kappa indices, including the Kappa for no information (K no ), Kappa for location (K location ), Kappa for location strata (K location strata ), and Kappa standard (K standard ) [51,104].Kappa statistical indicator can distinguish between quantity errors and location errors between two qualitative maps [52,107] and is computed employing Equation (10) [108].
where P 0 is the portion of cells correctly classified, and P C is the hypothetical probability of opportunity agreement between the actual LULC of 2013 and 2022 maps (Landsat images classification) and the projected LULC of 2013 and 2022 maps.The employment of Kappa indices in the computation determines the overall accomplishment rate and provides insight into the true causes of the strength or weakness of the outcomes.The Kappa values were categorized according to Aliani et al. [93], as illustrated in Table 5.Furthermore, Figure 5 demonstrates the methodology for LULC analysis and LULC projection employing the CA-Markov hybrid model.outcomes.The Kappa values were categorized according to Aliani et al. [93], as illustrated in Table 5.Furthermore, Figure 5 demonstrates the methodology for LULC analysis and LULC projection employing the CA-Markov hybrid model.

Accuracy Assessment of the LULC Classification
The obtained results showed that the overall accuracy and kappa statistics of all different land-use types in the Sohag Governorate area were acceptable for the different periods studied in 1984, 2002, 2013, and 2022 (Table 6).In the current study, the classification accuracy met the criterion that there must have been at least 80% accuracy for the sensor data [109].The overall accuracy of the LULC classification ranged from 87.5% to 95.5% (Table 6).According to Anderson et al. [82], these percentages represent high accuracy; thus, the LULC maps can be utilized with reliability for LULC change dynamics analysis

Accuracy Assessment of the LULC Classification
The obtained results showed that the overall accuracy and kappa statistics of all different land-use types in the Sohag Governorate area were acceptable for the different periods studied in 1984, 2002, 2013, and 2022 (Table 6).In the current study, the classification accuracy met the criterion that there must have been at least 80% accuracy for the sensor data [109].The overall accuracy of the LULC classification ranged from 87.5% to 95.5% (Table 6).According to Anderson et al. [82], these percentages represent high accuracy; thus, the LULC maps can be utilized with reliability for LULC change dynamics analysis and forecasting.Furthermore, the kappa coefficient values indicated that the strength of agreement ranged from good to very good according to criteria adopted by Aliani et al. [93], as the Kappa coefficient values ranged from 0.71 to 0.95 (Table 6).Furthermore, Kappa values were described as a considerable agreement to near perfect agreement [98].The assessment of overall accuracy and Kappa coefficient (Table 6) demonstrated the high capacity of the decision tree classifier (DTC) approach in integrating with different remote sensing-derived indices in producing multi-temporal LULC maps.Thus, this technique's findings are supposed to be compatible and reliable for modeling future LULC scenarios [27,[110][111][112].As a consequence, the categorized images are valid for examining and forecasting the changing dynamics of LULCs in the study area [50].

The LULC Classification
According to the results of the maximum likelihood algorithm of Landsat images supervised classification, there were four LULC categories recognized in the study region: urban areas (residential, commercial, industrial, and road constructions); cultivated lands (old cultivated lands, and newly reclaimed lands); deserts (bare lands, sand sheets, and rocky land); and water bodies (the Nile River, canals, drainage patterns, and wastewater treatment plants).These findings are in line with earlier studies [113][114][115][116][117][118][119][120][121], which found the same LULC categories in other areas of Egypt, such as the Northwestern Coast, Nile Delta, and others.Table 7 displays the measurable data of the four LULC categories over the different time nodes.Furthermore, Figure 6 illustrates the spatial distribution of LULC types (classified maps) derived from satellite images of the Sohag Governorate area for the examined dates of 1984, 2002, 2013, and 2022.On the created maps, the different LULC categories were tagged with different colors.
The LULC class percentages of the total study area for 1984, 2002, 2013, and 2022 showed that deserts and cultivated lands were consistently the most extensive, followed by built-up and water bodies, which were far less extensive (Table 7).Similarly, previous investigations have reported that the LULC classes in the investigated areas of Egypt are dominated by deserts and cultivated regions [113,116,117,119].Thus, in arid land regions, these LULC categories are dominant.Furthermore, throughout the 38 years from 1984 to 2022, the area of each LULC category in the Sohag Governorate changed significantly.According to the current study, cultivated lands and desert areas have been predominant from 1984 until 2022.According to Table 7, deserts and cultivated land constituted the majority of LULCs in the study region, accounting for 1398.9 km 2 (46.8%) and 1361.0 km 2 (45.5%) of the total area in 1984, respectively, and 703.8 km 2 (23.6%) and 1814.5 km 2 (60.7%) in 2022 (Figure 6).Moreover, urban and water bodies represented the lowest proportion of LULCs, as shown in Table 7. Also, urban areas more than doubled from 165 km 2 (5.5%) in 1984 to 369 km 2 (12.3%) in 2022 (Table 7).

LULC Change Dynamics
The results indicated that the built-up areas had a net gain of 67.74 km 2 between 1984 and 2002 due to the construction of new housing as the population increased (Figure 7).Similarly, cultivated lands have gained 152.99 km 2 .Furthermore, during this period (1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002), many desert land areas were subjected to reclamation processes.As a result, cultivated lands increased while desert lands decreased.This is obvious in Figure 7, which showed a decrease in desert lands with a loss of 217.13 km 2 .Concerning water bodies, some islands in the study area had vanished completely, while new islands had appeared in other locations.Furthermore, depositional processes cause the formation of new floodplains on the river's convex sides, as well as new islands and sand bars.As a result, between 1984 and 2002, the Nile River lost 3.6 km 2 .

LULC Change Dynamics
The results indicated that the built-up areas had a net gain of 67.74 km 2 between 1984 and 2002 due to the construction of new housing as the population increased (Figure 7).Similarly, cultivated lands have gained 152.99 km 2 .Furthermore, during this period (1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002), many desert land areas were subjected to reclamation processes.As a result, cultivated lands increased while desert lands decreased.This is obvious in Figure 7, which showed a decrease in desert lands with a loss of 217.13 km 2 .Concerning water bodies, some islands in the study area had vanished completely, while new islands had appeared in other locations.Furthermore, depositional processes cause the formation of new floodplains on the river's convex sides, as well as new islands and sand bars.As a result, between 1984 and 2002, the Nile River lost 3.6 km 2 .During the time series from 2002 to 2013, water bodies in the study area increased by 8.04 km 2 (Figure 8).This could be attributed to the construction of numerous wastewater treatment plants.These treatment units are located on the outskirts of desert areas outside the Nile Valley.Furthermore, this treated wastewater is destined to irrigate crops, forest trees, decorative plants, and landscapes of greenbelts and is not intended for food.Moreover, soil reclamation activities increased dramatically during this period, consequently increasing cultivated lands by 240.73 km 2 .Therefore, the desert lands had a net loss of 320.5 km 2 in the same period (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013).Additionally, during this period, a substantial increase in built-up settlements occurred at the cost of the most productive land in the investigation area (Figure 8).According to Figure 8, there was a high sprawl rate of builtup areas on cultivated lands, which is one of the main problems that threaten the limited fertile lands in the Sohag Governorate.During this period (2002-2013), urban settlements expanded by 71.73 km 2 .Figure 8   During the time series from 2002 to 2013, water bodies in the study area increased by 8.04 km 2 (Figure 8).This could be attributed to the construction of numerous wastewater treatment plants.These treatment units are located on the outskirts of desert areas outside the Nile Valley.Furthermore, this treated wastewater is destined to irrigate crops, forest trees, decorative plants, and landscapes of greenbelts and is not intended for food.Moreover, soil reclamation activities increased dramatically during this period, consequently increasing cultivated lands by 240.73 km 2 .Therefore, the desert lands had a net loss of 320.5 km 2 in the same period (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013).Additionally, during this period, a substantial increase in built-up settlements occurred at the cost of the most productive land in the investigation area (Figure 8).According to Figure 8, there was a high sprawl rate of built-up areas on cultivated lands, which is one of the main problems that threaten the limited fertile lands in the Sohag Governorate.During this period (2002-2013), urban settlements expanded by 71.73 km 2 .Figure 8  During the third investigated period (2013-2022), water bodies in the study area increased by 33.21 km 2 , as shown in Figure 9.At this period, the population of the Sohag Governorate was estimated to be around 5,193,052, accounting for 5.2 percent of the total population of Egypt.This could result in increased water consumption and, as a result, increased wastewater pumped to treatment plants.Both soil reclamation activities and During the third investigated period (2013-2022), water bodies in the study area increased by 33.21 km 2 , as shown in Figure 9.At this period, the population of the Sohag Governorate was estimated to be around 5,193,052, accounting for 5.2 percent of the total population of Egypt.This could result in increased water consumption and, as a result, increased wastewater pumped to treatment plants.Both soil reclamation activities and urbanization increased during this period but to a lesser extent than in previous periods.The cultivated lands and urban areas recorded an increase of 59.73 km 2 and 64.52 km 2 , respectively.In addition, the desert lands experienced a net loss of 157.47 km 2 (Figure 9).During the third investigated period (2013-2022), water bodies in the study area creased by 33.21 km 2 , as shown in Figure 9.At this period, the population of the Soh Governorate was estimated to be around 5,193,052, accounting for 5.2 percent of the to population of Egypt.This could result in increased water consumption and, as a res increased wastewater pumped to treatment plants.Both soil reclamation activities a urbanization increased during this period but to a lesser extent than in previous perio The cultivated lands and urban areas recorded an increase of 59.73 km 2 and 64.52 k respectively.In addition, the desert lands experienced a net loss of 157.47 km 2 (Figure  Overall, the findings of this study revealed a continuous increase in urban sprawl and expansion of newly reclaimed cultivated lands, with a reduction in desert areas over the time series analyzed.These findings are in agreement with previous studies, which reported similar LULC change trends [113][114][115][116][117][118][119][120][121].

Markov Chain Model Analysis
The data displaying the decreases and increases in transition probabilities over time series from 1984 to 2022 are shown in Tables 8-10.The gains were acquired by subtracting the persistence from the entire column for each group, while the losses were received by deducting the persistence from the whole row [122].For instance, the probability of remaining built-up areas as built-up areas from 1984 to 2002 is 69.75%, while the future change possibility of cultivated lands to urban areas is 20.36%, and so on for the rest of the LULC types.Desert lands and water bodies had high probabilities of 76.33 and 76.49%, respectively, as they did in 2002.For further understanding of the dynamic changes in LULC types in the Sohag Governorate, the annual change in these types was calculated for the three studied periods (1984-2002, 2002-2013, and 2013-2022), as shown in Table 11.According to the data in Table 11, cultivated lands increased annually by 8.5 km 2 , 21.88 km 2 , and 6.64 km 2 for the three studied periods, respectively.The greatest annual increase in cultivated land was recorded in the second period (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013), during which increasing desert land reclamation activities were noticed.Furthermore, over the three time periods, the desert lands shrank by 12.06 km 2 , 29.14 km 2 , and 17.50 km 2 per year, respectively (Table 11).Moreover, the urban zones expanded at an annual rate of 3.76 km 2 , 6.52 km 2 , and 7.17 km 2 for the three studied periods, respectively.Water bodies decreased by 0.2 km 2 per year from 1984 to 2002, then increased by 0.73 km 2 in the second period, and the highest increase (3.69 km 2 ) occurred between 2013 and 2022 due to the construction and startup of wastewater treatment units in the study area (Table 11).The results of the LULC class changes for different LULCs in the Sohag Governorate over the studied periods are displayed in Table 12.Furthermore, Figure 10 demonstrates the temporal distribution of LULC class changes in the Sohag Governorate in square kilometers over the years studied.The findings indicate that urban areas increased during the studied periods, and there was also a noticeable reduction in desert lands.The increase in urbanized areas can be attributed to population growth.The growth in urban areas (urbanization), which is associated with population increases, infrastructure development, and growing domestic product, is one of the primary drivers of LULC changes [24,[123][124][125][126][127].Moreover, cultivated lands, water bodies, and urban areas increased during the studied periods (Figure 10).

CA-Markov Model Validation for Predicting Future LULC Scenarios
The Kappa coefficients for the quantity and location of correct cells were derived based on a comparison of the projected and classified LULC maps at a resolution of 30 m × 30 m for model validation employing the IDRISI program.According to k-indicator statistics for the LULCs in 2013, the K standard , K no , K location , and K location-strata were 0.8703, 0.9142, 0.9336, and 0.9336, respectively, as shown in Table 13.Similarly, for 2022, k-indicator statistics revealed that the K standard , K no , K location , and K location-strata were 0.8402, 0.8942, 0.9012, and 0.9012, respectively (Table 13).These findings demonstrated high agreement between the predicted and observed LULC maps for 2013 and 2022.This revealed that there were minor quantification and location errors between the projected and actual maps [104,[128][129][130]. Therefore, employing the CA-Markov model can precisely specify the quantity and location of the LULC changes [100,118,128,131].Furthermore, based on the satisfactory agreement between the predicted and observed LULC maps, the data in Table 13 and Figure 11 established that the accuracy appraisal of categorized data for 2013 and 2022 is satisfactory and reasonable for simulation and prediction applications [132].As a result, the validated CA-Markov model is nearly perfect for simulating and projecting future LULC trends in arid land regions.These findings are in agreement with the results obtained from prior investigations [27,110,113,114,133,134].The modeling results showed that built-up areas would increase significantly compared to their zones in the past and present, expanding from 165 km 2 (5.5%) in 1984 to 369 km 2 (12.3%) in 2022 to 618 km 2 (20.7%) in 2030 and 691 km 2 (23.1%) in 2050, as shown in Table 14.This predicted increase in urban areas might be related to the existing expansion in the built-up areas due to population growth [27,49,124,125,135].Furthermore, cultivated lands would decrease slightly in 2030 by 67.9 km 2 and then increase slightly to reach  The modeling results showed that built-up areas would increase significantly compared to their zones in the past and present, expanding from 165 km 2 (5.5%) in 1984 to 369 km 2 (12.3%) in 2022 to 618 km 2 (20.7%) in 2030 and 691 km 2 (23.1%) in 2050, as shown in Table 14.This predicted increase in urban areas might be related to the existing expansion in the built-up areas due to population growth [27,49,124,125,135].Furthermore, cultivated lands would decrease slightly in 2030 by 67.9 km 2 and then increase slightly to reach 1851 km 2 (61.9%) in 2050 (Table 14).The predicted expansion in cultivated lands may be associated with the expansion of newly reclaimed soils due to ongoing land reclamation projects on arable lands.Moreover, desert land would continue to decline considerably from 703.8 km 2 (23.6%) in 2022 to 459.7 km 2 (15.4%), 312.4 km 2 (10.5%), and 245.4 km 2 (8.2%) in 2030, 2040, and 2050, respectively, as displayed in Table 14.Desert lands would be reduced due to urban sprawl and cultivated land expansion outside the Nile Valley through the reclamation of arable land in desert areas.Also, according to modeling results, the water bodies would increase slightly and gradually to reach 6.7% of the entire area of the study region in 2050.
Results of the simulation indicated that future scenarios represented by urban expansion growing, cultivated lands and water bodies increasing, and desert lands declining followed the past and current trends of LULC in the study region.This may be because the CA-Markov approach does not contain the variables that could influence or prevent the event of LULC changes.Thus, the resultant LULC trend updates in the coming years are expected to be linear.Projected maps can be identical to LULCs in 2030, 2040, and 2050 if the rate of the current changes is the same.However, because the current study was limited in its use of LULC change drivers and future LULC trends are unsure, additional factors should be considered to enrich our knowledge of LULC change trends [136].Therefore, future research studies are required to assess the accuracy of these predicted scenarios and identify the variables that cause the LULC changes.Furthermore, this study's findings indicate that the Egyptian government should focus on LULC changes, specifically urban sprawl on old fertile farmlands, to ensure sustainable development in the study area.
Sustained farming should use resources in such a manner that they can renew their productivity potential while minimizing detrimental effects on ecosystems [137].In this regard, FAO [4] has indicated that rapid population growth is causing increased land degradation to meet food demands.Overall, the current study provided a future vision of LULC trends that would be beneficial in land use management and offered a comprehensive understanding of ecosystem functions for establishing land use sustainability [138].

Conclusions
The current study was carried out to assess the performance of the CA-Markov hybrid model in predicting and modeling future LULC trends in arid regions using conventional remote sensing data.Therefore, this study focused on detecting and analyzing LULC change in the Sohag Governorate region over the different periods examined (1984, 2002, 2013, and 2022), as well as projecting and modeling future LULC trends for 2030, 2040, and 2050.The findings of the current study provided evidence that the CA-Markov hybrid model is an effective model for predicting and modeling the future LULC trends for 2030, 2040, and 2050 in the Sohag Governorate as an arid region using conventional remotely sensed data.Furthermore, results proved that the cellular automata filter was critical for improving the spatial distribution of various LULC classes in the study area.Concerning the accuracy assessment of the used model, an accuracy of greater than 70% was obtained in all stages.According to this study, CA-Markov modeling has delivered promisingly precise and reliable outcomes.
Moreover, the increased population and industrial and commercial activities in the Sohag Governorate are causing considerable urban sprawl.Although the newly reclaimed agricultural lands with moderate agricultural production potential are increasing, their productivity is insufficient to maintain sustainable food security, as the old farming lands with high production potential are dwindling.This investigation has shown that changing land use and land cover is a pervasive, rapid, and significant trend in Sohag Governorate.The results revealed that cultivated lands, built-up areas, and water bodies expanded from 1984 to 2022 at the expense of decreasing deserts.Furthermore, the most considerable increase in the area of cultivated lands and built-up occurred between 2002 and 2013, while the most significant decline in desert lands happened over the same period.The predicted 2030, 2040, and 2050 LULC results also presented that the trend from historical to future LULC change will be extended to be ongoing in the future.Furthermore, this work exhibited the flexibility of integrating remote sensing, GIS, and CA-Markov modeling that can be utilized as an effective technique for mapping and tracking the LULC changes.Moreover, the change detection and prediction modeling of future LULC trends achieved in this work has some limitations.These limitations stem from the adoption of the CA-Markov approach, which does not contain the variables or factors that could influence or hinder the event of the LULC change in the study area.Therefore, future research studies are required to assess the accuracy of these expected scenarios and identify the variables that cause the LULC changes.Furthermore, future research is needed to evaluate the performance of the CA-Markov hybrid model with diverse land use classes in various arid regions.The Egyptian government should work hard to prevent the loss of productive old lands due to urban sprawl and to reclaim more unused arable land.

Figure 1 .
Figure 1.Map of the investigation region.

Figure 2 .
Figure 2. Climate data for the study region.

Figure 1 .
Figure 1.Map of the investigation region.

Figure 1 .
Figure 1.Map of the investigation region.

Figure 2 .
Figure 2. Climate data for the study region.

Figure 2 .
Figure 2. Climate data for the study region.

Figure 3 .
Figure 3. Digital elevation model (DEM) map generated from ASTER data.

Figure 3 .
Figure 3. Digital elevation model (DEM) map generated from ASTER data.

Figure 4 .
Figure 4. Decision tree for the LULC classification in the study area.

Figure 4 .
Figure 4. Decision tree for the LULC classification in the study area.
depicts the changes in LULC categories in the study area from 2002 to 2013.

Figure 11 .
Figure 11.The comparison between simulated and actual LULCs in km 2 in 2013 and 2022.

3. 6 .
Prediction of the Future LULC Scenarios CA-Markov hybrid model was used to simulate and project future LULC trends for 2030, 2040, and 2050.The future scenarios of LULC classes for 2030, 2040, and 2050 are illustrated in Figure12as thematic maps of the predicted LULC spatial distributions.Furthermore, Table14displays the percentages and the areas in km 2 of the predicted LULC categories for 2030, 2040, and 2050.During all of the predicted periods, there would be increases in water bodies, cultivated lands, and urban areas.Desert lands, on the other hand, would decline in the study area during the same periods.

Figure 11 .
Figure 11.The comparison between simulated and actual LULCs in km 2 in 2013 and 2022.

Table 1 .
Types of LULC in the study site.

Table 2 .
The principal specifications of the used Landsat satellite images.

Table 4 .
The applied indices expressions.
P 11 P 12 . . .P 1n P 21 P 22 . . .P 2n P n1 P n2 . . .P nn ij =   1; S t is the current land-use status at time t, and S t+1 is the next land-use state at time t + 1.To obtain both the transition and probability matrix of LULC types, CA-Markov Chain analysis was performed using IDRISI 7.02 software for the 1984-2002, 2002-2013, and 2013-2022 time series.
2.6.Future LULC Prediction for 2030, 2040, and 2050 Markov chain model was used to compute transition probabilities matrices for 1984, 2002, 2013, and 2022; 2. The computed transition matrices were used to generate a set of conditional probability data for the different land uses from 1984 to 2022; 3. The transition probabilities matrices of 1984-2002, 2002-2013, and 2013-2022 for each LULC category, as well as conditional probability data and LULC, classified maps of 2013 and 2020, were integrated using the CA-Markov spatial operator in IDRISI 7.02 software, which is based on Markov chain analysis and multi-criteria evaluation (MCE), to simulate the LULC maps of 2030, 2040, and 2050; 4.

Table 5 .
Kappa coefficient values and agreement degrees.

Table 5 .
Kappa coefficient values and agreement degrees.

Table 8 .
Transition probability matrix derived from LULC-classified categories in Sohag Governorate from 1984 to 2002.

Table 9 .
Transition probability matrix derived from LULC-classified categories in Sohag Governorate from 2002 to 2013.

Table 10 .
Transition probability matrix derived from LULC-classified categories in Sohag Governorate from 2013 to 2022.

Table 11 .
Annual changes in LULC classes of Sohag Governorate.

Table 12 .
Temporal distribution in km 2 of LULC distribution.

Table 14 .
The future predicted changes in areas of LULC classes in km 2 .

Table 14 .
The future predicted changes in areas of LULC classes in km 2 .