Simulation of Urban Sprawl Factors in Medium-Scale Metropolitan Areas Using a Cellular Automata-Based Model: The Case of Erzurum, Turkey

Abstract

Urban development is the planned growth of cities that takes into account ecological issues, the needs of urban life, social and technical equipment standards, and quality of life. However, as a result of policies implemented by decision-makers and users, both planned and unplanned, urban space is expanding spatially outwards from the city, while also experiencing densification in vacant areas within the city and functional transformations in land use. This process, known as urban sprawl, has been intensely debated over the past century. Making the negative effects of urban sprawl measurable and understandable from a scientific perspective is critically important for sustainable urban planning and management. Transportation surfaces hold a significant share in the land use patterns of expanding cities in physical space, and accessibility is one of the main driving forces behind land use change. Therefore, the most significant consequence of urban sprawl is the increase in urban mobility, which is shaped by the needs of urban residents to access urban functions. This increase poses risk factors for the planning period in terms of time, cost, and especially environmental impact. Urban space has a dynamic and complex structure. Planning is based on being able to guess how this structure will change over time. At first, geometric models were used to study cities, but as time went on and the network of relationships became more complicated, more modern and technological methods were needed. Artificial Neural Networks, Support Vector Machines, Agent-Based Models, Markov Chain Models, and Cellular Automata, developed using computer-aided design technologies, can be cited as examples of these approaches. In this study, the temporal change in urban sprawl and its relationship with influencing factors will be revealed using the SLEUTH model, which is one of the cellular automata-based urban simulation models. Erzurum, one of the medium-sized metropolitan cities that gained importance after the conversion of provincial borders into municipal borders with the Metropolitan Law No. 6360, has been selected as the case study area for this research. The urban sprawl process and determining factors of Erzurum will be analyzed using the SLEUTH model. By creating a simulation model of the current situation within the specified time periods and generating future scenarios, the aim is to develop planning decisions with sustainable, ecological, and optimal size and density values.

1. Introduction

Understanding cities and designing their future requires a multidisciplinary approach. Simulation models based on information technology hold a significant place in urban research for analyzing and predicting the complex network of connections that are difficult to analyze and estimate with human expertise. Cities, like many biological entities, can be described as structures exhibiting complex system behaviors resulting from the simple interactions of the small units or cells that make them up. Therefore, Cellular Automata (CA), which operate on the principle of analogy, have begun to be used in urban research [1,2]. In cellular automata-based urban models, it is assumed that there is a push-and-pull force between cells with an equal-sized grid structure and different functions (residential, commercial, industrial, educational, transportation component, etc.), which is dependent on the distance between them; these models aim to project and simulate the spatial layout of settlements using remote sensing methods, based on the relationships between neighborhood units. Eventually, computational procedures utilize mathematical schemes towards casting projections of the urban scene into future temporal phases and predicting their future configurations [3].

The need for modeling tools able to deal with the complex patterns of urban systems has emerged increasingly, particularly after a comprehensive analysis of the dynamics linked with urbanization and sprawl. Urban growth refers to the increase in population, economic development, and physical spread of an urban area, commonly representing a quantitative increase [4]. On the other hand, urban development refers to the improvements and changes occurring within the social, economic, and spatial components of the city [5]. Unlike these two concepts, urban sprawl describes the spatial expansion of the city, often in an unplanned, irregular, and uncontrolled manner, toward its surroundings [6]. Urban sprawl can reduce the efficiency of infrastructure and services and lead to environmental problems [7]. Especially in metropolitan cities where the population has not shown a significant increase, urban sprawl resulting from planning decisions and urban policies leads to an increase in the area covered by the transportation network in its macroform and in the rate of mobility due to the need for access to urban functions. This situation results in an increase in the number of motor vehicles, traffic congestion, time spent traveling, and distance covered. This artificial urbanization, which is out of sync with real population growth, moves space away from the human scale and transforms it into a vehicle scale, while constantly putting pressure on the ecological environment and soil resources on the city’s outer edges. Factors contributing to sprawl include natural (land topography, climate), social (population growth, migration, demographic structure, legal statuses such as the expansion of provincial borders with Metropolitan Law No. 6360), economic (employment, income levels, investment), and spatial factors (infrastructure access, transportation networks, and land use). The combined effect of these factors creates the complex structure of urban growth processes.

The need for urban modeling stems from the size and complexity of cities’ internal dynamics, which cannot be understood through simple observations [8]. Simulating development processes is necessary for effective urban planning and decision support mechanisms [1,9]. Nowadays, computer- and statistics-based urban models are widely used. These models include economic models [10], settlement models [8], process-based models, and statistical/spatial analysis models [11]. Advances in technologies such as Geographic Information Systems (GIS) and artificial intelligence have enabled more precise modeling [1,12].

Among the available models, Cellular Automata (CA) is especially distinguished by its ability to explain the temporal evolution of complex systems through simple rules. Initially developed by John von Neumann in the 1940s, CA is a general tool for modeling both temporal and spatial dynamics [13]. Under CA systems, the spatial area is divided into discrete cells, where the future state of every cell depends on its previous condition and those of neighboring cells. Interactions within the neighborhood generate a high amount of complexity and order in the course of time [14]. Due to the complexity of the systems of a city, combined with spatial dynamics related to diffusion, there exists a suitable backdrop for the application of CA models [15]. Cellular Automata-based urban models are used to mimic changes in land use, urbanization processes, and spatial distribution patterns [8].

In this framework, the SLEUTH model is a state-of-the-art cellular automaton (CA)-based model designed to simulate urban growth and changes in land use. It receives its name from the five essential layers it employs: Slope, Land use, Excluded areas, Urban areas, and Transportation networks [16]. Developed in the mid-1990s by Michael Clarke and his team, SLEUTH [8] stands out for its distinctive layer-based data input system. It features four key growth mechanisms-diffusion, spread, reproduction, and path effect-along with comprehensive calibration processes and the ability to create future scenarios [16,17]. Thanks to its dynamic and multi-layered structure, various growth mechanisms, and high adaptability, the SLEUTH model offers significant advantages compared to other urban growth models [3,16] and stands out, particularly in simulating spatial diffusion patterns in a detailed and realistic manner [9]. However, the model’s complexity and high data requirements can make the implementation process difficult.

For this reason hybrid models have been developed to overcome the limitations of cellular automata (CA) models. Agent-based models (ABM) place the decision-making processes of individuals and institutions at the center, thereby simulating how micro-scale behaviors translate into macro-scale urban patterns [18]. In this respect, they offer the advantage of directly incorporating socio-economic factors into the model. However, their high data requirements and the difficulty of parameter calibration represent significant limitations.

The integration of artificial neural networks (ANN) with CA allows for data-driven learning of spatial transition probabilities. This approach has proven particularly effective in capturing complex patterns derived from satellite imagery [19]. ANN-enhanced CA models have been shown to achieve high accuracy in simulating multiple land use changes. Similarly, calibrating CA with neural networks offers significant advantages for simulating complex urban systems. However, the “black box” nature of ANN limits the interpretability of the model [20].

Markov chains provide a simple yet powerful approach for predicting future land use quantities based on transition probabilities. However, since they do not explicitly account for spatial dependence, they are often combined with CA in the form of CA–Markov models [21]. Such hybrids have proven particularly effective for short- and medium-term projections.

The SLEUTH model is a classical CA-based framework that relies on slope, land use, excluded, urban extent, transportation, and hillshade layers. Its parameters (diffusion, breed, spread, slope, and road gravity) control the specific patterns of spatial diffusion [16]. However, one of the most frequently cited criticisms in the literature is its inability to directly integrate socio-economic factors [22].

In conclusion, ABM is particularly effective in capturing socio-economic processes, ANN hybrids excel in learning complex spatial patterns, Markov hybrids are suitable for short-term projections, and SLEUTH stands out in testing compact and diffused urban forms through spatial diffusion rules. There is no single “best” model; rather, model selection depends on the research objective, data availability, and spatial scale [23].

Compared to other urban growth models, the most prominent advantage of the SLEUTH model is its scalability; regardless of the spatial extent of the study area, the model can be standardized and rendered analyzable within its data framework (

Table 1. Comparison of Urban Growth Models.

Model Type	Strengths	Weaknesses	Advantage over SLEUTH	Disadvantage Compared to SLEUTH
ABM	Incorporates human behavior and policy decisions	Complex, requires extensive data	Strong socio-economic dimension	Weaker in representing physical spatial processes
CA + ANN	High accuracy, strong compatibility with satellite imagery	Black-box, low interpretability	Data-driven flexibility	Transparency and interpretability issues
CA + Markov	Simple, effective in short-term prediction	Limited validity in long-term projections	Utilizes statistical transition probabilities	Weak in reflecting physical spatial processes
SLEUTH	Strong representation of physical processes, adaptable to policy scenarios	Neglects socio-economic factors	Effective in analyzing compact/diffuse urban forms	Limited in socio-economic dimension

).

Developing dynamic cities face complex challenges such as rapid population growth, unplanned expansion, and infrastructure deficiencies. In such cities, advanced modeling tools are needed to understand and manage urban sprawl [24]. The applicability of the SLEUTH model in these cities has also been demonstrated in various studies. In cities struggling with issues like rapid population growth and unplanned expansion, the model’s ability to simulate spatial and temporal dynamics in detail provides a significant advantage. However, limitations related to data accessibility and quality, rapid change processes, unplanned development, and management capacity may affect the success of the model [25,26,27]. Nevertheless, the flexible parameter structure and adaptability of the model increase its potential in such cities.

Finally, the model exhibits a high degree of parameter sensitivity. Even small adjustments in parameters such as diffusion, breed, spread, slope, and road gravity can lead to substantially different outcomes. This makes the accurate calibration of parameters a particularly challenging task [28].

However, in this study, the time-series data prepared for the selected case area are robust, directly measured, and vector-based, which eliminates the commonly cited disadvantage of the SLEUTH model regarding data requirements. On the contrary, given the large spatial scale of the study area, the model’s superposition capability and its advantage in handling large datasets have been effectively utilized.

Although the SLEUTH model is a powerful tool for explaining the spatial patterns of urban sprawl, it has several limitations frequently highlighted in the literature. Fundamentally, the model is built upon physical and spatial datasets. As a result, it cannot directly incorporate socio-economic factors such as population growth, migration, economic dynamics, or policy decisions. This leads the model to operate solely within spatial constraints, without accounting for human behavior or social processes, which constitutes a major point of criticism [22].

In addition, the calibration of the model requires the preparation of at least four different historical urban extent maps along with detailed layers such as slope, road, and excluded. This substantial demand for high-quality data makes its application particularly challenging in developing cities [23]. Furthermore, because SLEUTH involves running a large number of simulations during calibration, the computational cost is considerably high. For large study areas or scenarios where multiple policy alternatives are tested simultaneously, the model may require several days of processing, which represents another major limitation [16].

The outputs of the SLEUTH model are highly dependent on user-defined inputs. In particular, subjective choices made during the construction of the excluded layer can exert a direct influence on the simulation outcomes. Consequently, the spatial constraints specified by the user become critical in shaping the growth patterns predicted by the model [3]. Furthermore, SLEUTH does not estimate future land demand itself; rather, it is designed to simulate the spatial configuration of urban expansion. Thus, the model provides insights not into the total magnitude of growth, but primarily into where and in what form urban development is likely to occur.

One of the most important spatial factors determining the direction and speed of urban sprawl is transportation infrastructure and road networks [10,29]. The SLEUTH model uses the transportation layer as critical data and simulates the impact of road networks on sprawl [16,29]. In this case, the “road gravity” parameter controls the effect of roads on urban growth and enables planners to model the effect of different road investments on spatial distribution [3].

Being in a position to accurately predict urban sprawl and land use change is an extremely critical requirement of environmental sustainability in city planning and management. Urban sprawl models like SLEUTH are capable of reproducing how urban sprawl would occur in different scenarios and thus enable planning in advance, environmental sustainability, economic efficiency, and social benefit [7,17,27,30,31]. However, the model’s current limitations should not be overlooked [3,22]. While high spatial accuracy and strong scenario analysis capabilities are among the significant advantages [17], the need for high-quality data [25], the requirement for technical knowledge [2], and the limited modeling of socio-economic factors [22] are the main criticisms. In the future, integrating the model with current data technologies and artificial intelligence approaches [28,32] could be possible, particularly by using high-resolution satellite imagery and advanced optimization techniques [33] to improve its performance and realism.

In this study, in line with the stated propositions, Erzurum City Center was selected as the case study area, considering the rate of urban sprawl and population growth. The study aims to simulate the selected sample area using Cellular Automata-based simulation models and to evaluate the factors causing sprawl without population pressure, especially in relation to road networks/transportation, by comparing alternative development scenarios of the modeled city.

2. Materials and Methods

2.1. Study Area

Erzurum, located in the Turkish Eastern Anatolia Region, is one of the 30 cities that are considered to be of metropolitan size. It is the fourth largest province in Turkey by area and covers an area of 25,006 km² (General Directorate of Mapping, 2025) [34]. Despite being the largest city in the Eastern Anatolia Region, it ranks 31st in Turkey in terms of population size (Turkish Statistical Institute, 2024) [35]. Erzurum, which gained metropolitan municipality status in 1994, saw its powers and areas of influence expanded with the Metropolitan Municipality Laws that came into effect in 2004 and 2009. With the implementation of the Metropolitan Municipality Law No. 6360 in 2014, 17 districts with a rural character were also included within the boundaries of the Metropolitan Municipality; thus, service began to be provided to a total of 20 districts, including the three central districts (Aziziye, Palandöken, Yakutiye) (Erzurum Metropolitan Municipality Master Plan Explanation Report, 2015). This situation has made Erzurum the central hub for service delivery, both for district municipalities and across the region. As a result, despite no significant increase in the resident population, the city center has entered a process of spatial expansion and densification.

The area covered in this study includes the main urban area of Erzurum, encompassing its three central districts (Aziziye, Palandöken, Yakutiye), as well as potential development areas (Figure 1). The area also includes the three main transportation axs surrounding the city. The Northern Ring Road, built especially in the last 25 years, has served as a significant driving force in urban sprawl and planning decisions throughout the 2004–2024 period examined in the study.

In terms of model inputs, 2004 was chosen as the base year. This year represents a period of significant transformations in plan decisions, along with changes in legal status. Unlike most similar studies, this research covers a relatively short period from 2004 to 2024, a time when urban change was intensely observed.

The selection of the intervals within the time series was designed to enhance the calibration sensitivity of the model and to capture the distinct temporal dynamics of urban growth. In particular, choosing intervals that reflect breakpoints in urban expansion (e.g., periods of rapid growth or slowdown) improves the model’s predictive accuracy and ensures that the calibration reflects the observed development trajectory.

The study period was divided into five distinct intervals covering the years 2004, 2012, 2018, 2022, and 2024. The primary reason for selecting 2004 as the starting year is that detailed preliminary assessments indicated that prior to this date, land use changes did not show a significant or qualified increase. Furthermore, the satellite images available for the pre-2004 period lacked sufficient resolution, which could negatively affect the model’s accuracy. For this reason, the data corresponding to the period when True Orthophoto images began to be produced by local administrations were considered more reliable, and 2004 was chosen as the base year.

Additionally, since the target year for projection was determined as 2045, the use of datasets from 20 years prior to the present allowed for modeling approximately 20 years into the future. This approach not only increased the reliability of the model but also strengthened and systematized comparative analyses.

2.2. Modeling Approach

The modeling approach consists of two distinct phases: performing spatial analyzes to generate the input data required for the SLEUTH model, and identifying metric differences and change rates in spatial data across the model input years.

The modeling approach employed in this study differs from classical statistical or trend-based methods in that it is founded on spatial process simulation through Cellular Automata (CA). The SLEUTH model captures not only the magnitude of urban growth but also its spatial form and diffusion characteristics by incorporating the dynamic interactions of spatial data [8]. In this respect, it diverges from conventional approaches based on fixed regression coefficients and directly integrates spatial constraints (excluded areas), slope factors, and the role of transportation networks in shaping urban expansion.

This approach has the capacity not only to estimate the overall magnitude of urban expansion but also to predict the directions of growth, the corridors along which it is likely to occur, and the constraints under which it develops. In medium-sized and dynamic cities such as Erzurum, population growth alone is insufficient to explain urban sprawl; instead, factors such as service provision, changes in administrative status (e.g., the Metropolitan Law No. 6360), and the spatial influence of transportation infrastructure play a decisive role in shaping the urban form. For this reason, the SLEUTH model employed in this study was chosen not only to forecast future growth trends but also to evaluate policy alternatives under varying levels of spatial constraints and scenarios.

The modeling approach was designed in two stages. In the first stage, the spatial input datasets required by SLEUTH (urban extent, slope, excluded, transportation, hillshade, and land use) were generated, and annual land use/land cover data were utilized for model calibration. In the second stage, scenario-based testing, calibration, and projection processes were carried out, thereby enabling the identification of future urban growth probabilities under different levels of spatial constraints.

The model mechanism is built upon five key parameters that govern the probability of each cell’s urbanization: Diffusion (D), Breed (B), Spread (S), Slope Resistance (Sl), and Road Gravity (Rg).

2.2.1. Design of Model Input Data

During the preparation of the model input data, the years through which changes occurred in Erzurum’s urban planning decisions and plan revisions were carried out and the years during which True Orthophoto shoots at 30 m resolution were conducted were considered. Since satellite images were found to be insufficient for generating input data, these criteria were decisive in creating the dataset. In this regard, the review period has been divided into five separate time segments, covering the years 2004, 2012, 2018, 2022, and 2024.

The study area covers an area of 52,440 hectares and measures 27,600 m in the x-direction and 19,000 m in the y-direction. All data was prepared in NETCAD 8.5.1 software in vector format and at a 1:1 scale, and then transferred to the ArcGIS environment. In ArcGIS, attribute information has been added to the data and the layers have been made functional. All input data was prepared using the UTM coordinate system with a slice width of 3°, the ITRF96 datum, and slice number 42. The SLEUTH model is a C-language-developed, UNIX-based open-source software that runs the Urban Growth Model (UGM) and the Land Cover Deltatron (LCD) modules together [36]. Model inputs are recorded in 8-bit radiometric format and in accordance with the naming system predicted by the model. Pixel-based rasters were created with equal row and column widths; the data was processed by the SLEUTH model using classification, weighting methods, and color bands [29]. All data for the model was uploaded with the same pixel size and in raster GIF format. The data characteristics are given in

Table 2. Data layers and resources created for the SLEUTH model.

Data Quality	Data Source	Data Range
Slope	1/1000 Scale Current Topographic Map, Digital Elevation Model	-
Land Use	Nazım Imar Planı Analysis Maps, Soil Capability Maps, Agricultural Land Classes Maps, Areas to be Protected Maps	2004–2024
Excluded	Land Use Data	2024
Urban	Zoning Plans, TRUE ORTHOPHOTO Images	2004–2012–2018–2022–2024
Transportation	1/1000 Scale Zoning Existing Topographic Map, Digital Elevation Model Plans, TRUE ORTHOPHOTO Images	2004–2012–2018–2022–2024
Hillshade	-	-

.

In similar studies, the accuracy of model input data is often tested through the calculation of kappa coefficients and overall accuracy percentages, which is considered a standard procedure particularly for datasets obtained from satellite imagery and other open-access providers. This is because such datasets are typically generated through classification techniques and therefore may contain inherent uncertainties arising from factors such as spatial resolution, classification errors, or cloud cover. In particular, land use data derived from the classification of satellite images is prone to reduced precision in time series analyses, leading to variability in accuracy levels across different periods. Consequently, this approach may constrain the reliability of model outputs and exert a direct influence on the interpretation of the results.

In this study, all land use data were derived not from satellite imagery but from maps and plans that were directly measured and digitized at a 1:1 scale within CAD environments across the entire time series. Instead of relying on satellite classifications, true orthophotos and base maps produced by local authorities were utilized. The slope and hillshade layers were generated from a Digital Elevation Model (DEM), which itself was constructed by combining terrestrial and photogrammetric methods based on contour lines extracted from 1:1 scale topographic maps (Figure 2). Similarly, the transportation layer was digitized in CAD according to the temporal progression of city plans, adhering precisely to the classification system applied in the study (Figure 3). All datasets were originally created in CAD format and subsequently converted into raster format to ensure compatibility with the SLEUTH model. Furthermore, the land use datasets produced by the model for projection years were re-analyzed in the ArcGIS environment, confirming their accuracy through detailed examination of total area losses (Figure 4). These changes in land use were also systematically presented in tabular form for both the observed input years and the simulated years across all three scenarios.

2.2.2. Analysis of Current Data in Model Input Years

The slope layer was created in the ARCGIS environment using the spatial analysis module, based on contour lines from 1/1000 scale existing maps and Digital Elevation Model (DEM) data (Figure 5). The slope data prepared in 5 separate classes was weighted appropriately for these classes and converted to raster format. It was observed that 70% of the area has a slope value of 0–20% and is suitable for urbanization in terms of slope (

Table 3. Classification of the Slope layer created for the SLEUTH model.

Slope Class	Slope Value	Pixel Count	Area Size (Ha)
0	0–10%	16,818	33,639
1	10–20%	4747	9494
2	20–30%	3024	6048
3	30–40%	1872	3744
4	>40%	1452	726

).

The change in urban area was examined in 5 different periods (2004–2012–2018–2022–2024) between 2004, the starting year of the input data, and 2024, and it was observed that the urban area has a continuous increasing trend (Figure 6,

Table 4. Classification of the Urban layer created for the SLEUTH model.

	Urban Area		Non-Urbanized Area		Total
Years	Area Size (Ha)	Percentage Rate %	Area Size (Ha)	Percentage Rate %	Area Size (Ha)	Percentage Rate %
2004	5559.96	10.60	46,880.04	89.40	52,440.00	100.00
2012	7700.29	14.68	44,739.71	85.32	52,440.00	100.00
2018	9345.12	17.82	43,094.88	82.18	52,440.00	100.00
2022	9593.22	18.29	42,846.78	81.71	52,440.00	100.00
2024	9891.85	18.86	42,548.15	81.14	52,440.00	100.00

).

It is observed that there was a 78% increase in the urban area over a 20-year period. Land use data was examined in 12 separate classes between the starting year of 2004 and the ending year of 2024 (Figure 7).

As a result of the increase in urban areas, it has been observed that pasture areas, Class 1 agricultural lands, agricultural lands in other agricultural classes, and areas to be afforested are the most significantly decreasing areas. There has been a total decrease of 15% in these areas (

Table 5. The change in the Land Use layer created for the SLEUTH model over the years.

Land Use Classes	Year 2004		Year 2024
	Area Size (Ha)	Percentage Rate %	Area Size (Ha)	Percentage Rate %
Urban Area	5559.96	10.60	9891.85	18.86
Tourism Area	8492.44	16.19	8269.20	15.77
Military Areas	919.70	1.75	918.40	1.75
Airport	863.55	1.65	863.55	1.65
Cemetery	86.20	0.16	85.02	0.16
National Park Area	387.53	0.74	387.53	0.74
Forest Area	594.20	1.13	594.20	1.13
Areas to be Afforested	491.90	0.94	354.21	0.68
Wetlands	1522.10	2.90	1522.10	2.90
1st Class Agricultural Lands	4847.42	9.24	4110.88	7.84
Other Agricultural Lands	14,062.15	26.82	12,646.81	24.12
Grazing Areas	14,612.85	27.87	12,796.25	24.40
Total Area	52,440.00	100.00	52,440.00	100.00

).

The model limits land use classes to a limited number so that spatial complexity can be managed. The land use layer, defined and delineated vectorially as 12 classes in the input data design, was reduced to the codes because the SLEUTH model can take input data in raster format with classes coded between the range of 0–7. However, since there were no barren lands or water surfaces in the study area, the model was entered with 5 classes (0, 1, 2, 3, 6).

The roads (Transportation) from the SLEUTH input data layers were examined in three separate categories: First-Order Roads (width > 40 M), Second-Order Roads (25–40 M), and Third-Order Roads, and were entered into the model with different weights (Figure 8).

Again, the roads were included and the layer was arranged separately for the years 2004–2012–2018–2022–2024, and the total change was examined. Between 2004 and 2024, the total road length increased by 43% (

Table 6. The change in the Transportation layer created for the SLEUTH model over the years.

	First-Class Roads		Second-Class Roads		Third-Class Roads
Years	Length (km)	Percentage Rate %	Length (km)	Percentage Rate %	Length (km)	Percentage Rate %	Total %	Total Length (km)	Change Rate %
2004	132,966.00	52.89	47,284.00	18.81	71,159.00	28.30	100.00	251,409.00	0.00
2012	146,760.00	51.88	60,865.00	21.51	75,277.00	26.61	100.00	282,902.00	11.13
2018	165,609.00	48.62	76,590.00	22.49	98,402.00	28.89	100.00	340,601.00	16.94
2022	166,067.00	48.67	77,006.00	22.57	98,113.00	28.76	100.00	341,186.00	0.17
2024	169,823.00	47.29	85,136.00	23.71	104,113.00	29.00	100.00	359,072.00	4.98

).

2.2.3. Development of Alternative Scenarios

The External Area Layer, which plays a decisive role in guiding urban development, has been differentiated in three separate scenarios with the aim of presenting alternative development proposals. In this regard, three alternative scenarios have been created as follows:

• Normal Growth Scenario (NGS): In this alternative, Wetlands, Forest Areas, Areas to be Afforested, National Park Areas, Cemeteries, Tourism, Airports, and Class 1 Agricultural Areas, which must be protected due to their legal status, have been designated as protected areas, and it is anticipated that the city will not expand into these areas.
• Protective Growth Scenario (PGS): In the protective alternative, Wetlands, Forest Areas, Areas to be Afforested, National Park Areas, Cemeteries, Tourism, Airports, and First-Class Agricultural Areas, as well as all other agricultural areas, have been excluded from urbanization in the excluded layer prepared for the model. In this alternative, both legal boundaries are adhered to and ecologically protected areas are excluded from development.
• Urban Sprawl Growth Scenario (SGS): In the urban sprawl alternative, it is assumed that all areas outside of Wetland Absolute Protection Zones (Buffer Zones can be opened to urbanization), Forest Areas, National Park Areas, Cemeteries, Airports, and other areas protected by special laws can be urbanized. In this alternative, environmental concerns are not considered, and urban thresholds are significantly relaxed (Figure 9).

The Excluded layer used in the SLEUTH model plays a decisive role in guiding urban development. In this study, the layer was restructured according to three separate scenarios in order to evaluate the spatial effects of different development policies. Each scenario is based on the principle of differentiating the boundaries between areas that can be opened up for urbanization and areas that must be protected for legal and ecological reasons. Thus, a wide range of alternative development scenarios has been created, spanning from a protective approach based on legal boundaries to an expansionist approach that minimizes environmental constraints.

Table 7. External field layer according to alternative scenarios created for the SLEUTH model.

Scenarios	Urbanizable Area Size (Ha)	Excluded Area Size (Ha)	Total Area (Ha)
Excluded 1	35,734.19	16,705.81	52,440.00
Excluded 2	23,187.09	29,252.91	52,440.00
Excluded 3	48,674.40	3762.60	52,440.00

presents comparative values for the size of urbanizable areas, excluded areas, and the total study area for each scenario.

2.2.4. Model Construction

Land data in SLEUTH are represented as a grid structure consisting of cells, and a cell can switch between urban and non-urban status [16]. The probability of the cell turning urban is a function of the current state of the cell, neighboring cell states, environmental factors (e.g., slope, road access), model parameters (diffusion, spread, reproduction, attraction along roads, resistance due to slope) [8]. Neighborhood relationships are typically defined in the form of Moore (8-cell) or Von Neumann (4-cell) neighborhoods. The urban nature of neighboring cells influences the formation of the growth pattern by increasing the likelihood of urban development in the relevant cell [15].

SLEUTH models make preferential use of Moore neighborhood as it allows for a finer spatial spread analysis by including both the immediate and the diagonal neighbors of a cell [8]. The Moore neighborhood encompasses the eight cells surrounding the central cell (up, down, right, left, and four diagonal corners), thereby increasing the realism of the growth pattern through horizontal, vertical, and diagonal interactions [37].

Based on analyses performed on the input data, a literature review, and extensive testing of the model with different parameter combinations, study-specific parameters have been determined. Evaluations specifically for the city of Erzurum have revealed that it has grown in a compact manner and through edge growth over the past 20 years, without establishing new settlement centers. It has been observed that the slope factor is generally low; sloping areas are located around Palandöken Mountain, and these areas are already defined in the excluded layer. The city’s road network has a compact structure, excluding ring roads; however, it has been determined that road development has a moderate-to-high impact on urban sprawl. The city’s morphological structure was considered in determining the parameters, and scenarios for creating new centers or leapfrog growth were not included due to the settlement structure expanding without population growth.

After testing the model inputs, the calibration process was carried out in three stages: pre-calibration, fine calibration, and final calibration [29,36]. The other expressions in the parameters are SLEUTH’s five growth coefficients: path gravity, spread, diffusion, reproduction, and slope resistance. The road gravity coefficient determines development along road sections. The spread coefficient (spreed) describes urban development around previously and newly urbanized cells. The slope resistance coefficient determines the likelihood of urbanization in steeper areas. While the growth around existing and newly urbanized cells is established by the diffusion coefficient, breeding coefficient establishes the likelihood of spontaneously developed cells to create a new urban center, and the diffusion coefficient establishes the outwards limitation on urban expansion [38] (

Table 8. SLEUTH model parameters.

	Parameter	Start	Step	Stop	Explanation
Coarse Calibration Values	Diffusion	20	1	40	Controlled spread, no disconnected jumps.
	Breed	40	1	50	No new core, center-oriented.
	Spread	50	1	60	Balanced edge growth from the center.
	Slope Resistance	0	1	20	Flat space, less obstacle.
	Road Gravity	50	1	70	Environmental road extraction, compact structure.
Fine Calibration Values	Diffusion	20	1	30	Controlled, disconnected spread.
	Breed	40	1	50	Prevents new core formation.
	Spread	50	1	60	Balanced growth around the center.
	Slope Resistance	5	1	15	Light slope resistance, flat area.
	Road Gravity	20	1	40	Middle-level road effect, compact structure.
Final calibration values	Diffusion	20	1	20	Controlled spread.
	Breed	45	1	45	Center-oriented growth.
	Spread	55	1	55	Balanced edge fullness.
	Slope Resistance	10	1	10	Slight obstacle.
	Road Gravity	30	1	30	Controlled road extraction.
Other parameter values	Road Gravity Sensitivity	Value 10			Low shooting sensitivity.
	Slope Sensitivity	10			Slight slope sensitivity.
	Critical Low	0			Slope constraint lower bound.
	Critical High	100			Slope constraint upper bound.
	Critical Slope	30			Over 30% is completely excluded.

).

These parameters are expressed by the following equations:

• Spontaneous Growth (sng): P_sng = D/100;
• New Spreading Centers (breed): P_breed = (B/100) × P_sng;
• Edge Growth (og): P_edge = (S/100) × f(slope);
• Slope Resistance: f(slope) = 1 − (Sl × slope/100);
• Road-Influenced Growth (rt): P_road = (Rg/100) × g(distance to road).

Through this mechanism, the model simultaneously evaluates the emergence of random new urban seeds (diffusion), new spreading centers derived from existing settlements (breed), growth along the edges of current urban areas (edge/organic growth), the inhibitory effect of slope (slope resistance), and the attraction of road networks (road-influenced growth). Thus, the model not only measures spatial differences in the data but also generates future urban growth forms and scenarios, endowing it with a dynamic predictive capability.

Using the model input data, the test, calibration, and prediction stages were run separately for three different development scenarios in the SLEUTH model. The success of the urban growth model in the SLEUTH model is achieved through the stages of testing, calibrating, and making future predictions. These stages assess and improve how accurately the model simulates real-world urban growth.

Testing Phase

The model is tested by comparing it to historical land use data. At this stage, it is evaluated whether the model’s outputs are consistent with the real data. The testing process examines the accuracy of model parameters and assumptions.

Calibration Stage

Calibration is the process of changing the model’s parameters so that its predictions about urban growth are as close as possible to the real data. Most of the time, the process is performed over and over again, each time with a new set of parameters. In each iteration, the model’s accuracy is checked using different metrics, and the results are compared to real trends in how cities grow.

One of the important tools used in the calibration process of the SLEUTH model is cluster analysis and the resulting cluster table. This research attempts to measure spatial growth patterns in the outputs of models by clustering cells with similar spatial characteristics. Cluster analysis reveals patterns of urban development intensity and distribution, indicating whether growth is occurring in a more compact or scattered manner [17]. Thus, the effect of calibration parameters on urban sprawl can also be analyzed in terms of spatial density.

In the SLEUTH model, clusters tables are generated during the calibration stage by recording performance metrics (e.g., Leesallee, Compare, Population, Edges, Clusters) for each parameter combination. These tables quantitatively indicate which parameter ranges produce more realistic spatial patterns, thereby increasing transparency in the calibration process and supporting the identification of optimal combinations [16].

The control stats tables, on the other hand, summarize how well the simulated growth in each iteration matches the observed development. They enable systematic comparison between model outputs and empirical data while allowing individual evaluation of metrics such as Leesallee, Pop, and Edges to quantitatively test model accuracy [3]. Taken together, clusters tables reveal the details of parameter calibration, whereas control stats tables demonstrate the overall performance of the model.

During the model calibration phase, 13 different performance metrics are used to evaluate the similarity of its outputs to real-world data. These criteria reveal how accurately the spatial and structural characteristics of urban sprawl are simulated:

• Product Compare: A product-based comparison of the model output with actual land use maps. A high value indicates strong overlap.
• Pop: Measures the agreement between the model’s predicted urban area size and the actual urban area.
• Edges: Determines the similarity level of urban area boundaries; reflects similarity in shape and distribution.
• Clusters: Evaluates the degree of clustering in urban development. High value means a successful spatial clustering simulation.
• Size: This shows the model-real data agreement in terms of the average size of urban clusters.
• Leesalee: Measures the structural similarity of spatial patterns; reveals the spatial accuracy of the model.
• Slope: Indicates the degree of agreement between the model output and the terrain slope.
• %Urban: This compares the percentage of urban area predicted by the model to the actual percentage of urban area.
• Xmean and Ymean: The average coordinates of the spatial distribution of urban areas; indicate the accuracy of the growth’s positional center.
• Rad (Radial Distance): The average distance of urban areas from the center.
• Fmatch (F-Measure): A general measure of the overlap ratio in urban areas; a high value reflects the model’s level of success [16].

The combined evaluation of these criteria reveals how well the model captures not only the total urban area but also its spatial arrangement, density, distribution, and morphological characteristics.

The fundamental elements that demonstrate the accuracy and performance of the model are the metrics it generates internally, which establish the correlation between the real city and the simulated city.

The calibration process involves iterative refinement of growth coefficients. At the end of each stage, the model generates metrics such as composite score, compare, population, edges, mean cluster size, Leesallee, slope, and urban clusters, all of which reflect the degree of agreement between observed and simulated growth. Coefficient ranges are then selected based on these metrics, with Leesallee serving as the primary selection criterion in this study. Improved coefficients from each stage are carried forward to subsequent calibration and prediction phases [39].

In the calibration results of the study area, the Edges value (0.3835) indicates that urban growth occurred largely as contiguous expansion around existing built-up areas. The Edges metric typically yields higher values in scenarios dominated by fragmented satellite settlements, while lower values are associated with compact development [3,16]. In Erzurum, the relatively low edges value highlights a strong tendency toward compact growth without significant fragmentation.

The Cluster values obtained in the study area ranged between 0.6 and 0.8 across the three scenarios. This demonstrates that, despite varying spatial restrictions, the model consistently captured the overall dynamics of urban growth. In other words, although the excluded scenarios altered the spatial direction of expansion, the model successfully preserved the representation of compactness, continuity, and growth intensity [22].

With respect to the Leesallee metric, the study area produced values of 0.610 in the first scenario with moderate restrictions, 0.716 in the environmentally sensitive scenario with strict restrictions, and 0.576 in the low-restriction scenario (

Table 9. Final Calibration Metric Values.

Final Calibration
Scenario 1 (NGS)	Edges	0.24681
	Cluster	0.63685
	Leesale	0.61009
Scenario 2 (PGS)	Edges	0.38350
	Cluster	0.80818
	Leesale	0.71670
Scenario 3 (SGS)	Edges	0.29693
	Cluster	0.63798
	Leesale	0.57645

). These results confirm that stronger spatial constraints enhance model accuracy, whereas weaker restrictions lead to more scattered and less reliable spatial patterns [16,29].

2.3. Prediction Stage

Once the calibration is complete, the model simulates future urban growth scenarios. At this stage, spatial-temporal projections of urban sprawl are generated with optimized parameters (best fit). The forecasts include various scenarios for use in urban planning and policy development processes [36]. The best-fit values used in the prediction stage of the SLEUTH model are given in

Table 10. Best-fit values for the SLEUTH model.

Parameters	Best Fit
Diffusion	20
Breed	45
Spread	55
Slope Resistance	10
Road Gravity	30

.

3. Result and Discussion

Following the model’s calibration stages, prediction files were obtained for all three alternative growth scenarios (NGS, PGS, SGS) using the grow command. The model has generated future land use maps for each year between 2024, the final year of input data, and 2044, the year of prediction, as well as thematic maps showing potential urban sprawl areas and statistical results on urban growth. Thus, the growth amounts for each scenario and the changes in land use patterns have been simulated in detail.

When examining the statistical tables, it is evident that the model successfully simulated urban sprawl over a 20-year period, with realistic results for urban area growth, sprawl rate, and clustering patterns. The five fundamental growth coefficients of the SLEUTH model (Diffusion, Breed, Spread, Slope Resistance, Road Gravity) were modeled based on the directed values obtained from the coarse and fine calibration stages, assuming that the spread primarily occurs through the road effect. Indeed, in the prediction map for 2025–2044 (land and urban 2025–2044), it was observed that the spread consistently clustered around main roads and the number of growth pixels in the areas between roads increased.

However, it was understood that at the best-fit values of the final calibration, the path effect was somewhat lagging behind, and more realistic results were obtained by increasing the spread coefficient. This situation demonstrates that the urban sprawl of Erzurum city will occur “outwardly” in cells surrounding the previously urbanized cells. As the inputs to the model are examined, one understands that the number of roads increased by 47% over the 20-year period; however, this road influence was relatively limited compared with the 78% increase in urban areas.

As can be seen from the figures, in all three scenarios (NGS, PGS, SGS), although the growth pixels are focused on the main routes, the spread coefficient is salient in the simulation statistics due to the lack of additional external route investment in the last 20 years and the route expansion resulting from the dense route network expansion in marginalized developing areas. In addition, in the simulations for 2044 in all three scenarios, although there is no mention of the rise in a new principal center, small-scale sub-centers are projected to arise in previously unurbanized areas as a result of the breed coefficient.

In addition, it was observed that the diffusion coefficient in the city of Erzurum in 2044 also stood out to some extent, despite being kept low in the parameters. Diffusion, in the SLEUTH model, refers to the probability of random new development points emerging, which can be interpreted as a spatial representation of unplanned or sudden growth. As a result, the diffusion effect has led to urban sprawl occurring in an irregular and scattered manner.

The parameter values determined during the implementation of the model (Diffusion, Breed, Spread, Slope Resistance, Road Gravity) were repeatedly tested within various ranges by examining their relationships with the variables in the input data specific to the study area. By analyzing the resulting prediction tables and prediction maps, increases or decreases were applied to the parameter values to ensure that the model produced the most consistent results. The aim of this study is not to reveal how the city would behave under different parameters in the future, but rather to show how the city would be shaped in the future under different spatial constraints with the best parameter values obtained.

Land use change in the model was predicted separately for three different scenarios under varying urban constraints and thresholds. The land use and urban layers contained in the model’s prediction outputs were classified within the ArcGIS software environment for each scenario according to color categories. For each year, the increase in urban areas and the changes in natural areas (merged into four classes) and protected areas were calculated, and the differences between 2024 and 2044 were presented in tabular form.

3.1. Normal Growth Scenario (NGS)

When the first alternative, the NGS, was taken into account, the urban area was estimated to have a very large spread of 100.84% by 2044, even in the land classes divided into five categories, which conforms to the model setting of SLEUTH, and would result in a negligible decrease even in the reserved areas we classify as “unclass” (

Table 11. Scenario One (Normal Growth Scenario) Land use and urban area changes from 2024–2044.

NGS	2024 Land Use (Ha)	2044 Land Use (Ha)	Rate of Change (%)
Unclass (0)	10,523.70	9743.27	−7.42
Urban Area (1)	9891.85	19,866.80	100.84
Agriculture (2)	16,757.69	11,905.08	−28.96
Range Area, Forest, etc. (3)	13,744.66	10,068.26	−26.75
Wetland (6)	1522.10	856.59	−43.72
Total	52,440.00	52,440.00

). In the model, the areas shown in red represent urban areas, and it is estimated that the area between the Northern Ring Road and the main urban core will be completely urbanized, leading to the complete disappearance of agricultural and pasture areas remaining in this region (Figure 10). It is again observed that the city will develop compactly along the main roads in the western and southern axes, but the remaining areas will be filled in, leading to a reduction in agricultural land.

In the first scenario, even within the unclass areas designated by the SLEUTH model classification as strictly protected zones, some degree of urban expansion is projected, resulting in a 7.42% decrease by the forecast year 2044. Agricultural lands (across all classes) are expected to decline by 28.96%, while pastures, forests, and areas designated in the urban plan for afforestation (notation class 3) are projected to decrease by 26.75%. Despite the spatial restrictions defined in the excluded layer for this scenario, the spontaneous and organic growth metrics of the model indicate that these areas will also experience a reduction, suggesting that urban expansion cannot be entirely prevented. Likewise, encroachment into wetland areas is also anticipated.

3.2. Protective Growth Scenario (PGS)

In the second alternative, the PGS, the SLEUTH Model’s 2044 urbanization simulation predicts that the urban area will increase by 50.83% (

Table 12. Second Scenario (Protective Growth Scenario) Changes in Land Use and Urban Areas from 2024–2044.

PGS	2024	2044	Rate of Change (%)
Unclass (0)	10,523.70	10,152.47	−3.53
Urban Area (1)	9891.85	14,919.51	50.83
Agriculture (2)	16,757.69	15,326.08	−8.54
Range Area, Forest etc. (3)	13,744.66	10,989.31	−20.05
Wetland (6)	1522.10	1052.63	−30.84
Total	52,440.00	52,440.00

). In this alternative, due to the selection of excluded areas using an environmental approach, it is observed that the urban sprawl trend develops along the main road axes and growth pixels cluster along these roads, but the agricultural and natural areas in between are preserved (Figure 11). While in the previous alternative, many scattered urbanized cells are observed in natural areas due to the breed and diffusion effects, in this alternative, the diffusion effect is much less pronounced due to the restrictions. Nevertheless, a significant spread is predicted in the city due to the spread effect on natural areas, resulting in a decrease in these areas.

In the second scenario, despite being the most conservation-oriented, unclassified areas (0), which represent strictly protected zones, are expected to decline slightly by 3.53%, while agricultural lands (2) show a more significant reduction of 8.54%. Similarly, range areas, forests, and afforestation zones (3) are projected to decrease by 20.05%. Wetlands (6) also experience a decline of 30.84%; however, this high rate is largely due to the relatively small spatial extent of these areas, which makes proportional changes appear more dramatic.

3.3. Urban Sprawl Growth Scenario (SGS)

In the third scenario tested in the model, the SGS, the excluded areas were kept very limited, and the simulation showed the outcome of free, unconditional, and natural spread based on the working principle of cellular automata, which underlies the SLEUTH model. In other words, it was estimated how the city, left to its natural flow by removing thresholds and constraints, would achieve a spatial use pattern by 2044. In this scenario, the model predicts that the urbanized area of Erzurum in 2044 will increase by 116.04% compared to 2024, the starting year, resulting in a significant expansion (

Table 13. Third Scenario (Urban Sprawl Growth Scenario) Land Use and Urban Area Changes from 2024–2044.

SGS	2024	2044	Rate of Change (%)
Unclass (0)	10,523.70	8959.08	−14.87
Urban Area (1)	9891.85	21,370.73	116.04
Agriculture (2)	16,757.69	11,028.78	−34.19
Range Area, Forest, etc. (3)	13,744.66	10,265.63	−25.31
Wetland (6)	1522.10	815.78	−46.40
Total	52,440.00	52,440.00

). This situation predicts a 34% decrease in agricultural areas and a 25.31% decrease in other natural areas. Additionally, it is estimated that this alternative will result in the formation of many scattered and irregular built-up areas (Figure 12).

In the third scenario, the free development approach, where spatial constraints are kept to a minimum, comes to the fore. In contrast, losses of 34.19% in agricultural areas (2), 25.31% in pasture, forest, and planned afforestation areas (3), and 46.40% in wetlands (6) are estimated. There is also a 14.87% decrease in the unclass (0) category, which is designated as areas to be absolutely protected. These results show that with the relaxation of spatial constraints, urban sprawl has become rapid and dominant, leading to significant losses in natural and agricultural areas.

In spatial planning studies, the target year is generally considered to be 20–35 years after the plan’s completion date. Depending on the scale of the spatial plan, the target year range increases in large-scale studies, while in studies for small-scale implementation, the target year is considered to be closer to the present. The projected population for the target year is calculated using statistical methods, based on population growth rates from previous years. Since urban development areas need to be determined based on density (people/ha) depending on the future population size, the population projections obtained through statistical and mathematical methods often do not satisfy planners and decision-makers in cities with low population growth rates. In such cases, development scenarios that attract population, such as employment, education, and transportation investments, are designed, and population assumptions far exceeding the projected population are made, artificially determining the city’s future development areas. Thus, a legal framework is being created for urban sprawl.

In this study, population data obtained from the official address-based registration system, which ensures complete accuracy, were analyzed periodically for the neighborhoods located within the study area, and the population increases for each time series were calculated separately in the corresponding tables.

As a result, while the urban area expanded by 78% over a 20-year period (2004–2024), the population increased by only 6% during the same timeframe (

Table 14. Population and Density for each input data year.

Years	Population	Rate of Change (%)	Density (Person/Ha)
2004	389,619	0.00	70.08
2012	384,399	−1.36	49.92
2018	390,879	1.69	41.83
2022	405,645	3.64	42.28
2024	413,165	1.82	41.77

).

The population projection calculation for the city of Erzurum, the study area, using data on population changes between 2004 and 2024 (the base year for the model input data) and based on the main urban area, predicts a population of 456,108 people in 2045, (based on years ending in 0 and 5) an increase of 10.06% (

Table 15. 2045 Population Projection Table.

Years	Exp. Method	Least Squares Method	Compound Interest	Arithmetic Method	Average
2025	414,383	417,243	448,621	443,133	430,845
2030	420,530	424,077	455,266	448,484	437,089
2035	426,768	430,910	462,009	453,835	443,381
2040	433,098	437,744	468,852	459,187	449,720
2045	439,522	444,578	475,796	464,538	456,108

). This calculation was performed using four different valid population estimation statistical formulas. Therefore, the 78% increase observed in urban areas over the past 20 years and the increase of up to 116.04% predicted for 2044 in SLEUTH Model scenarios indicate that urban sprawl has no connection to population pressure (Figure 10).

It is well known that the SLEUTH model does not incorporate socio-economic variables, but rather predicts urban growth based on spatial patterns. For this reason, the fact that the model’s outputs (future time series) reproduce urban expansion with a high degree of accuracy relative to the input data (observed time series) demonstrates that the projected spatial magnitude of growth cannot be solely explained by socio-economic and demographic changes-such as population growth, migration, and employment conditions-that occurred in the study area between the initial and final years of the dataset.

Although the basic working principle of the SLEUTH model is based on the road-axis growth theory, the transportation infrastructure and road networks, which are the driving force of urban sprawl and are effective in determining the direction of development areas, have shown a decrease in road impact and an increase in edge-based sprawl, with road impact falling into the second-tier factors and failing to meet expectations. Because the existing road infrastructure in the city of Erzurum has not developed in parallel with the direction and speed of urban growth. If the road network is not sufficiently integrated with urban development areas or does not serve new urban sprawl areas, it leads to a weakening of the road effect in the model. In the urbanization process, social, economic, or legal factors that develop independently of road infrastructure, such as local planning policies, can overshadow the road effect (Figure 13).

The values shown in the graph represent the urbanized area and the size of the remaining natural environment (in hectares) across the entire study area for the year 2044, which is the end of the model’s projection period under three different scenarios. The red bars indicate the extent of urban areas for each scenario, while the green bars represent non-urban areas.

As seen in the model, the policies developed by the authorities tend to spread, but cities with proper decision-making mechanisms and legal constraints can limit the spread effect. As seen in the three alternative scenarios, increasing the protection status of natural areas on the urban periphery can help suppress sprawl. Although the same parameters were entered into the SLEUTH model, the prediction data generated from different excluded areas yielded different results. This indicates that while urbanization forces neighboring units to urbanize, environmental approaches and regulations can influence the direction and magnitude of the spread.

4. Conclusions

The rapid and often unplanned growth of urban areas is a critical planning issue for the conservation of natural resources, the sustainable management of land use, and the continuity of quality of life. In this context, spatial modeling and scenario-based forecasting studies provide important tools for understanding the future development directions of cities and making correct planning decisions. The city of Erzurum, chosen as the study area, has experienced significant urban sprawl in the last 20 years, with this process largely occurring through the marginalization and expansion of existing settlements. Despite low population growth rates, urban sprawl across large areas indicates that this spread is more related to morphological, political, and social and technical infrastructure factors than to demographic pressure. The main objective of this study is to contribute to sustainable planning approaches by analyzing the potential future growth trends and spatial development scenarios of the city of Erzurum. To this end, the SLEUTH urban growth model, which has a high capacity to simulate the temporal and spatial dynamics of urban growth, was used. Within the scope of the method, the model’s calibration stages (coarse, fine, and final calibration) were meticulously applied; land use, road network, slope data, and restricted areas were used as input data. The model has predicted urban sprawl up to the year 2044 based on three different scenarios (Normal Growth Scenario, Protective Growth Scenario, and Spillover Growth Scenario).

The results show that in all three scenarios, urban sprawl develops from the edges of existing settlements, concentrating especially along main road axes, but the road effect becomes secondary over time. The Normal Growth Scenario showed the highest spread, while the Protective Growth Scenario exhibited the lowest spread due to restrictions on the protection of natural areas. The Spillover Growth Scenario, however, showed that if there are no planning limits, a lot of natural and agricultural land can be lost. The model results show that population growth is not the main cause of urban sprawl. Instead, transportation infrastructure, settlement form, and planning decisions are the most important factors.

In future research, the SLEUTH model can be integrated with socio-economic variables, climatic effects, and transportation forecasts to generate a more comprehensive prediction. Additionally, by comparing different growth models, we can see how well sustainable development policies work in cities with slow growth but long distances, like Erzurum.