Analysis of land use/land cover changes and prediction of future changes with land change modeler: Case of Belek, Turkey

In the areas declared to be a tourism center by state planning, a rapid tourism-related development occurs depending on the investments in tourism, which causes a dramatic land use/land cover (LULC) change. Determining, monitoring, and modeling of LULC changes are required in order to ensure the conservation-use balance and sustainability within such vulnerable areas that are under development pressure. This study consists of four steps. In the first step, the Landsat images dated 1985, 2000, 2010, and 2021 were classified using the maximum likelihood method and the LULC of Belek Tourism Center located in Turkey were determined. The second step included the identification of areal and spatial changes between the LULC classes for the four periods. In the third step, the LULC changes in Belek Tourism Center for 2040 were modeled using the land change modeler. Last step evaluated the relationship between the modeled spatial development pattern and the current planning decisions. According to the results obtained during 36 years, the rates of built-up, forest, and water body areas have increased by 11.91%, 13.67%, and 0.82%, respectively, whereas the rates of barren land and agricultural areas have reduced by 22.25% and 4.15%, respectively. The LULC map modeled for 2040 predicts the built-up areas to expand by 8.25% and the agricultural areas to shrink by 5.42% by comparison with 2021. This study will contribute as a key measure for planners, policy-, and decision-makers to make decisions related to sustainable land use in the areas declared to be a tourism center.


Introduction
Land use/land cover changes consist of a complex and dynamic process that occurs under the influence of natural or anthropogenic activities, affecting the entire ecosystem. Coastal areas, which are a scarce resource, constitute an important part of the ecosystem. These areas need to be developed in a planned way in order to ensure their sustainable use and protection-usage balance. Therefore, ecologically vulnerable areas should be monitored for their change process over time and their land use models for future should be generated.
A sustainable development requires settlement systems to be comprehensively analyzed and modeled Abstract In the areas declared to be a tourism center by state planning, a rapid tourism-related development occurs depending on the investments in tourism, which causes a dramatic land use/land cover (LULC) change. Determining, monitoring, and modeling of LULC changes are required in order to ensure the conservation-use balance and sustainability within such vulnerable areas that are under development pressure. This study consists of four steps. In the first step, the Landsat images dated 1985, 2000, 2010, and 2021 were classified using the maximum likelihood method and the LULC of Belek Tourism Center located in Turkey were determined. The second step included the identification of areal and spatial changes between the LULC classes for the four periods. In the third step, the LULC changes in Belek Tourism Center for 2040 were modeled using the land change modeler. Last step evaluated the relationship between the modeled spatial (Sakieh et al., 2015). Urban development leads to intense changes in the spatial distributions, amounts, and types of land uses, and has an impact on the shrinkage and distortion of natural areas (Oliveira et al., 2018;Xu et al., 2016). As indicated in the final declaration of Rio Conference held in 1992 (Agenda 21), a sustainable spatial planning is one of the most important practical methods to support the planned development of settlements and to ensure environmental, social, and economic sustainability. Spatial planning processes are realized through collection and analysis of basic environmental, geological, socio-economic, etc., data regarding the areas to be planned, synthesis of these data in line with the aims and visions of the plans, and finally identification of the types of spatial use. In these processes, natural areas such as forests and coasts are considered to be a land cover which are generally ignored and the resources to be overused or polluted that affects sustainability. One of the primary topics of the planning process is the effects of the plans, key components of spatial development, on the area in which they are implemented. Changes in the natural areas through spatial development provide significant inputs in analyzing sustainability and identifying restrictive factors towards potential developments. Therefore, directing the spatial planning process accordingly would offer more rational results for protection-usage balance and sustainable development.
In this regard, management and land use policies should be developed for the sustainability of the areas, depending on the urbanization-oriented land use changes (Perminova et al., 2016). Considering the abovementioned point, natural areas should be managed by focusing on sustainability and the management process should be combined with spatial planning process in order to keep potential ecological distortion under control through spatial planning and to efficiently use the space (Perminova et al., 2016). From the sustainability-related perspective of urban development in an environmental context, predicting how spatial plans within the implementation area affect natural areas gains importance to make planning decisions (Sumarga & Hein, 2014). In the decision-making process for the type of use to be introduced through these plans, analyses of the results brought by development and determination of the growth model are among critical steps. The number of innovative and future simulation-focused studies on sustainable development and land use models (Batty et al., 1989) is gradually increasing (Onsted & Chowdhury, 2014;Zhou et al., 2020). Diversification of spatial data sets and the increase in their usability allows for computer-based software programs to be developed, various alternatives regarding land use planning to be compared, and simulations for determination of the best option to be created (Sakieh et al., 2015). Through simulation models, natural and anthropogenic factors affecting land use are included into system analyses in the form of inputs that help in prediction of outputs and explain the space (Çağlıyan & Dağlı, 2015). In order to maintain sustainable development, simulation results provide critical information to review the plans, to direct urbanization, to identify the areas open to urbanization in a spatial manner, and to support decision-makers and planners regarding the designing of new plans (Karakus et al., 2015).
Numerous studies addressing the inclusion of spatial planning policies into land use change models exist in the literature (Zhou et al., 2020). Simulations of changes in urban land use provide useful information for decision-makers, but on the other hand, the integration of the planning into land use change modeling is weak (Domingo et al., 2021). Within this framework, all planning partners, especially practitioners and decision-makers, should focus on studies towards learning from previous practices and providing sustainable development decisions. These changes should be investigated and monitored to ensure a land management in which sustainable urban growth and development exist. In order to integrate planning into simulations on changes of land use, the starting point can be the formation of a spatial modeling frame (Domingo et al., 2021).
Recently, understanding the concept of the land use system and modeling the changes in the LULC to take necessary measures against potential future problems have been among the most commonly discussed topics. Using the previous classification maps, it is possible to form a model to predict the trends regarding the changes in the LULC for a certain period of time. Formation of these change models is a necessity to predict the classes that have a dominant role in the future changes in land use/land cover and in sustainable planning studies that create visions for the urbanization process. These models simplify the Vol.: (0123456789) group of complex socio-economic and biophysical factors affecting the rate of changes in the LULC and provide the prediction of the effects of these changes (Hasan et al., 2020).
A number of studies used the geographic information systems (GIS) and remote sensing (RS) techniques to detect the changes in land use/land cover in a rapid and reliable manner (Adhikari & Hansen, 2018;Al Rıfat & Liu, 2022;Al-sharif & Pradhan, 2014;Butt et al., 2015;Mansour et al., 2020). The advanced techniques are helpful in analyzing the temporal and spatial changes in land use and the direction of these changes. Today, numerous simulation models including artificial neural networks (ANN) (Almeida et al., 2008;Martínez-Vega et al., 2017), cellular automation (CA) (Brown et al., 2012;Farjad et al., 2017), support vector machine (SVM) (Were et al., 2015), Markov chain (MC) (Pocewicz et al., 2008;Yang et al., 2012), SLEUTH model (Mahiny & Clarke, 2012), hybrid models (Guan et al., 2011;Subedi et al., 2013;Xu et al., 2022), and machine learning (ML) (Aburas et al., 2019) have been developed regarding the modeling of the changes in the LULC and urban growth. Several studies have noted that the land change modeler which combines the Markov chain and the cellular automata is a strong model to predict the changes in the LULC (Öztürk, 2015;Singh et al., 2022;Wang & Maduako, 2018). This is because outputs of neural networks, acquired through the weights of evidence technique (where a user can select and modify the weights) (Perez-Vega et al., 2012), more effectively show the transition of different types of land cover than do individual probabilities.
In Turkey, a "Tourism Center" and the "Culture and Tourism Conservation and Development Regions (CTCDR)" are the regions whose borders are identified to conserve and use the areas where historical, natural, and cultural values exist intensively or that have a high potential of tourism potential, and to ensure sectoral and planned development in these areas. In the determination process of these regions, the country's natural, historical, archeological, and sociocultural values, winter and water sports, health tourism, and other tourism-related potentials are considered. Investors are provided with support and promotions to ensure the development of tourism in these regions. In the areas declared to be a tourism center, a rapid tourism-related development occurs depending on the investments in tourism, which causes a dramatic land use/land cover (LULC) change. Being a barren area at first, the coast has been opened to settlement through investments in tourism. It has also triggered urban development in the coast and the areas behind the coast to meet the service needs of tourism facilities. The development of the second housing concept that is increased due to the need for shelter of the employees working at the tourism facilities and the attractiveness of the area depending on and the investments for infrastructure has also escalated the changes in the LULC. It is critical for decision-makers, planners, practitioners, and researchers to monitor the changes in the LULC in order to ensure the conservation-use balance in such vulnerable areas that are under the pressure of development.
This study aims to discuss the policies for spatial planning by examining the relationship between current planning decisions and the spatial development pattern created by the future simulation that is run based on the monitoring of the changes in land use and the current trends. It also aims to determine the potential efficiency of the land use modeling in terms of sustainable urban development and, through contributing to the planning process, to underline the effects of the plans on the space's environmental sustainability potential. Therefore, this study differs from other similar studies as it combines different perspectives from different disciplines.

Study area
The study area is Belek, a tourism center located in the district of Serik in Antalya, Turkey. It is 40 km away from the center of Antalya and is one of the tourism centers with the oldest date of declaration in the region (Declaration Date: 21.11.1984).
Belek Tourism Center is located in the south of Turkey between the north latitude of 36°54′3″-36°50′28″ and the east longitude of 30°55′34″-31°07′08″. Tourism center has Aksu River on the west, Acısu River on the east, Mediterranean on the south, and the Antalya-Alanya highway on the north (Fig. 1).
Belek Tourism Center is 8390.56 ha and its length of coastline is approximately 20 km. The built-up areas of Kadriye and Belek are located in this tourism center. The total population of these built-up areas is 16,191 for 2020. Belek Tourism Center has fortyseven hotels, fifteen golf courses, and various social and cultural facilities. It hosts approximately 1.5 million visitors from different regions of the world every year and is the leading tourism center that attracts the highest number of tourists in Turkey.
It is one of the most important seventeen nesting sites in Turkey for loggerheads turtles (caretta caretta) which are classified as an endangered species by the International Union for Conservation of Nature (IUCN). One study performed by the Belek Tourism Investors Union (BETUYAB) with several universities has indicated that Belek Tourism Center has 109 bird species, one endemic fish species (Aphanius anatoliae), and a total of 574 plant species belonging to 104 families including twenty-nine endemic families and one relict endemic family (Serik Pear) (Camgöz, 2008).
The study area has low values of altitude and slope, ranging from 0 to 37 m and from 0 to 5%. It has a Mediterranean climate, with hot and dry in summers and warm and rainy in winters. The mean yearly temperature of the study area is 18.8 °C. The dominant vegetation in the study region is dune vegetation and stone pine forests along the coasts and maquis and shrubs in the inland areas. Forests, especially stone pine trees (Pinus pinea) and Turkish pine trees (Pinus brutia), are created to prevent sand dunes in the eastern and western parts of the study region and cover a large area (Çakıcı, 2002).

Method process
This study consisted of four steps and aimed to determine LULC and the changes in the LULC of Belek Tourism Center, to simulate the land use/land cover model for 2040, and to assess the relationship between the model and the planning decisions (Fig. 2).
In the first step, land use/land cover of Belek Tourism Center was determined for 1985, 2000, 2010, and 2021. The second step included the identification of areal and spatial changes between the land use/land cover classes for the periods between 1985-2000, 2000-2010, 2010-2021, and 1985-2021. In the third step, a model was run regarding how the land use/land cover would change in 2040 in the event that the previous trends  would be sustained. The fourth step included the comparison between the planning decisions for Belek Tourism Center, the changes in the LULC from past to present, and the model for 2040, and the assessment of the relationship between the spatial development pattern which was formed through this comparison and the current planning decisions.

Data acquisition
Data used in this study were acquired through various resources. In order to determine the land use/ land cover of the study area, satellite images from Landsat 5 TM, the Landsat 7 ETM, and the Landsat 8 OLI were used (Table 1). These satellite images were obtained from the United States Geological Survey (USGS) data portal. All satellite images have a 30 m spatial resolution. With the aim of mitigating the effects of the seasonal changes on the land cover and obtaining less cloudy images, the satellite images were selected from the same season (summer) and within close dates. Digital elevation model (DEM), slope, distance to road networks, distance to river, and distance to built-up area were used as independent variables for the prediction of the future LULC of the study area. The DEM was created through the Shuttle Radar Topography Mission (SRTM) data. The slope data of the study area were generated using the DEM. The data of the distance to road network and the distance to river were obtained from the Open Street Map (OSM) and topographic maps. The data of the built-up area were digitized using the 1:25,000 topographic maps. All data were transformed into the Universal Transverse Mercator (UTM) Zone 36 N projection system.

Image classification and accuracy assessment
The satellite images were classified using supervised classification method. The method refers to the statistical categorization process of the reflection values of the pixels. The supervised classification method classifies an image considering the statistical characteristics of the reference pixels/points collected for each class determined by the user. Maximum likelihood classification (MLC) approach is being widely used for land use change assessment (Islam et al., 2018;Rawat & Kumar, 2015;Tadesse et al., 2017). This method hypothesizes that the bands of the image to be classified are normally distributed, and all classes have the same probability of classification. The MLC approach is based on the principle of determining the equal probability curves for the determined classes and assigning the pixels to be classified to the class with the highest probability. In this method, a candidate pixel's probability of being assigned to each class is calculated separately, and each pixel is assigned to the class that has the highest probability (Mather & Koch, 2011;Örmeci & Ekercin, 2007).
The LULC classes in this study were classified based on the land cover classification system identified by the Europe Environment Agency (EEA) and the characteristics of the region. The LULC was classified into five classes as agriculture, barren, built-up area, forest, and water body. Table 2 shows the description of the LULC classes. The control areas were determined for All cultivated, uncultivated and greenhouse agricultural areas such as farmlands, crop fields including fallow lands/ plots and horticultural land Barren Lands with exposed soil, sand or rocks, and never has more than 10% vegetated cover during any time of the year such as bare ground, bare exposed rocks, beaches, sandy areas other than beaches, strip mines, quarries, gravel pits, and transitional areas Built-up Area Residential, commercial services, industrial area, socio-economic infrastructure, transportation, roads, mixed urban, or built-up lands Forest Mixed forest lands and forest on customary land, protected forest, plantations, deciduous forest Water Body Seas, lakes, rivers, streams, permanent open waters, ponds and reservoirs, marshy land, and swamps the supervised classification. In determining the control areas, the training data were gathered by drawing polygons around the areas that represent each LULC class. The researcher, with their personal knowledge and experiences regarding the study area, collected the training data using high-resolution Google Earth data, and topographic maps. The images were classified using MLC supervised classification method. LULC maps belonging to four different dates were created (Fig. 4). The classification analyses were made using the ArcMap 10.5.1 software. The raster data were transformed into vector data and the area and change percentages of the land use classes belonging to each period were calculated ( Table 3). The accuracy analysis is a control method based on the statistical comparison of the pixel values determined through classification with the points considered to be reference (Gülersoy, 2013). The classification results and the reference data are statistically compared using confusion matrices. The accuracy analysis is performed by utilizing producer's accuracy, user's accuracy, overall accuracy, and the Kappa coefficient, which are generated from the confusion matrix. The producer accuracy shows how well each class can be classified with sampled points, while the user accuracy shows the degree to which the actual location of the user is correctly determined on the map . The overall accuracy expresses the compatibility of the classification result with the reality on the ground. Finally, the Kappa (κ) coefficient shows the true agreement between the reference data and the classified map, and again between the reference data and the randomly classified data (Congalton & Green, 2019). The Kappa coefficient was developed by Cohen (1960) and ranges from 0 to + 1. Landis and Koch (1977) characterized the Kappa values as follows: < 0 as indicating no agreement, 0-0.20 as slight, 0.21-0.40 as fair, 0.41-0.60 as moderate, 0.61-0.80 as substantial, and 0.81-1 as almost perfect agreement.
This study performed the accuracy analyses of the LULC maps from 1985, 2000, 2010, and 2021 and calculated their Kappa coefficients. A total of sixty reference data was generated for each year that was analyzed using the satellite images from Google Earth and topographic maps. The accuracy analysis of this study was performed by comparing each of the sixty reference points on the images that were classified separately. Table 4 shows the classification accuracy of the LULC maps.

Change detection analysis
The change analysis detects the areal change of gains and losses between the land use types, considering the types of land use at the same region in different periods of time. This analysis is frequently used in the identification of various changes regarding different classes of land use, such as an increase in the urban built-up areas and a decrease in the agricultural areas (Hassan et al., 2016;Sundarakumar et al., 2012). This study utilized land use maps from 1985, 2000, 2010, and 2021, and detected the changes between the types of land use by means of a "from-to" analysis that allows comparison on a pixel level. The changes in land use between 1985 and 2000, 2000 and 2010, 2010 and 2021, and 2021 and 1985 were calculated, respectively. Then, these changes were classified as spatial and areal (Table 3, Fig. 5).
LULC modeling and prediction of the future land use/land cover of Belek Tourism Center The Land Change Modeler (LCM) module integrated into the TerrSet software is designed to analyze the changes in LULC and to predict potential future changes. This model is based on the artificial neural network (ANN), Markov Chain matrices, and transition suitability maps, generated by training multilayer perceptron (MLP) or logistic regression (Ansari & Golabi, 2019;Megahed et al., 2015). The LCM utilizes historical maps of LULC to empirically model the association between LULC transitions and driver variables to map future LULC modeling (Eastman & Toledano, 2018). The LCM predicts the future changes in LULC in four steps: (1) analysis of the past changes in the LULC, (2) identification of the transition potential and driver variables, (3) modeling of the future LULC, and (4) validation of the model. This study predicted the LULC changes in Belek Tourism Center for 2040 using the LCM.
Transition potential and driver variables Transition potential maps determine the future change probability of a specific type of land use. These maps are generated by analyzing the artificial neural networks of a single transition sub-model or a group of transition sub-models. A transition sub-model can consist of a single land use transition or a transition group that is assumed to have identical driver variables (Addae & Oppelt, 2019). LCM allows choosing the optimal threshold for selecting the optimal number of sub-models, where the users can overlook small changes (Eastman & Toledano, 2018). This study generated the transition sub-models by ignoring the LULC changes below 110 ha (less than 1.5% of the study area). Three sub-models were used to predict the LULC for 2021 (agriculture to built-up, barren to built-up area, and barren to forest), whereas three sub-models were used to predict the LULC for 2040 (agriculture to built-up, forest to built-up area, and barren to forest).
For each sub-model, driver variables were identified as inputs to predict LULC maps (Motlagh et al., 2021). The driver variables are expected to have a significant impact on the future LULC changes. Cramer's V is used to calculate the importance of each variable. The value of Cramer's V ranges from 0 to 1. Jin et al. (2013) state that Cramer's values close to 0.4 and above of it are considered as the appropriate value for driver variables and values less than 0.15 are considered its weak ability to predict for a driver variable. However, Cramer's V does not guarantee a strong model performance because it cannot account for the mathematical specification of the modeling approach used as well as the intricacy of the relationship (Eastman, 2015). Cramer's V helps to determine whether a specific driver variable should be used for the prediction of the LULC Fig. 3 Maps of driver variables used for land change modeling changes. The driver variables in this study were identified considering the studies in the literature (Anand et al., 2018), characteristics of the region, and the experts' opinions. The driver variables used in this study were the DEM, slope, distance to road networks, distance to built-up area, and distance to river. Figure 3 shows the maps of the driver variables. Cramer's V values of the driver variables are introduced in Table 6. The DEM and slope are among the key topographic factors that influence the urban development. The urban sprawl is accepted to occur in regions where the values of altitude and slope are lower. The "distance to built-up area" variable was included in this study because regions that are surrounded by built-up areas but have not been opened for settlement yet have higher probability to turn into a built-up area within years. The distance to the river limits the urban sprawl. The distance to road networks has a key role in urban sprawl, as it facilitates the access to daily needs. Distance to built-up area and distance to road networks are modeled as a dynamic variable because they change and develop in the course of time.

Change prediction
Change prediction is analyzed using the Markov Chain model and the Cellular Automata (CA) algorithm. The Markov Chain model is a stochastic model used to predict LULC (Al-sharif & Pradhan, 2014;Singh et al., 2022). This model predicts the LULC starting from a t = 1 time to another t + 1 time, depending on the transition area matrix and the transition probability matrix between the LULC classes (Shawul & Chakma, 2019). The Markov Chain does not solely suffice by itself to predict the LULC changes because it does not take the spatial distribution of the LULC classes and the spatial aspect of growth into consideration (Ghosh et al., 2017). Therefore, the CA-Markov method that combines the Markov chain model, CA, and the multicriteria analysis (MCA) is adopted to predict future land use/land cover. The CA includes a regular grid of cells that manage how each cell's neighbors affect the future class of each cell (Aviv & Sipper, 1994). These models typically simulate changes in cells that are near the borders among classes (Benenson & Torrens, 2004). The CA-Markov method adds spatial This study used the CA-Markov method to predict future changes in the LULC. This method uses "the LULC data belonging to two different periods of time, the transition area matrix generated using the Markov Chain model, the transition potential map created using the MCA method, a 5 × 5 neighborhood filter" these data for the prediction of the LULC changes.

Validation
Model validation is required to test the reliability regarding the prediction of the land use/land cover changes. There are two different methods (visual and statistical) in the literature used for model validation (Pontius & Malanson, 2005). This study utilized the Kappa statistical method to assess accuracy. The first step of the model validation included the prediction of the land use for 2021 using land use maps from 2000 and 2010, transition potential maps, and transition probability matrix. Then, the model validation was performed through the analysis of the predicted land use map for 2021 and the actual land use map for 2021 using the Kappa statistical method. Following the validation of the model's predictive power for the land use/land cover for 2021, the LULC of Belek Tourism Center for 2040 was predicted using the land use maps from 2010 and 2021, transition potential maps generated for this period, and transition probability matrix.
Relationship between the LULC modeling for 2040 and planning decisions Planned developments within Tourism Centers may be possible if long-term spatial planning and capacity use decisions of these areas are considered. In this regard, planning process of the study area with lower and upper scale plans environmental plan, land use plan) were analyzed. These plans constituted the material of this step. Materials were obtained from the Municipality of Belek, Antalya Metropolitan Municipality, Ministry of Culture and Tourism, Ministry of Environment, Urbanization and Climate Change, individual interviews, and online scanning. Assessing the situation with a planning perspective, land use decisions of the plans regarding Belek tourism region were briefly identified.  Figure 4 shows the spatial distribution of the LULC of Belek Tourism Center for 1985, 2000, 2010, and 2021. Areal statistics of the LULC classes are indicated in Table 3. The total surface area of the region analyzed is 8,390.56 ha. According to Table 3, in 1985, 38.5% of the tourism center was agricultural area, 25.5% barren area, 24.4% water body area, 9.8% forest, and 1.8% built-up area. In 2000, 40.4% of the tourism center was agricultural area, 24.9% water body area, 15.8% forest, 11.3% barren area, and 7.6% built-up area. In 2010, 38.7% of the tourism center was agricultural area, 24.9% water body area, 23.3% forest, 9.1% built-up area, and 4.0% barren area. In 2021, 34.3% of the tourism center was agricultural area, 25.2% water body area, 23.5% forest, 13.7% built-up area, and 3.3% barren area. Table 4 shows the accuracy results. Accordingly, the overall classification accuracy values were 0.88, 0.86, 0.90, and 0.90 for 1985, 2000, 2010, and 2021, respectively. The Kappa coefficient values were 0.85, 0.82, 0.88, and 0.88 for these four periods, respectively. The producer's and user's accuracy values regarding the LULC classes ranged from 40 to 100% in 1985, from 60 to 100% in 2000, from 75 to 100% in 2010, and from 71 to 100% in 2021. These results indicate that the supervised classification is accurate and reliable for all four periods.

Change analysis
The LULC changes of Belek Tourism Center were grouped into four periods as 1985-2000, 2000-2010, 2010-2021, and 1985-2021. Table 5 shows the direction and rate of the changes between the LULC classes in detail. The LULC change maps for each period is demonstrated in Fig. 5.
When the change values are examined; -a shrinkage was detected in the barren area by 14.2%, whereas an expansion was observed in forests by 6.0%, in built-up areas by 5.8%, in agricultural areas by 1.9%, and in water body areas by 0.5% for the period of 1985-2000. The change analysis of the LULC classes showed that of the barren area, 732.50 and 396.96 ha were transformed into forests and built-up areas, so the expansion in the built-up and forest areas was arisen from the barren area. No changes were observed in the area with 6,381.17 ha within this period. -In the period of 2000-2010, the barren areas continued to shrink with a rate of 7.3%. On the contrary to the previous period, a decrease was observed in the agricultural area with a rate of 1.7% in this period (Table 3). The forest and built-up areas continued to expand with the rates of 7.5% and 1.5%, respectively. Within the same period, no changes were detected in the water body areas. It was found that the expansion in Transition analysis and validation of LULC simulation model Table 6 shows the potential explanatory power of each driver variable that is represented by Cramer's V coefficient regarding the LULC changes.
Results indicated that each variable had a sufficient level of explanatory power (Cramer's V > 0.15). The variable with the highest level of explanatory power was distance to road networks (39.35%). Other Table 5 Change of area between LULC classes for the years 1985-2000, 2000-2010, 2010-2021, and 1985-2021  variables were also determined to be a significant driving force behind urban growth. Following the selection of the driver variables, the determined land class transitions were modeled in a single transition sub-model and transition potential maps were generated. The accuracy of the transition potential maps ranged from 45 to 99%. This study created transition probability matrices for 2021 (using the LULC maps from 2000 and 2010) and 2040 years (using the LULC maps from 2010 and 2021). Table 8 shows the transition probability matrices for these periods. In Table 8, the rows represent the values of the land cover classes respectively for the previous period (2000 and 2010), and the columns represent the values respectively for the following period (2010 and 2021). Diagonal values indicate the probability of each class to remain unchanged, i.e., the resistance of one class of land cover to another class, while values outside the diagonal indicate the possibility of transition from one class to another Fig. 5 Changes between the LULC classes for the periods of 1985-2000, 2000-2010, 2010-2021, and 1985-2021   . The transition probability matrix for 2021 indicated that the most resistant to change class was water body areas (95.16%), whereas the most dynamic classes were barren (21.24%) and built-up (61.51%) areas. The LULC changes were towards forest, followed by the transition of barren areas into built-up areas. The transition probability matrix for 2021 showed consistency with the change analysis results indicated in the "Change analysis" section.
During the accuracy analysis of the model, the actual LULC map for 2021 and the simulated LULC map for the same year were compared. The accuracy analysis was conducted using the Kappa statistics and overall accuracy. This analysis calculated the Kappa value as 0.75 and the overall accuracy of classification as 80.0%. These results indicate that this model may be utilized for the LULC projection of Belek Tourism Center for 2040 (Kappa coefficient > 0.60). Figure 6 shows the actual and simulated LULC maps for 2021 and Table 7 indicates the area and change statistics of the LULC classes. Table 7 explains that in the simulated map, forest, and agricultural areas were overestimated by 15.76% and 0.24%, whereas built-up areas, barren areas, and water body areas were underestimated by 14.77%, 36.04%, and 2.34%, respectively.
The transition probability matrix for 2040 predicts 93.79% of the water body areas, 80.11% of the forests to be more resistant by 2040. Moreover, the barren areas (26.77%) are expected to be the most dynamic class (with the highest probability of change). A certain part of the agricultural areas (15.78%) is predicted to be under the pressure of built-up areas (Table 8). Table 9 indicates the results regarding the LULC changes of Belek Tourism Center for 2040. Figure 7 shows the simulated map of the spatial distribution of the region's LULC for 2040. This simulation map for 2040 predicts that Belek, Kadriye, and Kumköy built-up areas will expand towards agricultural areas and that the entire coastline except for the conserved areas will be transformed into built-up areas (housing, second housing, tourism facility, etc.).

Simulated LULC changes
The statistics for 2040 year predict that of the total area, 28.90% will be agricultural area, 25.08% water body area, 22.25% forest, 21.98% built-up area, and 1.79% barren area. Within the period of 2021 and 2040, it is expected that built-up areas will expand by 8.25%, whereas agricultural areas, barren areas, and forests will shrink by 5.42%, 1.45%, and 1.25%, respectively. The future changes in the water body areas (0.13%) are predicted to be insignificant. The results of the transition probability matrix show that the trend of change is expected to be mostly from barren areas to forests and from agricultural areas to built-up areas (Table 8). To summarize, this study indicated that the built-up areas in the inland and coastal zones would continue to expand within the next 20 years in the event that the LULC pattern and trend of Belek Tourism Center from 1985 to 2021 would be sustained.

Relationship between the LULC modeling for 2040 and planning decisions
The first planning in the region was conducted by the Scandinavian Planning and Development Associates on behalf of the State Planning Organization within the scope of the West Mediterranean (an international project) in 1967 (State Planning Organization, 1969). Being known as the Ole Hellweg Plan or the Western Antalya Project, this planning is considered the first tourism master plan in Turkey and covers a 1000-km part of the coastal area in Muğla and Antalya (Acar İnam & Ersin Ünal, 2010). This planning assessed the analysis of the tourism potential of the coastal areas in Antalya and Muğla, the infrastructure and superstructure studies, determination of the development areas, and preparation of the master plans. The Belek region was chosen as one of the primary development areas within the plan. Its total number of beds was determined to be 5000 (Almaç, 2005). This plan was followed by the Serik-Antalya-Alanya Environmental Plan approved by the Ministry of Public Works and Settlement in 1981.
The Law for the Encouragement of Tourism (law no.: 2634) that was put into force in 1982 in Turkey allowed the ongoing developments in the tourism industry to be supported with the legislation, the terms "tourism area, tourism region, and tourism center" to be added in the tourism planning terminology, and the mass tourism to rapidly increase. In accordance with this law, Belek was declared to be a tourism center in 1984. The border of the tourism center was determined by excluding the built-up areas (Kumköy, Kadriye, Belekköy) in 1984. After being declared to be a tourism center, the region raised its bed capacity to 13,000 (Acar İnam, 2009). Belek Tourism Center was announced to be a tourism investment area in 1986 (Acar İnam & Ersin Ünal, 2010). Within the same year, the Belek Tourism Center Land Use Plan (1/25,000) was approved. This plan separated Belek Tourism Center into two regions as east and west (Tezcan, 2008). In accordance with the planning decisions, the first tourism investments were made in 1987. The land use plan dated 1986 showed that land uses were arranged for their intended purpose, the zoning method was used to determine the borders of tourism-related development, natural values were conserved, and the needs of the local community were taken into consideration (regional park).
The Serik-Manavgat-Antalya Environmental Plan (1/25,000), the first environmental plan for this region, was approved in 1981. In 1990, the Serik-Manavgat-Antalya Environmental Plan was revised, and the Eastern Antalya Environmental Plan (1/25,000) was approved. In 2002, a planning revision was made for Belek Tourism Center and its surroundings through the Eastern Antalya Environmental Plan Belek Revision (1/25,000). A number of revisions were performed on the same plan in 2004, 2005, 2006(Acar İnam, 2009). Through the revisions on the plan and the changes in land use decisions, camping areas were transformed into tourism facility areas, forests were transformed into golf courses, and the bed capacity was raised.
The latest arrangement regarding Belek Tourism Center and its surroundings is the Antalya-Burdur-Isparta Planning Region Environmental Plan in 2014 (1/100,000) and the revisions made on this plan in 2019 and 2022 (Fig. 8) (Ministry of Environment, Urbanization and Climate Change, 2022). This plan also includes tourism, optional area of use, and housing uses. Within the scope of the changes made in the plan in 2022, a rearrangement was performed for the transfer of the 1/25,000-scale land use plan decisions approved and revised between 2017 and 2021 into the 1/100,000-scale environmental plan. After the plan in 2014, however, no planning decisions that would change the land use decisions for Belek Tourism Region were found. Development towards agricultural areas was recommended in the Belek Tourism Center region.
This study compared the planning decisions for 2022 with the modeling for 2040. According to the comparison results, the areas that continue to be used as agricultural area in the plan are predicted to be transformed to settlement in the modeling for 2040. Considering the current trends, the model predicts Kadriye and Belek built-up areas to be merged.

Discussion
When the reasons for the changes given in Table 3 are examined, it was determined that built-up areas expanded over time and their need for area to expand was mostly  (Hassan et al., 2016;Tahir et al., 2013). The main reason behind this expansion during the 36-year period is that this region was declared to be a tourism center in 1984 and became an attractive area, leading the tourism facility investments and second housing development to start rapidly. The increasing population and developing service areas (such as logistics) in parallel with tourism-related developments also affected the growth of the built-up area. The development dynamics of built-up areas are directly related to urban population growth (Khalifa, 2015;Mansour et al., 2020;Zhang et al., 2011). The continuous expansion of urban areas is caused by the acceleration of investments in residential, commercial, port, and entertainment facilities and road network (Hassan et al., 2016). According to the population census results, Belek and Kadriye had a population of 2586 and 3400 in 1997 while their current population is 8667 and 9126, respectively. Rapidly developing tourism and increasing population accordingly led to a bidirectional change in land use. On the one hand, structuring of the tourism facilities on the coastline affected barren areas. On the other hand, agricultural areas have been transformed into built-up areas in order to meet the housing demand. While the shrinking of barren and agricultural areas causes negatives for maintaining ecological balance and biodiversity, it increases the urban heat island effect and accelerating climate change (Al Kofahi et al., 2018;Mansour et al., 2020). Therefore, policymakers and planners should make decisions considering the spatial extent and distribution patterns determined for the land use type, especially in terms of issues such as urban sprawl, infrastructure services, and green spaces. Within the aforementioned period, changes in the water body areas (0.8%) were determined to be insignificant. This partial expansion in the water body areas was due to the construction of artificial ponds with landscape in golf courses. Land cover plays an important role in the sustainability of water resources (Alipour et al., 2017;Motlagh et al., 2021). The increase in agriculture, housing, tourism, and industrial areas will increase water consumption. Therefore, it is necessary to monitor and manage LULC changes for sustainable water management (Motlagh et al., 2021).
The forests were the land use class with the highest rate of growth between 1985 and 2021 (13.7%). Similar to the results in Belek, it has been determined that forest areas were increased in some studies in Turkey (Sivrikaya et al., 2007;Kadiogulları et al., 2014;. In addition, the successful forestry and afforestation prevent sand and coastal erosions works carried out throughout the country from the past to the present in line with the mission and objectives of the General Directorate of Forestry have made a great contribution to the expansion of forest areas. A transformation from barren areas to forests occurred through these planting works. The state-owned area in the Kumköy coast was declared to be a "strictly protected sensitive area" and was not opened to be a tourism facility. While this area was bare land in 1985, it has now turned into a forest area with the planting works. Inclusion of the golf courses in the forest class was another reason for the growth. The first golf course planning decision within Belek Tourism Center was introduced through the Eastern Antalya Environmental Plan approved by the Ministry of Public Works and Settlement in 1990. A total of five golf courses was planned for this area at first. Today, there are fifteen golf courses in the region. Agricultural areas constitute a dominating LULC class in the study area, despite showing a shrinkage at a rate of 4.2% during the 36-year period. Similarly, many studies in the literature were determined a decrease in agricultural areas Al-Kofahi et al., 2018;Hasan et al., 2020). According to Kaptan (2021), there was decrease in agricultural areas throughout Turkey, especially after 1980. This decreasing trend in agricultural areas is due to population growth, rapid urbanization, increase level of income, changes in living standards, Fig. 8 The 1/100,000-scale environmental plan of the study area and traditional farming methods (Wang et al., 2018;Hasan et al., 2020). A large part of the agricultural areas is located in the vast plains with arable fields in the north of the tourism center. Green house farming is common in this region. The convenient conditions of the Mediterranean climate increase the agricultural production in the region and directly meet the need for food of the tourism facilities. The agricultural areas expanded by 1.9% between 1985 and 2000, whereas they shrunk by 1. 7% and 4.4% between 2000-2010 and 2010-2021. This decrease within the last 21 years was due to overpopulation, tourism, and the increasing demand for second housing. These demands caused rural areas to turn into built-up areas with master plans.
In this study, LULC changes in the study area were simulated successfully via the integration of RS and GIS techniques with the CA-Markov model. When predicting the future, the model simulates the type of change in the periods specified by the transition probability values it has created for each class, based on the changes in the classes for the determined periods (Aksoy & Kaptan, 2021). The predicted LULC map of the study area for 2040 shows that the entire coastline except for the conserved areas and beach will be transformed to settlement through tourism. The predicted LULC map for 2040 indicates that a large part of the urban growth expected to occur between 2021 and 2040 will serve as an extension of the built-up areas in 2021 and that this growth will be in the core of the built-up areas (Belek, Kadriye). Gidey et al. (2017) predicted that forestry, settlement, grassland, and forest areas will increase, and water, agriculture, and floodplains will decrease. Zadbagher et al. (2018) predicted that built-up areas will increase, and forest areas will decrease in the study area. Aksoy and Kaptan (2022) also predicted that forest, settlement, and water areas will increase, and agricultural areas will decrease in their study. The emergence of diverse results in these studies can be attributed to the economic, ecological, and sociocultural differences of each study area.
The planning decisions and land use changes regarding Belek Tourism Center and the model for 2040 were compared; then, the relationship between trends and planning decisions was assessed. The land use changes predicted in this study will be a spatial guide to monitor future trends, taking into account the threats posed by the expansion of built-up areas. Integrating the land use simulations into the planning process will guide the decision-makers, practitioners, and planners to assess the accuracy of the decisions made for ensuring the sustainable urban development. The comparison between the planning decisions for Belek Tourism Region and the simulation for 2040 introduced the necessity to limit the current trends in terms of the conservation of the agricultural areas. The study includes innovations that can guide similar studies because the plan decisions include the process of preventing future negativities and evaluating potentials, in addition to evaluating the predicted land use changes. This output will help manage the urban growth and ensure the conservation-use balance regarding the spatial planning decisions.

Conclusion
The study examined the LULC changes that were caused by the declaration of the Belek region in Antalya as a tourism center from the year 1985 to 2040. The integrated approach including remote sensing, GIS, and the CA-Markov model was used to understand the spatiotemporal dynamics of LULC and prediction of future LULC change in Belek Tourism Center. During 36 years , the builtup areas that expanded to meet the needs of the population, increasing in parallel with the developments in the tourism areas showed an expansion on agricultural areas and barren areas.
The agricultural, barren, and forest areas are expected to shrink while built-up areas are expected to expand between the years 2021 and 2040. No significant changes are expected in water body areas. The urban development is predicted to be towards agricultural areas. This result indicates that the problem of urban sprawl, which is defined as the rapid expansion of cities towards their surroundings, will be a fundamental problem that needs to be solved in the future in Belek. It is necessary to take measures against negativities such as the decrease in agricultural areas and the increase in foreign dependency with the urban expansion, the deterioration of the ecological balance, the lack of access to healthy food resources, and the acceleration of the effects of climate change. By developing, land management approaches where the natural, economic, and cultural structure that forms the basis of the planning is evaluated holistically, the negativities encountered can be eliminated, and their effects can be reduced.
Integrating the land use simulations into the planning process will guide the decision-makers, practitioners, and planners to assess the accuracy of the decisions made for ensuring the sustainable urban development. The comparison between the planning decisions for Belek Tourism Region and the simulation for 2040 introduced the necessity to limit the current trends in terms of the conservation of the agricultural areas. This output will help manage the urban growth and ensure the conservation-use balance regarding the spatial planning decisions.
Studies on determining and modeling of the LULC changes will give the opportunity to determine the land use-related situations to occur in the event that the current trends will continue and to assess the relevant results in planning decisions. This study determined that the declaration of the tourism center created a pressure on the current LULC. The policymakers and planners should take this pressure into consideration when they make decisions regarding the region. The pressure caused by the human settlements increasing due to tourism-related investments is one of the primary problems to take measures against. In this regard, policymakers, planners, and practitioners should work coordinately for this problem, and the land use models should be generated considering the results of this study. These models to be generated using the RS and GIS techniques will help to reveal the negative activities from past to present, to prevent the repetition of the mistakes, and to eliminate the present problems.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
Code availability Not applicable.

Conflict of interest
The authors declare no competing interests.