Mapping and Predicting Land Cover Changes of Small and Medium Size Cities in Alabama Using Machine Learning Techniques

Abstract

In the southeastern US, Atlanta is always the focus of attention, despite the rapid expansion of small and medium-sized cities (SMSCs) in the region. Clearly, larger cities have more people, resulting in more loss during disasters. However, SMSCs also face natural calamities and must be made robust and sustainable. Keeping this in mind, this study chooses to focus on ten SMSCs in Alabama (Population > 40,000) which have encountered at least a 6% increase in population size between 1990 and 2020, out of which two large cities (Population > 180,000) which experienced loss during the same time. This paper examines the change in urban built-up area between 1990 and 2020 using the random forest algorithm in Google Earth Engine (GEE) and estimates future 2050 urban expansion scenarios using the Cellular Automata (CA) Markov model in TerrSet’s Land Change Modeler (LCM). The results revealed urban built-up areas grew rapidly from 1990 to 2020, with some cities doubling or tripling in size due to population growth. The future growth model predicted growth for most cities and urban expansion along transportation networks. The outcome of this research showcases the importance of proper planning and building sustainably in SMSCs for future natural disaster events.

1. Introduction

It is anticipated that during the forthcoming decades, there will be a consistent rise in the proportion of the world’s population that resides in urban settings, as opposed to rural settings. According to research published by the United Nations, by the year 2050, it is expected that 68.4% of the world’s population would be living in urban areas [1]. The percentage of the world’s population that lives in urban areas has climbed from 30% in the year 1950 to over 50% in the year 2018, with much of this urban population expansion being concentrated in developing nations [1]. It is also worth noting that most of the urban research is carried out on large, global cities such as Atlanta, New York, Tokyo, Kolkata, and Dhaka, amongst others [2,3]. Because of this, there is a scarcity of research on small and medium-sized cities (SMSCs), which frequently see unprecedented population expansion even though they are inadequate to manage it. In the United States alone, between 1982 and 1997, there was an increase of 34% in the total quantity of land that was used for urban and built-up purposes [4]. Following the urbanization trend, there was a significant jump of 13% urban population change between the year 1980 to 2020 as reported by the US Census [5]. There will likely be a continuation of urban growth over the course of the next 25 years, according to statistical forecasts; however, the extent of this development will differ between regions [4]. The United States Census Bureau has assigned a population-based ranking of each city in the United States. According to this ranking, cities with a number lower than 101 is regarded big city, between 101 and 200 are regarded to be medium-sized and higher than 200 as small city [6]. The population range for medium size city ranges from 98,000 to 210,000 residents in 2010. In Alabama, a southern state in the United States, there is only one large city (Birmingham, which ranks 100) and three medium-sized cities (Montgomery: rank 105, Mobile: rank 120, and Huntsville: rank 126), according to these numbers and the ranking methodology. The remaining ones are classified as small-sized cities and are ranked higher than position 200. Research has also shown that medium-sized cities have a faster population growth rate than a select few of the larger cities (such as Detroit, Cleveland, Pittsburg, Saint Louis, and New Orleans), which all lost more than 20 percent of their population during the 1990s [7]. Additionally, in another study, the southern and western states of the United States showed to have medium-sized cities with the highest rates of population growth [8], one of which is Alabama. The findings of above research point to a possible migratory pattern in which residents of numerous large cities move to a greater number of small and medium size cities (SMSCs). The Southeast US, in particular, is also highly vulnerable to the effects of climate change, such as sea-level rise and excessive heat. Temperatures in the Southeast are projected to rise by 2.2 °C during the next century, accompanied by a rise in the frequency and severity of droughts [9,10], providing a need to study these states further. Therefore, this study focuses on SMSCs in Alabama for Land Use Land Cover (LUCC) and impacts of the frequency of hazardous weather events, their surging population, and the presence of large populations of communities of color and those living in poverty [11]. There are few studies conducted in the cities of Alabama related to LUCC which includes integrating land use change with transportation model for Montgomery [12], studying LUCC in Mobile Bay [13], studying urban heat island study for Huntsville [14], improving LUCC classification for Huntsville [15], and exploring effects of LUCC on air quality for central Alabama [16]. However, none of these studies explore in-depth multiple decadal LUCC of the top 10 SMSCs of Alabama, using those to predict their “business-as-usual” future LUCC. This is important to understand as SMSCs in southeast US are particularly vulnerable to extreme events due to their exponentially growing population which includes large vulnerable populations of communities of color and those living in poverty [11].

Studying past and present trend combined with predicting future land use change is an important aspect of urban research. United Nations has forecasted that 68.4% of the population will reside in urban areas by 2050 referred to as city in this context [1]. A rise in the city’s population places a greater strain on the city’s infrastructure and overall quality of life. Because of this, there has been a natural increase in the urban built-up areas [17] leading to land use changes, which is necessary to support the rising population. It has been decided to utilize the amount of urban built-up as a proxy measure to indicate the expanding population. In this context, the term "urban built-up" refers to the geographic region that is bounded within a city by the human built impermeable surfaces. This region can be found described in recent research on remote sensing as impervious surfaces [2]. Using technical tools such as GIS and remote sensing to determine the expansion of urban centers coupled with land use changes in various parts of the world is becoming increasingly common. The advancement of remote sensing and digital image processing offers unparalleled opportunities for a broader range of locations to detect changes in land cover more accurately, with decreasing prices and processing times [18].

This literature review will focus on a few key research studies related to urbanization on a global scale using GIS and remote sensing. The transformation of land cover in Fez, Morocco, one of the most ancient imperial cities, was investigated and studied using satellite images and secondary datasets of thirty years, beginning in 1984 and continuing through 2013 [19]. Another study was conducted in Tamilnadu, a city in Chennai, India. The study was conducted to examine the consequences of increasing population and urban sprawl on productive agricultural areas and pristine forests using images from 1991 to 2016 and were used to project land cover for 2027 [20]. The change and urbanization expansion in Basrah Province, Southern Iraq was also studied using LUCC classification of images of Landsat TM in 1990 and Landsat ETM+ in 2003 [21]. In addition, for the purpose of designating LUCC, a supervised classification was carried out for the Northwestern coast of Egypt [22]. The United States has been the site of a significant amount of study into the use of remote sensing to better understand LUCC. Xiaojun Yang (2002) monitored the urban spatial growth in the Atlanta metropolitan region in 2002 using an unsupervised classification method that was based on Landsat TM data between 1973 and 1999 [2]. Similarly, Yuan et al. (2005) looked at land cover classification and change analysis in the twin cities of Minnesota [23]. The cities of Birmingham and Hoover, both located in Alabama, were analyzed by Trousdale (2010) using supervised classification. Over a span of thirty-four years, he analyzed the expansion of suburban sprawl (1974 to 2008). The findings indicate that there was a gradual loss in forests, agricultural lands, and green space over the course of the research period; in addition, there was an increase in urban and residential LUCC in the form of built-up in the metropolitan area [24]. With the advancement of technology, cloud-based platforms aiding remote sensing research such as Google Earth Engine (GEE) are being introduced. GEE is a cloud based geospatial analysis tool that enables users to visualize and study satellite imageries and other derived datasets of our planet. GEE is utilized by scientists and non-profit organizations for remote sensing research, disease outbreak prediction, natural resource management, and more [17,18,19,20]. Scientists have used GEE in many land use land cover classification studies as well. Phan et al. (2020) used GEE to examine the effect of various composition approaches and different imageries on classification outcomes [25]. Similarly, Tassi and Vizzari (2020) developed and tested an object-based classification approach using three techniques: the Simple Non-Iterative Clustering (SNIC) algorithm to detect spatial clusters, the Gray-Level Co-occurrence Matrix (GLCM) to determine cluster texture indices, and Random Forest (RF) or Support Vector Machine (SVM) for the categorization of the clusters in GEE [26]. Becker et al. (2021) used of Landsat-8 images in the GEE to automatically classify land use and land cover in the So Francisco Verdadeiro River hydrographic basin, western Paraná state [27]. Additionally, Liu et al. (2018) created multi-temporal global urban land maps based on Landsat imagery during the 1990–2010 era with a five-year interval ("urban land" in these maps refers to artificial cover and constructions such as pavement, concrete, brick, and stone) utilizing the power of GEE [28].

Remote sensing also serves as a tool to implement land use change models. Land use change models serve as analytical aids for determining the causes and effects of land use dynamics. Such data could serve as a foundation for scientific and efficient land-cover planning, management, and ecological restoration, as well as a guide for regional socioeconomic growth. Using land use classified maps from the past, it is feasible to construct a model to forecast trends in land cover (LC) changes over a specific time period. There are a variety of land use models available, each originating from a distinct academic discipline [29]. They can be categorized as analytical equation-based models [30], statistical models [31], evolutionary models [32], cellular models [33], Markov models [34], hybrid models [35], expert system models [36] and multi-agent models [37]. Currently, the most prominent models for monitoring and predicting land use change are cellular, agent-based, and mixed models [38]. The CA-Markov model combines the Markov model with the Cellular Automata model. This model combines the long-term predictions of the Markov model with the capability of the Cellular Automata (CA) model to simulate spatial variation in a complex system, and it can simulate LUCC change [38]. Both cellular automata (CA) and the Markov model have significant advantages in the analysis of land use change, as well as limitations. The Markov model for land use changes has been widely applied, although it is difficult to forecast the spatial pattern of land use changes with the classic Markov model. The CA model equipped with powerful spatial computing can be utilized to simulate the spatial variation of the system with precision. A CA–Markov model is a robust method for spatial and temporal dynamic modeling of land use changes because geographic information systems (GIS) and remote sensing (RS) can be incorporated effectively [39]. The CA–Markov model incorporates the advantages of the time series and spatial predictions of the Markov and CA theories, and it can be utilized to stimulate the Spatial–Temporal Pattern. The CA–Markov model also considers the suitability of land use changes and the impact of ecological, social, and economic factors on land use changes. Numerous research [3,33,39,40,41,42,43] have utilized the CA-Markov model to track and predict changes in land use and landscape. Hence, this study utilized CA-Markov model accessed through Terrset Land Change Modeler (LCM) to track and predict changes in urban built-up of Alabama’s cities.

Therefore, this study aims to quantify the changing dynamics of urban built-up expansion in Alabama over the past few decades, centering its attention on the ten cities in Alabama with the highest populations (Table 1) using GEE and Terrset LCM. These cities make up the state’s top ten population centers. Although Alabama as a whole is experiencing relatively slow population growth, specific cities within the state are experiencing radically contrasting patterns of urban development. With the exception of Birmingham and Mobile, the population of most of Alabama’s largest cities has increased significantly during the past few decades. On the other hand, the population of Birmingham and Mobile has decreased over the years. The numbers of people living in Madison, Hoover, and Auburn have all grown considerably over the course of the past three decades. Table 1 shows that major cities like Birmingham and Mobile are growing slowly (25.71% and 5.77% population were lost, respectively from each city from 1990 to 2020) compared to other cities like Hoover and Madison with more than 130% and 290% growth shown from 1990 to 2020, respectively. Birmingham and Mobile presence in the study aided in determining if population loss effects urban development and growth. Such exponential changes to population growth within a short span of time can led to unprecedented increase in impermeable areas leading to various sustainability issues. Therefore, it is extremely important to study the trend of land use changes and make projections on the future expansion of cities, especially Alabama SMSCs which has potential to grow more.

The following two research tasks have been carried out to conduct an analysis of the patterns of urbanization in the state of Alabama:

• Determine the expansion of urban built-up areas over time (1990 to 2020) for the ten Alabama cities (population change greater than 5% from 1990 to 2020 (Table 1)) using a supervised classification technique in GEE (Figure 1).

• Project future urban growth scenarios of 2050 for all ten cities using cellular automata (CA) Markov model combined with GIS based on the LUCCs in 2010 and 2020.

2. Materials and Methods

2.1. Land Use and Land Cover Analysis

2.1.1. Data

GEE provides a huge collection of selection for Earth observation data (EOD) encompassing satellite images from popular platforms such as Sentinel, Landsat, and MODIS, as well as other climate and demographic data. In this study, we used atmospherically corrected Landsat 5 and Landsat-8 surface reflectance Tier 1 data for years 1990, 2000, 2010, 2020, and 2021. Area of interest including city boundaries and surrounding rural areas was created in ArcGIS pro and imported into GEE using the shapefile upload option. The unit of analysis was the pixel, with each pixel in Landsat representing 30 m × 30 m. LUCC was divided into four major classes: water bodies, vegetation, barren land, and built-up areas. All green areas were considered vegetation, while rivers and ponds were considered water bodies. The study made use of visible bands – red, green, blue, and other bands such as near-infrared, and short-wave infrared for Landsat 5 and Landsat 8 imageries for the analysis (Table 2).

2.1.2. Methods

In order to classify the images into desired land use land cover classification the methodology shown in Figure 2 was used. The images were imported using ‘ee.ImageCollection’ function and the areas of interest created in ArcGIS pro was imported into the script using the ‘ee.featurecollection’ function. The images were filtered for dates from January 1 to December 31 for each year, no cloud and no cloud shadows. A composite image was then created with the filtered input images using the median function which resulted in a median value assigned to each pixel in the whole image stack, resulting in a single image for the entire image collection. The normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), modified normalized difference water index (MNDWI), and bare soil index (BSI) were calculated and added to the composite image as bands for improved classification. NDVI and NDBI were added to better differentiate between natural and built-up areas. BSI is traditionally used to differentiate bare areas such as houses, roads, bare open spaces, and eroded areas. It can prove beneficial to classify built-up area [45,46]. MNDWI is added for better classification of urban and water features as it can effectively suppress or remove built-up land noises and vegetation and soil noises [47]. It is necessary to classify water efficiently as they are used later to calculate distance from anthropogenic changes and distance from water for future LUCC predictions.

The NDVI [48] is the normalized difference between the NIR and red bands, the NDBI [49] is the normalized difference between the NIR and SWIR bands, the MNDWI [50] is the normalized difference between GREEN and SWIR bands, and the BSI [51] is the difference between the combination of RED, SWIR, BLUE and NIR bands as shown in Equations (1), (2) and (4):

NDVI=NIR−RED/NIR+RED(1)

NDBI=NIR−SWIR/NIR+SWIR(2)

MNDWI = GREEN – SWIR/GREEEN+SWIR(3)

BSI = ((RED+SWIR) − (NIR+BLUE))/((RED+SWIR) + (NIR+BLUE))(4)

The composite image was also used to create training samples for each type of land use class: water, vegetation, urban, and barren. In total 5038 polygons for all 10 cities for each year analyzed were created for training and loaded into GEE as feature collection. The training samples were created as polygons using visual interpretation in GEE. 100 polygons for each land use type for each city were created. The edge pixels were avoided to provide the algorithm with polygons containing pixel values belonging to each landuse type for best results. Since NDVI, NDBI, and MNDWI were added as bands in the composite image the other bands were also normalized. Sixty percentage of the training samples were randomly selected for classification and 40% for validation. The composite image was then classified using RandomForest (RF) algorithm available within GEE into the desired classes. There are other algorithms such as classification and regression tree (CART), support vector machine (SVM) available with GEE for LUCC. RF is the most commonly used classifier that builds an ensemble classifier [52] by combining many CART trees. Multiple decision trees are generated by RF utilizing a random selection of training datasets and variables [52]. RF is constructed by using a technique called bagging in which individual decision trees serve as parallel estimators [53]. Because of this increasing the number of trees in R, it does not cause overfitting. After a certain point, adding more trees does not improve the accuracy of the model, but it does not detract from the accuracy either [54]. For this analysis the number of decision trees used were 50. The classified images were then clipped for desired AOI and exported as .tif files to google drive to be imported into ArcGIS pro and TerrSet LCM for further analysis.

2.1.3. Accuracy Assessment

To understand the results of the classification, an accuracy assessment is required. It is vital that the thematic classification is accurate because important application decisions will be made using these data. The polygons were created using satellite imagery of 30 m spatial resolution and visual interpretation and were divided into training and validation sets. 60 percent, or 3022 polygons, were used for training and 30 percent, or 2015 polygons, were used as testing sets for all ten cities. GEE includes a confusion matrix method that validates and then evaluates the classification accuracy of the images. The following equations are used to compute the overall accuracy (OA) and kappa coefficient (k):

OA = (V_c/V_t) × 100 (5)

where V_c is the number of pixels classified correctly and V_t is the total number of pixels.

(6)$\mathrm{k}=\frac{\mathrm{N}{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{m}}_{\mathrm{ii}}-{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{C}}_{\mathrm{i}}{\mathrm{G}}_{\mathrm{i}}\right)}{{\mathrm{N}}^2-{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{C}}_{\mathrm{i}}{\mathrm{G}}_{\mathrm{i}}\right)}$

where i = class number, N = total number of classified values compared to true values, m_ii = number of values belonging to the truth class i that have also been classified as class i (i.e., values found along the diagonal of the confusion matrix), C_i = the total number of predicted values belonging to class i, and G_i = the total number of truth values belonging to class i. For each class, the consumer accuracy is derived by the percentage of correctly categorized pixels to the total number of classified pixels. Producer accuracy is also measured by the ratio of correctly classified pixels to total pixels in the reference data for each class. Classification errors are compared to errors in completely random classes to estimate the proportionate reduction in errors. The magnitude is often in the range of -1 to +1. If the value is more than +0.5 [52], the classification is considered acceptable.

2.2. Future Growth Analysis

Future growth predictions for urban areas are essential for understanding and mitigating the effects of rising human activities in cities. The future growth of all ten urban areas has been predicted using IDRISI TerrSet [55] Land Change Modeler (LCM) software for 2050 based on the results from the supervised classification conducted in GEE for the year 2010. The workflow of the process used for the prediction of land use change into the future is shown in Figure 3. According to Clark Labs, LCM is an integrated and innovative land planning and decision-making tool that is fully functional into the TerrSet software. LCM has an automated, user-friendly flow that simplifies the complexities of change analysis and rapidly analyze LUCC change patterns, predict the change, and validate the predicted outputs as well.

The land change prediction process in LCM moves in a stepwise fashion from (1) Change Analysis, (2) Transition Potential Mapping, to (3) Change Prediction. It needs two land cover maps of two different timelines to project future scenarios.

2.2.1. Change Analysis

Change is evaluated between time 1 and time 2 (2010 and 2020 in this study) between two land cover maps. It is done as transitions from one land cover state to another by evaluation of gains and losses, net change, persistence, and specific transitions both in maps and graphical format. This is done to identify dominant transition that can be grouped and modeled, also known as sub-models in the transition potential step. Each of these sub-models are modeled separately and at the end each of those sub-models are combined with all other sub-models for the final prediction. The results of change analysis of different land covers between year 2010 and 2020 is explained in the results section of this paper.

2.2.2. Transition Potential

After the change analysis the next step is transition potential modeling where the potential of the land to transition is identified. Transition potential maps are created which are maps of suitability for each transition. They can consist of a single or group of transitions that are believed to have the same underlying driver variables which are used to model the historical change process. A collection of such transition potential maps is created and organized within each transition sub-model identified in change analysis step that has the same underlying driver variables. The underlying driver variables which were used in this study are digital elevation model (DEM), slope, aspect, distance from roads, distance from water bodies, distance from settlements, and evidence likelihood layer. Driver variables used to evaluate these transitions potential map for Mobile is shown in Figure 4. The underlying driver variables utilized in LCM can be either static or dynamic and can be recalculated and reentered at regular intervals. The DEM was extracted from USGS Aster 30m resolution DEM. The slope and aspect were calculated and created from the DEM in ArcGIS Pro. For the distance from road, settlement, and waterbody, layers from OpenStreetMap data and datasets provided from the city governments for 2010 were used. The vector layers were changed to raster and combined to create a final layer for road, settlement, and waterbody. These layers were imported into TerrSet and the distance layer were created using the distance function in the software. The evidence likelihood layer is created from change analysis transition of the change analysis step and landcover of 2010. Multi-Layer Perceptron (MLP) neural network, Decision Forest (DF) machine learning, Logistic Regression, Weighted Normalized Likelihood (WNL), Support Vector Machine (SVM), or a similarity-weighted instance-based machine learning tool (SimWeight) are used to model the transition which are later used to predict future scenarios. Multi-Layer Perceptron (MLP) was used for this study to create the transition potential maps. The choice of MLP is based on the assumption that the driver variables for all transitions are the same, can accurately model all of the transitions that are gathered into a submodel, can model non-linear relationships, and can model multiple transitions simultaneously.

2.2.3. Change Prediction

Change prediction is the last stage of future land change study. On the basis of historical change rates from change analysis step and the transition potential model from transition potential step, LCM is able to anticipate a future condition for a suitable future date. Additionally, LCM permits the incorporation of incentives and limits, such as zoning maps and future infrastructure plans. This study estimated the landuse in 2050 using CA Markov, preserving three variables (distance from roads, distance from water bodies, and distance from settlements) as dynamic, and DEM, slope, and evidence likelihood as static. This research did not include any restrictions or incentives. LCM generates two types of predictions: (1) hard predictions and (2) soft predictions. Based on a multi-objective land allocation (MOLA) module [56], a hard prediction generates a predicted map [57] where each pixel is allocated one of the land cover classes based on its most likely likelihood. Soft prediction assesses the risk that a pixel may transition to another land category by developing a vulnerability map in which each pixel is assigned a value between 0 and 1 [57].

2.2.4. Model Validation

Model validation is a very important step in the modelling process [58]. The accuracy of the model can be assessed by validation panel in the change prediction tab of the LCM. It helps to evaluate the quality of the predicted land use map in relation to a map of reality. It is conducted using a 3-way crosstabulation between the later landcover map, the prediction map, and a map of reality and ROC statistic (also known as the Area Under the Receiver Operating Characteristic Curve - or AUC). 2021’s simulated map created in change prediction in LCM is compared with 2021’s actual land use land cover map created with supervised classification in GEE. ROC is used to determine how effectively a continuous surface predicts the locations given a Boolean variable’s distribution. It is calculated as a graph between rate of true positives on the vertical axis and rate of false positives on the horizontal axis. Its value ranges between 0 and 1, where 1 shows a perfect fit and values closer to 0.0 shows a random fit. For this analysis the threshold value for calculating the ROC statistics used was 100 and the soft prediction of landcover for year 2021 was used as input file and the actual change between 2010 and 2021 was used as a reference file. For 3-way crosstabulation the images of hard prediction of 2021 and classified image of 2021 were used as inputs. The output will illustrate the accuracy of the model results with a raster with green, red, and yellow pixels where:

A|B|B = Hits (green), i.e., Model predicted change and it changed

A|A|B = Misses (red), i.e., Model predicted persistence and it still changed

A|B|A = False Alarms (yellow), i.e., Model predicted change and it persisted.

3. Results

This paper had two parts, one is the LUCC of the ten most populated cities of Alabama from 1990 to 2021 and second, the future growth prediction of all ten cities of Alabama up to 2050. To quantify urban expansion and future growth, GIS techniques in GEE and TerrSet LCM were used, respectively.

3.1. Land Use and Land Cover Analysis

3.1.1. Land Use and Land Cover Classification

Many previous studies proved the efficiency of satellite image and remote sensing classifying LUCC. A supervised classification approach with random forest algorithm in GEE was used to classify the images. Random training samples were collected from each image for years 1990, 2000, 2010, 2020, and 2021 for each city and used to classify the images. After that, an accuracy assessment was performed to validate the classification of each image.

Figure 5 is a graphical representation of urban built-up area expansion from 1990 to 2020 for each city. As depicted, all the cities have been growing throughout the decade. Some of the steepest growth rates can be seen for Auburn, Dothan, Tuscaloosa, and Mobile. All these cities are small cities as compared to medium-sized cities such as Birmingham and Huntsville. The graph also represents a steady growth of cities up to 2010 and a steep increase in urban area in the last decade.

Different urban areas showed varied spatial patterns of growth due to influential factors like presence of transportation routes and water bodies which has been established in past studies [59,60,61]. Two basic forces that rule the development of an urban area’s functions, shape, and pattern are centrifugal and centripetal [62]. The former may clarify how functions and populations move from the center to the periphery of a city, while the latter keeps those functions in the center and makes it the gravitational center for the entire urbanized area [62]. Several urban functions and forces have resulted in different urban types (linear, grid, radial, etc.) [63]. A linear pattern runs parallel to a major urban transit route (interstate, highway, or railway) or physical infrastructure (such as a river) [63]. The grid pattern is the product of transportation routes being accessible and functions being available in areas that grow from restricted locations such as river or road junctions or islands [64]. Centrifugal forces along many transportation routes primarily create the radial pattern [64]. Most of the expansions of urban areas follow the lines of major transportation routes. As a result, sometimes, spatial pattern of growth was linear and sometimes radial or grid. Birmingham and Dothan having the interstate running through it has a linear growth; Mobile being around gulf shores had grid pattern of urban expansion and cities like Auburn, Montgomery, and Tuscaloosa had a radial pattern due to the growth being around university or water body. Table 3 shows the net addition of urban built-up area from 1990–2020 for all ten cities. It indicates that Dothan, Auburn, and Tuscaloosa have the greatest net addition in urban area study area and Birmingham, and Hoover have the least from 1990 to 2020.

Montgomery is the capital of Alabama. The city started growing at the intersection of I-65 and I-85 in 1982. Since then, it has spread towards the east and south. Gradually Montgomery took the form of a grid [64] filling in over the years. Based on Table 3 Montgomery added 54.79% of urban areas from 1990 to 2020. Urban expansion of Dothan, on the other hand, mainly followed a radial pattern. It spread from central part to periphery of the study area along US highways 431, 231 and 84 and state highways 1 and 53. This kind of expansion is mainly the consequences of centrifugal forces [62]. In 1990, the concentration was in mainly central part of the study area. From 2000, it started to spread towards periphery along several transportation routes. The net addition of urban built area was the highest for Dothan among the cities. Decatur experienced urban growth along water bodies. In 1990, urban expansion was limited to the river side (Tennessee River) and along I-65 which runs in a north to south direction with some in the western side of the I-65. From 2000 we can see the growth in all direction in Decatur. However, for Auburn, urban built-up expansion mainly follows north-east to south-west direction. From 1990 to 2010, it has taken place on both side of I-85. Urban expansion mainly concentrated in the central place of study area (due to presence of urban functions, one of the main being Auburn University) and expansion was more apparent in the southern part than the northern part of Auburn. Tuscaloosa on the other hand, though has similar urban function to Auburn with University of Alabama as the central growth pivot, experienced spatial expansion of urban built-up area around a water body (Black Warrior River). Most of the expansion occurred south of the water body. It did not follow any significant transportation route. From 1990 to 2010, water bodies decreased gradually although built-up did not increase significantly. The net addition in these ten years was 108.96 percent for Tuscaloosa urban area (Table 3). There are two adjacent urbanized areas. One is Birmingham and Hoover and another one is Huntsville and Madison. Though Birmingham is losing population, Hoover is gaining but for urban area expansion both are gaining at a different rate. For Birmingham, this addition was mainly concentrated in central parts following interstate 65 (north-east to south-west direction) such as downtown and university areas. Significant growth of Hoover took place in the north to south directions. The linear pattern of urban expansion was along interstate 65 (I-65) which also goes from north to south direction. Spatial expansion of urban or built-up areas of Huntsville intends to follow major transportation routes and is highly concentrated in central and western part of the study area. For Madison, in 1990 urban built-up areas were mainly concentrated near I-565. From 2000, it started spreading north. In 2010, it dispersed to all directions. Hoover and Madison have developed as extension satellite cities of Birmingham and Huntsville, respectively.

This study also reveals more conversions in certain categories. Mostly the non-urban category of LUCC has been encroached by urban built-up area. One of the main focuses of this study was urban expansion of small and medium-sized areas. Birmingham, Tuscaloosa, and Montgomery are the three largest urban areas in Alabama and have shown steady growth, as opposed to the excessive growth in the mid-sized urban areas (Auburn, Dothan, and Hoover). Below is the figure showing the LUCC changes for Mobile, AL from 1990 to 2020 (Figure 6). Mobile urban area mainly situated near the banks of several rivers (Alabama River, Mobile River, Tombigbee River). Initially it grew near the rivers and later it spread from east to west. Mobile shrunk 2.7 percent in terms of population from 1982 to 2010. However, it’s urban built-up area increased significantly which highlight urban expansion or sprawl. The net addition to urban built-up was 130 percent which is quite high.

3.1.2. Accuracy Assessment of Classified Images

Because of the limited availability of ground truth data, it was impossible to perform accuracy assessment for all images with authenticity. Therefore, the strategy adopted to assess the accuracy is to calculate it using stratified random sampling method and Kappa statistics. The Kappa statistic is a “discrete multivariate technique used in accuracy assessment” [2,65]. Kappa analysis produces the K^ statistic, which approximates the Kappa. It measures the accuracy or harmony connecting the classification map from remotely sensed data, and the reference data specified the chance agreement and the major diagonal, which is specified by the column and row totals [2,65].

Results (Table 4) revealed that overall accuracy met the minimum 85 percent accuracy level which determined by the Anderson image classification scheme [66]. Various literatures mentioned that the ‘vegetation’ land cover type caused most of the error because it contains different types of landuse (Table 5) [2,24].

3.2. Future Growth Analysis

3.2.1. Change Analysis

A change analysis was performed during period 2010–2020 for each city. Each of the ten cities experienced loss in non-urban areas and an increase in urban areas during 2010–2020. Auburn, Tuscaloosa, and Dothan saw the greatest increase in urban areas during 2010–2020 leading to 1118, 3915, and 3715 hectares which accounted for 35.22%, 36.83%, and 54.09% increase in urban areas in 10 years (Table 5).

Table 5 indicates that during 2010–2020 there was 21.48% and 22.59% net change in Birmingham and Hoover, and Huntsville, Madison, and Decatur. It was highest for Dothan (54.09%) and lowest for Montgomery (19.5%). Some of the ‘water’ category has also changed to urban land use. As a result, urban built-up increased continuously and the water and non-urban category decreased gradually for all cities.

Table 6 represents the contributors to the net change in percentage for each land use type used in the classification for the different cities. It is seen from the table that the biggest contributor for land use change to urban is the non-urban area (vegetation and barren) for each of the cities. Dothan is experiencing the biggest change in area from non-urban to urban combined. Table 6 also shows that mostly all other LUCC was converted to urban built-up area, but the rate of change varied from city to city. The ‘non-urban’ category here includes vacant land as categorized as barren LUCC and various types of vegetation; thus, it is understandable why this transition is most common. Urban built-up area mostly remained unchanged since it is very unusual and expensive to convert built-up area to vegetation or water body.

3.2.2. Transition Potential for the Cities

The transition potential maps for each city are created for all four transitions: toUrban, toVegetaion, toWater, and toBarren using the MLP neural network to predict the land use and land cover (LUCC) change for 2050. The transition probabilities were calculated using the Markov chain using the six driving variables shown in Figure 7 for Mobile, AL. The modeler also reports on the model’s accuracy and ability in predicting whether the validation pixels will adjust and, if so, to which class. The accuracy calculated minus the accuracy expected by change is the skill measure. Table 7 shows the model skill measure and accuracy rate to urban sub-model for each of the 10 cities.

3.2.3. Future Urban Built-Up Expansion

Based on these factors, the predicted LUCC of 2050 was produced for all the ten cities of Alabama. The 2050 growth projections in Table 8 indicates that built up area for all ten cities of Alabama will continue to grow at a fast rate. Figure 7 shows the LUCC classification (2010-2020) and land cover projection (2050) for Mobile city in Alabama. The other cities prediction maps are available in appendix. There is a significant increase of urban area from 2020 to 2050 with business-as-usual scenario.

From Table 8, the projected 2050 urban built-up area shows a significant increase as compared to water bodies and the non-urban category. The growth projections for 2050 show that Montgomery will have an increase of 9828 hectares. The two neighboring cities of Birmingham and Hoover will increase by 25,614 hectares. Similar trend is seen for the neighboring cities of Huntsville and Madison and Decatur projecting an increase of 12,854 hectares. The model output projected 15,618 hectares increase for Mobile, 21,085 hectares for Tuscaloosa, 12,267 hectares for Dothan, 9828 hectares for Montgomery and 4237 hectares for Auburn

Annual future growth in percentage will be the highest in Dothan with 4.05% (2010–2050) (Table 7). Auburn will grow fast (2.84% annual change rate 2010–2050). Birmingham and Hoover will grow at a rate of 1.92% (2010–2050). All these ten cities will see a significant growth in urban areas in the next 40 years compared to 2010.

Future prediction for 2050 showed that only Auburn will grow faster in next 30 years (105.93 ha annually) compared to last 30 years (70.06 ha annually). On the other hand, Madison, Hoover, Birmingham, and Mobile will grow at a slower pace in the future.

3.2.4. Validation of Future Growth Scenarios

The inbuilt validation module of the TerrSet software was used to validate the result of the prediction. 2021 imagery was classified beforehand with accuracy assessment shown in Table 4. The CA Markov model inside the LCM was then used to predict the landuse classification for 2021 using the same driver variables used to predict for 2050. The results were then compared using the model validation module inside the change prediction tab of the LCM. It provides a raster with hits, misses, and false alarms for each of the landuse conversion (Figure 8).

The validation results from ROC statistics which provides Area under curve (AUC) value for each LCM model for each city is shown in Table 9. All the simulations have value greater than 0.5 providing satisfactory results.

The results of the simulation indicate that there will be a significant urban built-up expansion in the future. Transportation and physical landform acted as driving forces for urban built-up expansion. Accessibility to main road, water, settlements, slopes, aspect, and altitude will also act as driving forces for urban built-up expansion in the future. Future growth prediction for urban areas is important to help plan and implement mitigation schemes to reduce impacts of increasing anthropogenic activities in cities. Thus, having knowledge of how the urban areas will look in next few decades will benefit planning the cities and informing policymakers.

4. Discussion

As seen from the results in Section 3, this study has demonstrated the usefulness of satellite remote sensing and digital image processing for LUCC classification. For future growth projection CA-Markov model proved its effectiveness. This study has also examined the evolution of urban spatial form for urban built-up areas in state of Alabama. Significant growth patterns of urban expansion were found in all ten cities. As aforementioned in Section 3.1.1, the spatial pattern of urban expansion in all the cities are influenced by the presence of transportation routes and water bodies. Birmingham, Montgomery, Dothan, Huntsville, Madison, and Auburn showed significant patterns of growth around transportation routes such as interstate highways and state highways. This expansion can be attributed to economic activities that highways attract and help develop the area along the highway lines. Examples of which are gas stations, hotels, and other service-related businesses. However, access to water bodies seemed to have dominated the growth of urban areas in Mobile, Decatur and Tuscaloosa, by encouraging development around them.

Some of the highlights of the classification of urban areas are:

• Mainly forest, barren land, and grassland have been urbanized.

• Most of the urban expansion took place along the interstates. As a result, most of the study areas exhibited linear pattern of urban expansion such as Birmingham, Hoover, and Auburn.

• Some areas are the results of centrifugal force [62] for instance Dothan, Mobile, Huntsville and Montgomery. Some urban areas exhibited dispersed patterns such as Tuscaloosa, Madison, and Decatur.

It was also noted as mentioned in Section 3.2.1, the major contributors to urban areas in all the cities are vegetation and barren land. It iterates the fact that natural pervious surfaces are converted to man-made impervious surfaces at exponential rates in SMSCs. This exposes the cities to different disasters making them extremely vulnerable since such cities receive limited funding and limited research. Predicted urban growth also represented the same kind of pattern in the study areas, highlighting growth centering the transportation routes and water bodies. However, it should be noted that the future growth scenario depends on the driver variables selected such as distance from road, distance from settlement, distance from water, likelihood to change, slope, and elevation among others. Cities development plans and population change are few of the variables which are not included in this study and should be considered for further studies.

5. Conclusions

In this paper, we apply random forest algorithm in GEE to classify four decades of satellite images for ten cities of Alabama and used the LCM of TerrSet software to conduct a future change study to 2050 for all the ten cities. This study has established a well-documented regional case focusing on Alabama. Findings of this study should be utilized in future urban planning strategies, managing resources, and providing direction in a rapidly changing environment. This study has provided the changing form and shape of the cities, past, present, and future which can help guide town planners, road network management and landuse.

The 5th Intergovernmental Panel on Climate Change (IPCC) report mentioned that many global risks of climate change are concentrated in urban areas and will be on the rise in future [67]. For example, heat stress, extreme precipitation, flooding, air pollution, drought, and water scarcity pose risks in urban areas for people. Therefore, it is important to understand the impacts, vulnerabilities, and adaptation measures suitable to improve life on Earth. Techniques like CA-Markov future growth model and GIScience are effective tools to aid sustainable planning and development because they can illustrate the foreseeable changes. A well-planned sustainable development strategy can decelerate the negative impacts of the modern era urban growth thus highlighting the importance and need of studies like the one presented in this paper.

Author Contributions

Conceptualization, C.M., M.S.; methodology, C.M., M.S.; software, Auburn University; validation, M.S., M.R. and C.M.; formal analysis, M.S.; investigation, M.S., L.M. and C.M.; resources, NASA and Clark Laboratory and other online resources; data curation, M.S.; writing—original draft preparation, M.S., C.M. and M.R.; writing, M.S., C.M. and L.M.; visualization, M.S.; supervision, C.M., L.M.; project administration, C.M.; funding acquisition, C.M. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

Not Applicable.

Acknowledgments

We acknowledge the support of Department of Geosciences, Auburn University. We would also like to acknowledge the contribution of posthumous Mahjabin Rahman, a graduate student in the Department of Geosciences, Auburn University.

Conflicts of Interest

The authors declare no conflict of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. Location of ten studied cities in Alabama.

Figure 2. Flowchart showing this study’s LUCC classification in GEE.

Figure 3. Flow diagram showing this study’s Change detection model.

Figure 4. Driver variables used for change prediction (Mobile City, Alabama).

Figure 5. Urban Built-up Area Statistics (1990–2020).

Figure 6. Land use Land cover map of Mobile, AL (1990–2020).

Figure 7. Predicted Urban Built-up Expansion, 2050 for Mobile city.

Figure 8. Validation output for Mobile, AL from validation module of the LCM, TerrSet.

Table 1

Growth (increase and decrease) in population for the ten cities in Alabama from 1980 to 2020. Source: United States Census Bureau, 2020 [44].

Study Areas (Based on City Population)	Population		Population Change, 1990 to 2020
	1990	2020	Number	Percent (%)
Birmingham	265,968	197,575	−68,393	−25.71
Montgomery	187,106	198,665	11,559	6.18
Mobile	196,278	184,952	−11,326	−5.77
Huntsville	159,789	216,963	57,174	35.78
Tuscaloosa	77,759	100,618	22,859	29.40
Hoover	39,788	92,589	52,801	132.71
Dothan	53,589	71,175	17,586	32.82
Decatur	48,761	57,804	9043	18.55
Auburn	33,830	78,564	44,734	132.23
Madison	14,904	58,357	43,453	291.55

Table 2

Landsat 5 and Landsat 8 band information used for the LUCC classification.

Data Layer	Source	Bands Used	Wavelength	Spatial Resolution (m)
Landsat-5 Thematic Mapper ™ surface reflectance Tier 1	Google Earth Engine (GEE), data accessed via the U.S. Geological Survey (USGS)	Blue (Band 1)	0.45–0.52	30
		Green (Band 2)	0.52–0.60	30
		Red (Band 3)	0.63–0.69	30
		Near-Infra-Red (Band 4)	0.77–0.90	30
		Short-Wave Infra-Red 1 (Band 5)	1.55–1.75	30
Landsat-8 Operational Land Imager surface reflectance Tier 1	Google Earth Engine (GEE), data accessed via the U.S. Geological Survey (USGS)	Blue (Band 2)	0.45–0.51	30
		Green (Band 3)	0.53–0.59	30
		Red (Band 4)	0.64–0.67	30
		Near-Infra-Red (Band 5)	0.85–0.88	30
		Short-Wave Infra-Red 1 (Band 6)	1.57–1.65	30

Table 3

Net Addition of Urban Built-up from 1990-2020 for Ten Study Areas.

Study Areas	1990		2020		Net addition (%)
	Area (ha)	%	Area (ha)	%
Auburn	1072	1.82	3174	5.38	196.08
Tuscaloosa	5087	2.46	10,630	5.15	108.96
Birmingham and Hoover	19,104	6.28	28,550	9.39	49.45
Dothan	1785	2.73	6868	10.49	284.76
Huntsville and Madison and Decatur	10,362	3.48	19,843	6.67	91.50
Mobile	8874	4.30	16,058	7.77	80.96
Montgomery	6795	3.29	10,518.21	10.07	54.79

Table 4

Accuracy Assessment and Kappa Statistics of Classified Image.

City	1990
		Vegetation	Water	Urban	Barren
Auburn
	Vegetation	65	0	2	0
	Water	0	63	0	1
	Urban	0	0	62	0
	Barren	0	0	4	11
	Overall Accuracy	96.63%	Kappa Statistics	0.96634
Tuscaloosa
	Urban	70	0	0	0
	Vegetation	1	45	1	0
	Water	0	0	200	0
	Barren	0	0	0	0
	Overall Accuracy	99.37%	Kappa Statistics	0.9937
Birmingham and Hoover
	Urban	108	3	4	3
	Vegetation	1	89	3	0
	Water	1	5	335	2
	Barren	5	2	0	59
	Overall Accuracy	95.32%	Kappa Statistics	0.9532
Dothan
	Urban	76	0	1	1
	Vegetation	1	50	1	0
	Water	3	0	105	4
	Barren	2	0	2	24
	Overall Accuracy	94.44%	Kappa Statistics	0.94444
Huntsville and Madison and Decatur
	Urban	125	1	1	29
	Vegetation	0	130	3	1
	Water	2	2	255	5
	Barren	5	1	2	189
	Overall Accuracy	93.08%	Kappa Statistics	0.9308
Mobile
	Urban	53	0	1	0
	Vegetation	0	76	0	0
	Water	0	1	110	1
	Barren	0	2	1	22
	Overall Accuracy	97.75%	Kappa Statistics	0.9775
Montgomery
	Urban	63	0	1	1
	Vegetation	2	195	0	4
	Water	0	1	110	2
	Barren	3	2	1	79
	Overall Accuracy	96.34%	Kappa Statistics	0.9634
	2000
		Vegetation	Water	Urban	Barren
Auburn
	Vegetation	67	0	2	0
	Water	0	60	0	1
	Urban	0	0	62	0
	Barren	0	0	4	11
	Overall Accuracy	96.63%	Kappa Statistics	0.9662
Tuscaloosa
	Urban	74	0	0	0
	Vegetation	1	49	1	0
	Water	0	0	200	0
	Barren	0	0	0	0
	Overall Accuracy	99.38%	Kappa Statistics	0.9938
Birmingham and Hoover
	Urban	110	0	4	3
	Vegetation	0	87	2	0
	Water	1	0	340	2
	Barren	5	0	0	59
	Overall Accuracy	97.23%	Kappa Statistics	0.9723
Dothan
	Urban	82	0	1	1
	Vegetation	1	52	1	0
	Water	0	0	107	2
	Barren	2	0	2	24
	Overall Accuracy	97.07%	Kappa Statistics	0.9707
Huntsville and Madison and Decatur
	Urban	127	1	1	29
	Vegetation	0	135	1	1
	Water	0	2	261	5
	Barren	5	0	2	190
	Overall Accuracy	93.82%	Kappa Statistics	0.9382
Mobile
	Urban	55	0	1	0
	Vegetation	0	81	0	0
	Water	0	1	101	1
	Barren	0	0	1	22
	Overall Accuracy	98.48%	Kappa Statistics	0.9848
Montgomery
	Urban	62	0	1	1
	Vegetation	2	200	0	4
	Water	0	1	107	2
	Barren	2	2	1	79
	Overall Accuracy	96.55%	Kappa Statistics	0.9655
City	2010
		Vegetation	Water	Urban	Barren
Auburn
	Vegetation	40	0	1	2
	Water	0	46	0	0
	Urban	0	0	42	1
	Barren	0	0	3	10
	Overall Accuracy	95.17%	Kappa Statistics	0.9517
Tuscaloosa
	Urban	49	0	1	0
	Vegetation	1	35	0	0
	Water	2	1	116	0
	Barren	0	0	0	0
	Overall Accuracy	97.56%	Kappa Statistics	0.9756
Birmingham and Hoover
	Urban	61	0	5	5
	Vegetation	1	64	3	1
	Water	2	1	219	2
	Barren	6	0	1	33
	Overall Accuracy	93.32%	Kappa Statistics	0.9332
Dothan
	Urban	51	0	3	0
	Vegetation	2	25	2	0
	Water	0	0	77	0
	Barren	1	0	0	12
	Overall Accuracy	95.38%	Kappa Statistics	0.9538
Huntsville and Madison and Decatur
	Urban	76	0	2	20
	Vegetation	2	87	1	1
	Water	1	0	153	6
	Barren	3	0	0	157
	Overall Accuracy	92.93%	Kappa Statistics	0.9293
Mobile
	Urban	45	0	0	0
	Vegetation	2	62	0	0
	Water	0	0	63	1
	Barren	1	0	0	10
	Overall Accuracy	97.83%	Kappa Statistics	0.9783
Montgomery
	Urban	30	1	1	5
	Vegetation	0	131	2	2
	Water	0	2	87	1
	Barren	2	1	2	50
	Overall Accuracy	94.01%	Kappa Statistics	0.9401
City	2020
		Vegetation	Water	Urban	Barren
Auburn
	Vegetation	36	0	0	1
	Water	0	36	1	0
	Urban	0	0	35	0
	Barren	1	0	3	8
	Overall Accuracy	95.04%	Kappa Statistics	0.9504
Tuscaloosa
	Urban	57	0	6	0
	Vegetation	1	32	4	0
	Water	0	2	130	0
	Barren	0	0	0	0
	Overall Accuracy	94.40%	Kappa Statistics	0.944
Birmingham and Hoover
	Urban	76	0	6	3
	Vegetation	2	58	4	2
	Water	3	2	232	1
	Barren	1	0	1	36
	Overall Accuracy	94.15%	Kappa Statistics	0.9415
Dothan
	Urban	57	0	2	2
	Vegetation	1	39	0	1
	Water	1	0	58	1
	Barren	1	0	3	18
	Overall Accuracy	93.48%	Kappa Statistics	0.9348
Huntsville and Madison and Decatur
	Urban	99	0	1	16
	Vegetation	1	82	0	0
	Water	1	1	167	8
	Barren	6	0	1	137
	Overall Accuracy	93.14%	Kappa Statistics	0.9314
Mobile
	Urban	43	0	1	2
	Vegetation	3	56	0	0
	Water	0	0	64	0
	Barren	1	0	5	4
	Overall Accuracy	93.30%	Kappa Statistics	0.933
Montgomery
	Urban	32	1	1	1
	Vegetation	1	113	0	5
	Water	0	2	76	2
	Barren	5	0	2	60
	Overall Accuracy	93.36%	Kappa Statistics	0.9336
City	2021
		Vegetation	Water	Urban	Barren
Auburn
	Vegetation	36	0	0	1
	Water	0	37	0	0
	Urban	0	0	35	0
	Barren	1	0	3	8
	Overall Accuracy	95.87%	Kappa Statistics	0.9587
Tuscaloosa
	Urban	56	0	7	0
	Vegetation	1	35	1	0
	Water	0	2	130	0
	Barren	0	0	0	0
	Overall Accuracy	95.26%	Kappa Statistics	0.9526
Birmingham and Hoover
	Urban	74	0	5	6
	Vegetation	2	58	4	2
	Water	7	3	226	2
	Barren	3	0	0	35
	Overall Accuracy	92.04%	Kappa Statistics	0.9204
Dothan
	Urban	58	0	2	1
	Vegetation	1	39	0	1
	Water	1	0	58	1
	Barren	0	0	2	20
	Overall Accuracy	95.11%	Kappa Statistics	0.9511
Huntsville and Madison and Decatur
	Urban	91	0	3	22
	Vegetation	1	82	0	0
	Water	0	0	169	8
	Barren	6	0	3	125
	Overall Accuracy	91.57%	Kappa Statistics	0.9157
Mobile
	Urban	44	0	1	1
	Vegetation	2	57	0	0
	Water	0	0	64	0
	Barren	2	0	5	3
	Overall Accuracy	93.85%	Kappa Statistics	0.9385
Montgomery
	Urban	30	0	0	5
	Vegetation	5	109	0	5
	Water	0	3	75	2
	Barren	3	0	3	61
	Overall Accuracy	91.36%	Kappa Statistics	0.9136

Table 5

Gains, losses, net change (ha), and net change (%) for the different land use and land cover for all ten cities for 2010-2020.

City	Class	2010–2020
		Gains (ha)	Losses (ha)	Net Change (ha)	Net Change (%)
Auburn
	Urban	1491	374	1118	35.22
	Vegetation	2750	9859	−7108	13.81
	Water	228	75	153	−18.85
	Barren	9172	3290	5837	34.39
Tuscaloosa
	Urban	5117	1201	3915	36.83
	Vegetation	9554	34,885	−25,331	−18.64
	Water	1283	1475	−192	−2.49
	Barren	32,212	10,605	21,607	48.62
Birmingham and Hoover
	Urban	9768	3634	6134	21.48
	Vegetation	12,457	39,425	−26,968	−12.38
	Water	1566	1428	138	2.38
	Barren	33,943	13,246	20,696	40
Dothan
	Urban	4163	449	3715	54.09
	Vegetation	4068	5612	−1544	−5.77
	Water	242	93	150	25.28
	Barren	4541	6861	−2320	−7.43
Huntsville and Madison and Decatur
	Urban	8096	3614	4482	22.59
	Vegetation	18,082	25,772	−7690	−5.32
	Water	3246	487	2758	16.14
	Barren	22,820	22,370	450	0.39
Mobile
	Urban	6494	1922	4573	28.48
	Vegetation	6102	17,785	−11,683	−10.66
	Water	1684	1141	543	1.00
	Barren	13,381	6814	6567	24.39
Montgomery
	Urban	3558	1678	1881	19.50
	Vegetation	6229	9295	−3066	−5.68
	Water	1215	303	912	18.27
	Barren	7891	7618	273	0.76

Table 6

Contributors to the net change in area for each land use type for all ten cities in hectares (2010–2015).

City	2010–2020
	Contributors (in%)
Auburn		Urban	Vegetation	Water	Barren
	Urban	0.00	1.17	2.24	5.52
	Vegetation	−25.54	0.00	−12.83	−58.02
	Water	1.04	0.27	0.00	0.08
	Barren	−29.87	14.41	−0.95	0.00
Tuscaloosa
	Urban	0.00	1.19	2.12	8.01
	Vegetation	−28.58	0.00	−3.45	−101.35
	Water	−2.50	0.17	0.00	−1.30
	Barren	−27.24	14.36	3.75	0.00
Birmingham and Hoover
	Urban	0.00	1.41	5.65	7.6
	Vegetation	−15.41	0.00	−9.01	−74.11
	Water	−1.42	0.21	0.00	−0.17
	Barren	−10.53	9.39	0.92	0.00
Dothan
	Urban	0.00	3.62	−8.77	8.13
	Vegetation	−32.55	0.00	−26.40	−1.20
	Water	1.23	0.41	0.00	−0.02
	Barren	−86.48	1.42	1.34	0.00
Huntsville and Madison and Decatur
	Urban	0.00	0.47	−1.56	3.44
	Vegetation	−4.69	0.00	−14.37	−4.24
	Water	1.45	1.35	0.00	0.41
	Barren	−25.94	3.23	−3.32	0.00
Mobile
	Urban	0.00	2.20	0.13	8.98
	Vegetation	−23.27	0.00	−1.10	−41.36
	Water	−0.63	0.49	0.00	0.12
	Barren	−15.91	6.94	−0.05	0.00
Montgomery
	Urban	0.00	1.23	−6.8	4.08
	Vegetation	−9.07	0.00	−11.41	−5.33
	Water	3.57	0.82	0.00	0.48
	Barren	−18.72	3.32	−4.15	0.00

Table 7

Model skill breakdown for all driving variable for all 10 cities for modeling change to urban areas sub model.

City	Skill Measure	Accuracy Rate (%)
Auburn	0.6393	72.95
Tuscaloosa	0.3412	50.59
Birmingham and Hoover	0.6703	72.52
Dothan	0.65	73.75
Huntsville and Madison and Decatur	0.6275	72.06
Mobile	0.5489	66.17
Montgomery	0.6692	75.19

Table 8

Predicted Urban Built-up by 2050.

Study Area	LUCC	Area 2010 (Hectares)	Area 2050(Hectares)	Difference		Annual Rate of Change (2010–2050) in %
				Hectares	(%)
Auburn	Urban built-up	2056	6293	4237	206	2.84
Tuscaloosa	Urban built-up	6715	27,800	21,085	314	3.62
Birmingham and Hoover	Urban built-up	22,416	48,030	25,614	114	1.92
Dothan	Urban built-up	3154	15,421	12,267	389	4.05
Huntsville and Madison and Decatur	Urban built-up	15,361	28,215	12,854	84	1.53
Mobile	Urban built-up	11,486	27,104	15,618	136	2.17
Montgomery	Urban built-up	7766	17,594	9828	127	2.07

Table 9

Area under curve value for LCM prediction model for each city.

Study Areas	AUC
Birmingham and Hoover	0.533
Montgomery	0.703
Mobile	0.606
Huntsville, Madison, and Decatur	0.64
Tuscaloosa	0.584
Dothan	0.683
Auburn	0.707
AUC is Area Under the ROC curve. Value closer to 1.0 points towards positive predictions

References

1. United Nations, Population Division. World Urbanization Prospects: The 2018 Revision (ST/ESA/SER.A/420); United Nations: New York, NY, USA, 2019; ISBN 978-92-1-148319-2

2. Yang, X. Satellite Monitoring of Urban Spatial Growth in the Atlanta Metropolitan Area. Photogramm. Eng. Remote Sens.; 2002; 68, pp. 725-734.

3. Islam, M.S.; Ahmed, R. Land Use Change Prediction in Dhaka City Using Gis Aided Markov Chain Modeling. J. Life Earth Sci.; 2011; 6, pp. 81-89. [DOI: https://dx.doi.org/10.3329/jles.v6i0.9726]

4. Alig, R.J.; Kline, J.D.; Lichtenstein, M. Urbanization on the US Landscape: Looking Ahead in the 21st Century. Landsc. Urban Plan.; 2004; 69, pp. 219-234. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2003.07.004]

5. US Census Bureau Section 1. Population. Available online: https://www.census.gov/library/publications/2011/compendia/statab/131ed/population.html (accessed on 19 September 2022).

6. US Census Bureau. Decennial Census of Population and Housing by Decades. Available online: https://www.census.gov/programs-surveys/decennial-census/decade.html (accessed on 19 September 2022).

7. Forman, B.; Vey, J.S. Demographic Change in Medium-Sized Cities: Evidence from the 2000 Census; Brookings: Washington, DC, USA, 2002.

8. US Census Bureau. Southern and Western Regions Experienced Rapid Growth This Decade. 2020; Available online: https://www.census.gov/newsroom/press-releases/2020/south-west-fastest-growing.html (accessed on 19 December 2022).

9. KC, B.; Shepherd, J.M.; Gaither, C.J. Climate Change Vulnerability Assessment in Georgia. Applied Geography; 2015; 62, pp. 62-74. [DOI: https://dx.doi.org/10.1016/j.apgeog.2015.04.007]

10. KC, B.; Shepherd, J.M.; King, A.W.; Johnson Gaither, C. Multi-Hazard Climate Risk Projections for the United States. Nat. Hazards; 2021; 105, pp. 1963-1976. [DOI: https://dx.doi.org/10.1007/s11069-020-04385-y]

11. National Assessment Synthesis Team. Climate Change Impacts on the United States: The Potential Consequences of Climate Variability and Change; US Global Change Research Program: Washington, DC, USA, 2000.

12. Clay, M.J. Developing an Integrated Land-Use/Transportation Model for Small to Medium-Sized Cities: Case Study of Montgomery, Alabama. Transp. Plan. Technol.; 2010; 33, pp. 679-693. [DOI: https://dx.doi.org/10.1080/03081060.2010.527178]

13. Ellis, J.T.; Spruce, J.P.; Swann, R.A.; Smoot, J.C.; Hilbert, K.W. An Assessment of Coastal Land-Use and Land-Cover Change from 1974–2008 in the Vicinity of Mobile Bay, Alabama. J. Coast. Conserv.; 2011; 15, pp. 139-149. [DOI: https://dx.doi.org/10.1007/s11852-010-0127-y]

14. Lo, C.P.; Quattrochi, D.A.; Luvall, J.C. Application of High-Resolution Thermal Infrared Remote Sensing and GIS to Assess the Urban Heat Island Effect. Int. J. Remote Sens.; 1997; 18, pp. 287-304. [DOI: https://dx.doi.org/10.1080/014311697219079]

15. Tadesse, W.; Coleman, T.L.; Tsegaye, T.D. Improvement of Land Use and Land Cover Classification of an Urban Area Using Image Segmentation from Landsat ETM + Data. Proceedings of the 30th International Symposium on Remote Sensing of the Environment Geospatial Applied Research; Beijing, China, 15–18 August 2022.

16. Superczynski, S.D.; Christopher, S.A. Exploring Land Use and Land Cover Effects on Air Quality in Central Alabama Using GIS and Remote Sensing. Remote Sens.; 2011; 3, pp. 2552-2567. [DOI: https://dx.doi.org/10.3390/rs3122552]

17. Lambin, E.F.; Ehrlich, D. Land-Cover Changes in Sub-Saharan Africa (1982–1991): Application of a Change Index Based on Remotely Sensed Surface Temperature and Vegetation Indices at a Continental Scale. Remote Sens. Environ.; 1997; 61, pp. 181-200. [DOI: https://dx.doi.org/10.1016/S0034-4257(97)00001-1]

18. Dewan, A.M.; Yamaguchi, Y. Land Use and Land Cover Change in Greater Dhaka, Bangladesh: Using Remote Sensing to Promote Sustainable Urbanization. Appl. Geogr.; 2009; 29, pp. 390-401. [DOI: https://dx.doi.org/10.1016/j.apgeog.2008.12.005]

19. El Garouani, A.; Mulla, D.J.; El Garouani, S.; Knight, J. Analysis of Urban Growth and Sprawl from Remote Sensing Data: Case of Fez, Morocco. Int. J. Sustain. Built Environ.; 2017; 6, pp. 160-169. [DOI: https://dx.doi.org/10.1016/j.ijsbe.2017.02.003]

20. Padmanaban, R.; Bhowmik, A.K.; Cabral, P.; Zamyatin, A.; Almegdadi, O.; Wang, S. Modelling Urban Sprawl Using Remotely Sensed Data: A Case Study of Chennai City, Tamilnadu. Entropy; 2017; 19, 163. [DOI: https://dx.doi.org/10.3390/e19040163]

21. Hadeel, A.; Jabbar, M.; Chen, X. Application of Remote Sensing and GIS to the Study of Land Use/Cover Change and Urbanization Expansion in Basrah Province, Southern Iraq. Geo-Spat. Inf. Sci.; 2009; 12, pp. 135-141. [DOI: https://dx.doi.org/10.1007/s11806-009-0244-7]

22. Shalaby, A.; Tateishi, R. Remote Sensing and GIS for Mapping and Monitoring Land Cover and Land-Use Changes in the Northwestern Coastal Zone of Egypt. Appl. Geogr.; 2007; 27, pp. 28-41. [DOI: https://dx.doi.org/10.1016/j.apgeog.2006.09.004]

23. Yuan, F.; Sawaya, K.E.; Loeffelholz, B.C.; Bauer, M.E. Land Cover Classification and Change Analysis of the Twin Cities (Minnesota) Metropolitan Area by Multitemporal Landsat Remote Sensing. Remote Sens. Environ.; 2005; 98, pp. 317-328. [DOI: https://dx.doi.org/10.1016/j.rse.2005.08.006]

24. Trousdale, M.G. Urban Growth Monitoring of Birmingham, AL Using Landsat MSS and TM Imagery. Master’s Thesis.; University of Alabama Libraries: Tuscaloosa, AL, USA, 2010.

25. Phan, T.N.; Kuch, V.; Lehnert, L.W. Land Cover Classification Using Google Earth Engine and Random Forest Classifier—The Role of Image Composition. Remote Sens.; 2020; 12, 2411. [DOI: https://dx.doi.org/10.3390/rs12152411]

26. Tassi, A.; Vizzari, M. Object-Oriented LULC Classification in Google Earth Engine Combining SNIC, GLCM, and Machine Learning Algorithms. Remote Sens.; 2020; 12, 3776. [DOI: https://dx.doi.org/10.3390/rs12223776]

27. Becker, W.R.; Ló, T.B.; Johann, J.A.; Mercante, E. Statistical Features for Land Use and Land Cover Classification in Google Earth Engine. Remote Sens. Appl. Soc. Environ.; 2021; 21, 100459. [DOI: https://dx.doi.org/10.1016/j.rsase.2020.100459]

28. Liu, X.; Hu, G.; Chen, Y.; Li, X.; Xu, X.; Li, S.; Pei, F.; Wang, S. High-Resolution Multi-Temporal Mapping of Global Urban Land Using Landsat Images Based on the Google Earth Engine Platform. Remote Sens. Environ.; 2018; 209, pp. 227-239. [DOI: https://dx.doi.org/10.1016/j.rse.2018.02.055]

29. Verburg, P.; Schot, P.; Dijst, M.; Veldkamp, A. Land-Use Change Modeling: Current Practice and Research Priorities. GeoJournal; 2004; 61, pp. 309-324. [DOI: https://dx.doi.org/10.1007/s10708-004-4946-y]

30. Orejuela, I.P.; Toulkeridis, T. Evaluation of the Susceptibility to Landslides through Diffuse Logic and Analytical Hierarchy Process (AHP) between Macas and Riobamba in Central Ecuador. Proceedings of the 2020 Seventh International Conference on eDemocracy & eGovernment (ICEDEG); Buenos Aires, Argentina, 22–24 April 2020; pp. 201-207.

31. Tayyebi, A.; Perry, P.C.; Tayyebi, A.H. Predicting the Expansion of an Urban Boundary Using Spatial Logistic Regression and Hybrid Raster-Vector Routines with Remote Sensing and GIS. Int. J. Geogr. Inf. Sci.; 2014; 28, pp. 639-659. [DOI: https://dx.doi.org/10.1080/13658816.2013.845892]

32. Termansen, M.; McClean, C.J.; Preston, C.D. The Use of Genetic Algorithms and Bayesian Classification to Model Species Distributions. Ecol. Model.; 2006; 192, pp. 410-424. [DOI: https://dx.doi.org/10.1016/j.ecolmodel.2005.07.009]

33. Hamad, R.; Balzter, H.; Kolo, K. Predicting Land Use/Land Cover Changes Using a CA-Markov Model under Two Different Scenarios. Sustainability; 2018; 10, 3421. [DOI: https://dx.doi.org/10.3390/su10103421]

34. Wang, S.W.; Munkhnasan, L.; Lee, W.-K. Land Use and Land Cover Change Detection and Prediction in Bhutan’s High Altitude City of Thimphu, Using Cellular Automata and Markov Chain. Environ. Chall.; 2021; 2, 100017. [DOI: https://dx.doi.org/10.1016/j.envc.2020.100017]

35. Munthali, M.G.; Mustak, S.; Adeola, A.; Botai, J.; Singh, S.K.; Davis, N. Modelling Land Use and Land Cover Dynamics of Dedza District of Malawi Using Hybrid Cellular Automata and Markov Model. Remote Sens. Appl. Soc. Environ.; 2020; 17, 100276. [DOI: https://dx.doi.org/10.1016/j.rsase.2019.100276]

36. Giuliani, G.; Rodila, D.; Külling, N.; Maggini, R.; Lehmann, A. Downscaling Switzerland Land Use/Land Cover Data Using Nearest Neighbors and an Expert System. Land; 2022; 11, 615. [DOI: https://dx.doi.org/10.3390/land11050615]

37. Ralha, C.G.; Abreu, C.G.; Coelho, C.G.C.; Zaghetto, A.; Macchiavello, B.; Machado, R.B. A Multi-Agent Model System for Land-Use Change Simulation. Environ. Model. Softw.; 2013; 42, pp. 30-46. [DOI: https://dx.doi.org/10.1016/j.envsoft.2012.12.003]

38. Liping, C.; Yujun, S.; Saeed, S. Monitoring and Predicting Land Use and Land Cover Changes Using Remote Sensing and GIS Techniques—A Case Study of a Hilly Area, Jiangle, China. PLoS ONE; 2018; 13, e0200493. [DOI: https://dx.doi.org/10.1371/journal.pone.0200493]

39. Sang, L.; Zhang, C.; Yang, J.; Zhu, D.; Yun, W. Simulation of Land Use Spatial Pattern of Towns and Villages Based on CA—Markov Model. Math. Comput. Model.; 2011; 54, pp. 938-943. [DOI: https://dx.doi.org/10.1016/j.mcm.2010.11.019]

40. Babbar, D.; Areendran, G.; Sahana, M.; Sarma, K.; Raj, K.; Sivadas, A. Assessment and Prediction of Carbon Sequestration Using Markov Chain and InVEST Model in Sariska Tiger Reserve, India. J. Clean. Prod.; 2021; 278, 123333. [DOI: https://dx.doi.org/10.1016/j.jclepro.2020.123333]

41. Anand, V.; Oinam, B. Future Land Use Land Cover Prediction with Special Emphasis on Urbanization and Wetlands. Remote Sens. Lett.; 2020; 11, pp. 225-234. [DOI: https://dx.doi.org/10.1080/2150704X.2019.1704304]

42. Leta, M.K.; Demissie, T.A.; Tränckner, J. Modeling and Prediction of Land Use Land Cover Change Dynamics Based on Land Change Modeler (LCM) in Nashe Watershed, Upper Blue Nile Basin, Ethiopia. Sustainability; 2021; 13, 3740. [DOI: https://dx.doi.org/10.3390/su13073740]

43. Mihailescu, D.; Cîmpeanu, S. Multi-Temporal Analysis of Land Cover Changes in Oltenia Plain, Using Terrset Land Change Modeler. AgroLife Sci. J.; 2019; 8, pp. 82-91.

44. U.S. Census Bureau QuickFacts: United States. Available online: https://www.census.gov/quickfacts/fact/table/US/PST045221? (accessed on 19 September 2022).

45. Wentzel, K. Determination of the Overall Soil Erosion Potential in the Nsikazi District (Mpumalanga Province, South Africa) Using Remote Sensing and GIS. Can. J. Remote Sens.; 2002; 28, pp. 322-327. [DOI: https://dx.doi.org/10.5589/m02-013]

46. Zha, Y.; Gao, J.; Ni, S. Use of Normalized Difference Built-up Index in Automatically Mapping Urban Areas from TM Imagery. Int. J. Remote Sens.; 2003; 24, pp. 583-594. [DOI: https://dx.doi.org/10.1080/01431160304987]

47. Xu, H. Modification of Normalised Difference Water Index (NDWI) to Enhance Open Water Features in Remotely Sensed Imagery. Int. J. Remote Sens.; 2006; 27, pp. 3025-3033. [DOI: https://dx.doi.org/10.1080/01431160600589179]

48. Pettorelli, N.; Vik, J.O.; Mysterud, A.; Gaillard, J.-M.; Tucker, C.J.; Stenseth, N.C. Using the Satellite-Derived NDVI to Assess Ecological Responses to Environmental Change. Trends Ecol. Evol.; 2005; 20, pp. 503-510. [DOI: https://dx.doi.org/10.1016/j.tree.2005.05.011]

49. McFeeters, S.K. The Use of the Normalized Difference Water Index (NDWI) in the Delineation of Open Water Features. Int. J. Remote Sens.; 1996; 17, pp. 1425-1432. [DOI: https://dx.doi.org/10.1080/01431169608948714]

50. Singh, K.V.; Setia, R.; Sahoo, S.; Prasad, A.; Pateriya, B. Evaluation of NDWI and MNDWI for Assessment of Waterlogging by Integrating Digital Elevation Model and Groundwater Level. Geocarto Int.; 2015; 30, pp. 650-661. [DOI: https://dx.doi.org/10.1080/10106049.2014.965757]

51. Rasul, A.; Balzter, H.; Ibrahim, G.R.F.; Hameed, H.M.; Wheeler, J.; Adamu, B.; Ibrahim, S.; Najmaddin, P.M. Applying Built-up and Bare-Soil Indices from Landsat 8 to Cities in Dry Climates. Land; 2018; 7, 81. [DOI: https://dx.doi.org/10.3390/land7030081]

52. Loukika, K.N.; Keesara, V.R.; Sridhar, V. Analysis of Land Use and Land Cover Using Machine Learning Algorithms on Google Earth Engine for Munneru River Basin, India. Sustainability; 2021; 13, 13758. [DOI: https://dx.doi.org/10.3390/su132413758]

53. Yıldırım, S. 6 Must-Know Parameters for Machine Learning Algorithms. Available online: https://towardsdatascience.com/6-must-know-parameters-for-machine-learning-algorithms-ed52964bd7a9 (accessed on 8 December 2022).

54. Ellis, C. Number of Trees in Random Forests. Available online: https://crunchingthedata.com/number-of-trees-in-random-forests/ (accessed on 8 December 2022).

55. TerrSet 2020 Geospatial Monitoring and Modeling Software. Available online: https://clarklabs.org/terrset/ (accessed on 19 September 2022).

56. Dzieszko, P. Land-Cover Modelling Using Corine Land Cover Data and Multi-Layer Perceptron. Quaest. Geogr.; 2014; 33, pp. 5-22. [DOI: https://dx.doi.org/10.2478/quageo-2014-0004]

57. Megahed, Y.; Cabral, P.; Silva, J.; Caetano, M. Land Cover Mapping Analysis and Urban Growth Modelling Using Remote Sensing Techniques in Greater Cairo Region—Egypt. ISPRS Int. J. Geo-Inf.; 2015; 4, pp. 1750-1769. [DOI: https://dx.doi.org/10.3390/ijgi4031750]

58. Adhikari, S.; Southworth, J. Simulating Forest Cover Changes of Bannerghatta National Park Based on a CA-Markov Model: A Remote Sensing Approach. Remote Sens.; 2012; 4, pp. 3215-3243. [DOI: https://dx.doi.org/10.3390/rs4103215]

59. Mubea, K.W.; Ngigi, T.G.; Mundia, C.N. Assessing Application of Markov Chain Analysis In Predicting Land Cover Change: A Case Study Of Nakuru Municipality. J. Agric. Sci. Technol.; 2010; 12, pp. 126-144.

60. Guan, D.; He, X.; He, C.; Cheng, L.; Qu, S. Does the Urban Sprawl Matter in Yangtze River Economic Belt, China? An Integrated Analysis with Urban Sprawl Index and One Scenario Analysis Model. Cities; 2020; 99, 102611. [DOI: https://dx.doi.org/10.1016/j.cities.2020.102611]

61. Araya, Y.H.; Cabral, P. Analysis and Modeling of Urban Land Cover Change in Setúbal and Sesimbra, Portugal. Remote Sens.; 2010; 2, pp. 1549-1563. [DOI: https://dx.doi.org/10.3390/rs2061549]

62. Colby, C.C. Centrifugal and Centripetal Forces in Urban Geography. Ann. Assoc. Am. Geogr.; 1933; 23, pp. 1-20. [DOI: https://dx.doi.org/10.1080/00045603309357110]

63. Furundzic, D.; Furundzić, B.S. Infrastructure Corridor as Linear City. Proceedings of the 1st International Conference on Architecture & Urban Design; Tirana, Albania, 19–21 April 2012.

64. Rodrigue, J.-P. The Geography of Transport Systems; 3rd ed. Routledge: London, UK, 2013; ISBN 978-0-203-37118-3

65. Jensen, J.R. Remote Sensing of the Environment: An Earth Resource Perspective; 2nd ed. Prentice Hall Series in Geographic Information Science; Pearson Prentice Hall: Upper Saddle River, NJ, USA, 2007; ISBN 978-0-13-188950-7

66. Anderson, J.W.; Hardy, E.E.; Roach, J.; Witmer, R.E. A Land Use and Land Cover Classification System for Use with Remote Sensor Data; Professional Paper; US Government Printing Office: Washington, DC, USA, 2006.

67. IPCC, C.C. Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change; Core Writing Team Pachauri, R.K.; Meyer, L.A. IPCC: Geneva, Switzerland, 2014; 151.