Land Use and Land Cover Simulation

Abstract

It is important to consider the dynamics of Land Use and Land Cover (LULC) patterns and how they affect society and the environment. Traditional LULC change evaluation methods can be inaccurate and require a lot of manual labor. However, AI techniques like deep learning and machine learning have shown a lot of promise for improving and automating LULC simulation processes. In this chapter, the importance of applying AI to LULC simulation is emphasized, along with the precision, efficacy, and scalability it offers. The importance of AI-based LULC simulation for making well-informed decisions in a variety of fields, including urban planning, agriculture, and natural resource management, is highlighted in the abstract’s conclusion. AI-based methods have shown great promise in LULC analysis, producing accurate classification results and paving the way for subsequent simulations. These findings demonstrate the importance. These results highlight the significance of utilizing AI approaches to assist sustainable development, solve environmental issues, and influence land management choices.

1. Introduction

Cities are currently home to nearly half of the world’s population, and over the next 30 years, most of the two-billion-plus person increase in global population is expected to occur in urban areas in the developing world. This represents a significant departure from the spatial distribution of population growth in the developing world over the past 30 years, which was much more evenly divided between urban and rural areas. The level of world urbanization today and the number and size of the world’s largest cities are unprecedented. At the beginning of the twentieth century, just 16 cities in the world, the vast majority in advanced industrial countries-contained a million people or more. Today, almost 400 cities contain a million people or more, and about seventy percent of them are found in the developing world.

Although the terms “Land Use” and “Land Cover” are frequently used interchangeably, each term has a distinct definition. The term “Land Cover” describes the material that covers the surface of the ground, such as vegetation, urban infrastructure, water, bare soil, and so on. Land-cover identification creates the baseline data for tasks like thematic mapping and change detection analyses [1]. The term “Land Use” describes a piece of land’s function, such as agriculture, wildlife habitat, parks, industry, or recreation. When the terms “Land Use” and “Land Cover” are used together, they typically refer to the grouping or classification of human activities and natural elements on the landscape over a specific period of time using recognized scientific and statistical methods of analysis of pertinent source materials.

Land Use and Land Cover patterns must be assessed and monitored to ensure/protect natural resource management, environmental sustainability, habitat conservation, and the development of sustainable living standards for human beings. Therefore, land-cover archives must be monitored, mapped, analyzed, and updated regularly. The rate of urbanization is greater than or equal to population growth rates. Several studies emphasize large metropolitan cities, where population growth is closely related to urban land expansion based on statistical records. However, the spatial distribution, trends, patterns, and degree of urban land transition were not included in these statistics. LULC research has been a well-established methodology for computing developments on the surface of the earth, such as mapping, monitoring, modeling, and forecasting urban growth, utilizing low- to very high-resolution satellite images.

1.1 What is LULC

LULC stands for Land Use and Land Cover, which is a commonly used application in remote sensing. Remote sensing is the science of obtaining, processing, interpreting, and storing images from the ground, airborne, or space devices for many years. LULC modeling involves identifying and classifying different types of Land Uses and Land Covers based on remotely sensed images. The use of remote sensing in LULC modeling is increasing due to the availability of new sensors and multi-sensor approaches that provide insight into changes in Land Use and Land Cover.

It is a term used in the field of remote sensing and GIS (Geographic Information Systems) to describe the categorization and classification of different types of Land Use and Land Cover features on the Earth’s surface [2]. Land Use refers to the human activities and purposes for which land is utilized, such as residential areas, agricultural fields, industrial zones, or commercial developments. Land Cover, on the other hand, refers to the physical and biological cover of the Earth’s surface, including forests, water bodies, grasslands, urban areas, and barren lands.

LULC mapping involves the analysis and interpretation of satellite imagery or aerial photographs to identify and classify these different Land Use and land-cover categories [3]. It provides valuable information for various applications, including urban planning, environmental monitoring, natural resource management, and land-use change analysis. By understanding and mapping LULC patterns, researchers and decision-makers can assess the impacts of human activities on the environment, monitor changes over time, and develop effective strategies for sustainable land management and resource allocation.

1.2 Why do we need LULC

LULC stands for Land Use and Land Cover, which is a commonly used application in remote sensing. Land Use refers to human activity on a piece of land, while Land Cover refers to the type of coverage over the land surface, whether it is driven by human or natural forces. LULC changes are increasingly affecting human needs, and there is a growing need to map and understand these changes [4, 5]. Mapping Land Use/Land Cover (LULC) changes at regional scales are essential for a wide range of applications, including landslide, erosion, and land planning, globally. LULC modeling involves the identification and classification of different types of Land Uses and Land Covers based on remotely sensed images. The use of remote sensing in LULC modeling is increasing due to the availability of new sensors and multi-sensor approaches that provide insight into changes in Land Use and Land Cover.

A society’s social and economic development is completely correlated with its rate of expansion. This is the main justification behind socioeconomic surveys. Datasets from both spatial and nonspatial sources are used in this kind of survey. The planning, management, and monitoring of programs at the local, regional, and national levels heavily rely on LULC maps. On the one hand, this type of information aids in a better understanding of land utilization issues, and on the other, it is crucial in the establishment of the policies and programs needed for development planning. Monitoring the ongoing pattern of Land Use/Land Cover through time is essential for ensuring sustainable development. Authorities involved in urban development must create these planning models so that every available piece of land can be used in the most logical and effective way in order to ensure sustainable urban development and prevent the haphazard growth of towns and cities [6]. Information about the area’s past and present land usage and Land Cover is necessary for this. We may study the changes in our ecology and surroundings with the aid of LULC maps.

We can create policies and start programs to safeguard our environment if we have inch-by-inch data on the study unit’s Land Use/Land Cover.

1.3 Detection of Land Use and Land Cover change

“The process of detecting differences of state of an object or phenomenon between two different dates of the same geographical region” is the definition of LULC change. It is broken down into three sections: (a) preprocessing; (b) suitable algorithm selection for change detection; and (c) accuracy evaluation. The changes, according to remote sensing terminology, are brought on by soil moisture, climatic conditions, and spectral, geographical, thematic, and temporal restrictions. There are typically two methods for change detection: (1) post-classification comparison and (2) change detection independent of categorization. Without any prior knowledge, change detection without categorization employs the difference function map to find differences in multi-temporal remote-sensing pictures. These include change vector analysis (CVA), band stacking of different times, time series modeling (trajectory-based), and so on. In the post-classification comparison, multi-temporal images are compared for change analysis. These include composite analysis, image rationing, image regression, and aerial difference computation. Post-classification is the method that is most frequently employed. Although being sensitive to the prior categorization accuracy, it is very simple to use, provides “from-to change information,” and decreases environmental differences and sensor influence. Each of these methods is used in turn to the monitoring of various LULC changes. As an illustration, consider vegetation, deforestation, monitoring of disasters, landscape fragmentation, ecological fluctuations, climate change, and urbanization. They function as a foundation for understanding the dynamics and connections between human actions and natural occurrences.

1.4 Land Use and Land Cover (LULC) classification

Classifying remote-sensing images is one of the most crucial methods for mapping LULC. Numerous categorization systems have been created and used in a variety of contexts, including environmental change research, LULC, land resource management, spatial-temporal modeling, and change detection, to name just a few: (1) Parametric (such as the maximum likelihood classifier (MLC), minimal distance to means and the box classifier, and K-Means) and nonparametric (such as the supervised classifier, decision tree, and artificial neural network (ANN)) approaches, as well as ISODATA methods; (2) pixel- and entity-oriented; (3) automatic and semi-automatic, and (4) spectral indices. The following table describes the different classification techniques with each of the pros and cons (Table 1).

Decision Tree (DT)	Nonparametric technique Makes no conclusions on how data can be distributed. For change and no-change classes can give a set of rules Disadvantages: (1) Sensitive to both quality and quantity of training; (2) Data can be over-trained; (3) Does not aim for an optimum match; (4) May increase much larger in size, making it harder to analyze
Support Vector Machine (SVM)	Nonparametric technique Capable of managing small training data sets. As compared to traditional methods produces high classification accuracy. A theoretically bigger dataset can be dealt with higher dimensionality. Outperformed RF and ANN. In terms of average precision and robustness, the use of radial basis functions or polynomial kernels outperformed ANN and RF. Computational involvedness is higher compared to traditional supervised methods (such as MLC) Relatively high classification accuracy Disadvantages: (1) Trouble in choosing the superlative kernel function; (2) With the data dimensionality, during the learning process, the computation time for classification and optimization increases polynomials.
Artificial neural network (ANN)	Does not make assumptions about the nature of data distribution, a non –parametric technique Estimates data properties based on training data. As compared to conventional supervised approaches, the computational complexity is greater (such as MLC) Disadvantages: (1) Complex architecture optimization; (2) Low computational robustness; (3) Exceeds the good mean classification accuracy due to tremendous training time; (4) The hidden layer is not well-known; (5) For network teaching, the amount of training data is essential
Random Forest (RF)	Bagging and random algorithm based on a decision tree Based on two parameters: (1) the number of trees, described by ‘n-tree’ and (2) in each break numerous features described by “m-try” Much more computationally expensive than SVM classifiers Classification trees give an individual choice of vote. Stable and robust Disadvantages: (1) Complexity; (2) Much laborious and time-consuming to construct several trees; (3) Overfitting; (4) No interpretability
Maximum likelihood classifier (MLC)	Parametric approach Involves assuming the selected signature groups in a normal distribution. Less computational complexity Exhibited inferior accuracies and higher variability. Disadvantage: Requires a large training area and is unable to resolve the interclass confusion

Table 1.

Summary of the different classification techniques.

1.5 Stages in the process of image classification

As shown, the method of classifying Land Cover from satellite pictures involves several stages, including data acquisition, information data preprocessing, feature extraction, training data selection, classification, and post-processing.

1.5.1 Data acquisition

Remote-sensing data acquisition involves collecting information about the Earth’s surface without physically being there. This is usually done by using instruments such as satellites, airplanes, drones, or ground-based sensors to capture images, spectra, and other data. The type of instrument used depends on the spatial and spectral resolution required for the specific application. For example, satellites are used for large-scale mapping, while drones are used for high-resolution imaging of specific areas. Data acquisition can involve active or passive sensing, with active sensors emitting energy and measuring its reflection, while passive sensors measure the energy emitted by the Earth’s surface. Specifically for remote sensing image classification, data acquisition involves obtaining remotely sensed images, which can be done using satellites, drones, or other aerial platforms. Pre-processing involves preparing the images for classification by removing noise, correcting distortions, and enhancing features.

1.5.2 Preprocessing

Satellite image preprocessing is a crucial step in the process of land-use and land-cover classification. The preprocessing stage involves a series of operations that aim to enhance the quality of the satellite data and prepare it for image classification. One of the most important preprocessing steps is image correction, which involves removing distortions and errors caused by the satellite sensor, atmospheric conditions, and other factors. This is done through techniques such as radiometric and geometric correction.

Another important preprocessing step is image enhancement, which aims to improve the visual quality and interpretability of the image. This can be achieved through techniques such as contrast adjustment, histogram equalization, and filtering. Other preprocessing steps include data calibration, image registration, and image fusion. The goal of all these steps is to ensure that the satellite data is accurate, consistent, and ready for image classification, which is the next stage in the LULC classification process.

1.5.3 Image classification

Image classification of satellite data is a process used in remote sensing to interpret images captured by satellites orbiting the Earth. This process involves the identification and labeling of different features or objects in the image based on their spectral characteristics, shape, and texture. The classification of satellite data is crucial in understanding the Earth’s surface and its changes over time, and it is used in various applications such as land-use and land-cover mapping, urban planning, agriculture, and environmental monitoring. The accuracy of image classification depends on several factors, including the quality of the satellite data, the choice of classification algorithm, and the expertise of the analyst performing the classification. Image classification involves the actual identification and mapping of land-use and land-cover types using various algorithms and techniques. This can be done using supervised or unsupervised classification methods.

1.5.4 Accuracy assessment

Satellite image accuracy assessment is a crucial step in the process of land-use and land-cover classification. It involves comparing the results of image classification with a reference dataset, such as ground truth data, to evaluate the accuracy of the classification. The accuracy assessment can be performed using various statistical measures, such as overall accuracy, user’s accuracy, producer’s accuracy, and kappa coefficient. These measures provide information on the level of agreement between the classified image and the reference dataset. The results of the accuracy assessment can be used to improve the classification accuracy and to assess the reliability of the classification results for different applications.

Finally, accuracy assessment involves evaluating the accuracy of the classification results by comparing them to ground truth data or other reference data. This helps to identify any errors or uncertainties in the classification process and improve the overall accuracy of the results.

2. Latest trends and innovations in future simulation modeling

In recent decades, modeling of Land Use has been of increasing importance as urbanization, and LULC changes have raised concerns among decision-makers and planners about the ecosystem’s and natural resources’ future consequences. The increasing trend of urbanization depends on various factors, such as socioeconomic development, demography, environment, geography, and culture. This specifies the increased importance of urban areas as the focal point of the populace and commercial concentration within a particular society.

The modeling of such dynamic systems is not an easy job. Many models have been created to date, and they are divided into categories: (1) Logistic regression and Markov chain models (mathematical equation based) (2) Statically, (3) System dynamic, (4) Expert system, (5) Genetic, (6) Cellular model, (7) Hybrid or agent-based, and (8) Embedded models are all examples. The most commonly used models are (1) LCM, (2) CA, (3) Markov chain, (4) CA-Markov, (5) GEOMOD, (5) conversion of Land Use and its effects (CLUE). However, it is difficult to compare which model gives a more accurate demonstration. According to the latest research and studies, models are particularly helpful in determining the effects of urban sprawl, designing and managing LULC, and determining the correct land-use transition patterns and trends.

2.1 CLUE and GEOMOD

Veldkamp and L. O. Fresco devised the CLUE dynamic model. The study of location logistic regression is a prerequisite for this model. This model’s main objective is to investigate LULC changes by including suitable driving elements including biophysical and human factors. There are three sections to this model. Regional land-use purpose, regional biophysical component, and municipal land-use distribution are the first three components. Using Costa Rica as a case study, they utilized this model to replicate changes in LULC at both the local and global levels and to illustrate how biophysical and demographic variables have impacted LULC. Land-use conversion only takes place in situations where the new Land Use boosts yield or value due to population increase and numerous biophysical factors (including food, technology, and socioeconomic conditions). This methodology’s drawback is that it requires the use of a different statistical model to predict the potential LULC.

The authors of the GEOMOD simulation model are Pontius, Cornell, and Hall. Three primary decision-making criteria are included in this model: nearest neighbors, political areas, and biophysical patterns.

2.2 Markov chain model

Markov model is a well-known and reliable model for tracking and ecological modeling, and simulating changes, trends, and predicting the future scenario at a diverse spatial scale. From one time cycle (t = 1) to the next (t + 1), this model forecasts potential LULC changes based on each LULC class’s transition potential matrix. The changes are considered a stochastic process in this model. The transition matrix, on the other hand, is critical for future LULC simulations. The major drawback of this model is that it is unable to give the spatial distribution of LULC change activities.

Using the below-mentioned equation, future simulation of LULC changes can be calculated:

$$\mathit{S}\left(\mathit{t}+1\right)={\mathit{P}}_{\mathit{ij}}\ast \mathit{S}\left(\mathit{t}\right)$$E1

$${\mathit{P}}_{\mathit{ij}}=\left(\begin{array}{ccc}{\mathit{P}}_{11}& {\mathit{P}}_{12}& {\mathit{P}}_{1\mathit{n}}\\ {\mathit{P}}_{21}& {\mathit{P}}_{22}& {\mathit{P}}_{2\mathit{n}}\\ {\mathit{P}}_{\mathit{n}1}& {\mathit{P}}_{\mathit{n}2}& {\mathit{P}}_{\mathit{nn}}\end{array}\right)$$E2

$$0\le {\mathit{P}}_{\mathit{ij}}<1\mathit{\text{and}}\sum _{\mathit{j}=1}^{\mathit{n}}{\mathit{P}}_{\mathit{ij}}=1,\left(\mathit{i},\mathit{j}=1,2,\dots .\mathit{n}\right)$$E3

Where S(t) denotes the state of the system at time t; S(t + 1) denotes the state of the system at a time (t + 1); and P ij denotes the transition probability matrix from current state i to next time state j. Several research studies have been conducted to assess the future LULC scenario using the Markov chain model.

2.3 Cellular automata (CA)

CA was developed by John Von Neumann and Stanislaw Ulam to ascertain the logical underpinnings of existence. Tobler utilized the CA model for spatial modeling for the first time in 1970. This first theoretical CA technique served as the foundation for subsequent models that initially surfaced in the 1980s and had the aim of simulating and forecasting urban expansion and LULC changes [7]. The LULC and urban dynamic frameworks in the 1990s used growth and advancement in the CA model (i.e., the capacity to figure things out added to the model). CA is a dynamic technique that can describe and manage intricate and nonlinear spatial trends. It offers a clear understanding of how LULC shifts from social behavior to universal patterns. The neighborhood type, Neighborhood size, Cell size, and Transition rules make up the majority of the CA model. These settings produce the best simulation outcomes. The transition rule, which controls the model and is dependent on the training data, is the most crucial parameter in this model. The condition of each subsequent step that depends on the current state of that cell and its surrounding neighborhood cells is referred to as the transition rule. The transition rule shows how the geographical and temporal complexity of Land Cover develops over time [8]. The ability of this model to simulate LULC and urban growth has grown. It is crucial to note that in this paradigm, space and time are separate entities. However, space is measured as a regular grid in two dimensions. This model’s crucial characteristics are that they illustrate the system’s complicated spatial dynamics. The CA model may be expressed using the equation below:

$$\mathit{S}\left(\mathit{t},\mathit{t}+1\right)=\mathit{f}\left(\mathit{S}\left(\mathit{t}\right),\mathit{N}\right)$$E4

Where N stands for the probability of the system’s state at any time and S (t + 1) stands for the system’s state at a time (t + 1). The CA model has been used in several research to evaluate changes in LULC and urbanization. These experiments show that this model is capable of accurately explaining and simulating the intricate process of LULC evolution, urban systems, and patterns [2]. This model has a key flaw in that it is unable to account for the macro-scale driving forces, such as social, economic, and cultural elements, which are in charge of LULC changes and urban sprawl. For simulation and prediction, some of this research do, however, depend on quantitative techniques like logistic regression (LR), the SLEUTH model, multi-criteria evaluation (MCE), and neural networks.

The potential of CA and Markov models to provide a computational method to give support to ecology and environmental planning, decision-making in urban, and the suitability assessment of land for construction, has been demonstrated. This is critical for the effective operation of major metropolitan areas. Many experiments, however, demonstrate the disadvantages of a single model. For LULC future prediction; integrated modeling methods are commonly used to solve the shortcomings of individual.

2.4 CA-Markov model

The CA-Markov model can address the drawbacks of a single mode by combining intricate simulation models with factual and observational models. The integrated model enhances LULC modeling, gives a better understanding, and supplements one another. In line with this, the LULC prediction model’s precision varies since each research location has a distinct collection of climatic and terrain factors.

A useful tool for evaluating and modeling LULC based on existing spatial and temporal patterns is the CA-Markov model. This model may be used to simulate and recreate spatiotemporal LULC data. Markov chain and (2) are the two halves of this model. In order to provide the right LULC planning, this model may decode the Markov chain model outputs through a CA model as a spatial distribution output. The Markov model regulates the temporal transition between LULC categories based on transition matrices [7]. While taking into consideration neighborhood structure and prospective transition maps, the CA model, on the other hand, employs local rules to address changes in spatial configurations.

One of the models that is most frequently used to calculate probable LULC changes at time “t + 1” using the progression from time “t-1” to time “t” is this one. The likelihood of a change in Land Cover from one class to another during two time periods is what determines it. The probability matrices are built using historical LULC data and anticipate future change. Numerous changes in Land Use category can be simulated, consequently, offering the ability to encourage the transition between LULC classes. Many research has recently attempted to employ the CA-Markov model to include socioeconomic and ecological variables into land-use models [9].

This will help cities better recognize and resolve a complex land-use structure and create better land-use management strategies that would better integrate urban growth and environmental protection. Compared to other models, which are also used for a similar task, this model poses benefits and drawbacks. The advantages of this model are (1) High proficiency, (2) Easy calibration, and (3) Strong capability of simulating a wide range of ground covers and dynamic patterns as compared to other models (e.g., GEOMOD and CLUE). The model’s main flaws are (1) as seen in agent-based models, its failure to incorporate human, social, and economic complexities in the simulation and (2) fails to identify the new developments occurring in the studied. Therefore, to address these limitations, this model needs to be combined with other models.

For potential land-cover estimation, Soe and Le used the multi-criteria (MCE) technique in CA-Markov. The development of criteria was based on the weight assignment to the drivers of LULC changes. The more relevant the driver, the higher the weight is assigned. A criterion was divided into two categories: (1) variables and (2) restrictions. Three criteria determine it: (1) Prioritization decomposition, (2) comparative decision, and (3) Prioritization synthesis.

2.5 Land change modeler (LCM)

LCM is an incorporated model designed by Clark Labs in partnership with Conservation International to monitor and forecast LULC changes and to monitor and predict biodiversity impacts. The three parts of the LCM process—transition potential simulation, change prediction, and interpretation—are all included in the IDRISI Terrset 18.1 program [6, 10]. This model uses two thematic raster images with the same number and order of LULC groups as data. LCM evaluates LULC changes of two different periods, and provides a comparative evaluation of improvements in various LULC groups in terms of gains, losses, swaps, net changes, and overall changes, with the outcomes shown in various graphs and maps.

LCM breaks down the LULC changes for different classes, calculates and assesses their trends and patterns, and then projects these changes to predict the future LULC. Given the past transition trend, the model envisions the land-use pattern. The explanatory variables, such as distance to roads, slope, aspect, and other variables, were added to the model as rater datasets. Using Cramer’s V, the influencing variables were chosen based on their availability, relative significance, and subsequent influence on LULC changes (a quantitative measure that shows the relationship between explanatory variables and land-cover classes).

The transitions are broken down into sub-models when the basic driving is taken to be the same as the ground cover transformations. For instance, the same variables that drive a change from vegetation to an urban area also cause a change from a forest to an urban region. Using MLP neural networks, the transition potential map of each sub-model is created, and explanatory variables are then allocated to each sub-model based on Cramer’s V values. MLP excels at multi-transition modeling, nonlinear processing connections among variables, and converting categorical data to continuous data. The time-specific potential for change is interpreted by the transition potential maps that were developed. Future LULC maps are projected by LCM using these prospective maps and Markov chain analysis [11].

In comparison with quickly changing Land Cover, LCM produced superior forecast accuracy over a short period of time, especially in the stable Land Cover. Compared to previous models that predict LULC changes using supervised methods, such the weighted model system, this model is more reliable. LCM developed more accurate change potential maps in which the user picks and modifies the weights. This is due to the fact that neural network outputs show changes in different LULC categories more effectively than individual probabilities obtained by weights of evidence. Furthermore, according to researchers that analyzed several approaches—including logistic regression, Bayesian analysis, weights of evidence, and a neural network—they came to the conclusion that neural networks produced predictions that were more accurate than those made by the other methods.

LCM has been utilized in several studies to forecast future LULC patterns, tropical expansion, urbanization, erosive processes, Mediterranean catchment, and habitat modeling. LCM is effective in simulating future LULC changes, patterns, and trends, according to regional case studies. This is because LCM helps local administrative authorities make decisions and gives a better grasp of LULC. All of these research findings demonstrated that LCM can provide extremely accurate LULC simulations.

2.6 Multilayer perceptron (MLP)

MLP is a feed-forward NN (Neural Network) that predicts outcomes employing the Back propagation (BP) algorithm. The core of neural net training is back-propagation (BP). It is a technique for improveming the accuracy of a neural network neurons weights using the error rate from the iteration. By fine-tuning the weights, it can reduce error rates and improve the model’s generalization, making it more accurate. The term “backward propagation of errors” is abbreviated as “Back Propagation.” It is a popular way to train artificial neural networks. This approach is useful for calculating the gradient of a loss function with respect to all of the network’s weights. It is a nonparametric algorithm that does not take multi-co-linearity into account. MLP is made up of three layers: (1) Input data, (2) secret (computing node sets), and (3) output. It signifies relationships between transitions of Land Use and their explanatory variables through a network of weighted relationships modified iteratively by the algorithm. In one direction, data flows from an input layer to an output layer via hidden layers, resulting in nonlinear relationships. LCM produces two kinds of predictions: soft and hard predictions. Centered on the multi-objective land allocation (MOLA) module, a hard prediction yields a forecast plot. This module assigns each pixel to one of the ground cover groups with a higher chance of being. A soft prediction represents a continuous mapping of susceptibility to change; it indicates the probability of a particular class pixel being changed to a specific pixel of Land Cover, over a T2–T1 and n-year cycle. How much land is assigned to a thematic class is determined by the transition probability matrix obtained by the Markov chain.

2.7 Artificial neural networks (ANN)

ANN is an enormously analogous distribution processor that possesses a natural storing investigational knowledge property available for use in two ways:

• Knowledge through a learning process.

• Interconnection strengths of a neuron.

A nonlinear relationship exhibited between the LULC and driving factors such as biophysical, socioeconomic, and topography such as nonlinear relationships is easily handled by ANN.

Artificial Neural Networks (ANN) are used to model complex systems such as urban sprawl, not assume the data, do not depend on exacting functional relationships, and find their way in biochemical modeling and cybernetics. ANN also reveals the relationships between future simulated urban growth probability and site attributes due to the system’s nonlinear complex behavior; therefore, it perfectly fits into the regression-type model category to forecast the change in the urban Land Cover. The ANN regulates those areas that are probably going to be changed in the future, but it could not determine how much change will occur.

3. Land Use and Land Cover change modeling and future scenarios

Human activities, climate change, and natural disturbances are just a few of the many elements that can have an impact on the complicated process of land-use and land-cover change. It is essential for efficient land management and decision-making procedures to accurately simulate these changes. Remote sensing, which involves utilizing satellite or aerial photography to identify and map Land Cover, is one method of simulating changes in Land Use and Land Cover. The use of agent-based modeling, which mimics the activities and interactions of distinct agents (such as farmers or land developers) within a broader system, is another strategy [12]. It is also possible to model changes in Land Use and Land Cover using artificial neural networks and evolutionary algorithms. In these methods, patterns and correlations within the data are found using machine learning algorithms, and the linkages are then used to anticipate future changes.

It is crucial to remember that thorough validation and accuracy evaluation are necessary for accurate modeling of changes in Land Use and Land Cover. These simulations may be made more accurate by including expert comments and employing cutting-edge methods like machine learning algorithms, but continual validation is required to make sure the models are properly reflecting changes in the actual world (Figure 1).

Figure 1.

Flow chart describing the process of LULC simulation by using earlier and later images.

3.1 Change analysis

The change between two separate time periods, time 1 and time 2, was calculated in the change analysis panel. By graphing gains and losses across various land-cover categories, the change analysis offers a fast assessment of quantitative change. Additionally, it assesses the information on Land Cover in both map and geographical representations in terms of net change, persistence, and the precise transition. These modifications are critical in locating the dominating transition from one class to another, which is subsequently categorized and targeted. The best-fit polynomial trend surface follows the pattern of change, and the spatial trend of change presents the trend as a map.

3.2 Transition potential modeling and driving forces’ determination

The area of change is determined by the transition potential. If it is assumed that the underlying sources of change are the same for each transition, land-cover changes may be categorized into sub-models. For instance, the same factors that affect the transition from forest to built-up land may also affect the transformation from agricultural to built-up land. As a result, sub-models were created for land-use and land-cover changes that shared a common set of driving factors. In order to estimate the relative frequency of various land-use and land-cover types that had taken place in the transitional areas, evidence probability was also chosen.

3.3 Selection of explanatory variables

The selection of driving factors is the most important step in modeling growth modeling that is associated with the LULC change [3, 13]. The model generates results that are close to reality. Several different factors responsible for LULC change have been identified and used by the researchers to predict the change, but still, no universal set has been formulated because each area is different from the other, and the factors responsible for change are not always common. Land transitions also vary from case to case, and the degree to which the driving factors contribute to the transitions also differs.

The urban growth prediction modeling involves computation of Cramer’s V that examines the best driving variable for modeling urban growth and measures the association strength between each pair of variables used in modeling.

$$V=\sqrt{\frac{{\phi }^2}{min\left(k-1,r-1\right)}}$$E5

Where, φ² = Mean square contingency coefficient. k = Number of columns. r = Number of rows.

For the calculations, Cramer’s V value falls between 0 and 1. V value ≥0.15 shows the influence of that particular factor, whereas >0.4 shows a strong association between land class and factors. For modeling purposes, the driving factors need to be declared as either static or dynamic. The factors that do not change over time are the static ones, whereas dynamic factors are time-dependent and they change with time. Dynamic variables are handled differently in the simulation process and are calculated temporally throughout prediction Multilayer Perceptron (MLP).

3.4 Change prediction

Future simulation model validation is the process of assessing the accuracy and reliability of a model that is used to simulate future scenarios. This is typically done by comparing the model’s predictions with real-world data that is collected after the simulation is complete [14]. The goal of this process is to determine whether the model accurately captures the dynamics of the system being simulated and whether it can be used to make accurate predictions about the future. Validation is an important step in the development of any simulation model, as it ensures that the model is fit for purpose and can be trusted to provide reliable predictions.

3.5 Future scenario

Two different types of forecasts are generated by the Land Change Modeler: (1) hard predictions and (2) soft predictions. A projected map is generated by a hard prediction using a multi-objective land allocation (MOLA) module. Each pixel is given a land-cover class based on the likelihood that it will be present. By creating a vulnerability map and assigning a value between 0 and 1 to each pixel, soft prediction calculates the likelihood that a pixel will transition into a different land category.

3.6 Landscape matrices

Landscape matrices serve as a quantitative correlation between trends in the landscape and biological and environmental processes. They show numerical data regarding landscape structure, configuration, and dimension, allowing for comparisons over time and assisting in developing future scenarios. Landscape metrics are essential tools for defining spatial dynamics and determining how landscape elements are arranged in space and time. Broadly protected reserves, forest dynamics, natural parks, and urban extensions are all studied using them in the literature.

The metric category should represent the pattern diversity seen through the landscape, but its usage must be limited, particularly in indexes closely associated with one another. They can be classified into three categories: (1) Patch, (2) Class, and (3) Landscape. Specific patches, which reflect distinct areas with identical characteristics, are measured at the patch level. To evaluate class-level metrics, all patches of a given class type, in this case, LULC classes, are used.

In other words, it depicts the spatial distribution and pattern of a single patch from within a landscape. In simple terms, it reflects the landscape level’s overall spatial pattern, taking into account all patch types at the same time. Many landscape metrics, such as the Euclidean Nearest Neighbor Distance (ENN), Largest Patch Index (LPI), and Percentage of Landscape (PLAND), can be used at the patch, class, and landscape levels [8, 15]. These metrics can be calculated at any of the three levels listed above. Complete definitions of these metrics, as well as the calculations used to calculate them, can be found in. Landscape measurements are widely used as indicators of LULC transition, ecology, water quality, and ecological health, and they are still evolving. They have a suite of spatial methods for studying whole environments and the arrangement and properties of the elements that make up such landscapes. These metrics will show you how fractured landscapes are and how patches are created. They also have numerical values that can be used to describe certain classes in general, unlike segmentation. Landscape matrices may provide additional knowledge to better understand the LULC transition because of these properties (Table 2).

Category	Metric Name	Acronym Unit	Level Used	Description	Range
Patch Size and Density	Patch Density (PD)	Number of patches per 100 ha	CL	The number of patches per unit area	PD ≥1, no limit
	Percentage of Landscape (PLAND)	%	CL	The aggregated area of landscape.	0–100
	Mean Patch Area (MPA)	ha	CL	An average patch size in each class	MPA > 0, no limit
Shape and Edge	Edge Density (ED)	m/ha	CL	Calculate the total lengths of all edge segments of corresponding patch type per unit area. Edge density explained the complexity of each patch shape.	ED ≥1, no limit
	Largest Patch Index (LPI)	%	CL	Ratio between the largest patch of the corresponding patch type and the total landscape area.	0 < LPI ≤ 100
	Area Weighted Mean Fractal Dimension Index (AWMPFD)	None	LL	Measure the average fractal dimensions of patches of a particulate patch type divided by the total sum of the patch area.	1 ≤ AWMPFD ≤2
Proximity	Mean Euclidean Nearest Neighbor Distance (ENN_MN)	m	CL	Measure the minimum edge to edge distance to the nearest neighbor same patch type. It explains isolation of corresponding patch type or landscape.	ENN_MN > 0
Diversity and Texture	Contagion (CONTAG)	%	LL	Measure the total probability that a patch of cells neighboring the same type of cells.	0 < CONTAG ≤100
	Shannon’s Diversity Index	None	LL	Indicate diversity in a landscape from the abundance of patch types. It increases as the number of different patch types increases or the distribution of area/land among patch types/classes becomes more equitable.	Shannon’s Entropy ≥0, no limit

Table 2.

The description of landscape metrics used for morphological analysis (where, CL = class level, LL = landscape level).

3.7 Model validation

Future simulation model validation is the process of assessing the accuracy and reliability of a model that is used to simulate future scenarios. This is typically done by comparing the model’s predictions with real-world data that is collected after the simulation is complete [16]. The goal of this process is to determine whether the model accurately captures the dynamics of the system being simulated and whether it can be used to make accurate predictions about the future. Validation is an important step in the development of any simulation model, as it ensures that the model is fit for purpose and can be trusted to provide reliable predictions.

4. Case study

4.1 Study area

The geographic coordinates of Islamabad, the capital of Pakistan, are 33° 28′ latitude and 73° 22′ longitude. According to the ICT master plan, it is organized into five zones (Figure 2). Zones I and II are separated into sections that are each around 2 km by 2 km long. Their names are assigned in the following ways: numerically from 1 to 18 from east to west, and alphabetically from A to I from north to south. Urban areas are divided into multiple equal-sized sectors by the horizontal and vertical interlacing of primary and minor highways. Included are residences, businesses, industrial centers, public and private services, foreign embassies, educational institutions, and so on. The majority of Zone-III is made up of mountains, forests, and natural landscapes.

Figure 2.

Case study area Islamabad.

4.2 Image classification

For this case study Landsat Images is being used and four major classes crafted details in as under (Figure 3 and Table 3).

Figure 3.

Land Use Land Cover image Islamabad.

LULC Classes	Description
Bare land	Areas with no prevailing vegetation cover on at slightest 90% of the area
Built-up	Residential, industrial, and utility areas, as well as recreational areas, public installations, and utilities, are all accessible. Pavement-based roads are also included in this class.
Vegetation	Reserved woodland and fields heavily populated with trees, as well as areas where grasses dominate the landscape. Since it was difficult to distinguish between fertile fields and sparsely located villages and roads made from earthwork, they were included here.
Water	Water sources such as the Indus River and streams, as well as lakes and main rivers.

Table 3.

LULC classes description.

4.3 Major steps to conclude future simulation

The major steps in conducting a Land Use and Land Cover (LULC) future simulation using AI are as follows:

1. Data Acquisition: Gather relevant data sources such as satellite imagery, climate data, and socioeconomic information that include (Remote Sensing, GIS layers, and supporting datasets).

2. Preprocessing: Clean, align, and integrate the acquired data for analysis.

3. Training Data Preparation: Create a labeled training dataset by manually classifying land-cover types in the data.

4. Feature Extraction: Use AI techniques, such as convolutional neural networks, to extract relevant features from the training dataset.

5. Model Training: Train the AI model using the labeled training data to classify land-cover types accurately.

6. Validation and Evaluation: Validate the trained model using independent datasets to assess its performance.

7. Future Projection and Simulation: Use the trained model to simulate future LULC scenarios based on anticipated drivers.

8. Uncertainty Analysis: Assess and quantify uncertainties associated with the future projections.

9. Scenario Analysis and Decision Support: Analyze the simulated scenarios to gain insights and support decision-making.

10. Documentation and Reporting: Document the methodology, findings, and limitations of the simulation study for communication purposes (Figure 4).

Figure 4.

Overall workflow for LULC prediction modeling.

5. Why future simulation of (LULC) using remote sensing and GIS is crucial

The future simulation of Land Use and Land Cover (LULC) using remote sensing and GIS is crucial for several reasons:

5.1 Urban planning

LULC future simulation helps urban planners and policymakers to anticipate and manage urban growth and expansion. By simulating different scenarios, they can assess the potential impacts of future land-use changes on infrastructure, transportation, and the environment. This information can aid in the development of sustainable urban development plans and policies.

5.2 Environmental management

LULC simulation allows for the assessment of potential environmental changes and impacts. By projecting future land-cover changes, such as deforestation or expansion of agricultural areas, environmental managers can evaluate the consequences on biodiversity, ecosystems, and natural resources. This knowledge is essential for conservation efforts, land-use zoning, and the preservation of sensitive areas.

5.3 Climate change adaptation

With climate change, there are anticipated shifts in precipitation patterns, temperature, and other climatic factors. LULC simulation can help analyze the potential effects of climate change on Land Use and Land Cover. It aids in understanding how changes in temperature and precipitation may alter agricultural patterns, vegetation distribution, and water resources, enabling policymakers to plan for adaptation strategies.

5.4 Disaster risk assessment

Future simulation of LULC can assist in assessing vulnerability and risk to natural disasters. By analyzing land-cover changes and their potential impacts, emergency management agencies can identify areas at high risk of flooding, landslides, or other hazards. This knowledge supports the development of early warning systems, land-use regulations, and disaster preparedness plans.

5.5 Policy and decision-making

LULC simulation provides valuable information for policymakers and decision-makers. It enables them to evaluate the effectiveness of current land-use policies, anticipate future challenges, and design appropriate strategies. By incorporating spatially explicit future scenarios, decision-makers can make informed choices to promote sustainable development, resource management, and socioeconomic growth.

6. Factor to consider for LULC simulation

Choosing data for LULC (Land Use and Land Cover) simulation involves considering several factors to ensure the accuracy and reliability of the simulation results. Here are some steps to guide you in selecting appropriate data:

6.1 Define the study area

Determine the geographical extent of your study area. Consider the size, location, and specific characteristics of the area to ensure the data you choose aligns with the study objectives.

Identify Data Requirements: Clearly define the data requirements for your LULC simulation. This may include spatial resolution, temporal coverage, thematic detail, and specific data formats. Consider the scale and level of detail required for your analysis.

6.2 Acquire Base data

Start by obtaining base data that provides foundational information about the study area. This may include digital elevation models (DEM), land-cover maps, topographic maps, satellite imagery, and existing GIS datasets. Sources for these data can include government agencies, research organizations, open data portals, or commercial providers.

6.3 Assess data availability

Evaluate the availability and accessibility of the data sources that meet your requirements. Check if the data is freely available, requires purchase or licensing, or if you need to request it from relevant organizations. Consider the data format, compatibility with your GIS software, and any processing or preprocessing steps required.

6.4 Data compatibility and consistency

Ensure that the data sources you choose are compatible with each other in terms of spatial reference system, coordinate system, and projection. Consistency across different data layers is crucial to maintain accuracy and avoid discrepancies in the simulation.

6.5 Data quality assessment

Assess the quality and reliability of the data sources. Consider factors such as data accuracy, resolution, completeness, and currency. Look for metadata and documentation that describe the data collection methods, validation procedures, and potential limitations.

6.6 Data integration and fusion

Depending on your simulation requirements, you may need to integrate multiple datasets to capture different land-use and land-cover categories accurately. Data fusion techniques, such as merging satellite imagery with ground-based data or combining different data sources, can improve the accuracy and representativeness of your simulation [17].

6.7 Validation and ground Truthing

Plan for validation and ground-truthing activities to verify the accuracy of your selected data. Field surveys, ground-based measurements, or comparison with existing ground truth data can help validate the simulated LULC outputs and refine the accuracy of the model [11, 18].

7. Conclusion

In summary, LULC future simulation using remote sensing and GIS is essential for informed decision-making, environmental management, urban planning, climate change adaptation, and disaster risk assessment. It helps us understand and anticipate the potential impacts of future land-use changes, facilitating sustainable and resilient development. Remember, the choice of data for LULC simulation should align with the specific objectives and requirements of your study. It is crucial to consider data quality, compatibility, availability, and validation to ensure reliable and accurate simulation results.

Acknowledgments

The authors would like to thank USGS Earth Explorer for providing Landsat data. The authors would also like to thank the anonymous reviewers for their insightful comments and substantial help in improving this chapter.

Conflicts of interest

The authors declare no conflict of interest.

Data availability statement

The data presented in this study are available on request from the corresponding author.