Cellular Automata-Based Artificial Neural Network Model for Assessing Past, Present, and Future Land Use/Land Cover Dynamics

Abstract

Land use and land cover change (LULCC) is among the most apparent natural landscape processes impacted by anthropogenic activities, particularly in fast-growing regions. In India, at present, due to the impacts of anthropogenic climate change, supplemented by the fast pace of developmental activities, the areas providing the highest agricultural yields are facing the threat of either extinction or change in land use. This study assesses the LULCC in the fastest-changing landscape region of the Indian state of Bihar, District Muzaffarpur. This district is known for its litchi cultivation, which, over the last few years, has been observed to be increasing in acreage at the behest of a decrease in natural vegetation. In this study, we aim to assess the past, present and future changes in LULC of the Muzaffarpur district using support vector classification and CA-ANN (cellular automata-artificial neural network) algorithms. For assessing the present and past LULC of the study area, we used Landsat Satellite data for 1990, 2000, 2010, and 2020. It was observed that between 1990 and 2020, the area under vegetation, wetlands, water body, and fallow land decreased by 44.28%, 34.82%, 25.56%, and 5.63%, respectively. At the same time, the area under built-up, litchi plantation, and cropland increased by 1451.30%, 181.91%, and 5.66%, respectively. Extensive ground truthing was carried out to assess the accuracy of the LULC for 2020, whereas historical google earth images were used for 1990, 2000, and 2010, through the use of overall accuracy and kappa coefficient indices. The kappa coefficients for the final LULC for the years 1990, 2000, 2010, and 2020 were 0.79, 0.75, 0.87, and 0.85, respectively. For forecasting the future LULC, first, the LULC of 1990 and 2010 were used to predict the landscape for 2020 using the CA-ANN model. After calibrating and validating the CA-ANN outputs, LULC for 2030 and 2050 were generated. The generated future LULC scenarios were validated using kappa index statistics by comparing the forecast outcomes with the original LULC data for 2020. It was observed that in both 2030 and 2050, built-up and vegetation would be the major transitioning LULC. In 2030 and 2050, built-up will increase by 13.15% and 108.69%, respectively, compared to its area in 2020; whereas vegetation is expected to decrease by 14.30% in 2030 and 32.84% in 2050 compared to its area in 2020. Overall, this study depicted a decline in the natural landscape and a sudden increase in the built-up and cash-crop area. If such trends continue, the future scenario of LULC will also demonstrate the same pattern. This study will help formulate better land use management policy in the study area, and the overall state of Bihar, which is considered to be the poorest state of India and the most vulnerable to natural calamities. It also demonstrates the ability of the CA-ANN model to forecast future events and comprehend spatiotemporal LULC dynamics.

1. Introduction

Among the significant elements of global transformation are land use and land cover change (LULCC), which are closely related to human activity. In pursuit of jobs, educational opportunities, and access to healthcare, more than 50% of the world’s inhabitants have moved to towns. Due to the growth of the economy and the construction of new infrastructure, urbanization is proceeding at an astounding rate. Habitat loss, global climate change, and growth in catastrophic events are just a few examples of environmental problems brought on by changes in LULC caused by natural or human action [1,2]. LULCC is occurring unexpectedly and uncontrollably due to pressure from an increasing socioeconomic necessity and population [3]. To be prepared for the future and to better handle natural resources, it is essential to recognize and assess the impacts on the environment as a result of growing LULCC, primarily caused by human activities [4].

Monitoring LULCC at various scales is necessary for both local communities and policy-makers. Many districts use this report to help them plan and create their agenda. The accessibility of new, enhanced multi-spatial-temporal data monitoring systems in close-to-real-time has recently been the subject of much discussion [5,6]. Several studies on LULCC and ecological deviations have been comprehensively described by applying various geographical and sociological approaches. Additionally, all spatial and temporal aspects have identified human-related land-use activities as a crucial indicator of ecological change [7,8].

Assessment, observation, and analysis of an area’s continual changes in LULC requires a sizable amount of data. The accessibility of satellite data from numerous satellite sensors is beneficial, according to studies on LULCC [9,10,11]. The combination of satellite remote sensing (RS) data with geographic information systems has generated much interest due to concerns about the alteration and transformation of LULC dynamics [12]. Research of model development for future circumstances forecasts, assessments, and analyses the present and past based on facts. However, scenario prediction is also closely related to developments from the present to the past [13]. Various software programs are available for modeling future LULC scenarios, as well as LCM (Land Change Modeler), CA-MARKOV, DINAMICA-EGO, and CLUE-S, which use empirical methods founded on historical LULC [14]. The Modules MOLUSCE (QGIS Plugin) was recently made available to evaluate current LULC changes and forecast future LULC. Using four alternative models—Multi-Criteria evaluation (MCE), Artificial Neural Networks (ANN), Logistic Regression (LR), and weights of evidence (WoE)—this module may employ the Markovian technique to build a transition potential/possibility matrix and train a simulation model. To construct the simulated land use map, a Monte Carlo cellular automata model technique is employed [15,16,17].

Numerous researchers have applied the Markov chain-cellular automata (MC-CA) approach for the spatio-temporal simulation of LULC. These studies have shown that it is an effective tool for land-use planning. Numerous improvements are being made to environmental planning, administration, and research [9,18,19,20,21,22,23,24,25,26]. In Varanasi, Uttar Pradesh [18] employed the multi-layer perceptron-Markov chain (MLP-MC), cellular automata Markov chain (CA-MC), and stochastic Markov chain hybrid models to forecast future LULC situations (ST-MC). In addition, [9] tracked and predicted LULC changes in the Indian region of Patna using a hybrid MLP-MC model.

This study was conducted in the Muzaffarpur district of Bihar, which has been facing uncontrolled growth in urban areas since 1990, aggravated by irregular new settlements. This work analyses how geographical drivers of unique LULCC processes can be converted into accurate land change simulation and prediction [27]. In the study area under investigation, no previous study has used the CA-ANN model to predict future LULCC. The current study’s objectives are: (1) to analyze the spatiotemporal change in LULC in the Muzaffarpur district from 1990 to 2020, (2) to utilize the CA-ANN model to forecast LULC for the years 2030 and 2050, and (3) to show how this shift would affect future land-use planning. Regulators, resource managers, and government entities should closely monitor these spatio-temporal situation models to forecast changes in landscape structure in order to use natural resources sustainably.

2. Materials and Methods

2.1. Study Area

The Muzaffarpur district is located between 84°50′ E to 85°45′ E longitude and 25°53′ N to 26°25′ N latitude, covering a total area of 3176 km² (Figure 1). The district is located in the Indo-Gangetic plain, which is a fertile area. This saucer-shaped, low-centered town is located on Bihar’s expansive Indo-Gangetic plains, where glacier-fed and rain-fed meandering rivers transport Himalayan silt and sand. The district’s soil is rich in nutrients, well-drained, sandy, white in color, and extremely soft. The terrain is lush and green all year.

2.2. Datasets Used

Multi-temporal satellite data was used in this research for the modeling and prediction of LULC. Landsat 5 TM data, dated 11 May 1990; Landsat 7 Enhanced ETM+ data dated 5 May 2000 and 26 May 2010; and Landsat 8 OLI data dated 5 May 2020 were downloaded from the website of USGS.

Table 1. Details of the data sets used in this study.

Satellite-Sensor	Path/Row	Date of Acquisition
Landsat 5-TM	141/42	11 May 1990
Landsat 7-ETM+	141/42	5 May 2000
Landsat 7-ETM+	141/42	26 May 2010
Landsat 8-OLI	141/42	5 May 2020

details the satellite data used in this research. Road networks, railway networks, and the DEM were used as secondary datasets in this investigation. The slope was derived using the ASTER DEM, with a spatial resolution of 30 m, downloaded from the website of NASA earth data (https://search.earthdata.nasa.gov/ (accessed on 31 December 2021)). The road, rail, and river stream vector layers were created using Google Earth Image data. In addition, the distance to the road, distance to the railway line, and distance to the river, based on these vector layers, and the distance to the settlement, were all calculated. These layers are dynamic in nature, and, therefore, might incorporate the area estimate uncertainties in future LULC scenarios. However, as a regional study, the area under these layers is much less. Therefore, such minor uncertainties in estimates are manageable when the purpose of the study is future planning and management of land resources. The details of the data sets used in this study are listed in Table 1. Overall methodology used in this study is shown in Figure 2.

2.3. Methods

2.3.1. Image Pre-Processing

Using the Reprojection and Transformation tool in the ERDAS Imagine software, the atmospherically corrected satellite data were spatially referenced to a common UTM map projection (zone 45, north) with datum WGS 84. The correct band combination must be used to transform each image into a false-color composite (FCC). The bands B4, B3, and B2 were combined to create FCC for Landsat 5 TM and Landsat 7 ETM+ data. The B5, B4, and B3 bands were used to create the FCC for the OLI Landsat 8 data. The training data (signature) for LULC classification were created using these FCCs.

2.3.2. Classification of Satellite Imagery

Landsat data for the corresponding years were classified to create the LULC maps for 1990, 2000, 2010, and 2020, using a support vector machine classifier, separately, in ENVI 5.1 software. The seven main LULC classes examined in this study are: built-up areas, vegetation, fallow lands, croplands, litchi cultivation areas, wetlands, and waterbodies.

Supervised classification is based on the spectral signatures of the designated LULC classes. We used a support vector machine algorithm for classifying the images of all the dates. The support vector machine (SVM) is a supervised non-parametric statistical learning approach that was previously designed for binary classification. The SVM is based on the hypothesis that the training set is linearly unique. The SVM detects the optimal line, which splits up the training set without errors and maximizes the gap between the objects of each class and the optima line. The SVM uses only those training samples that designate class boundaries (support vectors). The SVM essentially involves parameterizing a Support Vector Classifier (SVC) based entirely on the reference information and the classification of the image data. Training signatures for classification were selected based on the visual image interpretation, augmented with statistical training class estimates in the form of mean, variance, and covariance matrix. We chose approximately 400 spectral signatures, equally distributed among the designated LULC classes.

To evaluate the accuracy of the classification output, the confusion matrix was created. The accuracy was measured using a variety of metrics, including the overall accuracy (OA), user’s accuracy (UA), producer’s accuracy (PA), and kappa coefficient (Kc). Kappa coefficient ($\mathrm{k}$) is mathematically expressed as:

$$\mathrm{kc}=\{\mathrm{N}{\displaystyle \sum }_{\mathrm{i}=1}^{\mathrm{r}}\left({\mathrm{X}}_{\mathrm{ii}}\right)-\mathrm{N}{\displaystyle \sum }_{\mathrm{i}=1}^{\mathrm{r}}\left({\mathrm{X}}_{\mathrm{i}+}.{\mathrm{X}}_{+\mathrm{I} }\right)\}/{\mathrm{N}}^2-{\displaystyle \sum }_{\mathrm{i}=1}^{\mathrm{r}}\left({\mathrm{X}}_{\mathrm{i}+}.{\mathrm{X}}_{+\mathrm{I} }\right)$$

where

r represents the number of rows in the error matrix; X_ii represents the number. of observations in row I and column I; X_i+ is the total of observations in row I; X_+I is the total of observations in column I; N is the total number of observations included in the matrix.

2.3.3. LULC Change Analysis

The LULC modifications in the CA-ANN model are described using the modification potential from the ANN learning process. The MOLUSCE plugin is used to carry out this strategy. Several LULCC maps have been made, and area alterations were used to assess change in various LULC classes. Using the MOLUSCE plugin, change maps were produced for the period of 1990 through to 2020. During this time, the locations that belonged to the same change detection class and the conversion from every LULC type were made public. The MOLUSCE plugin is a set of tools for evaluating and simulating LULC modification. The inclusion of user-specified change causes enables users to map alterations, identify transitions across LULC classes, model, and predict future landscape conditions [28].

The uncertainty in the registration of the satellite images was assessed using the association between the topographical variables and was analyzed using joint information uncertainty, through Pearson’s correlation [29]. The technique generates a transition matrix, which shows the percentage of pixels that change from one kind to another. Pearson’s correlation coefficient gauges the strength of a linear relationship between two data sets. The greater the overall value, the higher the linear association. The relationship coefficient among the four decades is seen in

Table 2. Pearson’s correlation for different pairs of year.

Initial Year	Final Year	Pearson’s Correlation
1990	2020	0.83
1990	2000	0.79
1990	2010	0.93

. The findings indicate a very high registration efficiency between 1990 and the anticipated 2020, preceded by the years between 1990 to 2010, and 1990 to 2000.

2.3.4. Prediction of Future LULC

The LULC maps for the start year (1990) and end year (2010) are the first phases in this model. Figure 3 depicts several geographical variables included in the model, such as DEM, Slope, Euclidean distance to the railway line, Euclidean distance to a river, Euclidean distance to road, and Euclidean distance to existing settlements. This model utilized these data as inputs to produce a simulation map of land use and land cover variations, allowing the identification of the study area’s shifting pattern between 1990 and 2020.

The plugin also produces a transition probability matrix that shows the proportion of classes that have changed to other classes and calculates the proportion of area changes in a particular year. Additionally, the plugin creates a change map from 1990 to 2020 for each of the seven classes of built-up area, vegetation, cropland, litchi cultivation area, fallow land, wetlands, and waterbody. This was conducted using the ANN-(Multi-layer perception) plugin to forecast the variations in LULC [28,29]. Based on the raster maps from 1990 and 2020, the LULC alterations for 2030 and 2050 are also projected. Future LULC maps are anticipated based on the dynamics and pattern of the existing LULC.

The plugin works by generating the transition potential (TP) maps by feeding data from an ANN-TP model with information about relevant spatial drivers. These potential determinants of LULC shift from one place to another. Explanatory variables can be divided into two categories: constraints and factors. Future modifications to LULC will omit the Boolean map that describes the limitations. The urban class was assigned as a constraint in this research. It served as a proxy for other, more difficult-to-generate socioeconomic factors, including population, economic activity, and employment opportunities. A number of factors determine changes in a region’s appropriateness. The CA-ANN model included in ANN-TP was used to construct the transition potential map after associations between all of the components, and each land transition was tested. ANN of the feedforward variety uses a supervised backpropagation technique. It is composed of three layers: input, hidden, and output. The input layer is responsible for feeding information into the network’s neurons, whereas the output layer is where the network’s output may be accessed. There are three distinct layers: an input layer, an output layer, and a hidden layer that can identify a deviation in the input and output patterns. For this procedure, each neuron calculates a value that is the product of the values stored in the nodes of the layer above it and the network weights between them. To further train the neurons, the backpropagation technique was employed. This approach iteratively adjusts the neural network’s weights to reduce the gap between the node’s actual output and the desired output. To do this, CA-ANN generates a random sample of cells in each land transition submodel, some of which have undergone the LULC transition, while the rest have not. Using the critical land transitions and explanatory variables, a new network of neurons, with weights, was created. The samples in these cells were split in half and used for training and testing purposes, respectively. The training procedure involved adjusting the weights of each connection to decrease the inaccuracy. Once the MLPNN had been run for many iterations, a higher accuracy rate was achieved. It is suggested that the accuracy rate is greater than 80%. After ANN-TP was implemented, transition possibilities were created for all material land changes. These generated transition potentials were then used to make LULC predictions in the latter part.

The forecast of LULC is only considered reliable when it has been confirmed using the referenced LULC classes. Therefore, the validation process has been conducted in the MOLUSCE plugin for the actual LULC of 2020, with simulated LULC of 2020 using cellular automata. The same validation approach has been used to estimate the LULC map for 2030 and 2050. The four-kappa statistic metrics that the validation module calculates—kappa histogram, overall kappa, kappa location, and percentage of correctness—are used to judge the model’s accuracy and are listed below:

$$K=P \left(A\right) - P E/1-P\left(E\right)$$

(1)

$$K1oc=P\left(A\right) -P\left(E\right)/Pmax-P\left(E\right)$$

(2)

$$Kh i=iPmax i - iP E/1 - iP\left(E\right)$$

(3)

$$P\left(A\right)=\sum \_\left(i=1\right) ^c Pii,=\sum \_\left(i=1\right) ^c PiT PTi$$

(4)

P_Max =∑_(i = 1) ^c [(min (PiTPTi)]

(5)

Pij denotes the I jth cell of the frequency distribution, PiT means the total of all cells in the ith row, PTj means the sum among all cells in the jth cell, and c is the number of raster classes in the I jth cell of the frequency distribution. By contrasting the categorized raster with the LULC projected raster, the degree of agreement here between rasters and their probability may be assessed. The K location evaluates the simulation’s capacity to detect location, whereas K overall evaluates the simulation’s overall performance. A value of 100% in kappa statistics denotes seamless agreement, while a rate of 0% denotes no agreement.

3. Results and Discussion

This section is divided into four sub-sections to better understand the past, present and future spatiotemporal dynamics of LULC in the study area, viz, LULC estimation for 1990, 2000, 2010, and 2020; LULC change assessment for 1990–2010, 2000–2020, and 1990–2020; Simulation and validation of LULC for 2020 using the CA–ANN model; and prediction of the years 2030 and 2050.

3.1. LULC Maps and Accuracy Assessment

Support vector machines were used to classify Landsat data for 1990, 2000, 2010, and 2020.

Table 3. Area estimates of LULC during the period of 1990 to 2020.

Class Name	Year 1990		Year 2000		Year 2010		Year 2020
	Area	%	Area	%	Area	%	Area	%
Built-up	1534.71	3.08	3722.13	4.17	15,215.58	7.79	23,807.97	9.76
Vegetation	48,141.81	15.17	29,749.95	13.37	23,046.93	11.26	26,823.15	12.56
Cropland	98,709.39	31.10	136,799.01	33.11	109,475.19	28.50	104,292.99	25.89
Fallow land	152,496.89	48.06	133,043.58	41.14	152,218.80	38.51	143,913.42	35.79
Wetlands	10,151.47	3.10	7396.38	2.50	6501.42	2.04	6616.98	2.08
Waterbody	4179.94	3.31	3958.11	3.14	3137.31	3.03	3111.48	2.57
Litchi plantation	3759.34	1.18	4496.76	1.41	8168.71	2.57	10,597.95	3.35

shows statistics of the total area change during the year, and Figure 4a–d demonstrates the regional and measurable distribution of LULC classes for different years, such as 1990, 2000, 2010, and 2020. The data in this study has been divided into seven LULC classes, including built-up, fallow land, agricultural land, vegetation, water bodies, litchi cultivation area, and wetlands. As a result of the study of spatiotemporal data, it was discovered that from 1990 to 2020, several LULC classes showed deviations, including gains and losses. Figure 5 and Figure 6 shows the area estimates of various LULC from 1990 to 2020 and the associated change, respectively. An accuracy assessment was performed using the confusion matrix approach to estimate the utility and quality of classified images from the years 1990, 2000, 2010, and 2020. As part of the accuracy evaluation procedure, kappa statistics, producer accuracy, user accuracy, and overall accuracy were calculated [30,31,32].

Table 4. Percent change estimates in LULC during the period of 1990 to 2020.

Class Name	1990–2000	2000–2010	2010–2020	1990–2020
	% Change	% Change	% Change	% Change
Built-up Area	142.53	308.79	56.47	1451.30
Vegetation	−38.20	−22.53	16.38	−44.28
Cropland	38.59	−19.97	−4.73	5.66
Fallow land	−12.76	14.41	−5.46	−5.63
Wetlands	−27.14	−12.10	1.78	−34.82
Waterbody	−5.31	−20.74	−0.82	−25.56
Litchi	19.62	81.66	29.74	181.91

assesses the classification accuracy of LULC maps for the years 1990, 2000, 2010, and 2020.

3.2. LULC Change Analysis

The LULC maps produced by analyzing the Landsat TM/ETM+/OLI datasets for the years 1990, 2000, 2010, and 2020, respectively, served as the basis for assessing the LULC change in the study area.

3.2.1. Change in Built-Up

Multitemporal Landsat data analysis reveals that the Built-up area in the Muzaffarpur district has increased significantly between 1990 and 2020. The percent increase in the built-up was witnessed to be highest between 2000 and 2010 decade, and equaled approximately 308.9%, followed by the 1990–2000 decade (142.53%), and the 2010–2020 decade (56.47%). The overall increase in the built-up, in the district from 1990 to 2020, was approximately 1451.30 % of the area in 1990. The area estimates are shown in Table 4, and the percent area estimate is shown in

Table 5. Accuracy assessment of LULC maps of different years.

LULC Classes	1990		2000		2010		2020
	UA (%)	PA (%)	UA (%)	PA (%)	UA (%)	PA (%)	UA (%)	PA (%)
Built-up	98.21	95	94.44	94.44	84.37	90.75	96.43	83.07
Wetland	100.00	100.00	100.00	100.00	90.90	90.90	85.71	85.71
Fallow land	73.41	89.47	45.68	83.33	100.00	85.71	85.21	81.21
Cropland	83.24	100.00	89.65	100.00	90.49	81.82	85.00	92.31
Vegetation	87.64	93.36	92.88	92.76	69.23	81.82	53.85	77.78
Litchi plantation	81.26	84.98	87.76	84.79	96.50	88.74	81.66	99.56
Waterbody	100.00	96.56	99.11	99.82	95.29	84.08	94.37	100.00
Overall accuracy (%)	89.63		87.41		89.43		85.73
Kappa coefficient	0.79		0.75		0.87		0.85

.

3.2.2. Change in Vegetation

The percent decrease in vegetation was witnessed to be highest between 1990 and 2000, equaling approximately 38.2%, followed by the 2000–2010 decade (22.53%). However, the vegetation showed an increase between 2010 and 2020, equaling approximately 16.38% compared to 2010. The overall decrease in the vegetation in the district from 1990 to 2020 was 44.28% of the area in 1990. In some areas the decrease in vegetation corresponded to an increase in cropland area. However, these are observed in certain areas and verified from ground truthing.

3.2.3. Change in Cropland

The area under cropland witnessed an increase between 1990 and 2000, equaling approximately 38.59%. However, a decrease in its area was later witnessed, the highest decline seen in the 2000–2010 decade (19.97%), followed by a 4.73% decrease in the 2010–2020 decade. Overall, it has shown an increase in the area from 1990 to 2020, equaling an approximately 5.66% increase from the area in 1990.

3.2.4. Change in Fallow land

The percent decrease in the fallowland was witnessed to be highest in the 1990 and 2000 decade, and equaled approximately 12.76%, followed by the 2010–2020 decade (5.46%). However, the fallowland showed an increase between the 2000 and 2010 decade, that equaled approximately 14.41%, compared to 2000. The overall decrease in the fallowland in the district from 1990 to 2020 was observed to be 5.63% of the area in 1990. Rapid urban growth has led to a decrease in the fallowland, as well as an increase in the litchi cultivation area.

3.2.5. Change in Wetlands

The percent decrease in the wetlands was witnessed to be highest in the 1990 and 2000 decade, equaling approximately 27.14%, followed by the 2000–2010 decade (12.10%). However, the wetlands showed a marginal increase in the 2010 and 2020 decade, that equaled approximately 1.78%, compared to 2010. The overall decrease in the wetland in the district from 1990 to 2020 was observed to be 34.82% of the area in 1990. The wetland areas have been converted to urban colonies in the district, making these areas vulnerable to floods.

3.2.6. Change in Waterbody

The river is the prominent area under this LULC class. Three major rivers, Gandak, Budhi Gandak, and Bagmati, flow in the Muzaffarpur district. During this study, after the analysis of the spatiotemporal data, it is observed that the rate of change is slower than other LULC classes. The percent decrease in the waterbody class was witnessed to be highest between 2000 and 2010, equaling approximately 20.74%, followed by the 1990–2000 decade (5.31%), and 2010–2020 decade (0.82%). The overall decrease in the waterbody class in the district from 1990 to 2020 was observed to be 25.56% of the area in 1990. The decrease in the area under water body class was particularly observed along those streams that have dried due to climate change.

3.2.7. Change in the Litchi Cultivation Area

In this study, we take litchi as a separate class from vegetation, as litchi is an important cash crop in the study area. The litchi cultivation area has increased during the period because of its economic value. The percent rise in the area under litchi plantation witnessed the highest increase in the 2000 and 2010 decades, which equaled approximately 81.66%, followed by 2010–2020 decade (29.74%) and 1990–2000 decade (19.62%). The overall increase in this class in the district from 1990 to 2020 was 181.91% increase of the area in 1990.

The changing dynamics of the LULC in the study will have a pronounced impact on the flood management of the area, particularly the decrease in the wetlands. The state of Bihar constantly looms under the threat of floods from the Kosi, Gandak, and Bagmati rivers. The rampant urbanization, with a decrease in the natural sinks of the flood waters, has increased the threat to the lives and property of people in these regions. In recent years, the study area has witnessed huge floods and destruction of property, and loss of life. The floods in July 2017 destroyed huge farmlands and caused the devastation to the lives and property of people living in the study area.

3.3. Artificial Neural Network-Based Transitional Potential Modeling (ANN-TPM) in the LULC Change

Every method for calibrating and modeling LULC change uses LULC data as input. The transition probability matrix, which would be derived by different LULC from two dates, serves as the foundation for the change model in CA-ANN. The Transition probability matrix is constructed by determining the region from each LULC class for upcoming years and the level of changes for each transition. Each transition’s probability is calculated to assess the likelihood that something will change [33,34]. This approach is suitable when the algorithm needs to handle a lot of ambiguity or challenges to implementing input data. As a result, a constant index is produced that ranks the landscape from zero to one. As a result of CA-ANN using fuzzy logic, a fixed range, such as zero and one, is selected based on the terrain’s advantages. Interactions between linked neurons and changes in the strength of the connections supporting those interactions are the fundamental building blocks of CA-ANN. This adjustment is dependent on the network’s anticipated output as well as its input data. This is called neural network learning, as shown in Figure 7. A transition matrix that demonstrates the manner wherein land-use types are changed is the result of this learning (Table 5).

The classes most likely to remain the same from 1990 to 2020 are vegetation, water bodies, and fallow land, with probabilities of 0.78, 0.76, and 0.76, respectively (

Table 6. Transition probabilities of change among LULC classes for period 1990–2010.

	Built–Up	Vegetation	Cropland	Fallow Land	Wetland	Waterbody	Litchi Plantation
Built-up	0.10	0.04	0.09	0.02	0.01	0.01	0.04
Vegetation	0.30	0.70	0.80	0.75	0.02	0.01	0.20
Crop land	0.40	0.70	0.50	0.50	0.03	0.01	0.30
Fallow land	0.50	0.40	0.30	0.80	0.01	0.02	0.25
Wetland	0.09	0.07	0.10	0.03	0.30	0.70	0.01
Waterbody	0.05	0.02	0.01	0.08	0.04	0.70	0.02
Litchi plantation	0.20	0.40	0.30	0.09	0.05	0.01	0.80

). The built-up class has the highest transition probability (0.44), making it the most active. The movement in different land use classes between 1990 and 2020 is comparable with the earlier period; vegetation is the most stable class and has an 0.84 likelihood of urbanization. The most dynamic classes are vegetation and fallow land, with conditional probabilities of 0.23 and 0.37, respectively. These classes were primarily transformed into water bodies and grew with a probability increase of 0.60. This model helps support important planning and policy choices and analyzes how the land use/land cover system functions. Additionally, it may forecast potential LULC changes for future dates based on a number of variables [35,36,37]. The CA-ANN model has been used in various research for a wide range of applications [36,38,39,40,41].

3.4. Future LULC Simulation and Prediction Using the CA-ANN Model

The cellular automata-based method was used to project the 2020 raster between 1990 and 2010, with a phase extent of two years and five iterations. The simulated 2020 raster is compared to the actual 2020 LULC classified raster (Figure 8a,b). After obtaining agreement through validation, the same approach was followed for 2030 and 2050, with an increase in iterations. The predicted LULC map for 2030 and 2050 is shown in Figure 9a,b.

Table 7. Area estimates of predicted LULC in hectares.

	Actual	Simulated
LULC	Year 2020 (Area in ha)	Year 2030 (Area in ha)	Year 2050 (Area in ha)	% Change in Area 2020–2030	% Change in Area 2020–2050
Built-up	23,807.97	26,939.69	49,683.90	13.15	108.69
Vegetation	26,823.15	22,986.43	18,013.60	−14.30	−32.84
Cropland	104,292.99	102,624.32	100,627.52	−1.60	−3.51
Fallow Land	143,913.42	151,957.17	130,257.00	5.59	−9.49
Wetlands	6616.98	6461.19	4509.98	−2.35	−31.84
Waterbody	3111.48	3137.30	2890.48	0.83	−7.10
Litchi plantation	10,597.95	11,858.90	13,853.70	11.90	30.72

shows the statistics of predicted LULCC in hectares, and the aerial distribution of various LULC classes is shown in Figure 10.

As indicated in

Table 8. Model validation’s Kappa index (Simulated 2020 LULC with actual 2020).

Parameter	Value (%)
Kappa (Histogram)	0.93
Kappa (Overall)	0.75
Kappa (Location)	0.78
% of Correctness	78.22

, the validated module examined the consistency between the projected 2020 LULC and the actual LULC of 2020. A good promise among anticipated and actual 2020 LULC conditions is indicated by the kappa (histogram) estimation of 0.93, kappa (overall) of 0.75, and kappa (location) of 0.78, with a proportion of correctness of 78.22%. This shows that the model is good for Muzaffarpur. As a result, the LULC raster is predicted using the CA-ANN model with a step size of only two years and five iterations for 2030 and 2050. The validation information for the anticipated LULC in 2030 and 2050 is shown in

Table 9. Model validation’s Kappa index (Simulated 2030 and 2050).

Parameter	Value (%)
	2030	2050
Kappa (Histogram)	0.93	0.91
Kappa (Overall)	0.65	0.63
Kappa (Location)	0.69	0.67
% of Correctness	78.62	76.91

. The calculation for kappa (histogram) is 0.93 and 0.91, kappa (overall) is 0.65 and 0.63, and kappa location is 0.69 and 0.67 for forecasted 2030 and 2050, respectively. The percentage of correctness is 78.62% and 76.91%.

The future scenarios concerning the LULC changes for 2030 and 2050 showed built-up as a major transitioning land cover. In 2030 and 2050, the built-up increase will account for 13.15% in the area compared to 2020. Whereas in 2050, the built-up will increase by 108.69% compared to its area in 2020. Similarly, the litchi plantation will also show an increase in the area under its cultivation in the future. In 2030, the area under its cultivation will show an 11.90% increase compared to 2020, whereas in 2050, it will show an increase of 30.72%. For 2030, with respect to other LULC, the highest decline in area will be observed in vegetation, which will be approximately 14.30% compared to 2020, followed by wetlands (2.35%) and cropland (1.60%). For 2050, the highest decline in the LULC class will again be witnessed in vegetation, which will be approximately 32.84%, followed by wetlands (31.84%), fallow land (9.49%), and water body (7.10%). The continuous changing trends in the LULC will have adverse impacts on the land system processes; in particular, the increase in the built-up and the decrease in natural vegetation, wetlands, and water bodies [42,43,44]. India is facing recurrent heat waves and an increased number of annual urban heat island (UHI) events [45,46]. These adverse processes come as a result of rampant urbanization and the decrease in the natural landscape. Moreover, since UHI forms a positive feedback loop in the climate change cycle, it will also exacerbate the adverse consequences of climate change in the region [47,48]. The results shown in the present study for one region of India are applicable for the same trends in other parts of the country. The land managers and climate change risk assessors will find this work essential for devising appropriate strategies for mitigating the consequences of climate change [49,50,51].

4. Conclusions

The objective of this research was to detect variations in LULC from 1990 to 2020 and to predict the LULC for the years 2030 and 2050. In this study Landsat, TM/ETM+/OLI data for 1990, 2000, 2010, and 2020 were used to assess the changing LULC of the Muzaffarnagar district of the Bihar state of India. LULC has been classified into seven major classes using SVM (Support Vector Machine). The generated LULC for 1990, 200, 2010, and 2020 were validated using various accuracy assessment indices. After ground truthing and Google Earth-based historical imagery validation, the final LULC products for the years 1990, 2000, 2010, and 2020 were found to have kappa coefficients of 0.79, 0.75, 0.87, and 0.85, respectively. Subsequently, the LULC simulation was successfully performed for the years 2020, 2030, and 2050 using the CA-ANN model. The kappa index statistics were used to examine the validity of the prediction models for 2030 and 2050. The CA-ANN model shows a descriptive ability for future prediction based on validation findings. The prediction model not only describes previous changes, quantitatively and spatially, but similarly forecasts the pattern and intensity of future changes. The future scenarios concerning the LULC changes for 2030 and 2050 showed built-up as a major transitioning land cover. When compared to 2020, the built-up area is projected to expand by 13.15 percent in 2030. When compared to its 2020 size, the built-up area is expected to grow by 108.69 percent in 2050. In 2030, agriculture and wetlands will lose the most area, relative to other LULCs, at 2.35 percent, while vegetation will lose 14.30 percent of its current area (1.60 percent). In 2050, vegetation loss is expected to be the greatest in the LULC class, at approximately 32.84 percent, followed by wetlands loss, at 31.84 percent; fallow land loss, at 9.49 percent; and water body loss, at 0.01 percent (7.10 percent). Constant shifts in LULC are predicted to disrupt land system processes. The pattern observed in this study, conducted in one portion of India, is also apparent in other sections of the country. This research will be crucial for land managers and climate change risk assessors in developing effective plans to lessen the impact of climate change.