Correlation between alterations in degree of urbanization and the dynamics of agricultural land in Karawang regency

.


Introduction
Today, the world's population has tripled compared to the middle of the twentieth century.According to a report from the United Nations, the region with the largest population in the world is East Asia and Southeast Asia, which reaches 2.3 billion people.This covers 29% of the world's population.Indonesia is the fourth country with the largest population in the world, after China, India, and the United States, with a population reaching 275 million people in 2022.[1].
Based on the World Population Prospects 2018 Revision report, globally, most of the population in 2018 lived in urban areas, with a percentage of 55%.Urbanization projections show that by the middle of the 21st century, it is predicted that 68% of the world's population will be in urban areas.Ninety percent of the increase is concentrated in Asia and Africa.In 2018, 55% of Indonesia's total population of 267 million lived in urban areas.It is predicted that between 2018 and 2050, Indonesia will be one of the countries contributing to an increase in population in urban areas of 50 million people.[2].
Areas experiencing urbanization in Indonesia can be found in metropolitan cities such as Jabodetabek.However, other urban areas, including medium-sized cities, will also experience growth [3].City growth will cause more intensive land use around urban areas and shift production focus towards production with higher added value [4].As a result, this will create intense competition for land use around growing urban areas, which may increase socioeconomic vulnerability in affected areas.
One method designed to capture the urban-rural continuum is the degree of urbanization.Research regarding the degree of urbanization, especially in Indonesia is rarely carried out.Therefore, the research aims to (1) describe the degree of urbanization in Karawang and (2) determine the correlation between changes in the degree of urbanization and the dynamics of rice fields.The degree of urbanization in Karawang is classified using the level 2 classification by the European Union.This classification is based on the share of the local population living in urban clusters and urban centers, and it classifies local administrative units level 2 (LAU2).The utilization of the European Union's Level 2 classification enables the research to offer a holistic comprehension of the urbanization process in Karawang and its consequences on the rice fields.

Study Area
This research was conducted in Karawang Regency, West Java Province.Karawang Regency was formed in 1950 on a legal basis in Law No. 14 of 1950 concerning the Regency Regional Government within West Java Province.According to geographical data, Karawang Regency is at 107⁰ 02' -107⁰ 40' East Longitude and 5⁰ 56' -6⁰ 34' South Latitude.The area of Karawang Regency is 1,913.71km2.Most of the area is plain with 0-5 meters above sea level.Only a small part has hilly relief with a 0 -1,200 meters height above sea level.In the northern part of Karawang Regency it is a plain area; in the central part, the relief is hilly, and in the southern part, there is Mount Sanggabuana [8].

Data
This section provides a comprehensive overview of the datasets utilized for mapping the degree of urbanization.Firstly, we delve into the process of collecting the Landsat satellite dataset.Following that, we elucidate the methodology behind gathering population data.Launched for the first time in 1972, the primary objective of the Landsat program was to gather data relating to ecosystems, agricultural activities, urban centers, and the monitoring of clean water sources.Landsat data, characterized by its accessibility without charge, serves as a valuable resource for diverse analytical purposes, encompassing the examination of environmental transformations, agricultural management strategies, water resource allocation, and the assessment of natural calamities, among others.The utility of Landsat data is underpinned by a multitude of meticulously planned aspects inherent to both the satellite and its overarching mission, which have collectively enriched our understanding of the Earth's dynamics.[11].
Landsat data recording offers almost continuous image recording, but several differences exist between one satellite and another.The first group (Landsat 1-3) is equipped with a Multi-Spectral Scanner (MSS) that records data via four channels: visible and near-infrared.The next group (Landsat 4-7) carries Thematic Mapper (TM) or Enhanced Thematic Mapper (ETM+) sensors, which feature finer spatial resolution and increased radiometric resolution compared to the previous group.In this group, the number of spectral channels has been expanded by adding mid-infrared and thermal infrared channels.In addition, Landsat 7 adds a panchromatic channel by increasing spatial resolution.The third group (Landsat 8 and Landsat 9) is equipped with an Operational Land Imager (OLI) sensor.OLI's spectral resolution is complemented by dark blue and cirrus channels [12].

Population Data
The population data used in this research is published by the Indonesian Central Statistics Agency.Population data collected corresponds to the year of the population census conducted in Indonesia.The data collected is for 1990, 2000, 2010, and 2020.This data is used to obtain population density values in Karawang Regency.The population data collected is data from all villages in Karawang.Then, to make it easier to display the data, the population data for each sub-district will be displayed in Table 2.
Population data is a very important aspect to study.One thing that can be studied is population dynamics.
Population dynamics is the fluctuation in the population size of a particular species within a specific geographical area [13].To understand it, it's essential to consider interactions that occur within and between the same species.
This includes examining population characteristics, including population size, age distribution, birth rates, deaths, immigration, and emigration.Under normal circumstances, populations can exhibit exponential growth, which is characterized by a situation where the growth rate increases as the population increases [14].
Berke and Kaiser (2006) stated that one of the most important tasks in developing planning and policy is to carry out population projections.This will provide accurate estimates to help communities allocate resources and services appropriately [15].Especially in the planning sector, population projections are very important in determining future needs in the areas of land use, employment, housing, facilities, and infrastructure.In addition, population dynamics are important in strategies aimed at conserving biodiversity, which until recently have traditionally emphasised a single-species perspective [13].

Methods
The flow diagram of this research is represented in Fig. 2.This section consists of five subsections: satellite image processing, image classification, accuracy assessment, applying the degree of urbanization, and correlation assessment.In more detail, all of these subsections are described.

Satellite Images Processing
Landsat data often requires Pre-processing before analysis to account for sensor, solar, atmospheric, and topographic effects.Radiometric correction is carried out to improve two things, namely, improving the image's visual quality and correcting pixel values that do not correspond to the object's actual reflection or spectral emission.Sensors, sun, atmosphere, and topography.The term absolute is used to describe the process of obtaining "true" and comparable values [16]

Fig. 2. Research Flow Chart
In addition, the Landsat images are images with visible and infrared waves.As input in multispectral classification training, two indices were added: the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Built-up Index (NDBI).NDVI serves as an indicator of vegetation covering the Earth's surface, and NDVI spatial composite images are formulated to facilitate the enhanced differentiation between verdant vegetation and exposed soil [17].NDVI products, derived from Landsat Surface Reflectance data, are generated using data obtained from Landsat 4-5 Thematic Mapper (TM), Landsat 7 Enhanced Thematic  [19].
On the other hand, NDBI highlights constructed urban zones by using the ratio between Near Infra-Red and Sort Wave Infrared bands of Landsat Imagery [20].NDBI is also employed for automating the procedure of delineating built-up regions, offering superior precision and impartiality in mapping urban land in contrast to supervised classification methods [21].NDBI is a potent tool for detecting the presence of developed regions and is utilized for quantifying the extent of built-up areas [22].NDBI yields DN values that approach 0 for forests and agricultural areas, negative values for water bodies, and positive values for urban areas, thereby enabling the differentiation of urban land use from other land cover types [19].

Image Classification
Determination of land use is carried out using the Support Vector Machine (SVM) classification.SVM is a supervised learning model used in classification and regression analysis.The advantage of SVM is its ability to learn data classification patterns with a balanced level of accuracy and reproducibility.SVM has been widely used for classification.SVM is a versatile method that covers several data science scenarios.[23].SVM is a nonparametric classification, so data distribution assumptions are unnecessary [24], [25].SVM tries to find an optimal hyperplane that separates data groups into a predetermined number of classes.The maximum hyperplane is a decision boundary that maximizes the margin distance between class boundary hyperplanes that are parallel to the optimal hyperplane and is defined by the training samples closest to the boundary [25].
In this research, the SVM classifier operating at the pixel level is applied for classifying Land Cover using ArcGIS tools.The selected image is sourced from the United States Geological Survey (USGS), Level 1, and shows minimal cloud cover (less than 10% coverage).Initially, these images display upper atmosphere (TOA) reflectance values and then undergo atmospheric correction to convert them to lower atmospheric reflectance values.This correction process is carried out using the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) module, which is an integral component of the QGIS software package [21].The bands used in the classification are described in table 3.

Accuracy Assessment
The classification carried out will produce a land use map.The classification results will have a certain level of accuracy.These overall accuracy results can be measured quantitatively.Accuracy of incorrect classification is usually emphasised in the existing land use classification aspect [16].The accuracy test used in this research uses an error matrix.Matching is done by determining samples based on classification and field classification results.The sampling technique used is Stratified Random Sampling, where the area of each classification determines the number of classes, and the sample points are determined randomly for each land cover classification.Accuracy test calculations are carried out using Equation 1.

Applying the Degree of Urbanization
The degree of urbanization is the classification of local administrative units (LAU) as cities, suburbs, or rural areas based on the level of urbanization.This is influenced by geographical proximity and population density.The measurements were carried out using a grid of 1 km2.The minimum population threshold in a grid is used to determine which class each LAU can belong to [26].The Degree of Urbanization is a new method devised to capture the urban-rural continuum [27].
Determining the degree of urbanization uses built-up land data extracted from land use data and population data obtained from the Indonesian Central Statistics Agency.The steps employed for assessing the level of urbanization are elaborated as follows: 1. Generate a Population Grid A population grid is a uniform raster layer comprising pixels of consistent shape and size, containing data regarding the population residing in each pixel.These grids prove valuable in comprehending population distribution and density, aiding governments in making informed choices about resource allocation and strategic planning.Furthermore, they serve as a fundamental initial stage in assessing the level of urbanization within a given area [28].However, if it is not possible to obtain this data, then census data with administrative boundaries can be combined with spatial data such as built-up land [29].

Generate a Degree of Urbanization Grid
The population grid will be extended to create a degree of urbanization grid, commonly known as a geospatial settlement classification grid.This grid is utilized to classify 1-square-kilometer cells based on population density and size criteria for determining the type of settlement they are represented by.There are two levels of classification involved in this process.The initial level comprises three primary classes: urban center grid cells, urban cluster grid cells, and rural grid cells.The second level offers a more detailed breakdown with eight classes: urban center grid cells, densely populated urban cluster grid cells, semi-densely populated urban cluster grid cells, suburban or peri-urban grid cells, rural cluster grid cells, sparsely populated rural grid cells, very sparsely populated rural grid cells, and water grid cells.Grids will be created at both classification levels, and a comparative analysis will be conducted between them [28].However, only level 2 was used in the analysis, as shown in Table 4 [29].
Using this dataset, a map will be created that displays the predominant settlement type of each commune.The classification will merge the commune's geography with the population grid and degree of urbanization grid to ascertain the type of settlement in which most people in each commune reside [28].

Correlation Assessment
Regression analysis is a statistical technique for estimating the relationship between two or more variables with a cause-and-effect relationship and for predicting a topic by analyzing the relationship [30].When using multi-temporal data, the type of statistics that can be used is panel data regression.Panel data combines crosssection data and time series, so it can also be interpreted as data obtained from the same individuals observed within a certain period [31], [32].
There are three models used in panel data regression [32].The three models are the Pool Effects Model (PEM), Fixed Effects Model (FEM), and Random Effects Model (REM).PEM assumes that the coefficients of the  The area of rice fields represented 50.19%, 52.76%, 50.84%, 49.06%, and 47.09% of the total area of Karawang Regency in 1990Regency in , 2000Regency in , 2009Regency in , 2020Regency in , and 2023. .In 2000, the area of rice fields increased by 4919 ha or 2.57% compared to 1990.However, after 2000, the area of rice fields continued to decline.The decline in 2009 was 3676 ha (1.92%), 3424 ha (1.78%) in 2000, and 3772 ha (1.97%) in 2023 (Table 5 and Figure 3).The decline in the area of rice fields mainly occurred in areas close to cities and highways, especially in the central part of Karawang Regency [34].This is in line with the classification results, where the largest land conversion occurred in the central region.
Meanwhile, the built-up land area continues to increase from 1990 to 2023.The built-up land area was 14.5%, 18.48%, 22.84%, 25.85%, and 27.57% in 1990, 2020, 2009, 2020, and 2023.The largest built-up land area increase occurred between 2000 and 2009, covering 4.36% or 8351 ha.The smallest increase occurred between 2020 and 2003, covering an area of 3284 ha (1.72%).This increase mainly occurred in the central region, the center of urban and industrial activities [34].The increase in built-up land area is due to the high conversion of land from rice fields to built-up land, such as housing and commercial and industrial activities [35].
Forest & Plantation land cover also decreased each year of the study.The highest land conversion occurred between 1990 and 2000, covering an area of 14,594 ha.On the other hand, the land cover of water bodies & ponds and bare lands continues to fluctuate from year to year.

Accuracy
The overall accuracy calculation uses a confusion matrix by comparing sample points in the field and sample points on the map.The number of samples used was 50 samples obtained from calculations using the Slovin formula.The population used in Slovin's calculations is the number of pixels in the study area, namely 4,995,930.The expected margin of error value is 0.15.So, the number of samples produced is 44.44 samples.From these values, they were   Comparison between field samples and maps produces an error of five samples.The most errors occurred in the class of map samples in the form of builtup land, which turned out to be forests & plantations in the field with two samples.Next is an error in the map class in the form of rice fields, which turned out to be turned into built-up land in the field in one sample and turned into forests and plantations in one sample.Lastly, the land cover class in the form of bare lands on the map turns out to be a forest & plantation class in the field for one sample.(Table 6) The correct number of samples is 45 samples.This number shows that the overall accuracy value for land cover classification results in 2023 is 90%.According to Carletta (1996), identifying land use and land cover categories from remote sensing data must reach an accuracy level of at least 85% [36].Thus, the accuracy value obtained is still above the minimum accuracy threshold, and the classification model can be used.

Degree of Urbanization
Before applying the degree of urbanization, the population densities in each 1x1 km grid must first be calculated [29].This is because the classification of the degree of urbanization uses population density data.Population density data for each grid is obtained from a combination of built-up area and population data at each village administrative boundary.The resulting population density is depicted in Fig. 5.
Population density between 1990, 2000, 2010, and 2020 has changed.Significant changes are clearly visible in the central part of the district.This is in line with the increase in built-up land in the area.Meanwhile, areas with quite low additions of built-up land experienced quite low changes in population density.This shows that the development pattern in Karawang Regency is in the central area around the main road.
After the population density data is obtained, the data is then classified using the classes created by the European Union.The results show that most areas in Karawang Regency are included in the suburban or peri-urban class.The dominant degree of urbanization in the southern region is the Low-Density Rural area.The central part of the Karawang Regency area is an Urban Centers classification area.In 1990 and 2000, the resulting urban centers were still separated into three parts.Then this changed in 2010 and 2020 when the urban center areas, which were initially separate, became one due to expansion.The results of the classification of the degree of urbanization are depicted in Fig. 6.
The city's development in Karawang Regency cannot be separated from its position in the Jakarta-Bandung corridor.The growth of this area was triggered by its strategic location between Jakarta and Bandung, two big cities in Indonesia [37] [38].The development of urban areas in Karawang Regency is concentrated in the central region.In the central area, there is Jalan Pantura.This is what makes the growth of Karawang Regency focus on this area.In line with research conducted by Rustiadi et al., urban expansion is still occurring but is concentrated along road corridors [37].

Correlation Between Degree of Urbanization and Dynamics of Agricultural Land
To determine the correlation between changes in the degree of urbanization and the dynamics of rice fields in Karawang Regency, data on changes in the degree of urbanization and changes in rice fields in the three-year observation period were used.The three observation periods are changes from 1990 to 2000, 2000 to 2010, and 2010 to 2020.The data used is data with village administrative boundaries concerning 1990.The 1990 village boundaries were chosen because they have occurred several times-village expansion during the research period.
The number of villages in Karawang Regency in 1990 was 302 villages.However, in statistical calculations, only 301 villages were used.This is because there is one village, namely Cikampek Kota, where the area of rice fields is so small that it is considered non-existent.Then, statistical tests are carried out to obtain an appropriate statistical model.
The Chow test shows that the resulting probability value is 0.7959.This shows that the probability value is above 0.05.In the Chow test, it can be concluded that the CEM model is the most suitable to use.The Hausman test results show that the resulting probability value is 0.4025.This shows that the resulting probability value is also higher than 0.05.So, in the Hausman test, it can be concluded that the REM Model is suitable.The last test, the Lagrange Multiplier test, shows a probability value of 0.0024.This shows that the resulting probability value is lower than 0.05, and the appropriate model is REM.The conclusion that can be drawn from the three tests that have been carried out is that the REM model is the most appropriate model for describing the correlation value in the data used.
Then, calculations using the REM Model are carried out to determine the correlation.The REM model does not require classical statistical tests so that the results can be directly interpreted.REM eliminates classical testing because it assumes that individual components are considered random and unrelated [39].This is different from the FEM and CEM models.
The value resulting from the REM model (Fig. 7) shows that the probability value for variable X (change in the degree of urbanization) is 0.3382.This shows that the resulting probability value is more than 0.05.A probability value higher than 0.05 indicates no significant correlation between variable X (change in the degree of urbanization) and variable Y (dynamics of agricultural land).This aligns with research conducted by Rustiadi et al. that the decreasing level of suburbanization in the Jakarta-Bandung Mega Urban Area (JBMUR) is not in line with the decreasing level of agricultural land conversion [37].Several things can be suspected as driving the conversion of rice fields.Such as the low income of farmers and the distance between rice fields and roads [40].The ineffectiveness of government policies, increasing residential land, developing new cities, developing private industrial areas, and adding infrastructure can also encourage land conversion [37].Apart from that, local demographics, social and economic factors also play an important role in urban expansion, which can result in land conversion [41].

Conclusion
Analysis of land cover changes in Karawang Regency over several decades shows that there have been real changes in the region's landscape.There is a gradual reduction of rice fields and an increase in built-up areas, especially in the central part of the district.This trend aligns with the regions located in the Jakarta-Bandung Corridor and its strategic importance.
The study classifies the degree of urbanization in the Karawang Regency using European Union-defined classes.Most of Karawang falls into suburban or periurban categories, with the central part of the regency classified as Urban Center.The concentration of urban development is mainly observed around Pantura Road.
This research also investigates the correlation between changes in the degree of urbanization and the dynamics of rice fields.Analysis using the Random Effects Model (REM) shows no significant correlation between changes in the degree of urbanization and changes in agricultural land.This shows that factors other than urbanization, such as low farmer incomes, infrastructure development, government policies, and demographic and economic factors, have a greater impact on land conversion.
This study highlights the complexity of land use change in the context of urbanization.Urbanization is not the only factor.Effective land management and urban planning in Karawang Regency must consider various economic, social, and policy-related factors contributing to land-use changes.
These findings emphasize the importance of appropriate policies and strategies to address the challenges of urbanization, especially its impact on agricultural land.Policies to conserve agricultural land, support farmers, and sustain urban development are critical to maintaining the region's food security and economic stability.
Overall, the weakness of this research is there is no accuracy test on the data used in 1990, 2000, 2010, and 2020.The data accuracy test was carried out only on the latest data (2023), so the classification results' accuracy in years before 2023 was questioned.Furthermore, additional data (such as data regarding government policies and other data) that can trigger the development of a region can be added to find out the most influential factor in the dynamics of agricultural land in Karawang.
In summary, this research shows that the degree of urbanization changes does not significantly affect the dynamics of rice fields.Various factors influence these dynamics.Understanding and addressing this complexity is critical for the region's sustainable development and land use management.
independent variables are constant across all individuals and periods in the panel data set.The FEM model assumes that each individual's characteristics are different in various periods.The intercept values in the estimation models reflect the differences, which differ for each individual.The REM model assumes that in various periods, the characteristics of each individual are different; only in this model, these differences are reflected by errors from the model.To be able to determine the best model from the three models, a model test must first be carried out.The model selection used three types of tests: the Chow, Hausman, and Lagrange Multiplier.The Chow test is employed to ascertain whether the appropriate model for estimating panel data is the CEM or Fixed Effect FEM model.If Results H0: Select CEM (p> 0.05); H1: Select FEM (p <0.05).The Hausman test is a statistical examination used to choose between the more suitable Fixed Effect or Random Effect model for analysis.If Result:H0: Select REZM (p> 0.05); H1: Select FEM (p <0.05).The Lagrange Multiplier test (LM) is conducted to assess whether the REM outperforms the CEM method.If Result: H0: Select CE (p> 0.05); H1: Select RE (p <0.05).[33]

Fig. 3 .
Fig. 3. Land cover in Karawang Regency 1990, 2000, 2009, 2020, and 2023 to 50 sample points.The distribution of sample points is depicted in Fig. 4. Fifty samples were divided into five classes based on the 2023 land cover map classification results.The largest sample class is the rice field covered with 25 samples.Next are 13 samples that stand for built-up areas, five samples for forests & plantations, four samples for bare lands, and three samples for water bodies & ponds.

Fig. 7 .
Fig. 7. Calculation results with the REM Model

Table 3 .
[18]tral characteristics of bands used for image classification.Thermal Infrared Sensor (TIRS) scenes falling within both Collection 1 and Collection 2[18].NDVI is computed as the ratio between the Red and Near Infrared bands of Landsat Imagery.NDVI produces DN values with characteristics that teow positive values in dense vegetation areas.In contrast, water bodies and developed urban areas usually have values close to zero or negative

Table 4 .
The logic of the degree of urbanization hierarchical level 2

Table 5 .
Distribution of land cover classification each year.

Table 6 .
Accuracy assessment calculation of land use