The Classification of Land Use and Land Cover was conducted using the Manual Method in ArcGIS and using AI/ML in the Google Earth Engine for Salem District, Tamilnadu, India.

doi:10.21203/rs.3.rs-5088916/v1

This study examines the land use and land cover (LULC) classification in Salem District, Tamil Nadu, India, using both a manual method in ArcGIS and an AI-driven approach in the Google Earth Engine (GEE). The manual method offers high accuracy but is labor intensive, while GEE uses cloud computing and machine learning for greater efficiency and scalability. The study compares the accuracy, efficiency, and applicability of both methods across different scales and temporal resolutions. The results showed that ArcGIS is better for detailed, small-scale work, while GEE excels in large-scale and dynamic analyses. This study identifies the strengths and limitations of each approach and suggests future research directions for environmental monitoring, urban planning, and emerging technologies.

Land Use and Land Cover

Salem

ArcGIS

Google Earth Engine

AI – ML

Remote Sensing

Classification Methods

Environmental Monitoring

Urban Planning

Geospatial Analysis – Conclusion – Future Scope

The land use and land cover (LULC) classification is a critical aspect of understanding and managing natural resources, urban development, and environmental sustainability. Accurate LULC classification is essential for informed decision-making in areas such as agriculture, forestry, urban planning, and environmental conservation. Traditionally, LULC classification has been conducted using manual methods, which, although effective, are often time-consuming and prone to human error. With the advent of advanced geospatial technologies, the integration of artificial intelligence (AI) in platforms such as the Google Earth Engine (GEE) has revolutionized the process, offering faster, more precise, and scalable solutions.

This study focused on the classification of land use and land cover in Salem District, Tamil Nadu, India, by comparing the traditional manual method in ArcGIS with the AI-powered approach in the Google Earth Engine. Salem District, located in the southern part of India, is characterized by a diverse landscape, including agricultural lands, forests, and urban areas. The region's varied topography and land use patterns present a unique challenge for accurate LULC classification.

The manual method in ArcGIS involves the visual interpretation of satellite imagery, relying heavily on the expertise of the analyst. This approach, while detailed, can be labor intensive and subject to subjective bias. On the other hand, the Google Earth Engine leverages cloud computing and machine learning algorithms to automate the classification process, providing a more objective and efficient solution.

This study aimed to compare the effectiveness, accuracy, and efficiency of these two methods in classifying LULC in Salem District. The outcomes of this comparison will offer valuable insights into the advantages and limitations of traditional versus AI-driven approaches, contributing to the broader discourse on the modernization of geospatial analysis in environmental and urban studies.

Study Area

The Salem district is located in Tamil Nadu State, South India. Geographically, the region lies between 11°10′ and 12°10′N and between 77°35′ and 78°55′E. The study area is considered the largest district in Tamil Nadu and covers an area of 6016 km². The boundaries of the study area are marked by isolated hills: Nagaramalai in the northern part, Jarugumalai in the southern part, Kanjamalai in the western part, Godumalai in the eastern part, Shevaroy in the NE part and Kariyaperumal hill ranges in the southwestern part.

Sarabanga, Thirumanimuthar, Sweta and Vasista are the main river systems in the Salem district. The major river in southern India, Cauvery, passes through the study area and supplies water for domestic and agricultural purposes in the district. The Sarabanga and Thirumanimuthar are tributaries of the Cauvery River. Massive mineral deposits such as magnetite, magnesite, bauxite, limestone and soapstone are present in the Salem district. The study area is fast-developing city because of its favorable climate and abundant natural resources. According to census data, the population in 2001 was 30.16 people, and that in 2011 was 34.82 people, which indicates the rapid rate of urbanization in the study area.

Land use land cover (LULC) mapping is essential for understanding environmental changes and land management. Traditional manual LULC mapping involves expert interpretation of satellite images and field surveys, ensuring high accuracy but requiring significant time and effort. In contrast, modern techniques utilize the Google Earth Engine (GEE) combined with artificial intelligence (AI) and machine learning (ML) to automate and scale the mapping process. This approach leverages cloud-based processing and advanced algorithms to quickly classify large areas with improved consistency and accuracy. The following sections delve into these methodologies and compare their processes, advantages, and limitations.

For Manual Method

First, you start by collecting satellite images from the USGS (United States Geological Survey). For this project, data from the Landsat 8 and 9 satellites, which are well known for capturing detailed images of the Earth's surface, were chosen. After selecting the right satellite images, the next step is to download these data to your computer.

Once the data, are available, they need to be preprocessed. Preprocessing involves getting the data ready for analysis, which might include cleaning up the images, correcting any distortions, and ensuring that everything is lined up correctly. After preprocessing, the focus was narrowed to a specific area of interest (AOI) by using ArcGIS software. This means you will be working with just the part of the satellite image that covers the area you are interested in studying.

The next important step is selecting training sets. This involves manually identifying and marking different types of land cover, such as forests, water bodies, or urban areas, in an AOI. These training sets are used in the supervised classification process. Here, a machine learning algorithm, specifically the maximum likelihood algorithm, is applied to the entire image. The training sets were used to categorize every part of the AOI into different land cover types.

After running the classification, an output map that shows the LULC of your area is obtained. This map is initially in raster format, which is grid-based. For further analysis, these raster data were converted into vector format, where the land cover types are represented as polygons.

Next, the area covered by each type of land cover was calculated. This helps you understand the extent of different land covers within your AOI. To ensure the results are reliable, an accuracy assessment. This involves comparing the classified map with actual ground data or higher-resolution images to determine how accurate the classification is.

Finally, if you are interested in how the land cover has changed over time, you can perform a change detection analysis. This step involved comparing the current LULC map with maps from previous years to identify any significant changes, such as deforestation, urban expansion, or changes in water bodies.

This entire process (shown in Fig. 1.2), from data collection to change detection, provides a comprehensive way to map and analyze the land use and cover of an area, which is crucial for environmental management, urban planning, and resource monitoring.

Generated using the Google Earth Engine with AI and ML

The process of creating a land use/land cover (LULC) map using artificial intelligence (AI) methods begins with utilizing the Google Earth Engine (GE) to analyze Sentinel-2 satellite imagery. These high-resolution data are imported into the platform, where artificial intelligence and machine learning techniques are applied to classify the land into different cover types. The classification process involved training models using JavaScript and Python, to ensure that the algorithms accurately categorized the land cover. Once the classification was complete, the area covered by each land type was calculated, and an accuracy assessment was subsequently performed to validate the results. Finally, the LULC map, which visually represents the different land cover types, was exported for further use. This approach leverages advanced AI techniques to create precise and efficient land cover maps.

Advantages and Limitations of Both Methods

Advantages:

ArcGIS with Manually Trained Data
1. High accuracy due to user knowledge and local expertise.
2. Customizable classification tailored to specific needs.
3. Detailed control over the classification process.
4. Ideal for small or medium-sized areas requiring precise, hands-on control.
5. This approach is effective for temporal analysis with high accuracy for specific time periods.
GEE with AI-Generated Scripts:
1. This approach is highly scalable and suitable for large-scale projects, including global analyses.
2. AI-driven automation is effective at reducing the time required for LULC mapping.
3. Consistent classification across large datasets reduces subjectivity.
4. The internet is accessible and offers free access to extensive satellite imagery and tools.
5. This simplifies the classification process, decreasing the need for deep GIS expertise.
6. This approach is useful for dynamic mapping and temporal analysis, allowing for near real-time data use.

Disadvantages:

ArcGIS with Manually Trained Data
1. Subjectivity in accuracy due to dependence on user expertise can lead to potential biases.
2. This process is time consuming and labor-intensive, making it impractical for large-scale projects.
3. Limited scalability and difficulty extending to larger areas without additional effort.
4. Low automation, limiting efficiency for large datasets or repetitive tasks.
5. The steep learning curve for users unfamiliar with a GIS makes it less accessible.
GEE with AI-Generated Scripts:
1. The training data quality is dependent upon the training data quality, and inaccuracies may occur if the data are not representative.
2. The black-box nature of AI models limits the interpretability of the results.
3. There is limited customization compared to manual classification methods.
4. AI and machine learning expertise are needed, which may not be accessible to all users.
5. Internet dependency, with potential limitations in areas with poor connectivity.
6. Usage quotas and computational limits, particularly for free-tier users.
7. Ongoing maintenance, as required for AI models, may need retraining with new data.

Area calculation for ArcGIS generated maps

Classes	2018	%	2023	%	ñ/ò
Water Bodies	74.14	1.23	100.74	1.67	ñ
Trees	1312.62	21.81	1367.45	22.72	ñ
Crop and Vegetation	3520.21	58.50	2949.46	49.02	ò
Buildup Area	1110.39	18.45	1599.1	26.57	ñ
Total	6016.38		6016.75

Implications: An increase in water bodies generally represents positive development, as it may contribute to better water availability for agriculture and urban uses. However, these findings may also indicate potential issues such as waterlogging or flooding in certain areas.

2. Trees

Slight increase in Tree Cover: The area covered by trees increased from 1312.62 hectares in 2018 to 1367.45 hectares in 2023, representing a growth of 54.83 hectares (0.91% of the total land cover). This increase is a positive sign, possibly resulting from afforestation efforts, conservation policies, or natural regrowth.
Sustainability: An increase in tree cover contributes positively to biodiversity, carbon sequestration, and overall ecological balance. This reflects a move toward more sustainable land management practices, which is essential for long-term environmental health.

3. Crops and Vegetation

Significant decrease: The most notable change was the decrease in crop and vegetation cover, which decreased from 3520.21 hectares in 2018 to 2949.46 hectares in 2023, a reduction of 570.75 hectares (9.48% of the total land cover). This substantial decrease could be due to several factors, including urban expansion, land degradation, or a shift in land use practices (e.g., the conversion of agricultural land to built-up areas).
Agricultural Impact: A reduction in crop and vegetation cover may have significant implications for food security, local agriculture, and the economy. Moreover, climate change might also lead to a decrease in natural habitats for wildlife, increased soil erosion, and loss of biodiversity.

4. Built-up area

Substantial increase: The total built-up area significantly increased, growing from 1110.39 hectares in 2018 to 1599.10 hectares in 2023, an expansion of 488.71 hectares (8.12% of the total land cover). This reflects ongoing urbanization and infrastructure development.
Urbanization: An increase in built-up areas is often associated with economic development, population growth, and increased demand for housing and infrastructure. However, this expansion can also lead to the loss of agricultural land, increased surface runoff, and potential urban sprawl, contributing to environmental degradation and a greater urban heat island effect.

Area Calculation for Google Earth Engine-Generated Maps:

Classes	2018	%	2023	%	ñ/ò
Water	57.149	0.94	123.21	2.04	ñ
Trees	1917.179	31.86	2136.77	35.51	ñ
Crops & vegetation	3257.791	54.14	2701.19	44.89	ò
Built Area	784.713	13.04	1055.55	17.54	ñ
Total	6016.83		6016.83

1. Water bodies

Area increase: The area under water bodies increased significantly from 57.149 hectares in 2018 to 123.21 hectares in 2023, representing an increase of 66.061 hectares (1.1% of the total land cover). This increase could be due to the expansion of water reservoirs, the construction of new water bodies, or changes in water management practices.
Implications: This growth in water bodies is beneficial for water resource management and could indicate better availability of water for agricultural and domestic purposes. However, this may also suggest increased precipitation or flooding, which may require further investigation.

3. Crops and Vegetation

Notable decrease: The most significant change observed was the reduction in crop and vegetation cover, which decreased from 3257.791 hectares in 2018 to 2701.19 hectares in 2023, a decrease of 556.601 hectares (9.25% of the total land cover). This reduction could be driven by urban expansion, the conversion of agricultural land into other land uses, or changes in agricultural practices.
Agricultural Impact: A decrease in crop and vegetation areas may negatively impact food production, local agriculture, and rural economies. Additionally, this reduction could lead to habitat loss, reduced biodiversity, and increased soil erosion, particularly in regions dependent on agriculture.

4. Built-Up Area

Significant increase: The built-up area expanded considerably, increasing from 784.713 hectares in 2018 to 1055.55 hectares in 2023, an increase of 270.837 hectares (4.5% of the total land cover). This reflects ongoing urbanization, infrastructure development, and population growth.
Urbanization Trends: An increase in built-up areas is indicative of economic development, industrial growth, and increased housing demand. While this expansion supports economic growth, it also raises concerns about the loss of natural habitats, increased pollution, and the potential for urban

Sprawl. Effective urban planning is necessary to mitigate these impacts and ensure sustainable development.

When comparing LULC map generation via ArcGIS with manually trained data and via the Google Earth Engine (GEE) with AI-generated scripts, both methods offer valuable insights and tools for land cover analysis; however, they differ significantly in terms of their approach, strengths, and limitations.

ArcGIS with manually trained data excels in providing highly accurate and customizable LULC maps, particularly for small to medium-sized areas. This method is ideal when local expertise is available, allowing for detailed adjustments and corrections based on specific knowledge of the region. However, this approach is time-consuming, labor-intensive, and less scalable than other methods, making it less practical for large-scale or dynamic analyses. The accuracy of this method is strongly dependent on the user's expertise, which can introduce subjectivity and potential biases.
The Google Earth Engine with AI generated scripts offers a powerful, scalable solution for large-scale LULC mapping, particularly when working with extensive datasets or conducting temporal analyses. The integration of AI allows for consistent, automated classification across vast areas, making it highly efficient and suitable for dynamic monitoring. However, this method requires a solid understanding of coding and AI, and the results may be influenced by the quality and representativeness of the training data. Additionally, while providing broad consistency, manual methods may lack the fine-tuning and detailed customization possible.

Overall, the choice between these methods should be guided by the specific requirements of the project. For detailed, small-scale studies requiring high accuracy and expert input, ArcGIS with manually trained data is preferable. For large-scale, time-sensitive, or regularly updated analyses, the Google Earth Engine with AI-generated scripts offers unmatched scalability and efficiency. A combination of both approaches might be the best strategy in some cases, leveraging the strengths of each method to achieve a comprehensive and accurate understanding of land use and land cover dynamics.

Future Scope of This Study

The study of land use/land cover (LULC) changes using tools such as ArcGIS and the Google Earth Engine opens up several promising areas for future exploration:

1. Enhanced Environmental Monitoring:

Future research can build on this study to develop more advanced systems for real-time monitoring of environmental changes, helping to quickly identify and respond to issues such as deforestation, urban sprawl, or natural disasters.

2. Improved Urban Planning:

Insights from LULC changes can guide smart urban planning, ensuring that cities grow sustainably, with balanced development that minimizes environmental impact and preserves green spaces.

3. Climate Change Adaptation:

As climate change continues to alter landscapes, this study can provide a foundation for developing strategies to adapt land use practices, such as planning for changing agricultural zones or protecting vulnerable ecosystems.

4. Integration with Emerging Technologies

The methods used here can be integrated with emerging technologies such as artificial intelligence (AI), machine learning, and big data analytics to create more precise and predictive models for land use and environmental management.

5. Policy Formulation and Assessment:

This study provides a basis for evaluating the effectiveness of current land use policies and can inform the creation of new policies that promote sustainable land management practices.

6. Biodiversity Conservation:

By understanding how land cover changes affect habitats, future research can contribute to strategies that protect biodiversity, particularly in regions experiencing rapid development or environmental stress.

7. Global and Regional Studies:

The methodologies can be scaled to global or regional studies, allowing comparisons between different areas and the identification of global patterns in land use changes.

8. Education and Capacity Building:

The findings and techniques from this study can be used to educate and train professionals and policymakers in effective land use management and foster better practices at the local, national, and international levels.

9. Cross sectional Applications:

In addition to environmental science, the results of this study can be applied in sectors such as agriculture, water resources, disaster management, and urban development, supporting a wide range of societal needs.

Overall, this study has the potential to significantly impact how land use is managed and understood, leading to more sustainable and resilient communities in the face of ongoing environmental changes.

Rahmi KIN, Ali A, Maghribi AA, Aldiansyah S, Atiqi R (2023) Monitoring of land use land cover change using Google Earth Engine in urban area: Kendari city 2000–2021. Department of Geography, FMIPA, Universitas Indonesia, Kota Depok, Jawa Barat, Indonesia
Arulbalaji P (2019) Analysis of land use/land cover changes using geospatial techniques in Salem district. Tamil Nadu, South India
Zhao Z, Islam F, Waseem LA, Tariq A, Nawaz M, Islam U, Bibi I, Ur Rehman T, Ahmad N, Aslam W, Raza RW, D., Hatamleh WA (2023) Comparison of Three Machine Learning Algorithms. Using Google Earth Engine for Land Use Land Cover Classification
Yang L, Driscoll J, Sarigai S, Wu Q, Chen H, Lippitt CD (2023) Google Earth Engine and Artificial Intelligence (AI): A Comprehensive Review

The authors declare no competing interests.

The Classification of Land Use and Land Cover was conducted using the Manual Method in ArcGIS and using AI/ML in the Google Earth Engine for Salem District, Tamilnadu, India.

Status:

Version 1

Abstract

Figures

Introduction