Land Use and Land Cover Classification in Odisha Coastal Zone: A Remote Sensing Big Data Processing Practice Using Sentinel-2 and Random Forest

This article introduces the LULC_ODISHA project, which targets the coastal zone of Odisha, India (47,054 km²), using Sentinel-2 remote sensing imagery and balanced random forest algorithm to achieve automated land use and land cover classification, processing over 47 million spatial pixels. Developed by GOURGOPAL618, the project is open-sourced on GitHub (link: https://github.com/GOURGOPAL618/LULC_ODISHA) and was released on June 10, 2026. Its technical pipeline provides a reference example for large-scale remote sensing data processing.

Land Use and Land Cover (LULC) classification is a core application of remote sensing, crucial for environmental monitoring, urban planning, agricultural management, and ecological protection—especially in ecologically sensitive coastal zones. The coastal zone of Odisha, India, faces pressures such as urbanization expansion, agricultural land changes, and wetland degradation, which urgently require high-precision LULC data to support scientific decision-making. The LULC_ODISHA project develops a high-throughput geospatial processing pipeline, providing a reference for large-scale remote sensing processing.

Data Source: Uses the European Space Agency (ESA) Sentinel-2 Multispectral Instrument (MSI) Level-2A surface reflectance products, with advantages including:

• Spatial resolution: 10m (visible/near-infrared), 20m (shortwave infrared), 60m (atmospheric correction)
• Revisit cycle: 5 days (dual-satellite network)
• Spectral coverage: 13 bands (visible, near-infrared, shortwave infrared)

Processing Scale:

• Spatial coverage: Approximately 47,054.3 km²
• Number of pixels: Over 47 million (original matrix: 25604×18412)
• Grid resolution: 10m×10m

The project addresses large-scale data challenges through an innovative block processing strategy.

Classification Engine: Balanced random forest algorithm, featuring anti-overfitting, support for feature importance evaluation, and optimization of class imbalance issues.

Block Processing Strategy:

1. Grid blocking: Divide the original image into 512×512 pixel tiles to balance computational efficiency and memory usage;
2. NoData optimization: Discard deep sea/unmapped areas via np.all(block == 0) to reduce computational load by half;
3. NVMe cache: Write prediction results to high-speed storage in pages to avoid memory overflow.

Band Applications:

• Visible bands (B2 blue, B3 green, B4 red): Identify built-up areas, sandy formations, and coastal urban layouts;
• Near-infrared band (B8 NIR): Capture chlorophyll absorption peaks to identify delta mangrove groups;
• Shortwave infrared band (B11 SWIR): Distinguish muddy waterways, wet farmland, and exposed riverbeds.

Classification Results: A total of 8 land cover types, with area statistics as follows:

LULC Category	Area (km²)	Coverage
Bay of Bengal/Deep Water	17339.95	36.85%
Paddy Fields/Arable Land	12976.33	27.58%
Agricultural Wetlands/Mixed Areas	12784.59	27.17%
Rivers and Inland Waterways	1527.20	3.25%
Coastal Beaches/Spits	1277.09	2.71%
Mangroves/Dense Coastal Forests	796.56	1.69%
Residential Areas/Urban Built-up Zones	283.14	0.60%
Fallow Land/Open Areas	69.43	0.15%

The region is dominated by water bodies (36.85%) and agricultural land (54.75%), with small proportions of natural vegetation and built-up areas.

QGIS Field Verification: Classification results are verified via QGIS and real-time Earth observation basemaps; the boundary division of Chilika Lake (India's largest brackish lagoon) highly matches the actual terrain.

Data Governance: Includes DATA_GOVERNANCE.md recording sensor radiometric calibration traceability, and stac_catalog.json providing spatio-temporal database specifications to ensure data quality and reproducibility.

Technical Insights: Memory management (blocking + NVMe cache), computational optimization (NoData filtering), band selection, and verification strategy.

Regional Applications: Supports coastal zone management (beach erosion monitoring), agricultural planning (arable land distribution), ecological protection (mangrove monitoring), disaster assessment (flood mapping), and urban planning (built-up area expansion).

Outlook: The project framework can be ported to other coastal zones; open-source sharing promotes the development of remote sensing applications, providing an excellent practice example for environmental monitoring and other fields.