Mapping of Land Cover with Optical Images, Supervised Algorithms, and Google Earth Engine.

Area of Science:

• Agricultural Science
• Remote Sensing
• Environmental Monitoring

Background:

• Climate change and human activities pose risks to crop production and ecosystems, impacting food security and biodiversity.
• Developing strategies to maximize agricultural output while minimizing environmental degradation is crucial for sustainable land management.

Purpose of the Study:

• To develop and evaluate a strategy for identifying and mapping sorghum and corn crops, alongside land use and land cover.
• To assess the effectiveness of machine learning algorithms for crop and land-use classification in a tropical agricultural area.

Main Methods:

• Utilized Sentinel-2 satellite images and spectral indices for vegetation and water body detection.
• Employed machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), and Classification and Regression Trees (CART).
• Classified land use into water bodies, land in recovery, urban areas, sandy areas, and tropical rainforest over two growing seasons (2017-2019).

Main Results:

• Support Vector Machine achieved the highest overall accuracy (0.99%) and Kappa index (0.99%), outperforming Random Forest and CART.
• The Support Vector Machine model demonstrated the lowest false positive rate and margin of error.
• High accuracy was obtained in classifying soil types and identifying specific crops like sorghum and corn.

Conclusions:

• The proposed strategy using Sentinel-2 imagery and machine learning, particularly SVM, is highly effective for accurate crop and land-use mapping in tropical environments.
• Precise mapping supports sustainable agricultural practices, enhances food security, and aids in environmental monitoring and resource management.
• The study highlights the potential of automated remote sensing techniques for agricultural and land-use assessments.