Comparison of machine learning models for mapping Arecanut based agroforestry system in Goa by enhancing precision...

Area of Science:

• Agricultural Science
• Remote Sensing
• Environmental Science

Background:

• Agroforestry is crucial for climate change mitigation and rural livelihoods, especially in coastal zones.
• Traditional methods struggle to map complex agroforestry systems like arecanut-based ones.
• Accurate mapping is vital for effective land use planning and resource management.

Purpose of the Study:

• To identify and map arecanut-based traditional agroforestry systems in Goa, India.
• To evaluate the performance of machine learning models for agroforestry mapping.
• To assess the efficiency of Sentinel-2 satellite data in this application.

Main Methods:

• Utilized Sentinel-2 satellite imagery for mapping arecanut-based agroforestry.
• Employed machine learning models: Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM).
• Applied Boruta algorithm for feature selection to enhance model accuracy.

Main Results:

• Gradient Boosting Machine (GBM) achieved the highest accuracy (0.86 overall, 0.83 kappa).
• Boruta analysis identified NDWI2, B3, and SLAVI as key variables for mapping.
• GBM mapped 58.64 km² of agroforestry; the average of three models estimated 45.1 km² for Goa.

Conclusions:

• Machine learning algorithms, particularly GBM, are highly effective for mapping agroforestry systems.
• Accurate mapping supports land use planning, resource conservation, and coastal management.
• Future research with hyperspectral sensors could further refine these mapping techniques.