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

Agroforestry plays a pivotal role in mitigating climate change and supporting rural livelihoods. Especially in coastal regions, it aids soil conservation, provides biophysical protection, and diversifies income sources. Through conventional approaches, mapping the arecanut-based agroforests is difficult. This study utilised machine learning models to identify arecanut based traditional agroforestry systems using Sentinel-2 satellite data in Goa, India. A total of 374 non-agroforestry and 70 agroforestry locations were collected for model training and testing. Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM) were employed. The Boruta algorithm was applied for feature selection and to improve the model accuracy. The findings showed that the GBM model had significantly higher overall accuracy (0.86) and kappa (0.83) scores than other models during validation. The Boruta analysis revealed NDWI2, B3, and SLAVI as important variables indicating the importance of water indices and vegetation parameters in mapping agroforestry. The GBM model achieved high accuracy in identifying arecanut based agroforestry region amounting to 58.64 km² followed by the RF with 53.36 km² and the SVM with 23.21 km². Using the average of three models, the estimated area under arecanut based agroforestry in Goa was 45.1 km². The area of applicability (AOA) analysis based on GBM revealed that 97.32% of the total geographic area of Goa was inside AOA indicating better distribution of collected ground truth data. This paper demonstrated efficiency of machine learning algorithms for accurate mapping of agroforestry systems to support land use planning, resource conservation and management in coastal environments. Subsequent studies utilizing hyperspectral sensors can improve the efficiency of machine learning-based agroforestry mapping methods.