Development and Application of Coconut Vegetation Indices (CVIs) for Rapid and Accurate Coconut Mapping Using Sentinel-2 Images: A Case Study of Quezon Province, Philippines

Abstract. The existing methods for coconut mapping in the Philippines and globally are complex, necessitating the development of a simpler yet rapid and accurate classification technique. This study introduces the first spectral index for coconut mapping. Two Coconut Vegetation Indices were developed: one for dense Coconut Vegetation (CV) and another for sparse CV. CVI_dense utilizes three Sentinel-2 bands in its equation (NIR1-SWIR1)/(SWIR1-SWIR2) to map coconut areas with densities >2.25 x10^6 sq.m. per 1km pixel. Meanwhile, CVI_sparse incorporates four spectral bands in the equation (NIR1-Red)/(SWIR1-SWIR2) for areas with densities ≤ 2.25 x10^6 sq.m. per 1km pixel. The formulation of these indices is primarily based on previous studies involving band combinations and the analysis of spectral separability of the acquired coconut reflectance data. The extent of coconut vegetation was mapped using CVI_dense with a minimum threshold of 1.094, while CVI_sparse was applied using a threshold range of 0.4774 to 1.094. The Balanced Accuracy (BA) metric was used to assess the accuracy, accounting for the imbalanced reference data between coconut and non-coconut classes. CVI_dense proved highly effective with User’s Accuracy (UA) of 80%, Producer’s Accuracy (PA) of 88.89%, and BA of 88.90%, surpassing CVI_sparse, which had accuracies of 32.00% (UA), 53.33% (PA), and 74.10% (BA).