Abstract Mapping of tea plantations is vital for agricultural planning in Thai Nguyen city, Thai Nguyen province, Vietnam. This study applies an improved classification method using Landsat 8 imagery, integrating spectral reflectance, Normalized Difference Vegetation Index (NDVI), and phenological characteristics on the Google Earth Engine platform. Imagery from April and December was utilized to capture seasonal NDVI variations that enhanced land cover classification. In December, tea plantations exhibit lower NDVI values (0.38–0.47) due to dormancy, clearly separating them from evergreen forests (0.49–0.52) and recently harvested rice fields. In contrast, April imagery shows overlapping NDVI ranges among tea, crops, and forests, complicating classification. Using Landsat 8 December images, combining NDVI, spectral reflectance and seasonal ecological characteristics (phenology) in a machine learning algorithm analysis framework significantly improves the efficiency of land cover classification. The classification was performed using Support Vector Machine (SVM)and Random Forest (RF) algorithms, with the RF model achieving 91.3% overall accuracy and a Kappa coefficient of 0.89, outperforming SVM by 2.2%. These findings confirm the effectiveness of combining spectral features and NDVI with the Random Forest algorithm for accurate tea plantation mapping.
Phan et al. (Tue,) studied this question.