TY - JOUR AU - Hien Le AU - Luan Pham AU - Tuyet Hoang AU - Toan Dinh PY - 2022/06/30 Y2 - 2024/03/29 TI - Land-cover classification using Random Forest and incorporating NDVI time-series and topography: a case study of Thanh Hoa province, Vietnam JF - Science & Technology Development Journal: Science of the Earth & Environment JA - STDJSEE VL - 5 IS - S3 SE - Original Research DO - https://doi.org/10.32508/stdjsee.v5iSI2.681 UR - http://stdjsee.scienceandtechnology.com.vn/index.php/stdjsee/article/view/681 AB - Land cover/land use (LULC) mapping in the complex land cover area is a challenging task due to the mixed vegetation patterns, and rough mountains with fast-flowing rivers. Therefore, a new technique should be applied to improve the accurate classification of complex LULC. In this study, we applied a supervised machine learning approach to map land use in Thanh Hoa province, Vietnam utilizing multi-temporal Normalized Difference Vegetation Index (NDVI) data from MODIS, combined with topographic features. We used distinctive temporal features of land cover in 2015 as response variables and developed fifteen engineering features as predictors for automatic prediction. Then, we trained Random Forest classification (RFC) and conducted repeated cross-validation to identify the optimal RFC with the highest robustness on test data. RFC reached a total prediction accuracy of 91 % and Kappa coefficient (K) of 0.89 across eight different land covers including bareland, crops, rice paddy, forest, mangrove, urban and built up, grassland, and water. Besides, the results showed that the features extracted from time-series NDVI comprising the mean of yearly NDVI, the sum of NDVI, and the topography were the relative importance variables controlling the land cover classification. ER -