Abstract
In this paper, we present a land cover map of Ca Mau province in Vietnam, which was generated using three different classification methods: Random forest (RF), Support Vector Machine (SVM), and Gradient Boost (Gboost). The study area is a lowland region with a complex land cover in the Mekong Delta. To create the map, we used multi-temporal Sentinel-2 satellite images from January to December 2021, with 17 images remaining after cloud removal. A mean filtering method was employed to generate a single image for classification. The Google Earth Engine platform was used to perform the classifications, with sample data drawn from the 2014 land use map of Ca Mau province and Google Earth images. We collected nearly 4000 pixels for classification for all three methods, with an additional 3000 pixels used to evaluate accuracy. The SVM method produced an overall error (OA) of 79.5% and a kappa coefficient of 0.72, while the Gboost method achieved 85.6% accuracy with a Kappa of 0.79. The RF method had the highest accuracy, with an OA of 86.5% and Kappa of 0.81. We selected the land cover map generated by the RF method using the most accurate image for a scale of 1:50.000.