Forest tree species identification and distribution mapping using multi-temporal sentinel-2 imagery

Abstract

The forests of Uganda are home to a wide variety of plant and animal species, and the country is renowned for its astonishing biodiversity. More than 1,000 tree species and over 5,000 plant species, including rare and endemic species, are thought to exist in the nation's woods. The management and conservation of tree species diversity in Uganda has proven to be challenging due to the complexities and uncertainties associated with the methods employed to determine attributes of forest tree species. These methods are often time-consuming, laborious, and prone to inaccuracies. This study aimed at assessing the tree species distribution in the forest for two epochs, while examining the phenology characteristics for the forest tree species and also evaluating the changes in area covered by different tree species. Geolocation samples of various tree species within the forest were gathered to serve as training and validation data for the model. The collected field data was divided, with 80% used for training the model and 20% for validation purposes. A methodology was adopted to perform tree species classification that involved combining Sentinel-1 and Sentinel-2 data, along with machine learning techniques. The selection of image dates for the classification process was based on the phenological outcomes, which highlighted the most suitable seasons in the forest's phenological cycle. Classified maps were generated for two distinct time periods, specifically 2019 and 2022, resulting in overall accuracies of 79.17% and 75.0% respectively. These maps visually represented the distribution patterns of various tree species within the forest for both years, and statistical analysis was conducted to calculate the area coverage of each species. Based on the results obtained from this study, it can be inferred that this approach offers a reliable method for mapping species distribution with increased accuracy, reduced time requirements, and wider spatial coverage compared to traditional methods that have been conventionally employed.