INTEGRATION OF OBJECT-BASED CLASSIFICATION USING SENTINEL-2 IMAGERY AND IN SITU DATA TO IDENTIFY AND MAP FOREST FACIES IN THE HUMID AND HYPER-HUMID TROPICAL FOREST CONTINUUM OF TAI NATIONAL PARK, SOUTHEASTERN COTE DIVOIRE

Satellite remote sensing, particularly through high resolution satellite imagery, is a vital tool for mapping and sustainably managing tropical forests. While significant progress has been made through its application in various regions around the world, Tai National Park one of the last major tracts of primary forest in West Africa remains relatively undocumented using this approach. Existing data for this site are limited to outdated maps based on aerial photographs and a binary forest non forest classification, which fails to capture the diversity of its vegetation formations. In this context, the present study aims to update the forest formation map of the park and to propose a typology of ecological facies based on the physiognomy and heterogeneity of the environment.The data used consists of six Sentinel-2 scenes acquired in January and May 2020, pre-processed to provide surface reflectance values. Additionally, 9,287 observation points were collected in the field between November 2019 and April 2020 along 293 transects of 2 km each, in order to document the physiognomic characteristics of forest formations. These data were used both to train the classification algorithm and to validate the results. Object based image classification was carried out using the ORFEO TOOLBOX library in several steps: image segmentation using the LargeScaleMeanShift module, which generated 459,268 homogeneous spatial entities; training of the SVM algorithm with the Train Vector Classifier module; and classification of the segmented image using the VectorClassifier module. Validation was performed through confusion matrix analysis.The results identified four major forest formations: open-understory forest (61%), closed-understory forest (30%), forest on hydromorphic soils (6.9%), and shrublands or non-woody vegetation (0.02%).A grid-based analysis of spatial heterogeneity further enabled the mapping of eight forest facies, revealing ecologically significant transitional zones.Ultimately, this study confirms the effectiveness of object-based classification applied to satellite imagery for accurately mapping forest typologies in Tai National Park