Caracterización de Vegetación de Páramo A Partir de Imágenes Uav Hiperespectrales Y Algoritmos de Aprendizaje Automático - Estudio de Caso Parque Ecológico Matarredonda, Cundinamarca, Colombia

The páramo ecosystem in Colombia is very important due to its high biodiversity and its vital role in regulating water resources. Hyperspectral images carried on UAV platforms offer an opportunity for agricultural and forestry analyses due to their spectral richness (hundreds of bands) and high spatial resolution (centimeter-level), making them a trending topic in the scientific community. This project aims to characterize the páramo vegetation in the Mataredonda Ecological Park, an ecological sanctuary bordering the rural area of Bogotá (capital of Colombia). The project utilizes a 270-band orthomosaic with a spatial resolution of 3.7 cm. This project is divided into two phases. First phase focuses on the spectral characterization of the species present in the study area, which is carried out using an illustrated species catalog and photo-identification through visual photo-interpretation, identifying 13 species present in the image. Second phase aims to spatially map the distribution of species in the study area. This involves, firstly, dimensionality reduction and subsequently, supervised classification. Results shows that the best results are obtained with the Minimum Noise Fraction image and the Random Forest classifiers, achieving an overall precision of 81.6% and a kappa coefficient of 0.8 - with which the distribution of species in the park was spatialized - followed by Support Vector Machine, which achieves an overall precision of 80% and a kappa coefficient of 0.79.