Semantic Environmental Trajectories of Territories (SETT) represent the environmental evolution of territorial units over time by combining their spatial, temporal, and semantic information. In TRACES project, these territory trajectories are represented in Knowledge Graphs, where each territory is characterized by several thematic trajectories such as water, temperature, vegetation, snow, and urbanization. This representation is rich and expressive through KGs, but it holds many challenges in graph representation learning, as the data includes temporal, spatial, explicit relations and measurements in each thematic trajectory. This report studies how Graph Neural Networks (GNNs) can be used to learn meaningful representations from SETT Knowledge Graph.