Buried communicating sensor nodes are increasingly used in Internet of Underground Things to measure and transmit soil moisture and temperature information. This is particularly the case in agriculture to remotely monitor soil conditions and optimize cultivation interventions. However, as the radio waves are significantly more attenuated in the soil in comparison with in the air, the data collection of these nodes is problematic due to their very short-range communications. This paper focuses on the use of Unmanned Aerial Vehicles (UAVs) to collect the data. By flying at low height, UAVs can in fact come close to the nodes to collect the stored measurements. That requires however to consider the obstacles in the path planning algorithm and solve the Close Enough Traveling Salesman Problem (CE-TSP) in case many nodes need to be visited. To answer these problems, this paper first establishes the normalized Digital Surface Model (nDSM) based on an airborne High Density LiDAR sensor. This model is used to study the optimal path for the UAV having to visit a large number of nodes. To that end, the cost matrix defining the Euclidean distances of the trajectories connecting each pair of nodes is established. If one or more obstacles are detected between two nodes, two strategies are evaluated: the first one studies the possibility to fly above obstacles, the second one bypasses obstacles on the left or right sides. From the knowledge of this cost matrix, a Genetic Algorithm (GA) completed with heuristic rules is finally applied to solve the CE-TSP. The results are presented on a real use case, considering 100 buried sensor nodes distributed in our experimental farm and considering the real maximum time flight of our UAV. Theses results clearly highlight the relevance of the methods proposed and open the way for increased surveillance of underground spaces in agriculture.