Ground users suffer from severe uplink interference originating from high altitude Unmanned Aerial Vehicle (UAV) line-of-sight channels. Using multi-armed bandit, we propose a method aiming to find the best resource block and transmit power level for a UAV dynamically paired with a ground user using Non-Orthogonal Multiple Access (NOMA). It is done according to the UAV's location. It results in mitigating the UAV-uplink interference on its co-channel ground user and maximizing the sum of their data rate in the shared resource block. Performance is evaluated via simulating three explorationexploitation strategies, namely, epsilon-greedy, upper confidence bound and Thompson sampling.