Wildlife conservation efforts are constrained by a limited amount of resources available for surveillance activities. UAVs are used increasingly to assist rangers in patrol tasks. Effectively patrolling wildlife parks requires detailed knowledge of the environment and its threats, which is not always available. Previous work in Green Security Games (GSGs) that aims to develop defensive strategies to deter adversaries relies on historical poaching data to train machine learning models. Recent advancements in the field have led to the development of an online learning framework that does not require prior data. However, the defensive strategies resulting from this approach are focused on foot patrols by rangers, which do not have the same mobility as UAVs, or do not take into account spatio-temporal constraints associated with patrolling in a real-world situation at all. To address the desire of using UAVs for wildlife surveillance, this paper proposes MEOMAPP, a model that extends on the online learning approach by incorporating a patrol planning algorithm more suitable for aerial patrol. It also includes an evaluative algorithm that considers a human expert next to the online learning expert and balances the application of their strategies based on the observed performance of each expert. By simulating MEOMAPP in a realistic environment, the research demonstrates that the model is suitable to determine aerial surveillance strategies for wildlife conservation.