Smart farming concerns the usage of autonomous robots and sensors in the field to implement agro-ecology practices. Agriculture stake-holders are supported by supervision and control systems that allow for monitoring real-time data by means of Data Stream Management Systems (DSMSs), and remotely control robots in the field. However, processing in real-time streaming trajectory data that come from autonomous robots presents significant challenges due to the large volume of data generated by robots, and their bad quality since communication networks deployed in rural area can present several problems (e.g., instability and congestion). Furthermore, existing lightweight DSMSs do not support spatial data, and the proposed supervision systems are based on cloud-fog architectures, which do not provide effective solutions for trajectory data stream analysis within bad quality communication network. Therefore, in this paper, we extend [14], by proposing a hybrid edge-fog architecture and computation framework for robotic trajectory data analysis, which combines distributed and non-distributed DSMSs to manage trajectory data more efficiently. In particular, we extend the lightweight DSMS Esper with spatial operators, and dynamic frequency mechanism, at the edge level to process complex queries over trajectory data stream of the robots. Our architecture uses Geoflink at the fog, i.e., in the farm. A distributed computation approach for complex queries has therefore been implemented to split queries over the edge and the fog. We validate our proposals using experiment in real-life conditions.