Optically pumped magnetometers have opened many possibilities for the study of human brain function using wearable moveable technology. In order to fully exploit this capability, a stable low-field environment at the sensors is required. One way to achieve this is to predict (and compensate for) the changes in the ambient magnetic field as the subject moves through the room. The ultimate aim is to account for the dynamically changing noise environments by updating a model based on the measurements from a moving sensor array. We begin by demonstrating how an appropriate environmental spatial noise model can be developed through free-energy-based model selection. We then develop a Kalman-filter-based strategy to account for the dynamically changing interference. We demonstrate how such a method could not only provide realistic estimates of interfering signals when the sensors are moving but also provide powerful predictive performance (at a fixed point within the room) when both the sensors and sources of interference are in motion.