Data Assimilation for Agent-Based Modelling: An Implementation of the Ensemble Kalman Filter
Keiran Suchak, Nick Malleson and Jonathan Ward
One of the most common approaches to modelling how people move around their environment is agent-based modelling. Whilst there exist widely used methods for calibrating model initial states and parameters, there is a shortage of methods by which we can incorporate up-to-date data when simulating systems in real-time. This piece of work aims to demonstrate how we can borrow data assimilation techniques from the field of numerical weather prediction - specifically the Ensemble Kalman Filter - to combine the knowledge represented by an agent-based model of pedestrian movements with observations as they become available.