Recently there has been interest in the design of wind farm control systems that can coordinate individual turbine controllers to improve global plant performance. This improvement comes from accounting for the way in which turbines interact through wakes. Often however, controllers are designed assuming steady and known environmental conditions, without turbulence or wake meandering. This raises the concern that these methods will fail to perform well in practice because it could be difficult to apply methods based on steady wakes to a situation where wake locations are changing and not measurable. In this paper, a particle filter is used to continually estimate the wake locations in a stochastic setting by combining all of the available turbine measurements. The design of the algorithm is documented, and is shown to employ sensors that are available on modern turbines. Using a high-fidelity wind farm simulator, we show the effectiveness of the proposed framework using several multi-turbine scenarios and compare the wake locations predicted against the wakes observable in flow-field slices taken from the simulator output.