This is an experimental deployment of the FLEXPART air transport modeling system, running four times daily (every six hours) in the AWS EC2 environment.
In general the model takes forecasted meteorological input and applies it to MODIS satellite detected "hot spots" in a specified region, and calculates where particles emanating from the hot spots (smoke) will be transported under the forecasted weather.
The assumptions for the simulation are many, so I consider this to be a qualitative rather than quantitative simulation (you might notice there are no units on the colourbar).
Meteorological data comes from the US National Weather Service GFS 0.5 degree model output (available four times daily).
The FIRMS data is, roughly, a satellite-measured indication of radiance in a specific set of frequencies. This means that
Clouds and trees can block indications of fire
Things other than fires can emit detectable radiances. There are some notable power plants in Europe that indicate constant emissions, just because of their heat.
Although I've configured the model to produce output in three layers, the one that's shown in this graphic is the bottom layer, which goes from the surface to 300 meters AGL.
Under the assumption that previous model runs are available, a simulation starts with the smoke concentrations that were available from a previous simulation, to provide some continuity. This can get extremely noisy after a couple of days, so I "kill" particles after they've been around for a few days.