Hello Philip.
Unfortunately, scaling fire behavior gets complicated quickly. While froude based scaling does a reasonably good job of reproducing the temperatures in an open burning fire plume, it doesn't work well in under ventilated compartment fires. There are two main issues in this approach for compartment scaling.
First, time scales with length ratio to the 1/2 power. If you are primarily interested in the steady state response this might be ok. If you are interested in the transient response (time varying source term, secondary ignition of combustibles, time to flashover, etc), you need to do extra leg work to design your reduced scale experiment to account for the impact of time scaling.
The other issue is related to the flow of air through the door. The velocity according to froude scaling also scales with the length ratio to the 1/2 power. This works out if you are able to scale time correctly, but generally leads to issues experimentally since getting surrogate materials correct at the reduced scale is difficult. This generally leads to the equivalence ratio of the compartment not scaling correctly.
We recently completed a study looking at froude and surface energy based scaling approaches for fully developed fires, see below. We found for a post flashover fire that the surface energy approach with relaxed geometric similarity at the door provided better agreements of gas temperatures between the reduced scale and full scale. While the work in this report was based on simulation results, we recently completed a follow up experimental study and achieved similar results. The results of that study should be coming out in a few months.
To your original question of which scale to model, that depends on what the objective of the modeling is. Since it sounds like you are primarily interested in the full scale behavior, I would recommend running two models first before running any sort of parametric study. First, model the experiment at the scale it was conducted. This will allow you to verify your model configuration and better understand the model uncertainty in your application. Second, model the full scale using the scaling approach that you think is most appropriate. This will allow you to better understand the capabilities of the scaling approach to scale the specific quantities of interest in your study. If you only model the full scale, you will not know if the differences between the model and experiment are due to the scaling or model uncertainty.
Thanks,
Jonathan