I am running some annual daylighting (DA/sDA) calculations for a small floor plate. The model includes context, ground plane and appropriate materials (RAD File Verified). I am finding that across simulations (toggling the T/F into the “Run”) the results for sDA can vary by as much as 7-8%. I have “reload old results” set to false, which should start a fresh simulation and clear old results. I have been troubleshooting and testing the following modifications and their effect on the consistency of the results:
Radiance Parameters
Number of Workers
The Model Name Input
What I have found is that the results seem to be more consistent (within ~1% variation) when I change the model name from run to run (therefore generating a fresh HB Model), however I am not sure how this should affect the result (in such a big way). See the table below of my troubleshooting results and documentation.
Hi @johnsloane, can you share your model with us to be able to recreate this on our side.
The output folder gets cleaned at the start of each run and changing the number of workers should not change the results as much. Also the choice of some of the radiance parameters are not really appropriate for this study. There is no caching and aa will automatically be set to 0.
A higher value of ad should give you a more accurate results. You should also add a small value for lw. You can start with 1/ad. In some case I have seen Greg suggests a value of 1/(ad^2). See the discussions on the Radiance forum for more information.
My experience has told me that it’s usually the fault of the radiance parameters in a case like this. For a model with such a wide range of detail (from a window sill to a skyscraper), you should definitely be making some changes to the parameters. The key ones that matter for annual daylight are -lw and -ad as Mostapha says.
