DF verification study

Best all,
I am new to Radiance but I am using it as an executable from C++ in my thesis to evaluate rooms using the Daylight Factor (DF). I am trying to verify the DF however I haven’t succeeded.

In most cases, my ADF is +/- 1,3x too high.
I am not sure why my values vary so much.
If anybody knows what parameters or settings I should check, all the help is welcome.
Below I will present my results and process.
PS: I already tried more “accurate” rtrace settings, such as: -I -h -ab 5 -ad 2048 -as 512 -aa 0.1 -ar 256 -oov. However, this resulted in my computer running for almost 5 hours, instead of 20 seconds. The results remain comparable (0.05 difference of ADF).

Model setup:
First, I am creating materials with the corresponding reflectance values.
I am using polygons to create the 3D scene: scene.rad
my Gensky settings are:
. 3 21 12:00CET -a 52.2 -o -4.5 -c -B 55.865 -g 0.2 > sky.rad
. (55.865 = 10000/179)
with oconv I compile the scene:
. oconv sky.rad scene.rad > room.oct
my Rtrace settings are:
. -I -h -ab 4 -ad 256 -as 32 -aa 0.9 -ar 256 -oov < gridfile > results.txt
I calculate the illuminance values then with rcalc:
. -e ‘$1=179*(.265*$4+.670*$5+.065*$6)’ results.txt > lux.txt
. (resulting in Epoint, per point specified in the gridfile)
Then I calculate the DF: (Epoint/Eout)*100% = (Epoint/10000)*100%
Then I calculate the average of all points to derive the ADF.

I modelled the following rooms (vD) of: Compare effect of window placement
Room x: room name: ADF study : ADF my result : percentage difference
Room1: A4 : 7,2 : 11,69 : 162%
Room 2: A3 : 2,2 : 2,62 : 119%
Room 3: C2 : 2,9 : 4,4 : 152%
Room 4: B3 : 0,7 ; 0,54 : 77%

Furthermore, I am also comparing rooms from this paper (VK):
Salah Vaisi, Farzam Kharvari,
Evaluation of Daylight regulations in buildings using daylight factor analysis method by radiance
Room a = block 2, horizontal window
WFR: ADF study : ADF my result : percentage difference
WFR 12: 2,47 : 3,75 : 152%
WFR 15: 3,43 : 4,62 : 135%
WFR 18: 3,85 : 5,47 : 142%
WFR 24: 5,03 : 7,12 : 142%

And I also modelled a simple room in Velux Daylight Visualiser:
This allowed me to control more settings. Still with no correct result.
5x5x3 room, with window of 3x2 centrally placed in the wall.
ADF study : ADF my result : percentage difference
3,86 : 5,22 : 135%

Hi @quimara,

Welcome to the Radiance discourse forum. I’m quite rusty with daylight factor (we don’t use it anymore where I’m from), but I understand that it is defined based on an idealized unform sky (which doesn’t exist in reality). I don’t think your gensky settings using -c will achieve a uniform sky. Instead, you should use gensky with the -u setting or model your sky as a source without using gensky at all.

Hi all,

for DF calculations, the -c option is correct. It gives you an idealized overcast sky according to CIE, which is the used sky definition for daylight factor simulations.


My mistake, @David_Geisler-Morod1, thanks for the correction.

@quimara, I would also recommend that you check the number of ambient bounces in your simulations compared to the references you chose. The ab setting can have a significant effect on measured indoor illuminance as you move from 1 bounce to whatever limit is established by your -lr parameter. My personal recommendation is to use -ab 6 with rtrace and -ab 5 with rpict, and you’ll get lower illuminance readings with smaller -ab values. It looks like you tried increasing -ad already, which is also a good idea, but the effect of that won’t be so noticeable because you are taking an average of many points.

If your grid points are facing upwards on the work plane, and you have side lighting in your space, then results accuracy will depend heavily on the -ad and -as settings as a significant portion of the light comes from low angles through the window. As you are averaging together the results, you might get better accuracy with a strict MC path tracing, so try the following settings in rtrace:

-aa 0 -ab 6 -ad 10240 -as 2560 -lw 1e-6 -n 4

You can adjust the -n option to the number of processor cores available and your calculation should be reasonably fast. If there is still a large systematic error, like you are seeing in your VK paper comparisons, then I would check the surface reflectances, exterior description, and geometry very closely to see that it matches the publication. (Also, the -as option sometimes causes bias, so you can try -as 0 and see if that helps.)

By the way, which version of Radiance are you using. I.e., what does “rtrace -version” say?


P.S. The only gensky settings that matter in DF calculations are -c and -B. The rest you can leave off, since they only set the zenith brightness in the absence of a -B or -b option when you have a sunless sky.