What are the applications for five-phase method with high sun patches?

I do a test that compare the glare values derived by 5-PM with different sun patches and the ones calculated by conventional raytracing method, and found that when the sun patches was above 1278, the error of 5-PM have no signifcant change. Thus, it is sufficient accurate to calculate the glare metric for the 5-PM with 1278 sun patches. So I want to know if some applications require scenario for 5-PM with high sun patches (MF:4/5/6).

My results are presented here, I test two complex fenestration systems, which are the micro prismatic redirection fenestration and the spcular blind.

Errors of glare metrics for PDRF

Errors of glare metrics for specular blind

It is important to add to your claim: “Thus, it is sufficient accurate to calculate the glare metric for the 5-PM with 1278 sun patches.” for this particular scene and view points. Whether that is generalisable takes a bit more thought.

Higher sun patches reduce the angular error introduced by the gridded sun positions. Whether that angular error translates into photometric error depends on the complexity of your scene. Do you have mullions? overhangs? distant site obstructions? Partial height interior partitions? Is your CFS a micro structure (less sensitive), or does it cast discernible shadows (more sensitive)? Is it modelled with geometry (most sensitive), a tensor tree bsdf (more sensitive)? or klems (less sensitive)? If a tensor tree what is its resolution 2^5 (less sensitive)? 2^7 (more sensitive)?

You can see this with your specular blind vs. PDRF, with the clumps of misses that you’ve highlighted, as well as the DGP=1 values that differ between 5PM and RT. According to your plots, the number of these misses does go down as MF goes up, but because these are a small portion of your data this does not have a big impact on your summary error metrics. As you do, looking at error by category is a great way to isolate these types of errors from the bulk of conditions when direct sun is not present or near present in the field of view. In the limit, as MF gets even larger, these geometric errors would disappear, leaving only the differences in modelling procedure between you comparison.

As a broader aside:
How much does this matter? and what is the comparison to your reference actually saying? Assuming you are using hourly data, each point in time RT simulation is representative of 1 hour of solar transit, during which the sun will move ~15 degrees, across years, the exact location may vary by up to 0.5 degrees. So in some way what your result is showing is the limits of your reference data, as at MF:3 you have sufficient resolution to capture the hourly data. By MF:5 you are already calculating nearly as many sun positions as reference data, so unless you are testing multiple orientations and/or latitudes, using the exact sun positions starts to become more efficient.


Hi, Stephen

Thanks for your insight!

In my model, the tensor tree (2^6) BSDF was used for PDRF, and exact geomerty for specular blind.
The PDRF was insensitive to the solar position because it has no direct view of the sun. For specular blind, the outilers that deivates significant from RT simulated results was due to the fact that:
1.The inexact solar position in 5-PM leads to the moment the sun presented in the field of view mismatch the moment in the RT simulation.
2.The moment the specular reflected light occured was not consistent with the moement in RT simulation.

However, the probability that these two conditions occur is low, this leads to the error being not significant, especially in the glare rating based on the thresholed values. I also counted the frequency of the four glare levels based on the 5-PM and RT simulation glare metrics, and found that even for MF:1, the difference between 5-PM and RT simulation was not obvious. So I present the above question.

Actually, I used rpict to calculate the direct sunlight contribution based the exact solar position at each timestep instead of the sun-coefficient method, and obtain the results that were very consistent with RT simulation. But I thought if it was essential to do that work. Maybe I should test serval complex fenestration systems to see if the inexact sun position’s impact on glare calculation result would be enlarged. Below is the result of the modified 5-PM compared with RT simulation, let’s call it M5-PM for short.