General understanding of radiance / interpolation in rtrace


#1

Hi Radiance community!

I am an indirect user of Radiance, meaning that I have only used it through Honeybee, and unfortunately I think I lack some basic understanding about how things work. My question is probably simple, but I got a bit lost with all the information on the Radiance website and I thought that maybe someone can explain it to me in an easy way. :smiley:

I want to ask about the interpolation in rtrace. I want to calculate radiation on grid points that I defined on a building facade (specifically, I have 560 points). I have noticed that when I turn the interpolation off by setting aa to 0 the time reduces rapidly, but I get an error. Then, with aa 0 if I reduce the lw to something close to 0, I get accurate results in a very short time. Do you think that this makes sense? Or am I doing something wrong? I lowered the lw because I read some comments by @Greg_Ward on this forum saying that aa 0 might be a good idea for calculating a few points but you need to be careful about your other parameters.

Then, my general question is why do I need the interpolation when I want values on some points? If I understand it correctly, when it comes to picture rendering I want the interpolation so that I don’t calculate a value for every pixel. But when we are talking about grid based analysis what is the interpolation doing?

Thank you for the help!!


#2

Hi Myrta,

This is a good question. If your -ab setting is 1, then irradiance caching (-aa > 0) offers little benefit for calculating individual points, unless you have a great many you need to compute. In rare cases where you have many interreflections (-ab > 2), caching may reduce the calculation time as higher order interreflections are reused. Most times, though, you can get accurate calculations by balancing the -lw and -ad settings used with -aa 0 in less time than it would take with indirect caching on.

Whether you have caching on or off, it is always a good idea to check your calculation for convergence if you are uncertain of your other parameters. John Mardaljevic has a good example of how one might do this in his chapter on daylighting in “Rendering with Radiance.” If you do not have a copy handy, I believe he also has some of his tutorials online with the other Radiance workshop presentations available at radiance-online.org.


#3

Thanks @Greg_Ward!

The idea of irradiance caching is a bit hard for me to grasp but I think I understand enough. I found John Mardaljevic’s article.

I have one more question about ad and calculation time. I have noticed that when irradiance caching is on, decreasing the value of ad decreases the simulation time, which I understand. However, when I use aa 0 and then gradually decrease ad, the calculation time decreases until I reach a limit value of ad and then it starts increasing. Why is this happening? Is it expected or am I making a mistake?


#4

That does seem strange and unexpected. What are your settings for -ad, -ab, -as, and -lw when this happens? I can only guess without knowing your parameters.


#5

The settings are: -ab 2 -as 1024 -lw 0.000005. The time varies with the ad like this:


#6

It seems like you hit upon an interesting corner-case. Typically, the -as parameter is set to something less than the -ad parameter, and zero for -aa 0 or if -ad is less than 128 or so. By reducing -ad and keeping -as 1024, you are getting into a realm where the super-sampling is taking more of the time than expected. You probably should set -as 0 for your tests when -aa is 0. We don’t enforce this because there are rare occasions when it makes sense to send a few super-samples even with -aa 0.

It’s nearly impossible to keep the user from making settings that don’t really work, unfortunately.


#7

Ok I understand! The settings can get really confusing when you don’t know exactly what each parameter does. :confused: Thank you very much for your help!

I have one last question, not related with anything else. Do you have some study when Radiance results are compared against measured results? I have noticed that there are some differences when compared with Energy Plus and it would be interesting to know what is more accurate.


#8

I don’t know offhand of a 3-way comparison between Radiance, EnergyPlus, and measurements. Generally speaking, Radiance produces more accurate results for lighting and daylight calculations, because it supports more interactions and makes fewer shortcuts. It also takes longer, for that reason.

I can refer you again to John Mardaljevic’s thesis work, where he compares Radiance calculations to a full-scale model under a measured sky. This was one of the most comprehensive tests of Radiance for daylight simulation, and the agreement was quite good for the most part.

The first two references on my publication page may be of interest, though they focus more on matrix methods related to Radiance.

I hope others can suggest further comparison studies.


#9

There have been a huge number of studies comparing Radiance to physical measurements on the basis of sensor measurements, photographic validation, and occupant survey. In my work, I’ve done pixel-by-pixel image comparison, and I also cataloged 19 other published comparison studies, which is just a portion of the work that’s been done in this area. (If you need help getting through a paywall to see the publication, email me for an author copy.) For a model with reasonable geometric and material accuracy, you can expect Radiance to produce results within 20% of measured values. That’s about the best accuracy that can be hoped for, given variation in sky conditions and measurement equipment.

I would not expect high accuracy from EnergyPlus. While it can do some very basic ray tracing to see if the sky is visible, it does not do global illumination calculations.


#10

Thank you both for your suggestions! I will have a look at the papers that you are suggesting! :slight_smile: