[Radiance-general] Re: gendaylit

[John Mardaljevic]

In the Radiance validation I carried out using measured sky luminances, I
also tested a number of sky models including Perez (gendaylit).

Of the 754 skies, there were 41 for which the Perez model description could
either not be generated (outside parameter range - 27 skies) or which produced
negative vertical illuminances (14 skies). These were eliminated from the analysis
for this model leaving 713 skies. The negative vertical illuminances
resulted from distortions in the sky luminance distribution that can
occur unexpectedly for certain combinations of input parameters. These
parameter combinations were present in the data collected by the BRE 
but they were not encountered in the Berkeley data that were used to 
derive the model. This effect was noted by Littlefair and an 
adjustment to the model to prevent this distortion was advised by 
Perez. A routine examination of the gendaylit code showed this fix
to be present.  So it would appear not to correct all occurrences.

The presence of a distortion was taken to be a negative value for any
of the predicted vertical illuminances. The actual luminance distribution
for the sky was not examined. So the possibility remains that some of
the other skies may yet have exhibited some distortion.

The distortion manifested itself, weirdly, as a region of -ve luminance
sky that could be seen in rview as a conspicuous strangely coloured patch.
Note, Radiance (v2.4) didn't complain that -ve luminances were encountered
when rays 'hit' the glow material.  Which is fair enough, who'd expect -ve
luminances to be present?  Also, -ve luminances notwithstanding, the diffuse
horizontal illuminance predicted using rtrace was *correct*.  So, other
parts of the sky were no doubt high (+ve) luminance to compensate.  As
noted above, a minimum test that distortions were not present would be 
reasonable predictions for vertical illuminances from the Perez sky.  To
be certain, you'd really have to test individual rays that sampled the
hemisphere.

The above is a report of what happened. I'm not suggesting that Perez/gendaylit
should not be used.  In fact, as far as I recall, the parameters that led
to these distortions were also around sunrise/sunset with diffuse horizontal
illuminances of only a few klux - not a big deal for daylighting/energy calcs.
Rather, i'd caution that strange things can happen (or at least did) when gendaylit
is (was) routinely used to generate skies from measured data.

For desperate insomniacs, sky model tests (Chapter 5) and other stuff can
be downloaded from here:

http://www.iesd.dmu.ac.uk/~jm/zxcv-thesis/

Sleep well....

-John Mardaljevic

-----------------------------------------------
Dr. John Mardaljevic                     
Senior Research Fellow
Institute of Energy and Sustainable Development
De Montfort University
Scraptoft
Leicester LE7 9SU, UK
+44 (0) 116 257 7972   
+44 (0) 116 257 7981 (fax)

jm@dmu.ac.uk   
http://www.iesd.dmu.ac.uk/~jm