[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

The Lighting Survey Thing

> From: "Roger S. Miller" <rgrmill at rt66_com>

> I agree with Frank that Erik's survey results are thought-provoking and an
> interesting contribution to the hobby.  I have some observations and
> questions.
> 	First, the difference between the data from Amano's books and the
> Usenet data is striking.  I've only read one of Amano's three volumes and
> my recollection is that the smaller tanks feature a lot of stem plants
> with very dense plantings that shade each other, and that the larger tanks
> feature more moderate and low-light plants, with relatively little
> shading.  If my recollection is true and his other books are consistent,
> could the trend in Amano's data be a product of his unique style?

Possibly some influence.  I seem to recall, though, that he used a lot of
Glossostigma (high light plant) in all his tanks, irregardless of size.

> 	Second, when I visualize a trend through the four data points from
> commercial fixtures it parallels the trend in Amano's data, but is offset
> downward by a factor of about 4.

I'd say the four data points aren't really enough, but you are right, the
commercial fixtures would be determined by length (or approximated by
volume to the 1/3 power).  This is NOT a good fit to Amano's data in the
long run... I tried it, and it's much too shallow.

> 	Last, the data are plotted up in log-log coordinates, which is
> great when you're dealing with something (like chemical concentrations)
> that obeys logarithmic rules.  Is that the case here?  If not, then
> log-log plots will sometimes compress data to create trends that aren't
> evident in linear plots.

Log-log is great for data that spans several orders of magnitude (such as
the Amano tanks, which range from 1 L to several thousand L, and it's also
great for looking at "power" trends.  You'll notice that the watts/L is
linear, but so is the watts/(L^2/3), just with 2/3 the slope!  If I had
plotted watts/(L^1/3), it would also have been a straight line, with 1/3
the slope of the watts/L plot.  It would have been much more difficult to
spot this sort of trend if the data were plotted linearly.

The danger with log-log plots is that effects can be reduced at large
values and enhanced at small ones, but I don't think this is what's going
on here.

This is fun, I haven't gotten to do any of this stuff since the old
dissertation three years ago. :)

  - Erik

Erik Olson				
eriko at wrq.com