I’m working on analyzing and plotting the data from my paint test of Bare Wood -vs- Latex-vs- Whitewash this weekend. The dataset is huge due to me taking samples every 15 seconds on 4 probes. So the going is slow.
Here is a plot of the 4+ months of data with curve fits, as expected, the differences between the different screens gets less as solar irradiance reduces with the seasonal change.
Click for full sized image
I’ll have more later. I’m working on a Welch data depopulation scheme to make the data size a bit more manageable without losing peaks.
I’ll also plot some samples of daily data to show what happens at Tmax and Tmin
So what you are saying here is that as solar insolation increases due to the seasonal change (tilt of the earth) so does the bias in a non linear manner. One can not simply subtract a defined amount of bias without knowing the month of the year.
This also brings up another sticky issue, is the UHI effect similarly non linear throughout the year? Or worse, non linear through out the day? Or is the UHI bias pretty much dealt with by a simple comparison between a supposedly rural station and subtracting out the rate of change and bias?
REPLY: At this point I’m not saying anything, the analysis is incomplete. This was just a graphic to illustrate the size of data set and the change over season. I had hoped to be further along by now, so I simply provided this as demonstration of work in progress.
Do not draw conclusions from it about UHI or the paint test itself.
Considering that at the moment the daily and seasonal cycling is immaterial to the intial assessment, you should normalize on one of your measurements (air perhaps? – I’m guessing it was a control of some sort). If you then had a specific solar insolation reading, you could then correlate to that.
“This also brings up another sticky issue, is the UHI effect similarly non linear throughout the year? Or worse, non linear through out the day?”
Well that is how a heat sink operates, isn’t it?
OT.
http://www.iht.com/articles/2008/01/11/america/calif.php
Truth is stranger than fiction
Anthony:
At the scale and resolution which can be posted, little detail can be visually discerned. However, I see a couple of blue Tmins showing below the overlays, and would sure be curious as to your thoughts about that. I assume those are nighttime lows, and I am trying to think of a mechanism which would occasionally cause that. If it were often or always, instrument error would be the culprit, but an occasional lower than low? BTW, I watched (monitored, actually) the cold air drainage out of the hills around Pleasanton, CA one summer, and wondered if there were any of those effects in the location of the test screens.
Looking at the graph, two questions immediately come to mind.
1. What is the resolution of the individual readings? I assume they are digital and if so, what is the smallest reading increment?
2. How accurate are the measuring devices themselves? We appear to be looking at differences at the 1% level or less. Is this realistic?
Seems that the white paints are very nearly the same.
John Andrews
REPLY: Again this is graph is not for analysis, its just to show the extent and general trend.
The NIST calibrated dataloggers have .1 degree F resolution. The differences are there, but of course not visible in this graph. No more questions about this graph please. Wait for the results I’m compiling.
That plot gives me an idea for an experiment. I’d love to take one of the “interesting” temp stations (cr4 or 5) and set up a grid of thermometers, with automated frequent recording (like in your Stevenson screen paint test). Imagine a grid of, say 10 x 10 thermometers, with 2m spacing. (more thermometers with less spacing would be better, but this is already unrealistic.)
Record the temperature for a year, and then study the micro-differences as they are overlaid on an aerial image of the site. Is there a “stream” of higher highs downwind if the ac unit? Can you tell which days the BBQ pit was used? Is there a gradient in nighttime temps as you move away from asphalt and onto the grass?
It would be a big project, but the results would be fascinating, and go a long way towards quantifying the range of some of those errors.
Another interesting project Anthony!
We’re looking forward to seeing your report. Will you be testing for significance of differences?
It’s really amazing how data accumulate, the challenge presented by their analyses, and what technology allows! Your challenge reminds me of a Marine Ecology professor I knew who did his dissertation on recovery of marine algae communities following El-Nino destruction at UC’s research facility on Catalina Island. He continuously monitored a number of parameters at multiple locations and multiple depths, continuously for several years. Being pre-PC days, he had miles and miles of paper graph/curves accumulate.
After transporting the data rolls back to L.A., packed in sealed ice chests, roped together and attached to floats and flags, he manually (with scissors and friends) cut out all of the graph profiles to quantify (by weight of paper) the areas below the curves for each parameter. What he would have given for a USB-download!
John