Boxcare Smoothing
04/13/2011 – 14:16 — Tyson Howell – Hello all, I have done some preliminary analysis using the boxcar setting on our Jaz unit, and a comparison of that to a smoothing of the raw data done in excel. I am attaching an excel sheet to this post so that you can look at the data yourself if you would like. I used two different methods for boxcar smoothing in Excel, named “Excel boxcar 5” and “Excel boxcar 10”. The two sizes were used because I was not sure of the exact definition of boxcar width, i.e. whether boxcar 10 meant 10 pixels total, or 10 each side (21 total) of the pixel of interest. Based on this analysis I am confident it is the latter. The spectrum of the Jaz values for boxcar 10 looks very close to the spectrum for excel boxcar 10, in which I took the target pixel + 10 pixels from either side (for a total of 21 pixels). The excel boxcar 5 spectrum (11 pixels total) looks quite different, intermediate between no smoothing and the Jaz boxcar 10 smoothing.  Areas of high standard deviation are centered around the reflectance peaks (see graph in sheet “graph” in attachment. peaks are around 544 and 613nm). Despite the differences in the standard deviation, the % difference between the two methods is low (<2.5% for all but 3 points, located near the very end of the spectrum where we know it is more variable anyway). Higher standard deviations are expected in areas of higher intensity. The basic shapes of the spectra from each of the smoothing methods is mostly similar (peaks in the same places), but there is some change in the relative heights of the two main peaks. When no boxcar smoothing is used, the second (and more narrow) main peak present is of the highest intensity, while when using a boxcar smoothing width of 10, the first (and broader) main peak is the highest in intensity. Using excessively high boxcar widths will reduce the intensity of sharp peaks much faster than that of wide peaks (This is why it helps with reducing noise, but it may also result in data loss if smoothing is used aggressively). We may want to retain some of the smaller peaks, particularly if we will be developing new indices. We feel it would be good practice to keep the amount of initial (on the Jaz unit) boxcar smoothing low, probably a boxcar of 1 or 2 which would average the values over approximately .5 to .75nm depending on the part of the spectrum (a boxcar of 2 would average values over ~1-1.5nm). This would allow for some noise reduction in the spectrum without any additional processing, but would also allow us to apply further smoothing at a later stage if it was deemed necessary. I will also look into the effects of taking variable numbers of averages. This boxcar analysis was done by pointing the jaz unit at a colored piece of paper in my office (to get the highest amount of consistency between reads, reducing the amount of variation do to light fluctuations rather than boxcar smoothing methods). I will do the averaging tests either on grass or wheat plots outside to get a better idea of how averaging affects variation on actual plants though. I have so far established that the amount of time it takes for a read is mainly dependent upon the integration time, which will be variable depending on the amount of light (i.e. taking measurements will take longer on a cloudy day than a sunny day). In my office it took approx 7 seconds to take reads with an averaging value of 100, corresponding to 70ms for each averaged read, so averaging excessively large numbers of reads will be prohibitive, and may even lead to errors if the fiber optic cable is moved during that time.  The file is too big in .xls format and I cannot upload .xlsx files to this forum, so below is a link to the file. Please let me know if anyone has problems accessing it.

Height/Area Calculation
04/13/2011 – 04:31 — mgttri  – Would whoever has the spreadsheet that does the measurement height vs area calculation please post that spreadsheet so that we all have access to it?We need to resolve a measurement height, in part because this will set a minimum lateral distance from any support that we use to hold the sensors. Thanks, Mary

Measurement height and spot diameter
04/18/2011 – 13:30 — Tyson Howell – Attached is an excel file containing the values for the distance of the fiber optic to the target (reference/sample) and the corresponding spot diameter. For those of you that would like to calculated this yourself, this is the information we received from the Ocean Optics representative: