Sampling Bias

Now that you've seen the scanning method, its worth spending a moment considering sampling bias. All methods have it, and this is no exception.

A few (brief) thoughts on collector and sampling bias: A paper collector is not an approximation for a leaf- it does not have the same surface texture or chemistry, it reacts differently to air currents, et cetera. It can provide a way to measure relative amount of spray that was in a particular canopy zone and allow a researcher to relate it to other collections in other canopy zones. Orienting the cards on a horizontal plane and analyzing the "tops" and "bottoms" of the cards can be useful in understanding a spray technology's relative ability to penetrate and coat a canopy. Again, this is still not equivalent to leaf "tops" and "bottoms", but it might just be the best we can do...

The following commentary applies to collectors used in three dimensional canopy sampling, and generally would not apply to 2 dimensional bare field sampling such is commonly done to calibrate aircraft.
 
A quick look at any spray card and it becomes obvious that coverage, especially when canopy is involved, is NEVER truly uniform or homogeneous. The edges and ends of a collector are nearly always different from the middles. So what is the
right way to sample a spray collector card?

One way would be to scan the entire card. This comes with problems as the cards are often not perfectly uniform, making it necessary to scan each card and crop the image to fit. If not cropped, ImageJ will analyze the non-card area also. This is a time-consuming process, and cards are are usually held by a clothespin or clip of some sort, (as you can see in the water sensing paper image at left) adding more bias. Further, if using 4800 DPI, scanning the whole card, if, for example you are using a WSP 1" x 3" card, would be nearly 200 megabytes of data and about 10 minutes of scanning alone per card. That could get tedious very quickly.

The method proposed and used for sampling regions on the collector here is an imperfect compromise that gets a broad sample area, and preserves some productivity. By analyzing a relatively large 2 cm^2 area, there is a lot bigger sample size in each image than a typical microscopy photo. As you saw in the videos, by somewhat haphazardly placing the scanned zones on the platen, you would never really know where you were scanning on the cards, minimizing risks of human bias in selecting scan areas. It's not perfect though, as it is difficult to scan to the absolute edge of a collector.

More work could be done to improve this and it is hoped the community will provide feedback to improve this methodology in the future.
Comments