PYoungbaer wrote:Peter,
... where is this headed?
The goal is to take raw pesticide data and make it "visible"...
This screen grab from Google Earth shows what some of the raw pesticide data would look like when displayed with Google graphics. In the GE program you can click on the droplets and get a breakdown as to county, zipcode, toxin name, EPA registration number, etc. Unfortunately, this is pretty tedious. Just looking at a map with splotches representing toxic substance application areas takes a long time and begs the question "Where are bats feeding in relation to these splotches?" The end result ideally should be effortlessly understandable to someone looking for relevant leads to explore.
We need a better system that allows for mass processing. A good system should also allow for "adjustments" that will make it more or less sensitive, like fine tuning a shortwave radio to bring in a weak station. (Hopefully statistical analysis could also be applied, but for now let's just try to make discoveries.) Here is a sample of the use of the matrix techniques discussed previously, showing an example of such amplification:
The three images on top of the chart are the matrices for three bat locations. These foraging areas are identified as "A" "B" and "C".
Bacillus thuringiensis serotypes SA-11 and SA-12 (comparing all three locations for the years 1999 to 2005) were processed and the results placed in the tables below the respective images. No image is provided for
Bacillus on this chart, but here it is for you to look at:
This particular matrix is from the 2004 data, and clearly shows a high concentration close to area "A".
The amounts of
Btk applied were very low, only a few pounds, so sensitivity of the method in this case was strongly enhanced by applying the
Kriging interpolation . This caused an artificial spreading* of the pesticide data, clearly visible as a fog, brightest at the point of application and diminishing greatly in intensity with distance.
(Such methods are only acceptable when one is
exploring the data. In the specific case of this
Bacillus, spores are sprayed along with the actual toxin that the microbe produces. This spore is infective and capable of propagating the organism well out of the application area.)
The following graph illustrates the results (for Btk SA-11 only):
We can see quite clearly that in the year 2000, area "C" had a
Btk application nearby, and in 2004 another application was close to area "A". We can also see that in 2005 the trend was to apply
Btk farther away from all three areas. Three trend-lines were generated using 4th order polynomial regressions. These lines
suggest a four year pattern which may (or may not) match up with patterns of Gypsy moth flareups. Does the data also suggest that this form of insect control is very effective, and in 2005 the moth counts were considerably reduced? Possibly. Problem for bats? Maybe. Cause of WNS? Nope. Really doubt that very much.
One down and about 3000 to go.
*This spreading is an artifact of the math that goes into the Kriging algorithm. Other algorithms have artifacts as well and must be carefully applied so as to exclude any bias they may cause. Here, we use it creatively, to explore possibilities.
A big note of thanks goes to Al Hicks, John Chenger and their teams for providing the original bat emergence data, without which this analysis would not happen.
Technical details for creating the matrices and processing the data will be PM 'd to those who have requested it.
-p