Help with designing the analysis of radio telemetry study?

Question

I was wondering if there are any spatial ecologists out there who can provide some insight on the trouble I'm having with my Masters research. I am looking at the association a bat species to high density aggregations of a flowering plant. The bats were tracked throughout the night with a point taken at approximately 10 minute intervals. The bats were tracked over consecutive nights until an asymptote in their space use was reached. So now I have point patterns for each bat that pretty much covers the entirety of their space use at that period. When the data is displayed visually there is a clear pattern of the bats that were tracked during the flowering period moving into the aggregations of the plant however they also spend a lot of time foraging around their roost sites. I have simulated the point pattern of each bat 999 times by randomly shifting and rotating the points within the study window (the internal structure of the observed pattern was maintained in the simulations to account for autocorrelation). I then did randomisation tests to see if the average distance to the nearest plant aggregation for the observed data is within the lower 5% of the distribution of the averages for the simulated point patterns. However because the majority of the points collected are highly associated with the roosts the average distance is primarily effected by the position of the roost in relation to any of the plant aggregations. I've been thinking of ways to disentangle the association with roosts and the association with the plant. First I was thought about putting a buffer around the roosts and removing the points falling within these buffers from the analysis. The problem is deciding on a buffer. In the other areas I have been using a buffer equal to the circular error probable of the radio tracking error but placing this around the roost will only incorporate the points where the bat was actually inside the roost not foraging around it. All the bats have different foraging patterns around their roosts and I can't see how to assign a buffer distance that is not completely arbitrary. The second method we have been looking into is taking out the points that are more closely associated (closer in distance) with the roost than the nearest plant aggregation and then running the randomisation to look at the association of the remaining points to the plant. I want to get an idea of whether or not removing the association with the roosts in this way will potentially bias the analysis. Any thoughts or further suggestions?! Thanks in advance!

Answer 1

Hi Asked,

I don't exactly grasp what type of association you are attempting to quantify here, but there is perhaps a useful analogy from the different methods different manufacturers of HPLC (high-pressure liquid chromatography) systems use to assess their detector noise. A signal consists of a very high number of data (perhaps 10/second) per frequency band (these may be diode-array data for an entire UV range) sampled over perhaps 30 seconds. The issue to be solved is, whether to treat all the data (per wavelength) as a pooled group, and calculate the root-mean-square deviation from the average value, or to treat as a sequential (ordered) file, and determine a regressed fit (i.e. a drift rate) and then calculate the root-mean-square deviation from that regressed fit. The difference in calculation is considerable -- and what if the data have a domed structure (drift first up, then down), how could you compensate for that, let alone determine if it is significant, etc. One manufacturer (Perkin-Elmer, if I recall) chose to do something entirely different: calculate from the point(i) to the point(i+1) differences ONLY. The algorithm was extremely fast, and insensitive to drift.

The analogy here is, the UV data are like your bats -- how do you cluster data on the fly (so to speak) so as to obtain the most realistic signal (association of like plant-aggregation visits), in the presence of surrounding different value data (position away from all plant-aggregations, being bat noires or whatever)?

The behavior of the bats while wandering between aggregations will not perturb the measure of attraction to aggregations as much as is the case with the more conventional, all-data-pooled data reductions, perhaps?