# Assuming we can hear them….

I just came across these notes I took and never got around to blogging on – so here’s some more information on analysis from a couple weeks back:

One of the primary things we’ve been dealing with since the start of our field work has been finding the most accurate, and least difficult, means of surveying birds. Standard forms of sampling include strip transect sampling, which assumes detection of all objects along a strip, and is therefore better used for plants or relatively stationary animals; line transect sampling, which records the distance of objects away from a designated line, allowing for a greater sampling size; and trapping webs, which use mark-recapture methods to estimate populations. For our bird work, we used point counts, in which distances are estimated between the object (in this case, any bird we hear or see) and the observer, who is standing at a random point. The randomness of the point is important to ensure that the points represent the whole range of potential habitats. We selected our points based on those in riparian and upland habitats, yet within those parameters we experienced everything from briar-infested clear-cuts to muddy swamps in the middle of the woods.

Now that we’ve started the analysis components of the research, I’ve begun to pay more attention to the assumptions we’ve made in our work, and the steps that need to be taken to ensure those assumptions are met.

1. All objects directly on the point are detected. Not too much of a problem, seeing as how most birds are further away from us…

2. Objects are detected at their original location. This one is pretty hard to keep up with, especially as many birds fly about us the entire time we are surveying. We have to keep track of where they move, to make sure we don’t count them more than once. Sometimes it’s hard to tell if a song we hear is a second bird, or just the original one that’s flown somewhere else.

3. Distances are measured accurately. Also a challenge… (see previous post)

4. Objects are correctly identified. This was very difficult in the early spring, when I was only just learning the multitude of bird calls and songs. Though much easier now, it is still a challenge to differentiate between certain species. And I’m still hearing calls that I’ve never heard before!

Besides the challenges of accurately detecting the species, we also want to make sure we’re detecting all the birds within range of our point in the shortest amount of time. We started out with three ten-minute intervals. The separate intervals allow us to determine the detectability of a species via the mark-recapture method. If a species is detected in each time interval, as was often the case with cardinals or chickadees, it has a high detectability. If a species only shows up in one time interval, it has a lower detectability. Our ability to detect a species is then taken into account when determining species abundance. Just because we hear one species more often doesn’t necessarily mean that there are more of them.

However, after a few weeks of ten-minute intervals, we decided that we were detecting a majority of species after about 7 minutes. So we changed our sampling method to four seven-minute intervals. However, we only sampled 7-minute intervals for a few days, because there were still too many species showing up at the end of the point count. We finally settled on three 8-minute intervals. We recently analyzed the success of these intervals using excel, and found that we captured 97% of species by 7 minutes. This is pretty good! Most of our detections are at the beginning of each interval, and then level out to an asymptote towards the end, indicating that we’ve detected as much as we can in the allotted time.

We also broke down our data by observer, to see how time-of-detection varied from person to person. We found that Kelly and I detected about 95% of species by 7 minutes, which was definitely better than expected. I thought that this would turn out to be more of a problem for me, because I often spent so long just trying to think what a species’ song was, that I failed to accurately record the timing of other songs. It’s good to see that this wasn’t too much of a problem after all.

I think next year we’re going to continue the 8-minute intervals. It seems like just enough time to get in the majority of species, while still being short enough to maintain assumption 2 (the shorter the time, the less repeats).