Fitting boxes into curves

Is it really almost August? It seems so strange to be finishing up research for the summer. We just completed the last of our veg-work last week, and are now working on entering data. I’ve also spent the past few weeks learning how to use Distance, a program which determines the density and size of a population based on detection probability. Our method of point-count sampling has allowed us to generate a relative abundance index, with which we can estimate trends in bird abundance based on detection probability. The program can also be used to account for covariates by examining the effects of different factors, such as habitat type, wind speed, and time of day, on detection probability.

I’m still only getting used to Distance, and playing around with data. The actual analysis will come later in the school year, when we can pair our bird data with the vegetation data we recently collected. However, there is still a lot to gather from these early stages of analysis, particularly by looking at the curves of bird detection with respect to distance. The data is expected to fit a detection function, with the majority of our detections lying between 30 and 50 meters away from the observer. The function illustrates the probability of detecting an object with respect to a certain distance away from a point. Detection is expected to decrease with increasing distance from the point. While the bulk of our data, which includes over 5000 bird detections, forms a beautifully shaped half-normal cosine curve, the data is much less attractive when split up by species and observer. In fact, there seems to be a trend of bird detections lumped at about 40 meters away, such that the data does not fit the curve. This clumping is likely due, at least in part, to inaccuracy on the part of the observers. When we are in the field, we estimate the distance of any bird song or call to the nearest meter. Clearly, there are expected to be high levels of observer inaccuracy based on our ability to estimate distances. I find it hard to believe, however, that all four bird observers (myself, another student, my professor, and his grad student) consistently clumped data at 40 meters away. Another possible reason for the clumping is that the presence of the observer causes the birds to spread further away, so that they actually do end up clumped 30 to 40 meters away. Either way, we will have to test the accuracy of our distance estimations before continuing field work next spring. One way to do this is to go out into the field and guess the distance to a bird, then actually track down the bird and see how its actual distance away compares to our estimated one. Observers can also use tape recorders of bird songs, played at random distances away, to practice estimating distances to bird songs.

When we separated the detection curves by observer, we found even more variation in the resulting functions. Some of my own detection curves were laughable for their failure to even remotely follow the expected trend, and tended to look much more like rectangles than like curves. However, studying my box-like curves actually make for a very interesting lesson in detecting calls, which I will certainly take into account for next field season. For example, my detections are supposed to peak at about 40 meters away, yet they tend to drag on up to about 70 or 80 meters. This seems to suggest that I put too much focus on listening to further-away species, such that I fail to notice those that are closer to me. I think that some of this inaccuracy comes from my difficulty recognizing the call notes of certain species. I am far more comfortable with the songs, and find the various clicks and trills of the call notes to sound similar for each species. I’ll have to work on my call-identification in order to more accurately represent species-distribution.

All in all, getting to know the Distance program has taught me a lot about the assumptions behind field work, and how to account for the inaccuracies behind those assumptions. I’ve still got a lot of work ahead of me, particularly in order to better understand the mathematical functions behind the analysis. I’m planning on taking a bio-stats class at the University of Ghana next semester, which will hopefully provide me with further understanding of the analysis process that I’ll be able to apply to my research when I come back to campus next spring. Then we get to add vegetation variables into the mix!