Over-estimation of population densities

[Kah Ming Teo]

Hi all,

I'm currently Distance software on Windows to estimate the population densities of 5 species of birds in a certain region that was surveyed using 5-min point counts. But the results that I get look much higher than possible. I have a few questions and would like to seek someone's advice on how to fix the issue of over-estimation.

Each analysis is performed separately on each species individually using CDS. Data is being collected monthly over a span of 2 years. (survey effort = 24 per yr, 1 per mth). Data used for analysis is being sorted by month, then by point. survey effort has been put as 1. (Should survey effort be 1 or 24 if data is sorted by month, not by points? -> I noticed that Point 1.1 is being treated as a different point by software if they aren't arranged together in data. Since it's sorted by months, they are separated. I've tried sorting by points, and survey effort = 24 and I got similar values.) Distance data was binned, 10m, 50m, 100m, 200m. Intervals of 0 - 10m, 10 - 50m,  50 -100m, 100 - 200m were created.

All the possible combinations of Key Functions and Series Expansions have been used. (4 x 3 models). I've selected AICc for model selection, although selecting AIC didn't seem to fix the issue. I have stratified my data by month in the system, adjustment terms were automated selection, with max 5 terms, and Strictly monotonically non-increasing was selected for Constraints. For Cluster Size, means of observed clusters were selected.

Here's the result I got (pooled estimate). I also have the estimates for each month. I've looked at the Chi-square GOF for the model for each month and some of them were p = -1.0000 or p = 0.0000. (indicating over-fitting or perfectly fitting models? Either way is not a good model - please correct me if I'm wrong.)

I've tried truncating my data at 100m (As data is binned, 10% will automatically truncated all 200m data, which is equivalent to truncating at 100m.) But the estimates were no less ridiculous.

I would greatly appreciate any advice on how to fix such problem. Thank you for taking the time to read and troubleshoot!

 

[kmteo]

Hi all,

I would like to update regarding my query, following my discussion with Eric so far. (Thanks Eric for your advice!)

While the results were still larger than the expected density/abundance, there were some advice and comments that Eric offered which were useful.

1. To check the number of detections per month before using the monthly data for analysis. The number of detections has to be sufficient in order to analyse the data month-by-month. If the number of monthly detections is sufficient, it is not necessary to use the pooled detection function. Using pooled detection function would invalidate the monthly estimates.
2. The data with which Eric has used to troubleshoot was for a hornbill species, which calls loudly. Negative exponential function would not be biologically feasible since the function gives a detectability model that falls away too quickly. This will result in an over-estimation of the figure as the model presumes that you detect only a small percentage of birds.
3. The effects of overlapping points do not affect the outcome significantly according to Buckland (2006) so it's not required to adjust for overlapping points.
4. Large flocks that occur very occasionally create large uncertainty to the average group size.

Buckland, S. T. (2006). Point transect surveys for songbirds: Robust methodologies. The Auk, 123(2), 345–357. https://doi.org/10.1093/auk/123.2.345

Regards,

Kah Ming