This question references the Kery and Royle textbook, "Applied Hierarchical Modeling in Ecology", Volume 1.
I am trying to apply open HDS models to a 20 year point transect distance data set collected on Hawaii amakihi birds. Right now I am wrangling the data so it matches the way they are set up in the book, where they demonstrate applying the model to scrub jay data. Where I am encountering a problem is with the "y" matrix (pg 517) of distance class frequencies. For the simulated scrub jay data, there is only 1 detection per site per year, which makes it easy to set up the matrix; however, unlike the simulated data, I sometimes have multiple detections of different individuals at different distances per site, per year. So, I'm not sure the proper way to incorporate all of the information into the "y" matrix of distance class frequencies. My first thought was to simply add all of the individuals and average the distances, but i'm not sure how that would affect subsequent variability estimates or if that is an acceptable practice at all.
A pity that Hawaian birds are still being monitored by the discredited "point transect" method. The assumptions are difficult to check, previous claims to have done so are spurious, and most users rely upon a visual check on the raw data to reassure themselves. Considerable departure from the assumptions can remain hidden from such slack practice. This means that the effective area sampled is subject to large and unknown bias. At best, this method delivers an imprecise index of numbers or density, rather than the promised bias-free estimate. This is more important than configuring the data.
I gave up on Distance and just use counts now as a measure of comparison, rather than density
Hi,
despite all concerns about the proper used of the method, I'll answer about the data format.
For Bayesian computing, you just need to set up a matrix for each secondary occasion within a year with in row the number of point surveyed and in column your distance class. distance class is the same as using distance with bins.
even if you have 1 bird detected per point, it won't be at the same distance everytime, and so, you will use cutpoint to have a good fit of your model.
Perhaps, instead of diging directly in this model, you should go back to the chapter about Conventional distance sampling where they explained why to use this format.
Cheers,
Guillaume
- Badges
- Science topic
- Similar topics
- Biological Science
- Faunistics
- Ornithology
- birds