Replies: 4 comments 1 reply
-
Hi Olivier,
You need to specify the
I'm a bit confused about the setup. What is the unit of time in your index calculation? Day? Something longer? I don't think you'd want this as an offset, unless this is your measure of effort and you're summing your survey observations. The offset would be the (log) measure of effort for a given observation. E.g., (log) hours surveyed, assuming your response is the count and not already the ratio of count to effort. It is possible you would want a random intercept or something in time-varying to account for observations from the same 'survey' being correlated with each other, but again, I'm not sure I understand the setup. It's fine for some sites to have more observations than others (technically as long as your choice of where to sample isn't being driven by the underlying abundance dynamics themselves).
As defined by the 'grid' (or similar) supplied to
Did you set the |
Beta Was this translation helpful? Give feedback.
-
Hi Sean, Thank you for your reply. Let's see if I can be a bit more precise about the setup below...
This is indeed useful to know.
The study was initially set up as "point transects" for distance sampling. Each point is surveyed 6 times at 2 hours interval, measuring the distance and bearing to each animal group and the group size, within a ~ 5km radius plot centered on the survey point. Observations from a same surveys might be spatially correlated, but within a single survey, each group was observed only once. However, because of this design with 6 surveys, many of the animal groups were counted repeatedly across surveys throughout the day, within the same point transect.
Right. Now I understand better what was going on.
I guess I was doing something wrong... The estimates I get now seem to be much more "stable" across different grid sizes, which makes more sense. Thank you again for your help. Olivier |
Beta Was this translation helpful? Give feedback.
-
What is a row of data (observation) that you are fitting? Is it a count of individuals at one point in space and time from one survey?
Here, what would the 'offset' column in your data look like? |
Beta Was this translation helpful? Give feedback.
-
@seananderson Coming back to this after a break.... I'm still a bit confused. Please allow me to try to clarify a bit further, as I would really appreciate your input. Most plots were surveyed 6 times at 2 hours interval, but a few were surveyed only 2 or 5 times. This is coded in the data set as a factor column called facnumhour with levels 1 to 6 ; ie if the plot was surveyed only 2 times, then only levels 1 and 2 appear. The data also includes a column numhour , which is the numeric equivalent of facnumhour . There is also a column effort which gives the total number of surveys for each plot ; ie if the plot was surveyed 6 times, then effort contains I currently fit the model with :
My main question right now is about whether I need the Thank you in advance for your help. |
Beta Was this translation helpful? Give feedback.
-
Hi everyone,
We are trying to estimate the abundance of a group-living species in a fairly
large (~ 13000 km2) area, with repeated surveys (ie 6 counts in a day) at 101 points.
There aren't that many observations to work with though, with ~ 50 to ~ 120 (or ~ 300 to ~ 750 individuals) observed at each survey.
The model is fairly simple, with group_size modeled with a negative binomial
distribution, and for now, an intercept-only model with a spatial component to account for location
of observations, and a temporal component to account for repeated surveys, and because it is
kind of required by the get_index() function.
The model seems to fit reasonably well, and the sanity and residuals checks are OK.
I have 2 question (maybe for now) :
Not all points were surveyed 6 times. Is this already and properly 'accounted for' through the varying intercept by survey (ie through the temporal component of the model)? Or should I also use the actual number of surveys per point as an ofset?
As far as I understand in this context, the estimate provided by get_index() for each survey (time slice) is the abundance over the 'study area' as defined by the mesh. Before today, this estimate seemed to be extremely dependent on the cell size of the prediction grid. I had results ranging from small but sensical, to nonsensical crazy high estimates. Why would this estimate be so dependent on grid size, and what what would be the best way to find the "correct" grid size?
Disclaimer: as I am trying again now, this estimate seems much more stable across different grid sizes. I was likely doing something wrong, and this second question may not be relevant anymore. Please let me know if this is the case...
Thank you in advance for your help.
Olivier
Beta Was this translation helpful? Give feedback.
All reactions