Problems with too low detection data.

[hankyukim]

Hello everyone,

I'm currently working with national survey data of my country to figure out which species of forest birds are in decline. Unfortunately, the data is not so friendly for other methods lacking with some essential information in searcher's effort, survey distance, and the sites are not contingent each year(which makes me difficult to find out adjusting to multi-season analysis)

So, I've been tring to apply simplest occupancy analysis, for each species, each year's data was run by single season, single species, constant P analysis, and i was planning to line up all year's (about 14 years) data and run with simple linear regression to show their trend of incline/decline.

but I have troubles with some datas which make me difficult to use all other datas. In some species, which are rarely detected in that year, due to it's scarceness, the occupancy converge to 1.000 every time. even though there raw data seems to be almost zero.

what could be the solution? I do not feel familiar with the statistics, but in my opinion, I think it's because the total detection rate is too low.

I've also tried to fix the detection rate from the earlier year(first year's detection rate) to all of the years, assuming that it's not changing, but it brought out other problems with occupancy estimation.

Thanks!

[jhines]

Hi,

Sorry for not seeing this sooner. My suggestion when you have sparse data is to pool whatever data you can. It sounds like you've already tried combining years of data and estimating a single 'p' for all years. You might also try combining species which have similar characteristics. You might assume that they have similar detection probabilities, or even occupancy. If you can get an estimate of detection for one species, then apply that detection probability to another species which has sparse data, you should be able to get an occupancy estimate for the sparse species.

In the end, you might be right about not having high enough detection probabilities to estimate occupancy. If you have a species which cannot be combined with another species which has enough data, then there aren't any statistical methods which will produce estimates for the species with very little data.

Jim