Artificially low survival due to poor detections downstream?

questions concerning analysis/theory using program MARK

Artificially low survival due to poor detections downstream?

Postby Cmichel » Sat Feb 04, 2017 4:09 pm

Hi Phidot users,

I have a question that has been bugging me for a while. I use the a spatial version of the CJS model to estimate survival of outmigrating juvenile salmon in rivers. In one of my study years, I have reasonable (>0.5) detection probability at my furthest upstream detection location, then terrible (<0.1) detection probability for many detection locations in a row (12), then reasonable detection probability again in the lowest parts of the river. Program MARK estimates perfect survival (1) for all the river reaches that have poor detections on their downstream end, EXCEPT for the first one in that row, which was given an unusually low survival estimate (with somewhat reasonable confidence intervals). I think it's unrealistic to think that all the salmon survived through those 11 other reaches (total >150 km long), and I also feel that the survival estimate for the first reach is too low, so I'm wondering if the CJS model attributed all the mortality of the stretch of poor detection reaches to just the first reach? Is this a model behavior that others have noticed?

Finally, any suggestions to fix this behavior?

Thanks.

Cyril
Cmichel
 
Posts: 3
Joined: Tue Mar 02, 2010 7:07 pm

Re: Artificially low survival due to poor detections downstr

Postby aswea » Tue Feb 07, 2017 8:04 pm

I think that when you have really low DE, there just isn’t enough data to get reliable estimates. The model doesn’t know where to allocate the mortality. Also, CJS models don’t perform well near 0 or 1 so if your survival is quite high in the lower reaches, then the model could estimate that as 100%. (This might be exacerbated by the low DE so that the survival is bumped up too high.) Then the model would allocate the extra mortality to the first segment. The total survival over all your reaches will still be right (ie cumulative survival from release to the last detection site).

I usually exclude detection sites that perform really poorly.

I hope that's helpful!
aswea
 
Posts: 27
Joined: Sat Oct 17, 2009 3:32 pm
Location: Gander NL


Return to analysis help

Who is online

Users browsing this forum: No registered users and 17 guests