Geographic closure assumption

[jhart9]

I have been given a long-term (2000-2012) mark-recapture-resight dataset that may break the geographic closure assumption for open population models and would like to get some other opinions as to how it can be analyzed.

The problem:
There were 12 sites with banding efforts over the years with most effort at 3 sites. Not all of the sites were used consistently over the years; of the 3 main sites, 2 were used from 2000-2006, while the third was used from 2006-2012. The furthest points are 14 km apart and it is known that the birds banded at one location can be seen at another location, although most (84-99% at the 3 main sites) are seen again only at the initial banding site. So it is difficult to say if sampling at the 12 sites are sampling the same population or how much immigration/emigration is occurring at the site level.

My questions:
Is it valid to lump the data from all the sites and do one continuous survival analysis for all birds for all years from 2000-2012? Or do the different sites need to be analyzed separately? If not all the sites can be lumped, could sites in close proximity (<1-2 km) with an overlap in netting years be lumped?

More about the species:
The data are for a resident population of birds which are nomadic within their restricted and isolated range (i.e., the birds do not maintain territories at any time of year and there is no immigration or emigration from their range). The range of the species encompasses 1530 km^2 of which only 65 km^2 is covered by usable habitat. Both adults and juveniles are marked. Adults were given unique color combinations for resighting in the field. Juveniles were only given a metal band and had to be recaptured again before they were given unique color combinations.

Thank you for your feedback.