If you're going to use ud95 as the response you might as well average
the prop_land values per bird (i.e., aggregate the data down to a single
data record per bird); the within-bird variation in prop_land values
won't affect the model output at all (although the _number_ of
observations per bird will; if you have unbalanced information in this
way, you should incorporate weights proportional to the number of
observations as well).
With only 6 colonies you're going to have some difficulty estimating
the among-colony variance very well; if you end up with zero estimates
of among-colony variance, you might want to use blme or set the colonies
as fixed effects ...
If you use prop_land as the response variable you would indeed want to
put bird_id in as a random effect (and you might consider estimating the
proportional as a binomial (GLMM) response, _if_ you know the total
number of fixes for each bird)