banding area to two or more recovery areas, although the nature of the inferences
depend on assumptions about the constancy of survival rates among locations and
the permanence of migration “decisions” (Schwarz et al. 1988; Schwarz and
Arnason 1990). For example, do birds decide where to winter during their first
migration and thereafter consistently return to that same locality, or can birds visit
different wintering locations in different winters?
In the other sampling situation, banding and recovery occur on the same
areas. Winter banding of North American waterfowl occurs following the
hunting season in late winter, whereas early winter recoveries occur in the
same general locations. Under this sampling situation, multistate band recovery
models may be useful (Schwarz 1993; Schwarz et al. 1993). These models may
be viewed as special cases of multistate capture–recapture models and involve the
same kind of thinking and modeling as described in the previous section.
Covariate modeling and constrained models incorporating interesting biolo-
gical hypotheses can be developed using these band recovery models as with the
previous models. Tests and model selection criteria can again be used to discrim-
inate among competing models. Band recovery models have not seen much use
in estimating bird movement parameters, probably because large numbers of
recoveries are needed to obtain reasonably precise estimates.
5.5 Summary and general recommendations
Methods based on sampling marked birds exist for estimating parameters
associated with survival and movement. Utility of these methods depends on
consideration of the type of data resulting from sampling and the subsequent
modeling of these data in terms of parameters of interest that describe the
processes underlying data generation. When detection probabilities of marked
birds are 1 (when birds are detected at will on study locations), the modeling of
bird encounter histories can be based on the biological parameters of interest
(e.g. survival, fidelity, movement). When not all birds on sample areas are
detected during sampling, modeling is still possible, but the models become
more complicated as they must also include parameters corresponding to recap-
ture and resighting probabilities.
This distinction between data types and their associated models leads to the
simple observation that for studies of equal sample size (equal numbers of
marked animals), mark types (e.g. radios) for which detection probability is 1
will yield more precise estimates than studies using mark types (standard tags and
bands) with variable and unknown detection probabilities. However, radios are
much more expensive than conventional tags and bands, so in cases where it is
134 |Estimating survival and movement