As was the case for the single-location model in which the focus was on
survival estimation, the capture history data and the corresponding probability
model (each history has an associated probability as above) are combined to
form a likelihood function, and estimates are then obtained using software
such as MARK (White and Burnham 1999). The general model with time- and
location-specific parameters can be constrained in various ways. For example, it
may be that movement between pairs of locations is expected to be symmetric
( ). Covariate modeling can be used to investigate biologically interest-
ing hypotheses, as movement between two locations can be modeled as
a function of such factors as distance between the locations, the ratio of fitness
indicators between the locations, and density at the locations (Nichols and
Kendall 1995). Although the focus of this section is on movement, we note that
location-specific survival probabilities can also be estimated using multistate
modeling. These provide survival estimates in situations where animals move
among locations and where survival may vary over locations. Constraints involv-
ing survival (e.g. ) and covariate modeling of survival are also frequently
of biological interest. Again, competing models expressing different hypotheses
of biological interest about movement or survival can either be tested using
likelihood ratio tests or evaluated using a model selection approach (Burnham
and Anderson 2002).
The parameterization described above is most useful when bird movement
between locations occurs near the ends of the interval separating sampling peri-
ods. Although this approach seems reasonable for migratory birds (e.g. Hestbeck
et al. 1991; Spendelow et al. 1995) returning to breeding or wintering locations
at the end of each sample year, there are other situations where movement may
occur at any time during the interval. In such cases, it is possible to parameterize
with transition parameters that combine survival and movement, ,
thus requiring no assumption about the timing of movement. Another modeling
approach is to view time of movement as a random variable with known distribu-
tion (Joe and Pollock 2002), although user-friendly software for implementing
this approach is not yet available.
5.4.3Band recovery
Band recovery data can also be used to draw inferences about bird movement.
We generally envisage two sampling situations. In one, banding and recovery
occur at different times of the year and at different locations. For example, in North
America, it is common for banding of ducks to occur on the breeding grounds,
whereas the hunting season recoveries occur during the fall and winter. In this
situation, inference is sometimes possible about movement from a particular
iRSSiRiRSSiRSiSiRSiSRMovement| 133