The method used to identify the nodes of the graph, in this case representations of important areas for one or more species, is one of the most consequential steps in network analysis. Some papers have done this with predefined sites (e.g., ecoregions or planning units) and others have used more data driven methods (e.g., home ranges or clustering).
I think clustering methods are more compelling for our objectives for 2 reasons:
Since we are trying to identify sites of importance, these may transcend predefined boundaries such as ecoregions, and using a data driven method may yield more novel insight than ranking regions of importance.
Compared to other datadriven methods like home ranges, clustering makes handling datasets of differing resolutions more straightforward. Home ranges are likely more suitable for same species analyses, and when those species are likely to overlap their home ranges substantially, as is the case for flocking species. This may fail to yield insights into sites of shared importance for multiple species.
We will have to decide which scale is most important to our objectives when defining how many clusters we should split the data into.
Here I present the results of clustering using heirarchical methods with complete linkages. Its main advantage over other methods such as k-means clustering is that it is more stable (whereas k-means can produce different results if rerun).
Note! Each point represents the geographic center of one residency period for a given individual. These are colored by species. Obviously we have a greater representation of waterfowl, especially Northern Pintail, which we still need to grapple with.
See different results from same clustering algorithm, split into different number of clusters:
These maps are interactive, so please zoom in with your cursor to explore the clusters.
Definition of residency period: These clusters were all made from the centroid definition of FOUR resident locations followed by TWO transient locations, with each location representing a 12 hour timestep.
There has already been some discussion suggesting that a FOUR resident - ONE transient may be better suited to capture important stops of short duration particularly for Greater White-fronted Geese.
Included species: Cackling geese have also not been included here nor have American Avocets, the first because there have been some data issues that need resolving still, and the second because we have no full tracks from any of the 14 birds.
Ignoring the fact that the clusters themselves may change as we incorporate all species and agree on definition of the residency periods, the key question is:
Which of these options most closely captures the approximate scale we think will adequately address our objectives?
My initial feeling is the right scale is captured somewhere between 150 and 250, with 250 dividing things far too much, but 50 through 100 dividing into too great of an area.
. . . .