To investigate the hypothesis that the individuals of each species are distributed within the park in a characterist ic way, I estimated thedistribution of a number of species based on all the recorded sightings of individulas from the years 1990-1993. I used the R routine kde2d (2 dimensional kernel density estimation) on each species for which there were more than 150 recorded sightings. The figure shows 8 of these; higher densities are represented by yellower and whiter colors and lower densities by redder colors.
In these plots the horizontal coorinate is (roughly) the east-west location and the vertical coordinate the nort-south location, as measure in inches from the lower southwest corner on the standard map used on censuses to record bird locations.
The pairwise correlations c were found for the densities, and 1-c was used as a measure of distance for agglomerative clustering using average linking ( as implemented in the R routine hclust). The resulting dendrogram:
illustrates the “relatedness” of the geographic distributions of the species. Species with very similar distributions appear as close relatives in the tree while those with very different densities are presented as distant relatives.
Several checks on the reasonableness of this procedure were included. First, Song Sparrow (sosp) was included even though it is not a migrant species; the park usually contains one breeding pair (nesting in the southeast corner). The sightings of a breeding pair should produce a vary different distribution than sightings of migrants. As shown in the dendrogram, the sosp is the species with distribution least related to any of the others.
Records of Northern Flicker, Yellow-rumped Warbler, and Black-throated Green Warblers were each originally coded in two different ways (either as ysfl or nofl; as yrwa or mywa; and as bnwa or bgwa) by different observers or in different years. This original coding was retained in the analysis to make these species appear as two distinct taxa. Note that the dsitribution of yrwa is most closely related to mywa, nofl to ysfl, and bnwa to bgwa. Not only do these similarities suggest that the procedure used actually detects relatedness of the spatial distributions of the species, it also serves to calibrate the vertical scale (labeled Height) in the dendrogram by showing the magnitude of differences that can be attributed to random sampling from the same distribution. The exact pairing of the two codes representing the same species appears only when the clustering uses the average linking procedure; under single and complete linking the two repesentatives cluster near each other but are not always nearest neighbors.
Some of the clusters in the tree may be attributable to similarities in habitat selection, resource utilization, or other behaviors across species. The cluster of ground feeding birds found in open areas (chsp,deju,fisp) is such an example, as is the cluster of high canopy dwellers (bpwa,revi,tewa), and the cluster containing the mimids (grca, brth).
Other clusterings seem anomolous and may reflect the fact that other factors also influence the distribution of birds in the park. For example, the fact that wtsp and yrwa are similar might be partially due to the fact that many of the records for these two species are from early in the season when the location of all individuals may be influenced by strong winds and the relative lack of vegetative cover. The relatively isolated howr has a distribution that probably reflects the locations of several breeding paris (3 to 5 pairs nest int the park each summer) as well as migrants.
Try removing the effects of weather and season by blocking the data with respect to these variables.
Instead of estimating densities, try subdividing the park into regions that are relatively homogeneous wrt vegetation and looking at the distributions of each species across these regions. Such data could also be clustered, or if some characterization of the subregions were available, we could search for factors that best account for the presence of each species.
The distribution of one species could be used to predict the presence of another. By doing this on a census by census basis, we can search for negative effects.