Main content area

A canonical correspondence analysis of the effects of the exxon valdez oil spill on marine birds

Wiens, John A., Crist, Thomas O., Day, Robert H., Murphy, Stephen M., Hayward, Gregory D.
Ecological applications 2001 v.11 no.3 pp. 828-839
correspondence analysis, data collection, environmental factors, habitats, multivariate analysis, oil spills, summer, surveys, water birds, Alaska
To assess how the Exxon Valdez oil spill affected habitat occupancy by communities of marine‐oriented birds in Prince William Sound, Alaska, we conducted a canonical correspondence analysis (CCA) using data collected between 1989 and 1996 in 10 bays that had been exposed to different levels of oiling. CCA creates a multivariate space defined by a combination of environmental variables and measures of the abundances of the bird species present in the bays. The locations of individual sites (site scores) in this multivariate space indicate how bird‐community composition at the sites varies in relation to the combination of environmental variables, while the locations of individual bird species (species scores) indicate the mean values of the response curves (abundance variations) of species on the CCA axes. Relationships among the site scores of sites exposed to different levels of oiling may therefore be used to evaluate the effects of the spill on the bird communities occupying these sites, and the relationship of a species score to the site scores indicates the conditions under which the species is, on average, most abundant. In surveys conducted the summer after the oil spill, the site scores for unoiled or lightly oiled (“unoiled”) bays occupied a region of the CCA space that was largely separate from that occupied by the moderately or heavily oiled (“oiled”) bays, indicating consistent differences in bird‐community composition. Although the structure of the CCA ordination and site locations in that space varied among seasons and years, the separation of the unoiled and oiled site scores in the CCA space continued. Rather than indicating continuing spill effects, however, the separation of site scores was probably related to environmental differences among the bays and chance differences among bays in the reassembly of bird communities following the spill. Consideration of the locations of species scores in the CCA space clarified the patterns of spill impacts and subsequent recovery. The increase in the number of bird species present in early summer surveys between 1989 and 1990 was associated with the occupancy of previously oiled bays by species that were absent in 1989 or that previous analyses had shown to be initially impacted by the spill. Species scores for birds for which the previous analyses showed continuing spill impacts in early summer 1990 were associated most closely with the site scores of the unoiled bays. The species scores of several species that occurred too infrequently to be evaluated in the previous studies were associated with the site scores of the unoiled bays in 1989, suggesting initial spill impacts. In subsequent years, these species scores were associated with the site scores of the previously oiled bays, suggesting reoccupancy of those bays. Patterns for surveys conducted in midsummer generally were similar to those in early summer. Surveys conducted in midsummer 1996 suggested continuing recovery in the use of initially oiled habitats by birds, including some species that did not show clear evidence of recovery in 1991. Overall, these analyses showed that the Exxon Valdez oil spill had clear initial impacts on bird species and communities in the study bays in Prince William Sound, but they also provided clear evidence of increasing occupancy of previously oiled sites. Multivariate analyses such as CCA can provide valuable insights into the complex responses of environments and biological communities to large‐scale contamination events when precontamination data are lacking, especially when the analyses are repeated over time and the changing relationships of both species and site scores in the multivariate space are considered.