In the North Pacific, humpback whales feed in various locations along the Pacific Rim including in the US, Canada, Russia and eastern Asia during summer. In winter, they migrate south to mate and calve along Pacific coasts as well as the offshore islands of Mexico, Hawaii, and Japan (including Ogasawara and Ryukyu Islands). Fidelity to feeding areas is high, and is thought to be maternally directed; mothers take their calves to their specific feeding ground, and these offspring subsequently return to this region each year after independence.
This maternally directed fidelity is reflected in studies of maternally inherited mitochondrial DNA (mtDNA). In an ocean-wide survey of genetic diversity and subsequent analysis of population structure in North Pacific humpback whales (Structure of Populations, Levels of Abundance, and Status of Humpbacks; SPLASH), sequencing of the mtDNA control region resolved 28 unique mtDNA haplotypes showing marked frequency differences among breeding grounds (overall FST=0.106, p<0.001, n=825) and among feeding regions (overall FST=0.179, p<0.001, n=1031; Baker et al. 2008).
Despite genetic evidence of regional population structure in the North Pacific (i.e. separation of humpback whales into various stocks), there have been few studies to investigate the possibility of finer-scale structure within a single North Pacific feeding ground. For example, it is unclear whether maternally directed site fidelity at smaller scales within southeastern Alaska results in discernible differences in haplotype and sex frequencies.
For my final investigation in this course, I decided to look at fine-scale population structure of humpback whales in southeastern Alaska by exploring spatial patterns in haplotype and sex distribution. Specifically, I wanted to answer the following questions:
- Are haplotypes (A+, A-, E2) differentially distributed by latitude?
- Are males and females differentially distributed by latitude?
- Are certain maternal lineages more spatially clustered than others?
- Are males or females more spatially clustered?
It appears as though there is a peak in percent of sex and haplotypes observations between 56.6-58.5 degrees. After looking closer at this, I realized that this peak is a function of my bin selection. After visualizing my population distribution within each bin, it is clear that most of my encounters occurred between 56.6-58.5 degrees. However, there are some patterns in differential class variable percents. For example, more A+ haplotypes are found near 58 degrees than A- and E2 haplotypes. Also, the E2 haplotype seems to be more represented at lower latitudes than A+ and A- haplotypes. Males and females seem to be fairly similar in their latitudinal distribution.
Nearest Neighbor Analysis:
Hi Sophie,
Good stuff! I have three thoughts from your post. First, what if you did a similar haplotype and sex distribution analysis but only for the subset of points found between 56.6-58.5 degrees using a much finer scale for your bin selection? It would be interesting to see if that teases out any new patterns that you cannot see in your current results. Second, I agree that re-running your nearest neighbor analysis using a fixed spatial extent would be interesting. This is something I am considering doing, too and we could tackle it together sometime. Finally, I think it would be worth finding a bathymetry layer and looking for distribution patterns based on depth. If I remember correctly, southeastern Alaska has more interesting bathymetry than the relatively flat Gulf of Alaska region and this might produce some interesting results.
Cheers!
Dori
Nice work, Sophie!
I think I may have mentioned this in passing during your presentation last week, but I thought I’d follow up with more detailed information. You concluded your post by mentioning that each variable has its own study area, and that you would like to establish a standardized study area to investigate how your values vary across such an area.
Building off of Dori’s comment above, my understanding with Nearest Neighbor Analysis (and Cluster Analysis) within ArcGIS is that it generates a “bounding box” or “universal polygon” based on the maximum x & y extent of your data. There are many potential problems associated with this, particularly around the edges of the data extent, but also when comparing different variables with varying extents. One method you could try to standardize this area is to use the “Processing Extent” option under “Environmental Settings”. You’ll find the “Environmental Settings” button at the bottom of the Nearest Neighbor Dialog Box once you open it up in ArcToolbox.
There are a number of different ways to define the extent – my suggestion would be to start by creating your own rectangular polygon based on the complete extent of the entirety of your data. Make the polygon slightly larger than the complete extent, say by a margin of 1km, 5km, or 10km, and see how your results differ from the default setting that you’ve already run. Then, at that point, you could further tweak the size and extent of your customized “extent” polygon to evaluate your results.
I’m emailing you a screenshot of the “Processing Extent”, because I cannot seem to paste it in here…
-Doug
Hi Sophie,
Nice work and great observation of the potential bias when defining bins.
I also appreciate the discussion on the bounding box since I have the same issue. I’m also not sure on how to adjust for landforms or the way in which the encounter data was collected.
Any thoughts?