Salmon are a crucial cultural keystone species across most of the Northwest, of very high importance to many coastal and interior cultural groups for thousands of years. Many archaeological sites are chock full of salmon bones, and the oldest of these are around 10,000 years old. As a cultural story, the importance of salmon is obvious. Increasingly though, the archaeological data are also invoked to tell the history of salmon themselves. The very long-term view of the archaeological record provides knowledge of their ranges, their relative abundance, their life histories, etc. These data can then be harnessed as part of conservation and fisheries management of these threatened species of fish.
Each species offers different things to people: some run early, some late; some are more fatty, some leaner; some run in huge compact numbers, others tend to dribble by; some can be caught in large numbers in the open ocean, others can only be caught efficiently in streams. There’s just one problem with using salmon bones in archaeology: until recently, you really couldn’t tell one species of salmon from another based only on their bones.
This roadblock was lessened somewhat by recent advances in extraction of Ancient DNA from salmon bones, a research direction led by SFU’s Dongya Yang. This allows very confident assignment of a bone to a single species. Where applied, such as at Keatley Creek on the middle Fraser, it has strongly challenged previous analysis and indeed, essentially inverted some of these results (see this MSc thesis by Camilla Speller, for example, and table below; also Speller et al 2006 PDF). In the process, the aDNA analyses cast a lot of doubt on the use of x-rays to determine growth increments in vertebrae, which were thought to be reliable indicators of at least some species differences (mainly that Pink salmon spawn at 2 years of age, while other species mostly wait four – but aDNA showed many 2 and 3 year old sockeye at Keatley Creek).
Meanwhile, only a few head-bones can be identified to species, and many archaeological assemblages are typified by an hyper-abundance of vertebrae, reflecting processing practices and seasonal storage of the meaty bits,. Identification of these head-bones is difficult and of arguable reliability.
Anyway, aDNA extraction is expensive, and while it may be great to know that some of the bones from your site are, say, sockeye and others are pink, typically this might amount to a hundred or so total bones which you could assign to species. However, it is no exaggeration to say that you might actually have on the order of tens of thousands of salmon bones in your assemblage, and the identification of the aDNA sample tells you nothing of all these other specimens.
So, now the point: a new paper in the Journal of Archaeological Science shows how very simple measurement of two variables on salmonid vertebrae enables confident division of your salmonid bone assemblage into four groups based on length:width ratio (not absolute size). This is enabled through a statistical technique known as Classification and Regression Trees (CART). The four groups it can differentiate with success rates of between 92% and 100% are:
- Chum and Coho Salmon and Steelhead Trout
- Sockeye and Pink Salmon
- Cutthroat Trout
The paper by Harriet Huber et al. is not available online except via subscription, though you can see the abstract here. I usually try to avoid posting on material not freely available on the web. However, these robust results, while still leaving some unfortunate lumping of distinct species, seem to me to be potentially a massive step forward in zooarchaeological methods on the Northwest Coast. Salmonid vertebrae can themselves be distinguished from other fish very easily and reliably with little training, and the measurement protocol seems simple and replicable. This method should, therefore, be a simple and inexpensive way to extract much more information about this crucial fish taxon, which is currently under so much pressure.
Harriet R. Huber, Jeffrey C. Jorgensen, Virginia L. Butler, Greg Baker and Rebecca Stevens 2011.
Can salmonids (Oncorhynchus spp.) be identified to species using vertebral morphometrics?
Journal of Archaeological Science, Volume 38, Issue 1, January 2011, Pages 136-146
Thanks Quentin for sharing the news about this new morphometric approach to salmon vertebra species identification. Jeff Jorgenson, the stats guru who led the way with the complex statistics, has posted some information about the free downloadable software, “R” that one can use for this, along with the data file you’ll need. see : http://conserver.iugo-cafe.org/
If you type in a few obvious keywords (“salmonid”, “vertebrae”, or “identification”, for example) you should navigate to it pretty quickly. Jeff put two files there, one is an R workspace that has the two CART models for Types II and III vertebrae identification. The other file is a text file with some explanation of things, and an example of how to extract vertebrae sample identifications from the models.
To learn more about using “R”, there are actual you-tube tutorials.
Hi Virginia! Thanks for stopping by the NW blog – I think we’ve discussed your work more than once, most recently here:
I’m really looking forward to hearing more about this salmon vertebrae study – simple salmon speciation is one of the holy grails of NW Coast archaeology, and even while I discourage people from being blinded by salmon as the be all/end all of NW Culture, I’m not about to claim they aren’t the amazing keystone species which they clearly are. The more we can tell about them from the archaeological record, especially in this more nuanced way of some finer taxon distinctions, the better.
Oh, and I’ve tinkered with the R stats package, so thanks for the links.
For the uninitiated, “R” is free statistical package which does pretty much everything that better known, very expensive packages such as SPSS and SAS can do. It is open-source (think R=Firefox vs SPSS=Internet Explorer) as well, so is constantly updated and tweaked. (SPSS, for example, costs about 200$ even with academic discount, for a mac, and in OS/X it was as buggy as a salmon fry in the Broughton archipelago. Institutionally, they have switched to licences, meaning the package always seems to be expired just when you need it, and maybe the department only buys one licence so it is on one computer. So, R is very welcome.) I can’t claim to know much about it but since it seems to be gaining a lot of momentum, it’s worth talking about for a second.
There is a large community of stats geeks who, because it is open source feel a sense of ownership in the project and develop add-ons and tweaks.
While it is still true that SPSS and SAS dominate academia, and it is always true that the best package to use is the same one as the friendly stats nerd down the hall uses, I do have some links to dump here, in particular, people have devoted time to creating user-friendly graphic front ends (GUIs) for the R stats engine:
LINK DUMP ALERT
The R-project itself: http://www.r-project.org/
You can download a copy here, among other places: http://cran.stat.sfu.ca/
A good R blog for SPSS and SAS users: http://www.statmethods.net/
And another one, more advanced: http://www.r-statistics.com/
R for SPSS and SAS users is a free 80 page book, basically a draft of a longer, 75$ book: http://sites.google.com/site/r4statistics/free-version
Deducer GUI looks like maybe the best one?: http://www.r-statistics.com/2010/10/r-gui-now-offers-interactive-graphics-deducer-0-4-2-connects-with-iplots/
R-commander GUI: http://socserv.mcmaster.ca/jfox/Misc/Rcmdr/
JGR – a Java GUI: http://www.rforge.net/JGR/
Thanks for the interest in our work, and your nice post about it in your blog. I wanted to let interested folks know that I have updated the files posted to the website:
The R workspace now includes species ID models where species were grouped according to morphological similarity. Users of these models will notice that model predictions from these models will come in the form of a number rather than a species name, and each number corresponds to a species or species grouping:
Group # Species or species grouping
1 Chinook salmon
2 Chum, coho, steelhead
4 pink, sockeye
This is covered in the R script file available at the above web site.
Thanks for posting this – having analyzed a massive fish collection from Kodiak Island, I am anxious to try this method on our salmon species. I have used Bruce Finney’s work that derives sockeye abundance from lake cores on Kodiak to compare to the archaeological record, but have always had to make the assumption that his records generally represent all the salmon populations. Now I can actually test this idea!
Pingback: Mysterieus bot gevonden - Bodynet.nl - Bodybuilding & Fitness discussieforums