Isotopes have become an increasingly important tool in bioarchaeology, allowing anthropologists to reconstruct diet and origin for archaeological remains. Given the recognized importance of strontium isotopes in geolocation, mapped models of Sr data provide an opportunity to reference predicted data and make preliminary conclusions regarding a population’s origin and migration patterns. Sr values differentiate with geologic patterns in the earth, and Sr is incorporated in human dentition during development when consuming food grown in a certain region. Sr data for this region was collected through meta analysis of archaeological and geological publications, and reflects the “known” points of Sr values. But what about the unknown? This project highlights a predictive isotope model, or isoscape, for the Maya region of Mesoamerica including Belize, Honduras, Guatemala, and El Salvador. This isoscape was created through universal kriging, a form of geostatistical interpolation, and maps predicted Sr values smoothed across the Maya region.
If interested in downloading this poster, please visit this link for the pdf
Hello! I found an interest in bioarchaeology, and specifically in stable isotope analysis and its uses, after working for the past year in Vanderbilt’s BSIRL (bioarchaeology and stable isotope research lab). I wanted to take these interests and apply my newfound GIS knowledge to them, and thus this isoscape was created. Isoscapes have such significant potential for the future of archaeology and also modern forensic analysis, and this preliminary model exemplifies how much data is already out there, just waiting to be analyzed spatially. I found it especially exciting to see my data collection take shape in the form of a predictive model, and am happy to discuss the tools (Strontium isotope values) and methods (kriging) I used to get to this product. Let me know if you have any questions.
Hi Alyssa, How great to see all of those strontium isotope data presented visually. Well done! The difference between the Pacific and the Atlantic (Gulf) coasts is quite dramatic! That bodes well for detecting non-locals from either zone. What do you make of the great geologic diversity in some of those areas, yet the fairly constrained range of strontium isotope values from the humans, animals, and plants?
Hi Dr. Tung, I think there are two important parts to this answer. First, the range of naturally occurring strontium values (which we find by either testing the dentition/bones of archaeological remains, or from ecological sources like plants, animal teeth, or soil) is not very big. For my area of interest, the samples ranged from about 0.70001 to 0.72000, and most ecological sources in this region only range from 0.70 to 0.73, so that is part of the reason there is little variation in broad terms of Sr isotope analysis. Looking at the entire region, there is significant geologic diversity (there are dozens of different patterns represented on that map). Given that Sr values in bedrock, and therefore in soil, plants, and the things that eat those plants, are tied to both the geologic age of the area and the composition of other elements, one might expect different Sr values in every geologic unit. The reality is that these differences are subtle, but most definitely spatially dependent, as seen in your notice of the differences between the Gulf and Pacific coasts, which have fairly different geologic compositions. Further, the uptake of Sr in humans is dependent on the ability to grow food on the land where the Sr is present, so there are likely limits to the geologic units actually included in most studies. Thank you!
Hey Alyssa!
Really fantastic poster! I’m curious about the isoscape methodology and particularly how sensitive it is to less-than-ideally distributed samples used to build the predictive model? I noticed you definitely had some slight clustering in the distribution of your Strontium values (which is totally expected/normal when working with archaeological data) and I was wondering if you felt like that impacted the final isoscape product:?
Separate question – you used both human and ecological samples and I’m curious whether uptake has any variation between species with respect to Strontium (I admittedly know the least about Strontium of all the isotopes we utilize in stable isotope analyses)?
Hi Sylvia! So I originally plotted over double the points you see here, and ended up cutting out any of them in Mexico (ranging from the upper Yucatan to some burials at Teotihuacan), in order to keep my values close together for the analysis. Having big gaps was more of the concern for me compared to clustering, because kriging essentially relies on distance based regression to smooth the surface of the isoscape. I do think however the best possible future isoscape, especially if I were to expand this, would have more equidistant points, or at least make up for the clustered areas with point values within ever ~25 km or so.
To your 2nd question, Sr isotopes do not have a known significant fractionation effect. The Sr value in human dentition is therefore considered to be an accurate reflection of the dietary sources, and those plants/animals are a reflection of the geology underlying them, so that is why I was able to plot ecological samples alongside animals and humans. It is also why Sr is one of the most abundant data sources out there for isoscape production, because you can draw upon both ecological and anthropological publications! Thank you for your interest!!
Fantastic poster! This is not an area I know anything about, so my questions are very rudimentary. How does the outlier principle work in this case? I see that the idea is to remove the outliers from the analysis on the basis that they are non-local in order to look at migration patterns. This seems logical. But are there other reasons why someone might be an outlier? Is there a danger of missing new information by rejecting outliers, or is this method so established that it has already reached the point of being the only/best explanation for differences? Thanks for sharing your terrific work.
Hi Professor Ramey! It is fairly established to operate under the assumption that the “mean” value of an archaeological site is the best representation of that site’s food source, and that outliers are likely migrants or had special diets/do not reflect the average individual. But you are right- there are other reasons to be an outlier, and that is why increased sample sizes are always better; the outlier in a small group (say 5 to 10 burials), could actually be the only “local,” while others migrated there, or could be someone of a higher status/lower status, consuming foods grown elsewhere. Also, there is always the chance that samples are corrupted in the lab, and if we have a really out-there value (say, an outlier with a Sr value of 0.74), we know that something went wrong. I think its safe to say this is the best current process to achieve my goal, but there are always opportunities to ensure better results, such as going out and collecting even more samples! thank you for stopping by and seeing my research!!
Great poster Alyssa! I am curious what led to your decision to go with universal Kriging vs ordinary Kriging.
Hi! I chose universal kriging specifically based upon previous research in Sr isoscapes by Dr. Beth Scaffidi. Her work produced a isoscape for Peru, and similar to my data, showed a global trend in Sr values going to east to west. Because this method was already tested and proven successful, and because I also saw similar trends in my data, I thought it best to use universal kriging. Thank you for stopping by my poster!
Hi Alyssa
Great map, the information you have visually presented here is so important not only for archaeological research on populations but to identify all those victims of the civil war in Guatemala who have been found in clandestine collective burials in different locations of the country. How the data was collected?
Hi Maria! Thank you so much for noting that aspect of isoscapes, their potential in modern forensic situations is one of my primary research interests as I move through school, and I hope that this kind of research can become more accurate and applicable for those scenarios. All of the data was collected via meta analysis of currently published articles, either from anthropological publications on the individual archaeological sites, or from ecological publications in which they did field study (you can definitely see this pattern of walking/driving and collecting samples in Guatemala). Thank you for your comment!
The big picture trend is pretty clear: lower strontium nearer to Pacific/to south and west and higher on the Caribbean. You have strong linear trends in your sample data, which complicates the kriging, but I wonder what your outlook is for producing what you would consider a more definitive reference isoscape model? If you were to design a sampling strategy, how would you do it?
Hi Professor Wernke- I agree that the linear/clustering patterns present in my analysis, simply due to the means and patterns of data collection in both archaeology and ecology is not the perfect input for kriging analysis, and that more random/spread out sampling is necessary to produce a final, definitive model. The lovely thing about Sr though is that ability to draw from ecological sources, so in future isoscape production (which I hope to continue with), I am not limited to the location of archaeological sites to collect samples, and I can even test my samples against ecological data from the same region. If I were able to start from scratch, and use the data collected from meta analysis and also collect my own samples, I would want to take into account the ‘breaks’ in the model, where the predictions have higher likelihoods of error because they are too far from other samples. In a previous reply, I discussed considering simply going out and ‘filling in the blanks,’ to make sure there was a sample within every ~25km radius of another point. I think if I were to present such a strategy to say, someone I wanted funding from in order to complete it, I would need to look further into ArcMap’s tools, as I am sure there is something out there that would pinpoint where I am lacking, even before the kriging process is done. I really think this particular region is suitable for more definitive Sr isoscape production in the near future, given all the work done there already, but continued sampling is 100% necessary and not likely going to happen organically in a non-linear or non-clustered fashion.
Thank you for all your help this semester!
Great job Alyssa! Way to dive into the world of Kriging. It’s a confusing and convoluted but also magical place!
And it looks like you did a great job. The ability to include non-archaeological samples really opens up the possibilities for getting a thorough and valid sampling across the region. I look forward to seeing what you do with it in the future! Please don’t hesitate to reach out as you keep working on it.
As you move ahead, I have a couple of questions/things you might want to think about.
1) How are you thinking about precision when you’re modeling? You say in your comments above that the values across the whole region only vary from 0.70 to 0.73, but your data was 0.70001 on the small end. Can we actually measure strontium ratios to that degree of precision (+/-.00001)?
This is important because if so, 0.70 to 0.73 is a HUGE range, with 3000 possibilities between the top and the bottom value. Alternatively, if we can only measure to the +/-0.01 range then that’s a very tiny range. In this case, the *best* you can really hope to do is to divide the landscape up into 3-4 regions (corresponding to 0.70, 0.71, 0.72, and 0.73). Any more precision than that would be the result of your modeling procedure, rather than real data because we can’t measure the values to that level.
2) Professor Ramey and your discussion actually covered my second question in the comments section above, but I’d like to add a couple of thoughts. You’re absolutely right that using the average for a site is currently standard practice, and it is definitely what you should have done for this preliminary project. But it may be worth exploring other options. The “nugget” in the semi-covariogram when you’re doing kriging is an estimation of how much variation you would expect there to be at a single location. In theory, it would be 0, in practice, of course, it isn’t. By taking the average of all or most points at a site to get a single value you make kriging work, but mathematically you’re hiding a lot of variation. There might be alternatives and if you could work out a good one it would be a significant contribution!
For example, it might be possible to turn a single site into a cloud of points randomly distributed across a small area (say 500 meters), with each point representing a sample from that site. The kriging could then use all of the data, rather than just the mean, and could better estimate the “nugget.”
Hi James, thank you so much for all your help and the encouragement! To answer your first question, the appropriate/reliable measurement of Sr is to the fifth decimal place (so in anthropology and ecology, publications typically report say 0.71121, but if they include more digits, its pretty safe to round off because any more precision is considered unnecessary or unreliable). My original kriging output automatically divided the data into ten natural breaks, so ten groups, but you are right that in measuring to the fifth decimal, that range is technically pretty large, even though the net difference in values is small. I think ten groups of ranges was appropriate for this preliminary model, but in the future that number may change based on larger/smaller ranges in larger/smaller models, and also as more samples are added.
To your second point, I actually really would be interested to see these models displayed in different ways, such as your point “cloud” idea, given that there is natural variation even in the collection of samples from one site, and that would be a more accurate method of not only reporting the values, but referencing them. It would be easier for those trying to use an isoscape for geolocation to be working with a “range” that their value could fit, especially in areas where the values are different from the surrounding region (such as those red ‘peaks’ in my model throughout Guatemala and Honduras.) I look forward to delving into this farther, and hopefully producing something not only accurate but actually useful to those in the discipline!
Oh wow! Okay, so the precision is actually pretty good then! That’s great because it means that there is real potential for differentiating between regions even with such a seemingly small amount of variation.
I look forward to seeing what you do with it! See you around!