Apus, or sacred mountains simultaneously considered earthly spirits, local deities, and ancient ancestors, were one key component in the Andean conception of a vibrant and sacred natural landscape. Often existing as parts of hierarchical networks with other apu and huacas, studies have shown that apu have also been used to manufacture political legitimacy by building sites in places that maximize the visibility of apu networks (Williams and Nash, 2006). However, there has not been much written on the treatment of apu in the early colonial period. Because Spanish and church officials sought to break Andean spiritual traditions during the early colonial period, it stands to reason that one tactic would be removing access to apu, both physically and visually. This poster aims to help answer this question by comparing the viewsheds of Late Intermediate Period (1000-1438CE) and Late Horizon (1438-1532CE) sites near the apu Coropuna with the viewsheds derived from randomly generated points.
9 thoughts on “A Preliminary Viewshed Analysis of Archaeological Sites Near Coropuna from GeoPACHA Data”
Comments are closed.
I also wanted to add that there are many interesting and necessary future steps for this project! The first of which is to run a Monte Carlo test with the data. Other future steps include creating new suitability polygons by which to generate random viewpoints and viewsheds in this region, generating individual viewsheds from local reducciónes to Coropuna, and ground truthing the current results to establish if the analysis correctly predicted visibility.
Hi Sam, Great to see this viewshed analysis, especially with the emphasis on the importance of the apus. You note that the LIP/LH sites could not be easily seen by neighboring sites and suggest that that may have been an intentional strategy to avoid visibility by neighbors. However, what do you make of the suggestion that sites near each other in this valley (and other similar Andean valleys) needed to see each other in order to “communicate” (eg, with smoke signals, per Arkush and per Wernke)?
Hi Tiffiny, thank you so much for your comment. I do think that Arkush and Wernke have strong arguments for the necessity of local-level visual relationships. That is why I was so confused when the RPV returned with 101098.5km2 as the total area seen by between 1 and 9 viewpoints. This is something like 35 times the area generated by the LH/LIP viewshed’s 1-9 viewpoints. The fact that the SRPV value was also much lower at 9051.35 km2 makes me think that I need more random sample data. Is there an outlier? If so, which one might it be, as the LH/LIP viewshed seriously surprised me in the other direction by containing the most area seen by 25 viewpoints or more. I had theorized initially that one of the random viewshed tests would see more of the mountain as LH sites may have been built to minimize viewsheds to apus. My data is so preliminary it doesn’t preclude this hypothesis, but it does make me take a hard look at my data (for example the interdeterminancy of whether a site dates to the LIP or LH) and the methods by which I constructed my suitability polygon. There are plenty of other parameters with which I can calculate a polygon, such as Marsh and Schreiber’s 2015 viewshed suitability polygon restricted to 2900-3800masl (and no slope affordance).
Hi Sam! This was super interesting to read! I’m curious if you have any hypotheses about why the SRPV had lower visibility than the RPV when you expected it to be better? It’s an intriguing anomaly, and I’m wondering if it also ties in to why the visibility between LIV/LH sites was also low…is there something about the geography of that area where sites that were more ‘suitable’ (flatter and not too high up – also good attributes to consider when choosing settlement location) unfortunately just didn’t have the visibility to other sites that may have been desired (per Tiffiny’s comment above per Arkus/Wernke)?
HI Sylvia, thank you for your question. Yes, in nearly every way this data did not do what I expected!
You can see the upper left hand corner of the RPV has those radial beams that extend into the northwest. There is a chance that, as it can happen when points are generated randomly, two or three points were placed in that area at the top of a rise and able to look down over a very large and gently sloping planar feature. That area isn’t seen in the LH/LIP or SRPV viewsheds, so I wonder if it might be something to do with that. Or, 2300masl was too low a lower limit for my suitability polygon and too many points were inadvertently in the streambeds of higher elevation streams and thus unable to see larger features like Coropuna. This all said, my data seems like a great motivator to go back to the drawing board with my suitability analysis and try parameters used by scholars in other regions!
Hi Samantha! This project is very interesting, and I enjoyed reading your response to the above comments, as I had some similar questions regarding your thoughts on these surprising results. My other question has to do with your choice of technology, though; did QGIS offer anything different or additive in regards to viewshed analysis, or was the choice a personal preference? Given the QGIS is open source and therefore more accessible to the average person, I was wondering what your experience was and choice for use in your specific project. Thank you!
Hi Alyssa, thank you for your question! I started the project in Arcmap but had a litany of issues when running the viewshed tool- namely GPU/processing errors. I moved to QGIS to create my suitability raster and run my viewsheds instead, and I had no errors there of any kind. (Plus, when you do have errors in Q the error screen is usually quite helpful.) For this project, I used QGIS to run all my data and then moved it into Arc at the end to create the layouts. While I like QGIS overall more, I have to admit that Arc has a better interface for the “make pretty” part of map-making, especially for people who are still learning. You should try it out!
You have a good start to a systematic viewshed analysis, as you are comparing against a null hypothesis distribution. I have many questions, but will restrict them to top level q’s here: 1) Why might late prehispanic settlements be keying to visibility of Coropuna? What about other apus in the region? You are capturing settlements from more than one valley and some are more proximate to other regionally-paramount apus such as Hualca Hualca and Ampato. What might happen if you added those to the analysis (i.e. included all regionally paramount apus–which would be a little tricky to define, but could be done)? Why run it against both suitable and non-suitable null hypothesis distributions? If an area is unsuitable for human settlement (implied in the suitability analysis) then it doesn’t seem warranted to run a viewshed that includes areas unsuitable for habitation). Or perhaps I’m misreading it? Your tables get the point across, though a visualization showing the distributions would be preferable, such as boxplots of viewshed areas of settlements vs. SRPV points. What is Table 2 presenting? I’m not sure I understand what “average number of points able to see Coropuna means” if each site is one point. Or did you use a polygon (circle) around a point and this is the average number of points per site from which Coropuna is visible?
Hi Steve, thank you for the great questions and comments! I will try to address each larger question below.
1. My guess is that late prehispanic sites were located within Coropuna’s sightline for a number of possible reasons: capitalizing on traditions of mountain worship as way to generate legitimacy, desire or convenience of building near settlements from previous periods that had good sightlines, and the general visibility of Coropuna from all nearby flatland areas around 3000masl. The list goes on. Comparing Coropuna’s visibility statistics to other apus in the region is a really good suggestion. I would be particularly interested to know what sites see all three as well as how visible each apu is to, say, the sites within a 20km or 30km radius. How much does visibility improve the close you get? Does it improve at all? Is there enough of a change for use to draw any real conclusions?
2. I ran a non-suitable null hypothesis distribution fairly late in the game when I had already gotten the results for the suitable and late prehispanic viewsheds. Those two viewsheds seemed so wildly dissimilar (and in the opposite direction of my hypothesis) that I became concerned that something might have gone wrong in the creation of the suitability raster. I created the non-suitable viewshed in an attempt to establish how far apart my two results actually were. For example, did the non-suitable viewshed have even fewer areas seen by 25+ points? Did it cover less land overall than the other viewsheds? On both counts I found that the non-suitable viewshed was in between the values of the late prehispanic and suitable viewsheds, which indicated to me that I might need to go back to the drawing board in terms of my suitability raster.
3. Table 2 is presenting a zonal statistics calculation of a polygon drawn along a high-elevation contour line of Coropuna. I then calculated the average number of points within the polygon for each of the viewsheds (since they were cumulative, not binary). The table shows the results of this operation: in the late prehispanic viewshed of 105 points, only an average of 13.75 were able to see Coropuna. The null hypothesis viewsheds fared much worse at 3.49 (non-suitable) and 0.96 (suitable). Thank you to James for the suggestion to use zonal statistics!