Or, Most Maya buildings are on hills. Some aren’t, which is interesting

Luke Auld-Thomas and Marcello A. Canuto

Coming up on one year ago, we published an analysis of landforms and archaeological settlement in the interior central Maya lowlands (Canuto and Auld-Thomas 2021; article access via Academia.edu), with special reference to the Corona-Achiotal region of the northwest Petén. The paper argues that Maya settlement archaeology is strengthened by a quantifiable and generalizable understanding of how archaeological features are distributed across the landscape. In this post, we’d like to elaborate some on that argument, without retreading too much of the ground covered in our paper (besides, the paper’s pretty good. You should read it, or at least glance at the figures and then cite it wherever feels right).   

Mayanists working in the Petén and adjacent parts of the central/southern lowlands have had good descriptive models for where settlement does and does not occur from the inception of the discipline roughly a century ago. While this understanding has been nuanced and contextualized and qualified, the model basically goes like this: “Most Maya buildings are on high ground.” Here is O.G. Ricketson writing in the 1930s:

No Maya constructions (with one exception to be noted later) are found in bajo. Though house-mounds are scattered throughout the area, they always occur on high ground (Ricketson and Ricketson 1937:9).

O.G. Ricketson at Uaxactun. Apparently, in a bajo.

And here is William Bullard, writing in the Year of Our Lord that “Itsy Bitsy Teeny Weeny Yellow Polka Dot Bikini” sold a million copies:

In broad view, settlement fringing lakes and bajos and where aguadas and water sources are common often seems to be virtually continuous. Actually, the seeming continuum is divided into large and small segments by breaks in the terrain and the availability of suitable building sites. Preferred locations were the comparatively level, well-drained, hill and ridge tops at medium elevations, with houses also found on level spurs on hill slopes, the tops of semi-isolated knolls, level areas at the foot of escarpments, and similar places. Where the ground surface has a slight undulation, the houses typically occupy the high points. Areas of steep slopes and rugged terrain were either not settled or very lightly so. Bajos and other areas of swamp or poor drainage were not inhabited, although houses may be found at their very edges. In sum, the distribution of settlement appears conditioned principally by the occurrence of sufficiently large tracts of well-drained relatively level terrain within a kilometer or so of a water source. (Bullard 1960:365)

William Bullard.
Image Source: Willey, Gordon. “WILLIAM ROTCH BULLARD, JR., 1926-1972”. American Antiquity 38(1), 1973.

These descriptions are powerful and useful, and each is cited in basically every grant application and graduate thesis chapter dealing with survey and sampling methodology right down to the present. But they do have their limitations, and these turn on items like Ricketson’s “one exception to be noted later.” That is, there are departures and deviations from every rule, and these are often subjects of great analytical interest. Why are there some buildings in bajos? Are there other landforms that were avoided as building sites—and if so, why? What can the distribution of ancient buildings across landforms tell us about ancient Maya society? Most buildings are on hills, some aren’t: so what?

That brings us to geomorphometry, “the science of quantitative land-surface analysis” (Pike et al. 2009:3). Now, “quantitative” needn’t imply anything terribly sophisticated. We don’t necessarily advocate for archaeologists to rush headlong into Bayesian topographic inference (yes, it’s a thing). But it is important to measure and model things, because it allows for direct comparison, explicit hypothesis testing, and so on. Mayanists have been thinking and writing about forest type and soil class in this regard for a long time, using each as a quantifiable guide to where and how people might have lived. But it warrants noting that these are both dependent variables, controlled primarily by topography and climate (and neither, unfortunately, is particularly generalizable beyond small study areas). A pair of quotes from a classic ecological study of the Maya forest captures the relation between topography, soils, and vegetation succinctly:

In this manner topography and associated edaphic characteristics combine to create impressive changes in the vegetation over short distances. As soil characters and topographic position are strongly correlated vegetation characteristics are typically similar along the same regions of the topographic gradient at different sites, tempting designation of “forest types” (Schulze and Whitacre 1999:174).

To which they add: “To a large degree, forest type [is] merely a condensation of topographic positions” (180).

Geomorphometry, then, is the basis of other conditions often used to explain
settlement patterning.

What kinds of things can we do with geomorphometry?

Let’s begin with a land surface parameter that some Mayanists have worked with already: ruggedness (or roughness). Obviously, this is an important condition for settlement growth and land use considerations. Here, something like John Lindsay’s wonderful Multiscale Roughness analysis (Lindsay et al. 2019) gets to the heart of the matter: in extremely rugged landscapes, mesoscale surface roughness imposes a major constraint on settlement suitability. At a place like Caracol, for example.

Source elevation data: GLO-30 DEM, courtesy of ESA. Computed with WhiteboxTools.

For those who prefer a map legend: the yellow bits are rugged, the black bits are not, the white lines are causeways, the part that says “Caracol” is downtown Caracol.

Or, maybe you’re, like, really into bajo-margin agriculture and wish there was a straightforward way to map colluvial deposits along footslopes. Friend, don’t reach for the spectral classifier, because geomorphometry is here for you. Here’s a map of footslopes, in cyan, extracted using the geomorphons classification. You could do something similar using any number of surface curvatures.

I assure you that everything is not as ixim -s

Source elevation data: GLO-30 DEM, courtesy of ESA. Computed with WhiteboxTools.

You can repurpose that same geomorphons classification to extract other relevant landforms, like…wait for it…”level spurs on hill slopes, the tops of semi-isolated knolls, level areas at the foot of escarpments, and similar places.” Or, just highlight the flat areas, which has some analytical utility in itself.

Source elevation data: GLO-30 DEM, courtesy of ESA. Geomorphons computed with WhiteboxTools.

So, we now come to our settlement suitability model; we gridded the lowland Maya landscape into four broad landform categories, demonstrating that ancient Maya builders consistently and across a very wide area conferred the same suitability rank to these landforms. We then showed that settlement growth led to “spillover” into marginal landforms faster than it led to the colonization of higher-ranked landforms at greater distance. There are some interesting implications for the environmental vulnerability of Maya cities, and the social vulnerability of their inhabitants, although we lacked the space to get into these fully.

Our landform classification was a straightforward adaptation of the Topographic Position Index, originally published by Weiss (2001) and implemented in the handy Relief Analysis Toolbox, which you can download here (it has a few other fun tools, like a scale-optimized hillslope position classifier for high resolution DTMs).

TPI is essentially a local elevation z-score that is highly sensitive to analytical scale. What exactly does that look like in practice?

This is a part of the Achiotal region in northwest Guatemala, visualized with Sky View Factor and a simple DEM color ramp (this is a bajo, encircled on three sides by a low escarpment and sprinkled with residual karst hills throughout).

Source elevation data: 1m lidar-derived DTM, courtesy of PACUNAM.
Computed with RVT and WhiteboxTools.

Here is a TPI model of the same region, computed from the same elevation data with a small analytical window of 101 x 101 meters.

Source elevation data: 1m lidar-derived DTM, courtesy of PACUNAM.
Computed with WhiteboxTools.

And here is a TPI model computed using a 1001 x 1001 meter window.

Source elevation data: 1m lidar-derived DTM, courtesy of PACUNAM.
Computed with WhiteboxTools.

As you can see, the small analytical scale picks out local dips and rises but doesn’t really provide any distinction between broad-scale uplands and lowlands. The larger scale captures these but blows out most of the finer distinctions. By combining them, following Weiss (2001), we can have the best of both.

Source elevation data: 1m lidar-derived DTM (downsampled to 5m), courtesy of PACUNAM.
Computed using Relief Analysis Toolbox.

For our analysis, we downsampled our lidar-derived DTM to 5m and calculated TPI using 300 and 3000 meter radii. We also computed slope using the same 5m DTM. We converted both TPI rasters into 3-part categorical models using standard deviations and combined them with the slope model to produce a 10-part landform classification (that’s 3 small-scale TPI classes times 3 large-scale TPI classes, with one class further subdivided using a slope threshold; for those who are so inclined, you can do the whole thing in a raster calculator using logical operators, and Weiss’s poster provides the code to do so. The Relief Analysis Toolbox provides a GUI for the same thing, and the back-end math is a little simpler). We then simply collapsed these 10 landforms into 4 “super-classes” that reflect the way people in Peten talk about their own landscape. That extremely simple procedure revealed that about 90% of the ancient buildings in northwest Peten are found on landforms representing less than 20% of the total land surface (and yes, we checked on the ground).  Put another way, we found that only about 20% of the region was deemed “desirable” as places to live and gather by ancient Maya people.

La Corona, bustling metropole, lidar data courtesy of PACUNAM.

For us, this raises some interesting questions about the famously “dispersed” and low density nature of Maya cities, which is often chalked up to a combination of climate and agricultural practice. But raw topography clearly plays a very important role in structuring settlement across spatial scales, and this deserves more sustained and explicit attention.

At risk of belaboring the point, we close by noting that people live on the surface of the
earth, and if we want to understand the processes and historical contingencies that
resulted in (some) people living in some places and not others, then we need
geomorphometry to standardize our comparisons and make our analyses spatially

Besides, geomorphometry is also fun! No, really:

This is not a lidar-derived geomorphometry model.
It is in fact a visualization of your own consciousness.


If you’d like to start smashing things together to see what you learn about settlement and landscape in your own study area, we recommend the following software to get you started.

WhiteboxTools, created by John Lindsay at the University of Guelph, is loaded with analytical tools and features excellent documentation; it’s also open-source and available for free with the option to pay what you like. It can be called from R and Python and has great, user-friendly plugins for ArcGIS and QGIS.

GRASS has a number of excellent geomorphometric analyses, including the original implementation of geomorphons. It has its own GUI, and also integrates closely with QGIS.

SAGA-GIS is an unsung hero of geomorphometric and hydrologic analysis, with tools that are not replicated in other software packages.

Finally, the Relief Analysis Toolbox provides (for ArcGIS users) an easy GUI for TPI-based landform classification and hillslope position classification.

Cited references

Bullard, William R.
1960       Maya Settlement Pattern in Northeastern Peten, Guatemala. American Antiquity 25(3):355-372.

Canuto, Marcello A, and Luke Auld-Thomas
2021       Taking the high ground: A model for lowland Maya settlement patterns. Journal of Anthropological Archaeology 64:101349.

Lindsay, John B, Daniel R Newman, and Anthony Francioni
2019       Scale-optimized surface roughness for topographic analysis. Geosciences 9(7):322.

Pike, R. J., I. S. Evans, and T. Hengl
2009      Chapter 1 Geomorphometry: A Brief Guide. In Geomorphometry – Concepts, Software, Applications, pp. 3-30. Developments in Soil Science.

Ricketson, Oliver Garrison, and Edith Hill Bayles Ricketson
1937        Uaxactun, Guatemala. Group E–1926-1931. Carnegie institution of Washington Publication, Vol. no 477. Carnegie institution of Washington, Washington.

Schulze, Mark D., and David F. Whitacre
1999        A Classification and Ordination of the Tree Community of Tikal National Park, Peten, Guatemala. Bulletin of the Florida Museum of Natural History 41(3):169-297.

Weiss, Andrew
2001      Topographic Position and Landforms Analysis. Paper presented at the ESRI Users Conference, San Diego, CA.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s