Uncovering Spatial Patterns in Archaeological Data through the Integration of Self-Organizing Maps, Principal Component Analysis and Hierarchical Clustering
This study demonstrates that integrating Self-Organizing Maps (SOM) with Principal Component Analysis and hierarchical clustering enhances archaeological landscape analysis by not only confirming robust settlement groupings but also revealing continuous spatial gradients and transitional sites that traditional rigid partitioning methods often overlook.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Archaeology is fundamentally an act of sorting. For centuries, researchers have taken the vast, messy collection of things found in the ground—pottery shards, stone tools, the locations of ancient villages—and tried to arrange them into meaningful groups. This process helps them understand how past societies were organized, how they moved, and how they interacted with their environment. However, when the data becomes complex, involving many different measurements at once, such as the height of a hill, the steepness of a slope, and how far a site is from a river, traditional methods of sorting often struggle. These standard techniques are excellent at finding clear-cut categories, but they sometimes miss the subtle, gradual shifts that exist in the real world, where one type of landscape slowly blends into another rather than stopping abruptly.
In a new study, researchers from Newcastle University and the Universitat de Barcelona tackled this challenge by applying a powerful new tool to the study of ancient landscapes. They focused on the Upper Arlanza Basin in northern Spain, a rugged, mountainous region that was home to early medieval communities between the fifth and thirteenth centuries. This area is a perfect laboratory for understanding how people chose where to live, as it features a mix of high peaks, deep valleys, and river networks. The team gathered eleven different measurements for fifty-seven ancient sites, ranging from altitude and distance to water to how visible a site is from the surrounding land. To make sense of this data, they used three different analytical approaches: two well-established methods that archaeologists have relied on for decades, and a newer technique called a Self-Organizing Map, which is a type of computer program designed to find patterns in complex data without being told what to look for.
The researchers wanted to see if this newer computer method could reveal something the older methods missed. They started with Principal Component Analysis, a technique that simplifies a large table of numbers by finding the main directions in which the data varies. They also used Hierarchical Clustering, which builds a family tree of the sites, grouping the most similar ones together and then merging those groups into larger clusters. Finally, they applied the Self-Organizing Map. Imagine a grid of tiny tiles, where each tile learns to represent a specific type of landscape profile. As the computer processes the data, similar sites are drawn to the same tile, while different sites move to different tiles, creating a map where the distance between tiles reflects how different the landscapes are.
When the team compared the results, they found a reassuring consistency. All three methods agreed on the existence of four main groups of settlements. This convergence suggests that the classification is robust and not just an artifact of one specific mathematical trick. However, the way these methods presented the results revealed a crucial difference in how they handle the messy reality of the landscape. The traditional methods forced the sites into distinct boxes. While this is useful, it can sometimes feel artificial, as if it is cutting through a smooth gradient to create sharp edges where none truly exist. The Self-Organizing Map, by contrast, preserved the smooth transitions between these groups. It showed that while there are four main types of settlement, they do not exist in isolation; they flow into one another.
The power of the Self-Organizing Map became even clearer when the researchers looked at the details. Because this method keeps the original measurements visible, the team could create heat maps that showed exactly what each group looked like in the real world. For instance, they could see that one group of sites was characterized by high visibility and dominance over the surrounding area, while another group was defined by steep, rough terrain. The map also highlighted sites that sat right on the boundary between groups. These "transitional" sites, which possess a mix of features from two different clusters, were often the most difficult to classify using the older methods. The Self-Organizing Map treated these ambiguous sites not as errors to be forced into a box, but as valuable clues that the landscape itself is a continuous spectrum.
The study concludes that the best approach is not to choose one method over the others, but to use them together. The traditional techniques provide a solid, familiar framework for grouping the data, while the Self-Organizing Map adds a layer of depth by showing how those groups connect and where the boundaries are fuzzy. By revealing the continuous gradients and the transitional sites, this new approach offers a more nuanced understanding of how early medieval people navigated and settled their world. It suggests that the decisions of the past were not always made in black and white, but often in the complex, shifting shades of gray that characterize the natural landscape.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.