South Arabia’s prehistoric monument landscape shows social resilience to climate change

PLOS One | https://doi.org/10.1371/journal.pone.0323544 May 28, 2025 10 / 24 We further used ANOVA with Tukey post-hoc tests, to compare total stone volumes between monuments that included one or more large stones and those that did not, repeating the test within each monument type. We also compared total stone volumes between two categories of construction-- -episodic and accretive--for insight into the relationship between labor investment and collective action. We already understand the temporal sequence: episodic construction pre-dates accretive construction. Neolithic platforms had to be assembled in one episode to stand. Bronze Age HCT served as tombs, and their architectural and stratigraphic characteristics suggest they were also constructed in one episode. SCABs could have been accretive, but the documented examples are all complete, hinting that they never were left unfinished as might happen in accretive, sequential, low investments of labor. Accretive triliths were built with ele- ments that could be assembled by few people in short time but that could be replicated thereafter by newcomers; indeed, there are examples of triliths with a lone element and examples with dozens. boat graves are similar. To test whether large, simultaneous groups built early episodic monuments, we also tested the association between size of the largest stone and technological requirements, using a chi-squared test of independence. Results Environmental factors influencing monument placement

Monuments occur in a region now hyper arid, characterized by sparse water availability and uneven graze. Our survey yielded areas with high monument concentrations—around the oasis of Mudhai, the Wadi Haluf-Hanun corridor, and two tributaries of Wadi Dhahabun. (Fig 3) Analytical results show how local environmental factors explain the positioning of monuments in the landscape and whether cultural factors like routes and extant monuments were significant. Supported by machine learning, our analysis revealed a strong spatial patterning in the environmental characteristics of 371 monument locations. Results from bootstrap aggregating show that (Fig 4a) proximity to springs was important, whilst a partial dependence plot shows that the probability of monument occurrence decreases substantially beyond 5 km from springs (Fig 4b). Most monuments were also placed in low topographic positions, within wadis or depressions (Fig 4b). The probability of any monument occurrence decreases with increasing distance from routes (Fig 4b), with most mon- uments located within a few hundred meters of routes. Furthermore, the probability of monument occurrence increases sharply with vegetation (max-NDVI) in arid areas (S1 Fig). Environmental differences are apparent across the locations of different monument types characteristic of different time periods. A multinomial logistic regression preserved ten non-zero variables to predict monument type with coefficients (Fig 5) for each environmental variable for each monument type. Coefficient path plots show changes in the coefficient values

with decreasing λ (S6 Fig). Our results in Figure 5 show that platform monuments tend to occur at lower elevations, close to springs and in low topo- graphic positions in localized context (r = 200 m), such as within small wadis and depressions. However, platforms do not show an association with any particular topographic position at larger geographic scales. HCT monuments are more frequent in areas with high fog densities, with low NDVI values, on north-facing slopes. They tend to occur close to routes but farther from springs. Moreover, HCTs occupy steep slopes in high topographic positions of large landforms, such as in plateau areas and on the shoulders of large wadis. Trilith monuments occur in low topographic positions within wadis and depressions. They are associated with areas that have low fog densities and low vegetation cover, on north-facing slopes close to springs and routes. We find that boat grave monuments are associated with areas at higher elevations, with high fog densities and high vegetation cover. They tend to occupy south-facing slopes in high topographic positions of small landforms. Eleven of the twelve environmental variables were preserved in the minimum lambda model to assess whether labor input responded to environmental factors. The variables with the highest coefficients entered the model early and have the greatest predictive power (S7 Fig). Notably, labor input (total stone volume) increased with fog density and topographic

position at large scales, but decreased with maximum NDVI (S8 Fig). This can be explained by social responses to popula- tion densities, such that people in the mountains had frequent enough contact that monuments were not needed as social