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 19 / 24 to those that enter the model later. When a new variable enters the model, it affects the slope of the coefficient paths of the other predictors, depending on the magnitude of collinearity. (PDF) S7 Fig. The elastic net regression coefficient paths for largest stone volume. Penalized coefficients are col- ored grey. The regularized multiple linear regression assessed whether labor input responded to environmental factors; however, using the ‘one-standard error’ rule, no variables were preserved in the model. Eleven of the twelve environ- mental variables were preserved in the minimum lambda model. The variables with the highest coefficients entered the model early and have the greatest predictive power. Notably, labor input (total stone volume) increased with fog density and topographic position at large scales but decreased with maximum NDVI. Like the results of bootstrap aggregat- ing, the minimum lambda result suggests a threshold in population dispersal at which humans engage monuments for messaging-while-absent. At the higher population densities supported in the vegetated mountains, people had sufficient direct contact that monuments were not needed as social touchstones. Fog density is high in the near Nejd backslope of the plateau, where mobile families would readily find seasonal grazing and yet be infrequently in contact, using monu- ments to communicate. (PDF) S8 Fig. The regularized multiple linear regression model coefficients show the relationship between total stone

volumes of monuments and the environmental variables. A positive coefficient indicates that stone volume increases with that variable, whilst a negative coefficient shows an inverse relationship between stone volume and that variable. We also analyzed whether labor input responded to environmental factors within the most common monument types; HCTs (n = 100) and Triliths (n = 138). No variables were preserved in the ‘one-standard error’ rule models, but in the minimum lambda models, total stone volumes for HCTs increased with fog density (coef = 0.0798), whilst for Triliths, total stone volumes decreased with increasing slope steepness (coef = -0.0677). These results are also consistent with bootstrap aggregating and with an interpretation that HCT (episodic) lay within a zone where labor could be convened, (closer to the vegetated mountains), while the accretive triliths decline in size where labor expenditures rise in carrying stones up and down steep inclines. (PDF) Acknowledgments The authors are grateful to the Ministry of Heritage and Tourism, Sultanate of Oman; The Ohio State University; General Organization of Antiquities and Museums, Republic of Yemen; the Office of the Advisor to HM The Sultan for a reconnais- sance visit; and especially to the teams of colleagues, students, and locals who helped accomplish the work. Thanks also to Shane Scaggs, who completed Fig 2. Author contributions Conceptualization: Joy McCorriston. Data curation: Joy McCorriston, Lawrence Ball, Michael J. Harrower, Matthew J. Senn, Abigail F. Buffington.

Formal analysis: Lawrence Ball, Ian M. Hamilton. Funding acquisition: Joy McCorriston, Ian M. Hamilton. Investigation: Joy McCorriston, Michael J. Harrower, Matthew J. Senn, Tara Steimer-Herbet, Abigail F. Buffington, 'Ali Ahmad Al-Kathiri, 'Ali Musalam Al-Mahri. Methodology: Joy McCorriston, Lawrence Ball, Michael J. Harrower, Ian M. Hamilton, Matthew J. Senn, Tara Steimer- Herbet, Abigail F. Buffington.