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 18 / 24 large, social gatherings of contemporaries living in a more abundant environment. Furthermore, the appearance of Ara- bian genealogical organization as adaptive social resilience belongs to a larger pattern: over time, ever-smaller groups fragmenting into culturally distinct regions began to build accretive monuments to maintain collective social identities among persistent pastoralists who seldom, if ever, coalesced. Conclusions Derived from the metrics of labor scheduling and location, and testable in new data sets, our empirical model theorizes eras of monument building as the exercises of social capacities to absorb risk in the dynamic climates and environments of the arid sub-tropics of Arabia. By decentering analysis from a single era or type of monument, we show how monument classes are proxies for social behavior of mobile, persistent pastoralists. Our model highlights a reliance on monuments as touchstones of social belonging and as flexible technologies of social resilience in a changing world. While built on archaeological observations of monuments in Arabia, our model may be applicable and adaptable to assess social resil- ience in other regions, such as Saharan, Mongolian, or the high Andes. Supporting information S1 Table. Codes and Metrics of 371 monuments in Dhofar used in this analysis. (XLSX) S2 Table. Description of monument types with age ranges of archaeological periods. (XLSX)

S3 Table. Samples, contexts, and radiocarbon determinations shaping the chronology of monuments in Dhofar. For comparison with published Bayesian analysis of radiocarbon determinations [38], see horizontal bars in Fig 2. (XLSX) S1 Fig. Partial dependence plots showing the relationship between each environmental variable and the pre- dicted probability of monument occurrence. (PDF) S2 Fig. Multinomial deviance across values of log λ from 11 models with different α parameter settings. We used the model with α = 0.4 (elastic net regression), which had the lowest multinomial deviance. (PDF) S3 Fig. Multinomial deviance across values of log λ with α = 0.4 (elastic net regression). We used the ‘one-standard error’ rule, to select the model with λ = 0.049 (log λ = -3.023). (PDF) S4 Fig. Mean-squared error across values of log λ from 11 models with different α parameter settings. We used the model with α = 0.1 (elastic net regression), which had the lowest mean-squared error. (PDF) S5 Fig. Mean-squared error across values of log λ with α = 0.1 (elastic net regression). We selected the model with the minimum λ = 0.129 (log λ = -2.047). (PDF) S6 Fig. The elastic net regression coefficient paths for each monument type. As λ decreases, the maximum permis- sible value of the L1 norm increases and more coefficients enter the model. Variables enter the model based on their true linear regression coefficient, and therefore variables that enter the model early have a higher predictive power, compared