Bayesian regression versus machine learning for rapid age estimation of archaeological features identified with lidar at Angkor.

Lidar (light-detection and ranging) has revolutionized archaeology. We are now able to produce high-resolution maps of archaeological surface features over vast areas, allowing us to see ancient land-use and anthropogenic landscape modification at previously un-imagined scales. In the tropics, this has enabled documentation of previously archaeologically unrecorded cities in various tropical regions, igniting scientific and popular interest in ancient tropical urbanism. An emerging challenge, however, is to add temporal depth to this torrent of new spatial data because traditional archaeological investigations are time consuming and inherently destructive. So far, we are aware of only one attempt to apply statistics and machine learning to remotely-sensed data in order to add time-depth to spatial data. Using temples at the well-known massive urban complex of Angkor in Cambodia as a case study, a predictive model was developed combining standard regression with novel machine learning methods to estimate temple foundation dates for undated Angkorian temples identified with remote sensing, including lidar. The model's predictions were used to produce an historical population curve for Angkor and study urban expansion at this important ancient tropical urban centre. The approach, however, has certain limitations. Importantly, its handling of uncertainties leaves room for improvement, and like many machine learning approaches it is opaque regarding which predictor variables are most relevant. Here we describe a new study in which we investigated an alternative Bayesian regression approach applied to the same case study. We compare the two models in terms of their inner workings, results, and interpretive utility. We also use an updated database of Angkorian temples as the training dataset, allowing us to produce the most current estimate for temple foundations and historic spatiotemporal urban growth patterns at Angkor. Our results demonstrate that, in principle, predictive statistical and machine learning methods could be used to rapidly add chronological information to large lidar datasets and a Bayesian paradigm makes it possible to incorporate important uncertainties-especially chronological-into modelled temporal estimates.

Does the Locally-Adaptive Model of Archaeological Potential (LAMAP) work for hunter-gatherer sites? A test using data from the Tanana Valley, Alaska.

We report an assessment of the ability of the Locally-Adaptive Model of Archaeological Potential (LAMAP) to estimate archaeological potential in relation to hunter-gatherer sites. The sample comprised 182 known sites in the Tanana Valley, Alaska, which was occupied solely by hunter-gatherers for about 14,500 years. To estimate archaeological potential, we employed physiographic variables such as elevation and slope, rather than variables that are known to vary on short time scales, like vegetation cover. Two tests of LAMAP were carried out. In the first, we used the location of a random selection of 90 sites from all time periods to create a LAMAP model. We then evaluated the model with the remaining 92 sites. In the second test, we built a LAMAP model from 12 sites that pre-date 10,000 cal BP. This model was then tested with sites that post-date 10,000 cal BP. In both analyses, areas predicted to have higher archaeological potential contained higher frequencies of validation sites. The performance of LAMAP in the two tests was comparable to its performance in previous tests using archaeological sites occupied by agricultural societies. Thus, the study extends the use of LAMAP to the task of estimating archaeological potential of landscapes in relation to hunter-gatherer sites.

Multiple hominin dispersals into Southwest Asia over the past 400,000 years.

Pleistocene hominin dispersals out of, and back into, Africa necessarily involved traversing the diverse and often challenging environments of Southwest Asia1-4. Archaeological and palaeontological records from the Levantine woodland zone document major biological and cultural shifts, such as alternating occupations by Homo sapiens and Neanderthals. However, Late Quaternary cultural, biological and environmental records from the vast arid zone that constitutes most of Southwest Asia remain scarce, limiting regional-scale insights into changes in hominin demography and behaviour1,2,5. Here we report a series of dated palaeolake sequences, associated with stone tool assemblages and vertebrate fossils, from the Khall Amayshan 4 and Jubbah basins in the Nefud Desert. These findings, including the oldest dated hominin occupations in Arabia, reveal at least five hominin expansions into the Arabian interior, coinciding with brief 'green' windows of reduced aridity approximately 400, 300, 200, 130-75 and 55 thousand years ago. Each occupation phase is characterized by a distinct form of material culture, indicating colonization by diverse hominin groups, and a lack of long-term Southwest Asian population continuity. Within a general pattern of African and Eurasian hominin groups being separated by Pleistocene Saharo-Arabian aridity, our findings reveal the tempo and character of climatically modulated windows for dispersal and admixture.

Rainfall, temperature, and Classic Maya conflict: A comparison of hypotheses using Bayesian time-series analysis.

Studies published over the last decade have reached contrasting conclusions regarding the impact of climate change on conflict among the Classic Maya (ca. 250-900 CE). Some researchers have argued that rainfall declines exacerbated conflict in this civilisation. However, other researchers have found that the relevant climate variable was increasing summer temperatures and not decreasing rainfall. The goal of the study reported here was to test between these two hypotheses. To do so, we collated annually-resolved conflict and climate data, and then subjected them to a recently developed Bayesian method for analysing count-based times-series. The results indicated that increasing summer temperature exacerbated conflict while annual rainfall variation had no effect. This finding not only has important implications for our understanding of conflict in the Maya region during the Classic Period. It also contributes to the ongoing discussion about the likely impact of contemporary climate change on conflict levels. Specifically, when our finding is placed alongside the results of other studies that have examined temperature and conflict over the long term, it is clear that the impact of climate change on conflict is context dependent.

Climate change, not human population growth, correlates with Late Quaternary megafauna declines in North America.

The disappearance of many North American megafauna at the end of the Pleistocene is a contentious topic. While the proposed causes for megafaunal extinction are varied, most researchers fall into three broad camps emphasizing human overhunting, climate change, or some combination of the two. Understanding the cause of megafaunal extinctions requires the analysis of through-time relationships between climate change and megafauna and human population dynamics. To do so, many researchers have used summed probability density functions (SPDFs) as a proxy for through-time fluctuations in human and megafauna population sizes. SPDFs, however, conflate process variation with the chronological uncertainty inherent in radiocarbon dates. Recently, a new Bayesian regression technique was developed that overcomes this problem-Radiocarbon-dated Event-Count (REC) Modelling. Here we employ REC models to test whether declines in North American megafauna species could be best explained by climate changes, increases in human population densities, or both, using the largest available database of megafauna and human radiocarbon dates. Our results suggest that there is currently no evidence for a persistent through-time relationship between human and megafauna population levels in North America. There is, however, evidence that decreases in global temperature correlated with megafauna population declines.

Radiocarbon dating uncertainty and the reliability of the PEWMA method of time-series analysis for research on long-term human-environment interaction.

Statistical time-series analysis has the potential to improve our understanding of human-environment interaction in deep time. However, radiocarbon dating-the most common chronometric technique in archaeological and palaeoenvironmental research-creates challenges for established statistical methods. The methods assume that observations in a time-series are precisely dated, but this assumption is often violated when calibrated radiocarbon dates are used because they usually have highly irregular uncertainties. As a result, it is unclear whether the methods can be reliably used on radiocarbon-dated time-series. With this in mind, we conducted a large simulation study to investigate the impact of chronological uncertainty on a potentially useful time-series method. The method is a type of regression involving a prediction algorithm called the Poisson Exponentially Weighted Moving Average (PEMWA). It is designed for use with count time-series data, which makes it applicable to a wide range of questions about human-environment interaction in deep time. Our simulations suggest that the PEWMA method can often correctly identify relationships between time-series despite chronological uncertainty. When two time-series are correlated with a coefficient of 0.25, the method is able to identify that relationship correctly 20-30% of the time, providing the time-series contain low noise levels. With correlations of around 0.5, it is capable of correctly identifying correlations despite chronological uncertainty more than 90% of the time. While further testing is desirable, these findings indicate that the method can be used to test hypotheses about long-term human-environment interaction with a reasonable degree of confidence.