Brandon Lee Drake
My long term interest is on better understanding how global climate change will affect human societies. To this end, the best way to understand our own vulnerabilities to climate change is to understand how to past human societies have been affected by climate processes. Archaeology provides the only approach to synthesize a history of both paleoclimate and human behavior.
However, there is a methodological stumbling block on this path - archaeologists are dependent upon regional records of climate change. This is fine, but additional specificity is sometimes desired. For my dissertation research, I've focused on developing an approach to use Δ13C from radiocarbon-dated charcoal to facilitate the development of paleoclimate records at the same resolution as the data archaeologists use to infer past human behavior. This approach has been spun off to pollen analysis as well - Δ13C from radiocarbon-dated pollen provides a regional record of paleoclimate that has some predictive power over mean annual precipitation.
An additional focus is on the use of Bayesian inference to tie together change in paleoclimate records. All to often, paleoclimate time series data is a set of wavy lines over time with no statement of significance for hypothesized points of change. Bayesian change-point analysis can help develop probability statements as to the significance of a given change-point. This has some interesting implications for how paleoclimate records can be compared and contrasted.
I also worked on e x-ray fluorescence in archaeological data collection, particularly in assessing whether or not the portable varieties hold up to the standard laboratory devices.
Supervisors: W. H. Wills
However, there is a methodological stumbling block on this path - archaeologists are dependent upon regional records of climate change. This is fine, but additional specificity is sometimes desired. For my dissertation research, I've focused on developing an approach to use Δ13C from radiocarbon-dated charcoal to facilitate the development of paleoclimate records at the same resolution as the data archaeologists use to infer past human behavior. This approach has been spun off to pollen analysis as well - Δ13C from radiocarbon-dated pollen provides a regional record of paleoclimate that has some predictive power over mean annual precipitation.
An additional focus is on the use of Bayesian inference to tie together change in paleoclimate records. All to often, paleoclimate time series data is a set of wavy lines over time with no statement of significance for hypothesized points of change. Bayesian change-point analysis can help develop probability statements as to the significance of a given change-point. This has some interesting implications for how paleoclimate records can be compared and contrasted.
I also worked on e x-ray fluorescence in archaeological data collection, particularly in assessing whether or not the portable varieties hold up to the standard laboratory devices.
Supervisors: W. H. Wills
less
InterestsView All (35)
Uploads
A 12,000 year record of pollen collected from packrat middens across Chaco Canyon were analyzed using a new normalization procedure to produce a Holocene record of piñon and ponderosa pine abundance. The normalization procedure, species occurrence, enabled statistical analysis of the data. Simple linear models indicated that piñon and ponderosa pollen were strongly correlated with each other. Bayesian change-point analysis was run, and a period between 5,440 and 5,102 cal. yr BP was identified as a point of expansion of piñon. This expansion of piñon was contemporaneous with increased storage and territoriality of populations in the San Juan Basin around the same time period.
In the Lower Alentejo of Portugal, a series of δ13C values from radiocarbon-dated pollen were used to calculate rates of 13C discrimination (∆13C). These ∆13C values indicated a period of stability from AD 600 - 1000, increased arid conditions during the first half of the Medieval Warm Period (AD 1000 - 1100), and a return to normal conditions from AD 1100 - 1150. Following this, the rural Lower Alentejo was abandoned for almost two centuries. The data suggest a climate that was stable during a period of population growth followed by heightened variability during the Medieval Warm Period. This would have caused some drying out of soil, potentially contributing to later soil erosion as wetter conditions returned prior to the rural abandonment. Importantly, the study paired modern observations of ∆13C with archaeological data, establishing an approach to the study of data commonly available to archaeologists.
Finally, in the Eastern Mediterranean, a set of regional paleoclimate records were used to better understand climatic conditions during the time of the Late Bronze Age Collapse (1200 - 1000 BC). At this time, most urban centers in the region were destroyed and abandoned during a period of depopulation. Alkenone-derived sea surface temperature records and warm species dinocyst/formainifera ratios indicate a cooling of Mediterranean waters at the time. Terrestrial paleoclimate records, including a biome-wide measure of Δ13C derived from radiocarbon-dated bulk pollen, indicate that conditions may have been more arid at this time.
This study of paleoclimate reconstruction includes new techniques developed to make use of under-utilized data gathered by archaeologists. These techniques include the use of of radiocarbon-derived Δ13C as a local indicator of aridity, the use of biome-wide Δ13C from pollen as a regional paleoclimate record, a new normalization technique for pollen records, and the use of Bayesian change-point analysis to assess the significance of changes in a paleoclimate record. These new techniques can compliment archaeological research questions that require information about paleoclimate.
""
This is the promise and potential of machine learning and artificial intelligence as used in the calibration and interpretation of XRF data. This chapter presents an overall outline of considerations for how to deploy this new suite of technologies, and their common pitfalls. First, a discussion of machine learning principles and model types will be used to lay common groundwork. Of these, there two quantitative examples (brines and ceramics) and two qualitative examples (wood and fossils) will be used to illustrate how these principles interact with different data types.
Quantification: a necessary step in converting spectral data to stand- ardized scientific units (%, ppm, mg L1, etc.)
Hypothesis testing: testing hypotheses while incorporating uncertainty.