A new machine-learning modeling method was used by Alexander Cox and Brenhin Keller, computer geologists at Dartmouth College, in hopes of uncovering what led to the dinosaurs’ extinction.
Machine-learning models are powerful tools that have been increasingly employed in research in recent years, specifically to make predictions. They operate by discovering patterns or making decisions based on unlabeled data sets the model has been fed. Different types of models exist and can be used depending on the question a research study seeks to answer.
Unlike other research approaches, Cox and Keller’s aim was to remove human bias in identifying the possible cause of death. The main two ideas of contention in this paleontology debate are whether the dinosaurs were killed by an asteroid strike to earth, or a series of volcanic eruptions that lead to the end of the Cretaceous Period.
The asteroid strike theory is based on the idea that a large asteroid hit the coast of the Yucatán Peninsula, resulting in a cooled state of earth. On the other hand, the volcanic theory dictates that eruptive activity led to a disruptive climate emitting carbon and sulfur dioxide.
To approach the question objectively, Cox and Keller used a statistical model called the “Markov chain Monte Carlo” approach. This approach used 128 independent processors—electronic circuits that run algorithms—to run the scenarios in parallel, comparing the decisions at the end. The benefit of parallel computing is that it reduces the time it takes to run these simulations from years to a couple of days.
The data sets the model was fed were deep-sea sediments from 65 to 67 million years ago containing foraminifera—marine microorganisms with different isotypes within their carbonate shells. The foraminifera build their shells using minerals from the seawater around them, meaning the compositions of these shells are reflective of the ocean chemistry at the time of their creation.
Over 300,000 scenarios were run to calculate the different carbon and sulfur dioxide levels that might be needed to produce a climate matching the existing data from the shell fossil records. This greatly differs from other computational models since it works backwards with a prediction already set in place rather than using data to make its own, new hypothesis.
The results from these simulations added more nuance to the previously black and white debate. It appears that volcanic activity was already occurring, releasing copious amounts of carbon dioxide and sulfur dioxide which led to widespread global warming and oceanic changes prior to an asteroid strike.
These uneven levels of gasses in the atmosphere lead to the gradual extinction of many organisms, altering the earth’s carbon cycle. Once the asteroid hit the earth, the release of energy triggered the eruption of the Deccan Traps.
The Deccan Traps are large volcanic areas, located in what is now western India, that were capable of massive releases of gasses and lava. Their eruption led to the end of the Cretaceous Period.
From the work of Cox and Keller, it seems there’s not a clear winner from either side of the debate, as results suggest both circumstances worked together to cause the downfall of the dinosaurs.
Geology has greatly benefitted from the development of new computational models and predictive analysis. It provides the ability to analyze large sets of data and is capable of understanding complex processes on Earth more rapidly than traditional methods. Statistical models can uncover other subtleties within data sets researchers can dismiss.
At Queen’s, the Geology Department has a Geomechanics Computation Laboratory hosting a multitude of research and design analysis software that are applied towards understanding the Earth’s processes.
Tags
AI, dinosaurs, machine-learning, paleontology
All final editorial decisions are made by the Editor(s)-in-Chief and/or the Managing Editor. Authors should not be contacted, targeted, or harassed under any circumstances. If you have any grievances with this article, please direct your comments to journal_editors@ams.queensu.ca.
