A Catalyst for Complexity
XSEDE Programs Help Archeologists Master HPC, Model Human Origins
by Ken Chiacchia, Pittsburgh Supercomputing Center
April 22, 15
Picture an endangered species. Climate change has forced it into a small refuge. The population has crashed, to thousands or fewer.
Many species have faced this grim scenario. Some simply go extinct; a lucky few adapt and survive.
But what if a species changes the rules? What if it begins to cooperate and innovate at such a dizzying level of complexity that it adapts the environment instead of itself?
Research based at Arizona State University (ASU) and Nelson Mandela Metropolitan University (NMMU), South Africa, and elsewhere in the U.S., South Africa, Australia, Israel and France has begun to shed light on how modern human behavior emerged from such a crisis. Concentrating on the crucible of the Cape Floral Region of South Africa during a time when the earth was primarily in a glacial state, they are offering clues as Homo sapiens survived—and how our species became unique in its complexity.
The work was made possible by a fortuitous confluence of XSEDE initiatives that helped a large, complex collaboration of experts with very different skillsets merge their efforts seamlessly with advanced computing resources.
Coaching through the Complexity: XSEDE Champions and Fellows
"I applied to be a Campus Champion in my first year as a professor at Kent State University," says Eric Shook, who is an assistant professor of geography as well as an XSEDE campus champion there. "When I heard that I was going to be teaming up with Curtis Marean, the principal investigator of the project, I was really excited, because it was something I'd never seen before."
As an XSEDE campus champion, Shook had volunteered to help colleagues at his institution interface with the XSEDE system, matching them with high-performance computing (HPC) resources and generally familiarizing them with HPC. But Shook also decided to participate in one of XSEDE's groundbreaking initiatives: the Extended Collaborative Support Service (ECSS) Campus Champions Fellows Program. In this competitive program, campus champions (currently more than 230 people representing over 120 US universities and colleges), apply to join one of the collaborative projects in which one or more experts from the ECSS staff assist the members of a research team to make their programs work as efficiently as possible on the leading-edge systems coordinated by XSEDE.
"We see the Campus Champions Fellows program as an important connection and a way to share our experience conducting the in depth, project-based work in ECSS with campus champions," says Nancy Wilkins-Diehr, of the San Diego Supercomputer Center, co-principal investigator of XSEDE and co-director of ECSS. "The champions then serve as experts in their own campus communities."
In the 2013 competition, Shook applied to join ECSS expert David O'Neal from the Pittsburgh Supercomputing Center (PSC) in collaborating with Marean's team. Shook's technical and "people" skills, and his enthusiastic essay, won him one of six fellowships awarded in 2013. Now, he and ECSS mentor O'Neal worked with the members of the Marean team to find his role in the collaboration. The first step was to understand the paleoanthroplogical motivation and background of Marean's project.
Evidence in our genes, in the ground
"Humans cooperate with non-kin at spectacular levels of complexity," says ASU's Marean. "So what we wanted to know is what are the contexts of evolution for those special features of humans? When did they arise, and why did they arise?"
Two avenues of research argued that the Cape Floral Region had a special role in human evolution. For one thing, DNA sequencing suggests everyone on Earth today may descend from 15,000 or even fewer survivors of a great population crash. Based on genetic mutation rates, this crash happened roughly 100,000 to 300,000 years ago.
The other line of evidence is archeological. During the last glacial maximum, at about the same time, most of Africa would have been so arid that human habitation was difficult if not impossible. The Cape Floral Region is one of the few on the continent that show evidence of human habitation at that time.
More interesting, at this point people began to display behaviors that separate "modern" from "early" humans. They carried out complex thermal treatments of stone to make better tools. They engraved geometric patterns in ochre and bone. They improved their hunting tools with innovations such as atlatls—a kind of lever that allows a spear to be thrown farther and harder.
"We not only have a big brain, it's wired in way that allows us to think in complex analogies, plan for the future, understand mathematics," Marean says. "That proclivity is imbedded in our genes. If we're trying to understand that event, which leads to all modern humans, we need to understand the environment at that time."
Bringing It Together: Modeling Ancient Climate and Human Ecology
One archeological finding in particular caught the researchers' eyes. The Cape Floral Region is rich in shellfish, with a long, rich coastline, as well as tubers—plants with fleshy roots. This combination of foods could provide the protein and carbohydrates humans need to survive. But were they available enough to humans traveling on foot?
"Our project began as a straight archeological dig," Marean says. "Then I realized that we needed much better climate and environmental contextual data to understand the archeological record we were excavating."
The key, he reasoned, was being able to model the human remnant population at the time and its local environment. In addition to providing further evidence that the Cape Floral Region could sustain humans, such a tool could also help explain how that environment might have sparked the revolution in human behavior.
Marean's team envisioned coupling a regional climate model that could "hind-cast" conditions between 50 and 150 thousand years ago. In concert with the climate model, they would run an "agent-based model" in which virtual "agents" represent plants, animals and human hunter-gatherers interacting under the conditions computed in the paleoclimate model. Marean's collaborators Francois Engelbrecht of the Council for Scientific and Industrial Research in South Africa and botanists Richard Cowling and Alastair Potts at NMMU suggested adapting an atmospheric physics code that is used for weather forecasting in the region. Researchers at ASU had already used agent-based modeling to study modern day hunter-gatherers in their ecosystems in Paraguay and Ache, Indonesia. But neither model had been used in computers powerful enough to meet the challenges of the Cape Floral project, nor had they been made to work together.
Managing the Complexity: the Role of XSEDE
As it happened, Sergiu Sanielevici of PSC, manager of the ECSS Novel and Innovative Projects program (NIP), visited Marean's dig at Pinnacle Point on a vacation trip in July 2012. NIP, like the Campus Champion Fellows program, is a signature initiative of XSEDE ECSS whose mission is to foster projects from fields of science and communities that have not traditionally used HPC in past decades.
Sanielevici and Marean agreed that XSEDE resources and services ought to enable the proposed modeling with the high degree of accuracy required to confront the computational results with the archeological evidence. Marean followed up by requesting an XSEDE Startup allocation along with collaborative assistance from ECSS.
PSC's O'Neal had worked for decades on optimizing atmospheric physics codes for HPC systems, so he was the obvious choice for helping the South African team to scale up and optimize their climate code on the XSEDE resource Blacklight, a large shared-memory system at PSC. When Shook joined the team, his expertise and interest in agent-based models made him the perfect match for adapting the Arizona hunter-gatherer model to HPC.
Shook soon realized that most of his collaborators, while rich in many scientific disciplines, needed particular help transitioning to being HPC users.
"The PI of the project is an archeologist," Shook explains. "They did have climate modelers. They had an agent-based modeler on the team, but not in large-scale modeling that would require a supercomputer. And there are a lot of ecologists, botanists and everybody else who does field work. So I would say it's definitely not the traditional XSEDE project, filled with teams of computational chemists, high energy physicists, or computer scientists."
Changing Roles, Emerging Models
Marean agreed that they needed Shook's role to widen. As Shook says, "I became a little bit more of a communicator in terms of how XSEDE resources could be used and how all of these pieces could be put together on a computational platform."
Shook's catalytic role proved a big success. He suggested adding the Trestles system at SDSC to the XSEDE allocation to amplify the scope of the agent-based model, and designed a new system to bring the agent-based models to supercomputing scale. Under the guidance of David O'Neal, his ECSS Mentor, Shook designed a workflow that couples climate, vegetation, and agent-based models on XSEDE. Based on the work done during the startup phase, Marean's team secured a full XSEDE research allocation—with Shook as a co-principal investigator.
The team has now created a "paleoclimate" model to simulate the local climate in the Cape Floral Region during the glacial maximum. For the first time, this model simulated the Cape region at the level of detail needed to shed light on whether the region could sustain sufficient shellfish and tuber populations—a major accomplishment in the field. Even better, the initial results suggest that the climate would indeed have supported the food sources humans needed, at a time when virtually no place else in Africa did.
Shook is now helping the researchers add their agent-based models to the simulation to see how people may have interacted with that environment. The project promises enough predictive power for the models and the archeological evidence to be tested against each other, which would be another first.
"We certainly have debates over exactly how small the modern human population was at this point," Marean says. "And other people have argued that the progenitor population was in North Africa, or the Maghreb area ... It's going to be a while before we can say one way or another; but I think right now the Cape Floral Region hypothesis is a strong one. And like all good hypotheses, it's generating an enormous amount of good science."
He adds, "In my opinion there are three evolved behavioral traits that separate modern humans from other animals: an advanced cognition, a dependence on social learning, and hyper-prosociality, or the tendency for everyday and prolonged cooperation between unrelated individuals. Humans solve super-complex problems through collective intelligence—teamwork—and this is only possible if a species has an evolved proclivity to hyper-prosociality. Solving the great questions of science, such as how and why hyper-prosociality evolved, rests on that self-same evolved proclivity, which we see in XSEDE itself."