Models that project how the climate will change in the future rely heavily on information about what happened in the past, including how humans used land. Scientists feed models data to create algorithms that estimate everything from weather to vegetation to land use. But this approach has one fundamental flaw.
According to archaeologists—who have dedicated their lives to collecting the data about the history of the world—the models used by climate scientists to describe past land use (known as “Earth system models”) are, in many ways, superficial. Specifically, they tend to overlook the impact humans had on land throughout the millennia, often miscategorizing land as untouched when it wasn’t. That’s why more than 200 researchers from all over the world have come together to create one of the world’s first archaeological climate databases. Though Earth system models are advancing without archaeological data, these scientists believe that the historical record they’re putting together could radically improve the way we think about the future.
“It is fair to say that these models have never really taken archaeological insights into account,” said Kathleen Morrison, a historical anthropologist from the University of Pennsylvania. She specializes in using pollen to reconstruct vegetation records, and has spent decades understanding the connections between changes in social life, land use, and biodiversity in southern India over the past 5,000 years. When she first came across a model-estimated representation of the places she’d studied for so long, she was shocked.
“This was the ‘ah ha’ moment for me. All the rice fields, grazing lands, canals, towns, and cities my colleagues and I studied seemed to have disappeared, and the model assigned these times and places to primarily natural vegetation,” Morrison said. “I knew that was wrong. Many of us in archaeology had never even heard of these models. But we knew we had a contribution to make.”
Two of the main Earth system models used by scientists to understand prehistoric land cover and model climate today are the History Database of the Global Environment or HYDE, and a less recent one called KK10 (the name comes from the initials of the scientists who created it and the fact that it was published in 2010). These models take modern data about how the land is used and what it’s made up of, and they essentially “hindcast” what they think would have happened in the past. The hindcasts are based on estimates of past population density multiplied by “stock” values to represent land use per capita for all parts of the world and all early time periods, Morrison explained, but these models ignore that humans used land in the past in ways that are not analogous to the present at all.
“The models currently used are problematic,” Morrison said. “But the nature of the error isn’t always the same.” Since the models are “off” in different ways in different regions, there isn’t a simple, one-size-fits-all fix to the issue.
For example, models rely on assumptions about how much cropland per capita was needed in order to feed populations and they assume a constant of how much food was needed and produced, although that has changed substantially through time, Emily Hammer, another anthropologist at the University of Pennsylvania, said. The models also assume people only produce a enough food to subsist, with no overconsumption or waste.
Different groups of people have changed their land-use practices, the crops that they farm, and how they live—for example, whether they lived in concentrated large settlements or dispersed villages and hamlets. In 1500, for example, Western European farmers growing wheat and barley used around 2.5 acres of land per person to feed themselves, explained Hammer. During the same period, farmers in southern China growing paddy rice used closer to 0.8 acres per person to feed themselves. The same “stock” values can’t be plugged into the formula for both. Humans are complicated, and all of the land-use choices different societies made in the past, in turn, affected the climate (yes, ancient human activities likely influenced the climate, though not at the scale they do today).
Given these disparities and the fact that, Hammer said, “we actually had a lot of data concerning” different types of land use, researchers launched the LandCover6k project in 2015. They aim to aggregate, synthesize, and harmonize all evidence of land use and make reconstructions of vegetation through time with actual paleoenvironmental and archaeological data, not algorithms.
In a paper published recently in the journal PLOS One, the researchers laid out a hierarchical database, a sort of five-mile (eight-kilometer) grid over the surface of the Earth that shows what humans were doing in a given square at a particular point in time. “Were people farming in this area, or were they hurting animals, hunting, and gathering. Were they burning the landscape? Were they producing pottery, which requires a lot of fuel in order to run the kilns?” Hammer said. “Were they doing other types of technologies, like glass making or smelting or plowing the land?”
This builds off of a previous project called ArchaeoGLOBE, which operated at a much coarser scale of resolution of roughly country-sized regions. “Reaching global coverage, especially in areas that have seen less archaeological investigation, will be more difficult and will surely take many years,” Lucas Stephens, the creator of ArchaeoGLOBE, said. For now, iLandCover6k has only created maps for 6,000, 4,000, and 2,000 years ago for Mesopotamia and the Arabian Peninsula.
These investigations will in turn inform climate modeling, making them more accurate by improving the model inputs, such as the per capita land use figures, according to Morrison. All of this data is crucial if scientists want to understand vegetation, and therefore climate, in the past—and what comes next as the planet heats up.
“Our ultimate goal, however, is to develop data-based maps derived from archaeological evidence and avoid the models all together,” said Morrison. “This is a huge task that builds on more than a century of research all around the world, and this is just the first major explanation of this approach.”
The researchers actively using land-use projection models are also itching for that archaeological data to plug into the algorithms and increase accuracy—but some say it’s been a long time coming, and there will probably be no getting rid of models completely in the long run.
“This is just a classification of things. It hasn’t really broken new ground yet,” Erle Ellis, an environmental scientist at the University of Maryland, Baltimore said about the LandCover6k database. “We have been holding our breath for products to come out of this research. What they’re trying to do is a lot harder than making a model, right? But there hasn’t been a huge amount of progress there.”
Ellis is also involved in the LandCover6k research, but he stressed that models like HYDE are already improving significantly. There’s a classic phrase amongst modelers that all models are wrong but some are useful because a model is always an interpretation of facts, not the facts themselves. “But that will also be true for local archaeological data [like Morrison’s],” Ellis said. “There’s really not the potential to have a 100% empirical global map of land-use history. It’s always going to involve some type of model.”
In fact, climate modeling is already adjusting to increasingly factoring in how humans affected land throughout history. According to Stephens, some of the information from ArchaeoGLOBE is already being incorporated in the next version of HYDE. The same week the LandCover6K paper came out, he and Ellis published research in the Proceedings of the National Academy of Sciences using the latest version of the HYDE modeling system. They demonstrated that, while previous climate models used to say 82% of Earth’s land in 6,000 BCE was “wild,” the most up-to-date algorithm indicated that most of the terrestrial biosphere showed signs of human transformation already 12,000 years ago.
“That really shocked us, right? That’s a lot more than we expected,” Ellis said.
He added that the archaeologists he’s provided these findings to feel this, too, is still a reductive figure, and inputting more accurate data will show there was land use was even more extensive. But help could be on the way as scientists build the archaeological database out further.
“It will be exciting to see if, and if so, how our findings differ from model-based expectations,” Morrison, who acknowledged there’s still a lot of work to be done on LandCover6k, said. “We all have the same goal: to use understandings of the past to address the challenges of the present and the future.”
Sofia is a freelance science journalist working between Italy, the UK, and the U.S. Her work has appeared in Inverse, Quartz, Wired, the Guardian, National Geographic, Atlas Obscura, and more.