Understanding Climate Change—One Tree at a Time
By Rob Knies
June 12, 2008 11:00 AM PT

Forests contain two thirds of Earth’s biodiversity, and as much carbon is stored in trees as is present in the atmosphere. Such facts are critical to understanding the ramifications of climate change—and how we can best respond.

But unlike other key factors such as ocean dynamics, the effects of forest dynamics—how tree populations interact and impact the environment—remain a mystery.

That is changing, though. A recent influx of data from far-flung nations, together with recent advances in modeling and data-analysis techniques, is enabling Drew Purves and his collaborators to shed light on what occurs to forests over time—and what the implications are for other terrestrial life forms, including humans.

Drew Purves
The world-leading journal Science accepts only a small fraction of the thousands of papers submitted each year, but the latest issue contains two by Drew Purves of Microsoft Research Cambridge.

On June 13, Purves, a scientist in the Computational Ecology and Environmental Sciences group, part of the Computational Science group at Microsoft Research Cambridge, had the rare honor of having two papers published in the renowned journal Science.

One, entitled Predictive Models of Forest Dynamics, authored by Purves and Stephen Pacala, of the Department of Ecology and Evolutionary Biology at Princeton University, is an excellent treatment of the state of the art in modeling forest dynamics and its impact on climate prediction. Microsoft Research Cambridge has prepared a video explaining the overview.

The second, Animal Versus Wind Dispersal and the Robustness of Tree Species to Deforestation, examines the response by tree species to the destruction of forests. That paper was based on work conducted at Microsoft Research Cambridge by Daniel Montoya, of the Departamento de Ecología of Spain’s Universidad de Alcalá, under Purves’ supervision, and written with Miguel A. Rodriguez, of the Universidad de Alcalá, and Miguel A. Zavala, of the Centro de Investigación Forestal at Spain’s Institito Nacional de Investigación y Tecnología. This paper, too, is featured in a video.

Both papers appear in a special issue of Science focused on forest dynamics.

“By applying various methods in computational data analysis to a large source for forest data,” Montoya says, “we have confirmed that, in Spain at least, plants with animal-dispersed seeds are less vulnerable to habitat loss, because animals provide trees with an intelligent dispersal mechanism, travelling and distributing seeds between areas of remaining forest. In contrast, a wind-dispersal method is more susceptible to habitat loss, as seeds are more likely to fall in inhospitable environments.”

The first paper is a bit more high-level, as Pacala explains.

“Until now,” he says, “one of the most important pieces of the climate-change jigsaw has been missing. We argue that we can significantly further our understanding of forest dynamics if scientists work together to use new computational techniques and data sources. … We feel that these discoveries could unlock the climate-change mysteries of forests on a global scale in as little as five years.”

Coming at a time when concern about climate change and carbon-dioxide emissions is acutely heightened, the need for such work, Purves says, is obvious.

“I’ve realized that we have a real responsibility as ecologists to do whatever we can to address the problems of climate change and biodiversity,” he says.” This really is an important problem.”

The Purves-Pacala paper makes that clear at the outset, with a laundry list of forest-related developments that could have profound effects: deforestation, logging, pollution, nitrogen deposits, the loss of pollinating and seed-dispersing animals, and increased carbon dioxide in the atmosphere.

The latter has gotten much attention in the past few years, and greater understanding of the role forests play in storing and regulating a large portion of the world’s carbon would play a significant role in determining how to address the concern.

“Forests are made of trees, and trees are made of carbon,” Purves says. “And we know that forests are processing and storing large amounts of carbon. When trees grow, they remove carbon from the atmosphere and store it in wood. But eventually, every tree dies, and the wood rots or burns, releasing the carbon back into the atmosphere. There’s a potential for a huge effect of that cycling and storage of carbon on the future of the earth’s climate.”

The research he and his associates are pursuing holds the promise to address fundamental scientific questions that could help conservationists and governmental officials decide how to protect the climate, such as:

  • What kinds of ecological strategies are employed by different tree species?
  • Given current data constraints, what is the optimal model for predicting forest dynamics?
  • How quickly does the biodiversity of forests shift in responses to changes in climate?

The answers, Purves suggests, lie in dynamic global vegetation models (DGVMs), which simulate the distribution, the physiology, and the biogeochemistry of forests and other vegetation at global scales, under present, historic, or future climates.

“The way that forests process water and energy has immediate effects on everything from cloud formation to air temperatures to seasonal patterns of air flows,” Purves says. “This is an important part of modeling the global climate, along with things like ocean currents and atmospheric movements.

“DGVMs are dynamic in that they predict short-term patterns and changes in things like photosynthesis and evaporation, but the dynamic part really refers to longer time scales, a century or more, where the actual vegetation can respond to the climate. If the world warms up, for instance, we’d expect to see changes in where we find grasses, forests, and other kinds of vegetation. This would then affect the climate.”

DGVMs predict how, over such long time scales, those distributions might change across the globe. Rain forests, savannas, or grasslands could shrink or expand, and such models could predict how much carbon is exchanged between vegetation and the atmosphere.

The problem with such models, though, is that, by definition, they incorporate a degree of uncertainty.

“Some of the models say that the forest will soak up gigatons of extra carbon,” Purves says. “In doing so, they act as a brake on climate change, because they’re soaking up a lot of our excess emissions.

“But other models say that the opposite will happen and that, because of the increase in temperatures, we’ll get a lot of those forests turning into savanna, all the carbon that was in tree trunks will be released into the atmosphere, and the forest will act as an accelerator on climate change. So we have a real dilemma at the moment. We’re relatively sure the forests are going to play an important part in the future of the climate—in particular, how it responds to CO2. But exactly how is a complete mystery.”

Therein lies the challenge—and the opportunity.

“We need to move on to a situation where we reduce this uncertainty and decide one way or the other what these effects are going to be and their magnitude,” Purves says. “That’s what we tried to do in the [first] article: give a vision for how we might do that.”

His vision? Develop a new generation of DGVMs based more closely on how forest ecosystems actually work—using lots of data.

“We suggest that the convergence of recently developed mathematical models, improved data sources, and new methods in computational data analysis could produce a step change in the realism of these models,” he says. “That would give us truly invaluable information to help manage the world’s forests and understand their impact on our climate.”


In recent years, a number of nations have invested in assessing their timber stocks, with an eye toward production forestry. But Purves and colleagues are using such data for more far-reaching purposes.

“They’re a hugely underutilized resource,” he says, “some of the largest ecological data sets in the world in terms of sample sizes and potential impact. These are a bit like the Human Genome Project equivalent for ecology. They’re just enormous, millions of individual trees. It’s really exciting.”

What’s more exciting is the possibility that such data could help produce more accurate models.

“If we could just put them together properly,” Purves says, “then we could develop models of forest dynamics that are defensible, properly constrained with data, so that when someone asks, ‘Are your model predictions correct?’ you can say, ‘Yeah, we really think they are broadly correct.’

“On one hand, it’s about using powerful computers to run large numbers of simulations. But on the other hand, we also have to do a lot of fundamental thinking, for instance, about what we call scaling. How can we integrate processes at fine spatial and temporal scales to predict dynamics at larger scales? If leaves start to photosynthesize more quickly, what does that mean for the forest ecosystem in 100 years’ time? It’s a very common problem across a lot of different areas.

“And we’re talking about the fact that forests are composed of different species and they’re very different. Some species are fast-growing, others are slow-growing, or fast-dying and slow-dying. We’ve thought quite a lot about statistical techniques that are able to deal with those sorts of heterogeneities in the system.”

The novelty of this approach is what has led to the pair of Science articles, which are contributing to the advancement of the state of the art.

“One,” Purves says, “is that we’ve worked out a way to scale up from the behavior of individual trees to the behavior of forests, the short-term behavior of individual trees that grow and die up to the large-scale and long-term behavior of forests. It’s what we call the scale transition.

“It’s a mathematical problem that we think we’ve cracked. What that lets us do is use the measurements we have at the tree level, then scale up and implement those models at large spatial scales over the long term. I think those sorts of scaling methodologies are going to be important in a lot of other fields.”

There are other contributions, as well.

“The other part of the problem is different statistical algorithms for constraining the structure and parameters of these models, given the data,” Purves adds. “Although there is a lot of data, as soon as you start to break it down, you look at all the different processes you have to constrain. If you don’t do it intelligently, you’ll find you don’t have enough data. At all stages you need to be constraining the shape of the way you inform the model structure and the species parameters with data.

“There’s a lot of work to be done. It’s thinking of efficient, usable methods for finding the best model, given the data. We have a way that implicitly captures the variation in parameters in the forest, which is a result of the variation in the species, without having to model every species separately.”


For Purves, the ability to examine such issues at Microsoft Research Cambridge is a bit surprising. Having dabbled in programming as a youngster—the son of a computer engineer—he became entranced by biology, finishing atop his class in plant sciences at the University of Cambridge before getting a Ph.D. from the University of York and spending five years as a postdoctoral researcher with Pacala at Princeton.

“The default career path for me would have been as an academic,” he says. “I had some job offers at U.S. universities just around the time that this job was advertised. I was very intrigued, excited, and surprised that Microsoft was doing anything in science, doubly surprised that they would recognize ecology as important.

“What’s great in [Director of Computational Science] Stephen Emmott’s vision at the lab in Cambridge is he recognized the critical importance of ecology in the challenge of understanding climate change and the vital need for radically new kinds of computational science to better understand the role of ecology in the earth and climate system. So here I am, still programming like I was when I was 7 years old and back in Cambridge like when I was 18, but now I’m working at Microsoft.”

And working with people who share his passion. In addition to his collaboration on the two Science papers, he cites a number of colleagues he met at Princeton, such as Jeremy Lichstein, Nikolay Strigul, John Caspersen, and Paul Moorcroft, and also with colleagues Lindsay Turnbull and Andy Hector at the Universität Zürich. These collaborators are making strides from which, you could argue, everybody in the world could benefit, particularly forest managers, environmentalists, conservationists, and governments that will need to form environmental and even economic policy based on predicted climate change. Developing the right models is critical.

“The more certain our model predictions,” Purves says, “the more we’re going to know exactly how big a problem climate change is and the time scales of when the various problems might arise. That’s going to feed into the more realistic discussion of what to do about CO2 and how much it’s going to cost and over what time scale.”

Perhaps, given the magnitude of the challenge that climate change portends, such work might encourage other young scientists to pursue similar paths.

“The young people who have quantitative ability, mathematically able and interested young people, if they do go into natural sciences, they tend to go into cell biology,” Purves observes, “genomics and drug design, which are very important. But if we could manage to divert a few of those people onto some of these problems, which are, by anyone’s standards, at least as important for the future of humanity, that would be a fantastic outcome of this work.”

The stakes, he underscores, are high.

“It is imperative,” he concludes, “that we create the tools and science to accurately understand the reaction of ecosystems to climate changes and other forces—not just for plants and animals, but for our children and succeeding generations.”