Warnell’s award-winning researcher Cristian Montes uses machine learning to find trends in forest growth. There’s a lot of information embedded in a tree stand. There’s elevation. Temperature changes. Rainfall. The age of the trees. At some point, insects might settle on one. A windstorm could blow one of them down. A fire could burn it up. This data could seem overwhelming, but if harnessed the right way, it can be tremendously useful. Similar to how the insurance industry uses actuarial science to make coverage predictions, tree scientists like Cristian Montes are using large sets of data to make predictions about trees. Montes’ work was acknowledged last year by his peers, as the recipient of the 2021 Alumni Association Faculty Research Award. More recently, Montes received the 2021 Award of Excellence in Research and Development from the Southeastern Society of American Foresters. “We’re not only modeling how trees are growing into the future, but we’re also modeling the risks over time,” said Montes, an associate professor of natural resources biometrics, co-director of the Plantation Management Research Cooperative and the 2021 recipient of the Alumni Association Faculty Research Award for the Warnell School of Forestry and Natural Resources. “For example, what are the chances a particular forest will get a particular disease in a certain spot? Because if there’s a 50% chance, you could get 50% of the yield you expect.” This applies to a broad spectrum of scenarios, such as hurricanes, tornados or fire. To do this, Montes uses machine learning—also called artificial intelligence—to collect and analyze huge sets of data that contain a range of data points and many variables describing a particular forest. This may include satellite information, weather data and forest stand growth. It’s essentially the same technology that social media uses to predict what ads you might like to see, or what stores use to determine what coupons to send you. By capturing a large amount of data and then relating that data to one particular trait, you can make an informed prediction. And, said Montes, this data is everywhere. “There are databases all over the states with that information. For example, we recently finished a paper on the depth of the water table, which is an important ecological aspect. It was data just sitting there,” said Montes. “We were thought it was about time to use that data to pair with our extensive environmental database using some machine learning.” The goal is to create maps summarizing production risk models that point to certain catastrophic events so that, in the future, landowners and managers can use these models to make better informed decisions. This represents the next generation of forest modeling, and the formulas take a much more dynamic approach to the science. Current models, by contrast, are considered more static, where tree stands are akin to islands and are evaluated separately. Now, with machine learning, Montes and other Warnell scientists can incorporate not just one tree stand, but all the nearby tree stands. It would be an impossible task to tackle from the ground, but satellite and drone technology, plus access to supercomputers to crunch the numbers, makes the process accessible to forestry. Another bonus to Montes’ work: His access to tree plantations across the Southeast through the PMRC, where he and his research teams can test and monitor sites, collect data and try out new ideas. It’s a win for landowners and PMRC members too, who get first access to these new tools just as they are developed. “We have several projects laid out all over the South that we are measuring regularly, and we present those models to the members of the industry,” he said. “That’s the good thing about working with the PMRC, in that everything we do gets implemented almost immediately. We’re working with people that are using the products.” Montes also uses this technology for a variety of international projects related to machine learning and forest production. In some cases, he’s collecting data on the scale of an entire country and working with a supercomputer on the UGA campus to crunch the numbers. Google and Amazon provide similar services for cloud computing that companies in other industries are using as they become aware of the technology. “I hope the forest industry will make the switch to these new ways of processing data soon, too,” said Montes. While Montes has been plugged into this process for about two decades, it’s still fairly new for the forestry industry in the southeastern United States. He said he hopes this experience will give students a leg up when they graduate from Warnell and begin working in the field. Landowners want these types of models as fast as they can be produced, he said. “So many things are being used already, especially the risk maps,” he said. Last year, Montes launched risk maps based on tornado probability. This summer, the PMRC will debut risk assessments based on tornados. “I’m not investing in something new; we’re just investing in a new application for something that has been done in different branches of science.”
