New Tree Tech: Data-driven reforestation methods match trees to habitats

This story is the second article of a four-part Mongabay mini-series exploring the latest technological solutions to support reforestation. Read Part One, Part Three and Part Four.

Nations aim to restore hundreds of millions of hectares of lost and degraded forest over the coming decade to help meet Paris Agreement carbon emission targets and sequester more carbon as a stopgap as the world transitions away from fossil fuels. This requires repairing degraded ecosystems by helping forests regrow naturally, but also necessitates large-scale tree planting — reforestation carried out on a planetary scale.

But the word “reforestation” prompts hard questions: How can massive tree plantings be done effectively? When faced with abandoned cattle pasture or degraded croplands in the Amazon or elsewhere, how do conservationists decide what tree species to seed in which habitat? Which ones will take hold, grow strong, and store carbon for the long haul? Also key: Which trees will thrive now and in a warmer world?

For the world’s reforestation ambitions to be realized, careful planning is clearly needed. Reforestation projects that don’t use diverse native species, or that don’t account for local ecology, the evolving climate, and social environment in their plantings can at best waste time and money, and at worst do more harm than good.

Reforestation project failures across the globe show that traditional forestry methods aren’t working in the planning of many large-scale restorations. That’s where new tech solutions can help: Complex computer models, aided by artificial intelligence and laboratory testing, can assist in selecting tree species that will thrive now and in the future.

Understanding the trees

The first stage of any tree-planting project is deciding what to plant. Native tree species need to be identified that will thrive in the climate and ecology of the site to create healthy, diverse ecosystems, sequester carbon and support biodiversity. But with some 73,000 species of trees worldwide, gathering and crunching the mountains of data needed to understand each species’ needs and habitat preferences is a daunting task.

Some commercially important tree species have been extensively studied, and researchers are working to gain a better understanding of less-studied species, but humanity knows despairingly little about the majority.

“The problem is, almost nothing is known about these trees,” says Raymond Menard, founder of Minnesota-based biotechnology company Spades. “Probably 20% of them or more don’t even have a name.” This is especially true in the tropics, even as NGOs embark on gigantic reforestation efforts in the Brazilian Amazon and elsewhere.

Traditional forestry practices assess the tree species most suitable for a location based on what’s already growing in the area. As a result, foresters often default to recommending non-native species for replanting because these are the best-studied, and are the trees already growing in degraded locales. “The system is set against reforesting with indigenous species,” says Menard.

Where research data is lacking, foresters may set up test plots to see which species will survive in a particular location. But this trial-and-error approach is slow, and often doesn’t provide much meaningful information. “The [species] that lived you don’t know why they’re alive. [Those] that died, you don’t know why they died,” Menard explains.

Whether a tree grows and thrives in a particular spot, or wilts and dies, is largely determined by the availability of the scarcest resource in that habitat. Plant growth is determined by factors like water, sunlight, and nutrients such as nitrogen and phosphorus. A plant may also be constrained by factors that exceed its tolerance limit, such as acidity, salinity or pollutants.

It doesn’t matter how plentiful the supply of one factor is; if another factor is insufficient or excessive, a plant will be stunted, or die. “In any given location, there’s going to be two or three limiting factors that are dominant,” Menard notes. Determining these limiting factors for each species is crucial to identifying where it can grow.

Seeding diverse ecosystems, not monocultures

Habitats aren’t uniform; their soil characteristics, topography, light exposure and wind vary over kilometer to meter scales, and these microenvironments add to the challenge of selecting the appropriate species to plant. Though a given species may grow well in a particular climate, that doesn’t tell you whether it will grow at a specific location — the soil may be too moist, might contain too much clay, or receive too little sunlight.

Compounding the problem is that reforestation projects are often sited in places where soils have been severely degraded by decades of intensive management, meaning indigenous species that might otherwise thrive there can’t grow due to lack of nutrients.

It’s not just the non-living components of a location that matter when selecting trees; interactions with other species also matter. Which is why tree plantation monocultures, an accepted way of earning carbon credits, are often a bad idea, whereas planting a mix of trees can double forest carbon storage.

“In order to get healthy forests that capture carbon in the long term, we must have a natural diversity of species that are necessary for that ecosystem to survive,” says ecologist Tom Crowther of ETH Zürich in Switzerland. Not only do monoculture plantations capture less carbon than diverse natural forests, “they do not survive well in the long term because they are very susceptible to the same threats or diseases.”

Pests and diseases are often adapted to attack one or a few species, so diverse forests make it more difficult for a pest to jump to its next host. In contrast, “when you have hectares and hectares of the same species, this pest can just come and eat everything up and really ravage an entire area very quickly,” says Cynthia Gerlein-Safdi, an ecohydrologist at the University of California, Berkeley.

To properly restore a site, conservationists need to select complementary groups of species to create diverse functional ecosystems. That requires still more data.

Setting up a natural process of succession

Add to all these complexities yet another: Time. Natural ecosystems don’t go from bare ground to dense forest overnight, so it’s foolish to expect reforestation projects to work this way.

Ecological succession is the process by which one community of species paves the way for new species. For example, bare rock might be colonized by mosses and lichens, which create soil that allows grasses to take root. Over time, grasses may be replaced by shrubs and trees, creating a shaded understory that offers habitat for shade-loving trees and understory vegetation.

Tree-planting projects often come in partway through this process, so pioneer species that tolerant bright sunlight need to be selected, rather than shade-loving species more characteristic of the mature forest that once stood there. However, not all pioneer species are able to tolerate the degraded soils found in many deforested areas, making it doubly challenging to find the right species to kick-start a succession process.

Artificial intelligence sheds light on tree preferences

Science has recently seen advances to address this set of problems. To provide more quantitative data, researchers have developed environmental niche models that estimate the suitable habitat for species based on their current distribution. These models analyze data on climate, soil conditions and other characteristics within a species’ range to calculate the tree’s requirements. This approach for predicting habitat suitability has been in development in research labs around the world for more than 20 years, but the models have only recently begun to be applied on the ground to guide reforestation projects at the planning stage. For example, researchers at the Alliance of Bioversity International and the International Center for Tropical Agriculture (CIAT), in collaboration with reforestation practitioners, developed a free online tool, Diversity for Restoration, that recommends a diverse set of tree species for tropical forest restoration projects by modeling habitat suitability.

“Niche modeling is a great way of understanding where different species can naturally exist,” says Crowther, but these top-down models then need to be paired with local knowledge, such as past management interventions and prevalent invasive species. often comes from traditional and Indigenous communities, serving as a powerful adjunct to computer modeling.

An additional complexity is created by climate change, since what thrives now may not survive in decades to come. To hit this moving target, niche models can be combined with climate forecasts to determine how suitable habitats might move with climate change.

“Ecologists now have exceptional data that allows us to predict which species might be suitable for future climate conditions,” says Crowther. But the future isn’t set in stone — society’s actions will determine the climate trajectory Earth takes — and the further into the future forecasts go, the more uncertainty is introduced into the models.

With so many tree species to consider, and so many other variables, scientists are turning to artificial intelligence to sort through the data and make better-informed predictions about the best species suited to an ecosystem. Machine-learning algorithms are adept at identifying patterns in large, complex data sets, allowing for detailed analysis of ecological niches for tens of thousands of species. For example, researchers at the Ministry of Forests in British Columbia, Canada, developed a climate change-informed tree species selection (CCISS) model that uses machine learning to model the future distribution of different habitat types under 90 climate change scenarios, and makes site-specific predictions about the suitability of native tree species.

“Machine-learning approaches give us unparalleled ability to predict” what trees will thrive where, says Crowther. However, these algorithms usually don’t explain the mechanisms behind their predictions. “It is not always easy to understand why those predictions look like [they do].”

Run the labs

This lack of mechanistic understanding of exactly what makes a species suited to a habitat has driven the Spades team out into the field to gather vital real-world data. Spades offers tailored ecological assessments for forest landscape restoration through its commercial service, Ecofit. The team visits tree-planting sites and collects tissue samples from all the plant species found locally. By “running lab tests on those samples, we can make predictions about what that tree likes,” reports Menard. “It’s like having a blood test.”

This detailed analysis can determine each plant species’ minimum, maximum and optimal conditions for life, which can then be matched to current data and future predictions about site conditions. “We can do a quantitative measure across a wide variety of issues like soil salinity, soil toxicity, types of soils, access to water, drought tolerance,” and more, he explains.

These tests are time-consuming and expensive, but all the effort pays off. “In three months you know more about these trees than you know about a pine tree that’s been planted all over the world,” said Menard. This sort of in-depth analysis is already being applied to reforestation projects worldwide and could help make mega-reforestation projects more successful in the future.

The enormous complexity of selecting the ideal tree species for reforestation sites is one reason so many tree-planting projects fail to live up to their initial hype, and why it’s cheaper and simpler to assist ecosystems to regenerate naturally, if possible.

Nevertheless, large swaths of degraded land will need a helping hand to regrow trees in the future if global carbon storage efforts are to succeed. The ongoing work of scientists and entrepreneurs to better understand the needs and limitations of different tree species will be essential if humanity is to reforest the Earth.

Banner image: A reforestation project in the Democratic Republic of Congo. Image by Axel Fassio/CIFOR via Flickr (CC BY-NC-ND 2.0).

Citations:

Cazzolla Gatti, R., Reich, P. B., Gamarra, J. G., Crowther, T., Hui, C., Morera, A., … Liang, J. (2022). The number of tree species on Earth. Proceedings of the National Academy of Sciences, 119(6), e2115329119. doi:10.1073/pnas.2115329119

MacKenzie, W. H., & Mahony, C. R. (2021). An ecological approach to climate change-informed tree species selection for reforestation. Forest Ecology and Management, 481, 118705. doi:10.1016/j.foreco.2020.118705

van Tiel, N., Lyu, L., Fopp, F., Brun, P., van den Hoogen, J., Karger, D. N., … Pellissier, L. (2021). Mapping tree species for restoration potential resilient to climate change. bioRxiv, 2021-06. doi:10.1101/2021.06.04.447113

Chen, Y., Wu, B., Chen, D., & Qi, Y. (2019). Using machine learning to assess site suitability for afforestation with particular species. Forests, 10(9), 739. doi:10.3390/f10090739

Hengl, T., Walsh, M. G., Sanderman, J., Wheeler, I., Harrison, S. P., & Prentice, I. C. (2018). Global mapping of potential natural vegetation: An assessment of machine learning algorithms for estimating land potential. PeerJ, 6, e5457. doi:10.7717/peerj.5457

Castro, J., Morales‐Rueda, F., Navarro, F. B., Löf, M., Vacchiano, G., & Alcaraz‐Segura, D. (2021). Precision restoration: A necessary approach to foster forest recovery in the 21st century. Restoration Ecology, 29(7), e13421. doi:10.1111/rec.13421

FEEDBACK: Use this form to send a message to the author of this post. If you want to post a public comment, you can do that at the bottom of the page.