Machine learning shows how climate extremes are changing global vegetation
For many years, experts have studied how changes in climate patterns drive major changes in vegetation around the world. Indeed, as various regions experience unprecedented changes in temperature and rainfall, plants will gradually migrate to places where they could not grow before, or even disappear altogether, causing serious impacts on wildlife and humans. , in addition to preparing the ground. for systemic feedback loops that will further accelerate and aggravate the climate crisis.
Typically, scientists now collect data on these vegetation patterns which are analyzed with what is called a global dynamic vegetation model (DGVM), a kind of computer-simulated model that helps them predict the type of changes that may occur.
But plant-climate interactions — especially at such large scales — can be quite complex, and existing techniques such as global vegetation dynamics models have limitations that could prevent them from accurately predicting future outcomes.
To fill this gap, a team of researchers from the University of Oslo, the University of Helsinki and the Norwegian Institute for Nature Research proposes to use machine learning as a way to improve these simulations. computers, with particular emphasis on the impact of climate extremes on the future. predictions.
“Climate variation is the main factor determining the distribution of vegetation around the world,” explain the researchers in their recently published article in the journal Biology of global change. “As the world faces climate change, large-scale future dynamics of vegetation distribution are expected, which in turn could exert strong biophysical and biochemical feedback on climate. However, predicting the distribution The future of vegetation in response to climate change is particularly challenging, requiring a detailed understanding of how large-scale vegetation distribution relates to climate.
As the team also points out, the use of machine learning in science is nothing new. Machine learning has become increasingly popular in biogeoscience, the interdisciplinary field that studies the interactions between biological and geological processes, and such models that are built on observational data present an approach that is both granular and global.
“Models built on observational data offer the possibility of combining higher resolution while keeping surveys at the largest possible scale,” the team said. “In this study, we used a decision tree approach from machine learning to explore available climate and vegetation data, and to systematically re-examine long-standing and re-emerging scientific questions regarding the relationships between climate and vegetation This approach allowed us to analyze whether new climatic thresholds affecting the large-scale distribution of vegetation types could be detected, in particular extreme climatic thresholds that had been overlooked in previous studies.
The team chose to focus on weather extremes because these are the types of conditions that are becoming more frequent and severe, due to climate change. These extremes are statistically diverging from average climate records and are now occurring more often, which can have huge impacts on how the dynamics of vegetation patterns ultimately play out. For example, extreme lack of rainfall or extreme cold are crucial for the proliferation of savannah and deciduous broadleaf forest.
“The predictive performance of species distribution models increases when mean climate predictors are complemented by climate extremes,” the team noted. “Climate change influences the duration, frequency, intensity, timing and spatial extent of climate extremes. For example, daily extremes of temperature and precipitation, in particular, have been observed to increase in frequency and intensity due to global warming with a spatial pattern distinct from average climate changes.
The team’s method involved the use of decision tree models, also known as classification trees or regression trees, which is a form of machine learning that involves a series of queries or hierarchically structured tests. The decision made on a request or test level will influence the next test, and these progressive series of tests will affect the final result.
Since the mechanisms and reasoning behind AI-based predictions are often unclear, researchers chose to use decision tree models because they are easily interpretable, meaning it is easy to determine why a certain classification was made. The team then trained their decision tree models with publicly available current global climate and vegetation data, and tested their ability to predict which type of vegetation would be dominant in a region, given its variables. climatic.
Curiously, the team’s work found that their AI-assisted approach made much more accurate predictions about future vegetation distribution than baseline models. Additionally, the team pointed out the limitations of the “hard-coded climate thresholds” that reference models typically rely on.
“To the best of our knowledge, no attempt has yet been made to use machine learning to understand the threshold conditions that govern and separate globally dominant vegetation types,” the team said. “The results of our decision tree, however, underscore the importance of using climate extremes, particularly day-to-day extremes, to define climate thresholds for different vegetation types.”
Ultimately, the team’s findings indicate that it will be essential for next-generation DGVMs to incorporate variable climate thresholds, drawn from average climate conditions, rather than static thresholds for the entire globe, as is the case. is generally the case now. Such a change will help experts improve their tools and, in doing so, humanity will be better informed – and therefore also better equipped – to face the terrible challenges of climate change.
Read more in the team diary.