Predicting world’s vegetation under climate change scenarios: maps of forest areas most threatened by climate change

8 min readFeb 8, 2023

Prepared by: Valentina Delconte (OpenGeoHub) and Carmelo Bonannella (OpenGeoHub)

Biomes, large communities of vegetation and wildlife adapted to a specific climate, are like many living systems dynamic and in constant flux. Recently, there has been an increasing in interest in understanding effects of climate change on biomes of the future (e.g. 2040, 2060 and beyond). Climate change is one of the biggest threats to human civilization, with slowly accumulating effects and unknown instabilities in front of us and future generations. It will also affect global distribution and resilience of biomes. But not all biomes will be affected equally by climate change. Machine Learning can tell us where and when climate change will alter ecosystems the most: by the end of the century we may observe less forests on our Planet, a new study reports.

Rising temperatures due to anthropogenic activities are reshaping ecosystems and the distributions of living communities on Earth for generations. Scientists have documented innumerable cases of local extinctions caused by diminishing resources, or the fierce competition with invasive species and, ultimately, by changing temperatures and precipitation regimes. When mending these stories together in a global context, experts have been able to track shifts of entire geographical ranges for individual species or whole biomes — highlighting that the extent of these transformations could crack new boundaries.

In a recent study now available as a pre-print led by Carmelo Bonannella, PhD candidate at OpenGeoHub, it emerged that forests will undergo a major transformation by 2080, regardless of how much temperatures are predicted to rise: “The biomes expected to shift the most are the tropical and subtropical biomes, particularly the tropical rainforests: these biomes are expected to experience a decrease in their potential distribution in the future 60 years towards savannah and grassland biomes, a process called ‘savannization’”.

Spatial location of biome transitions for scenario RCP 8.5 epoch 2061–2080. In red: all the
areas that will change according to our model with a margin of victory value ≥50%. In yellow: all the
transitioning areas with margin of victory values <50%.

In contrast, biomes located at higher latitudes, such as boreal forests, are expected to experience an expansion in their potential distribution in the future time periods “at the expense of the polar biomes” according to the results of the research.

When designing this study, the authors intended to create “maps that can be used to locate hot spots of change, from which the shift in ecosystem can then expand”. However, there are many factors impacting the reliability of the projections, as Bonannella explained, such as the limited number of studies using Machine-Learning for biomes, the lack of ground data from less surveyed regions and the overall uncertainties related to climate models: “We wanted to provide consistent projections of future natural vegetation under different climate scenarios, so we included the uncertainty estimates for each projection to invite a careful interpretation of our results”. For the purpose of this study, the team considered natural vegetation to be the vegetation that would cover an area if no human related activities (buildings, agriculture, roads, pollution, etc) had impacted it, “hence why there’s no mention of cities in the maps, they are considered as virtually non-existent.

Boreal forests: trapped in a warming ground

The projections of 20 biomes in the three most commonly used climatic scenarios (RCP 2.6, 4.5 and 8.5) have shown specific emerging trends in biome shifts in precise locations of the globe. While differing in size and depending on the time range considered, such transformations are evident across all three climatic scenarios — shedding light on those areas that are most vulnerable to climatic changes.

One of these emerging trends is the transition from a polar to a boreal forest biome in the global North, around the Arctic Circle, according to Bonannella. In all time ranges analyzed (present–2040, 2041–2060 and 2061–2080), the replacement of the boreal steppes (also known as tundra) by the forested ecosystem is one of the most evident and consistent transitions detected by all three climatic scenarios. This trend goes hand in hand with consistent observations of tree line to advance towards the North Pole as a consequence of rising temperatures: it is estimated that between 20 and 40% of the Arctic tundra is turning literally greener, i.e. into forest, since the 1980s, a trend supported by ground observations of shrub cover increases.

Predicted shifts from tundra to boreal forest.

Bonannella pointed out that the results of this study, “regardless of the scenario, the predictions show that while the change will happen, it won’t be uniform across North America, Asia and Europe: North America seems to be the one that will be most affected, while only in a few areas of Siberia all scenarios agree in detecting a biome shift.

This is why the study also provides ‘uncertainty maps’ indicating the error of the prediction and to guide towards a more “careful interpretation” of the results. The issue with lack of ground data from some boreal regions, in this case Siberia, is that “it affects the precision of our model predictions as some locations are seriously underrepresented.” The author highlighted therefore the importance of gathering more comprehensive data in the future to improve the model’s accuracy and prediction abilities also for these underrepresented regions.

Predicted probability of occurrence of ”warm temperate evergreen and mixed forest” class,
zoom in on the area around the Pyrenees. The probability values over time show that the class is slowly
shifting towards northern latitudes. Only the RCP 4.5 scenario is shown since it is considered as the
”middle of the road” scenario. Points indicate training points from the BIOME 6000 dataset.

This phenomenon is explained by warmer temperatures causing longer seasons which are favorable for more trees to grow. Meanwhile, temperatures along the southern edge are becoming too hot to sustain forest growth, leading to die-backs.

Despite the uncertainty in less surveyed areas, these maps can provide a robust indication of where the largest transformations are more likely to occur in all climate scenarios. Even if boreal forests could appear to be escaping the unbearable temperatures in their southern range, scientists are warning, ‘the rate at which trees die in the lower-latitude edge is much faster than the rate at which they’re able to expand north-wise’ — emphasizing the slow, but steady downsizing of the boreal forest.

What is certain for scientists is that the effects of warming temperatures in the Circum-Arctic region are resulting in further permafrost thawing, biodiversity loss and carbon release in the atmosphere, deeply impacting local infrastructures and communities. Without immediate action, these damaging effects will continue, or even exacerbate through positive feedback loops. “We recommend that this information is used by policy makers and land managers to make informed decisions about the management and conservation of these ecosystems, and to take action to mitigate the negative consequences of climate change on communities and economies” concluded Bonannella.

Predicted “savannization” in Europe under the most extreme climate change scenario RCP 8.5. Predictions of future distribution of biomes (including probabilities per class) under different climate scenarios are available as open data via:

A paper cut in the Amazon

More than one third of what’s left of the Amazon forest is threatened by deforestation, human activities and droughts according to the recent analytical reviews published in Science.

These findings are part of a long series of studies showing “alarming signs of an incoming process of savannization in the southern edge of the forest” , commented Bonannella, which is redrawing the forest’s range. His computer models confirmed the dire trend, exacerbated by the impacts of climate change. The projections detected a consistent transition from rainforest to tropical grasslands, also known as savannahs, especially in the regions between North Bolivia and in the Southern edge of Mato Grosso, Brazil. In that same areas, the models forecasted that with just an increase of 2°C in temperatures “we may face a loss of about 59,000 km2 of Amazon forest”, whilst in a 5°C (RCP 8.5) scenario, the lost forest surface would expand up to 1,1 million km2.

In fact, experts have been long warning that rainforests are highly sensitive to changes in rainfall and moisture levels — both impacted by a changing climate. Additional disturbances such as human-caused fires and prolonged droughts can result in forest areas losing trees and shifting to a savannah-like mix of woodland and grassland: recent satellite-based research is showing that more than 75% of the untouched forest has lost stability since the early 2000s and that as much as 40% of the existing Amazon forest is now at ‘point where it could exist as a savannah’.

Predicted shifts from rainforest to savannah.

The consequences of climate change on the Amazon rainforest are at the center of scientific, political and economic debates given the critical role in global climate regulation, biodiversity conservation and in globally-demanded resources such as timber, materials, food and carbon. Additionally, the loss of the tropical rainforest could have “significant ecological implications for the distribution and diversity of plant and animal species, impacting also local communities that depend on these ecosystems.”

While the full extent of these changes in the Amazon basin has not yet been completely realized, Bonannella’s prediction maps add up to the overall picture by identifying areas where these impacts could be particularly significant — and where to focus policy efforts.

Even though the use of Machine Learning for this task is pioneering new ways to support evidence-based decision making and to track global sustainability goals, the limited number of studies and lack of long-term data still challenge the accuracy and reliability of the global prediction models. In the future, Bonannella wishes to see “more collaborations with local research networks and communities to collect ground observations which can aid our models, as well as opening up already available data, making it accessible and usable to us modelers following the principles of free and open research.

  • Bonannella, C., Hengl, T., Parente, L. et al. “Biomes of the world under climate change scenarios: increasing aridity and higher temperatures lead to significant shifts in natural vegetation”, 12 January 2023, PREPRINT (Version 1) submitted to PeerJ,

Note: OpenGeoHub has previously contributed to the NatureMap project helping map potential natural vegetation (“planet without people”). Carmelo Bonannella’s work is part of the Open-Earth-Monitor project, that has received funding from the European Union’s Horizon Europe research an innovation programme under grant agreement №101059548. The Open-Earth-Monitor project aims to accelerate the uptake of environmental information and create active user communities. By applying open data principles and collaborating with ground observation networks, the consortium led by OpenGeoHub plans to create easy-to-use monitoring services for environmental trends at regional, national and global levels.





Not-for-profit research foundation that promotes open geographical and geo-scientific data and develops open source software.