A machine learning technique that can identify optimal spots for tree planting to inform climate mitigation strategies is being unveiled by scientists from the European Centre for Medium-range Weather Forecasts.
Afforestation – establishing forests on previously non-forested land, or where forests have not existed for a long time – is one of the nature-based and cost-effective solutions for climate change mitigation because it offsets carbon emissions through carbon storage and can help reduce the effects of flooding.

The European Union’s Biodiversity Strategy for 2030 targets converting at least 10% of agricultural land into forest.
However, previous modelling studies suggest that while afforestation can help lessen the impact of flooding, conversely, it can exacerbate water scarcity and exacerbate wildfires when implemented on a large-scale. Studies also show the benefits of afforestation depend on factors such as species choice, soil and landscape conditions, regional climate and spatial configuration.

“Afforestation is a careful balancing act between reducing flooding and managing water scarcity,” said Fredrik Wetterhall, Senior Hydrologist at ECMWF. “Our study examines the hydrological impacts of converting abandoned croplands in Europe into forests. Where, and how much to plant, matters. Optimised afforestation, guided by ecological and hydrological criteria, can reduce river peaks while preserving groundwater.”

Publishing in Nature Communications Sustainability (10 April 2026), Fredrik and his ECMWF colleagues outline how a smart, data-driven algorithm can help identify optimal sites, minimising the known negative effects of afforestation on water availability.

The study shows that, with a targeted selection of afforestation sites, river flooding could be reduced by up to 43% with a median reduction of 3.1%. Random selection only leads to flood reduction in a few cases and has mostly no effect. Optimised afforestation also curbs evapotranspiration, the transfer of water from plants to the atmosphere, reducing water loss and thereby preserving available groundwater by up to 60%, a critical save, given the growing need to conserve water in a changing climate. The combined effects leads to a “sweet spot” of optimal afforestation between 40-80%.

Siham El Garroussi, the machine learning scientist who developed the genetic algorithm used in the study, explains that the model was optimised to enhance water retention. Its flexibility, however, allows other criteria to be incorporated, such as reducing wildfire risk or improving ecological biodiversity. “The constraints guiding the machine learning model’s decisions are defined by a physical discharge model, showing the power of combining physical understanding with machine learning.”

The team also tested how both strategies perform under a +2°C warming scenario and found that both are similarly affected. “This means the optimised approach would remain the most effective option and is likely to stand the test of time in a warming climate,” added Fredrik Wetterhall.

The study highlights that implementing nature-based solutions can be more complex than expected and may involve trade-offs. Data-driven technologies can support large-scale, smarter planning—delivering benefits for both carbon mitigation and sustainable water management.

You can read more in the published paper here: https://rdcu.be/fcN8m