Seeing the Forest and the Trees: The First Global High-Resolution Canopy Height Map

Redacción HC
27/10/2023

For decades, scientists and policymakers have faced a major blind spot in forest monitoring: knowing how tall trees are across the globe. Canopy height isn’t just a measurement of tree size—it’s a vital proxy for understanding carbon storage, biodiversity, and ecosystem health. Yet, until recently, data on global canopy height was coarse, fragmented, and incomplete.

Now, a team of researchers from ETH Zurich and Yale University has developed the first global canopy height model at a 10-meter resolution, offering unprecedented detail and coverage. This breakthrough, published as a preprint on arXiv in 2022, leverages satellite imagery and deep learning to map the vertical structure of forests with remarkable accuracy.

Why Canopy Height Matters

The height of a forest canopy is directly linked to its biomass and carbon sequestration potential. Taller trees usually mean more stored carbon, richer biodiversity, and more resilient ecosystems. Canopy height is also key for climate modeling, conservation planning, and sustainable forest management.

But traditional methods for measuring canopy height—such as airborne LiDAR or NASA’s GEDI spaceborne laser system—are limited. GEDI, for example, only covers about 4% of Earth’s surface and at relatively coarse resolution (~1 km). What’s been missing is a model that combines broad coverage with fine detail.

A Deep Learning Breakthrough: Mapping Every Tree at Scale

To solve this, the researchers developed a multimodal convolutional neural network (CNN) that fuses data from two satellite sources:

• GEDI (Global Ecosystem Dynamics Investigation): LiDAR data providing precise, though spatially sparse, height measurements.
• Sentinel-2: Optical imagery with 10m resolution and frequent global coverage.

Using over 600 million GEDI measurements and cloud-free Sentinel-2 composites, the model was trained to predict canopy height globally, excluding Antarctica. Importantly, the CNN also estimates predictive uncertainty, allowing users to assess where results are more or less reliable.

The output? A canopy height map with 10m spatial resolution and 6m average error, along with a corresponding uncertainty map that flags regions with lower confidence—crucial for scientific transparency.

Key Findings from the Global Model

1. Mapping the Vertical Dimension of Forests

The resulting map reveals detailed tree height information across the planet, detecting canopies as high as 50 meters in tropical rainforests. This is the first time that such fine-grained global data has been made available at this scale.

2. Quantifying Uncertainty

The model doesn’t just give a number—it estimates how confident it is in each prediction. By integrating Bayesian techniques, it reports both epistemic (model-based) and aleatoric (data-based) uncertainties, enhancing its utility for decision-makers.

3. Forest Height and Conservation Gaps

Only about 5% of Earth’s surface is covered by tall forests (canopies >30m), and two-thirds of these forests lie outside protected areas. This finding highlights urgent conservation gaps and the need to prioritize protection for intact yet unprotected ecosystems.

4. Annual Monitoring Potential

Because Sentinel-2 captures imagery every five days, the model could be updated regularly to track deforestation, regrowth, or forest degradation over time. This creates new opportunities for dynamic forest monitoring in support of carbon markets, biodiversity tracking, and restoration efforts.

Transforming Forest Management, Climate Science, and Biodiversity Conservation

For Climate and Carbon Tracking

Accurate canopy height measurements are critical for calculating aboveground biomass and refining estimates in global carbon budgets. This model fills a key data gap, helping scientists and policymakers track changes in forest carbon stocks with improved precision.

For Conservation Planning

The map enables fine-scale detection of high-value conservation areas, especially in regions like the Amazon, Congo Basin, and Southeast Asia. It can also support the design of ecological corridors, buffer zones, and climate refugia.

For Biodiversity Research

Tall canopies often harbor complex ecosystems and rare species. This model helps identify biodiversity hotspots that may not be visible in standard land cover maps. Researchers can use the data to correlate canopy structure with species distribution and ecosystem function.

For Communities and Transparency

The team has made all model code and datasets publicly available, ensuring that scientists, NGOs, governments, and indigenous communities can use and adapt the data for local needs—from reforestation projects to carbon offset validation.

What’s Next? Scaling, Validating, and Improving

The researchers recommend several next steps:

• Expand validation in underrepresented regions such as boreal forests and drylands.
• Improve model resolution by integrating terrestrial or aerial LiDAR datasets.
• Automate updates to detect changes as frequently as Sentinel-2 imagery becomes available—approximately every five days.

These improvements could enhance real-time forest change detection, making this tool even more valuable for early warning systems and sustainable land use monitoring.

Conclusion: Measuring Forests with Unprecedented Clarity

This high-resolution canopy height model marks a turning point in remote sensing for ecology. By combining advanced deep learning with open satellite data, the researchers have provided the clearest view yet of Earth’s forests—tree by tree, meter by meter.

It’s not just about pixels or predictions; it’s about empowering conservation, advancing science, and bridging the gap between data and action.

Topics of interest

Climate Technology