Foundation-Model Earth Representations Enable Regional-Scale Forest Aboveground Biomass Monitoring Across the Northeastern United States

Imagine the Earth as a giant, breathing lung. Every time this lung inhales, it pulls in carbon dioxide from the air and stores it inside its tissues—trees, leaves, and roots. This stored carbon is called "aboveground biomass," and it's the secret weapon forests use to fight climate change. But here's the tricky part: to know how much carbon a forest is holding, scientists usually have to go out, measure every single tree, and do some heavy math. It's like trying to count the grains of sand on a beach by picking them up one by one. It's slow, expensive, and often impossible to do everywhere at once.

In recent years, scientists have tried to use satellites to take a shortcut. Instead of measuring trees directly, they look at pictures from space to guess how much wood is there. But traditional satellite pictures are a bit like looking at a forest through a foggy window; they can see the green canopy, but they can't tell you how tall the trees are or how thick the branches are. That's where a new kind of "super-satellite" tool comes in. Think of these tools as a magical translator. Instead of just showing a picture, they turn the entire landscape into a secret code—a long list of numbers that describes the forest's shape, health, and history all at once. This paper asks a big question: Can we use this new "secret code" from a global AI system to count the forest's carbon storage accurately, even in places where we don't have the expensive, high-tech 3D maps of the trees?

---

The Forest's Secret Code

In this study, two researchers from the University of Connecticut decided to put this new "secret code" to the test in the forests of the Northeastern United States. They wanted to see if they could use a powerful new tool called Google Satellite Embeddings (GSE) to estimate how much wood is stored in the trees.

To understand the challenge, imagine you are trying to guess the weight of a backpack.

• The Old Way (LiDAR): You have a 3D scanner that measures the exact height and shape of the backpack. It's super accurate, but you can only use it if you have the scanner right there. In the real world, this is LiDAR, a laser technology that creates detailed 3D maps of forests. It's great, but it's rare and expensive, like having a 3D scanner for only a few backpacks in a whole school.
• The New Way (GSE): You have a super-smart AI that has looked at millions of backpacks from space, weather reports, and maps. It doesn't see the backpack directly, but it has learned a "secret code" (a list of 64 numbers) that describes what a backpack usually looks like based on its surroundings. This is the Google Satellite Embedding. It's available everywhere, every year, but it's a bit of a mystery because it's a "black box"—we know it works, but we don't always know exactly which number means "tall tree" and which means "thick leaves."

The Experiment: Mixing the Tools

The researchers gathered data from the Northeastern Forest Inventory Network (NEFIN), which is like a massive logbook of forest measurements taken by humans over decades. They had two main goals:

1. Could they use the "secret code" (GSE) alone to guess the weight of the forest?
2. Could they mix the "secret code" with the 3D laser maps (LiDAR) to get an even better guess?

They fed all this information into three different types of computer "brain" models (called Random Forest, XGBoost, and ExtraTrees) to see which one could learn the best.

The Results: A Winning Combination

Here is what they found, and it's pretty exciting:

• The 3D Laser Maps are Still the King: When they used only the 3D laser maps (LiDAR), the computer models did a great job, reaching an accuracy score (called R²) of 0.79 after fine-tuning the settings. This means the model could explain about 79% of the changes in forest weight.
• The Secret Code is a Strong Runner-Up: When they used only the "secret code" (GSE), the models still did surprisingly well, reaching an accuracy of 0.67 with the smaller dataset. This proves that the AI's secret code actually knows a lot about the forest, even without the 3D laser maps.
• The Power of Teamwork: The absolute best result came when they combined the 3D laser maps with the "secret code." This team-up pushed the accuracy up to 0.79 as well, showing that the two methods complement each other perfectly. It turns out that the 3D maps tell the computer how tall the trees are, while the secret code tells it what kind of forest it is and how it's changing over time. Together, they are a perfect match.

The Big Breakthrough: Filling in the Gaps

The most surprising discovery happened when they stopped worrying about the 3D laser maps. Because the "secret code" (GSE) is available every single year for the whole world, the researchers were able to use way more data. They expanded their study from a small group of 589 forest plots to a massive 6,801 plots!

By using this huge amount of data and letting the computer learn from the "secret code" alone, they achieved their highest accuracy score yet: 0.82.

What This Means for the Future

This study suggests that we don't always need expensive 3D laser maps to monitor forests. If we have good ground measurements (like the NEFIN logbook) and we use this new global "secret code," we can create accurate maps of forest carbon storage for entire regions, even in places where we've never flown a laser scanner.

The researchers found that by using this method, they could reduce model bias by over 70% compared to models that didn't use the annual growth adjustments and GSE data. They also showed that by tweaking the computer's settings (a process called hyperparameter optimization), they could make the results much more stable and reliable.

The Bottom Line

While the 3D laser maps are still the most precise tool for small, detailed areas, this new "secret code" from Google's AI is a game-changer for large-scale monitoring. It allows scientists to fill in the gaps where laser maps don't exist. The study concludes that by combining these new AI tools with traditional forest data, we can finally keep a much better, year-by-year eye on how much carbon our forests are storing, helping us understand and protect our planet's climate more effectively than ever before.