Forests interact with seismic waves. Tree roots couple to the ground; trunks and canopy resonate at frequencies set by their height, stiffness, and mass. Ambient seismic noise — the constant low-level vibration from ocean waves, wind, and human activity — passes through forested ground differently than through open terrain.
The signal lives in the 35–60 Hz range (arXiv:2602.18485). Below 35 Hz, the wavelengths are too long to interact with individual trees. Above 60 Hz, attenuation dominates. In between, forests leave a detectable imprint on the seismic noise spectrum: specific amplitude patterns, spectral shapes, and phase relationships that differ from grassland, bare soil, or urban cover.
Machine learning classifies vegetation type from seismic data with 86% accuracy. But the more interesting finding is topological: the researchers use persistent homology — a tool from topological data analysis — to extract features from the seismic signals that are invariant to rotation, scaling, and small perturbations. The topological features capture the shape of the spectral pattern rather than its absolute values, making the classification robust to seasonal changes, weather, and sensor calibration differences.
The practical implication is monitoring forests with existing seismic infrastructure. Seismometer networks designed for earthquake detection are already deployed worldwide. They are running continuously, recording ambient noise, discarding it as irrelevant. That discarded noise contains information about the forests above the sensors. Deforestation, regrowth, species composition changes — all should alter the 35–60 Hz signature.
You don't build new sensors. You listen to what the old ones have been hearing all along.