How Can Satellite Light Measure the Amazon Forest's Revival?

Recovering degraded areas of the Amazon is a huge challenge. Researchers have found that satellite images can help track this process without costly field trips. They used Sentinel‑2 photos, a satellite that scans Earth every five days, to calculate "spectral diversity." This measures how many color tones appear in an area, similar to counting species in a forest.

In four compensation zones and three rehabilitation sites in Pará, they compared spectral diversity with the Restoration Index (RI), which combines tree height, ground cover, and species composition. The results show that two spectral metrics – Shannon diversity and spectral richness – are most closely linked to the RI. An increase of 1 % in the RI corresponds to a rise of 0.0198 in spectral richness. When the RI reaches 70 %, the predicted spectral richness is 2.29.

Analyses also proved that spectral diversity stays stable across different seasons, making it reliable for monitoring forest evolution over time. In practice, scientists can now observe large forest stretches remotely, saving time and money, and provide transparent data to assess the success of restoration projects.

Spectral diversity is a concept borrowed from biodiversity theory and applied to multispectral imagery. Each Sentinel‑2 pixel contains information from several wavelength bands (visible, near‑infrared, short‑wave infrared). These bands interact differently with vegetation: chlorophyll absorbs red light, while canopy structure reflects near‑infrared. By calculating the variety of reflectance combinations across bands, we obtain a measure of the area's spectral complexity.

In this study, four metrics were tested: spectral richness (number of unique combinations), Shannon index (weighing the frequency of combinations), Rao’s Q (incorporating distances between combinations), and functional composition indices (simulating ecological traits). Correlation analyses highlighted Shannon and spectral richness as the most responsive to the Restoration Index (RI), likely because they capture both variety and distribution of spectral classes.

The linear regression model provided a quantitative estimate: each percentage point increase in RI adds 0.0198 units of spectral richness. This value was validated on two sites surveyed twice, confirming the relationship's robustness despite seasonal variability. Moreover, moderate correlation between images from different months suggests spectral diversity is less affected by temporary phenological changes, making it suitable for long‑term monitoring.

Operationally, using the BiodivMapR package in R automates the workflow: downloading Sentinel‑2 scenes, atmospheric correction, computing spectral metrics, and regressing against RI. This approach scales to landscape level, cutting monitoring costs from hundreds of thousands to a few tens of thousands of euros per year, depending on area coverage. In summary, spectral diversity bridges remote sensing data and ecological indicators, offering a transparent, reproducible solution to assess forest regeneration progress in complex environments like the Eastern Amazon.