Over three million hectares of forests and woodlands span Australia, storing vast amounts of carbon, supporting unique biodiversity, and regulating water resources. Researchers announce that they will soon be able to track variations in these ecosystems in near real-time using a combination of satellite data and artificial intelligence, ushering in a new era of environmental monitoring.
Near real-time forest tracking: satellite data and AI
The core of the new system relies on Earth observation satellite constellations, including European Copernicus Sentinel missions (Sentinel-2 for optical imaging, Sentinel-1 for radar) and a panel of high-resolution commercial satellites. Daily, these platforms transmit optical images capturing soil and canopy reflectance, as well as radar images able to penetrate clouds and probe the three-dimensional structure of forest cover. This acquisition covers the entire Australian territory, from tropical forests in Queensland to vast savannas in the northwest.
The volume of data generated is enormous: several terabytes of raw imagery are available each week, making manual processing impossible. To transform these massive flows into usable information, research teams use convolutional neural networks (CNNs) trained on decades of historical observations. These models learn to recognize the spectral signatures of healthy foliage, differentiate eucalyptus species, identify areas where canopy density decreases, and detect early signs of drought stress or tree mortality.
Additionally, forecasts from the European model ECMWF and products from the Copernicus Climate Change Service are integrated to relate observed changes to climate variables (temperature, precipitation, drought index). This multi-source approach creates a dynamic dashboard where each forest surface pixel is updated in near real-time, providing resource managers unprecedented visibility and the ability to trigger corrective actions within hours rather than weeks.