AI × remote sensing applied to food, land and environment. One thread assesses pasture and vegetation health from Sentinel-2 multispectral imagery, with spectral signatures feeding feature extraction, model training and validation, biomass estimation and, ultimately, grazing strategy, replacing manual field monitoring with satellite- and drone-based sensing.
Another maps rice at scale: vision transformers applied to multi-temporal Sentinel-1 SAR time series, covering 42,481 km² across nine regions of southern and central Brazil and outperforming prior state-of-the-art approaches.
The environmental thread targets wildfire. Canada sees more than 8,000 wildfires per year, burning an average of over 2.1 million hectares. Today’s risk rating relies on the Canadian Forest Fire Danger Rating System (CFFDRS) and its Fire Weather Index, a largely manual workflow. We combine multi-modal remote sensing, landcover classification and FIRMS fire-hotspot data to predict and monitor forest-fire risk.