Spatial patterns of cattle densities across the Brazilian Amazon revealed by very high-resolution satellite imagery

Abstract

Cattle ranching is a sustainability challenge worldwide, and in the Amazon, the planet’s largest tropical forest, it remains the main driver of deforestation. Yet, cattle numbers have typically been estimated from coarse census data or indirect proxies, limiting our ability to monitor land-use change at finer scales. Here, we introduce a novel approach that applies deep learning-based density estimation to very high-resolution satellite imagery to detect individual animals across the Brazilian Amazon. Our cattle data set covers over 12,000 km² in four states and is integrated with pasture maps to analyze property-level stocking rates. We find patterns of extensive land use, deriving conservative stocking rate estimates of 0.73 head per hectare in 2018–2019, with lower cattle stocking rates on properties with higher recent deforestation and properties further away from slaughterhouses. While the use of VHR imagery presents challenges of coverage and detection, our framework establishes a foundation for advancing livestock monitoring and supports strategies to address deforestation and promote sustainable resource management.