COMPARING TREE DETECTION FROM SATELLITE AND DRONE IMAGES FOR VEGETATION MAPPING NEAR ASTANA

Abstract

In rapidly urbanizing regions such as Astana, where glass towers rise beside aging Soviet blocks and green spaces struggle within concrete grids, urban vegetation plays a role far beyond ornamentation - it becomes a critical agent of sustainability and climate moderation. Managing these green lifelines requires precision: regularly updated, high-resolution tree maps produced through automated detection and georeferencing. This study compares two approaches: satellite imagery and aerial photographs captured by drones hovering 100 meters above the urban canopy.

Training data and annotations were prepared on the Roboflow platform, while detection relied on a YOLO neural network architecture adapted to the local urban fabric. Our focus extended beyond detection accuracy to include processing speed and operational limitations of each method. The contrast was striking: drone imagery, with its ~5 cm/pixel resolution, achieved accuracies of 70 – 80%, offering a detailed portrait of the city’s vegetative texture. Satellite data, limited to ~50 cm/pixel, delivered a more modest 50 – 60% accuracy. However, its advantage lies in broader coverage and faster processing, making it suitable for regional-scale monitoring rather than fine-detail inventory.

All detections were geolocated and displayed on an interactive map using ArcGIS and Folium, with each tree pinpointed and linked to its crown image-transforming data into a navigable urban landscape.

Drone data thus proves superior for fine-grained tree inventory, its fidelity unmatched at the micro-scale, while satellite imagery retains value for macro-level surveillance. Together, they form a complementary toolkit for green infrastructure planning, biomass estimation, and ecological governance within urban GIS systems. By integrating both, cities can pursue resilient, data-driven strategies rooted in green.