Machine learning and remote sensing for mapping land cover and conditions

This project involves the development of advanced machine learning (ML) frameworks based on transfer learning that integrate multi-sensor remote sensing data. These frameworks support detailed land cover mapping and the assessment of agricultural plant health. The approach leverages pre-trained ML models and adapts them systematically across different agro-ecological regions, management systems and sensor configurations. Specific consideration is given to high-spatial-and-temporal-resolution data from the Sentinel-2 and Planet constellations. The project aims to improve the robustness and generalisation capacity of ML models under varying environmental conditions by combining complementary spectral, spatial and temporal information. There is a particular focus on transferability across sites and seasons to reduce the need for extensive local training data while maintaining high classification and diagnostic accuracy. The resulting frameworks are designed to be scalable and operational, enabling the consistent monitoring of vegetation status, crop conditions and land cover dynamics. They will support decision-making in precision agriculture, ecosystem monitoring and sustainable land management.