Café com Física

Machine Learning Detection of Invasive Plant Species in Portugal Using Sentinel-2 Multispectral Satellite Imagery

by Siddique Baig (LIP)

→ Portugal
Sala de Conferências (Departamento de Física FCTUC)

Sala de Conferências

Departamento de Física FCTUC

Universidade de Coimbra
Description

Invasive alien plants, such as Acacia longifolia, A. dealbata, A. melanoxylon, and A. saligna, threaten native ecosystems in Portugal, and monitoring them over large areas requires scalable, low-cost methods. This study combines field observations, phenology-based spectral analysis, and machine learning to detect invasive plants using Sentinel-2 imagery. Field data provided by the RESCUE project on plant density, plant size, presence and number of Trichilogaster galls, and GPS accuracy were analysed, including correlations between gall abundance and both plant density and plant size, to understand how biocontrol pressure relates to plant characteristics. Sentinel-2 Level-2A surface reflectance products, distributed as tiles, were used with the 10 m, 20 m and 60 m bands. Spectral signatures were extracted for the pre-blooming and post-flowering seasons, and changes in the phenological spectral curves and in Sentinel-2 vegetation indices were evaluated to identify the period when invasive plants are most separable from native vegetation and plantations. Two Random Forest models were then trained: one using the four 10 m bands and one using eight 20 m bands. Their predictions were compared using overall accuracy, F1-score under spatial validation, and the resulting maps were summarised by district to compare the extent of invasion across regions. Results show that the 10 m / 20 m model achieved significant accuracy and provided the strongest spectral contrast. This approach offers a reproducible workflow to support invasive plant monitoring and management in Portugal.

Brief Bio-data of presenter:

Siddique Baig, PhD, is a geoinformatics and remote sensing researcher based at the Laboratory of Instrumentation and Experimental Particle Physics (LIP), University of Coimbra, Portugal. He holds a PhD in Geo-Informatics from Universiti Teknologi Malaysia and an MSc in Photogrammetry and Geoinformatics from HFT Stuttgart, Germany. His career spans an assistant professorship at COMSATS University Islamabad, postdoctoral work at TU Eindhoven, and a guest researcher at NOVA University Lisbon’s GeoBioTec. His research covers the application of geoinformatics to bio- and abiotic natural resources, habitat suitability, machine learning, and city modelling, with over a dozen journal articles.

Organised by

Paulo Silva, Marcos Gouveia