Harnessing remote sensing and machine learning techniques for detecting and monitoring the invasion of goldenrod invasive species

Invasive alien species, such as goldenrods (Solidago spp.), pose significant threats to biodiversity and ecosystem services across Europe. Effective monitoring of these species is essential for early intervention and informed management, yet traditional ground surveys are often labor-intensive and limited in scale. This study aims to evaluate the potential of remote sensing and machine learning for detecting and monitoring Solidago spp. in Kampinos National Park, Poland, using multitemporal imagery from Sentinel-2 and PlanetScope satellites. We compared the performance of Random Forest and One-Class Support Vector Machine classifiers across 17 classification scenarios incorporating spectral bands, vegetation indices, and temporal statistics. Our results showed that Random Forest consistently outperformed One-Class Support Vector Machine (OCSVM) by 1%–15%, achieving the highest F1-score of 0.98 using multitemporal Sentinel-2 data and 2%–29% using PlanetScope imagery. Sentinel-2 data, with its broader spectral range, provided better large-scale detection accuracy, while PlanetScope's higher spatial resolution enhanced local detail. Goldenrod patches are distinctive even in autumn and winter due to living or dry biomass that persists the whole year. In our study autumn imagery (October–November) yielded the most reliable detection due to distinct phenological characteristics of Solidago during this period. Importantly, our analysis demonstrates that the added complexity of vegetation indices does not necessarily improve classification accuracy for goldenrod detection. Our findings present high-accuracy invasive species monitoring approach and highlight the critical role of phenological timing in remote sensing-based ecological assessments.