Sorting the Weeds from the Grass: Using Plant Phenology to Improve Discrimination of Serrated Tussock Cover from Satellite Imagery

Serrated tussock (Nassella trichotoma) is an invasive weed species in Australia, impacting both agricultural production and native biodiversity. Rapidly evolving advancements in remote sensing and spatial analytics have expanded tools available to land managers for weed detection and monitoring. This paper explores the extent to which plant phenology can be used to improve the classification of serrated tussock cover from satellite multi-spectral imagery. Using a random forests approach and multi-seasonal Sentinel-2 imagery, serrated tussock cover was classified with an overall accuracy of 64.2%, and cover class F1-scores ranging between 0.49 and 0.74. Results showed that using imagery from all four seasons (autumn, winter, spring and summer) improved the overall classification accuracy by 7.0-11.3 percentage points when compared to analyses that considered imagery only from a single season. This study demonstrates the usefulness of factoring plant phenology into these types of approaches.