Moropane, Mmasechaba L.Dube, TimothyShoko, CletahMasocha, Mhosisi2026-05-292026-05-292026Moropane, M.L., Shoko, C., Masocha, M. and Dube, T., 2026. The analysis ready multispectral data and machine learning algorithms for mapping groundwater-dependent invasive alien plants in Heuningnes Catchment, South Africa. Geocarto International, 41(1), p.2631854.10.1080/10106049.2026.2631854https://hdl.handle.net/10566/22953Field monitoring of groundwater-dependent invasive alien plants (GDIAPs), though accurate, is spatially limited and prone to misclassification bias. Satellite imagery and machine learning algorithms (MLAs) offer scalable alternatives, yet their effectiveness in mapping GDIAPs remains underexplored. This study assessed Sentinel-2 and Landsat-8 data using five MLAs with pixel- and object-based approaches to delineate GDIAPs within groundwater-dependent ecosystems (GDEs) of the Heuningnes Catchment. Pixel-based random forest (RF) achieved the highest overall accuracies (98% with Sentinel-2, 94% with Landsat-8), while object-based gradient tree boosting (GTB) using Sentinel-2 achieved 85%, outperforming RF (82%). The class-specific F1-scores for GDIAPs and native plants exceeded 90% when pixel-based RF classifications were used. The Sentinel-2 pixel-based RF results indicated that GDIAPs dominated 26.8 km² (60%) of GDEs in 2022, compared to 13.6 km² (31%) of GDEs occupied by native fynbos. These findings demonstrate the value of satellite-based machine learning for monitoring GDIAPs and supporting invasive species management, ecosystem protection, and water resource planning.enEcosystem functiongroundwatermachine learningmultispectralplantsArticle