Mapping invasive alien trees in water towers: A combined approach using satellite data fusion, drone technology and expert engagement

Abstract

Operationalising advances in remote sensing for invasive alien plant management remains a major challenge, despite its proven value globally. There is also a lack of detailed remote sensing studies in water towers, for example the southwestern Cape of South Africa, where invasive alien trees threaten water security. There is a need for cost-effective, easy to use, inclusive and repeatable approaches to map invasive alien plants to inform management. We use a novel, transdisciplinary approach, combining Google Earth Engine's processing power, freely available Sentinel imagery (fusion of Sentinel-1 and -2), expert engagement (including researchers, managers and decision makers), drone technology and field trips, to provide an accurate and up-to-date understanding of the occurrence and density of invasive alien trees in an important water tower for the southwestern Cape of South Africa at a 20 m resolution. We explored the efficacy of combining bands and indices with the highest alien tree discriminatory power based on a statistical spectral analysis of 10 of the 13 Sentinel-2 bands and 39 relevant indices with various other data inputs (e.g. Sentinel-1, topographic information) in six classifications using robust and advanced non-parametric image classifiers. All six classifications performed well, with invasive alien trees discriminated from surrounding shrublands with 89-92% accuracy, and alien tree groups discriminated with 74-84% accuracy. Data fusion of Sentinel-1 and -2 and inclusion of topographic information (elevation and landform) marginally improved the accuracy statistics. However we caution against over-reliance on accuracy statistics given the relatively small sample sizes typically used in multispectral classifications. The rich spectral information contained in the red edge and shortwave infrared parts of the spectrum were critical for alien tree discrimination as the key traits distinguishing these alien trees from indigenous shrubland vegetation are differences in biomass and water usage. Discrimination between different types of alien trees and discriminating alien trees from native vegetation with high water use (e.g. wetlands and mountain forests) remained a challenge for the given spatial and spectral resolutions of Sentinel-2 imagery, despite reasonable accuracy statistics. Though the results present a significant advance for the region, given currently available out-dated maps, engagement with decision makers showed that managers require even more detailed products for alien tree management.