Detection of wilding pines with Sentinel-2 and WorldView-3 satellite data

Authors: Sebastian Klinger, Nicolò Camarretta, Grant D. Pearse, Benjamin S.C. Steer and Thomas Paul
Publication: New Zealand Journal of Forestry, Volume N.Z.J.For. 2024, Issue N.Z.J.For. 69(1) 2024, pp Pages 17 - 24, May 2024
Publisher: New Zealand Institute of Forestry

---

Abstract: Invasive species are a threat to rare ecosystems and can cause adverse effects to indigenous biodiversity, productive lands and natural landscapes. In New Zealand, wilding pines introduced in the last century have invaded around 2 million ha. The New Zealand Government spent nearly $100 million over the last five years to stop further spread and control large infestations in many parts of the country. Remote sensing offers the ideal tool to monitor the development of existing and new infestations and track the success of the national management control programme efficiently. Here, we tested freely available Sentinel-2 satellite data to detect wilding pine infestations. We developed pixel-based classifiers to detect infested areas in Sentinel-2 imagery and compared their accuracy against that of classifiers developed from very high-resolution imagery (WorldView-3). A 98% accuracy was achieved in the classification of healthy invasive conifers, controlled unhealthy conifers and native forests with Sentinel-2. Conversely, the accuracy obtained with WorldView-3 was 85%. The higher accuracy of Sentinel-2 was attributed to the availability of spectral bands in the shortwave infrared and red-edge spectrum and to the higher purity of pixels in the training data. Classification with Sentinel-2 imagery is therefore a promising tool to detect and monitor advanced infestations of exotic conifers.

---