Remote sensing tools for plant invasions assessment in native tropical forests of a volcanic island

Abstract

Surveying and monitoring biological invasions by plants can greatly benefit from the application of remote sensing techniques. In tropical insular enviroments however, multiple challenges makes the task difficult. Invasive species can be diverse and many, so that evaluating an overall invasion level may help to distinguish uninvaded native forests from forests invaded at various levels. Also, accessibility can limit the availability of field-based data documenting the ground truth of plant invasions. Here we present a methodological study with the doublefold objective of first estimating an overall index of invasion and level of native forest canopies and second detect two particular exotic (Cinamomum camphora, Lauraceae and Cryptomeria japonica, Taxodiaceae) planted in monospecific stands. The index of overall invasion level comprises four classes derived elsewhere from field observations and expert knowledge. A thematic map provided polygons representing monospecif plantations. Textural features of sub-regional regions of interest and polygons were derived using Haralick coefficients on co-occurrence matrices, and morphological profiles. Classification algorithms were applied to produce invasion classes (k-means) or detect target species (Support Vector Machine, SVM) based on pixels features.