Assessing the effectiveness of RapidEye multispectral imagery for vegetation mapping in Madeira Island (Portugal)

Abstract

Madeira Island is a biodiversity hotspot due to its high number of endemic/native plant species. In this work we developed and assessed a methodological framework to produce a RapidEye-based vegetation map. Reasonable accuracies were achieved for a 26 categories classification scheme in two different seasons. We tested pixel and object based approaches and the inclusion of a vegetation index band on top of the pre-processed RapidEye bands stack. Object based generally showed to outperform pixel based classification approaches except for linear or highly scattered classes. The addition of a vegetation index to the workflow increased the separability of the Jeffrey-Matusita least separable class pairs, but not necessarily the overall accuracy. The Pontius accuracy assessment highlighted class specific accuracy tradeoffs related to different combinations of the inputs and methods. The approach to be used, in conclusion, should be carefully considered on the basis of the desired result.