UAV-mounted hyperspectral mapping of intertidal macroalgae

Highlights

•

We use a drone-mounted hyperspectral sensor to map intertidal macroalgal species.

•

Classification accuracies of two sources of spectral endmember data are compared.

•

The target species of interest, Ascophyllum nodosum, can be accurately mapped.

•

Other common intertidal species can be spectrally distinguished between.

•

High-resolution RGB imagery is an effective tool for feature identification.

Abstract

Intertidal macroalgal communities mark the boundary of the marine realm and are faced with many direct and indirect anthropogenic pressures. The effective and sustainable management of these resources must be underpinned by accurate, efficient and cost-effective environmental data collection. Traditional field survey methods, whilst accurate, are time-consuming and limited in the area that can be covered. Remote sensing permits large areas to be rapidly surveyed but the effectiveness of satellites and aircraft for mapping fine-scale intertidal macroalgal mapping is limited by their coarse spatial resolution and restricted operational flexibility. The rapid development of unoccupied aerial vehicle (UAV) and sensor technology can address these issues and provide a potential alternative to established remote sensing platforms. Here, a detailed methodology is presented for the assessment of the commercially and ecologically important intertidal brown macroalga Ascophyllum nodosum using a multirotor UAV and pushbroom hyperspectral sensor. Two different classifiers, Maximum Likelihood Classifier (MLC) and Spectral Angle Mapper (SAM), were compared along with two different sources of spectral profiles, one collected in-situ with a spectral radiometer and the other derived from hyperspectral imagery. Of the classifiers compared, both trained using image-derived spectra, MLC more accurately classified A. nodosum, and other common intertidal species and substratum (Overall Accuracy (OA) 94.7%) than SAM (OA 81.1%). In addition, SAM, trained using in-situ spectra, was the least accurate of the three classifier workflows used (OA 71.4%). The low accuracy of the spectral radiometer approach was likely due to high levels of noise present in the hyperspectral data, a result of the relative instability of the UAV platform causing vibration. The accurate mapping of non-target species also highlights the applicability of this methodology for a broader range of intertidal macroalgal species and communities. This research clearly demonstrates the potential of UAV-mounted hyperspectral remote sensing for mapping the spatially and spectral complex macroalgal habitats found within the intertidal zone.