Automated mucilage extraction index (AMEI): a novel spectral water index for identifying marine mucilage formations from Sentinel-2 imagery

Abstract

Marine mucilage that threatens marine habitats is one of the natural disasters, mainly resulting from global warming and marine pollution. Monitoring sea surface mucilage formations and mapping their spatial distributions provide valuable information to the local authorities and decision-makers in developing prevention and rehabilitation strategies. This study proposes a new spectral index called Automated Mucilage Extraction Index (AMEI) that allows effective and accurate detection of surface mucilage aggregates using Sentinel-2 satellite imagery. The index uses four bands of Sentinel-2 Level-2A imagery (Bands 3, 4, 8, and 12) covering visible, near-infrared, and shortwave infrared regions. The index was formulated considering the image acquired on 19 May 2021, when mucilage formations were most intensively observed in the Sea of Marmara. The performance of the developed index was then evaluated using the images acquired on 14 and 24 May and 13 June 2021. The Jenks Natural Breaks (JNB) algorithm was applied to estimate the threshold values for separating mucilage formations from the water surface background and classify index maps into two classes: 'mucilage' and 'others'. To statistically analyse the effectiveness of the indices in distinguishing water background and mucilage formations, the proposed index was compared with 21 widely used water indices by applying Bhattacharyya Distance, Jeffries-Matusita Distance, and M-Statistic measures. Results confirm the robustness of the proposed spectral index, offering superior separation performance (above 1.5 in terms of M-Statistic) compared to other water indices on both cloud-free images and images with cumulus clouds. Visual interpretations also verified that boundaries of the mucilage formations in the cloudless and thin cloudy images were accurately identified by the proposed index, and different types of mucilage (i.e. yellow and white) can be identified when an appropriate histogram thresholding is applied.