Change-mapping of estuarine intertidal seagrass (Zostera muelleri) using multispectral imagery flown by remotely piloted aircraft (RPA) at Wharekawa Harbour, New Zealand. (5th November 2020)

• Record Type: Journal Article
• Authors: Martin, Ross; Ellis, Joanne; Brabyn, Lars; Campbell, Marnie
• Is Part Of: Estuarine, coastal and shelf science. Volume 246(2020)
• Publication Date: 2020-11-05
• Subjects: Seagrass -- Mapping -- Multispectral -- Condition -- Change -- Remotely piloted aircraft -- Drone; Estuarine oceanography -- Periodicals; Coasts -- Periodicals; Estuarine biology -- Periodicals; Seashore biology -- Periodicals; Coasts; Estuarine biology; Estuarine oceanography; Seashore biology; Periodicals; 551.461805
• DOI: 10.1016/j.ecss.2020.107046
• Languages: English
• ISSNs: 0272-7714

Abstract

Abstract: This research assesses a survey method for measuring the change in density of intertidal estuarine seagrass ( Zostera muelleri ) at Wharekawa Harbour, New Zealand, using an autonomous remotely piloted aircraft fitted with a narrowband multispectral camera. Image classification was modelled using the random forest classifier trained with ground observation data sourced from 63 photographed quadrat stations upon three parallel transect lines. Seagrass coverage in the georeferenced and rectified ground photography was estimated by visual interpretation on a three-tier (low, medium and high) density scale, then exact leaf area additionally calculated from digitised seagrass leaf coverage visible in rectified ground photography. Three replicate aerial surveys across four months were conducted to compare predicted change in seagrass density with actual measured change, during austral summer growth and autumn decline. Classification of the resulting image mosaic (2.5 cm pixel size) achieved up to 90–93% overall accuracy across multiple surveys when attributing density class, 93–96% accuracy for prediction of seagrass presence, and 81–91% in terms of detection of seagrass on the ground. Change-maps allow regions of growth and decline to be visualised. Correlation (r) between actual and predicted change for 48 independent test grid squares was 0.89 and 0.61 for the summer and autumn change periods respectively. Rapid visual interpretation of classification end-member classes yielded change measurement equivalent to that of accurately measured seagrass leaf area. The research demonstrates that RPA survey using a multispectral camera is viable for monitoring change in seagrass condition. Graphical abstract: Image 1 Highlights: Seagrass density change-mapping by remotely piloted aircraft (RPA). High accuracy in revisit positioning by RPA flight program. Map classified at up to 90–93% overall accuracy in replicated surveys. Rapid visual interpretation of ground reference classes at high mapping accuracy. Moderate-high correlation between predicted and actual change for summer and autumn.