A Methodology for National Scale Coastal Landcover Mapping in New Zealand

Abstract

Satellite earth observation data has become fundamental in efforts to map coastal change at large geographic scales. Research has generally focussed on extracting the instantaneous waterline position from time-series of satellite images to interpret long-term trends. The use of this proxy can, however, be uncertain because the waterline is sensitive to marine conditions and beach gradient. In addition, the technique disregards potentially useful data stored in surrounding pixels. In this paper, we describe a pixel-based technique to analyse coastal change. A hybrid rule-based and machine learning methodology was developed using a combination of Sentinel multispectral and Synthetic Aperture Radar composite imagery. The approach was then used to provide the first national-scale pixel-based landcover classification for the open coast of New Zealand. Nine landcover types were identified including vegetation, rock, and sedimentary classes that are common on beaches (dark sand, light sand, and gravel). Accuracy was assessed at national scale (overall accuracy: 86%) and was greater than 90% when normalised for class area. Using a combination of optical and Synthetic Aperture Radar data improved overall accuracy by 14% and enhanced the separation of coastal sedimentary classes. Comparison against a previous classification approach of sandy coasts indicated improvements of 30% in accuracy. The outputs and code are freely available and open-source providing a new framework for per-pixel coastal landcover mapping for all regions where public earth observation data is available.

Keywords

• coastal change
• multispectral
• synthetic aperture radar
• classification
• mapping
• google earth engine