Method for national mapping spatial extent of southern beech forest using temporal spectral signatures

Abstract

Characterisation of native forests is essential for sustainable forest management and for maintenance of ecological and socio-economical functions. In New Zealand, knowledge of forest composition and extent informs predator control measures to protect native bird life, particularly in forests dominated by Southern beeches (Nothofagaceae). As high-resolution (> 1:50,000) maps of beech cover do not exist at national scale, we present a method to identify and map beech cover that combines multi-temporal spectral signatures from ESA’s Sentinel-2 satellite with forest plot survey data. A temporal stack of satellite imagery from 2016 to 2019 is used to derive annual metrics (mean and standard deviation) of vegetation indices which are used as input to a pixel-wise classification. A random forest classification, discriminating between beech/non-beech areas (with a beech relative cover threshold of 25%), and trained using 880 forest plots from the Land Use and Carbon Analysis System (LUCAS) natural forest network, achieved an accuracy of 87.7% (± 2.2%). This spectral classification captures both large- and local-scale spatial patterns of beech cover, which is confirmed by field visits and multi-source species occurrence information.

Keywords

• Forest type mapping
• Machine learning
• Sentinel-2
• Spectral indices
• Forest survey
• Beech forest