Upscaling a native woodland survey using Supervised Classification

Abstract

Forests have historically been at the forefront of scientific research to understand complex ecological processes and to help mitigate the effects of climate change on ecosystems. There has been an increase in the availability of accessible data, in addition to the development of data-analysis methods like machine learning techniques which provide advantages over traditional statistical methods for assessing ecological datasets. Scotland’s forestry and woodland sector depends upon meeting goals within the Scotland’s Forest Strategy 2019-2029, however missing data within the ecological dataset can cause misinformation and accuracy in the Strategy. Using freely available Sentinel-2 data, and forest inventory data in three diverse parts of Scotland, supervised classification is used to predict native and non-native trees. Findings suggest that supervised classification can help upscale missing forest inventory plots most accurately in the lowlands (86% accuracy) but less in urban areas there is less potential for this (60% accuracy). Higher resolution datasets may prove to be more beneficial in increasing the accuracy, with training datasets that are not imbalanced. Future work could combine data fusion of active remote sensing e.g., LiDAR, radar to help distinguish and upscale native trees that have missing data through supervised classification.