Naturalness Indicators of Forests in Southern Sweden derived from the Canopy Height Model

TL;DR

This paper develops a methodology to extract interpretable features from canopy height models and applies machine learning to assess forest naturalness in Southern Sweden with high accuracy, aiding ecological and conservation efforts.

Contribution

It introduces a novel feature extraction approach from canopy height models and demonstrates effective machine learning predictions of forest naturalness.

Findings

Prediction accuracy ranges from 89% to 95%.

Features are human-interpretable and useful for stakeholders.

Method is applicable to ecological monitoring.

Abstract

Forest canopies embody a dynamic set of ecological factors, acting as a pivotal interface between the Earth and its atmosphere. They are not only the result of an ecosystem's ability to maintain its inherent ecological processes, structures, and functions but also a reflection of human disturbance. This study introduces a methodology for extracting a comprehensive and human-interpretable set of features from the Canopy Height Model (CHM), which are then analyzed to identify reliable indicators for the degree of naturalness of forests in Southern Sweden. Utilizing these features, machine learning models - specifically, the perceptron, logistic regression, and decision trees - are applied to predict forest naturalness with an accuracy spanning from 89% to 95%, depending on the area of the region of interest. The predictions of the proposed method are easy to interpret, something that various stakeholders may find valuable.