NZJFor Home Search Join Author instructions NZIF website NZJFor Home NZJFor


New Zealand Journal of Forestry (2021) 65(4): 19–25
©New Zealand Institute of Forestry

Feature article
Identifying post-harvest soil disturbance using satellite imagery

Jim Walsh *,1 and Rien Visser 2

1 Final year Forest Engineering Honours student, 2020. Corresponding author:
2 Professor Rien Visser, Director, Forest Engineering programme, School of Forestry, University of Canterbury, Christchurch.
*Corresponding author.

Abstract: Minimising the overall level of soil disturbance during forest operations is a cornerstone of sustainable forest operations. Soil disturbance assessments are generally carried out using plots or line transects that are both labour-intensive and time-consuming, and hence currently rarely done except for research purposes. The increasing availability of higher resolution satellite imagery and improved image classification tools means there may be an opportunity to efficiently estimate soil disturbance as part of a performance assessment tool. Seven harvest sites in the South Island were used to assess the accuracy of using satellite images for measuring soil disturbance. Satellite images obtained through PlanetScope were collected for each site (3 x 3m resolution). The images were processed in ArcMap using two supervised classification tools: Maximum Likelihood Equation (MLE) and Support Machine Vector (SVM). Ground-truthing was carried out creating two lines of 15 points at 10 m intervals where land cover type was determined by visual inspection (e.g. bare soil, slash or vegetation). The accuracy assessment compared classification methods and techniques. The supervised classification techniques were able to easily identify large disturbances (such as roads and skid sites), but struggled to pick up smaller disturbances due to the effects of ‘mixed’ pixels, where the pixels contain more than a single land cover class. The average overall agreement for MLE and SVM with the ground-truth measures was 64% and 65%, respectively. For best case scenarios, average overall agreement for MLE and SVM was 68% and 72%, respectively, confirming that the SVM classifier outperforms the MLE. This project highlights that it is feasible to achieve realistic measures of soil disturbance from satellite images. Higher resolution imagery from daily satellite images, or drones and fixed-wing aircraft, presents an opportunity to increase the accuracy of the classifications.
(no keywords)

Issues > 65(4) > Abstract

Download article as 724 KB PDF file

As an issue ≤ 3 years old, access to this article is restricted to subscribers. (All articles from issues > 3 years old are free.)

(You can read PDF files with the free Adobe Acrobat Reader)