eCite Digital Repository

Modelling LiDAR derived tree canopy height from Landsat TM, ETM plus and OLI satellite imagery - A machine learning approach

Citation

Staben, GW and Lucieer, A and Scarth, P, Modelling LiDAR derived tree canopy height from Landsat TM, ETM plus and OLI satellite imagery - A machine learning approach, International Journal of Applied Earth Observation and Geoinformation, 73 pp. 666-681. ISSN 0303-2434 (2018) [Refereed Article]

Copyright Statement

2018 Elsevier B.V. All rights reserved.

DOI: doi:10.1016/j.jag.2018.08.013

Abstract

Understanding ecological changes in native vegetation communities often requires information over long time periods (multiple decades). Tropical cyclones can have a major impact on woody vegetation structure across northern Australia, however understanding the impacts on woody vegetation structure is limited. Woody vegetation structural attributes such as height are used in ecological studies to identify long term changes and trends. LiDAR has been used to measure woody vegetation structure, however LiDAR datasets cover relatively small areas and historical coverage is restricted, limiting the use of this technology for monitoring long-term change. The Landsat archive spans multiple decades and is suitable for regional/continental assessment. Advances in predictive modelling using machine learning algorithms have enabled complex relationships between dependent and independent variables to be identified. The aim of this study is to develop a predictive model to estimate woody vegetation height from Landsat imagery to assist in understanding change through space and time. A LiDAR canopy height model was produced covering a range of vegetation communities in northern Australia (Darwin region) for use as the dependent variable. A random forest regression model was developed to predict mean LiDAR canopy height (30 m spatial resolution) from Landsat-5 Thematic Mapper (TM). Validation of the random forest model was undertaken on independent data (n = 30,500) resulting in an overall R2 = 0.53, RMSE of 2.8 m. Assessment of the RMSE within four broad vegetation communities ranged from 2.5 to 3.7 m with the two dominant communities in the study area Mangrove forests and Eucalyptus communities recording an RMSE value of 2.9 m and 2.5 m respectively. The model was also applied to Landsat-7 Enhanced Thematic Mapper Plus (ETM+) resulting in an R2 of 0.49, RMSE of 2.8 m. The model was then applied to all cloud free Landsat-5 TM, Landsat-7 ETM+ and Landsat-8 Operational Land Imager (OLI) imagery (106/69 path/row) available between the months April, May and June for 1987 to 2016 to produce annual estimates (29 years) of canopy height. A number of time traces were produced to illustrate tree canopy height through time in the Darwin region which was severely impacted by cyclone (hurricane) Tracy on the 25th December 1974.

Item Details

Item Type:Refereed Article
Keywords:random forest, LiDAR, canopy height model, vegetation structure, Landsat
Research Division:Engineering
Research Group:Geomatic Engineering
Research Field:Photogrammetry and Remote Sensing
Objective Division:Environment
Objective Group:Ecosystem Assessment and Management
Objective Field:Ecosystem Assessment and Management of Forest and Woodlands Environments
UTAS Author:Staben, GW (Mr Grant Staben)
UTAS Author:Lucieer, A (Professor Arko Lucieer)
ID Code:131171
Year Published:2018
Web of Science® Times Cited:1
Deposited By:Geography and Spatial Science
Deposited On:2019-03-06
Last Modified:2019-05-13
Downloads:0

Repository Staff Only: item control page