On Friday, 20 November, Temitope Olaoluwa Omoniyi defended his doctoral thesis at the Estonian University of Life Sciences, entitled “Spatiotemporal modeling and assessment of forest growing stock volume using national forest inventory and remotely sensed data”.
His research explored how satellite data and machine learning can improve the accuracy of estimates of forest growing stock volume and its changes over time.
The doctoral thesis was supervised by Associate Professor Allan Sims from the Estonian University of Life Sciences. The opponent was Dr Alex Appiah Mensah from the Swedish University of Agricultural Sciences.
Growing stock volume, which refers to the volume of wood in the stems of living trees, is an important indicator for assessing forest conditions and timber resources. In Estonia, this information is collected through the National Forest Inventory (NFI), but the estimates are based on measurements conducted over several years. As a result, some of the data may already be outdated by the time the results are published. Conducting new field surveys is costly and time-consuming, creating a need for faster and more accurate ways to update information about forests.
Satellite data and machine learning support forest growing stock estimation
In his doctoral research, Omoniyi investigated how National Forest Inventory data could be combined with remote sensing and machine learning methods. He used Sentinel-2 satellite imagery, Sentinel-1 radar data, airborne laser scanning (LiDAR) data, and various environmental variables.
The study compared three machine learning algorithms: Random Forest, Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost). The results showed that the Random Forest algorithm provided the most consistently accurate estimates, while the other methods performed well only under certain conditions.
Omoniyi also compared the use of satellite data from a single point in time with data from two different points in time for predicting forest growing stock volume. The results showed that both approaches produced similar estimates.
New methods help reduce uncertainty in forest inventory estimates
Omoniyi also investigated methods for estimating changes in forest growing stock volume. He compared different statistical approaches and found that estimates supported by remote sensing data reduced uncertainty compared with estimates based solely on forest inventory measurements.
The study also showed that remote sensing data with spatial resolutions of 10 and 20 metres produced similar results when estimating changes in growing stock volume. However, models that performed most accurately at the individual sample plot level did not necessarily provide the most reliable estimates for the entire forest area.
Omoniyi's doctoral thesis demonstrates that combining remote sensing data, machine learning, and statistical methods can improve the accuracy of forest growing stock volume estimates and reduce the uncertainty associated with them.
The doctoral thesis is available in the digital archive of the Estonian University of Life Sciences EMÜ DSpacelink opens in new page.
Editor:
Research Communication Specialist
Rector's Area of Responsibility
Department of Marketing and Communication
53585680
53585680More information:
Junior Research Fellow of Forestry
Institute of Forestry and Engineering
Chair of Forest Management Planning and Wood Processing Technologies