- Title
- Applying high-resolution remote sensing to quantify baboon damage at a sub-compartment level in pine stands in the Mpumalanga escarpment region of South Africa
- Creator
- Ferreira, Regardt
- Subject
- Environmental sciences -- Remote sensing
- Subject
- Geographic information systems Remote sensing
- Date Issued
- 2020
- Date
- 2020
- Type
- Thesis
- Type
- Masters
- Type
- MSc
- Identifier
- http://hdl.handle.net/10948/49082
- Identifier
- vital:41599
- Description
- Managing risk in intensively managed monoculture plantation forests is an essential task to ensure sustainable yield and a continuous flow of forest products. However, since risks can be either biotic or abiotic, not all of them have a predictable pattern of spread, which can cause severe losses if management does not have the chance to implement mitigation action. Monitoring the change in forest health is vital as this provides the opportunity for preventative management and quantifies the amount of damage that management has to deal with. To provide this window of opportunity for appropriate action, constant monitoring is required. Until recently, forest health was measured through field surveys which provided adequate data. This procedure, however, is time consuming. Remote sensing has become very popular as a monitoring tool, due to its ability to provide assessment data in a fraction of the time. In this study, baboon damage in plantations along the Mpumalanga escarpment area of South Africa was monitored using remote sensing methods. While there are many methods of forest health monitoring using remote sensing, some approaches are less suitable as they either monitor damage caused at a plantation level, use lower spatial resolution (>10m) datasets or map damage using one available time period. The purpose of this study was first to establish the impact of baboon damage through time, using Sentinel-2 satellite imagery with all vegetation indices available, and the Extreme Gradient Boosting (XGboost) algorithm. The second part focused on analysing the damage at a tree level using PlanetScope imagery using a deep Learning approach. Overall, the study found that the use of Sentinel-2 data and PlanetScope data could accurately distinguish between the varying severity of baboon damage, achieving an accuracy of 95% and 82%. The processing time of the deep learning Artificial Neural Network (ANN) was greatly affected by the number of hidden layers and neurons used. Implementation of techniques used in this study has the potential to improve the accuracy of forest health monitoring in compartment forestry in South Africa.
- Format
- ix, 44 leaves
- Format
- Publisher
- Nelson Mandela University
- Publisher
- Faculty of Science
- Language
- English
- Rights
- Nelson Mandela University
- Hits: 1076
- Visitors: 1156
- Downloads: 134
Collections
NMU School of Environmental Sciences
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | SOURCE1 | Ferreira, R.pdf | 1 MB | Adobe Acrobat PDF | View Details Download |