Tools
Tools
Gyamfi-Ampadu, E., Gebreslasie, M., & Mendoza Ponce, A. 
ORCID: https://orcid.org/0000-0002-9069-0652
(2020).
Mapping natural forest cover using satellite imagery of Nkandla forest reserve, KwaZulu-Natal, South Africa.
Remote Sensing Applications: Society and Environment 18 p. 100302. 10.1016/j.rsase.2020.100302.
Abstract
Natural forest ecosystems are vital environmental resources that provide multiple benefits to society, making it imperative for them to be monitored and mapped for practical management purposes. Satellite remote sensing technology is a new source of data and information for forest management and conservation. This study, therefore, applied Support Vector Machine (SVM) and Random Forest (RF) algorithms to Landsat 8 image for mapping a natural forest in South Africa. The objectives were to classify the forest into specific thematic cover classes that indicate its condition, compare the classification performance of the two algorithms based on their default parameters, and determine the most important variables that contributed to the mapping accuracy. The closed canopy forest was determined as the dominant thematic class, followed in descending order by the open canopy forest, grassland, and bare sites. Both algorithms obtained high classification accuracies of above 95%, though the SVM was slightly superior to the RF. The McNemer test indicated that the difference in performance between the two algorithms was statistically insignificant. The most important variables that contributed to the accuracy were the red, blue, green, Near Infrared and Short-Wave Infrared bands, which is attributed to their sensitivity to vegetation. The information provided through the study can be utilized for the planning, management and prioritization initiatives aimed at the protection and conservation of the forest reserve and similar forest ecosystems. The mapping approach could be used for other natural forest ecosystems to ascertain the spatial coverage of the specific thematic cover for conservation purposes. The SVM is recommended for forest ecosystem mapping as it optimally utilized the capabilities of the spectral bands that reflect their actual importance in the mapping of each cover class. The bands identified as important variables can be incorporated as part of input variables when using Landsat 8 satellite imagery for natural forest mapping.
| Item Type: | Article |
|---|---|
| Uncontrolled Keywords: | Forest ecosystem; Remote sensing; Support vector machine; Random forest; Mapping; Conservation |
| Research Programs: | Ecosystems Services and Management (ESM) |
| Depositing User: | Luke Kirwan |
| Date Deposited: | 19 Mar 2020 16:01 |
| Last Modified: | 27 Aug 2021 17:32 |
| URI: | https://pure.iiasa.ac.at/16362 |
Actions (login required)
 |
View Item |