Eucalyptus plantations are an important contributor to the South African economy and are geographically concentrated in the productive Zululand region. However, this area is becoming increasingly vulnerable to various forest disturbances such as insect attacks and drought. Information of the location, extent and duration of insect infestation is therefore crucial for sustainable timber production and forest management planning, and remote sensing approaches have been successful in the retrieval and analysis of such information along with field surveys. Sappi Forests frequently conduct field surveys to establish the nature of disturbances to improve the performance of trees, as these may be difficult to distinguish from other disturbances such as droughts. Great effort has been made to remotely detect the impact of G. scutellatus using single-date high-resolution commercial satellites, whose application is restricted by high costs. Here, we compare the performance of spectral indices for detecting the impact of G. scutellatus infestation in KwaMbonambi plantations using publicly available Landsat data and also apply anomaly detection framework to further test their performance for detecting the impact of insect damage on trees and other disturbances that occurred on the same compartment over time. Our results showed high performance of short-wave infrared (SWIR) band for detecting the impact of insect damage, followed by the normalized difference infrared index (NDII) and normalized difference vegetation index NDVI, when anomaly detection was applied. Without anomaly detection, the spectral indicators showed changes even to drought impact in 2015, which did not when anomaly detection was used. We believe that indices containing the SWIR would perform better for detecting anomalous changes, and further research is required to explore this. Anomaly detection only showed the impact of insect damage and clearcutting as anomalous changes. Overall, we demonstrated the benefit of identifying vegetation anomalies caused by the impact of insect damage and other disturbances using freely available Landsat time-series data, particularly in data-poor regions that have restricted computational resources.
Read full abstract