Mapping the remnant KwaZulu-Natal sandstone sourveld grass patches in the Ethekwini Municipality using a high resolution multispectral sensor.
Abstract
The indigenous KwaZulu-Natal sandstone sourveld (KZN SS) grassland is highly endemic
and species-rich, yet critically endangered and poorly conserved. Ecological threats to this
grassland are further exacerbated by the occurrence of woody plant encroachment, a form of
degradation that has severe negative environmental and economic consequences. In this regard,
understanding the distribution of the KZN SS fragments is critical for implementing conservation
and management strategies. Advances in remote sensing technologies allow for accurate and
precise mapping, hence the aim of this study is to identify the remnants of the KZN SS within
the eThekwini Municipality using high resolution multispectral RapidEye data.
The first part of this research seeks to assess the capability of RapidEye satellite imagery in
mapping the indigenous KZN SS using support vector machines (SVM) and maximum
likelihood (ML) classifiers. Although both techniques were successful in mapping the KZN SS,
results show that ML was slightly outperformed by SVM, which yielded an overall accuracy of
74.4%. In addition, SVM were more accurate in distinguishing the KZN SS class with a score of
74.4%, compared to that of ML, namely 72.1%. The study underscores the importance of high
resolution RapidEye data in detecting and mapping the remaining fragments of the KZN SS
within the eThekwini Municipality.
The second part of this research zoomed into discriminating between indigenous and alien
woody plant encroachment within the KZN SS. The random forest (RF) algorithm was applied to
the image and successfully mapped the two types of vegetation with an overall accuracy of 86%.
In addition, an overall accuracy of 74% was obtained in estimating the five dominant tree species
within the two classes. The results obtained highlight the potential of new generation RapidEye
satellite data in combination with new advanced machine learning techniques in predicting the
distribution of woody cover in a grassland ecosystem.
Overall, this study successfully mapped the KZN SS patches, as well as bush
encroachment patches. The strategic bands in the new generation RapidEye image were critical
in species mapping.