Object-based classification of semi-arid vegetation to support mine rehabilitation and monitoring

Bao, Nisha, Lechner, Alex M., Johansen, Kasper and Ye, Baoying (2014) Object-based classification of semi-arid vegetation to support mine rehabilitation and monitoring. Journal of Applied Remote Sensing, 8 (1). 083564. ISSN 1931-3195

Full content URL: https://doi.org/10.1117/1.JRS.8.083564

Full text not available from this repository.

Item Type:Article
Item Status:Live Archive


Mining activities result in significantly modified landscapes that require rehabilitation to mitigate the negative environmental impacts and restore ecological function. The aim of this study was to develop a remote sensing method suitable for monitoring the vegetation cover at mine rehabilitation sites. We used object-based image analysis (OBIA) methods and high-spatial resolution SPOT-5 imagery to identify discrete land-cover patterns that occur at fine spatial scales. These patterns relate to spatial processes that are important drivers of successful restoration of mine sites. SPOT-5 imagery of the Kidston Gold mine tailing dam in semi-arid tropical north Queensland was acquired in July 2005, comprising four 10-m spectral bands and a 2.5-m panchromatic (PAN) band. The classification scheme used in this study was adapted to the spatial scale of SPOT-5 imagery from mine closure criteria cover requirements, according to a mine rehabilitation plan. Four land-cover classes were identified: tree cover, dense grass, sparse grass, and bare ground. First, textural layers (contrast, dissimilarity, and homogeneity) were derived for each vegetation class except for bare ground from the PAN and multispectral bands. Of all textural layer combinations, homogeneity and contrast in the PAN band were identified using a Z -test as the most useful for differentiating between multiple land-cover classes. Next, an optimal segmentation scale parameter of 15 was identified using an analysis of spatial autocorrelation. Finally, the SPOT-5 image bands, derived textural layers, and normalized difference vegetation index (NDVI) were used in an OBIA fuzzy membership classification approach to map vegetation land-cover classes. The classification results were assessed with the traditional error matrix approach and the object-based accuracy assessment method. The overall classification accuracy using the error matrix was 92.5% and 81% using the object-based method. The relatively high-classification accuracy demonstrates the potential of SPOT-5 imagery for monitoring mine rehabilitation. The complete spatial coverage associated with remote sensing data at fine spatial scales has the potential to complement field-based approaches commonly used in rehabilitation monitoring. Furthermore, SPOT-5 data along with OBIA can characterize vegetation spatial patterns at spatial scales appropriate for monitoring rehabilitated landscapes, providing an important tool for landscape function analysis.

Subjects:F Physical Sciences > F810 Environmental Geography
Divisions:College of Science > School of Geography
ID Code:42035
Deposited On:14 Oct 2020 15:06

Repository Staff Only: item control page