Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Afrique du Sud → 2014 → Assessing multi-temporal remote sensing imagery for discriminating savannah tree species.

University of KwaZulu-Natal (2014)

Assessing multi-temporal remote sensing imagery for discriminating savannah tree species.

Madonsela, Sabelo.

Titre : Assessing multi-temporal remote sensing imagery for discriminating savannah tree species.

Auteur : Madonsela, Sabelo.

Université de soutenance : University of KwaZulu-Natal

Grade : Master of Science (MS) 2014

The advent of new multispectral sensors such as Worldview-2 with very high spatial resolution (VHR) has presented new opportunities for mapping vegetation at species-level. However the use of VHR data for tree species mapping is often confronted with issues of within-canopy spectral variability. The prevailing intraspecies variability in southern African savannah limits our ability to accurately map the distribution of tree species. These challenges necessitate the development of new methods for tree species mapping. This study investigated i) the utility of object-based image analysis (OBIA) for tree species mapping in the savannah environment using Worldview-2 image, ii) the spectral capability of WV-2 for species mapping and iii) the ability of multi-temporal data to enhance spectral separability between tree species in southern African savannah. Using Random Forest (RF), the study could not establish any statistically significant difference between OBIA and pixel-based approach towards savannah tree species classification (zobt < zcrit). However OBIA successfully improved classification accuracy of Sclerocharya birrea and Acacia nigrescens which makes it an appropriate alternative for classifying big trees in the savannah environment using WV-2 image. Moreover, the spectral configuration of WV-2 with the inclusion of yellow and red-edge bands enhanced the discriminatory power of WV-2 sensor. The WV-2 image achieved higher classification accuracy (74.5% with object-based and 76.4% with pixel-based) than simulated IKONOS image (58.6% with object-based and 67.9% with pixel-based). The difference was statistically significant (zobt > zcrit). The use of multi-temporal data enhanced spectral variability between species and achieved the highest classification accuracy (80.4%) than March and April dates (72.9% and 76.4%, respectively). Multi-temporal data mitigated the spectral confusion between Sclerocharya birrea and Dichrostachys cinerea and achieved producer’s and user’s accuracy of above 60% for these tree species. The results highlight the opportunities available to biodiversity managers due to advances in remote sensing technology. The ability to accurately map tree species is the key element in the management of savannah biodiversity.


Version intégrale (2,38 Mb)

Page publiée le 11 octobre 2017