Mapping ephemeral wetlands: manual digitisation and logistic regression modelling in Nelson Mandela Bay Municipality, South Africa

• Published in: Wetlands ecology and management 2017-06, Vol.25 (3), p.313-330
• Main Authors: Melly, Brigitte L., Schael, Denise M., Rivers-Moore, Nick, Gama, Phumelele T.
• Format: Article
• Language: English
• Subjects: Abundance; Accuracy; Aerial photographs; Aerial photography; Area; Arid regions; Biomedical and Life Sciences; Capacity; Coastal environments; Conservation; Conservation Biology/Ecology; Digitization; Distribution; Encapsulation; Environmental Law/Policy/Ecojustice; Environmental risk; Flats (landforms); Flood management; Floodplains; Freshwater & Marine Ecology; Geographic information systems; Geologic depressions; Geomorphology; High resolution; Hydrology/Water Resources; Identification; Indication; Information systems; Land cover; Life Sciences; Logistics; Mapping; Marine & Freshwater Sciences; Modelling; Municipalities; Original Paper; Polygons; Prediction models; Probability theory; Rain; Rainfall; Regression analysis; Seepages; Variables; Vegetation; Water conservation; Water Quality/Water Pollution; Wetland management; Wetlands
• ISSN: 0923-4861; 1572-9834
• DOI: 10.1007/s11273-016-9518-7

Until recently, little research has been conducted on the distribution and structure of ephemeral systems in semi-arid areas. This information is critical for appropriate wetland management and conservation. The Nelson Mandela Bay Municipality is a semi-arid area along the south-eastern coastline of the Eastern Cape Province of South Africa. The Municipality encapsulates a wide range of geological and geomorphological features as well as vegetation types within an area of some 1950 km 2 , providing an ideal area for such research. The distribution and abundance of wetlands were defined, and a logistic regression (LR) model was used to establish whether this modelling technique is viable in semi-arid areas with highly variable rainfall patterns. Wetlands were delineated manually using geographical information systems, high-resolution aerial photographs and environmental data. More than 1700 wetland polygons were identified, with 80% of the systems being categorised as depressions, seeps and wetland flats. Unchannelled (8%) and channelled (7%) valley bottom wetlands and floodplain wetlands (5%) were also identified. The wetland database was then used to create a wetland occurrence probability model. There were 19 environmental variables used to develop the LR model, with eight variables used in the final model output. The predictive capacity of the model was good, with an area under curve value of 0.68 and an overall accuracy of 66%. This indicates that probabilistic wetland models are useful in highly variable environments with high numbers of small (