Mapping private gardens in urban areas using object-oriented techniques and very high-resolution satellite imagery

Abstract

Gardens remain the least studied and least understood habitat in urban areas. With the recent exception of the URGENT funded urban domestic gardens project in the UK, there is a notable lack of research on the ecological character and contribution of gardens to the wider urban biodiversity. This is despite the fact that gardens usually comprise the largest vegetated component of the urban greenspace resource. In part this omission has been due to the difficulties inherent in obtaining ecological data on gardens and the lack of a methodology for classifying and analysing garden data. This paper presents data from a study undertaken in the city of Dunedin, New Zealand. The study developed a methodology using object-oriented classification techniques and very high-resolution multispectral Ikonos imagery to automatically map the extent, distribution and density of private gardens in the city. The focus was on the vegetated garden area which was calculated as comprising 46% of the residential area or 36% of the total urban area. Rigorous accuracy assessments were undertaken. When using the automated classification technique, a total of 90.7% of the private gardens were correctly identified. Discrimination of garden types (e.g. trees or grass dominated) was encouraging, but still requires improvement. Our results indicate the great potential that the methodology has in providing a quick method for obtaining good quality ecological data on garden habitats in urban areas.