Forest species classification using Object Oriented Image Analysis – does CIR aerial imagery and Object Oriented Methods offer new possibilities?

Abstract

National Forest Inventories (NFI) in Britain such as the Forestry Commission’s National Inventory of Woodlands and Trees (NIWT) and the Native Woodland Survey of Scotland (NWSS) provide a synoptic view of the current extent, condition and location of forest resources. The NFI’s not only need to be accurate and precise but also up-to-date. Current methods of producing NFIs in Britain rely on manual photo-interpretation analysis of digital aerial photographs combined with field-based surveys. Development of a cost effective and reliable method to identify tree species as a semi- automated process would be of major benefit to forest management.

This research explores object oriented image analysis methods for Forest Inventory classification on three levels: 1.Woodland/Non woodland, 2. Interpreted Forest Types (IFT) and 3. Species using 40cm resolution Colour InfraRed Imagery and Definiens Developer Software.

A 17km2 study area within Clocaeanog Forest, Wales, was chosen with the intention of developing a robust classification ruleset. The area was flown throughout 2006, resulting in variations across the study area of sun angle, cloud cover, vegetation, intensity etc. However, a hierarchical image object classification scheme based on three levels has been generated and analysed to identify feature characteristics associated with individual species recognition and thereby aid production of NFIs