Classification of late seral coastal temperate rainforests with high spatial resolution QuickBird imagery

Abstract

Over the last decade, the spatial resolution of satellite imagery has approached the fine resolution of aerial photographs. However, the suitability of high spatial resolution satellite imagery in providing detailed forest inventory information remains an area of ongoing research. In this paper, we assess the relative roles of high spatial resolution satellite imagery and ancillary landscape positional data (elevation and potential soil moisture) to detect the fine-scale variability in late seral coastal temperate rainforests in British Columbia, Canada. Using an object-based classifier, a broad-scale classification first delineated late seral forests from younger forests. This was followed by a second, finer-scale classification of late seral vegetation associations. The results of the broad classification indicated that late seral forests can be well delineated from younger forests, with an accuracy exceeding 90%. Results from the finer-scale classification indicated that the ability to discriminate different forest associations was lower and highly variable depending on the location and type of class. In particular, results indicated that spectral and textural information was less useful in discriminating among late seral forest associations than were landscape positional variables. High spatial resolution satellite imagery has potential for use in ecosystem inventories because of its ability to map and monitor forest cover at multiple scales. However, further testing of the capabilities of QuickBird imagery in heterogeneous forested landscapes is necessary.