Landscape structure assessment with image grey‐values and object‐based classification at three spatial resolutions

Abstract

The analysis of landscape pattern through remote sensing data is relatively widespread in landscape ecology and landscape planning. However, the lack of comparability of results between different image-processing methods and across spatial resolutions limits the potential usefulness of landscape pattern indices. In this study, 96 sampling plots in Switzerland were investigated covering land-use intensities ranging from old-growth forest to intensive agricultural landscapes. The sampling plots were captured using fused Landsat ETM-IRS, Quickbird and aerial photograph data. In order to quantify landscape patterns, seven patch indices (derived by object-oriented classification) and six grey-value indices were extracted from the sampling plots. Principal component analysis was applied to the datasets, with the amount of variance in the first four axes compared among the sampling plots. Biplots of indices and sampling plots derived from all datasets were investigated with respect to land-use intensity patterns. PCA results indicated that increasing spatial resolution corresponded to a slight increase in explained variance. Moreover, image grey-values explained more variance between the sampling plots than segmented patch indices. Furthermore, biplots of grey-value indices were capable of grouping sampling plots according to the land-use intensity gradient, while segmented patch indices failed to adequately represent these.

Object based image analysis for remote sensing

“…This approach was applied to the Congo River basin to accurately estimate deforestation at regional, national and landscape levels. Ivits and Koch (2002) and Ivits et al (2005) analysed landscape patterns for 96 sampling plots in Switzerland, based on OBIA-derived patch indices for land-use intensities ranging from old-growth forests to intensive agricultural landscapes: landscape patterns could be quantified on the basis of merged Landsat ETM-IRS, QuickBird and aerial photographic data. Yan et al (2006) compared per-pixel and OBIA classifications for land-cover mapping in a coal fire area in Inner Mongolia, and found the differences in accuracy, expressed in terms of proportions of correctly allocated pixels, to be statistically significant.…”

Object-based target search using remotely sensed data: A case study in detecting invasive exotic Australian Pine in south Florida

“…Recently, an object-oriented approach has been increasingly used in remote sensing image classification (e.g. Herold et al (2003), Hodgson et al (2003), Laliberte et al (2004), Ivits et al (2005), Robinson et al (2005), Chubey et al (2006), Yu et al (2006), Mallinis et al (2008), Lafarge et al (2008)) with the coming of commercially available image segmentation (object extraction) algorithms and software (e.g. Definies Professional 5.0, or formerly eCognition).…”

Comparison of relative radiometric normalization methods using pseudo-invariant features for change detection studies in rural and urban landscapes

“…26 For example, both the 6S (second simulation of the satellite signal in the solar spectrum) radiative transfer code 36 and the modtran 4þ radiative transfer code 37 require visibility estimates. 38 Atmospheric data such as water vapor are often complex to acquire. Field-based methods include ground-based LIDAR, weather balloons, and surface meteorological observations.…”