1 Multiscale Object - Specific Analysis ( MOSA ) : An Integration of Ecological Theory , Remote Sensing , and Spatial Modelling

It is now widely recognized that landscapes are complex systems composed of multiscale hierarchically organized entities that interact within unique spatial and temporal scales. These interactions result in scale-dependent spatial patterns that visually change, depending upon their scale of observation. Remote sensing platforms represent the primary data source from which such landscape patterns can be observed and assessed, but suffer from the modifiable areal unit problem (MAUP). The clearest way out of MAUP is by using objects, as objects constitute a non-arbitrary representation of space. Consequently, their aggregation and scaling contains implicit ecological meaning. Therefore, to appropriately monitor, model, and manage our interaction within landscapes, we require a multiscale approach that judiciously integrates ecological theory, remote sensing data and spatial modeling capabilities for the automatic delineation, hierarchical linking, evaluation, and visualization of dominant landscape objects through scale. Furthermore, this approach should be guided by the intrinsic scale of the varying sized, shaped, and spatially distributed image-objects that compose a remote sensing scene. In an effort to achieve this, we present Multiscale Object-Specific Analysis (MOSA) as a novel approach for automatically upscaling and delineating multiscale landscape structures from a high-resolution remote sensing image. MOSA is composed of three primary components: Object-Specific Analysis (OSA), Object-Specific Upscaling (OSU) and Marker Controlled Watershed Segmentation (MCS). OSA is a multiscale approach that automatically defines unique spatial measures specific to the individual imageobjects composing a remote sensing scene. These object-specific measures are then used in a weighting function to automatically upscale (OSU) an image to a coarser resolution by taking into account the spatial influence of the image-objects composing the scene at the finer resolution. Because imageobjects, rather than arbitrary pixels, are the basis for upscaling, the effects of the modifiable areal unit problem (MAUP) are reduced. MCS is then applied to the newly upscaled data to automatically segment them into topologically discrete image-objects that strongly correspond to visually defined image-objects. The elegance of utilizing MCS as a feature detector is that it requires inputs that are automatically and explicitly met by the OSA/OSU outputs. Analysis is performed on an IKONOS-2 image (acquired August, 2001) that represents a highly fragmented agro-forested landscape in the Haut St-Laurent region of south-western Québec, Canada. 1 An extended version of this work appears in: Hay, G. J., and Marceau, D. J. 2004. Multiscale Object-Specific Analysis (MOSA): An integrative approach for multiscale landscape analysis. In: S. M. de Jong & F. D. van der Meer (Eds). Remote Sensing and Digital Image Analysis: including the spatial domain. Book series: Remote Sensing and Digital Image Processing. Volume 5. Chapter 3. Kluwer Academic Publishers, Dordrecht (in press). 2 Please reference as: Hay, G. J. and D. J. Marceau, 2004. Multiscale Object-Specific Analysis (MOSA): An integration of Ecological Theory, Remote Sensing, and Spatial Modelling. Proceedings of Bridging Scales and Epistemologies: Linking Knowledge with Global Science in Multi-Scale Assessments. Bibliotheca, Alexandrina, Alexandria, Egypt. March 17-20. 3 Corresponding author