Geostatistical methods for analysis of multiple scales of variation in spatial data

Scale-related effects in digital images result from interaction between the measurement support size and multiple scales of landscape variation. Analysis of such effects requires consideration of two distinct scale-related concepts. First, most spatially varying properties exhibit multiple characteristic scales, or typical spatial periodicities. Second, observation scale, equivalent to sensor resolution, controls the degree to which scene properties can be detected and monitored. While characteristic scales are a fixed property of a scene, observation scale must be chosen appropriately to detect the pertinent scene variation.; Multiple characteristic scales are modeled as effects arising from Analysis of Variance applied to spatial hierarchies. The nested hierarchical model of landscapes is based on ideas from landscape ecology, and incorporates elements of the theory of regionalized variables. Decomposition of variance into scale-specific components is reflected by a similar decomposition of the scene variogram. The variogram, in turn, is the central element of a model relating image variance with the spatial response of a sensor. Combination of this model with the scale decomposition of a variogram allows use of geostatistical methods to analyze the relationship between characteristic scales and observation scales.; The models developed support an analytical approach to understanding scale effects in remote sensing. Several applications are presented. First, the model relating variance and sensor response is used with the indicator variogram of a geophysical field to describe the sensitivity of cover estimates to chosen threshold values. Estimation of snow cover in a forest scene is found to be highly sensitive to thresholds even at relatively fine resolutions. Monitoring forest clearcutting, however, is fairly robust even at relatively coarse scales. Another application investigates scale-related properties of change indicators derived from remote sensing in temperate conifer forests. Comparison of scale-specific variograms reveals that different types of change are manifest at distinct characteristic scales.; Such applications indicate the benefits of geostatistical modeling of scale effects for evaluating image information content as a function of spatial resolution. These methods can guide data processing strategies and help planning of future remote sensing systems.