A Detection Threshold Model for Multi-Scale Multi-Temporal Spectral Mixture Analysis

Christopher Small, Columbia University, cs184@columbia.edu (Presenting)

Spectral Mixture Analysis (SMA) can be extended to multi-temporal SMA (MTSMA) to represent changes in land surface properties in terms of the fundamental physical classes of land cover (e.g. vegetation, water, snow/ice, rock & soil substrates). Because SMA represents the mixed radiance field in terms of spatial abundance of endmembers with distinct physical properties it is inherently scaleable. However, the physical model on which it is based is inherently scale-independent but still accommodates the dependence of the signal on both sensor resolution and characteristic scale of individual land cover components. A major benefit of SMA is its ability to represent both apparent LC change and actual LC change in the mixed radiance field. This provides a basis for estimating detection thresholds to represent the magnitude of apparent change below which actual change cannot be resolved - for each spectral endmember. Examples and estimation procedures are developed for the primary sources of apparent change influencing imagery at multiple spatial scales. One important benefit of MTSMA is its ability to derive land surface physical properties directly from optical radiance measurements without the loss of information or introduction of error inherent in parameterizations of thematic classifications.