Πects of spatial aggregation approaches on classi ed satellite imagery

Spatial aggregations of raster data based on the majority rule have been typically used in landscape ecological studies. It is generally acknowledged that (1) dominant classes increase in abundance while minor classes decrease in abundance or even disappear through aggregation processes; and (2) spatial patterns also change with aggregations. In this paper, we examined an alternative, random rule-based aggregation and its eŒects on cover type abundance and landscape patterns, in comparison with the majority rule-based aggregation. We aggregated a classiŽ ed TM imagery (about 1.5 million ha) from 30m (4231 3 3717 pixels) incrementally to 990m resolution (132 pixels 3 116 pixels). Cover type proportion, mean patch size ratio, aggregation index (AI ), and fractal dimension (FD) were used to assess the eŒects of aggregation. To compare landscapes under diŒerent resolutions, we assumed that the landscapes were least distorted if (1) the cover type proportions and mean patch size ratios among classes were maintained, and (2) all cover types responded in the same way for a given index as aggregation levels increased. For example, distortion is introduced by aggregation if some cover types increase their AI values with increasing aggregation levels while other cover types decrease. Our study indicated that the two spatial aggregation techniques led to diŒerent results in cover type proportions and they altered spatial pattern in opposite ways. The majority rule-based aggregations caused distortions of cover type proportions and spatial patterns. Generally, the majority rule-based aggregation Ž ltered out minor classes and produced clumped landscapes. Such landscape representations are relatively easy for interpreting and, therefore, are suitable for land managers to conceptualize spatial patterns of a study region. By contrast, the random rule-based aggregations maintained cover type proportions accurately, but tended to make spatial patterns change toward disaggregation. Overall, the measurements of landscape indices used in this study indicated that Internationa l Journal of Geographica l Information Science ISSN 1365-8816 print/ISSN 1362-3087 online © 2002 Taylor & Francis Ltd http://www.tandf.co.uk/journals DOI: 10.1080/13658810110075978