Characterising land cover and detecting land-cover changes using spatial methods is an area of research that has been attracting increasing attention recently. We compare performances of selected pattern recognition methods for characterising different land covers using unclassified Landsat Thematic Mapper (TM) data for a lowland site in north-eastern Costa Rica. Two spatial statistics (fractal dimension, using the isarithm and triangular prism surface area (TPSA) methods, and Moran's I index of spatial autocorrelation) and selected landscape indices (Shannon's diversity index, contagion, and fractal dimension from perimeter/area) were investigated. Mean values of each metric for each cover type were calculated for subset areas representing forest, agriculture, pasture, and scrub, for all seven Landsat-TM bands and the Normalized Difference Vegetation Index (NDVI). Fractal dimension (DTPSA) and Moran's I were found to be useful for characterising spatial complexity of Landsat-TM data, whereas the standard landscape indices were not. Values of DTPSA decreased along a gradient of increasing human disturbance: old-growth forest-scrub-pasture-agriculture. These results can be further applied locally using moving windows for change detection in global environmental change studies. Moreover, in this information era characterized by increasingly abundant imagery, these spatial statistics could serve as metadata for content-based data mining of imagery.
ASJC Scopus subject areas
- Earth and Planetary Sciences(all)