Landscape metric performance in analyzing two decades of deforestation in the Amazon Basin of Rondonia, Brazil

Sixteen landscape metrics were evaluated with respect to the effects of spatial aggregation on six different years of Landsat data for a deforested area in Rondonia, Brazil. Spatial aggregation was performed by two methods. The first method involved varying the window size in texture mean co-occurrence filtering prior to classification. The second method involved aggregating the data post-classification by resampling with a majority filter. The Landscape Shape Index (LSI) and Square Pixel (SqP) metric showed the most predictable behavior of the shape complexity metrics having strong decreases with each increase in aggregation. The Edge Density (ED) and Patch Density (PD) metrics showed the most predictable behavior among the edge and patch metrics, decreasing with increasing aggregation. The Mean Nearest Neighbor (MNN) metric also behaved as expected but its results were less consistent than those of ED and PD. Many of the remaining metrics gave inconsistent and unpredictable results with respect to spatial aggregation.