Feasibility of pattern type classification for landscape patterns using the AG-curve
Context Ecological data often contain spatial structures that are latent indicators of ecological processes of interest. The emergence of spatial pattern analysis has advanced ecological studies by identifying spatial autocorrelation and testing its relationship to underlying processes. Spatial point pattern tests such as Ripley’s K function were designed for identifying spatial patterns, however they are not without their limitations. Objectives Recently another graphical technique, AG-curve, was proposed. This paper examines its suitability for classifying disturbance patterns in remote sensing scenery containing tens of thousands of pixels. Methods To answer the question, Is there a significant pattern of disturbance or decline present?, landscapes that were subject to disturbance from mining, wildfire and logging activities were analyzed and compared using the AG-curve technique, which classifies spatial patterns in a window as either random, aggregated, or regular (dispersed). 40 9 40 km windows of NDVI data covering the three prototypical disturbance landscapes and one undisturbed landscape were analyzed for the presence of patterns. Results From a raster representing the net change in NDVI spanning 18 years, the AG-curve correctly classified the spatial pattern of disturbance in the three disturbance landscapes as a pattern of aggregation among the net-loss in NDVI pixels. In contrast, the undisturbed landscape was classified as random. Conclusion The AG-curve is a descriptive classification technique useful for identifying spatial patterns in remote sensing imagery and discerning clustered from dispersed patterns. Results highlight that information about the spatial scale of the pattern is also apparent when interpreting the AG-curve graph.