Spatially Consistent Landcover Maps for Reliable Landscape Monitoring: an Object-based Disturbance-inventory Approach

Landscape monitoring studies often employ metrics to quantify changes in landscape composition (e.g. proportional forest area) and pattern (e.g. edge density, mean patch size), through the analysis of land cover maps arranged in a time series. Remote sensing plays a key role in such studies; however, the approaches for generating multi-temporal time series are not yet fully developed. For example, spurious differences can arise from errors in classification, image registration, scene illumination, resolution, and segmentation; and can substantially impact the monitoring of trends. While much progress has been made in refining classification and change detection in an object-based environment, separating real from spurious changes is still a complex task, and one that is often performed manually. In this paper, we present a summary of our new disturbance-inventory approach to creating spatially consistent landcover maps through the automated updating and backdating of an existing object-based reference map. With this approach, the final maps are only altered in confirmed regions of change, i.e. detected dynamic objects, and spatial consistency is enforced through boundary-conditioning rules. This approach constitutes an innovative, efficient and transparent framework that is able to produce spatially consistent time series with landscape dynamics ranging from feature appearance, disappearance, succession, expansions and shrinkage, without the need for manual editing.