GEOBIA systems for massive data processing

Large portions of the Earth’s landscape are now captured by high-resolution remotely-sensed datasets and turned into corresponding thematic data. Despite these advancements the number of comprehensive, high-resolution land-cover maps is surprisingly low. The value of high-resolution land-cover data in landscapes that are increasingly fragmented and heterogeneous is great, but so are the challenges associated with turning these disparate datasets into information. We argue that effective geographic object-based image analysis (GEOBIA) system design, while rarely discussed in the literature, is perhaps the most important factor in determining the success of projects whose focus is on broad-area mapping. Human resources, data, hardware, and software must be tightly integrated to make the system efficient and effective. At the same time, the object-based approaches used by such systems for land-cover mapping must try to replicate the human cognitive process as much as possible, using stable, context-based approaches to feature extraction that leverage the strengths of the various input datasets while minimizing their weaknesses. Drawing on our experience deriving 12 terabytes of high-resolution land cover for more than 232,000 km2 in the United States, we describe the design considerations for GEOBIA systems that are capable of processing huge volumes of data. In addition, we provide examples of the techniques and approaches deployed within these systems that overcome the challenges associated with mapping land cover from massive, disparate datasets.