One may even move beyond events to understand spatiotemporal interactions among event types and underlying processes. For example, a terrorism monitoring and prediction center could use fusion to estimate the parameters of a social-cultural model, which could then be used to assess risks of terrorist attacks at particular locations.
Knowledge and Skills
Fusion draws on many disciplines, including geographic information science, spatial statistics, remote sensing, computer science, electrical engineering, and physics. The concepts are taught at the university level under a variety of topics, such as map conflation (Saalfeld, 1988; Kang, 2009; Longley et al., 2010); spatial statistics (Bivand et al., 2008; Cressie and Winkle, 2011); spatial data mining (Shekhar and Xiong, 2008; Shekhar et al., 2011); data, sensor, or image fusion (Hall and Llinas, 1997; Pohl and Van Genderen, 1998; Hyder et al., 2002; Mitchell, 2010a, b); semantic web (Antoniou and Harmelen, 2004; Allemag and Hendler, 2011); and data, information, or schema integration (Batini et al., 1986; Sheth and Larson, 1990; Lenzerini, 2002; Dyché and Levy, 2006; Halevy et al., 2006). Increasingly this means using an interdisciplinary approach, especially as new data sources (e.g., sensor webs, social network data) are added to existing data sources (e.g., remote sensing). Searching for structure within large volumes of complex, multitheme, and multitemporal data (e.g., big data) also requires interdisciplinary skills, which will become increasingly important as data input sizes continue to grow. “Big data” are often defined by data volumes, variety, and uptake rates that are so large that they challenge the accepted methods of data aggregation, description, visualization, and analysis. Big data present important challenges to GEOINT fusion where current approaches are not scalable. Skills for dealing with these massive data agglomerations may require recruitment of data specialists.
A variety of skills are necessary to handle the workflow to produce GEOINT fusion. For situation awareness, for example, the workflow may include tasks such as identifying relevant sources, georegistering new information (e.g., aerial images), detecting and resolving inconsistencies and uncertainties across sources, characterizing new phenomena from data sources using models, and making cartographic and visualization decisions for presenting the information. Based on common workflows, the necessary skills for fusion include the following:
• Task-relevant source identification. During the 1980s, there were few geospatial intelligence data sources and most of the effort was dedicated to processing. However, advances in sensing, communication, and data management have greatly increased the number of potential sources. As a result, fusion is now leveraging an increasingly diverse array of information sources, including new physical sensors (e.g., videos from unmanned aerial vehicles), social media, and data sets gathered by governments, businesses, and scientists.
• Knowledge of common geospatial intelligence data sources. Data fusion often starts by merging data from multiple sources, which may have different data formats, geographic coordinate systems, geographic resolution, accuracy, and timeliness and are commonly handled by different domain experts. Knowledge of these differences is needed to load data into software systems, to merge data from multiple sources, and to resolve conflicts across data sources.
• Georegistration methods. Fusion often adds new information to a geospatial data set. For example, georegistering information from sources such as aerial imagery, Global Positioning System (GPS) tracks, and cell phones allows information on current locations of friends and foes to be added to a base map. Aerial imagery may be georegistered by identifying several landmarks common to the image and the base map and applying photogrammetric principles. A GPS track may be georegistered to a roadmap in an urban area by identifying the closest roads.
• Deriving new information from sources and managing uncertainty in a complex multisource environment. Some phenomena cannot be fully characterized from observations. Statistical and data-mining methods are used to remove anomalies, identify correlations across data sources, find clusters or groups, and classify or predict specific features using data sources as explanatory features. Evidential reasoning methods such as Bayes’ rule or the Dempser-Shafer theory of evidence may be used to estimate the most likely location and shape of a feature from the information available.