Click for next page ( 36


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 035
3 Emerging Areas of Geospatial Intelligence T he National Research Council (NRC, 2010a) GEOINT FUSION report New Research Directions for the National Geospatial-Intelligence Agency: Workshop ­Report GEOINT fusion is concerned with combining identified five emerging subject areas that could poten- geographic information from multiple sources, whether tially improve geospatial intelligence: ­ eospatial intel- g structured or unstructured (e.g., sensor networks, data- ligence (GEOINT) fusion, crowdsourcing, human ­ bases, documents), to assess spatial or spatiotemporal geography, visual analytics, and forecasting. 1 Although phenomena for purposes such as tracking, prediction, or human geography goes back more than a century, providing a common operational picture. For example, technological and analytical developments have so a situation assessment of an ongoing event such as the changed the field that it is treated as an emerging area 2011 Arab Spring may fuse location-aware data from in this report. Among the emerging areas, there is airborne or satellite sensors, social media (e.g., Twitter, an emphasis on crosscutting concerns such as three- blogs), news wires, and reports from observers on the dimensional and spatiotemporal visualization, as well ground. Fusion is important because assessments of a as linkages between geolocation, social media, crowd- phenomenon from multiple sources of information are sourcing, and spatial analysis. GEOINT fusion covers likely to be better than those from a single source. the linkages, while each of the emerging areas shares the cross­ utting concerns. c Evolution The five emerging areas are relatively computation- ally oriented and interdisciplinary, with concepts and Research findings on GEOINT fusion began to skills taught across academic departments. Few are sup- be published in the 1980s. Early work provided a clas- ported by degree programs or academic infrastructure sification of use cases (White, 1999) for common tasks (e.g., professional societies, journals), although these such as object refinement (e.g., observation-to-track will come as the fields develop. This chapter describes association, target type and identification), situation each of the five emerging areas, including its origin, assessment (e.g., identification of force structure, com- the knowledge and skills that are taught, and the scope munications, and physical context), impact assessment of existing education and professional preparation (e.g., consequence prediction, susceptibility and vul- programs. nerability assessment), and process refinement (e.g., adaptive search and processing, resource management). These use cases have two dimensions—geographic footprint and temporal extent—as shown in Figure 3.1. The simplest use case (level 0, subobject) fuses data 1 at the granularity of a single location, such as a pixel Note that these terms differ slightly from those used in NRC (2010a). in a remote sensing image. For example, a new image 35

OCR for page 035
36 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE FIGURE 3.1  The complexity and methods used in GEOINT fusion depend both on the size of the geographical area (horizontal axis) and the length of the time period (vertical axis) being covered. In this figure, the classification of use cases is shown by these dimensions. may be georegistered to a reference map by aligning quality on input quality. For example, fusion may be specific image pixels to corresponding map landmarks. used to reconfigure the locations and trajectories of At level 1 (object/entity), information from multiple sensor platforms (e.g., satellites, aircraft, vehicles) to sensors with overlapping sensing ranges is combined to closely monitor an event (e.g., hurricane) or high-value estimate properties (e.g., location, shape, size, type) of target in order to improve the quality of fused output an identifiable entity, such as a vehicle or building. For estimates of interest. example, a national air-traffic monitor room may track The late 1990s brought the establishment of the every aircraft using information collected from local International Society for Information Fusion as well air-traffic controllers. At level 2 (situation assessment), as two journals dedicated to information fusion: Jour- information from all sources is combined to estimate nal of Advances in Information Fusion and Information the impact of a recent event or behavior on a geographic Fusion. Conference discussions and publications have area of interest. For example, an emergency manager refined the use cases in new directions. For example, may fuse weather prediction data sets, plume simula- a long time series of snapshots may enrich traditional tion maps, population density maps, and transportation fusion with concepts from time-space geography maps to identify emergency evacuation routes. (­ ägerstrand, 1967) and the dynamics of geographic H The subobject, object, and situation assessment domain (­ ornsby and Yuan, 2008), leading to a new H levels are often concerned with a single point in time use case (level 5). At the location (e.g., pixel) geo- (snapshot). However, multiple time frames can be con- graphic footprint, a past time series of measurements sidered at any level. At level 3 (impact assessment), a can be used to determine a statistical distribution, recent image may be compared with an older image to which, in turn, can be used to evaluate future values detect major changes in an object or geographic area of for anomalies, regime-change points, and other fac- interest. For example, the impact of a forest fire may be tors. At an identifiable-entity geographic footprint, a assessed by comparing remotely sensed images before time series of locations produced by multiple sensor and after the fire. At level 4 (process refinement), the measurements for an object can be fused to estimate process of data collection and fusion is refined using the object’s trajectory, which can be processed further what could be considered “control law” that depends on to identify its frequent locations, routes, schedules, and a utility function expressing the dependence of ­usion f other spatiotemporal patterns (Shekhar et al., 2011).

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 37 One may even move beyond events to understand characterizing new phenomena from data sources spatio­emporal inter­ ctions among event types and t a using models, and making cartographic and visualiza- under­ying processes. For example, a terrorism moni- l tion decisions for presenting the information. Based toring and prediction center could use fusion to esti- on common workflows, the necessary skills for fusion mate the parameters of a social-cultural model, which include the following: could then be used to assess risks of terrorist attacks at particular locations. • Task-relevant source identification. During the 1980s, there were few geospatial intelligence data Knowledge and Skills sources and most of the effort was dedicated to process- ing. However, advances in sensing, communication, and Fusion draws on many disciplines, including geo- data management have greatly increased the number of graphic information science, spatial statistics, remote potential sources. As a result, fusion is now leveraging sensing, computer science, electrical engineering, and an increasingly diverse array of information sources, physics. The concepts are taught at the university including new physical sensors (e.g., videos from un- level under a variety of topics, such as map conflation manned aerial vehicles), social media, and data sets (­ aalfeld, 1988; Kang, 2009; Longley et al., 2010); spa- S gathered by governments, businesses, and scientists. tial statistics (Bivand et al., 2008; Cressie and Winkle, • Knowledge of common geospatial intelligence 2011); spatial data mining (Shekhar and Xiong, 2008; data sources. Data fusion often starts by merging data Shekhar et al., 2011); data, sensor, or image fusion from multiple sources, which may have different (Hall and Llinas, 1997; Pohl and Van Genderen, 1998; data formats, geographic coordinate systems, geo- Hyder et al., 2002; Mitchell, 2010a, b); semantic web graphic resolution, accuracy, and timeliness and are (­Antoniou and Harmelen, 2004; Allemag and Hendler, commonly handled by different domain experts. 2011); and data, information, or schema integration Knowledge of these differences is needed to load data (Batini et al., 1986; Sheth and Larson, 1990; Lenzerini, into software systems, to merge data from multiple 2002; Dyché and Levy, 2006; Halevy et al., 2006). sources, and to resolve conflicts across data sources. Increasingly this means using an interdisciplinary • Georegistration methods. Fusion often adds approach, especially as new data sources (e.g., sensor new information to a geospatial data set. For example, webs, social network data) are added to existing data georegistering information from sources such as aerial sources (e.g., remote sensing). Searching for structure imagery, Global Positioning System (GPS) tracks, and within large volumes of complex, multitheme, and cell phones allows information on current locations multi­ emporal data (e.g., big data) also requires inter­ t of friends and foes to be added to a base map. Aerial disciplinary skills, which will become increasingly imagery may be georegistered by identifying several important as data input sizes continue to grow. “Big landmarks common to the image and the base map data” are often defined by data volumes, variety, and and applying photogrammetric principles. A GPS track uptake rates that are so large that they challenge the may be georegistered to a roadmap in an urban area by accepted methods of data aggregation, description, identifying the closest roads. visualization, and analysis. Big data present important • Deriving new information from sources and challenges to GEOINT fusion where current ap- managing uncertainty in a complex multisource envi- proaches are not scalable. Skills for dealing with these ronment. Some phenomena cannot be fully character- massive data agglomerations may require recruitment ized from observations. Statistical and data-mining of data specialists. methods are used to remove anomalies, identify corre- A variety of skills are necessary to handle the lations across data sources, find clusters or groups, and workflow to produce GEOINT fusion. For situation classify or predict specific features using data sources awareness, for example, the workflow may include tasks as explanatory features. Evidential reasoning methods such as identifying relevant sources, georegistering new such as Bayes’ rule or the Dempser-Shafer theory of information (e.g., aerial images), detecting and resolv- evi­ ence may be used to estimate the most likely loca- d ing inconsistencies and uncertainties across sources, tion and shape of a feature from the information avail-

OCR for page 035
38 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE able. Optimization techniques from operations research In addition, database courses discuss schema integra- are often required to develop solutions to complex tion and data integration; signal processing courses combinatorial optimization problems across all fusion discuss sensor fusion; and statistics, data mining, and levels. Simulation models may be used to project phe- spatial computing courses discuss spatial statistics nomena, such as trajectories of chemical plumes. and spatial data mining. • Geospatial intelligence information presenta- Graduate degrees in related areas (e.g., geographic tion. Fusion results are often presented on maps. Prepa- information science, remote sensing, computer sci- ration of paper maps requires traditional cartographic ence and electrical engineering) allow a specialization skills, and preparation of electronic maps requires skills in fusion through research projects and coursework to leverage animation and interaction in context of from relevant disciplines. Degrees with a fusion computer screens, tablets, and cell phones. specialization are available from universities with • Workflow management. Workflow manage- strong ­ ducation and research presence in geographic e ment systems may be used to specify fusion tasks and infor­ ation science, remote sensing, spatial statis- m their interdependencies as well as to help keep track tics, computer ­cience, electrical engineering, and s of progress and facilitate communication among team p ­ hysics. Examples include George Mason University; members. Workflows also enable fusion tasks to be G ­ eorgia Institute of Technology; Ohio State Univer- handled within a data collection-analytical context, sity; Pennsylvania State University; Purdue University; thus increasing the operational value of the fused data. University of California, Santa Barbara; University of Minnesota; and the University of Southern California. Education and Professional Preparation Programs Some professional programs in related broader areas (e.g., geospatial intelligence, geographic informa- Although no degree programs are offered in tion science, security technologies, dynamic network G ­ EOINT fusion, two universities have a research analysis) provide opportunities to specialize in fusion by center in fusion: the State University of New York, allowing students to choose a fusion-related capstone Buffalo (Center for Multisource Information ­ usion) F project and enroll in fusion-related elective courses. and ­ ennsylvania State University (Center for P Such training opportunities are available at several uni- N ­ etwork-Centric Cognition and Information ­ usion). F versities, including George Mason University; Georgia In addition, some universities offer courses in various Institute of Technology; Pennsylvania State University; aspects of fusion, largely in computer science depart- Redland University; the University of California, Santa ments (e.g., Table A.6 in Appendix A). Semantic web Barbara; and the University of Minnesota. courses are offered by many universities, including Johns Hopkins University, Georgia State University, CROWDSOURCING and Lehigh University. Database interoperability and data integration courses are offered by the University of The term crowdsourcing was introduced by Jeff Southern California and by industry (e.g., Oracle, SAS, Howe in a 2006 article in Wired Magazine (Howe, Sybase, IBM). Courses in multisensory data fusion are 2006) and is defined in the 2011 Merriam-Webster offered by a few universities (e.g., Pennsylvania State dictionary as “the practice of obtaining needed services, University; Arizona State University; Georgia Institute ideas, or content by soliciting contributions from a of Technology; University of California, Los Angeles; large group of people and especially from the online State University of New York, Buffalo) and by industry community rather than from traditional employees or (e.g., Objectivity Inc., Applied Technology Institute). suppliers.”2 Crowdsourcing is related to participatory In addition, fusion topics are commonly discussed sensing, which shares the same principle of collecting for a few weeks in courses on broader topics at many data from a set of users working collaboratively (Estrin, research universities. For example, geographic informa- 2010). The two terms are often used interchangeably, tion science courses often discuss map conflation, and but the committee prefers the term crowdsourcing, remote sensing and photogrammetry courses discuss image-to-image and image-to-reference (map) fusion. 2 See .

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 39 which implies not only data collection but also other Cheng, 2009; Hoh et al., 2012), available today from types of group activities, such as using performing numerous companies (Google, INRIX, NAVTEQ, work. Spatial information contributed by crowd­ Waze, BeatTheTraffic.com). The concept was soon sourcing is often referred to as volunteered geographic extended to enriching maps with other user-generated information. Because such information is collected by content, either through location-based services or volunteers, it comes with challenges of accuracy, cred- posting from public records. Examples include maps ibility, and reliability (Goodchild, 2007; Flanagin and of crime in Oakland,6 geolocalized real estate data Metzger, 2008). As the use of crowdsourced data grows, (e.g., Zillow), photographic geolocalized postings (e.g., issues of data quality, uncertainty, trust, and conflation Flickr), pedestrian and sports GPS traces (e.g., Nokia7), at the semantic level will increase in importance. and earthquake information (e.g., Figure 3.3). The explosion of location-based services has led Evolution to the emergence of users sharing personal informa- tion (e.g., Facebook), professional information (e.g., Perhaps the earliest example of crowdsourcing was LinkedIn), location (e.g., presence in a restaurant, at the Longitude Prize, a reward offered by the United a landmark location; FourSquare), and social network Kingdom in 1714 to anyone who could develop a prac- activities (e.g., placing Facebook activity on maps; tical method to precisely determine a ship’s longitude. Loopt). This new information complements traditional Another early example is the 19th Century Oxford cell tower information, which is already used in opera- English Dictionary, whose editors asked the public to tional contexts (e.g., tracking al-Zarqawi by the U.S. index all words in English and provide example quota- military; Perry et al., 2006), by enriching available feeds tions for them (Winchester, 1998). The pace and scale using attributes disclosed knowingly or not, willingly of these volunteer initiatives has increased in recent or not, by the user. years with the emergence of the Internet and social Finally, new concepts of collaborative work are networking. Among recent high visibility efforts were emerging. Wikipedia created a completely crowd- the DARPA Network Challenge to collaboratively find sourced encyclopedia on a voluntary basis. It was fol- marker balloons deployed by DARPA in the United lowed by numerous services provided by volunteers, States,3 and the Netflix Prize to develop algorithms for such as Facebook translation (Hosaka, 2008) and predicting how well users would like a film, based on Yahoo! Answers. Amazon’s Mechanical Turk8 enables their movie preferences.4 Openstreet Map,5 an editable workers to remotely perform tasks at a distributed and map of the world, has been used by numerous compa- large scale for money. This model represents a new nies (e.g., Waze) as their backbone mapping system. trend in which the crowdsourced workers are active Openstreetmap.org had a remarkable success following and follow directions. This type of activity has been the Haiti earthquake of January 2010, when volunteers used successfully for tagging, identification, labeling, worldwide created a new map from donated imagery parsing, clustering, and recognition. in a few days. The crowdsourced map became the most accurate base for relief efforts (Zook et al., 2010). Knowledge and Skills Today, crowdsourcing plays a major role in creating information-rich maps, collecting geolocalized human Developing the technology for a crowdsourcing activity, and working collaboratively. The convergence system requires knowledge of the problem domain of sensing, communication, and computation on single as well as skills in computer programming (including cellular platforms and the ubiquity of the Internet and parallel programming), data visualization, database mobile web have allowed maps to be enriched with design and management, operating systems, service- a variety of data. Early applications included traffic oriented architectures, Internet applications, and the information collected from smartphones (Figure 3.2; 6 See . 3 See . 7 See the Nokia Sportstracker program at . 5 See . 8 See .

OCR for page 035
40 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE FIGURE 3.2  Example of crowdsourced GPS data, which were obtained by collecting tracks from San Francisco taxis through the Cabspotting program. Each point represents one GPS recording, sampled at an interval of one minute. Three different magnification levels show the detail obtainable from the data. The San Francisco Bay area is shown in red, the approach road to San Francisco Inter- national Airport is shown in blue, and the lanes on the Highway 101 intersection by the airport are shown in black. The road map for the Bay Area can be reconstructed from only one day of data. SOURCE: University of California, Berkeley, Mobile Millennium project. ability to work with various types of data feeds. The • Statistics, machine learning, and large-scale data technology has been developing rapidly, but a generic analytics. Pattern matching, data mining, and statistical set of tools for implementation across applications has inference are needed to extract information from the yet to emerge. large volume of data. Building a crowdsourcing system requires the fol- • Communications, cellular technology, mobile lowing knowledge and skills: computing, and human-computer interaction, which are necessary because numerous crowdsourcing systems • Sensing, including hardware knowledge (any are based on cellular devices. type of sensor), device knowledge (using phones or • Cloud computing and high-performance com- other devices to collect data), and software knowledge puting, which power most crowdsourcing applications. (e.g., collecting data from Internet activity). • Signal processing and filtering, which are needed The knowledge and skills needed to analyze crowd- to remove noise from the data. sourced data as well as the crowdsourcing process are

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 41 FIGURE 3.3  Example of crowdsourced data for earthquakes. The U.S. Geological Survey’s “Did you feel it?” program creates earth- quake intensity maps from user responses. The top figure shows the geocoded intensities for the 2011 Virginia earthquake (magnitude 5.8). The bottom figure shows the intensity collected from user input as a distance from the epicenter (dots). The crowdsourced data is compared to model-based predictions (line). SOURCE: U.S. Geological Survey.

OCR for page 035
42 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE markedly different from those required to develop of crowdsourcing requires a doctorate, although the technology. At a minimum, basic statistical and implementation skills can be obtained at the master’s graphing skills are needed. Additional skills are needed level. For institutions such as the Massachusetts In- to deal with data tagged with location and temporal stitute of Technology, which has a thesis as part of information, including econometrics, error estimation, its master’s program, or the University of California, geo­patial analytics, geospatial visualization, dynamic s Berkeley, which has a project as part of its master’s analysis, temporal clustering, social network ­ nalysis, a of engineering program, students will gain exposure dynamic network analysis, data mining, and text to the topic through the research or project. In addi- mining. tion, many people involved with crowdsourcing are self-taught and learn by doing. Experts at this level Education and Professional Preparation Programs are worldwide and often fall across the age spectrum. Two-year colleges have started to offer curricula to Crowdsourcing is not an established academic dis- a ­ ttract these casual practitioners, such as Android cipline. Students generally gain skills and knowledge in phone programming and scripting for web data crowdsourcing through special projects carried out as scraping. part of a graduate curriculum. Most of the knowledge Finally, with the rise of Web 2.0 and the social web, required for crowdsourcing lies outside traditional geo- numerous companies have trained engineers in house, spatial domains, as illustrated by the skills listed above. enabling them to develop most of the skills needed for For this reason, training is distributed among academic crowdsourcing. Several types of companies now have departments and programs, including engineering crowdsourcing skill sets, including the following: (aerospace, civil, computer science, electrical, environ- mental, mechanical), statistics, geography, urban plan- • Companies which collect vast amounts of ning, and architecture (e.g., Table A.7 in Appendix A). crowdsourced data by the nature of their products, such A few multidisciplinary research institutes at universi- as Google, Facebook, Twitter, and FourSquare. Each ties offer knowledge and skills aligned with training in of these companies has divisions or at least groups that crowdsourcing, including the following: focus on the internal development of data analytics tools for crowdsourced data. • Center for Embedded Networked Sensing at the • Companies that provide back-end support for University of California, Los Angeles, which was one of systems which rely on crowdsourced data, such as the first centers to make academic contributions in the infrastructure systems companies (e.g., IBM, HP) and field and to offer a doctorate in participatory sensing traffic information companies relying on smartphone (Estrin, 2010). data (e.g., NAVTEQ, Waze, INRIX). • Computer Science and Artificial Intelligence • Companies that have developed a busi- Laboratory at the Massachusetts Institute of Technol- ness around crowdsourced data analytics, such as ogy, which has a diverse faculty spanning most of the S ­ enseNetworks or Sensor Platforms, which were start- fields required for crowdsourcing. ing up when this report was written. • Wireless Information Network Laboratory at Rutgers University, which focuses on privacy and wire- HUMAN GEOGRAPHY less information aspects of crowdsourcing. • Algorithms Machines People at the University Human geography concerns the mapping of people, of California, Berkeley, which focuses on building groups, organizations, sentiments and attitudes, norms, systems that connect people to the cloud to solve hard belief systems, social activities, and “ways of doing problems using large data analytics algorithms and business” over space and time (Figure 3.4). It has been massive amounts of crowdsourced and other data. r ­eferred to by many names, including cultural geogra- phy, human terrain, rich ethnography, cultural map- In most cases, acquiring thorough knowledge ping, social mapping, sociocultural context, and social

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 43 FIGURE 3.4  Different types of visualization used in human geography. (Top right) A social network (lines between the locations of participants) superimposed on a NASA Worldwind visualization of the globe. Such images are used to show the relation of social network ties to physical space. (Middle left) Map showing the density of a particular activity. Each dot indicates the location where an actor of interest has been seen. The background image is a standard ARCGIS shape file. (Middle right) Tracking information for two ships, used to track who or what was where when and to identify common paths. Solid lines show known movement between locations (colored columns) and dashed lines indicate inferred movement or lack of movement. Time is vertical and locations are horizontal. (Bottom left) Locations of actors of interest (dots) and secondary information about the spatial density of the betweenness of the nodes (clouds; e.g., Freeman, 1977). Such images are used to identify critical locations. Background image is from NASA Worldwind. (Bottom right) Heat map image of Afghanistan using a standard ARCGIS shapefile. Each region is colored by the number of times actors of interest have been in that region. The brighter the red, the higher the level of activity. Such images are used to understand the region of activity and identify points of intervention. SOURCE: All images were produced using ORA. domain. The use of new technologies and methods, Evolution such as network analysis, graph-based statistics, and evolutionary agent-based modeling, distinguishes the Although human geography has been around for emerging area of human geography from its roots as a more than a century, the decision to build a human subfield of geography, sociology, and anthropology. terrain program for the wars in Iraq and Afghanistan

OCR for page 035
44 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE led to the rethinking of the role of human cultural foods, habits, religions, etc., which are increasingly knowledge. The human terrain program brought ­taking the form of web-based mashups. Such overviews t ­ogether information technology and a vast array and the tools for analyzing them formed the basis of of regional sociocultural information that had been human terrain efforts during the Iraq and Afghanistan scraped from the web, provided through social media, wars. gathered from other open sources, and collected in 4. Sociolinguistic ethnic characterizations—­ the field. The data were analyzed using search and mapping which families, clans, and tribes are where comparison techniques, social network analytics, (e.g., the tribal sociolinguistic heredity network). geographic visualization, and statistics. The aim was to provide up-to-date, accurate information about the Each of these areas requires different expertise. Some general sociocultural environment, current opinion areas require technical expertise (e.g., programming, leaders and persons with power, and climate, eco- scripting) while others require the mastery of advanced nomic, and political conditions. conceptual frameworks and approaches (e.g., agent- Increasingly, sociocultural information, both his- based modeling, network analysis). These skills are torical and current, is being placed on maps. New tech- not generally acquired in traditional courses on sensor nologies that admit location capture (e.g., modern cell assessment, cartography, or map interpretation. phones) are increasing the amount of location-based An important skill in human geography is text data on social and social interaction. Crowdsourcing, mining: the process of deriving high-quality informa- Ushahidi-style data captures (e.g., reports submitted tion from textual sources for analysis. Text, such as by local observers via mobile phone or the Internet), news articles, books, twitter feeds, and blogs, contain location-based twitters, and so on are providing un- information about differences in the human condition precedented levels of sociocultural information that is across locations. Techniques for mining text are reason- at least partially spatially tagged. With new data come ably accurate for extracting the names of people, orga- new research opportunities and the ability to under- nizations, and locations from English texts. However, stand how space constrains and enables social and cul- challenges remain in interpreting multiple languages, tural activity. Illustrative new areas of research include identifying the location of places, and distinguish- geotemporal social media sampling, location identifi- ing between place and person names (e.g., the city of cation from texts, and geonetwork analytics. The next ­Dorothy Pond, Massachusetts) and place and organiza- decade will likely see major changes in the quality of tion names (e.g., the White House). Both geographi- sociospatial data presentation and new technologies for cal expertise and text-mining expertise are needed to capturing, assessing, visualizing, and forecasting social address these problems. data with a spatiotemporal context. Education and Professional Preparation Programs Knowledge and Skills A comprehensive human geography program Human geography involves four main components: covers five core elements: (1) collection and coding of geomarked human data, (2) geo-enabled text analysis, 1. Geo-enabled network analysis—mapping the (3) geo-enabled network analysis and dynamic network network of who, what, how, why, and when to locations analysis, (4) computer simulation of human ­ eography g (e.g., the al-Qaeda social network). data and forecasts, and (5) geocultural analysis and 2. Sentiment and technology dispersion—­mapping overviews. Each of these has an associated set of the movement of ideas, activities, technologies, and m ­ ethods and tools that students need to learn, includ- beliefs as they move from location to location (e.g., the ing (1) tools for collecting social media and news data spread of revolution in the Middle East during the Arab (e.g., TweetTracker, REA); (2) tools for natural lan- Spring). guage processing, text mining, and sentiment mining 3. Cultural geography overviews—compendiums (e.g., AutoMap); (3) tools for metanetwork analytics of diverse information on current leaders, languages, and visualization (e.g., ORA, R); (4) tools for develop-

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 45 ing agent-based and system dynamic simulations (e.g., central to the analyst’s task of drawing conclusions MASON, Construct, Dynamo), with particular atten- from a disparate set of evidence and assumptions. The tion to the diffusion of information and the dispersion objective of visual analytics is to derive insight from of beliefs and activities; and (5) qualitative ethnographic voluminous, changing, vague, and often contradictory assessment, sociolinguistic characterization, sentiment geospatial data and other information while avoid- analysis, text mining, and questionnaires. These ele- ing human information overload (van Wijk, 2011). ments are rarely taught at the undergraduate or master’s Some examples of information graphics used in visual level. Most of the education is at the doctorate level a ­ nalytics are shown in Figure 3.5. (e.g., Table A.8 in Appendix A) or is offered through professional development or specialized training pro- Evolution grams such as the Center for Computational Analysis of Social and Organizational Systems (CASOS) Sum- The growth in the quantities of information that mer Institute. Although many universities cover one or require visual representation and analysis by humans two of these elements in their doctorate programs, only and the increasing complexity of the associated two (Carnegie Mellon University and the University of data and analytical problems have given rise to visual Arizona) cover all five. a ­ nalytics as a new scientific discipline (Andrienko In addition, a number of universities are adding et al., 2010). Visual analytics has formalized only courses in the human-geography area to their doctor- recently, with a key publication in 2005 (Thomas ate programs. For example, the sociology programs at and Cook, 2005) and more recently a series of special Cornell and the University of California, Irvine, and i ­ssues in journals (e.g., Keim et al., 2008; Stapleton the computer science program at the University of et al., 2011). Arizona all cover network analysis with courses related Visual analytics has origins in cartography, geo- to geo-enabled network analysis. The George Mason graphic information science, computer vision, infor- University Center for Social Complexity and the Uni- mation visualization, and scientific visualization. In versity of Michigan Center for Complex Systems cover general, cartography deals with maps and geospatial agent-based modeling that takes account of the spatial data, geographic information science deals with spatial aspects of human behavior. relations and spatial query and analysis, scientific visu- Programs that teach social network analysis alization deals with data that have a natural physical or (Box 3.1) are beginning to cover geo-enabled network geometric structure (e.g., wind flows), and information analysis. Some programs that teach computer modeling visualization deals with abstract data structures (e.g., are beginning to teach the programming and data ac- trees, graphs). Choice and reasoning are central to quisition techniques needed to create and use maps as a visual analytics. way of displaying human behavior. Two-year and com- Research and new directions in visual analytics in- munity colleges have been among the first academic clude creating new information visualization methods, institutions to teach some of the basic skills needed to virtual imaging, semantic search, data fusion, dynamic use and develop social networking tools and, to some network visualization, and user testing. In particular, extent, basic tools necessary for network analysis of methods that focus on how to integrate graphics into s ­ocial data, such as reading GPS signals. These pro- the problem-solving process itself has become a key grams are loosely based in media studies and computer research interest. science programs and are widespread across the nation. Knowledge and Skills VISUAL ANALYTICS Visual analytics deals with amplifying human cog- Visual analytics is the science of analytic reasoning, nitive capabilities facilitated by interactive visual interfaces integrated with computational power and database capacity • by increasing cognitive capacities and resources, (Thomas and Cook, 2005). Analytical reasoning is such as memory;

OCR for page 035
46 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE BOX 3.1 Social Network Analysis There has always been an implicit link between social network analysis and human geography. For example, proximity is a strong basis for individuals forming relations, with most relations weakening with distance. Social network analysis examines the structure of the relations connecting nodes (e.g., people, organizations, topics, events). Many of the earliest studies looked at networks of people connected by relationships such as kinship, mentoring, and works-with. These networks are represented as graphs (e.g., Figure), and matrix algebra or nonparametric network statistics are often used to assess these networks; to identify key nodes, critical dyads, and groups; and to compare and contrast networks (Wasserman and Faust, 1994). Social network analysis is a key methodology in the human geography toolkit. Evolution. Social network analysis emerged prior to World War II, with early advances in fields such as anthropology, sociology, and communications (Freeman, 2006). The past 10 years have seen a movement to broaden the field of social networks. Changes include the transition from graph-theory-based metrics to a combination of graph-based and statistical measures, the expansion from small networks to very large-scale networks, the increased atten­ tion to communication and social media data, and the shift to geotemporal networks. This broader field is often referred to as dynamic network analysis and it is characterized as the study of the structure and evolution of complex sociotechnical systems through the assessment of weighted multimode, multilink, multilevel dynamic networks that are geo-embedded. The field is supported by the quarterly journal Social Networks, the online Journal of Social Structure, and an increasing number of specialty journals such as Social Network Analysis and Mining. Knowledge and Skills. The study of social networks is integral to fields such as statistics, sociology, organizational science, communication, computer science, and forensic science. However, the ubiquity of networks, the value of graphs as a representation, and the strength of structural thinking has increased the interest in networks in almost every scientific discipline. For example, network analysis has been used in sociology to study social and communications networks (Wasserman and Faust, 1994), in biology to study animal behavior (e.g., Krause et al., 2007), and in geography, civil engineering, ecology, and other disciplines to extend graphs to real or abstract space (Haggett and Chorley, 1969; Urban and Keitt, 2001; Adams et al., 2012). This increased interest has led to a proliferation of theories about how these networks form, evolve, and affect behavior. It has also led to new methods, such as dynamic networks techniques for sets of networks through time, and meta-network metrics for multimode, multilink data. Statistical approaches for assessing dynamics, information loss, and error provide the foundation for social network analysis. Social science approaches are used to study the dynamics within social networks (e.g., reciprocity, social influence, power) and the social, institutional, and historical contexts in which network ties are formed and broken. Education and Professional Preparation Programs. Classes in social networks are taught in a number of U.S. universities, usually at the doctorate level. However, undergraduate textbooks and courses are starting to appear. Universities with multiple courses in this area include Carnegie Mellon University, University of Kentucky, Northeastern University, Northwestern University, Harvard, Stanford, Indiana University, and the University of California, Irvine. Courses are taught primarily in business and sociology departments, but also in anthropology, communication, manage­ ment, organizational behavior, organizational theory, strategy, public policy, statistics, information science, and computer science departments. Network analysis in the geometric sense is taught in geography, mathematics, transportation science, computer engineering, and operations research programs. Continuing education programs provide a primary venue for training in this area. For example, didactic seminars are conducted at the main social networks conference (the International Network for Social Network Analysis) for 2 days prior to the conference. Half-day and full-day training programs are often offered at management science, organization theory, sociology, and anthropology conferences. In addition, there are numerous multiday or week-long training programs, including the CASOS Summer Institute, the Lipari summer school, and the East Carolina University program for marine biologists. • by facilitating search; • by providing methods that support exploration • by enhancing pattern recognition, often by re- and discovery. structuring relations within data; • by supporting perceptual inference of structures Methods in visual analytics are based on principles and patterns that are otherwise invisible; drawn from cognitive engineering, design, and percep- • by improving the ability to monitor large num- tual psychology (Scholtz et al., 2009). These methods bers of sensors and events; and provide a means to build systems for threat analysis, prevention, and response. Visual analytics therefore

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 47 FIGURE  Social network analysis is used to show changes in criticality of topics—protests and demonstrations, war and conflict, and Internet and social networking—for the Arab Spring countries. The degree centrality of the three topics (the extent to which a node is connected to other nodes) is based on tags for Lexis-Nexis news articles. The figure shows that the coverage of protests and demonstrations did not spread geographically, and that the change in relevance of the Internet and social networking did not spread in the same way as the revolutions. SOURCE: Courtesy of Jürgen Pfeffer and Kathleen Carley, Carnegie Mellon University. expands the methods available to analysts but also cre- reason spatially, and knowledge of user-centered design ates a need for new sets of skills (Ribarsky et al., 2009). principles. For example, programming or scripting skills Many of these methods are highly dependent on the are needed to develop visualization tools or to extend Internet and on graphics systems and standards. existing tools, which are commonly targeted to particular The suite of skills necessary for research and prac- applications. Searching for structure within large vol- tice in visual analytics includes an ability to program in umes of complex, multitheme, and multi­emporal data t scripting and numerical computing languages, an under­ (e.g., big data) requires inter­ isciplinary skills. ­ evering d S standing of maps and graphics, the ability to think and (2011) noted the importance of moving beyond spe-

OCR for page 035
48 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE FIGURE 3.5  Examples of some information graphics used in visual analytics. (Top) Multimethod display. SOURCE: Screenshot from GeoViz Toolkit developed by Frank Hardisty, GeoVISTA Center. (Center left) A semantic landscape of the Last.fm Music Folksonomy using a self-organizing map. SOURCE: Joseph Biberstine, Russell Duhon, Katy Börner, and Elisha Hardy, Indiana University, and André Skupin, San Diego State University, 2010. (Center right) Heat map of wireless connections. SOURCE: Sense Networks. (­ ottom B left) Synchronized time-series display. SOURCE: Hannes Reijner, Panopticon Software. (Bottom right) Debris objects in low Earth orbit. SOURCE: European Space Agency.

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 49 cialization in one field and teaching interdisciplinary InfoViz, Where2.0, and the Institute of Electrical and flexibility when dealing with big data (in his case for Elec­ ronics Engineers Symposium on Visual Analytics t bioinformatics). These skills are rarely available in one Science and Technology. person, so teaching and research in visual analytics is commonly carried out by groups of collaborative scholars FORECASTING with different disciplinary backgrounds. Forecasting is a technique that uses observations, Education and Professional Preparation Programs knowledge about the processes involved, and ana- lytical skills to anticipate outcomes, trends, or future Closest to a formal education in visual analytics are behaviors. Forecasts are related to predictions and interdisciplinary graduate and undergraduate programs a ­ nticipatory intelligence. In general, forecasts attempt that have evolved from communications, visual arts, to estimate a magnitude or value at a specific time media studies, geography, computer vision, and human- (such as 3-day forecast of temperature), whereas pre- computer interaction research. For example, at the dictions estimate what may happen and the odds of it University of California, Santa Barbara, it is possible to happening (such as predicting what fraction of people earn a Ph.D. in multimedia arts and technology while will develop skin cancer). Anticipatory intelligence ­ doing a considerable amount of coursework in visual combines computational methods (e.g., agent-based analytics. Universities that offer suites of graduate- modeling, system dynamics, Bayes network models) level classes in visual analytics include the University of with role playing and applications of game theory to North Carolina, Indiana University, the University generate integrated time-based simulations. of Washington, and the Georgia Institute of Technol- In the geospatial domain, forecasting needs to ogy (Table A.9 in Appendix A). The Georgia Institute a ­ ddress what, where, when, and how events will unfold of Technology also hosts an online library of materials and how processes will evolve in space and time. Geo- (e.g., videos, recorded lectures, sample exams) intended spatial events and processes are a result of interactions for use in higher education in visual analytics.9 among the natural and built environments as well as Methods used in visual analytics are often taught social and cultural systems across global, regional, and in discipline programs—such as information visualiza- local scales. tion, cartography, GIS, computer gaming, and com- puter graphics—although not as a central focus. Many Evolution 2-year and community colleges offer basic preparation in visual analytics through media technology, computer The ability to forecast future behavior is central programming and scripting, graphic design, imag- to many scientific disciplines. Among the first dis- ing and graphics, and human-computer interaction ciplines to embrace quantitative methods for fore- programs. casting were meteorology and economics. Weather Research and on-the-job training in visual analytics forecasts were made from data, charts, and maps until are also offered by online businesses, gaming compa- the late 1950s, when empirical methods began to be nies, and the open-source programming community. replaced by numerical weather forecasting (Lutgens Visual analytics research has bases at both the Pacific and ­ arbuck, 1986). Similarly, economic forecasts T Northwest National Laboratory and at the Oak Ridge transitioned from methods using stationary and deter- National Laboratory. Private companies involved ministic assumptions in the late 1960s (Khachaturov, in ­ isual analytics include Northrop Grumman and v 1971) to probabilistic or stochastic methods, then to Oculus, Inc., a Toronto-based company working on complex simulations of dynamic, adaptive economic the visual display of time-space tracks. The primary systems in 1990s and 2000s (Clements and Hendry, avenue for discussing visual analytics is national con- 1999; ­ asparikova, 2007). G ferences, most based in the United States, such as Recent advances in computational methods, econo- metrics, simulation, system dynamics, agent-based 9 modeling, and game theory have allowed forecasters See .

OCR for page 035
50 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE to generate a range of possibilities to support decision ample, nowcasting systems to project the development making or scenario-based planning. The International and dissipation of convective storms 2 hours ahead were Institute of Forecasters (IIF) was founded in 1981 tested during the 2008 Beijing Olympics (Wilson et al., to promote forecasting through multidisciplinary re- 2010). Nowcasting is considerably more challenging search, professional development, bridging theory and than forecasting. It is one thing to forecast population practice, and international collaboration among deci- growth of a city over the next year; it is quite another sion makers, forecasters, and researchers. A majority to nowcast the population distribution downtown for of its members are from the economics, business, and emergency evacuation. Nowcasting demands rapid as- statistics communities. IIF publishes two journals: the similation of massive amounts of data from multiple International Journal of Forecasting (a peer-reviewed sources into model runs; scientific understanding of academic journal started in 1985), and Foresight: The event evolution, the environment, and their interac- International Journal of Applied Forecasting (a journal tions; and the ability to deal with measurement errors, for practitioners, started in 2005). incomplete data, or uncertain information in real time. Advances in sensor technologies and the increas- Moreover, research shows that both computational ing availability and timeliness of information have models and human judgment are required to optimize opened new opportunities for forecasting. Forecasts are the nowcast (Monti, 2010). now being made in areas ranging from ecology (Luo Geospatial intelligence forecasting can play a key et al., 2011) to technology (NRC, 2010b) to sports role in informing a variety of decisions for military or (­ iannakis et al., 2006). The concept of nowcasting— Y security operations. Examples include determining op- forecasts of local events in near-real time—has emerged timal clothing based on weather forecasts (Morabito et for both physical and socioeconomic systems. For ex- al., 2011), strategic planning based on forecasts of politi- FIGURE 3.6  Example of geospatial intelligence forecasting. Data extracted from various sources, including structured data sets or unstructured texts, provide information about people, activities, and events. The information is analyzed using computer models to reveal the potential connections among people, activities, and/or events and to project possible future events. SOURCE: Reprinted by permission from Macmillan Publishers Ltd on behalf of Cancer Research UK: Web of War, Weinberger (2011).

OCR for page 035
EMERGING AREAS OF GEOSPATIAL INTELLIGENCE 51 cal instability (Goldstone et al., 2010) or other events tions brings the field a step closer to short-term and that may threaten liberal democracies (­Anderson, 2010), near-real-time forecasts of event progression, such as and anticipating social or political change through the spread of wildfire or disease, or of social dynamics, cyber-empowered political movements, social disrup- such as perception or activities planning. tions, or cultural conflicts (Bothos et al., 2010; Paris et al., 2010; Weinberger, 2011; Figure 3.6). However, rig- Education and Professional Preparation Programs orous methods for forecasting social patterns and social changes have not yet been fully developed. No university programs offer degrees in fore- casting, and many science-based or business-based Knowledge and Skills curricula emphasize modeling instead of forecasting. Courses in advanced methods for spatial and domain- Robust forecasting methods build on a solid specific processes are taught at senior undergraduate or understanding of the composition and structure of a graduate levels in a wide range of disciplines, includ- system and the embedded interactions among system ing statistics, computer science, information science, components and between the system and its environ- electrical engineering, civil engineering, meteorology, ment (Boretos, 2011). Geospatial forecasting requires geography, economics, ecology, criminology, epide- both deep domain knowledge and advanced skills miology, and urban and regional planning. Geospatial in spatiotemporal analysis, modeling, and synthesis. forecasting requires an integrative treatment of spatial Examples include regression statistics, spatial and tem- and temporal data and is still considered an advanced, poral interpolation techniques, space-time prisms and specialized area of research. The few advanced spatial trajectory models, cellular automata and agent-based modeling courses available are commonly tailored to modeling, artificial neural networks, evolutionary and the faculty’s research interest, rather than providing genetic algorithms, computer simulation and ensemble a comprehensive coverage of analytical and model- techniques, and scenario-based planning that antici- ing techniques. Examples of universities with strong pates multiple possibilities. programs in agent-based modeling include Carnegie Forecasts in the context of geospatial ­ntelligence i Mellon University, George Mason University, and the need to integrate both geospatial processes and ­domain University of Michigan (see Table A.10 in Appen- processes to reveal patterns, relationships, and mecha- dix A). The Massachusetts Institute of Technology has nisms that drive state changes. For example, activity- a strong program in system dynamics. based intelligence—the predictive analysis of the Time-series analysis is the foundation for fore- activity and transactions associated with an entity, casting, and relevant courses are commonly taught in population, or area of interest—depends on an under- meteorology, geography, geology, ecology, economics, standing of environmental, social, and cultural factors; political science, and other departments that emphasize individual space-time behaviors; and the spatiosocial modeling and projections. Students learn how to detect processes that move and regulate activities of groups temporal trends and to project them into the future and the society. using techniques such as harmonic analysis, wavelet New methods and analytical tools emerging from analysis, and historical event modeling. Examples of the computational social sciences are changing the programs that offer courses in these areas include the education and skills needed for geospatial intelligence University of Oklahoma and the University of Wash- forecasting. For example, new approaches are being ington (meteorology); the University of California, developed to address the validation and calibration Santa Barbara, and the State University of New York challenges of agent-based and other complex systems at Buffalo (geography); and Harvard University and m ­ odels. Tools such as the Integrated Crisis Early Princeton University (economics and political science; Warning System have been developed to predict see Table A.10 in Appendix A). political events such as insurgency, civil war, coups, Space presents another important dimension or invasion. The increase in volunteered geographic of forecasts. In human geography, spatial diffusion information and geotagged images or communica- theory, central place theory, and time geography offer

OCR for page 035
52 FUTURE U.S. WORKFORCE FOR GEOSPATIAL INTELLIGENCE both conceptual and mathematical bases for spatial Utah; the University of Maryland; and Ohio State prediction, such as spatial interpolation, spatial gravity University. modeling, spatial regression, and spatial optimization. Some community colleges or technology centers These traditional analog and mathematical modeling (e.g., GeoTech Center) offer basic statistics courses or techniques are commonly taught in geography, geology, computer modeling tools (such as STELLA), which epidemiology, criminology, civil engineering, trans- can provide foundation training for beginners. Oppor­ portation science, urban and regional planning, and tunities for professional training in forecasting are landscape architecture departments. A few universities limited. Workshops or summer schools, such as those offer advanced geocomputational methods for spatial offered by the Spatial Perspective to Advance Cur- prediction, such as Monte Carlo simulation, Markov ricular Education program,10 the Center for Spatially chain modeling, cellular automata, agent-based mod- Integrated Social Science,11 and the University of eling, geographically weighted regression, spatial self- Michigan, are perhaps the main form of training for organizing maps, spatial trajectory modeling, spatial advanced space-time methods or geocomputational niche modeling, spatial Bayesian statistics, and spatial techniques. Many of these workshops cover only the econometrics. Example universities offering courses in fundamentals. For economics and business, the IIF the spatial aspects of forecasting include Arizona State frequently offers training workshops for practitioners University; Clark University; the University of Texas, at their conferences. Dallas; San Diego State University; the University of 10 See . 11 See .