From Layers to Objects: Evolving the GEOINT Analytic Tradecraft
By Todd S. Bacastow, Ph.D; Dennis Bellafiore, Ph.D.; Susan Coster; Stephen Handwerk; Lisa Spuria; and Gregory Thomas, Ph.D.
This article was originally published in USGIF’s State & Future of GEOINT Report 2017. Download the full report here.
This article addresses the tradecraft implications of moving from layer-based geospatial intelligence (GEOINT) to object-based GEOINT.[1] The terms “layer-based GEOINT” and “object-based GEOINT” are used to capture a way of thinking (i.e., a paradigm, including tradecraft, analytic methods, and technologies). The focus is how analysts routinely think about solving a problem. In such paradigm thinking,[2] actions are taken according to the dominant perspective, which may not be reflected wholly in the documented tradecraft. As a result, the needed changes in tradecraft, training, and thinking may be a struggle for the GEOINT Community.
The community is evolving toward the object-based GEOINT paradigm because of the emerging array of object technology, data, and information. Significantly, object-based GEOINT is not any particular tool or a technology, but is enabled by developments such as activity-based intelligence (ABI) and object-based production (OBP). OBP is the technology and production capability that creates a conceptual “object” for people, places, and things. The object becomes the single point of convergence for all information and intelligence produced. Objects also become the launching point to discover information and develop intelligence.
Objects and object-based analysis are not new. What is different is the imperative to move the GEOINT tradecraft beyond the prevailing layer-based thinking. Object-based analysis enables analysts to more accurately model the real world in the way humans naturally interact with it. There are many benefits of object-based GEOINT, but the tradecraft must adapt.
Intelligence analysts have observed objects for more than a century. For any given object, the who, what, when, where, and why have been collected, analyzed, recorded, and reported. An object received a unique identifier, and attributes were assigned to the object. As automation was introduced into the workplace, the data were captured in layers, or spreadsheets, where each row was a discrete object, and the columns identified the attributes of the objects. These are known as relational databases. An inherent shortcoming of this approach is the difficulty to see how different objects might be related to one another as well as how time might affect these relationships. In the relational database realm, analysts must use cognitive abilities to study object relationships over time — a difficult and time-consuming undertaking. Further, only a limited set of objects and data can be comprehended. In the mid-’80s, programmers developed algorithms for graph-structured objects and CAD applications, and the term object-oriented database was created.[3]
Layer-Based Analysis
Geospatial education and analytic thinking are heavily influenced by an early method of environmental planning in which Earth data were graphically displayed on Mylar sheets assembled in various combinations to determine areas of environmental constraint.[4] Educators often cite Ian McHarg’s 1969 book, Design with Nature, as having influenced the basic overlay concepts that later developed into geographic information systems (GIS). GEOINT’s use of layers is rooted with environmental planning’s use of transparent acetate map overlays as implemented in the Intelligence Preparation of the Battlefield (IPB) process.[5] GIS allowed direct automation of the IPB process, through which layers are stored electronically rather than on Mylar sheets, and different layers are combined and computers calculate constraint areas.
A layer-based focus is still dominant in geography education in general and GEOINT in particular. The layer-based analysis examines the differences between the maps or layers. Location is the primary construct for analysis when using layers of data. Layers, whether Mylar or electronic, offer an orderly but fixed-in-time means to think about the Earth and the relationship among features. Critically, layer-based GEOINT analysis begins with heavily pre-filtered observations and proceeds to a conclusion in light of the generalized evidence.
Object-Based Analysis
Introducing object-based analysis brings actor, time, location, and action together as the analytic framework. In object-based GEOINT, the analyst’s methodology changes from looking at differences between layers to examining changes of many small things in space and time. Understanding these changes exposes forms, patterns, functions, and diffusion — it is a discovery paradigm. Technologic advancements beyond organizing data into layers allow analysts to handle the smaller elements that make up a layer. These elements are objects. Objects have been likened to creating electronic “baseball cards” for individual items.[6] These objects include attributes such as location, time, extent, etc. The objects can be related, indexed, searched, and updated with respect to location and time or to another set of objects. The analyst can then construct a narrative, such as a hypothesis, to better understand activities and events. Seemingly isolated activities can be cast as a narrative exposing interwoven threads. Objective-based GEOINT is deductive reasoning in which the analyst begins with the assertion and proceeds to a specific conclusion. The object framework is in n-dimensional space rather than the two-dimensional space of geospatial layers.
Struggling to Change
Technological changes create anxieties for analysts and managers. Such is the case with changing from a layer-based analytic framework to one that is object-based. There are two key problem areas that contribute to the anxiety: tool mismatch and data complexity that increases cognitive loading.
First, technologies and tools are never perfectly aligned with the work they are intended to support. The tools can try to be a “one size fits all” solution, but the competent professional ultimately tailors either the tool or its outputs/inputs to compensate for any mismatch in tool interface to data availability, formats, or desired work. This is the case for existing tools in the analyst’s toolbox — they don’t interface well with object-oriented data structures and work. These tools were originally created to support traditional, layer-based GIS methodologies and tradecraft. The work implicit in these tools revolved around layer manipulation techniques and structures that had evolved through necessity from limited computer memory and processing power resources. A lot of the problem-solving “heavy lifting” was done by professional reasoning and the tools supported the creation of a standardized output product format. Instead, the tools need to evolve and provide logical support during the actual problem analysis.
Analysis was provided on a backdrop of a fused map, image, and occasionally other data layers as annotations or text. The tools provided a context enabler for the analyst to tell the intelligence narrative; they didn’t usually provide decision process maps or technical subject matter expert experience that informed an analyst’s decision-making process. The “heavy lifting” of logical, multidimensional problem-solving that took place in an analyst’s head was actually object-based thinking techniques not yet supported by existing data storage structures and tools. Technology continues to limit this type of analysis; however, new 3D and immersive screen displays may pave the way for real change. To update traditional analysis to object-based analysis will require technologies and tools be developed that include the inputs of new object data structures and provide additional problem-solving frameworks that can support analysis algorithms and ground truth models to compare and contrast against.
Second, object-based analysis derives its power from the creation of a different type of data manipulation. Historically, one of the main contributing reasons for the data layer-based approach has been limited computing power available for desktop manipulation of data. Data were often filtered early in the workflow via accepted GIS data combination techniques and tradecraft to simplify the problem space. The data analyst worked to minimize excess data, merging multiple hypotheses down to minimal complexity levels so an analyst could then identify a narrative, working hypothesis, or model that was logically supported by the remaining data. This was necessary as the human brain has limitations on working memory and maximum cognitive load.[7] Data were “lost” during this filtering process in the sense that there was no longer a need to revisit it for new insights once a plausible narrative was determined. The data layers were difficult to change once they were established and to manipulate once created. They were, in essence, “static” data forms.
One can compare the pre-filtered approach of GIS layers to the incremental-build approach of object-based GEOINT workflows. As the workflow unfolds, the pre-filtered approach deals with less and less unfiltered data, and the incremental-build approach creates alternative models of the space-time activities and can accommodate more and more data.
The comparison between the pre-filtered and incremental-build approach can be summarized as follows:
1. The starting points are different:
a. The pre-filtered approach interprets the information selected as an accurate record of reality and defines this as data.
b. The incremental-build approach captures all data inputs and the algorithms used to capture the information and store all of this as data.
2. The filtering of data is different:
a. The pre-filtered approach uses relational database pre-defined categories and layers.
b. The incremental-build approach uses the first objects to filter and define the space-time activities, but all data is kept as the data set grows.
3. The author’s methods are different:
a. The pre-filtered approach employs expected mental models and arrives at a best fit.
b. The incremental-build approach uses multiple hypotheses and performs iterative analysis techniques to create more hypotheses and better and better fits.
4. The intent of the interpretation is different:
a. The pre-filtered approach follows a narrative structure that aggregates the interpretation, shapes understanding, and best conveys the intent.
b. The incremental-build approach dynamically creates the interpretation of the intent as data is added and the complexity of objects grow.
5. The map design is different:
a. The pre-filtered approach is limited by the foundational layer and the tools used for presentation and visualization.
b. The incremental-build approach interprets more complex objects as data are added and utilizes the presentation and visualization that are best for the analyst and potentially the decision-maker.
6. The analysis provides a different perspective:
a. The pre-filtered approach produces a static report that balances the biases and interpretation of the analyst and relies on the decision-maker’s understanding. This operates much like a fixed-focus lens — parts of the report are in focus while other parts are blurred.
b. The incremental-build approach tells a story over time utilizing the network of dynamic objects thus able to provide ongoing analysis as more data are added to the model. This operates much like a variable-focus lens — all parts of the report are in focus.
In contrast, the object-based model seeks to retain all data and iteratively restudy it to forge new connections between evolving objects. A GEOINT object can be defined as a digital data construct that holds intelligence data points that are relatable by a shared geographic location on Earth or in space. These data points, associated by a shared geographic location, do not need to be in the same data formats nor at the same points in a timeline. An object is allowed to “evolve” over time as new data are added and the analyst discovers the object has relationships with other objects. These object relationships are stored as digital tags in the database. As the object relationships are discovered and expand over time, networks of objects are created.
The ultimate power of an object is the data within the object are digitally tagged as “related” to each other and can therefore be manipulated at a more detailed level by advancing computer technologies. Data no longer needs to be pre-filtered and reduced to layer states for exploitation as a result of low computational power. Objects can exist either as unconnected containers of location-based information or as part of a larger data network of objects and relationships created either with or without filtering and selection criteria. This available computer-enabled option to filter or select criteria-based data first is dependent on the type of problem the analyst is trying to solve — at times there are known selection criteria (e.g.,, tanks usually stay on land); other times (especially with respect to big data analysis), the intent is to search for previously unknown and new patterns (e.g., financial trends, statistical traffic information via Google, or unconstrained growth in a market).
Both unconnected objects and object network types of information can promote tipping and queuing, a starting point for pattern or trend recognition (layers usually don’t carry the change over time component) or prediction modeling. The exploitation options therefore are enhanced in the object paradigm. The computational power available today allows the data to be stored indefinitely and re-evaluated against different scenarios of interest — even scenarios that occur over extended periods of time (known as activity-based intelligence or ABI). Algorithms that will be developed in object-based GEOINT analysis will take this tagged data and search for known pattern matches. They will study object relationships that have been built in the database and compare with ground-truthed intelligence information.
Multiple intelligence hypotheses will be carried within computer memory simultaneously as they are investigated with no limitations caused by human cognitive constraints. As new data are added to the database, some scenarios will be proven false and others will be supported and become part of a static snapshot in the intelligence report. The ability to save and manipulate data in this fashion will enable analysts to more efficiently investigate multiple narratives, to share objects with other analysts, to build known pattern libraries over time to compare against, to examine intelligence patterns and activities over larger time periods, and to revisit earlier reports and re-analyze intelligence situations. This object-oriented data structure allows data to grow and change over time, and the motivating power for change is data knowledge builds on itself and can be shared with the entire analysis community via the digital cloud framework currently under construction.
Creating Change in Tradecraft, Training, and Thinking
In addition to understanding the object construct’s motivating power for change, it is also significant to appreciate how technology has driven work structures, workflows, and tradecraft. Technology has a significant and often unappreciated influence on the workforce’s belief of how work should be done. These beliefs can linger long after the introduction of new technology or work structures. This is the case with respect to the concept of “layers” in a GIS, the key technology and approach for today’s GEOINT practitioner.
The tradecraft and learning implications of object-based analysis are far-reaching. Object-based analysis is an integrating methodology that promotes “making sense” of relations among different components of human behavior. The expert analyst creates models, which help to make the connection. Models are frameworks to help understand the nature of the problem, derive potential solutions, and anticipate constraints. Technology aids in the evaluation of the models. Object-based analysis aids the sense-making process by filling the gap between the conceptual and the logical levels. This approach involves observing, recording, and acting upon the world at small levels of detail. Models are built of objects and logically put together. Analysts assemble the world as they see it in space and time. Layer-based analysis is focused on planned-problem-solving. Planned-problem-solving relies on well-understood conceptual models and predefined logical map layers. “Predefined” poses challenges when applied to messy, real-world problems for which the conceptual and logical models cannot be completely predefined.
The implications for GEOINT training and education are significant. Cognitive research has identified “dimensions of difficulty” that require increased expertise.[8] The dimensions are situations requiring increased mental effort on the part of learners. A few of the key dimensions in object-based GEOINT are:
· Static map layers versus dynamic objects that are not organically organized as a map
· Discrete data represented as a static map versus continuously changing discrete objects
· Discrete organized data themes that are fixed at a time versus interactive data elements that can be constantly organized
· Homogeneous layers of an area of interest versus heterogeneous objects that comprise an area of interest
The “dimensions of difficulty” mirror the differences between analyzing layers of data versus objects. Object-based GEOINT analysis will require the analyst to meet the challenge of reasoning from objects, seeking and finding cues of actions within and across individual objects and time. Technology enablers will be developed that will aid the analyst, but this will take time and money and doesn’t replace the fundamental shift in problem-solving culture that will be necessary to solve the more complex, multi-scenario problems facing the Intelligence Community.
There are many differences between layer-based and object-based GEOINT analysis. The following list of attributes demonstrates the differences:
1. Base Data Organizational Structure
a. Layer-based is grounded in relational database management systems (RDBMS) based on the relational model invented by Edgar F. Codd at IBM in 1970.
b. Object-based is grounded in object-oriented database management systems (OODBMS) that support the modeling and creating of data as objects created by Michael Stonebraker and Lawrence A. Row — later defined by Malcolm Atkinson in 1985. All conceptual entities are modeled as objects.
2. Supporting Technology
a. Layer-based GEOINT technologies include electronic light tables, GIS software such as ArcGIS, and a number of specialty tools.
b. Object-based GEOINT technologies include OODBMS, big data technologies (search and match algorithms), technologies that data search, tag, and manipulate subsections of data that meet specific criteria (time window, location buffer on ground, etc.). Tags, once applied, are saved back into the object structures for future use.
3. Flexibility
a. Layer-based GEOINT placed in a standard layer format can be “stacked” in a viewer. It is usually a time static data view unless the layer is continuously updated and recreated.
b. Object-based GEOINT databases allow dynamic new object data updates and either snapshots in time or additional ongoing pattern matching, sorting, and new relationships building and saving data sifting via background algorithms that could run autonomously. Models can be formulated (object networks) and used as basis for hypotheses — “What if?” scenarios. As data is added, it will attempt to be “related” to existing models and objects, and if it doesn’t match anywhere, it’s not “thrown away” but saved for later comparisons.
4. Data Complexity
a. Layer-based GEOINT layers are usually “like-spec’d” types of data categorized together.
b. Object-based GEOINT objects are multi-INT, multi-spec data structures tied relationally via location on the ground. If two objects are “time coincident” at the same place on the ground, they are “suspect” for a relationship. Due to the multiple types of data that can be related, these objects contain lots of complex data types, however, the base object concept retains its relational simplicity. Standards for objects are being created at this time.
5. Time Orientation
a. Layer-based GEOINT layers usually contain data that is within certain windows of time to keep them relevant to each other for a report.
b. Object-based GEOINT objects are intended to grow over time and gain more and more data and relationships. Technology enablers allow analysts to take a “time slice” across the data objects in a geographic area or to follow objects across time (ABI) to tell a story or narrative. It’s up to the reporter to pick the “view” of the data to present.
6. Data Tagging/Indexing
a. Layer-based GEOINT standard spec’d data contains metadata fields that can be searched by tools, and most imagery can be thumbnail represented in JPEG format.
b. Object-based GEOINT object data structures can grow and expand for any new types of spec’d data “related” by space/time to the existing object data. The key to the object is the word “relationship.” Is the data linked in some way? It doesn’t have to be spec’d the same in its raw input format — the key is to retain the fact that they are related, the detailed analysis within the data can happen later.
7. Extensibility for New Sensors
a. Layer-based GEOINT requires either tool updates to handle the new spec or new tools that know how to manipulate and display the data format when developing new layers for new sensors.
b. Object-based GEOINT tagging means any new sensor type can be related within an object structure. Calibrating the data for internal object comparison and fusion is still a tool-based task that requires some tool specialization, but, in the framework of the cloud and services, new algorithms may “plug and play” for situations such as these.
8. Hard Problems
a. Layer-based GEOINT layers have limits on how well they could present relationships in the data. Many hard problems had to be solved in the analyst’s headspace and a textual dialogue presented in the report. These reports were difficult to relate to any earlier or later reports (tagging not very effective), so the continuity of time was usually hard to maintain.
b. Object-based GEOINT data is tagged and organized within a computer database structure so it can constantly (as need) be re-sifted digitally and new patterns or trends sought after — aka models matched. This presents huge benefits to the analysts as they don’t lose previous history or information, they continue to add to the data in the relational model, and they can tag related data forward and backward in time. Taking different “views” of the data can present new insights that would not have been realized in the past.
9. Analyst Thinking Style
a. Layer-based GEOINT layers usually mean sifting “down” through the data. “Like go with like” types before the analyst can find a narrative that “fits” what is left. This leaves open room for biases when it comes to constructing narratives.
b. Object-based GEOINT objects retain “all” the data attached to them. They create relational links that can be manipulated across space and time and added to as new data appears. This can be a more complex 3D or 4D type of thinking than layers as the analyst has to remain cognizant of carrying multiple threads or hypotheses at once until some or all are ruled out — either due to newly discovered relationships that don’t fit the models or new data that disproves a hypothesis. New hypotheses can always be created if the original ones don’t work, as analysts keep the original data relationships intact in the database and can iteratively revisit the data with new thinking paradigms and models.
Conclusions
There is an imperative to move the GEOINT tradecraft beyond layer-based thinking. Layers of data will continue to exist, but an evolutionary change to object-based analysis is unavoidable. Object-based GEOINT provides richness not possible in the layer-based paradigm and enables the analyst to more accurately model the real world. No longer constrained by layers, analysts will more readily handle the complexities of today’s problems.
The conceptual move from layers, the larger volumes of intelligence data, and the new analytic options will challenge the analyst. Anticipating this change is important. The geospatial analyst needs education and training to move to the object paradigm. The academic community must prepare GIS and GEOINT students to examine changes in objects in space and time. Students must learn how to identify forms, patterns, functions, and the diffusion of effects. Most importantly, the GEOINT Community needs to prepare analysts to cope with the increased complexity.
[1] We specifically do not use the abbreviation OBG to avoid confusion with established programs such as object-based production (OBP).
[2] J.K. Swindler, review of Simplicity: A Meta-Metaphysics by Craig Dilworth. The Review of Metaphysics, 68, no. 3 (2015), 649.
[3] T. Atwood, “An Object-Oriented DBMS for Design Support Applications.” Proceedings of the IEEE COMPINT 85 (September 1985), 299–307; N. Derrett, W. Kent, and P. Lyngbaek, “Some Aspects of Operations in an Object-Oriented Database.” Database Engineering, 8, no. 4 (December 1985), IEEE Computer Society; D. Maier, A. Otis, and A. Purdy, “Object-Oriented Database Development at Servio Logic.” Database Engineering, 18, no.4 (December 1985).
[4] Ian McHarg, Design with Nature. New York: Doubleday, 1969.
[5] R. Glinton, J. Giampapa, S. Owens, K. Sycara, C. Grindle, and M. Lewis, (2004). “Integrating Context for Information Fusion: Automating Intelligence Preparation of the Battlefield.” Proceedings of the 5th Conference on Human Performance, Situation Awareness, and Automation Technology, iAA, 224.
[6] Biltgen, P., and Ryan, S. (2016) Activity-Based Intelligence: Principles and Applications, Boston, MA, Artech House, p. 154.
[7] W. Huang, P. Eades, and S. Hong, “Measuring Effectiveness of Graph Visualizations: A Cognitive Load Perspective.” Information Visualization, 8(3), 139–152.
[8] R.R. Hoffman et al. (2010) Accelerated Proficiency and Facilitated Retention: Recommendations Based on An Integration of Research and Findings from a Working Meeting, Air Force Materiel Command, Report AFRL-RH-AZ-TR-2011–0001 p. 40.
