Modeling Outcome-Based Geospatial Intelligence

The application of GEOINT models enables decision-making in myriad sectors


Society has tremendous capabilities to prepare, respond, and recover following natural disasters, emergency events, and security incidents—but these capabilities each require time, space, and energy to mobilize and focus. Geospatial intelligence (GEOINT) is a key element to engaging response capabilities in the right way, at the right time, and in the right place. In particular, the connection between GEOINT and modeling has emerged as a capability that decision-makers and response teams can rely upon to increase the correctness, reliability, and timeliness of their decisions.

Today, the most notable example of models in action occurs in the realm of hurricanes and severe weather. The impact of hurricanes on the United States and across the globe continues to be a challenging part of our natural hazard landscape. We experience not only direct threats to life and property from such storms but also increased second and third order effects. Important tools to focus action (while simultaneously garnering public attention) include the ubiquitous European and United States hurricane models that we are accustomed to seeing. They enable us to track and predict where a hurricane may travel and its potential impacts. Similarly, we subsequently rely on flood and inundation models to predict storm surge and intensity.

As GEOINT systems collect additional data and decision-makers are exposed to more complex and useful models, the demand for modeled outputs in a variety of applications will likely grow. They present data from spatial and dynamic sources as a unique combination of time and space in a format that decision-makers, the public, first responders, non-governmental organizations, and recovery experts all use to enhance and support their actions. This reflects a shift from simple data collection and display to a world where we ask and expect an answer to the question: “What does it all mean?”

The Integration of GEOINT and Mathematical Modeling

From a GEOINT perspective, the output of modeling can be used to analyze the projected path of a hurricane or wildland fire, depict and assess the impact of the event and on structures within that path, and prioritize evacuations. The objective of modeling is to create a simulation of real-world events or conditions. By using modeling to focus on real-world outcomes such as population shifts or emergency services response times, the GEOINT Community is beginning to transform geospatial data from discrete data elements into the direct language of decision-makers and operations personnel.

While many think of modeling in terms of weather/hurricane models and Monte Carlo simulations, it is the application of those models into predictive situations that improve accuracy and the available time for decision-making. Delivering outcome-based modeling as a product of GEOINT allows policy-makers to more rapidly assess both risks and opportunities. Models allow us to investigate complex things by applying our knowledge of simpler things. Once a model is proven consistent with supporting evidence and therefore accepted, it can be confidently used to make reliable predictions.

“What if?” analysis is one of the most recognizable modeling outputs. Often called “predictive analytics,” these models encompass a variety of modeling and statistical techniques ranging from machine learning (ML) to linear regression to multivariate analysis. These models allow policy-makers to assess various decisions, predict uncertainty, and assess the impacts of one attribute versus another. They present insights into what may happen based on trends, combinations of data and patterns, or rule-based behaviors. Predictive models generally focus on outcomes rather than numbers. Decision-makers are far more likely to be asked when a river will reach flood stage, and what to do about it, than what the reading is on a particular stream gauge. The goal is to answer not only what may happen in the environment, but also what would be the impact of various decisions on the outcome.

Applying Models to Decision-Making 

The majority of models are developed in academia or other technically sophisticated environments. Transitioning them from expert users to emergency service practitioners poses a variety of challenges. The development and transition of models such as Hazus or the European and U.S. hurricane models into common tools used by responders requires consistent mentorship and stewardship by both the academic community and the targeted user community, typically including a centralized authority such as the Federal Emergency Management Agency (FEMA).

As demand for actionable GEOINT information increases, academia and GEOINT producers need to focus on ways to speed up the long-term adoption process (which exceeds five to eight years in some cases) that typical models take to transition from science into practice. One example of a recent effort to streamline this adoption cycle is the efforts of the State of Colorado to integrate a fire behavior model directly into fire operation. The Colorado Center of Excellence for Advanced Technology Aerial Firefighting (CoE) recently entered into a partnership with the National Center for Atmospheric Research (NCAR) to transition a weather-based wildland fire predictive model to operational use. As part of the project, the CoE not only provided training material to firefighters, but also engineered a separate training module to allow for multivariant simulations based on specific geographic and meteorological case studies. These enhancements improved the usability of the modeling system for non-experts and hastened potential adoption by the wildland fire community.

  • This article is part of USGIF’s 2018 State & Future of GEOINT Report. Download the PDF to view the report in its entirety. 

Current Uses of Modeling 

Emergency Services Delivery: Increasingly, fire and EMS leaders are applying quantitative standards, confronting complex problems, and seeking innovative deployment solutions. There is growing demand within fire and emergency medical services (EMS) communities for decision support systems that can answer “What if?” questions and allow for deployment planning based on future incident predictions. This confluence of factors is spurring the development of the kinds of hybrid GEOINT solutions described above.

A number of cities around the country are using hybrid modeling, geospatial data, and ML techniques to evaluate performance and determine efficiencies. For example, the city of Palo Alto recently faced a daunting problem: optimizing paramedic unit placement, scheduling, and staffing to optimize five competing performance metrics within budget and physical constraints. A unique software solution that paired ML algorithms with discrete simulation and geospatial-temporal data identified eight near-optimal solutions out of millions of alternatives for command staff to evaluate. Nearby Redwood City addressed the problem of unprecedented growth by utilizing a hybrid system that combined a machine learner with geospatial-temporal analysis tools to generate an accurate, validated model of future incident profiles. This model generated future scenarios using discrete event simulation, enabling command staff to assess key performance metrics such as response time, utilization, budget impact, and system resiliency on both reasonably anticipated and extreme versions of future scenarios.

Fire Behavior: Fire behavior modeling projects the behavior and effects of fire activity to inform and guide prevention measures, response tactics, resource management, and safety decisions across all levels of the fire service—from wildfire to structure fire, from initial attack to sustained attack, and throughout the cycle of land and forest management. As a fire response increases in complexity, wildfire modeling is used for tactical planning, the assessment of future resource requirements, and evacuation planning.

Wildfire behavior models range from empirical fire spread estimation to physical models. They combine geospatial data (slope, fuel/land cover data) and dynamic data (weather measurements) to provide specific decision products including fire spread, intensity, and behavior. The State of Colorado is currently in the second year of a multiyear effort to deploy a wildland fire prediction system based on improved weather data provided by the High-Resolution Rapid Refresh (HRRR) model developed by the National Oceanic and Atmospheric Administration (NOAA). Rather than relying on a single point forecast, as was the case in previous generations of fire models, the HRRR model provides a gridded forecast at three-kilometer spatial resolution across the state. The HRRR model creates a new 18-hour forecast every hour and uses radar data to model the propagation of existing storms. The fire prediction system creates a hyper-local forecast over 36 square kilometers at the location of the fire. This forecast is suitable for operational and tactical decisions of ground personnel based on micro-terrain and local winds.

In Southern California, a system called Wildfire Analyst simulates 10 million fires daily to evaluate the potential impact of wildfires on values at risk and utility lines, using a two-mile resolution local model and high-resolution five-meter fuels analysis. This massive analysis allows local utility managers to prepare in advance to address the potential impact of wildfires up to three days in advance. Similar systems are used in Chile by Corporación Nacional Forestal (CONAF), which uses real-time observations to calibrate modeling of large-scale incidents such as the Las Maquinas Fire in January 2017. CONAF uses the modeling not only to support incidents but also to communicate to the public what is intact after every big fire, comparing the resulting burned area versus simulated fire progressions that would have occurred without suppression efforts.

Land Use: Land use change modeling projects historic patterns into the future and visualizes alternative futures as a tool for decision-making by local or regional government officials. This type of modeling helps reveal the causes, mechanisms, and consequences of land use dynamics by modeling the interaction in space and time between humans and the environment. GEOINT data in the form of satellite images and maps play a key role.

One example of a land use change model is SLEUTH, which stands for Slope, Land Cover, Exclusion, Urban, Transportation, and Hill Shade—the input data to the model. For more than 20 years SLEUTH has been used in 18 countries to study land use change. The impacts to urban growth can also be examined. One such study assessed declining water quality in the Chesapeake Bay estuary due in part to disruptions in the hydrological system caused by urban and suburban development. Land use change models will continue to be used to study the complex interactions of urban dynamics and can be used by local and regional governments to inform policy decisions.

Watershed: Providing water for human and ecological needs remains a challenge for local and regional government officials worldwide. As populations grow and demand for water increases, land and water resource management is evolving from simple, local-scale problems toward complex, regional ones. Such problems can be addressed with models that can compute runoff and erosion at different spatial and temporal scales. In 2002, the U.S. Environmental Protection Agency, U.S. Department of Agriculture, the University of Arizona, and the University of Wyoming first developed an automated, GIS-based watershed modeling tool. Now under continual development, the Automated Geospatial Watershed Assessment (AGWA) helps decision-makers manage and analyze water quantity and quality. AGWA utilizes the Kinematic Runoff and Erosion (KINEROS2) hydrologic model and the Soil and Water Assessment Tool (SWAT) to evaluate watersheds with varying soils, land uses and management conditions, and their related environmental and economic impact. AGWA has also been used to analyze land impacts of coalbed methane extraction, management of impacts from military training activities, and the evaluation of flow in streams on military bases in the southwestern U.S.

Sea-Level Rise: The effects of rising sea levels range from large-scale population displacement to critical infrastructure degradation due to saltwater intrusion. Coastal erosion is evident in many areas of the world, in the U.S. notably along the Louisiana coast. In 2014, the U.S. Geological Survey and the University of San Francisco published a new marsh accretion model, WARMER, to assess the risk of sea-level rise to salt marsh parcels around San Francisco Bay. The aim of this model is to provide site-specific sea-level rise predictions to land managers through the intensive collection of field data and innovative predictive modeling. WARMER indicates that most salt marsh around San Francisco Bay will transition from high to mid marsh by 2040, to low marsh by 2060, and to mudflat by 2080, however, there is a great deal of variation around the bay.

What Makes a Good GEOINT Model?

A scientific model must not only generate predictions, but also generate results that are used and accepted by decision-makers. As observers of the natural world, we will only accept a model if its predictions stand up against outcomes we can observe. Although specific fields and disciplines may accept and use models with varying attributes, models that can be integrated as a GEOINT product must share some specific attributes. Here are some characteristics of good models:

Output that is linked to decision or analytical objectives: Models that support specific identifiable decisions or outputs support the entire GEOINT cycle. In this sense, they are “products” and must be aligned to an information need or decision point. They are the automated counterpart of manual analysis and reports.

Consistent, identifiable, and available data: The data that support a model can turn a good model into one that is inconsistent or irrelevant. GEOINT models are used to support decisions at all levels, from analysts to the public, and at frequencies that range from one to three times to common, automated runs. As a result, they should be aligned to consistent, accurate, and standardized geospatial data for which an analyst or automated system has a reasonable expectation of availability.

Ability to assess and compare the impact of inputs: The sensitivity of a model to changes and variation in input data is directly linked to decision-maker understanding, trust, and adoption. If a model appears to have wide swings in output based on small changes to inputs, it can limit the adoption and trust of outputs by decision-makers.

Consistent outputs: A model must produce output that is consistent with inputs. Although one of the benefits of modern modeling is that it goes beyond pattern assessment, decision processes and trust in systems begin to wane when similar inputs result in different outcomes.

Ability to assimilate real-time observations: Operational models need to provide answers in near real time and respond to the influx of massive amounts of data that can be captured from operators, citizens, and the wealth of sensors available through the Internet of Things (IoT) and remote platforms.

Produce results for advanced visualization platforms: Traditional GEOINT results in maps, a tool for specialists to be used in planning and operations. But our audience (GEOINT professionals, decision-makers, the public) demands easy-to-understand information in a variety of formats (e.g., 360-degree videos, 4D immersive environments, augmented and virtual reality).


Better and expanded application of modeling as a GEOINT product has the potential to enhance and focus the work of traditional analysis. To expedite the adoption and improve the relevance of modeling, the GEOINT Community should begin to revamp and refocus modeling advances on customer needs. The low adoption rate of modeling as a GEOINT product for some of the examples cited in this article can be refocused by addressing the core attributes listed above. Modeling should produce immediate, consumable results. Simultaneously, models need modernization to take advantage of new practices and approaches as well as new and improved data sources. Modeling needs to decrease reliance on hard-to-find or outdated information, such as inconsistent, manually collected land cover data, and transition to the application of methods, such as automatically extracted data from satellite remote sensing systems.

Policy-makers must be able to depend upon a reliable, integrated, and continuously improving GEOINT framework to address the increasing challenges. Our future GEOINT framework should expand to include modeling tools as well as more data, processes, and visualization as we strive to support today’s decision-makers.

Headline Image: The Automated Geospatial Watershed Assessment (AGWA) tool is an automated, GIS-based watershed modeling tool that uses nationally available GIS data layers to parameterize, execute, and visualize results from the RHEM, KINEROS2, KINEROS-OPUS, SWAT2000, and SWAT2005 watershed runoff and erosion models. 

Data On Demand

The cloud and the future of geospatial data

Time-Dominant Fusion in a Complex World

Defining time-dominant fusion and its interdependent relationship with airborne ISR capabilities and Air Force DCGS

Ground Control Points

The importance of accurate, standardized GCP collection