Written June 2020, revised 2026.
This essay was written four months into the COVID-19 pandemic, drawing on the 2001 Dark Winter exercise. Its structure — medical surveillance, violence and terrorism, thermal and atmospheric anomalies, natural disasters, transportation, public sentiment — is the same set of threat vectors DeepWatch now covers.
“We’d have been much more comfortable with a terrorist bombing,” Nunn later said in congressional testimony. Former Senator Samuel Nunn was testifying on the terrifying results of a senior-level exercise that simulated a biological weapons attack — an outbreak of deadly smallpox — on the United States: Dark Winter. The simulation was conducted to gauge how senior State and Federal leaders would respond to a biological threat posed by smallpox. Smallpox’s profile, extremely contagious, no cure, and an extremely high historical death rate, made it an ideal candidate to support a simulation designed to show potential stress fractures in Federal and State emergency response to a national epidemic. In many ways, Dark Winter seemed to be a dress rehearsal for the difficulties faced with Federal and State response to what the New England Journal of Medicine calls a “Once-in-a-Century Pandemic”: COVID-19.
To be fair, the Dark Winter simulation has some significant differences from our national pandemic response to COVID-19, however, from an emergency response perspective, there is more that ties these cases together than separates them. The participants of Dark Winter testified to the lack of data-driven expertise when it comes to understanding the spread of the disease, public sentiment associated with government response, and the complete breakdown of analytics associated with surge capacity and supply chain. In these ways, the COVID-19 national response bears an eerie resemblance to Dark Winter. Many of the difficulties faced by government could have been in some respects resolved quicker, and potentially avoided if the data and analytics to support situational awareness were addressed.
Dress Rehearsal
On June 22–23, 2001, the Center for Strategic and International Studies, the Johns Hopkins Center for Civilian Biodefense Studies, the ANSER Institute for Homeland Security, and the Oklahoma City National Memorial Institute for the Prevention of Terrorism hosted a senior-level war game examining the national security, intergovernmental, and information challenges of a biological attack on the American homeland. For the next 13 days John Hamre of CSIS and General Dennis Reimer of the MIPT bore witness to lengthy discussions, heated debates, and forced decisions focused on the public health response, lack of an adequate supply of vaccines, medical supplies and the roles and mandate of federal and state governments, civil liberties associated with quarantine and isolation, the role of DoD, and potential military responses to what was to become a national medical emergency.
In many ways, the lessons drawn from Dark Winter provided a preview of what state and local officials would later face: the unfamiliarity of governing officials with where to get good public health data and the containment options available to address them; a lack of situational awareness in the earliest moments of the crisis to include the location of vulnerable populations and the location of the supply chain necessary to support them. There is no doubt that 2020 was its own Dark Winter, not necessarily stemming from bioterrorism, however still consistent with a Global Pandemic Model. Also consistent were the tragic outcomes that came from the lack of data and analysis in the areas of transit, supply, and logistics. Without a clear picture of where supplies were located, transit routes for logistics and relief operation in areas under quarantine, emergency management options were severely limited and prevented an effective response strategy.
Preparing for the Next Storm
A crisis will always lead to disruption. Unavailable essential supplies, contractions in spending for important projects, disruptions to supply chains and supply routes, and reduced availability of critical employees will create a rough sea for all boats. There will be wrecks, and every enterprise should be prepared for them. However, greater situational awareness can reduce the effort spent on analyzing on-going emergencies, distributing correct information and prioritizing the lessons learned from Dark Winter.
Understanding the data that’s publicly available reduces the effort often associated with the strategic understanding of situational threats, and the course of action necessary to mitigate them. Organizing analytics across medical, thermal, natural-disaster and public-sentiment vectors provides an opportunity to source, analyze and leverage data in the hopes that past mistakes can be averted and prior successes can be duplicated.
Medical Surveillance
Conducting surveillance is essential to derive situational awareness, especially from a medical standpoint. Medical surveillance is a collection, analysis, and interpretation of death, illness, and injury data. This form of surveillance helps us determine the extent of the impact of a medical crisis on the affected population. Typically, medical surveillance falls under two categories–mortality and morbidity. Each of these categories requires us to identify and track risk factors and disease trends, and ultimately, determine a necessary course of action. Situational awareness in this sector allows us to assess the public health impacts of a medical crisis while evaluating potential roadblocks to future planning and prevention.
Terrorism, Violence, and Activism
While the universally accepted definition of violence remains in question, situational awareness requires us to understand the definitions of violence and terrorism on a broad spectrum. That means using open source intelligence (OSINT) to analyze current and emerging violent or terrorist threats. Aggregated data reveals risk and surfaces anomalies, providing early warning through machine analysis. Organizations like the Global Terrorism Database (GTD) use systematic data on both international and domestic terrorist acts around the world. The GTD provides information from the location of the incident to the weapons used during the attack.
Thermal & Atmospheric Anomalies
The technology behind thermal and atmospheric anomaly detection is paving the way to create situational awareness in efforts to minimize catastrophic disasters like wildfires and other natural hazards. Satellite technology specifically is often the first to detect things like wildfires burning in remote regions, atmospheric anomalies and high levels of pollution, or even solar storms.
Natural Disasters
Natural disasters can occur at any moment. However, technological advancements with situational awareness in mind have made it possible for effective emergency and disaster planning and overall predictability. Though natural disasters are not within our control, data and analytics on seismic variance, thermal events, floodplains and weather patterns narrow the uncertainty around them. Location is also a primary factor in determining the right preventative measures, as natural hazards typically tend to occur in the same geographic area. Situational awareness gives us the ability to monitor specific patterns and provide early warnings and ample time for response planning.
Global Transportation
The pandemic made plain the importance of transportation on a local, national, and global level. It’s a critical component of our infrastructure, but during a crisis, it tends to be the most vulnerable and impacted.
Public Sentiment
Situational awareness requires us to understand the emotions and interests of the public. Social media platforms and search engines are widely used, especially during a time of crisis or disaster. These platforms are not only a tool for frequent updates and fact sharing but can also be utilized to increase public awareness of current and future events, and provide different forms of informational and emotional support. Situational awareness requires us to understand the data and information being produced through these platforms and search engines, so it can potentially be applied to disaster relief and emergency responders.
Update — September 2026
The structure of this essay — medical surveillance, violence and terrorism, thermal and atmospheric anomalies, natural disasters, transportation, public sentiment — treats each vector as its own watch problem. That was the honest description of what was achievable in 2020. It is no longer the limit.
The capability that changed is inference across streams. Models can now ingest many large, heterogeneous feeds at once and reason about the relationships between them, rather than surfacing each one to a human who must do the correlating. Recent work demonstrates exactly this: cross-modal correlation producing context-aware risk inference — a physiological signal read against spatial adjacency to confirm an emergency — with layered inference cutting false positives while raising sensitivity to real events. Large language models have proven effective at guiding cross-modal representation learning, improving robustness precisely where sensing is heterogeneous and imperfect.
This is the difference between six dashboards and one picture. A thermal anomaly, an unusual transport pattern and a shift in local sentiment are three alerts on three screens in the 2020 model. Correlated, they may be one event — and the correlation is the finding, not the inputs.
The caution from the original argument still applies, and applies harder. Inference across everything is also the ability to be confidently wrong across everything. A correlation engine that cannot show which streams drove a conclusion produces a faster version of the problem Dark Winter exposed: decisions no one can reconstruct, made under time pressure, on evidence nobody can audit.
Sources: Multisource sensor fusion with LLM integration for explainable state perception, 2026 · Adaptive multimodal decision fusion for safety monitoring, 2026
What we built from this. DeepWatch organizes analytics across exactly these vectors and turns them into a persona-based operational picture: a standing watch floor with alert triage and a persistent shift log, multi-INT fusion that resolves one event arriving through four sources into one event, and operation at forward and disaster-impaired sites that reconciles when the link returns. Explore DeepWatch.
An earlier version of this essay was published on Medium in June 2020. This version has been revised.