The ability to successfully predict a crisis is about more than bragging rights. Anticipating a crisis can mean saving lives, limiting damage, or even averting disaster altogether. Geospatial intelligence can play a crucial role in helping to identify a future crisis before it unfolds.
It’s with those stakes in mind that competitors in the second annual GEOINT Innovative Tradecraft Competition faced off to predict a chosen crisis leveraging geospatial data and related capabilities. But amidst the persistent COVID-19 pandemic, it seemed fitting to ask USGIF and Open Geospatial Consortium (OGC) members how they would leverage geospatial technology to address this challenge.
Of course, a pandemic is just one sort of crisis. The competition’s field of entrants addressed a broad range of potential catastrophes, from weather-related disasters to geopolitical conflicts and human-made ecological calamities.
The judges evaluated presenters according to their creativity and innovation, as well as their emphasis on advancing tradecraft and mission, interoperability and data sharing. Adjudicating the competition’s preliminary round on August 18 were Ronda Schrenk, CEO, USGIF; Nadine Alameh, CEO, OGC; Hon. Sue Gordon, former principal deputy director of National Intelligence; and Dawn Meyerriecks, deputy director for science and technology at the U.S. Central Intelligence Agency.
Each of the four finalists will present again on October 6 at the GEOINT 2021 Symposium in St. Louis. Here’s an overview of the four entries that emerged as finalists from the preliminary round.
Identifying bad actors at sea is perhaps the ultimate “needle-in-a-needle stack” challenge. Sure, container ships are large needles, but the world’s oceans are enormous needle stacks, and the odds of catching a vessel in a nefarious act are frustratingly slim. To narrow the search, analysts have traditionally relied on characteristics such as a vessel’s country of origin or region of operation, but that’s a very limited approach. Machine learning (ML) offers a better way.
That better way is what Amazon Web Services (AWS) set out to build with partner Hawkeye 360, a geospatial analytics company based in Herndon, Virginia. Hawkeye 360 uses formation-flying satellites to gather radio frequency (RF)-powered geolocation data. That data has extraordinary potential, but its volume can also prove to be overwhelming—in 2019, for example, Hawkeye 360 geolocated nearly one million signals per month. It needed a way to make sense of everything and approached the Amazon ML Solutions Lab to create an algorithm that could analyze the data its satellites were collecting, with the goal of better identifying law-breaking vessels.
“Machine learning is a powerful tool in this scenario because it allows us to automate the analysis of all of the data, to find interesting trends and patterns, and to drastically narrow down the search area,” said Kate Zimmerman, senior manager of Amazon ML Solutions Lab.
Working together using Amazon’s SageMaker Autopilot tool for quickly building and training algorithms, the companies crafted a solution that identifies patterns and behaviors typical of nefarious vessels, such as rendezvousing with one another at sea. The model then generates a risk score for each vessel based on its characteristics and behavior. Those scores provide maritime analysts with an enhanced view of illicit activity, such as smuggling and human trafficking, enabling them to deploy their enforcement resources more efficiently.
At the end of the joint project, Hawkeye 360 folded the new capability into its Mission Space software platform and continues to refine the model.
Wildfires have burned through millions of acres in California this year, and the same thing happened in 2020. The wildfire season is lasting longer, too, expanding by more than a month over the past three decades. Fighting those fires is dangerous, but it’s also critically important.
Firefighters and other first responders do have a few things working in their favor, such as imagery captured from drones and light manned aircraft that can help them stay abreast of a fire’s progress. The resulting insight can be a literal lifesaver, with one caveat: imprecision. The telemetry from aerial footage can be inaccurate by up to 150 meters.
That’s where Edgybees comes in. The Gaithersburg, Maryland-based company started five years ago as a gaming company that fused aerial video with augmented reality. But it quickly discovered that its technology had powerful, real-world applications as well. By using its software to geo-register drone footage using landmarks such as roads, homes, and schools, Edgybees can produce augmented reality-overlaid images that are accurate to within two meters. The technology is also fast, processing raw imagery in 140 milliseconds.
“When you need to do a search and rescue, or you have planes that are dropping fire retardant, you want to be as accurate as possible,” said Trisha Kinman, Edgybees’ senior director of product marketing. “Accuracy matters, especially when you need to get to someone or home in on a specific target or area of interest.”
The other primary benefit of Edgybees’s software is clarity: Its AR overlays can translate the chaos of smoke and fire into actionable insight. The software works with multiple image types, but in fighting wildfires, infrared imagery is especially useful because it is well-suited for nighttime capture. It shows both where fire is actively burning and where the ground is hot. Egybees’s overlays can include landmark data and dynamic signals, such as a GPS device carried by a stranded first responder. The result is an image that is not only accurate but intelligible and intuitive.
Egybees’s AR overlays also can be modified, allowing response teams to mark locations of interest or measure the size of an at-risk area. The effect is to take aerial footage—so valuable in combatting wildfires, yet so often underutilized because it is either inscrutable or imprecise—and make it a flexible, useful tool in the hands of firefighters.
“The biggest problem people have is contextualizing this in real time. You’re looking at an image or video, and it’s ‘What the hell am I looking at?’” said Adam Kaplan, Egybees’s CEO and its presenter at the USGIF Innovative Tradecraft Competition. “Our software connects the dots through a combination of geo-registration and augmented reality to give people situational awareness in real time.”
Overfishing—catching fish and other marine wildlife faster than those populations can replenish—poses a serious threat to the world’s oceans, one that is too difficult to address through policy alone. That’s because illegal fishing constitutes up to 30% of the total global catch, according to the World Wildlife Fund, earning bad actors up to $36.4 billion annually.
The devastating ramifications of a collapsed fish population aren’t only ecological. Such a crisis would also affect the food security of developing nations, and disrupt a significant source of income for vulnerable populations.
Those are the stakes that prompted a team from Guidehouse, a consulting firm based in Tysons Corner, Virginia, to craft an entry that uses machine learning and geospatial intelligence to identify at-risk fisheries. The team crafted its model by analyzing available data, such as fishing vessel density, sea surface temperature, fish price index, and the catch statistics reported by local fisheries, to track changes in fish population.
The Guidehouse team settled on three key metrics that would indicate a fish population in peril: a total biomass threshold (the weight of all fish caught at a given fishery); a fish mortality threshold (the number of fish caught at a given fishery); and a traffic threshold (fishing vessels docking at a given fishery at an elevated frequency over an extended period).
The greater challenge was identifying a classification model capable of spotting fisheries that hit those thresholds. Fishery data around the world can be spotty, and Guidehouse worked to find an approach that would perform well despite the uneven character of the data while also controlling for sample and measurement bias.
“One of the challenges that we faced was data quality, so we were keen to find techniques that can still work with sparse or incomplete data,” said Nateé Johnson, data scientist and senior consultant and the Guidehouse team’s presenter at the Innovative Tradecraft Competition.
The team’s testing revealed two machine-learning techniques that were up to the task: the “random forest” and the “linear stochastic gradient descent” models. The results left the team encouraged that it was possible to use currently available geospatial data to gain crucial insight into far-flung fisheries and identify areas where illegal, unreported, or unregulated fishing may be rampant—especially if used in combination with other indicators, such as vessel density and movement.
“When I think about the economies that are built on and based off of the [fishing] industry, it would be globally detrimental to have the sort of collapse that we’re at risk of experiencing to have some confidence in what steps we can take to intervene and to mitigate some of these risks is important,” said Johnson. “And it’s important that we act while we still have a chance.”
The COVID-19 pandemic has proven adept at keeping humanity on its heels—reacting and responding to outbreaks more often than anticipating them. Whitespace, an Alexandria, Virginia-based company specializing in location-focused analytics, is working to change that with its Data Pandemos project. Data Pandemos aims to equip government officials with a tool to better predict and mitigate outbreaks.
Data Pandemos uses aggregated, anonymized geolocation data from mobile devices to identify proximity patterns: the times and places when large groups of people are likely to be within six feet of one another, and thus at greater risk of COVID-19 transmission. The project is a collaboration between Whitespace and Yale biostatistician Forrest Crawford, with the State of Connecticut serving as the pilot client.
Whitespace gathered its data through a partnership with location data provider Outlogic, gaining access to a sample size of about 3% of Connecticut’s population. It then built and trained an algorithm to flag instances when multiple devices gather in close proximity. The focus on proximity was different from the two prevailing strategies aimed at mitigating outbreaks, both of which were flawed: contact tracing was reactive, while the mobility-based approach that informed interstate travel restrictions failed to account for the fact that an activity, such as crossing state lines to go camping, was far less likely to influence COVID transmission than attending an indoor concert in one’s own community.
“The real driver of disease transmission for COVID is coming into contact with other people,” said Samantha Leung, a spatial data analyst at Whitespace and the company’s presenter at the Innovative Tradecraft Competition.
So that’s what Data Pandemos measures. Its dashboards display trends and spikes in contact, including a meaningful correspondence between contact and positive cases over the last year. That connection gives public health agencies a tool to help focus testing resources before there is a surge in cases and guides messaging about masking and social distancing protocols. For example, as Connecticut colleges and universities prepared to reopen this fall, the data showing the correlation between contact rates and positive cases informed the decisions to require masks indoors at several institutions. The insights were also useful in guiding deliberations about reopening plans at shopping malls and restaurants as the state moved to ease its “Phase 3” restrictions.
Whitespace is now experimenting with using the device-proximity algorithm to assess crowd dynamics in different contexts, such as sporting events and political rallies.
“The most interesting thing for the company as a whole is getting to use the skill sets that we already have, and the capabilities that we’ve been building, on different use cases” such as venue and facility operations,” said Leung.
Who do you think will win? Each finalists will present their cutting-edge technologies again Oct. 6 on the Main Stage at the GEOINT 2021 Symposium in St. Louis where the champion will be determined by audience vote. Register for the Symposium here.
Featured image: Edgybees technology can produce augmented reality-overlaid aerial images that are accurate to within two meters. (Still image from Edgybees capability video shown during the first round of the GEOINT 2021 Innovative Tradecraft Competition.)