Introducing DistanceML: Estimating Social Distancing with Satellites and AI
Social distancing has been encouraged worldwide to combat the spread of COVID-19, but nations struggle to maintain it in practice. We introduce a new AI tool — DistanceML — to automatically infer neighborhoods of cities with higher and lower levels of human movement using satellite data on a weekly basis. We hope this will be a useful tool to aid public health officials in encouraging social distancing and slowing the spread of COVID-19.
I typically haven’t written articles about my own projects, but these are unprecedented times, and they call upon all of us to do our part. Read further to learn more, in the usual format of my blog!
1. The Problem
No human being had immunity to COVID-19 prior to its emergence, and thus all of us are susceptible to it. One of the best ways to prevent its spread is to engage in social distancing, a term that many of us had never heard about just a few months ago, but is now part of everyday conversation almost everywhere in the world. However, in many places, like the United States, a legal framework to force a complete lockdown may not exist. Even in places where it does exist, it is exceedingly difficult to enforce. People still crowding beaches during spring break, and protesting the shelter-in-place orders in several American states, showed us that it is a huge challenge to bring everyone on board.
Additionally, it is likely we may have to continue social distancing in some form for the foreseeable future. The extent to which we will have to socially distance depends on the precise nature of the virus, SARS-COV-2, and that is still not well understood. It seems that we are learning something completely new about it every single day. Furthermore, the initial impacts of the virus have been in developed countries where levels of travel are much higher, compared to developing countries. It is only a matter of time until large outbreaks emerge in the developing world and recent UN estimates suggest that there will be as many as 10 million cases in Africa in just 6 months. In many developing countries, public health infrastructure is not very strong, so this would be a real disaster. The only way to truly mitigate the damage would be through social distancing. But how do you initiate and continue to enforce social distancing on such a large scale?
2. Existing Approaches
Countries around the world have had mixed success in enforcing lockdowns. For example, in China, strong policing and surveillance has facilitated the lockdown. Recently, the New York Times revealed that citizens receive a ‘Health Code’, autogenerated in the popular AliPay digital payments app, which algorithmically dictates if they are to be quarantined or not, based on their personal information, location data and more. The citizens were never asked for any consent before their private data was harvested and this system was created — this model is not only impossible in liberal democracies, but also very troubling from a human rights perspective.
In most liberal democracies, such as the US, India, and many EU countries, lockdowns and shelter-in-place orders are typically enforced through the voluntary compliance of the citizenry and a much smaller amount of policing. However, even this good-will can eventually falter — we have already started to observe friction in the United States with protests against lockdowns, and even some groups in India who do not appreciate the need for a lockdown and attack healthcare workers. Most liberal democracies do not have enough police to enforce lockdowns in every corner of a country once people become more restless. Furthermore, the police put themselves at risk of succumbing to COVID-19 by working to enforce lockdowns.
Hence, if we had a more automated and simpler manner of bringing about social distancing, which also respects civil liberties to the greatest extent possible, we would be much better equipped to face the pandemic.
3. Related Developments
The last decade has seen the rise of Artificial Intelligence (AI), largely due to improvements in computing power. Already, AI has begun to impact various facets of our lives, and there are countless people trying to use AI to ‘solve’ coronavirus. To date, most of the efforts have been on directly modeling public health data, and there are even competitions to do this. However, the applicability of AI is definitely not limited to just this facet, and there may be countless outside-the-box applications that no one has even thought about yet.
There is also an often overlooked source of data about the world that is growing exponentially: satellite data. The number of satellites around the Earth is expected to grow five-fold this decade, and petabytes of data are already produced on a daily basis by satellites that observe the earth. These earth observations can span the entire electromagnetic spectrum, from gamma rays to radio waves, producing a variety of data every single day. In fact, the top satellite constellations (groups of satellites) can resolve objects up to just a few meters on earth’s surface and can image every location on earth every single day. My own company, ConserWater, which I founded in 2016, uses some of this data along with AI to predict soil moisture and nutrient levels in the soil.
4. Connecting the Dots
How can we tackle social distancing with satellites? After some thought, you may think about automatically detecting and counting the number of individuals moving outside their homes. That would be the ‘conventional’ approach, but it is also impossible to do — most satellites do not monitor a single location continuously to notice people leaving their homes and gathering. Even if they did, they do not have sufficient resolution to discern individuals — experts say a 5cm imaging resolution is required, and no satellites are legally permitted to operate at that level. There may still be some military spy satellites operating with that resolution, but they are quite unlikely to be made public even during a devastating pandemic.
Hence, we need a true outside-the-box solution. If we are able to measure social distancing with satellites and AI, we would be able to greatly simplify its implementation on a global scale, in a cost-effective manner. We would be able to quickly identify which areas are not complying, and concentrate public messaging and enforcement in those regions.
We propose DistanceML, an AI algorithm using 10 meter (30 feet) resolution satellite data to infer relative levels of NO2 on a neighborhood level, on a weekly basis. Why NO2? Because it is a great proxy for human motion — most of us get around by using fossil-fuel powered vehicles, and these vehicles inevitably emit NO2, a major pollutant that is already measured by some satellites. However, we can’t use data directly from these satellites that measure NO2, as they are of relatively poor resolution. A single pixel on the images taken by these satellites can be 1 kilometer (0.6 mile) wide or larger. With such a bad resolution, we can’t use this data to draw conclusions on the scale of neighborhoods in a city — this is the minimum scale we would need to operate on to truly measure social distancing. In DistanceML, we have trained an AI model to infer NO2 levels directly from 10 meter images, using years of satellite data, and the results are very promising.
Here’s a sample of some initial data from Madrid and Washington DC. We have found similar trends from cities around the world. We continue to fine-tune our model, and hope to generate more results for major cities in developing nations in the very near future.
Now, armed with these images, public health officials in any city around the world can get a quick understanding of neighborhoods where further messaging and enforcement is required to bring about social distancing. It is another powerful tool for their arsenal, in our war against the pandemic.
You can learn more about the DistanceML project by contacting me, Abuzar Royesh or Olamide Oladeji at Stanford University.
Do you know anyone at the local, state, or central levels of government anywhere in the world? Contact your government officials today, and let them know about DistanceML! Let’s work together to fight COVID-19!
Like this content? Sign up by clicking below to receive an email every month with the latest articles from Out of Box!
