Speakers at the AIxSPACE online artificial intelligence (AI) conference Tuesday (Jan. 19) pointed to the need to consider rapid access for users, along with ethical discussions, when using AI and machine learning for Earth observation (EO).
The rapid expansion of CubeSats, along with more data collection capability by larger EO satellites such as the RADARSAT Constellation series, are creating an interesting problem for providers of terrestrial data. With so much information coming at companies and organizations, selecting actionable information is crucial to respond to rapidly evolving natural disasters or security threats.
“Simply downloading is no longer an efficient way to access the data. Therefore, we need new platforms and new tools to deal with the large volume of datasets,” Samuel Foucher, a senior researcher and remote sensing specialist at Montreal’s CRIM (Computer Research Institute of Montréal), said in a prerecorded keynote speech.
“One possible technical solution that has taken off in the last few years is to build observational ‘data cubes’,” he continued. “The data set are pre-processed, and becoming part of an infrastructure we call ‘data as a service’. So instead of downloading the data, processing is performed close to the data.”
In other words, the newer satellites and their sensors are pivoting not only to gathering information, but attempting to process at least part of that information before sending down the most interesting findings to researchers on Earth. This reduces the problem of data overwhelm and allows researchers to move immediately into taking action, as opposed to spending time crunching numbers.
On the other hand, however, transparency is required as to how the satellite computing devices make their decisions (a classic “black box” problem) and who has the ability to act on that information, warned an AIxSPACE panel participant.
Industry requires a “different conversation on ethics,” as opposed to just how AI does the data-gathering, said Brig.-Gen. Michael Adamson, Director General of space and the commander of the Joint Force Space Component for Canada’s Department of National Defense.
“That is a grave concern and interest for us as well,” he continued. “We want to make sure the right people maintain the right checks and balances on what we’re doing.”
Adamson said many people have a vision of military technology and AI built upon popular depictions that emphasize “robots, science fiction and sci-fi” (perhaps like the Terminator franchise’s Skynet). Adamson added, however, that AI is not the end itself, but a means to an end, as AI “facilitates decision-making loops and processing of information.”
A typical example in Canada is considering the sheer number of vessels typically making approaches to the western and eastern coasts and increasingly, northern regions as well. While most ships play nice and use the mandated automatic identification systems (AIS) that provide information on their port of origin, nationality, and destination, bad actors likely would not, Adamson said. “Typically the people we are looking for are not the ones being cooperative, or doing the things that make them easily found.”
Another aspect of protecting Earth assets comes from Hydro-Québec, which is engaged in at least 10 projects that use AI for aspects like predictive generation. Examples include using LIDAR to cut trees threatening hydro poles and other infrastructure, or monitoring alternative sources of electricity such as solar panels, said David Murray, chief innovation officer of Hydro-Québec and president of Hydro-Québec Production. The organization now has about 30 people working on AI projects, a quickly growing team working closely with Mitacs to serve industry.
Space assets are also quickly growing in their use of AI, such as the GHGSat series. The Montreal-based company is about to launch their third satellite and increasingly uses AI to transform measurements or data into actionable insights for customers, which seek to address methane leaks rapidly in infrastructure such as pipelines.
Typical applications of AI, said GHGSat CEO and president Stephane Germain, include finding small emissions in large “scenes” that are typically just under 150 square kilometres across, finding emissions in data from third-party satellites, and improving the coverage and “revisit” rates by GHGSat satellites. A selection of these AI analytics are available for free via the regularly updated PULSE methane emissions map available to the public.
Finding methane plumes is no small endeavour, Germain added, which is where AI’s subset capability of machine learning (ML) comes in helpful once it knows what to look for. “The plume changes shape, and every plume is different given the wind conditions and atmospheric conditions of the time … it’s a challenge to pick out these emissions.”
ABB is another space company looking to use AI to enhance its sensor capabilities, said Frederic Grandmont, technology and business development manager of ABB’s space and defence systems. “Most of AI is done underground,” he said, meaning that most of the compression and analysis currently takes place on the ground after collection is complete. But in the near future, ABB plans to have sensors pre-select information through ML and send curated data down to Earth.