Many companies in the aerial and satellite imaging sector probably dream of being an AI innovator ‘par excellence’. They would love to leverage artificial intelligence to rapidly convert high-resolution multi-view imagery into high-definition vector maps. With advanced AI technology, they could even map an entire continent in a matter of months, or keep a digital twin continuously up to date with the real world. This could be a possible scenario in the future, but the industry is not quite there yet. It takes a dedicated strategy to achieve future-ready critical success factors, according to Ecopia AI.
Artificial intelligence-based systems stand out in their ability to leverage human-like image interpretation and reasoning to create map features that accurately represent the real world. In a simple example, AI-based systems can infer when a pavement is obscured by tree canopy and produce a full pavement vector layer, whereas other computer-based mapping methods would produce a fragmented layer. To achieve these results, a neural network must be rigorously trained to mimic what a human can detect in imagery and subsequently deduce from their contextual knowledge and sound reasoning.
Beyond what is humanly possible
Toronto-based Ecopia AI has built up a reputation for superior AI innovation in the geospatial imaging sector. The majority of the Canadian company is comprised of R&D teams entirely dedicated to training and evolving AI-based systems to ingest new types of imagery and extract highly accurate features for mission-critical applications, such as transportation planning, climate resilience and public safety. Abigail Coholic, senior director at Ecopia, is confident in the company’s future-proof mapping capabilities. “In the last ten years we’ve developed a robust system. Our AI-powered systems make it possible to scale image interpretation beyond what is humanly possible while retaining the human-level reasoning that is required for accuracy,” she says.
When creating a 3D feature, the standard process is to take reference of all the points across the feature and then determine its height. AI-based systems can take into account millions of reference points across multiple images over time. At this scale, a human would struggle to keep track of all of this data in their brain. Additionally, leveraging AI results in a normalized data output that is not always possible with manual digitization due to inevitable human error. Two GIS professionals could interpret an image and all of the world’s complexities differently, producing inconsistent results. In the case of AI, a computer-based system extracts data across the full area and routinely applies the same definitions and interpretation style. This normalization of feature extraction results is an important benefit of AI. “Coupling this normalization with 3D extraction means that you can now truly track precise changes of your map data at scale instead of re-extracting each time you get imagery and then trying to compare those results,” Coholic explains. Ecopia envisions this AI technology powering a digital twin of the entire world, updated at least every year.
Value for all types of imagery
In Coholic’s experience, the new possibilities that AI and machine learning bring do not change the relative competitive position of aerial photogrammetry and satellite images. “The ideal geospatial ecosystem includes satellite imagery, aerial imagery and AI, as they all provide additional value to each other. For satellite imagery specifically, AI-based mapping is beneficial because of how frequently imagery is captured. It is not humanly possible to manually digitize all of the images being constantly captured by satellites. However, this frequent imagery capture and the subsequent AI-powered feature extraction are essential for understanding the dynamic nature of our world. We can start to monitor change in near-real time,” she continues. “And the same goes for aerial imagery. One of its main benefits is its high resolution, which provides an extremely granular view of the planet. This leads to a similar challenge in manual digitization; there is so much detail to capture that it is not feasible to do manually. So in both cases, and also with drone or street view captures, AI affords imagery providers with the ability to double down on their unique advantages.”
Traditional geospatial data creation was so resource-intensive that most organizations would only update their databases every three to five years. By creating and maintaining map content at scale, geospatial data becomes more accessible. The positive feedback loop that comes with more frequent data creation, plus the increase in end users leveraging this data, leads to increased demand for even more frequent and detailed data creation. “However, there is still a learning curve when it comes to assessing and implementing new geospatial technology. This remains a big bottleneck for growth in our industry,” expresses Coholic. “We need not only to develop innovative technology, but also enable organizations to use it, which comes down to education. End users now have so much to choose from that they have to spend a lot of time evaluating different sources and determining what innovations are useful. Geospatial companies need to empower their users by sharing new implementation strategies and inform actively on market trends relevant to their core responsibilities.”
Land cover
Ecopia invests heavily in thoroughly understanding the reasons behind the user applications. What are the most essential vector features for the analysis their customers need to transform data into actionable information? And can they extract them at national or continental scales? As a result, the company has off-the-shelf high-resolution mapping products that include 3D land cover across the USA, Canada, Italy, Australia and Ireland, as well as in major metropolitan areas around the world. 2D content also includes off-the-shelf building footprint and road data for all of sub-Saharan Africa.
Last year, Ecopia completed the delivery of land cover data to support the Coastal Change Analysis Program from the US National Oceanic and Atmospheric Administration (NOAA). Historically, the data was produced at a resolution of 10-30m, but Ecopia produced a 1m-resolution vector dataset across over 1.5 million square miles. While NOAA leverages the vector version of this land cover dataset for its own operations, any organization can access the 1m raster dataset for free to support climate resilience analysis. Ecopia is currently expanding the coverage to empower US coastal communities to understand climate risks and develop mitigation strategies to enhance resilience and sustainability. For example, the State of Washington leverages the data for hydrological modelling and flood mapping, and also to optimize the placement of electric vehicle (EV) charging stations.
Transportation networks
Over the past year, the company has standardized the extraction of highly detailed transportation features, resulting in planning data solutions for transportation authorities. For instance, the Illinois Department of Transportation and the Chicago Metropolitan Agency for Planning engaged Ecopia to provide transportation agencies across the state with access to routinely updated information about their infrastructure. Information on pavements, pedestrian crossings, intersections, lanes and more is used to advance multimodal analysis, compliance with the Americans with Disabilities Act, Vision Zero objectives, performance-based capital programme monitoring and storm-water and flood management. “I am glad to think that, in doing so, transportation plans – and therefore regions – become safer, more sustainable and more accessible, ultimately improving residents’ lives,” Coholic comments.
Another recent transportation project was in the UK to map Network Rail’s railways and surrounding land: over 20,000 miles of track and more than 30,000 bridges, tunnels and viaducts. Network Rail wants frequent updates on land use changes within a specified buffer around rail infrastructure to assess the impact on travel safety and sustainability. Ecopia highlighted where changes took place. The AI-based mapping systems ingested and compared ESA and Airbus imagery from the first and second halves of 2022, delivering a vector dataset of classified feature layers in the space of just six weeks.
Critical success factors
The monumental success of AI in general in 2023 begs the question, ‘Why isn’t every company in the mapping business doing something like Ecopia?’. It seems easy to ask ChatGPT for some code to create an algorithm for the specific use case. However, as anyone who has tried that in practice knows, actually implementing those algorithms and getting them to improve efficiency is not that simple – particularly in the geospatial industry, where data is used to make highly detailed decisions such as where to plant trees to reduce urban heat islands, or where pedestrian infrastructure is needed to reduce the risk of traffic fatalities. A poorly trained, low-accuracy product will not provide enough value and improvement to surpass traditional methods.
Coholic sees three critical success factors: “First, you need expert knowledge of the domain you are trying to implement a solution for and the AI technology you are applying. Second, you need to invest time in training the algorithms that power that AI, accounting for all edge cases. Third, you need to make a huge financial investment in servers, cloud infrastructure and training data. The imagery datasets used to create high-quality maps are enormous, and processing them requires advanced and resource-intensive infrastructure. It is this commitment to thoroughly understanding the power of map creation, becoming an authority in computer vision, and investing massive amounts of time and money into our systems that have resulted in our industry-leading 3D AI map creation, which is not so simply replicated.”
Additionally, the internal workflow for an average project has evolved immensely over the last two years, according to her. “Each dataset we create requires much less customization than previous projects. We have executed projects in more than 100 countries around the world, and with every one we further advance our AI-based systems to map new features and geographies with greater accuracy and efficiency.”
Meaningful
Looking to the near future, Abigail Coholic is hopeful that the next couple of years are going to be particularly meaningful in the field of optimizing geospatial data analysis – and not only in terms of business opportunity or growth. “As the power of AI is harnessed more fully by governments, it will enable improved decision-making that benefits all individuals. Some of the biggest geospatial themes we’ve seen being funded this year are equity and accessibility. High-quality and regularly maintained data is one of the foundations for remedying historic inequities and exclusions in society. I think successes like these will contribute to new professions looking to integrate geospatial data and to existing industries broadening the scope of their work, which will benefit every stakeholder,” she states.
Lastly, she also believes that the Future of Life Institute’s ‘Open Letter to pause AI’ demonstrates the new potential for the public and private sectors as well as non-profits and academia to act as partners. “Together, we can pursue solutions that prioritize the flourishing of humanity over technological competition.” Coholic concludes: “The fact that so many of the concerns around this technology are being discussed well in advance of its full development, by all leaders of our world, means that we have the opportunity to properly account for all externalities of these advancements and make decisions about where we should never take these technologies collectively,”
