AiThority – Artificial Intelligence | News | Insights | AiThority
While most people may think of ChatGPT or deep fake images when they hear “artificial intelligence”, AI methodologies are proving beneficial for a variety of applications, especially in data-heavy industries such as Earth observation. Using new government and commercial models, customers are able to remotely monitor large ship fleets, measure poverty at a granular level, and track natural disasters to provide rapid, targeted aid to people in need.
This NASA Landsat-8 image shows the wildfires that raged across Maui on Aug. 8, 2023. Government agencies and private companies are developing new AI tools to prioritize satellite image processing for natural disasters and other events where a rapid response is essential. [Credit: NASA Earth Observatory image by Lauren Dauphin]
Satellites in orbit are producing more data than ever before, and operators are turning to sophisticated processing techniques to enable faster and more meaningful analysis across a range of applications. For Earth observation (EO) satellites, especially those producing optical imagery, one tool is quickly becoming essential: artificial intelligence (AI).
On Aug. 3, NASA and IBM released the first open-source foundation model for satellite image analysis, hosted on the Hugging Face platform and trained on Harmonized Landsat Sentinel-2 (HLS) data. The release includes demo models trained to identify geographical features such as burn scars and flooding.
Rahul Ramachandran, project manager and senior research scientist at NASA emphasized the importance of foundation models to advancing geospatial analysis in a press release for the partnership. “AI foundation models for Earth observations present enormous potential to address intricate scientific problems and expedite the broader deployment of AI across diverse applications,” he said.
NASA and IBM’s foundation model isn’t the only AI milestone reached by the space industry in recent years, either. AI and space have long associations, ranging from satellite maneuvering and spacecraft telemetry to Mars-based robotic navigation and exoplanet analysis. As commercial EO satellite constellations grow and produce more images every day, companies are developing AI-based software to compile data and provide usable, easy-to-understand products to their end users.
The market for commercial EO data is growing as commercial and government customers — including national security agencies with their own high-resolution satellites — find benefits in the satellite constellations and analytical tools. Markets and Markets estimates that the geospatial analytics market (valued at $78.5 billion in 2023) could grow to $141.9 billion by 2028.
The EO industry has come a long way from the early space race days of strapping cameras to suborbital rockets. Now, commercial and government satellites take millions of photos of the Earth every day from a variety of orbits.
The number of EO and remote sensing satellites orbiting Earth is growing alongside overall payload launch trends. According to Space Foundation analysis, 265 EO/RS satellites were deployed in 2022, almost 40% more than in 2021. Of these satellites, the majority (67%) were optical, producing high-resolution images of the Earth. A decade earlier, only 18 of 193 satellites deployed were for remote sensing or environmental monitoring.
However, with frequent, high-resolution imagery comes millions of large files that need to be transmitted to Earth and stored for analysis. NASA’s Earth science data archive held 72 petabytes (one petabyte is equivalent to 1 million gigabytes) of satellite imagery by the end of fiscal year 2022. That amount is expected to grow rapidly over the next few years as more EO satellite missions are launched. NASA projects the archive could reach 600 PB before the end of the decade. Europe’s Copernicus program had 34 petabytes in its data archive at the end of 2022 and is expected to grow to 80 PB in six years.
Because many EO companies utilize constellations of small satellites rather than one large one as many government missions do, those companies also produce massive amounts of data. Planet Lab’s constellation has photographed every location on Earth 1,300 times on average. Maxar Technologies, founded on Oct. 5, 2017, has more than 125 petabytes in its own data archive.
The AI field is so versatile that different techniques can be used throughout the lifecycle of satellite data analysis for increased speed and accessibility. Esri, the geographic information system (GIS) company behind the ArcGIS software family, identifies three key ways to apply AI to geospatial analysis: automate tasks and repeat them quickly at scale, look at past patterns to make predictions, and search for patterns hidden in large amounts of data.
NASA and IBM’s foundation model is built to make the initial data cleaning step easier and more accurate so that researchers can focus on their actual analysis of the identified features. This AI model has demonstrated a 15% improvement over traditional models while only needing half the typical training dataset. For EO image datasets, which require manually identified features from which the model learns, the smaller training dataset requirement is a massive timesaver.
UKG Survey 95 Percent Of Indian Employees Believe AI Can Improve Their Quality of Life
Leverage Microsoft 365 Enhanced Security to Stay Ahead of Evolving Threats
IGC Pharma Announces Master Agreement with Leading South American University to Advance AI Initiatives 
The European Space Agency’s Φ-lab, in collaboration with Oxford University and Trillium Technologies, utilized the RaVAEn machine learning (ML) model to speed up an even earlier part of the satellite data lifecycle, transmitting large files to ground stations on Earth. This model is hosted and trained directly on satellites, allowing them to detect changes in images of the same location and to prioritize the downlink of the anomalies that are deemed important.
On the commercial side, most EO companies develop proprietary software that utilizes AI to clean and present analysis-ready data to their customers. Maxar’s DeepCore Suite consists of more than 100 models that process satellite imagery and identify object types, helping users build training data and AI models.
Similar to the RaVÆn model, BlackSky’s Spectra AI software processes satellite imagery, detects changes and prioritizes analysis of important events such as natural disasters or military movements. However, instead of identifying events for the model to photograph, the BlackSky team taught the software to monitor events itself, according to Chief Innovation Officer Patrick O’Neil.
“Tasking is automated,” he said. “Our AI reads the world’s news, including hyper-local foreign language news sources all the way from the Associated Press and BBC. It identifies emerging events around the world and automatically tasks our satellites to take an image.”
One of the biggest advantages of foundation models for satellite imagery analysis is that they can be adjusted for different use cases. NASA and IBM’s model includes demo applications such as categorizing types of crops or mapping flooded areas at a higher resolution than the original satellite image. Because the model is open-source, researchers can train it on different types of data to identify features of interest.
One of the most recent demonstrations of the versatility and effectiveness of AI is Synthetaic’s use of existing software, which identified and traced the path of a Chinese balloon as it flew over the United States in February. Synthetaic, founded in 2019, created the Rapid Automatic Image Categorization (RAIC) tool to identify features from a single reference image instead of relying on a large training dataset. Paired with unlabeled Planet Labs satellite imagery, Synthetaic CEO Corey Jaskolski located the balloon in less than two minutes using RAIC and a rough RGB sketch.
Many recent improvements in satellite image analysis focus on accessibility. New AI tools and user interfaces are enabling a wider range of individuals — including social scientists and journalists who may not have specialized training in geospatial analysis — to gain insights from satellite data and share the results.
One particularly useful application for optical satellite imagery is measuring poverty at a granular level within regions of interest. These AI poverty models, like the one developed by Stanford University’s sustainability and artificial intelligence lab, typically combine high-resolution daytime imagery with nighttime light intensity data that can represent how developed a location is.
While using satellite data is promising for research applications where traditional data is difficult to predict or collect, there are concerns that not all research is adhering to the best practices that are being established in this emerging field. A review published in October 2022 examined 32 papers that utilized similar ML methods and found that many lacked at least one of the three main requirements for AI models: transparency, interpretability, and domain knowledge.
In response to the growing demand for geospatial analysis, many EO satellite companies are making strategic partnerships with — or acquisitions of — AI software companies. After Synthetaic’s success using Planet Labs imagery, the companies announced a formal partnership in April.
Maxar Technologies partnered with Blackshark.ai in 2022 to combine satellite imagery with 3D modeling tools, and later that year, the company acquired Wovenware to further grow its AI and 3D capabilities.
One opportunity for the industry came on Aug. 7, when the Office of Space Commerce modified licenses for many commercial EO constellations, lifting restrictions that included image resolution limits for the companies. Without these restrictions, private companies can provide higher-quality analysis to their non-government customers and continue to expand use cases for AI satellite imagery analysis.
Zoe Hobbs is an economist and data analyst at Space Foundation.
Prev Post
Nokia Supercharges AI with Ultra Ethernet
Next Post
Retail Crypto Traders ‘Increasingly Interested in Altcoins’
UKG Survey 95 Percent Of Indian Employees Believe AI Can Improve Their Quality of…
Leverage Microsoft 365 Enhanced Security to Stay Ahead of Evolving Threats
IGC Pharma Announces Master Agreement with Leading South American University to…
Narwal Appoints Prashant Malvadkar as VP and Global Head of Data
Comments are closed.
AiThority.com covers AI technology news, editorial insights and digital marketing trends from around the globe. Updates on modern marketing tech adoption, AI interviews, tech articles and events.
Copyright © 2023 AiThority. All Rights Reserved. Privacy Policy
