The advent of large constellations of small satellites for earth observation are generating an increasing torrent of data. Despite downlink speeds increasing 10x every 5 years from small satellites and reaching 1Gbps as of right now from 3U satellites, the ability to download all data still presents a bottleneck in the full process chain from data capture to solution delivery to the customer.

On earth, the same problem has appeared in the Internet of Things, with large number of ever more capable sensors producing too much data for backhauling to a central location. The solution developed and deployed is called Computing at the Edge. It pushes the analysis of the data towards the individual sensors (“the edge”), with only the relevant information being backhauled when needed.

Faced with sensors that can produce upward of 100Mbit/sec - or 1TB per day - Spire Global started looking at bringing a powerful “Computing at the Edge” capability to space. However, we did not want to only think about just processing of the data, but take advantage of terrestrial improvements in microprocessor development for Machine Learning and AI, including parallelized GPU-capability in small, low-power microprocessors.

One particular use case for our existing customers is the extraction of information from high-density, highly overlapping RF signals over high-traffic maritime zones (e.g. the South-Chinese Sea). In this areas, RF traffic is commonly dense enough that it’s hard to recover individual ship messages with traditional signal de-interference (“decollisioning”) algorithms.

Spire has found that deep learning techniques show excellent improvements and has deployed this with success. The first, very simple deep learning algorithms already yielding up to 30% improvements in information extraction. However, the bottleneck for training these and more advanced AI and ML algorithms is access to the raw RF-IQ files, which are prohibitively large to download for anything larger than a few seconds. The solution would be a fast on-orbit AI /ML processor which would use the raw, real-time data stream generated to train a model in situ, without ever a need to download all of the raw data. There are many other types of in-situ processing applications for the wide range of data collected by Spire satellites, include live maritime or aviation traffic pattern detection (e.g. illegal fishing), decollisioning of aircraft message, ionosphere and weather data processing and others.

With the support of ESA, Spire is launching in November just such a “AI Brain” into space. It will provide over 1 TFLOPS and a 256 core GPU processor. Created specifically for AI at the edge for terrestrial application, we believe bringing this capability to space is the next step in Earth Observation and the next step in Computing at the Edge.

The presentation will cover the challenges, concept of operations, and initial applications of Spire’s ‘AI Brain’ (codename Sabertooth) and give an outlook of additional applications and development planned for the following 12 month.

With the advancement of space hardware technologies such as smaller spacecraft, component and instrument miniaturization and high performance space processors, and with the advancement of software technologies in artificial intelligence, big data analysis and autonomous decision making, Earth Science is looking at novel ways to observe phenomena that previously could not have been studied or would have been too expensive to study with traditional missions. In particular, the New Observing Strategies (NOS) component of the NASA Earth Science Technology Office (ESTO) Advanced Information Systems Technology (AIST) Program aims at leveraging these novel technologies as well as low cost and easy access to space to acquire multi-temporal or simultaneous multi-angular, multi-locations, multi-resolution and multi-spectral observations that will provide better multi-source measurements and will build a more dynamic and comprehensive picture of Earth Science phenomena that need to be studied and analyzed. For applications such as water resources management, air quality monitoring, biodiversity studies or disaster management, NOS will integrate the use of small instruments, small spacecraft, constellations of spacecraft and networks of sensors to design new missions that will provide the necessary measurements to improve future forecast and science modeling systems.

Measurement acquisition will therefore be approached as a system of systems rather than on a mission basis, and a system of this complexity should not be expected to work without full integration and experimental characterization. Although most of the individual technologies enabling to link and coordinate multi-source observations are more or less mature, a few technologies need to be developed and all of them need to be integrated and tested as a system. In order for this validation to occur, the AIST Program is developing the NOS Testbed that includes 3 main goals:
1. Validate novel NOS technologies, independently and as a system
2. Demonstrate novel distributed operations concepts
3. Socialize new Distributed Spacecraft Mission (DSM) and SensorWeb (SW) technologies and concepts to the science community by significantly retiring the risk of integrating these new technologies.
The NOS Testbed will consist of multiple sensing nodes, simulated or actual, representing space, air and/or ground measurements, that are interconnected by a communications fabric (infrastructure that permits nodes to transmit and receive data between one another and interact with each other). Each node will be supported by hardware capabilities required to perform nodes monitoring and command & control, as well as intelligent “onboard” computing. The nodes will work together in a collaborative manner to demonstrate optimal science capabilities. The testbed will enable to validate technologies such as inter-node communication models, techniques and protocols; inter-node coordination; real-time data fusion and understanding; planning; sensor re-targeting; etc. Additionally, the testbed will have the capability to interact with various mission design tools, OSSEs and one or several forecast models. More details about the NOS Testbed will be presented at the confererence.

Capella Space is an aerospace and information services company that provides Earth observation analysis-ready data on demand. Capella is the first U.S. commercial Synthetic Aperture Radar (SAR) satellite constellation capable of providing hourly monitoring of the dynamic and changing conditions on the surface of the Earth. In December 2018, Capella launched its first pathfinder satellite that is successfully operating on orbit. Following missions will be incrementally deployed to build a 36-satellite constellation, with three satellites each operating along 12 symmetrical polar orbits, progressively reducing time to revisit for a given point on the globe. This large constellation guarantees constellation redundancy, reliability, and responsiveness.
Capella is making space-based remote-sensing more persistent, through data acquisitions day or night and in all weather; accessible, through direct tasking, web ordering, and API integration; and more affordable through tiered pricing and service levels. Capella’s high revisit time, and Very High Resolution (VHR), up to 0.5 meters, allows global satellite users to exploit multi-temporal series for unprecedented change detection capabilities, interferometric SAR techniques, precise target detection and tracking, bio-physical parameter estimation and mapping.
We present updates on our mission and goals to support the community focused on future EO applications. We focus on Capella’s SAR capabilities, product specifications and data provisioning strategy to deliver high value data to both SAR experts (in form of Single Look Complex (SLC) and Ground Range Detected (GRD) images) and new users who seek information streams to solve problems.
Moreover, we present how high-revisit VHR SAR collects will represent a game changing factor for classical radar signal processing methodologies, such as object detection, change detection and interferometry. The high revisit detection of changes is going to enable several new applications in across market verticals including risk assessment for insurance, environmental monitoring, oil and gas and maritime monitoring.

EO-ALERT is an EC H2020 project that proposes the definition and development of the next-generation EO data processing chain, based on a novel flight segment architecture that moves optimised key EO data processing elements from the ground segment to on-board the satellite, with the aim of delivering the EO product to the End User with very low latency (quasi-real-time).
EO-ALERT aims to provide the technological building blocks for this next-generation EO data processing chain, with a view to future autonomous EO satellites. The EO-ALERT satellite concept is based on the establishment of an on-board data handling and processing capability, that enables the generation of EO products on-board the satellite and their direct transfer to the End User. This overcomes the typical bottleneck problem with the raw data transfer to ground in the classical EO data chain, avoids the transfer of EO data that is not of value, and allows for the provision of EO products with very low latency. EO-ALERT targets the provision of high-priority EO products from observations, made globally, to the ground and End User, within 5 minutes.
In this manner, the EO-ALERT concept proposes a new breakdown of functions between the flight and ground segments, where the ground segment is apportioned the task of managing the mission and its objectives, while data processing is moved partially or totally onboard the satellite, with the processed data and EO products transferred to ground, directly to the End User.
The EO-ALERT satellite concept particularly supports those EO scenarios and mission concepts that are enabled or enhanced by the provision of EO products with very low latency. While such mission concepts are becoming ever more common in order to meet more demanding End User needs, in the EO-ALERT project, two such scenarios are considered in detail, being maritime surveillance and extreme weather monitoring.
The EO-ALERT project considers the challenge of providing very low latency EO products considering operational scenarios and requirements, through the incorporation of End Users in the scenarios definition and the system requirements specification. While the technological building blocks of EO-ALERT are generic, the EO-ALERT developments are made for two VHR missions, in the DEIMOS-2 VHR optical satellite and the TerraSAR-X VHR SAR satellite. These two satellite missions are used in the project for verification of the technologies and the overall data chain using real VHR satellite image data. The two satellites are also used as experimental validation cases, to demonstrate the effectiveness of the EO-ALERT concept in real EO scenarios.
This paper provides an overview of the results achieved by the EO-ALERT project (http://eo-alert-h2020.eu/). The paper presents the proposed flight segment architecture, its performance, the hardware employed and the technologies developed, for both the two different user scenarios: ship detection (maritime surveillance) and extreme weather observation/nowcasting.
The proposed EO-ALERT satellite data handling architecture is shown to solve the very low latency EO data challenge through a combination of innovations in the on-board elements of the data chain and the communications. Namely, the architecture introduces innovative technological solutions, including on-board reconfigurable data handling, on-board image generation and processing for the generation of EO products and alerts, on-board data compression and encryption, high-speed on-board avionics, and reconfigurable high data rate communication links to ground. This communication link includes separate chains for raw data (typically characterised by large volumes) and for products and alerts (typically featuring small volumes and calling for fast delivery), to allow the transfer of the latter with minimum latency and global coverage.
The EO-ALERT project is currently in its second year of execution, moving towards the verification of the individual technological building blocks in a real-time environment. The results to date show that, when implemented using COTS components and available and upcoming communication links, the proposed architecture can deliver EO products and alerts to ground with latencies lower than five minutes, for both SAR and Optical missions, demonstrating the viability of the EO-ALERT concept.
The EO-ALERT project pursues the development of the very low latency capability of the EO data chain at a system level, in order to achieve a medium-high TRL level (TRL 5/6) of the concept, supporting the uptake of the concept and its technologies in upcoming EO missions. While the employment of AI-based technologies is not a basic aim of the EO-ALERT project, the project does showcase some areas where AI-based technologies have been employed and provide benefits, as well as some of the challenges associated with their employment. As such, AI technology employment is not required in EO-ALERT. Rather its use is encouraged in an organic manner, to be exploited when this allows for improved system capabilities vs. cost, mass, power and TRL/risk.
Finally, the paper also discusses the implementation of an avionics test bench for testing the flight segment architecture with real EO data, from both the optical DEIMOS-2 and SAR TerraSAR-X missions. This testing with real EO data serves to demonstrate that the EO-ALERT concept, as a complete system and data chain, can meet the requirements of the considered scenarios in terms of detection performance, and critically latency. When proven, this will provide technological building blocks for upcoming European and international missions, and open new opportunities for the exploitation of civil EO products based on this novel flight segment architecture.

During this session BlackSky and e-GEOS will expand on the theme of reliably monitoring sites and areas of interest with multiple sensor types including, but not limited to SAR and EO capabilities, to better detect Pattern of Life and on the ground anomalies. This session will demonstrate how the real world application of new space economics – many tens of satellite(s) constellations and multiple launch ride share services, combined with the rapidly decreasing costs of Machine Learning compute power and the massive scaleability and reach of cloud platforms have converged to now deliver previously unparalleled levels of persistent and deep stack monitoring globally from space. BlackSky will present the operational status of its multiple satellite 1m EO constellation and the very high revisit and time of day diversity of the constellation. BlackSky and e-GEOS will further discuss low latency (from tasking to delivery) applications and SAR/EO hybrid solutions. All presented to users over a powerful AI/ML platform at high scale and with rapid delivery to the end user.

GHGSat is an SME developing satellite technology and capability for integrated data and analytics solutions specifically designed to identify, quantify, attribute and help address greenhouse gas emissions. This capability is targeted at multiple sectors, including oil and gas, hydroelectric power, mining, waste management and agriculture. In June 2016, GHGSat launched GHGSat-D, the world’s first and only satellite capable of measuring greenhouse gas emissions (CO2 and methane) from targeted industrial facilities around the world.

When the local column density for the gas of interest exceeds the background concentration by more than our detection threshold anywhere in our 12km x 12km field of view, our retrievals display the emission plumes in a spatially resolved image which makes geolocation of the sources of emissions possible.

We present single pass measurements taken with our satellite in the Short-Wave Infrared (SWIR) band, showing evidence of point source emission plumes (e.g., underground coal mine vents and oil and gas facilities). Third-party measurements and estimates are being used to validate our methodologies.

We also describe time-averaged results combining several single pass measurements which on their own may not have shown emissions, thereby demonstrating the promise of the technique to evaluate persistent emitters with a rate of emission just below the detection threshold of the instrument, for example wide area diffuse sources like landfills.

The lessons learned from the GHGSat-D demonstration satellite have been incorporated into our second satellite GHGSat-C1 scheduled for launch in August 2019. If possible, our aim is to present early results during Phi Week 2019.

At GHGSat, we understand that the end use of our analytics is the opportunity to serve our commercial and public sector customers that make operational decisions for industrial facilities, regulate such facilities or have responsibility for UN Sustainable Development Goal reporting. With this in mind, we introduce some of our innovative products and applications.