Thanks to the advent of open source satellite data coming from the Copernicus and Landsat programs, as well as the emergence of IA technologies, it is now possible to measure economic development indicators from Space, as continental scales, and thus to optimize resources allowance.

Based on that collected data, Earthcube has used proprietary Deep Learning segmentation technologies and Siamese artificial networks to measure the evolution of road infrastructures in South-East Asia between 1995 and 2018 and eventually quantify the impact of the associated development program led by local governments.

Built on a concrete business case, the presentation will go through the technical aspects of the architecture used to support this analysis, and the challenges linked to data collection and labelling but also algorithms’ performances measurements thanks to a comparison with Open-street map data.

Eventually, we will illustrate the positive impact of the technologies used and of open-data policies.

CLEOS is the online Satellite Data Platform to access e-GEOS Geoinformation Data, Products and Services allowing an efficient and cloud-based access to COSMO-SkyMed data (COSMO-SkyMed Cloud Hub) and to the Vertical Platforms.
CLEOS is therefore the online “platform” of platforms providing the backbone of the digital transformation process of the e-GEOS portfolio, in fact, CLEOS is a secure and structured infrastructure allowing the e-GEOS Vertical and Technological Customers to exploit e-GEOS Digital Services and, more in general geoinformation data and capabilities in a modern way.
Moreover, CLEOS provides APIs allowing Premiere Partners to access the full e-GEOS data catalog, image analysis capabilities, and reports producer tools to facilitate the integration and adoption of such tools and Information products into customer-specific business processes. Customers can use CLEOS as a heavy-duty platform to generate Analysis Ready Data or Decision Ready Products.
CLEOS is an open environment that allows the onboarding of Third Party capabilities
CLEOS is fully connected to the e-GEOS Salesforce System to manage users with a secure and API based SSO process, monitor the usage of a single component by transaction schema and finally to manage the billing and delivery process supporting customers in all the step through a success funnel and support team.
As part of the CLEOS development strategy, e-GEOS has set up a partnership with MEEO, an Italian SME focused on Data Cube technology. In the framework of this partnership, an innovative approach to Data Cube Services has been launched under the brand of CUBEO which is offering scalable and affordable massive satellite EO data pre-processing and data access services and, more in general, efficient access to distributed and heterogeneous geospatial datasets.
e-GEOS is the first customer of CLEOS capabilities powering the Vertical Platforms (braint, SEonSE, AWARE, ASKme!, Agrigeo) and to build customized services to respond to the customer requirements in terms of data acquisition, processing, and delivery of geographic information based maps, reports, and indexes or to grant a simpler and immediate access to COSMO-SkyMed data.

In recent years the amount of data collected by remote sensors has accelerated rapidly, complicating manual processing. In parallel, the computer vision field and in particular the object recognition, object tracking, semantic segmentation and motion analysis subfields have improved considerably, due to the advances in deep learning techniques and the popularization of new frameworks by big companies. The coupling of these 2 facts suggests complementing the old earth observation techniques with these new machine learning algorithms to obtain the best of both worlds.

Recent publications about this interaction provide extraordinary results and are laying the foundations for a full automated exploitation of the satellite imagery in the future. Nonetheless there are still many considerations to be addressed for achieving an efficient processing of large area extensions using AI. This processing can be challenging due to the interaction between worlds as dissimilar as the EO and the AI ones. The convergence of these broad disciplines involves huge-data manipulation, geospatial calculations, machine learning inference and training; which have a plus of difficulty when they need to be done in an efficient way, as all of them are computationally intensive and require supercomputing resources.

Thales Alenia Space in Spain has been working on a scalable platform for processing satellite imagery in clustered computing systems using artificial intelligence. During this presentation, the results and lessons learned of the research phase, using Sentinel II imagery will be presented.

Among others, the following topics will be discussed:
• Area trimming: Search areas may involve multiple parts of different products, thus after recovering the original products the search area should be cropped from the union of the resulting EO products to bypass unnecessary computing.
• Zoom adjustments: Appropriate zoom levels should be adjusted before processing the search area with an AI algorithm.
• Parallel processing: As search area could involve huge extensions once the appropriate zoom levels are achieved the resulting image may be splitted in chunks for an efficient processing in parallel. This way, each chunk can be processed separately (if the problem conditions and AI algorithm selected made this possible) by a cluster computing system like Apache Spark and compose later the results of each scan.
• Coordinate system adjustment: Coordinate transformation between the original product coordinate system and the generated image coordinate system must be done before and after processing the search area with an AI algorithm.
As a plus, all the steps of the processing pipeline will be clarified, starting from an area of interest and finalizing with the inferences performed by a deep learning algorithm from this search area.

Scientists rely on Algorithm Theoretical Basis Documents (ATBDs) to understand NASA's data products. ATBDs describe the scientific and mathematical theory underlying the algorithms used to create NASA Earth observation data products, and scientists depend on them to scope research methodologies. While ATBDs are required for every standard NASA Earth observation data product, there is no governing standard format or structure for the ATBDs themselves. Furthermore, while they are technically publicly accessible, each ATBD is relatively isolated. Searching across these siloed resources is limited, and varies depending on the hosting website’s own storage mechanism. Scientists conducting research with released data products need to be aware of the scientific choices made in the creation of those products, as those decisions will inevitably inform and influence any research that follows. Making the provenance of that data more transparent will benefit not only those scientists, but anyone using the results of their research.

We are building the Algorithm Publication Tool to unlock and connect ATBD components by using modern, optimized infrastructure to model and expose the links these ATBDs already share in theory. We have reimagined ATBD documents as structured content, and created a metadata model to describe and define the requirements of that content. To facilitate this new way of assembling ATBDs, we’ve created a user-friendly web application which walks authors through filling out each of the necessary fields. The resulting documents, complete with figures, references, and equations, are made available in a single centralized repository, allowing for search, links, and references across and between ATBDs.

This talk will describe the technical considerations behind reconfiguring ATBD documents as content categorized by a unifying metadata schema, and creating an intuitive interface to construct them. We will also discuss managing the tension between a metadata standard rich and nuanced enough to create readable, searchable documents of this complex type without forcing an authorship process so onerous as to discourage uptake. Finally, we will explore the value of simple, discoverable records of data and algorithm provenance as sources of each proliferate as a part of the movement towards open science in the cloud.

We are enhancing our capacity to retrieve, extrapolate from, and analyze more data from remote-sensing instruments, and scientists are increasingly able and encouraged to share their own algorithms to process and analyze these products. It is vital the movement toward open science in the cloud include improving and expanding our ability to communicate the provenance of both the original data products and the increasing number of resulting ones. The Algorithm Publication Tool will not only make those communications more discoverable and readable, but will also unlock deeper connections between ATBDs and their corresponding data products through a searchable, linked, and reproducible metadata model.

The European Copernicus program provides information derived from earth observation (EO) in an operational manner at planetary scale via its Sentinel satellite missions. The available Sentinel data reach the 14 PB strorage volume by the end of 2018. Approximately 5 PB of EO data was generated in 2018 alone and the growth rate is expected to increase in the future due to the expansion of the programme with additional Sentinel satellites put into orbit and the future sensor development of the upcoming next generation of the Sentinel satellites. Moreover, historical and contributing missions such as the ESA Earth Explorers as well as airborne and terrestial in-situ data contribute to the further demand of storage capacities. Alongside the raw data, higher level data products need to be served to the wider scientific, public and private user community. Aside from its technical attributes, EO is part of a wider technological revolution as society moves into the Information Age, an age defined by Big Data, cloud computing and artificial intelligence. The full potential of these EO data sets are still untapped as they require massive storage and processing infrastructure as well as complex analytical method developments.

The European expertise in the analysis of EO data is at a very high level and wide-spread witin Europe. However, in contrast to the big US IT companies, Europe lags significantly behind with its current IT capabilities. In order to tackle the big challenges in respect to IT provisioning based on European assets the European Commission (EC) started to act. The EC funded the Copernicus Data and Information Access Services (DIAS) in 2018. DIAS has the goal to make Copernicus data accessable worldwide and shall establish a long-term business model for data provisioning and value-adding service provision. Four DIAS projects have been kicked off under the umbrella of ESA, while DIAS number five (WEkEO) is established by EUMETSAT, ECMWF and Mercator Ocean. Additionally, the European Open Science Cloud (EOSC) was launched in 2018, focusing on scientific users with a much wider thematic scope covering data and processing intensive research disiplines. The EOSC has the potential to become a viable European alternative to the US service offerings with a dedicated focus for the scientific EO community. In contrast to DIAS, EOSC is based on a federated approach. Within the current H2020 project EOSC-hub IT resources for the scientific EO exploration are already established within the EOSC.

Next to the afore mentioned activities driven by the EC, the EODC (Earth Observation Data Centre) was established as a public-private-partnership (PPP) in Austria already in 2014. EODC’s ambition is to operate a data centre dedicated to the efficient handling of local, regional and global big EO data. EODC's infrastructure development is driven by requirements of the EO sector, both with respect to hardware and software. The infrastructure encompasses a multi-petabyte storage facility for storing different levels of EO data sets. The storage facility is further interconnected to different processing facilities. Both storage and processing capacities are offered to EODC’s broad national and international partner and customer base including scientific, public, as well as private organisations. Currently, EODC serves 19 partner organisations within a long-term service contract and several further national and international organisations. Similar to EOSC, EODC is following a federated infrastructure approach by connecting different existing infrastrcture capabilities. EODC is currently supporting the following activities:
• EOSC: EODC participats in the H2020 project EOSC-hub, where EODC leads the EO-Pillar activities coordinating the initial EO services provided within the EOSC
• WEkEO: EODC federated with the WEkEO DIAS and offers within WEkEO the access to EODC’s global Sentinel-1 and Sentinel-2 data archive and free (limited) processing resources that can be expanded via a comercial offering (see: https://www.wekeo.eu/dec_2018_update).
• OpenEO: EODC is member of the project consortium of the H2020 project OpenEO (https://openeo.org/) that develops an open API to connect R, python, javascript and other clients to big EO cloud back-ends in a simple and unified way.
• DMA: EODC is a member of the Data Market Austria (DMA, https://datamarket.at/) project that aims to develop a sustainable data-services Ecosystem in Austria.

All these activities shell contribute to serve the different local and global users from different scientific, public and private organisations with a wide range of services. Instead of one single infrastructure the service offerning shall be provided by a federated network of different providers with their specific domain knowledge. EODC’s activities within the federated EOSC cloud have a dedicated focus on the scientific user community. However, even within the scientific community, different user levels, i.e. expert users, advanced and basic EO users, need to be adressed and different tailor-made services are required.

Within this contribution a general overview of the existing EO services integrated into the EOSC will be provided. Moreover, a detailed insight of the EOSC services running on the EODC infrastructure, e.g. the EODC Jupyter Hub service, the data catalogue service and the hosted geo-hazards service operated by Terradue, is presented. Moreover, possible future EOSC service offerings like high-performance computing (HPC) via its integrated and co-located HPC envirement, i.e. the Vienna Scientific Cluster (VSC, see http://vsc.ac.at/) and the Austrian Data Cube service providing analysis ready Sentinel-1 and Sentinel-2 data for the Austrian user community are presented next to a general overview of the EODC activities.

The ESA Network of Resources and EU Open Clouds for Research Environments are examples of initiatives to stimulate the use of existing resources. Clearly, when it comes to selecting a platform to access data and perform processing one could claim that there is an overwhelming number of platforms. For instance, the DIAS platforms offer the Copernicus and other data and processing services, TEPs offer a specific community viewpoint and are evolving to offer services for Analysis Ready Data, collaborative ground segments present a national focus and commercial platforms offer yet other possibilities. At the same time we should recognise there is a growing need for integrating non-EO data or even applications where EO data is just one small part of the total processing chain with the need to perform processing on multiple platforms.

The user community is quite rightly asking which platform should I choose, which platform offers the best match of services to support the development and execution of my application.

For DIAS, EARSC offers a comparison (http://earsc.org/news/dias-comparison) and ESA is benchmarking the performance, but, what the users want is a means to try-out the multitude of platforms without having to invest time and effort to do so. Furthermore, to avoid being locked in to an initial platform choice users want to know how easy is to swap platforms?

In response to the evolving needs of users of platforms to access and use data Space Applications Services has developed and provides the Automated Service Builder Framework (ASB). ASB is a framework providing an application agnostic solution for implementing complex processing chains over globally distributed processing and data resources. ASB provides a “low coding” solution to develop a data processing chain based on orchestrated workflows.

Users define, configure and run algorithms embedded in workflows. ASB provides highly automated methods to register new processes, import and dockerize user defined algorithms, graphically edit workflow definitions, execute processors with user-defined parameters, and access the results. ASB has implemented a customisable ontology-based mechanism that verifies the consistency of the dataflow between processes. Only compatible process parameters may be connected to pass data within workflows. Users do not have to create interfaces because a Web-based interactive interface is automatically generated for inputs not resolved in the workflow.

ASB’s Generic Flexible Orchestration means that processor tasks are orchestrated by the workflow engine independent of the location of the actual executable files and independent of the underlying programming languages and related technologies. User algorithms are packaged, deployed and executed on user-selected platforms which can be where the data is located.

Algorithm developers benefit from ASB’s Process Import Tool that ingests user's own algorithms and libraries to be included in the ASB knowledge base. Imported algorithms can be used by other users providing a platform to co-develop workflows for faster development cycles, easy substitution of modules in processing chains for comparative studies and scalability at processing platform and algorithm levels to tackle big data issues.

Driven by the need to assign a process to a GPU or HPC/HPDA to gain performance improvements ASB allows a user to associate a process in a workflow to a named platform and if necessary to a processing node type. ASB manages the interoperability issues for the user providing to users the possibility to develop and run application truly globally orchestrated across multiple platforms.

ASB is used to federate across multiple resources (NoR, OCRE) or can be integrated into as part of a larger domain specific platform as demonstrated in the H2020 EOPEN project (https://eopen-project.eu).

Our paper will elaborate further on the benefits and use of ASB in EOPEN and include a demonstration. ESA may wish to consider including our paper in a side event if a side event was to focus on platform technology for the scientific community.