The DTE-S2GOS project

Digital Twin Earth Synthetic Scene Generator and Observation Simulator

The Digital Twin Earth Synthetic Scene Generator and Observation Simulator (DTE-S2GOS) project consists in the development of a new component of the Destination Earth initiative to be implemented in the ESA DestinE Platform as a pre-operational service. The primary objective of DTE-S2GOS is to develop a comprehensive and accurate simulation framework that can generate physically realistic synthetic 3D scenes of the Earth and simulate ground-based or spaceborne remote sensing observations, among other with the Eradiate radiative transfer model.

DTE-S2GOS will be designed to operate at multiple spatial scales, from meter-scale locale simulations that include intricate 3D representations of vegetation, cities, and coastal waters to global-scale simulations at kilometre resolution, suitable for large-scale atmospheric modelling. These capabilities will support current and future ESA missions, ensuring compatibility with a broad range of Earth Observation applications. This framework aims to support Earth Observation by providing physically realistic simulations. A key objective of DTE-S2GOS is its seamless integration into the DestinE Platform, a cloud-based infrastructure designed to facilitate Earth Observation relying on digital twin components. By leveraging OVHcloud and Kubernetes, the service will provide researchers with a scalable and easily accessible computational framework, removing the technical barriers associated with software management and high-performance computing demands. This approach will not only improve accessibility but also promote collaboration, reproducibility, and data sharing, aligning with open science principles.

Simulation of a satellite view over a 3D synthetic scene with the Eradiate model: without atmosphere (top) and with atmosphere (bottom).

To maximize its scientific impact, DTE-S2GOS will be developed as an open-source and flexible service, following FAIR (Findable, Accessible, Interoperable, and Reusable) principles. The software architecture will be modular, allowing users to adapt and extend the system based on specific research needs. By maintaining public repositories and comprehensive documentation, the project aims to foster a broad community of users who can contribute to and benefit from its development.

A critical aspect of the project is the demonstration of the service’s capabilities through real-world use cases. These will be carefully selected in collaboration with stakeholders to highlight the practical benefits of DTE-S2GOS. The use cases will include applications such as AI-based satellite image training, retrieval algorithm benchmarking, and satellite calibration/validation, among others. Additionally, the system will undergo rigorous scientific validation, comparing its simulated data with satellite and ground observations from reference networks such as HYPERNETS and RadCalNet. To ensure the accuracy and reliability of the simulations, a Monte Carlo uncertainty propagation approach will be implemented using the CoMet toolkit, enabling a robust assessment of potential errors.

Finally, DTE-S2GOS will establish a roadmap for future enhancements, including the potential integration of active sensors and microwave simulations. This forward-looking strategy will outline the steps needed to transition the service from a pre-operational research tool to a fully operational component of the DestinE Platform. The roadmap will also explore options for long-term sustainability, addressing possible service level agreements and operational funding models. Through these efforts, DTE-S2GOS aims to become a pivotal tool in the Earth Observation community, supporting advanced research and operational services for future ESA missions. By achieving these scientific objectives, the DTE-S2GOS project aims to advance the state-of-the-art in Earth Observation simulations, support the Digital Twin Earth initiative, and provide valuable tools for researchers.

DTE-S2GOS overall design

High-level diagram of the DTE-S2GOS components.

The overall design of the S2GOS service is structured around four high-level concepts distributed across two layers. The first layer is the front end, which consists of the user interface—either a notebook or a UI derived from it. This user interface communicates with the application, a use-case-specific software component. Each application will expose different interfaces tailored to specific requirements.

The application translates user inputs into RESTful requests that are sent to the backend. This communication is handled by a web server, which redirects the calls to the backend portion of the application software. Based on the user input—and any potential modifications introduced by the application—scene generation and simulation jobs are orchestrated and launched

DTE-S2GOS Use Cases

Three Use Cases will be implemented to prototype this pre-operational DestnE service.

Synthetic satellite observations for ACIX-IV

The Atmospheric Correction Inter-comparison eXercise (ACIX) is an international initiative with the aim to analyse the Surface Reflectance (SR) products of various state-of-the-art atmospheric correction (AC) processors. The Aerosol Optical Thickness (AOT) and Water Vapour (WV) are also examined in ACIX as additional outputs of AC processing. The ACIX initiative went already through 3 phases and it now starting the fourth one.

The purpose of this use case is the simulation of synthetic satellite images, e.g., CHIME, at a spatial resolution of about 30 m over different types of land cover types and atmospheric conditions to benchmark atmospheric correction algorithms. The land cover types should include vegetated surfaces (e.g., forest, crop land), urban, arid or polar regions. The benchmarking will be performed in the framework of ACIX phase IV. Different atmospheric properties like aerosol and water vapour concentrations will be considered. Finally, simulation of bottom of atmosphere surface reflectance (HRDF) will also be performed for each pixel.

DTE-S2GOS will be used to simulate TOA reflectance or radiance (orthorectified L1c data) observed by a hyperspectral instrument like CHIME over the selected ROIs. The atmospheric correction algorithms will be benchmarked using these simulated data as input to derive the surface HDRF that will be compared with simulated HDRF. Simulated HDRF will be estimated for each pixel.

Byproducts will be made available during the result analysis such as the total column AOD and water vapour over each pixel. Prior information that should be make available to the users is still to be decided. A maximum of two seasons (winter and summer) will be simulated per ROI Atmospheric conditions will be taken from CAMS data.

Benchmarking of land parameter upscaling methods

Upscaling methods are essential in Earth observation for bridging the gap between high-resolution ground measurements and coarse-resolution satellite observations. These methods vary in complexity and underlying assumptions and can be broadly categorized into several types depending on whether they are empirical, statistical, deterministic, or hybrid. This Cal/Val activity is crucial for meaningful validation and comparison between field measurements and satellite-derived estimates as performed in the context of Ground-Based Observations for Validation (GBOV) of Copernicus Global Land Products. This service relies on a few upscaling procedures will be implemented to provide reference values (the so-called LPs) representative of areas large enough to cover over several pixels of typical mid-resolution satellite imagers. The GBOV service allows the quality control of the main land products Copernicus Gobal Land Service (CGLS) products (top-of-canopy reflectances, surface albedo, fAPAR, LAI, fCover, LST and soil moisture) providing collections of multiple years of ground-based RMs and derived land Products (LPs). Assessing the uncertainties resulting from upscaling method is thus challenging.

The purpose of this use case is to propose a rigorous framework for the evaluation of the uncertainties resulting from ground observation upscaling for land parameters such as surface albedo, LAI or FAPAR over two types of land cover types, simulating both ground and satellite observations.

Satellite image simulation over bright desert PICS

Bright desert Pseudo Invariant Calibration Sites (PICS) play a critical role for the routine monitoring of radiometric quality of EO data. These sites have stable and predictable reflectance properties over time. Traditionally, satellite images are acquired over this site. Pixels surrounding the site are averaged to estimate the mean TOA radiance or BRF and its standard deviation. These averaged observations are next compared with simulated calibration reference relying on 1D RTM. Within this use case, a new approach will be proposed by which pixel values are not averaged, meaning that each pixel individual value is kept.
Data acquired by the Gobabeb HYPERNETS Namibia (GHNA) will be used for that purpose . The site is well characterised as it is very close to an instrument already recognised as a radiometric calibration site (GONA) as part of the RadCalNet network. GONA and GHNA are 650 m apart. These simulated images will be compared with actual ones on a pixel-per-pixel basis, accounting for the uncertainties of both simulation and observations.