DTE-S2GOS 3D Synthetic Scene Generator

Mon, 01 Jan 0001 00:00:00 +0000

General approach

The 3D Synthetic Scene Generator (SSG) is one of the core components of DTE-S2GOS. It relies on scene configuration parameters and external datasets to generate a scene description and the associated 3D assets. The level of realism and detail of the generated scenes depends on the availability and quality of the input information.

The SSG leverages a variety of global datasets to provide worldwide coverage. Users can specify any valid geographical location using latitude and longitude coordinates. The generator then produces a scene description file and the necessary 3D assets, including terrain meshes and vegetation, which can be used to reconstruct the scene within a radiative transfer model. The SSG also supports the generation of seasonal scene variants, particularly summer and winter configurations.

Schematic of the scene components.

Scene Components constitute the various parts of the 3D scene. All components have an impact on the radiative signal and must therefore be represented to produce a faithful representation of the earth. Those components are represented by 3D assets that are generated by the Resources. The identified Scene Components that are represented by the SSG are:

Ground:
- Topography
- Vegetation and Bare soil
- Snow and Ice
- Built-up areas
Water:
- Coastal water
- Open water
- Inland water
Atmosphere:
- Molecular Atmosphere
- Aerosols
- Hydrometeors: Ice and water clouds

Realistic atmospheric conditions are incorporated using data from the Copernicus Atmosphere Monitoring Service (CAMS). The scene description integrates the atmospheric and surface components into a consistent representation, enabling coupled atmosphere–surface radiative transfer simulations rather than relying on approaches that treat these components separately.

Users can customise the scene generation process through a set of configuration parameters, allowing the resulting scenes to be adapted to specific scientific requirements and use cases.

Scene configuration

The configuration objects are the main way the user will parametrize their scene generation. Here is a non-definitive list of configurable objects:

Location and Date: includes the location and date to be generated, along with target, buffer, and background sizes.
DEM Config.: includes the elevation dataset and related parameters.
Landcover Config.: includes the landcover dataset, corresponding materials, and related parameters.
Atmosphere/Cloud/Aerosol Config.:
- Atmospheric Profile datasets if provided or standard profiles and discretization.
- Cloud location, density and phase data
- Aerosol location, density and phase data.
Vegetation Config.: Includes tree meshes, statistical datasets or parameters, and placement methods.
Water Config.: includes water masks, water properties etc…
Built-up Config.: includes building/road footprint and building/road generation parameters.
User Asset Config.: includes additional user assets to be added to a scene, e.g. a HYPSTAR mast.

Input data

SSG relies as much as possible on open licence data sets.

Type	Data set
Topography	Copernicus GLO-30 DEM
Land cover	ESA WorldCover 2021
Atmosphere composition	CAMS EAC4
	CAMS EGG4
	CDS ERA5
Buildings	OpenBuildingMap
Buildings and roads	OpenStreetmap
Optical properties	ECOSTRESS spectral library

Scene description generation

Scene generation involves a sequence of processing steps, each progressively increasing the complexity and realism of the generated scene.

Topography

Once the region of interest has been selected, the first step consists of preparing the terrain topography. By default, the Copernicus GLO-30 DEM is used for this purpose.

Extraction of the Digital Elevation Model above the selected area of interest.

Land cover

Land cover type extracted from ESA 10m WorldCover 2021 is projected next on the terrain. This data set distinguishes only 11 different land cover types.

Application of the land cover over the topography.

Buildings

Building information is obtained from OpenBuildingsMap, a global building dataset that includes building heights and occupancy types. It integrates information from several sources, including OpenStreetMap, Google Open Buildings, and Microsoft ML Building Footprints.

Adding the buildings.

Roads and train tracks

Roads and railway tracks, where present, are extracted from OpenStreetMap and incorporated into the 3D scene. The terrain topography is subsequently adjusted to account for their presence.

Adding roads and train tracks.

Water mask

The presence of water bodies is initially determined from the land cover raster dataset. This information is subsequently refined using vector-based water masks extracted from OpenStreetMap.

3D vegetation

Vegetation architecture is subsequently incorporated into the 3D scene based on the vegetation land cover type. For forested areas, representative tree architectures and their structural characteristics are derived from terrestrial laser scanning (TLS) data. Users can customise several vegetation parameters, including tree species, tree density and tree height.

Adding 3D vegetation.

Spectral Angle Mapping (SAM)

Optical properties are assigned to vegetated areas using a spectral angle mapping (SAM) technique based on the analysis of Sentinel-2 observations.

Adding roads and train tracks.

Buffer area

To account for adjacency effects, a buffer zone is added around the selected region of interest. This surrounding area is represented at a lower level of detail to reduce memory requirements.

Buffer area.

Atmosphere

The last step consists in adding an atmosphere above the scene. The properties describing the atmospheric composition and its vertical structure are taken from CAMS data corresponding to the time and location of the simulations

Atmosphere.

Observation Simulator

Mon, 01 Jan 0001 00:00:00 +0000

General concept

The DTE-S2GOS simulator relies on Eradiate, a state-of-the-art 3D radiative transfer model, to accurately simulate observations acquired by satellite- and ground-based radiometers. It supports both multispectral instruments, such as Sentinel-2/MSI, and hyperspectral instruments, such as CHIME and HYPSTAR-XR, offering users a high degree of flexibility in configuring instrument characteristics and observation conditions.

The simulator is also being extended to generate Level-2 products. In particular, satellite simulations can provide the reference (“true”) surface reflectance, making DTE-S2GOS a valuable tool for developing and validating retrieval algorithms.

For certain instruments, such as HYPSTAR-XR, the simulations achieve a high level of geometric detail, accounting for characteristics such as the circular field of view (FOV) and the presence of the instrument’s supporting structure. These features may influence the recorded radiance and are therefore important for accurately reproducing real measurement conditions.

Satellite observations

DTE-S2GOS supports the simulation of multispectral and hyperspectral satellite observations, accounting for instrument characteristics such as radiometric noise, spectral response functions and the point spread function. Simulations typically produce Level-1C images, with orthorectification applied when requested by the user.

S2GOS currently supports the following radiometers:

Platform	Radiometer
Sentinel-2 A,B,C	MSI
Sentinel-3 A,B,C	OLCI
EnMAP	HSI
PRISMA	HYC
CHIME	CHIME
New radiometers can easily be added.

Simulated Sentinel-2/MSI RGB image of the Frascati (Italy) region of interest.

Ground-based observations

S2GOS allows the simulation of different types of measurement techniques/radiometers:

Digital Hemispherical Photography (DHP)
HYPSTAR-XR instrument
Photosynthetically Active Radiation (PAR) sensor

Simulated Digital Hemispherical Photography at the San Rossore ICOS test site close to Pisa (Italy).

DTE-S2GOS components | S2GOS