The EarthSignature service is the centerpiece of the SnapEarth project.

It produces the data that the pilots of the SnapEarth project will use.

The goal of EarthSignature is to

• periodically download new products from various data provider
• perform precise land cover classification
• make the classification results (Table 1) available through a gRPC API

To make the accessibility of satellite imagery easier, the European Commission funded five cloud-based platforms to access Copernicus data and information.

These platforms are called Data and Information Access Services (DIAS) and are the data provider we use.

In the SnapEarth project, we mainly use Sentinel-2 optical data. Other DIAS platforms such as Sobloo[1], Onda-dias[2] or CREODIAS[3] also offer access to Copernicus Sentinel data.

The SnapEart Project pilots connect to EarthSignature API to download the land cover classification data and integrate it in their professional applications such as EarthPress, EarthAgriculture, EarthClimate or Earth FoodSecurity.

 

^{Figure 1: Architecture overview of the EarthSignature service}

This figure shows an overview of the components of the EarthSignature service. The service is divided into three main parts:

• The Monitor service is responsible for downloading and pre-processing the data. This service will connect to any DIAS daily to download new Sentinel-2 products. For the needs of the project, it only downloads products imaging Europe. The Monitor service is the central orchestrator of EarthSignature. It handles the products since they are downloaded from the DIAS until stored in the database. The downloaded product cannot be used for land cover classification right away. We must ensure the images are appropriately pre-processed (for additional information please refer to D3.2) before being fed to the models, which is also a task of this service.
• The Model service is responsible for efficiently serving the artificial intelligence models used in the service. EarthSignature performs land cover classification on the products using semantic segmentation, detailed extensively in Section 2.2.1. This family of models needs a lot of computing power, and it is the role of this service to efficiently provide access to the models in the most efficient way possible. The models available in the service are:
  • A semantic segmentation model to classify into the CORINE Land Cover level 3 classes.
  • A semantic segmentation model to classify into the CORINE Land Cover level 2 classes.
  • A model that performs burnt areas detection.
• The Database service stores the data and makes them available to the pilots. It is composed of two components, a database for the storage, and it exposes a gRPC API to allow the pilots to query the database. Direct access to the database is not allowed for security and data integrity reasons. The API exposed to the pilots is secured using a private token that ensures only permitted users can access the content of the database.