On this page you can find all publicly available deliverables and reports from GEM project, once they have been accepted.
Deliverable 2.4 - Meteo/climate service
Model data and measurements are made available to GEM project and public through meteoblue dataset API and measurements API, described in this report. An adapter has moreover been developed to access meteoblue data directly through the Sinergise eo-learn library, bringing Earth Observation and weather data together. A temperature downscaling to 10 m resolution for built-up areas has been developed.
This deliverable provides moreover a description of:
-
new measurements sets added during the course of the project and the implemented gridding methodology;
-
description of the eo-learn adapter;
-
new functionalities added to the dataset API to serve machine learning applications;
-
gridded operations designed to seamlessly access multi-resolution models;
-
using the meteoblue advanced API to obtain climate variables;
-
API access to high-resolution (10m) temperature fields in built-up areas.
Deliverable 2.6 - DEMO datasets
This deliverable provides an overview of the data, available within Global Earth Monitor (GEM) for the areas of interest within the project use cases and demonstration activities. Data and data management plays an essential role within GEM and was defined in the first phase of the project to ensure a high level of data quality and accessibility for end users and stakeholders and to enable the application of machine learning techniques.
In general, several types of data are used in GEM. Within this report we categorized the data as Earth Observation (EO) data, weather/climate data, and EO-derived data. We provide an overview of the different data used in the projects use-cases and present a collection of Python notebooks showcasing how to access said data with eo-learn, which is the main building block of the GEM framework.
Deliverable 2.5 - Integration of VHR sources
New data gateways have been developed to allow the ingestion of very high-resolution data via Sentinel Hub. A list of supported sources is:
• Planet's PlanetScope data
• Airbus's Pléiades data
• Airbus's SPOT data
• Maxar's WorldView data
The deliverable shows a detailed description of how to use, purchase, order, and access data for the supported sources. One or more images are added for each supported source, with an example of how the data looks like. A tutorial on how to search and order data from third-party providers such as PlanetScope or Pléiades using Sentinel Hub API with Requests Builder is also added.
Deliverable 3.5 - Front-Office Services
This deliverable reports on Front Office Services exposing the new capabilities of Sentinel Hub services, and GEM framework building block – eo-learn and eo-grow.
The user-facing services are linked mostly to usage of GEM framework or parts of it with Python, be it via scripts (using eo-learn and eo-grow), or Jupyter notebooks (standalone/local or via cloud deployment on Euro Data Cube - EDC). Additionally, the report outlines the QGIS Sentinel-Hub plugin, as an extension of the popular GIS tool facilitating access to Sentinel-Hub datasets and Adjustable Data Cubes.
Several user-facing services, connected either to Sentinel Hub services, or meteoblue weather API, are not part of this deliverable, but are constantly being improved to facilitate the new/updated functionalities of the services. EO-Browser and Request-Builder should be mentioned explicitly, as they bring most of the Sentinel-Hub functionality to the user as web browser applications.
Deliverable 4.2 - Cloud masking
As part of the Machine Learning Package (WP4), this deliverable (D4.2: Cloud masking) demonstrates the development and implementation of cloud masking algorithms that were included in eo-learn to enable pseudo-probability cloud masking of Sentinel-2 data. The cloud masking technique is integrated into the Sentinel Hub's pre-processing chains and made readily available for the complete Sentinel-2 archive.
We have upgraded the eo-learn library with two cloud masking algorithms: s2cloudless for single-observation cloud masking and InterSSIM for multi-temporal cloud masking. The deliverable also shows how the s2cloudless model is used to produce cloud masks and accompanying cloud mask pseudo-probabilities accessible directly through Sentinel Hub service. This gives users a more streamlined access to the data, directly suitable for value added services, since they do not have to deal with cloudy pixels anymore.
For a quick demonstration of the cloud masking capabilities, reader can go to EO-browser application using this link http://bit.ly/eobrowser.
Deliverable 4.3 - eo-learn Gateways
In this deliverable we show the integration of eo-learn with popular machine learning and deep learning frameworks. In particular, we demonstrate the integration with conventional ML frameworks already in place, and is being used extensively.
From the numerous deep learning frameworks, we explain why we decided to interface eo-learn with two: TensorFlow and PyTorch. In order to avoid dependency, maintenance and implementation issues, we have decided that the gateways – interfaces between eo-learn and deep learning frameworks – will be developed as standalone packages. The integration with eo-learn and implementation of a number of deep learning models using TensorFlow framework has already been released in the eo-flow package. In the continuation of the Global Earth Monitor project, PyTorch interface will be added in a similar fashion, extending eo-learn workflows to another large deep learning community.
Deliverable 5.7 - Demonstration
This deliverable provides an overview of the GEM demonstration phase. During this phase, all 5 GEM use cases were run on the selected DEMO domains. For each use case scope area, the ML model, input data and prediction results are provided.
A large-scale, cost-effective continuous monitoring service is also presented in this deliverable. The Continuous Monitoring service aimed to explore the use of the GEM framework for continuous monitoring, with a specific focus on water monitoring in the Sahel region as a showcase.
The results collected during the demonstration phase are also visible via dedicated viewer and/or client applications. In addition, 4 demonstrators from meteoblue are presented.
The report also provides a comprehensive list of the public collections shared by the GEM Consortium that represent the results of the demonstrations.
Deliverable 6.1 - GEM portal
The Global Earth Monitor portal will make all available public products and documents accessible to a wider user community than those already involved in the project. Sinergise has set up the project website https://globalearthmonitor.eu and will later manage it throughout the project.
