Periodic Reporting for period 2 - CANDELA (Copernicus Access Platform Intermediate Layers Small Scale Demonstrator)
Reporting period: 2019-11-01 to 2020-10-31
Initial objectives of the CANDELA application-oriented platform can be grouped into the following activities:
• Generic Big data analytics building blocks allowing the analysis of large volume of Earth Observation (EO) Data.
• Tools for the fusion of various multi-sensor EO and non-EO data in order to create new applications and services.
• Compatibility between CANDELA and existing European assets and namely the DIAS (Copernicus Data and Information Access Service).
• Develop realistic reference scenarios that demonstrate the platform capabilities and showcases its functionalities to new external users.
According to the final assessment, not all foreseen tools reached same level of relevance at the end of the project. Optical change detection, Data Mining, Data fusion and semantics are clearly scientifically relevant. For CANDELA as a platform this is less clear (technologically), and the same for SAR change detection.
The Jupyter-like interface resulted too limited, and the project should have explored alternative solutions if the resources would have permitted. The need for a graphical interface is something to consider for the future.
The CANDELA platform was developed in the CREODIAS (DIAS operated by CloudFerro) but not ported to Mundi although the effort required to do it has been technically evaluated. There are two major dependencies to the CREODIAS cloud platform: the access to the Openstack API, and the access to the products and catalogue, both were tested and validated on the Mundi platform.
Four main blocks of tools were developed in CANDELA:
• Earth Observation data mining for classification and change detection, allowing users to refine their query by iteratively specifying a set of relevant and non-relevant images.
• Deep Learning for Change Detection on time series for optical and radar Earth observation data. For optical data, it provides generic change detection maps for every couple of images and transform these maps in a temporal curve of more interpretable change indicator. In the case of radar data, it provides a classification of each image and a change detection map for each couple of images.
• Semantic search and indexation on the output of the Earth observation library and non-image data, to allow users to make requests using multi-criteria classification.
• Data fusion techniques to merge pre-processed data that came from various sources, enabling to combine multiple image sources for classification.
CANDELA provides a geoprocessing environment for data scientists, on top of a cloud infrastructure with built-in scalability. Docker and Kubernetes technologies are used respectively as containerization and orchestration solutions. The architecture solution based on containers also allows to enrich the platform building blocks easily with a simplified and standardized integration process. The computation engine deployed on CANDELA platform is Geoserver. All processing services are packaged into Docker containers and made available as standard OGC WPS services. A frontend layer has also been installed (Jupyter notebook environment) to interact with the workflow execution engine.
Apart from maintaining up to date and active the project website (online at http://www.candela-h2020.eu) and the twitter account (https://twitter.com/CANDELA_h2020) additional material consisting of the project brochure and newsletters, were produced in order to efficiently disseminate project results and activities. All 6-monthly newsletters (5 in total) published during the project lifetime are available at project website. Special attention was put as well on tutorial videos and these can be found in the project YouTube channel.
Besides developing the tools, the validation process was an important part of the project. It allowed to catch potential errors and make the necessary changes during the development process. Four so-called use-case scenarios were prepared:
• Urban Expansion and Agriculture, aimed at studying the effect of urban expansion on agricultural areas due to the continuous development of human settlements and climate changes.
• Change Detection in Vineyards – The effects of natural hazards estimation.
• Abrupt natural disasters over the forest vegetation.
• Forest health monitoring.
Land cover and land use changes driven by climate changes and population growth are crucial factors that affect economy, agriculture and decision-making strategies among others. This project sub-use case aimed at closely studying urban expansion and the resulting agricultural surfaces shrinking. To achieve the latter objective, robust and generic data analytics tools and various remote sensing data sources were used to detect changes of interest.
Wine-making is one of the largest agricultural industries in many countries over the world including France. However, natural hazards such as frost and hail cause large damages in vineyards resulting in enormous financial loss. This second sub-use case aimed at using data analytics tools with remote sensing data in order to quantify the damages caused by natural hazards, which is of great interest for winemaking syndicates, farmers and insurance companies.
The abrupt natural disasters use case was expressed by State Forests in Poland after windfall that happened in August 2017. The disaster was a result of a bow echo weather phenomenon that brought winds blowing at 100-150 km/h and sweeping off forest stands on its path ranging from the Baltic Sea coast to the region of Lower Silesia. The disaster affected whole ecosystems and forest stands (rather than individual types and/or species and/or habitats of trees), including 22 natural reserves, 134 “Nature 2000” areas, 15 protected bird habitats and breeding areas covering, inter alia, the areas protected under the Bird and Habitat Directive. In the CANDELA project satellite images were used to train algorithms that could quickly identify areas affected by damage. This would allow foresters to efficiently manage tree cut and restoration of forest.
Complementary to the above, a fourth use case was expressed by people related to forest management and nature protection. Knowledge of the condition of forests and its monitoring is important for many reasons. Forests are one of the most relevant renewable resources, both economically and socially. Therefore information about the state of health of the forest and its ongoing monitoring is so important. Knowledge about the location of tree stands weakened or attacked by diseases and insects, allows for the application of appropriate preventive measures.