Day 4

Detailed paper information

Back to list

Paper title Satellite Rapid Damage Assessment service for flooded and burned areas delineation
Authors
  1. Edoardo Arnaudo LINKS Foundation Speaker
  2. Fabio Montello LINKS Foundation
  3. Federico Oldani LINKS Foundation
  4. Claudio Rossi LINKS Foundation
  5. Maria Luisa Quarta MEEO srl
  6. Marco Folegani MEEO srl
  7. Aitziber Egusquiza Ortega FUNDACIÓN TECNALIA RESEARCH & INNOVATION
Form of presentation Poster
Topics
  • D2. Sustainable Development
    • D2.12 Cultural and Natural Heritage
Abstract text The detection and assessment of damages caused by violent natural events, such as wildfires and floods, is a crucial activity for estimating losses and providing a prompt and efficient restoration plan, especially in cultural and natural heritage areas. Considering major wildfire or flood events, a typical assessment scenario consists of the retrieval of post-event EO-based imagery, derived from aerial or satellite acquisitions, to visually identify damages and disruptions. The challenge of this task typically resides in the complex and time-consuming activity carried out by domain experts. Usually, assessments are produced manually, by analyzing the available images and, when possible, in-situ information. We automated these tasks, by implementing a ML-based pipeline able to process satellite data and provide a delineation of flooded and burned areas, given a specific region and time interval as input. Sentinel-1 and Sentinel-2 satellite imageries from ESA Copernicus Programme have been exploited to respectively train and validate flood and burned areas delineation models. Both the approaches are based on state-of-the-art segmentation networks and are able to generate binary masks in a given area and time interval. An extensive experimental phase was carried out to optimize hyperparameters, leading to optimal performances in both the flood mapping and the burned areas delineation scenarios.
One of the objectives of the Rapid Damage Assessment service proposed here is the detection and delineation of burned areas, caused by wildfire events. Our approach consists of a deep learning that performs a binary classification to estimate the areas affected by the forest fire. The model obtains an average F1-score of 0.88 on the test set. Another main objective of the Rapid Damage Assessment service is the delineation of flooded areas, caused by the overflow of water basins. To tackle this task, we implemented a deep learning solution which utilizes pixel-wise binary classification of an image. Several training iterations of models have been tested, starting from different datasets and architectures and the average F1-score produced is 0.44.
The Rapid Damage Assessment service is currently deployed within the SHELTER (Sustainable Historic Environments holistic reconstruction through Technological Enhancement and community-based Resilience), an ongoing project funded by the European Union's Horizon 2020 research and innovation programme. The project aims at developing a data driven and community-based knowledge framework that will bring together the scientific community and heritage managers with the objective of increasing resilience, reducing vulnerability, and promoting better and safer reconstruction in Historic Areas.
Among the different Copernicus-based solutions developed in the context of the SHELTER project, the above-mentioned services represent the most mature ones, but further developments are foreseen. The different Copernicus core services in fact have already internally the relevant sources of satellite imagery (such as the Sentinels and the Contributing missions), models and in-situ data sources to cover a large part of the user requirements expressed by cultural and natural heritage user communities. Nevertheless, the development of specific products and/or adaptation of existing ones is needed to respond to specific requirements of the SHELTER use cases.