Spatio-temporal deep learning workflows for transforming remote sensing data into geo-indicators for environmental policy support (GEO-INDICATORS)

Data-driven environmental indicators are necessary to tackle the challenges of the climate and biodiversity crises and to measure progress towards global targets such as the UN Sustainable Development Goals (SDGs). At present, governmental agencies mainly invest in ground-based networks to monitor the status of our ecosystems and the environmental pressures they face. In Flanders, ground-based monitoring networks have been designed recently for habitat quality (2013), species (2016) and abiotics (2019). 
This project aims to develop the scientific basis needed to integrate Earth Observation (EO)-based information into operationable geo-viewers for the Flemish environmental policy sector. EO repeatedly provides spatially-explicit, area-covering snapshots of (a part of) the Earth’s surface obtained in a verifiable and transparent way. The current combination of unprecedented amounts of EO data, increased computational power and recent developments in artificial intelligence - more precisely ‘deep machine learning’ - allows for the development of monitoring workflows of accurate, spatially explicit environmental indicators. 
Generic deep learning (DL) building blocks will be developed that can handle the spatiotemporal data challenges associated with EO data. They will be accompanied by generic cal/val procedures required for training the DL models. Workflows compiled from these generic components rely on free and open EO data sources only. 
Thematically, the project focuses on surface water, vegetation and environmental pressures. Our co-creation approach involves societal actors at multiple steps in the design and the optimization of the workflows and the geo-viewers. This is the best guarantee for an end-product that will be used by the main actors at all levels of government. In a spatially fragmented region like Flanders, where policy instruments are strongly intertwined across sectors and spatial scales, a central platform  is of primordial importance.
EXPERT PANEL SBO-expert panel Applied biological, environmental & earth sciences Main research topic Earth and spatial sciences Secondary research topic Environmental and ecological sciences: the thematic application domain is mainly environmental as we want to develop models for environmental policy. The main research topic is however earth and spatial sciences; as we use satellite data as the main source of input data and the models have specific spatio-temporal characteristics because of that. The workflows we aim at developing can be transferred to other application domains as well.
Non-confidential summary (in Dutch)