5th round of Open Calls: Meet the Winners!
Meet the winners of the 5th Open Call for Micro-Projects, a new additional round that has been launched in the context of the AI4Copernicus Open Calls! The 5th Open Call for Micro-Projects is addressed to Technology-advanced SMEs, start-ups, micro-enterprises from EU Member States and/or Associated Countries (single-beneficiary micro-projects) with a total funding up to €30.000 for each micro-project (10 micro-projects to be funded / the end output should be at min TRL-5).
AQQA – CAMS Air Quality Question Answering
Abstract: Recent developments in chatbot technologies and the reports about ChatGPT underline the potential of language as an interface to technology. The AQQA project aims to adapt and improve our earlier heatwAIve developments. In the heatwAIve project we developed an AI-based voice assistant to inform citizens about the risk of upcoming heatwaves and high air pollution levels and to provide them with personalised behavioural measures. The related information was derived from the Copernicus Atmosphere Monitoring Service (CAMS) and other data sources. However, the use of Amazon Alexa technology induced technical and especially privacy limitations. In AQQA we will investigate the use of the Health Bootstrapping service for downscaling CAMS AQ data. This will deliver AQ information at unprecedented detail. Moreover, multiple tools for linked geospatial data for language-based user interaction will be tested. The use of linked geospatial data enables us to develop a service that is hardware independent and especially webbased. This increases the potential group of users. Independence from legacy providers allows us also to develop tailored health services with personalised behavioural recommendations which is, to our knowledge, still not available on the market.
FLORA4COP – Flora and fauna data analysis with AI algorithms enhanced with AI4C services
Abstract: 3Bee aims to use AI4 Copernicus services for enhancing its tool FLORA. FLORA, supported in R&D phase by the European Space Agency, is an earth observation application for terrestrial biodiversity mapping. FLORA is an innovative and quantitative method to define changes in biodiversity in a precise, scalable and continuous way. 3Bee uses different devices to measure terrestrial biodiversity by monitoring flora and entomofauna (considering insects, especially pollinators, as a proxy for terrestrial biodiversity): IoT devices (Hive-Tech – for monitoring the health status of honeybee colonies – and Spectrum – for monitoring and census of fauna starting from the analysis of the sound spectrum); Satellite images (FLORA). For the time being, 3Bee measurement costs are nearly 200 €/ha (hectare), since it is based on high density IoT installation. With ESA, 3Bee developed a mechanism to classify raw images with different supervised learning algorithms (SGD Classifier, Random Forest, Logistic Regression, Deep Learning classifier). From this processing, 3Bee achieves 4 indexes: land use, land cover, vegetation diversity and nectar potential with high resolution (10X10 m) and high frequency of update (1 month). The next step, in order to improve FLORA and complete pollinator mapping, is to correlate images with an index that shows the likelihood to find pollinators in a given area (pollination abundance index). With these improvements, 3Bee would be able to lower biodiversity monitoring costs to 30-40 €/ha, reducing the number of sensors installed, and using artificial intelligence and satellite technology. To reach this goal 3Bee needs to perform a training phase of a neural network that predicts pollination abundance index starting from Satellite data (land use, land cover, vegetation diversity and nectar potential). By having access to AI4C services, 3Bee may improve its algorithm training model and make use of a time series database of Sentinel 2 data in the case of crop fields. For this reason, 3Bee shows interest in two AI4C’s services: 1) “Deep network for pixel-level classification of S2 patches”, in order to test new ways of training a Sentinel 2 patch pixel level classifier; 2) “TimeSen2Crop”, for analyzing a different topic (crop fields instead of other types of plants used by pollinators) and inspiring new opportunities for the development of new products and services based on Sentinel-2 data. 3Bee already has a network of 4000 IoT sensors installed on the field that would represent the backbone for training the prediction model for pollination abundance index. This data would be the perfect match with AI4C services. Biodiversity mapping enabled by satellite would set a new benchmark in terms of pollinator data availability for EU researchers and policy makers. From a market potential, our services would be of interest for different verticals, interested in having a scalable and precise service to measure biodiversity: Agriculture, energy, infrastructure, nature based solution.
LIFT Sentinel – LIFT Sentinel AI Terrain Detector
Company: Flycom Technologies d.o.o.
Abstract: This project aims to develop an automated classification system for satellite images using deep learning to identify water, urban, rural, and forest areas. The system will be integrated into the existing HazMap module in LIFT software and validated using real-world datasets. The project is led by Flycom Technologies, a company specializing in spatial data and location intelligence, and will utilize the AI4Copernicus service “Deep network for pixel-level classification of S2 patches” for training the deep neural network. The project team will also evaluate the usefulness of the tool for other potential use cases. The proposed system has the potential to improve the accuracy and efficiency of land use mapping and support environmental monitoring and management efforts.
NOEMI – high resolution NO2/NO prEdiction using Machine learnIng
Abstract: WaltR’s mission is to fight against Climate Change and improve Air Quality. To enable policies’ and economical transition towards this end, it is imperative to have adequate and valid information. NOEMI project will use/test AI4Copernicus services to propose an affordable gap filler between S5P/TROPOMI satellite data and local in-situ measurements to provide hourly maps of near-surface NO2 and NO concentrations at high resolution at the regional scale. Moreover, it will demonstrate that NO2 and NO fields are consistent with the so-called “indicative measurements” uncertainty level of air quality directive 2008/50/EU.
AIMPSI – Assistant for the identification, maintenance and planning of solar installations
Abstract: We are living an exponential growth of distributed solar installations. However, there is still a gap of assistant tools for determining optimal locations for new assets, detect maintenance needs, or assess the impact of growth in the power network. In this project, we propose a new method for solving these issues by leveraging both single and temporal satellite images, along with other structural and weather data. In comparison with conventional methods, it enables a simpler, proactive and cheaper approach to promote, adopt and provide higher predictability and planning capacity for future power networks.
SemiLake – semi-supervised representation learning-powered urban lakes and algae monitoring system
Company: Raniarose Technology Limited
Abstract: Urban lakes are natural places for citizens to relax and enjoy mindfulness. A key obstacle is their algae blooms which have harmful impacts on such in-land water bodies. Particularly, algal blooms in urban lakes not only cause unpleasant odours and health risks, but also negatively affect the economy. Here, we aim to develop a novel machine learning pipeline for urban lakes and algae monitoring. This will be based upon state-of-the-art semi-supervised contrastive learning, Sentinel-2 MSI data and AI4Copernicus preprocessing services. In our experiment, we will evaluate this new approach on multispectral image patches of urban lakes in Ireland and UK by comparing it with a baseline machine learning model.
SandMap – SandMap system enabling AI and EO technology
Company: Sense Space Informatics
Abstract: SandMap is a map simulation interactive platform which incorporates a series of exercises for students to familiarise themselves with geomorphology and geography. The student may reform the sand and the system responds in real time to show the geographic information adapted to the new morphology of the sand. As an educational tool, it promotes geographic awareness and will prove beneficial for a wide range of applications like natural disaster and emergency management, public safety, urban planning, etc. THe integration of AI4Copernicus services will facilitate three areas: The “Sentinel-2 Change detection”, service e.g. (data before and after a fire or flood) will be utilised to design a series of exercises in order to prove on the consequences of fires or floods to the environment and society and increase the awareness of the students for the environment protection and climate change and being responsible citizens. Another series of exercises for high-school students will be developed for olive-grove and vineyard crops classification in the area of Crete utilising the “Deep network for pixel-level classification of S2 patches” service. Finally, the “Long Short-Term Memory Neural Network for NDVI prediction” service will be also utilised to help students estimate metric indices such as NDVI which aid the farmers and the organisations of the agriculture sector to make refined decisions for the crops of their local area. The market potential and services of our product will be sky rocketed, since AI enabling technologies, EO and Copernicus services will provide a new area for education and spatial intelligence establishment.
THRUST-4RESST – Remote Sensing via Satellite Technology
Abstract: Forests are a vital component of Europe’s natural environment, providing a range of ecological, social, and economic benefits. Covering more than 1 billion hectares (around 46% of land area) within the EU, forests have a strong impact on biodiversity conservation, carbon sequestration, water management, recreation & tourism, and, of course, timber production. However, forestry resources are being continuously challenged due to deforestation (both legal and illegal), damage by wildfire and other natural causes (such as flooding, storms, etc.), as well as inefficient, manual maintenance & control processes that cause serious time lag in decision making. This lag in decision making is caused by lack of current technologies ability to provide large-scale, high-resolution data at high refresh frequency: while satellites can provide large-scale data with high-frequency, the resolution is not sufficient to detect small changes, like forest pest hotspot in an early stage. Needed resolution can be provided via means traditional aviation, however due to high cost such data is gathered every 3-4 years, which is too long interval for timely decision making. With the help of AI4Copernicus we will bridge the gap between high-resolution, low-frequency aerial data and low-resolution, high-frequency satellite data to create a decision support system for foresters.
AI4EW – Artificial Intelligence for Early Warning
Company: GECOsistema SRL
Domain: Environmental, Energy
Abstract: Real time flood and water levels forecasting in the next hours/days across rivers is crucial for flood prevention and civil protection activities, as recent flood events in Germany in 2001 as clearly demonstrated . Europe has a dense monitoring network of gauging stations, that could be coupled with rapid flood modelling to quickly obtain flood scenarios and support civil protection activities. With this goal, our company develops and maintains the global flood mapping platform SaferPlaces , that provides flood risk intelligence based on Geospatial, Satellite, Climate Data and AI-based models combined into a cloud computing environment. The platform democratises access to flood risk intelligence to a wide range of decision makers, including non-experts users, professionals, urban planners, insurances, multiutility companies, promoting the transition of cities towards greater resilience under current and future climates. It is already a commercial platform that exploits both Data As A Service (precomputed flood maps) or Software As A Service (capability to perform on the fly flood simulations) business models. Our platform could benefit considerably by integrating a data driven forecast of incoming floods for river gauging stations, to run flood simulations in advance and offer added value the civil protection authorities /first responders, for example in improving early warning and Disaster Risk Reduction DRR which is among the priority areas of the Copernicus program. We want to test AI4 Copernicus bootstrapping services, particularly those providing access to ERA5 datasets from CDS (from which we will retrieve also forecast of meteorological forcings), and LSTM algorithms (short term times series forecasting services, to be adapted from other domains, provided among I-NERGY Services), to produce local forecast of incoming water levels to a generic river section, where water levels are monitored and publicly available, assessing performances and accuracy of the provided forecast for flood mapping applications. The forecasted river water levels can be easily converted by our proprietary raster-based flood models in flood hazard and damage maps at high resolution exploiting the available DTM/DEM (Copernicus DEM) or Lidar datasets.
AI-quaFarm – Optimising Fish Farm Location Selection using AI
Company: Heuristic Data
Abstract: AI-quaFarm’s goal is to showcase how the use of AI can assist on the process of selecting optimal locations for the installation of new aquaculture sites. The proposed methodology aims to test different AI4Copernicus services that will enhance the evaluation in different levels. Satellite product pre-processing and harmonisation is used, along with training and usage of AI models for image classification and fine-grained air quality data generation that will be incorporated into the evaluation process of new areas of interest.