Conference Agenda

We would like to present TRIPS, a web application developed as an ESA 5G for l'ART Demonstration Project in collaboration with the Turin Municipality, Municipal Police aimed at integrating advanced ML/DL technologies to address key challenges in urban safety. It has three core modules:

Road Markings Quality Assessment: Using state-of-the-art Computer Vision algorithms on Very High-Resolution (VHR) data, TRIPS provides a diagnostic of road markings quality through a robust pipeline. A domain-adapted neural network is used to extract a visible road mask, which is then used to assist in training the markings model. This process helps the model concentrating on the key elements of the road. Two other helpful features are that the models have been trained under supervision through the laborious process of manually labeling many VHR pictures, and a customized shadow detector model has been created to manage the shadows.

Car Risk Accidents Forecasting: TRIPS incorporates a sophisticated predictive model capable of forecasting car accidents across the city for the next 5 days by analyzing traffic patterns, weathers and street characteristics like intersection type or number of pedestrian crossings. This approach empowers authorities to identify high-risk areas and implement targeted interventions to prevent accidents, ultimately enhancing overall road safety.

Drone-Based Accident Surveillance: During the pilot phase, TRIPS integrates a module that harnesses the capabilities of drones deployed by municipal police for accident surveillance. This module facilitates real-time collection of images and data at accident sites, enabling prompt emergency response and comprehensive post-accident analysis.

In terms of accident prevention and emergency response, TRIPS offers a paradigm change in urban safety management. Through cooperation between the Police, the Turin Municipality, and ESA, TRIPS is an example of innovation using EO data and ML technology to address smart mobility in urban areas.

The African Development Bank (2014) reports that 53% of African roads are unpaved, with these roads being vital for the continent's economic and social development but requiring efficient maintenance strategies to remain motorable. This study, supported by the European Space Agency, responds to the challenge of irregular and inaccurate traditional condition surveys by introducing an innovative, cost-effective machine learning (ML) solution aimed at aiding local road authorities to monitor and plan road maintenance more effectively. This research builds on earlier initiatives by TRL, which demonstrated the efficacy of classical ML models in analysing Tanzania's unpaved roads through high-resolution satellite imagery. Despite the potential economic advantages of the proposed ML methods over traditional techniques, stakeholders considered high resolution imagery to be expensive. This trial involves leveraging both medium and low-resolution satellite images to assess road conditions in Madagascar and Malawi to make significant savings on imagery costs. Our approach involves Multimodal ML using classical models trained on values derived from image statistics. Our findings indicate that Multimodal ML achieves commendable accuracy (87%) with high-resolution imagery, which declines by 8% and 7% for medium and low resolutions, respectively. This study underscores the potential of ML technologies to significantly enhance the assessment and maintenance of unpaved roads through optical satellite imagery analysis, presenting a promising path for cost-effective road management strategies.

The PRIMARY project aims to enhance air quality monitoring, especially in urban areas, using data from the PRISMA (PRecursore IperSpettrale della Missione Applicativa) mission. By analyzing PRISMA's hyperspectral data, the project seeks to understand atmospheric aerosols content and composition, crucial for assessing environmental and health impacts, particularly in cities. Spatial resolution of PRISMA data (30 m) and artificial intelligence play a key role in overcoming challenges such as spatial resolution limitations and the complexity of the inverse problem in satellite-based atmospheric studies.
To train neural networks for estimating aerosol characteristics using PRISMA data, a synthetic PRISMA-like dataset was generated with support from the Copernicus Atmosphere Monitoring service (CAMS). The training dataset generation exploited the libRadtran radiative transfer model. Trained AI modules are then applied to PRISMA images to produce atmospheric aerosol products.

Field campaigns were conducted in Rome (autumn 2022) and Milan (winter to summer 2023) to validate the PRIMARY project's outcomes. Moreover, drone measurements are being integrated to support validation activities. Preliminary results are encouraging and seems aligned with ground based measurements.
Data fusion products combining PRIMARY results with air quality data from missions such as Sentinel-5P, Sentinel-3 and GCOM-C are also under consideration.

In many cities the population is exposed to elevated levels of air pollution. Often, the spatial distribution of local air quality throughout urban areas is not well known due to the sparseness of official monitoring networks, or due to the inherent limitations of urban air quality models. Satellite observations and emerging low-cost sensor technology have the potential to provide complementary information. An integrated interpretation, however, is not straightforward. The CitySatAir project (part of ESA’s EO Science for Society program) investigates how satellite data of atmospheric composition can be better exploited for monitoring and mapping urban air quality at scales relevant for human exposure.

Focusing particularly on the nitrogen dioxide product provided by the TROPOMI instrument on the Sentinel-5P platform, we investigate different approaches for combining this data with other information such as from models and air quality monitoring stations. We choose four contrasting study sites across Europe (Madrid, Oslo, Rotterdam, Warsaw) differing in size, pollution levels, dominant emission sources, and cloud cover.

For Oslo and Warsaw, we use the Sentinel-5P NO₂ data in conjunction with the urban dispersion model EPISODE to bias-correct the underlying bottom-up emission dataset. The results indicate that, when the model is run with the satellite-corrected emission dataset and validated against air quality monitoring stations, the model error (RMSE) decreases for all stations by up to 20%. The updated model dataset is then used to assimilate observations from monitoring stations and low-cost sensors. In addition, we exploit the synergy of TROPOMI and EPISODE data by deriving surface NO₂ data and carrying out geostatistical downscaling to provide a satellite-based surface NO₂ dataset at scales relevant for human exposure.

For Madrid, Rotterdam, and Warsaw we developed a versatile urban dispersion model able to calculate both surface concentrations of NO₂ at street level and NO₂ column concentrations matching the TROPOMI observations. Urban emissions are described by proxies taken from open data, where emission factors are updated periodically to best match the observations from either ground or space. Compared to the CAMS regional ensemble, local biases are reduced considerably, especially if in-situ measurements are also assimilated in the simulated concentration fields using optimal interpolation.

The multi-annual reanalysis of hourly urban air pollution concentrations at street level provides a very rich data set, which demands special user-friendly tools for exploration and analysis. The data sets for the different cities are showcased in the Lobelia Explore viewer. Lobelia Explore is based on a serverless architecture: as a result of the user interacting with the viewer, the web application requests air pollution data from the cloud as static files and uses this data to render maps, display charts and aggregate data over user-defined areas, all of this browser-side. This architecture eliminates the need of on-demand data processing and reduces maintenance costs.

This work describes preliminary attempts aimed at creating a dataset of daily averages of aerosol optical depth (AOD) on an intraurban scale (300m) using MODIS MAIAC AOD, SEN3 SYN, and AOD from ERA5 reanalysis models. Our preliminary efforts were aimed at understanding the quality of available AOD products by comparing them with daily average measurements provided by the AERONET network for the Italian peninsula during the reference period 2019-2023. MODIS MAIAC AOD proves to be state-of-the-art in satellite AOD reconstruction, while SEN3 SYN correlates less and shows a significant bias when compared with AERONET.

Our efforts are oriented in two directions: a) evaluating whether SEN3 SYN is mature enough to deliver unbiased AOD products on an intraurban scale on a daily, weekly, and monthly basis, possibly using other sources of information such as DEM, latlon, and LST; b) performing data imputation of missing observations using AOD from reanalysis models such as ERA5.

Regarding AOD correction on an urban/intraurban scale, we are evaluating pixel-based approaches such as linear/nonlinear/GAM regression algorithms fueled by the combined use of SEN3 SYN, MODIS MAIAC, ERA5 AOD, and auxiliary data such as MODIS land surface temperature and climatic data.

Our findings demonstrate that there is room for further improvement of AOD products by imputing missing AOD values and by further calibrating AOD using regression models fed with available AOD estimates and auxiliary data.

This work is part of a collaborative project funded by ASI and called APEMAIA (Assessment of PM Exposure at the intra-urban scale in preparation for the MAIA mission). The project is designed to investigate the potential of MAIA by developing a multi-modular system for extracting PM concentrations at the intra-urban scale using Artificial Intelligence techniques.

The environMENTAL EU project aims to investigate the impact of major global challenges on mental health and brain health across the lifespan, including climate change, urbanization, and psychosocial stress. The project also seeks to develop prevention techniques and early interventions in this context.

Earth observation data is used to provide a comprehensive set of spatial information layers that may influence mental health and behavior. Our environmental datasets focus on urbanicity, greenness, water bodies, and elevation information, which will be adjusted to the geographic regions of the studied cohorts and linked to geographical positions and their corresponding data. We examine the relationship between environmental factors and mental health, utilizing global datasets such as the TanDEM-X Digital Elevation Model (DEM), the World Settlement Footprint (WSF) 3D data, night time lights data, local sun incidence angel corrected sun energy data, multi-spectral data, atmospheric data (NO2, SO2, CO, O3, and CO2 concentrations), cloud cover, air temperature, precipitation, and air pollution data.

This work presents first results from combined analysis of UK Biobank cohort data and spatial urban neighborhood metrics on mental well-being indicators such as the neuroticsm score. The neighborhood metrics are calculated for suitable geodata within different diameters ranging from 300 m to 7500 m to include spatial context information in point analysis. From geodata, the corresponding values are extracted at the patient coordinates of the cohorts and a cohort population density normalized analysis of defined bins above UK Biobank cohort data value thresholds is performed. First results indicate correlation of green indices, night time light data and building volume metrics with mental health scores within the UK Biobank cohort data. We hypothesized that environmental factors could also serve as proxies for the social environment and significantly influence mental health, particularly when coupled with the presence of urban green and blue spaces.

Overall, our research provides first insights into the link between geo-environmental factors and mental health outcomes, providing valuable information for policymakers, urban planners, and public health professionals aiming to create healthier and more sustainable living environments.