Weather events like heat waves or droughts are frequently accompanied by severe air pollution. For example, high temperatures and strong solar irradiation during heat waves favor the production of ground-level ozone (O3) and can thus lead to extreme O3 concentrations posing a threat to humans and the environment. In addition, stable wintertime anticyclones cause extreme particulate matter (PM) concentrations due to increase residential heating and reduced dynamic mixing of the air mass. Hence, high-resolution air quality forecasts and analyses are key for national and regional environmental agencies to understand the underlying phenomena, the cause of air pollution, and potentials to initiate appropriate measures to limit air pollution during such extreme events.

Figure 1: EURAD-IM analysis of the daily maximum concentration of PM10 on 7 February 2023 at ground level. European air quality thresholds require that PM10 concentrations must not exceed 50 µg/m³ on more than 35 days in a year and that the annual PM10 average concentration must nowhere exceed 40 µg/m³.

Use Case Overview

In this context, the DestinE air quality use case developed an interactive air quality forecasting analysis system to assess the population exposure. The demonstrator system implemented a web-based user interface that can trigger high-resolution air quality analyses and forecasts produced by machine learning methods as well as the chemistry transport model EURAD-IM (European Air pollution Dispersion – Inverse Model). The machine learning approaches allow fast and precise air quality forecasts at the location of air pollution monitoring stations by fusing weather forecasts and ground-level observations. At the same time, EURAD-IM provides detailed forecasts and analyses of air quality in Europe on the spatial scale governed by the digital twin. The system further allows for detailed emission modulation illustrating the effect of mitigation scenarios on air pollution.

Figure 2: DestinE Air Quality use case system. Here, ML stands for machine learning.

The use case combined the advantages of machine learning applications and numerical modeling with EURAD-IM. The machine learning modules are based on neural network approaches exploiting longtime time datasets and allowed for air quality predictions at the location of European observation stations and downscaling of coarse resolution numerical predictions to high resolution ground-level fields. EURAD-IM is a chemistry transport model that can be executed to include three-dimensional variational data assimilation to combine the best available information on the atmospheric state. As base for air quality simulations serve, among others, meteorological data from the DestinE Extremes Digital Twin, in situ observations, and high-resolution anthropogenic emission data and temporal emission profiles. The ensemble generation features a modular emission control panel in the user interface that allowed detail modifications of the emission data on an emission sector basis (e.g. industry, road transport, agriculture etc.). All use case components developed are flexible and interchangeable. Machine learning tools were implemented to run on the service front end, while the EURAD-IM were directly coupled to the model data of the DestinE Digital Twin.

Figure 3: Machine learning downscaling for high resolution air quality predications.

To demonstrate the information gained, two demonstrator simulations were delivered to this use case's core users LANUV (North Rhine-Westphalia Office of Nature, Environment and Consumer Protection) and UBA (German Federal Environment Agency). Here, the focus was placed on high-resolution simulations in the areas of North Rhine-Westphalia and Berlin-Brandenburg. The first demonstrator prediction covered an O3 episode in July/August 2018, while the second demonstrator simulation analysed a PM episode in Winter 2016/2017. Both demonstrators were designed in close exchange with the core users to incorporate the required output formats, ease of simulation control (i.e. triggering and output handling) and emission scenario design. These scenarios were available to be selected via the user interface. Besides, a flexible tool to modulate the emissions on a sector and/or regional basis was included in the demonstrator as a prototype. This would also include the modulation of temporal emission profiles to show the potential of, e.g., the implementation of potential stay-at-home orders during air pollution episodes.

Example of a winter-time PM episode (Source: David Karich via Pixabay).