CyI scientists develop a deep learning framework that leverages multiple data sources for improving air quality forecasts

Cyprus Institute scientists, including CARE-C Associate Professor Theodoros Christoudias, have presented a novel deep learning framework that fuses multiple data sources in order to improve accuracy in air quality forecasts – leading to a wide spectrum of downstream applications that carry significant impact to public health and policy. The work has received a Best Paper Award at the 2022 Workshop on Machine Learning for Earth Observation (MACLEAN’22), held in conjunction with the European Conference on Machine Learning ’22 (ECML/PKDD).

Air pollution is detrimental to human health and its contribution to the global
burden of disease is now well known. Accurate prediction of the local surface concentrations of atmospheric pollutants is the key to mitigating these harmful effects on human health and the environment.

Motivated by the challenge, the CyI team has developed a deep learning framework that leverages publicly available data sources including satellite observations, ground station measurements, and elevation and land-use maps in order to improve the modelling accuracy of atmospheric gas and particle pollutants by up to 57%.

Leveraging these results, the Cyprus Institute spin-off company GAIA is creating a Hybrid Geospatial Artificial Intelligence Analytics platform, aimed at providing robust solutions that facilitate critical decision-making supported by data-driven evidence to enhance fundamental operational tasks in the insurance and adjacent industries.

This work has been supported by the Cyprus Research and Innovation Foundation under contract number PRE-SEED/0719/0042, and a grant for computational resources on the Cyclone High Performance Computer at the Cyprus Institute.

Figure 1 description: Prediction of pollution concentrations from EU state-of-the-art model CAMS (left) compared to new method (right) over North Italy (center). Circles indicate ground-station accurate readings.


Mihalis Nicolaou, Computation-based Science and Technology Research Center
Theodoros Christoudias, Climate and Atmosphere Research Center