The rapid spread of the Covid-19 pandemic over Spain
recently forced the Spanish authorities to implement drastic
measures of social distancing through a strict lockdown of the
population starting on March 14th. As hospitalizations were
still strongly increasing, a second and more stringent phase of
the lockdown was implemented from the March 30th to April
9th with workers from all non-essential economical activities
forced to stay at home.
This situation has impacted numerous activity sectors,
including road transport, air traffic and part of the industries.
As a consequence, air pollutant emissions have been greatly
reduced. Although such a large change of emission forcing is
expected to reduce the air pollutant concentrations in Spanish
urban areas, the extent of such reductions remains uncertain.
Key to this uncertainty are the highly variable meteorological
conditions that can either attenuate or amplify changes of air
pollution concentration originally driven by changes of
emissions. Thus, assessing the impact of the Covid-19
lockdown solely based on the analysis of the concentration
time series can often be misleading since at least part of the
variability is driven by the meteorology.
In this study, we explore the use of machine learning
algorithms for estimating the business-as-usual NO2
concentrations that would have been observed without the
Covid-19 lockdown based on ERA5 meteorological data and
additional time features. Trained on past data, these ML
models can learn the complex relationships between
meteorology and NO2 concentrations, indirectly assuming an
average emission forcing. By using these ML models to
predict the NO2 concentrations under the current situation
(with very different emission forcing), we expect the
discrepancies between predictions and observations to be
related to a large extent to the reduction of emissions induced
by the lockdown regardless of the meteorological conditions.
In this study, the reduction of NO2 pollution is investigated
in all 50 provinces of Spain.
