Accurately judging whether clean air policies are working is harder than it sounds — the weather often gets in the way. Rain, wind, and temperature changes can hide or exaggerate pollution trends, making it tricky to see if improvements come from cleaner transport or just better weather. To tackle this, scientists often use machine learning weather normalisation (ML-WN), which separates the effects of weather from human emissions.
However, this new study by Dr Yuqing Dai and colleagues at the University of Birmingham shows that ML-WN may underestimate the impact of short-term air quality measures, such as temporary traffic restrictions or emergency pollution controls. Using both simulated scenarios and real data from London’s Marylebone Road during the COVID-19 lockdown, the team found that ML-WN can miss up to 42% of the actual reduction in nitrogen oxides (NOx) during the first week of an intervention.
To fix this, the authors developed a refined approach called MacLeWN, which better captures rapid emission changes and reduces underestimation errors by over 90%. For longer-term policies, both methods perform similarly, but for short-term actions, MacLeWN provides a more realistic picture of policy effectiveness.
“Our findings have significant implications for the evaluation of air quality interventions and formulation of environmental policies… Underestimation of benefits may lead to underappreciation of policy measures and improper resource allocation.” the authors wrote.
This work supports more accurate policy evaluations and fairer assessments of interventions designed to protect public health.
Read the full paper here: