European Central Bank working paper finds satellite NO2 data can improve real-time oil demand nowcasting

The European Central Bank published a Working Paper assessing whether satellite observations of nitrogen dioxide (NO2), a short-lived pollutant largely emitted by fossil fuel combustion, can improve near-term forecasting of oil demand. Using cleaned and aggregated daily satellite data, the paper finds that adding NO2 measures to standard forecasting setups materially improves oil demand nowcasts across a set of advanced and emerging economies. The study builds daily NO2 indices from Sentinel-5P TROPOMI observations and tests them in linear and non-linear forecasting models for monthly oil consumption across ten economies (Australia, China, France, India, Italy, Japan, South Korea, Spain, the United Kingdom and the United States), together representing around 60% of world GDP and NO2 emissions. Relative to models using autoregressive terms and conventional predictors such as industrial activity, prices, weather and vehicle registrations, incorporating NO2 reduces forecasting errors by roughly 20–25% on average, with larger gains during crisis episodes but statistically meaningful improvements also outside the pandemic period; non-linear approaches, including neural networks, deliver the biggest incremental improvements, consistent with a non-linear relationship between pollution and energy demand. The paper also benchmarks NO2 against alternative high-frequency indicators used in the literature (Google Mobility, Google Trends and Oxford Stringency indices), finding that pandemic-specific indicators can perform better during the COVID-19 period while the satellite NO2 proxy retains more predictive value in more stable periods, supporting its use as a globally consistent, near-real-time monitoring tool where official oil data are delayed.