Enhancing air quality forecasts by geomatic downscaling: an application to daily PM10 concentrations in France

Abstract

Public health institutions need high-resolution next-day forecasts so they can order appropriate measures when there is a risk of air pollution exceeding regulatory thresholds. The MOCAGE model, the chemistry transport model developed by Meteo-France, forecasts hourly surface PM10 concentrations at a resolution of 0.1° throughout France (7.6 km). To obtain more efficient forecasts, a downscaling method is applied using topographic data (250-m resolution) and inventory data (2.2 km). All these disparate inputs are spatially standardized in a geographical information system to construct continuous daily fields at 250-m resolution. This method is suitable for large territories with widely varying environments (mountains, lowlands, coastlinessnap, urban areas, etc.) and areas with a low density of monitoring stations. The parameters used to improve MOCAGE forecasts are derived from “global” and “local” regressions describing the links between the daily PM10 concentration averages collected at 325 monitoring stations and seven explanatory variables (three topographic and four emission-inventory variables). One of the main results shows that the topographic and emission variables, respectively, explain 6% and 13% of PM10 variance in France. Analysis by local regression accounts for 74% of the spatial variation of PM10 concentration, while the global regression accounts for 49%. The results show above all that if the authorities responsible for human health protection had used the downscaling method instead of MOCAGE raw forecasts in 2016, they would have informed or alerted ten times as many people about the information and recommendation threshold (50 μg m−3) and alert threshold (80 μg m−3) being exceeded.