Search

Air quality over cities is mainly monitored by in-situ surface measurements. However, these stations are too sparse to properly capture the inhomogeneity of pollutant concentrations over urban areas. The need for high-resolution concentration estimate has grown in recent years, together with the awareness of the harmful effects of air pollution. In this study, we develop a Bayesian scheme that combines the high-resolution (3 × 3 m2) Particulate Micro SWIFT SPRAY numerical air quality simulations (PMSS) with operational surface measurements. The goal is to improve NOX and PM10 PMSS concentrations estimates over monitoring stations and within their vicinity. For this purpose, we simulate pollutant concentrations over the city of Paris for ten days over the period of March 2016. The Bayesian model provides an enhanced estimate of pollutant concentration in space and time. At the monitoring stations location, these estimates are characterized by lower temporal dispersion compared to the simulated data. Within the vicinity of the monitor stations, enhanced concentration estimates are closer to observations. For NOX, the improvement is stronger and occurs in a larger area for urban background stations than for traffic stations. Overall, NOX improvement is higher than PM10 improvement. The initial PMSS model prediction is more biased for NOX than for PM10 due to large uncertainties in NOX emissions over the traffic network. Keywords: Urban air pollution, Surface monitoring networks, High-resolution modeling, Spatial representativeness, Bayesian modeling

Ambient pollutant concentrations in Paris, France are routinely measured by the local surface-monitor network of the AIRPARIF agency. Such networks, however dense, have a limited spatial representativeness around the monitoring-site and are not capable to represent the strong horizontal gradients of pollutant concentrations over urban areas. High resolution models simulate 3D pollutant concentration fields at a spatial resolution as fine as a few meters over the urban area by integrating the underlying emission sources and accounting for the effect of buildings on the dispersion patterns. These models, provide a good spatial variability over the urban area but suffer from uncertainties related to the emission inventories, meteorological fields and parametrizations of the physical and chemical processes.In this paper, simulations conducted by ARIA Technologies with the Parallel Micro-Swift-Spray (PMSS) model (http://www.aria.fr/projets/aircity) are used to assess NOX and PM10 representativeness areas around five urban background and five traffic-oriented monitoring-sites of the AIRPARIF network during ten days in March 2016. Commonly, the spatial representativeness of a monitor site is defined through homogeneity areas, namely the area around a monitoring-site where pollutant concentration is above 20% of the concentration at the location of the monitoring-site. Here, we propose a novel approach that uses similarity areas to define the spatial representation of monitor sites. Similarity areas integrate points that respect the additional condition to be highly correlated in time with the concentration at the monitor station. Thus, the criterion to select similarity areas is a combination of a high value of the correlation coefficient and a small value of the normalized root mean square error with regards to the concentration at the grid-cell corresponding to the location of the monitor. Criteria thresholds are determined through an iterative analysis and a representative area is defined through image processing that selects all the connected pixels that satisfy criteria thresholds and incorporate the grid-cell of the monitor.Daily similarity areas estimated around each monitor are compared against homogeneity areas with regards to their shape, spatial extent, and urban specific characterization. Around urban background sites they are of the same order of magnitude, whereas around traffic sites similarity areas are generally larger than homogeneity areas. PM10 representativeness areas are found to be 2.2 times larger than the NOX ones. Urban background areas are representative of the broad neighborhood around the monitoring-site, whereas traffic-oriented monitoring-sites are representative of specific urban features such as sections of roads and sidewalks along the road. Averaged over the 10 days of the study and across all monitoring-sites, representativeness areas for urban background monitoring-sites are about 8 times larger than traffic representativeness areas (0.6 km2 vs. 0.07 km2). Keywords: Urban air pollution, Surface monitoring networks, High resolution modeling, Spatial representativeness