Constraining chemical transport PM2.5 modeling outputs using surface monitor measurements and satellite retrievals: application over the San Joaquin Valley

Abstract

Advances in satellite retrieval of aerosol type can improve the accuracy of near-surface air quality characterization by providing broad regional context and decreasing metric uncertainties and errors. The frequent, spatially extensive and radiometrically consistent instantaneous constraints can be especially useful in areas away from ground monitors and progressively downwind of emission sources. We present a physical approach to constraining regional-scale estimates of PM₂.₅, its major chemical component species estimates, and related uncertainty estimates of chemical transport model (CTM; e.g., the Community Multi-scale Air Quality Model) outputs. This approach uses ground-based monitors where available, combined with aerosol optical depth and qualitative constraints on aerosol size, shape, and light-absorption properties from the Multi-angle Imaging SpectroRadiometer (MISR) on the NASA Earth Observing System's Terra satellite. The CTM complements these data by providing complete spatial and temporal coverage. Unlike widely used approaches that train statistical regression models, the technique developed here leverages CTM physical constraints such as the conservation of aerosol mass and meteorological consistency, independent of observations. The CTM also aids in identifying relationships between observed species concentrations and emission sources. Aerosol air mass types over populated regions of central California are characterized using satellite data acquired during the 2013 San Joaquin field deployment of the NASA Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality (DISCOVER-AQ) project. We investigate the optimal application of incorporating 275 m horizontal-resolution aerosol air-mass-type maps and total-column aerosol optical depth from the MISR Research Aerosol retrieval algorithm (RA) into regional-scale CTM output. The impact on surface PM_2.5 fields progressively downwind of large single sources is evaluated using contemporaneous surface observations. Spatiotemporal R ² and RMSE values for the model, constrained by both satellite and surface monitor measurements based on 10-fold cross-validation, are 0.79 and 0.33 for PM_2.5, 0.88 and 0.65 for NO₃ ^-, 0.78 and 0.23 for SO₄ ^2-, and 1.01 for NH⁺, 0.73 and 0.23 for OC, and 0.31 and 0.65 for EC, respectively. Regional cross-validation temporal and spatiotemporal R² results for the satellite-based PM₂.5 improve by 30 % and 13 %, respectively, in comparison to unconstrained CTM simulations and provide finer spatial resolution. SO₄ ^2- cross-validation values showed the largest spatial and spatiotemporal R² improvement, with a 43 % increase. Assessing this physical technique in a well- instrumented region opens the possibility of applying it globally, especially over areas where surface air quality measurements are scarce or entirely absent.

Figure 1.

San Joaquin study area shows the ground elevation, EPA AQS and CSN monitors, and AERONET sites during the NASA DISCOVER-AQ flight campaign.

Figure 2.

Methods flow chart connecting satellite-retrieved AOD to modeled AOD, PM_2.5 mass, and PM_2.5-speciated mass. The parenthetical terms are defined in their respective steps in the methods section.

Figure 3.

Scatterplot comparisons of AERONET coincidences with MISR-RA, MAIAC, gap-filled MISR-RA, and CMAQ results within ±15 min of Terra overpass time. The MAIAC and AERONET AOD comparisons are plotted at 550 nm, whereas the MISR-RA and AERONET AOD comparisons are at 558 nm; the dotted lines indicate the 0.15 AOD threshold; 1 : 1 dashed lines are shown for reference.

Figure 4.

Scatterplots of the ratios of daylight averages to the Terra overpass time vs. AERONET AOD retrievals within ±15 min of Terra overpass time. Two ratios are shown for CMAQ: daytime average-to-hour ratio and diurnal average-to-hour ratio. The FCMAQ ratios shown are the FCMAQ diurnal to CMAQ hour values. The dashed unity lines are included for reference.

Figure 5.

Theoretical plot of fused and optimized dataset weights as a function of distance from surface observations. Three regimes identify the contribution of each dataset towards improving concentration held estimates stratified by distance. As applied here, the exponential decay rate that reflects the temporal Pearson correlation between ground observations as a function of distance is species- specific. The temporal variations between ground observations and CMAQ or FillSAT are more consistent (shown as constant), independent of ground observations, and therefore are not functions of distance to monitor. FCMAQ (surface measurements + model) and OPT (surface + satellite measurements + model) curves show how the strengths of the ground observations and other datasets are maximized using temporal correlations as each grid cell is a function of distance from a ground observation.

Figure 6.

PM_2.5 FRM calculated concentration maps with monitor observations (filled circles) and RBG images for the 3 days during the study period with highest AOD. The resolution of the concentration maps is 275 m, whereas the size of the observation markers is ~ 0.1° (~ 11.1km)

Figure 7.

NH₄, SO₄, and NO3 calculated concentration maps and monitor observations (filled circles) for 3 February.