Remote sensing to improve air quality forecasts
Fine particles suspended in the air (referred to as aerosols), such as occur in smoke and haze, are known to pose a significant health risk. The US Environmental Protection Agency (EPA) sets strict guidelines for exposure to ‘fine particulate matter’ (PM2.5), defined as measuring 2.5 microns or less in diameter.1 Various government agencies accordingly try to issue air quality predictions of fine particulate pollution, in a manner similar to regular weather forecasts. To do this, they employ sophisticated computational models that integrate meteorological simulations, estimates of particle emissions from different sources, and complex air chemistry and physics to understand and quantify likely pollutant events.
To correctly predict the transport and dispersion of aerosols, such models must take into account complex meteorological mechanisms beyond the first order (simplified) wind fields. It is especially important to understand the various advection mechanisms (e.g., convective heating and turbulent mixing) that transport particles from the surface into the lowest part of the atmosphere, a region called the planetary boundary layer (PBL). Because the behavior of the PBL is directly influenced by contact with the ground, its dynamics are challenging to simulate.
As an example of the difficulties that can arise, a retrospective comparison between forecasted and measured PM2.5 mass concentrations found sharp overestimates (spikes) in the New York City area during both the predawn and post-sunset periods in the summer.2 Those spurious predictions would have caused false air quality warnings to be issued to the community.
There are two possible explanations for such forecast spikes. One is that the emission factors (data inputs quantifying the amounts of pollutants discharged into the atmosphere by various primary sources) were themselves overestimated. The other is that the overall emissions were correct but their advection away from the surface was not realistically modeled, producing an incorrect vertical aerosol profile. To determine which of these two explanations applies, and make appropriate adjustments to the model, it is necessary to check its output against direct measurements of the vertical distribution of particulate matter.
The air quality model in question is one of the most widely used in the community. It uses meteorological predictions computed by the Weather Research and Forecast (WRF) model, which takes into account complex surface-atmospheric interactions, including energy and mass transport modeled via turbulent mechanisms and driven by various thermal sources and fluxes. One especially important factor affecting mass transport during convective mixing of the atmosphere is the PBL, i.e., the lower layer of the atmosphere that interacts convectively with the surface. If the dynamics of the PBL are not accurately modeled, the resultant vertical particulate distribution will be skewed, producing large errors. The meteorological data from the WRF model (including cloud cover, which affects radiation and hence the atmospheric chemistry) then drives and constrains the Community Multiscale Air Quality (CMAQ) model,3 which generates the particle pollution forecast.
To diagnose the cause of the WRF-CMAQ particle overestimates, we used a combination of lidar (light detection and ranging) and ceilometer readings to directly quantify the vertical structure of the aerosols and PBL height for comparison against the forecasts.4 The lidar and ceilometer measure an optical scattering property that can serve as a proxy for particulate concentration. While the lidar was used to assess the PBL height forecasts during the daytime when the PBL is very high, this instrument has a blind spot in the first 500 meters. Unfortunately, this is precisely the altitude region where the WRF-CMAQ model discrepancies appear, so we also employed a 24/7 ceilometer that can see near the surface (∼30m) but requires significant pulse averaging to improve the signal-to-noise ratio.
Our results, discussed more in detail elsewhere,4 indicate that the model yielded a skewed vertical aerosol profile. Figure 1 shows the cumulative path-averaged PM2.5 concentration as predicted by CMAQ and as measured by ceilometer backscatter over the diurnal cycle for different altitude ranges. Panel (a) shows the forecasted early-morning spike, with the particle distribution markedly compressed near the surface. However, the ceilometer backscatter measurements in panel (b) do not show any corresponding compression of particles, even for the lowest-altitude bins. On the other hand, if we look at the entire altitude column, the total particulate concentration matches the general diurnal trend predicted by CMAQ, which is consistent with the hypothesis that the forecast spikes are mainly due to incorrect vertical apportionment of the particulates.
The underlying reason can be traced to incorrect WRF predictions of PBL height, shown overlaid on the CMAQ-forecasted vertical particulate profile in Figure 2. In panel (a), the high particulate concentrations near the surface correspond to unrealistically low PBL heights. Panel (b) instead shows the results of an altered WRF boundary layer model (2010), with more realistic PBL altitude levels that generate significantly improved predictions of particulate concentration.
In summary, we found that active remote sensing is a valuable tool for better interpreting and constraining atmospheric transport models to obtain more realistic surface pollution predictions, in support of EPA mandates for air quality forecasts. In future, we envisage using multispectral passive and active optical remote sensing to assess the microphysical properties of aerosols (e.g., size distribution, refractive index, and hygroscopic behavior), which could then be assimilated into forecasting models5 to improve their accuracy.
Barry Gross is a professor at CCNY and a research scientist with NOAA-CREST (National Oceanic and Atmospheric Administration - Cooperative Remote Sensing Science and Technology Center). His research interests are in active and passive remote sensing of aerosols and clouds. He has published over 50 articles and is a member of SPIE and IEEE.
Chuen-Meei Gan is a post-doctoral researcher with the National Research Council, Atmospheric Model Development Branch. Her past research has explored urban PBL, PM2.5, and aerosol optical properties using multiple remote sensing instruments. Her current work focuses on combining a variety of approaches to study the physical and optical properties of aerosols and clouds, with a special emphasis on sensor techniques (e.g., in situ and spaceborne), for the validation and assessment of the WRF-CMAQ model.
