Multiscale Air Quality Simulation Platform Research Articles

An empirical assessment of Bayesian melding for mapping ozone pollution

This paper reviews the Bayesian melding approach while a companion technical report cited in the paper gives technical details about the Gibbs sampling algorithm used to implement the model and software developed for the research reported in this paper and available online can enable the method's use in other applications. This paper critically assesses the use of melding for mapping ozone concentration fields for possible future use in setting regulatory standards. This assessment has two stages. First a simulation study validates the computer code and goes on to investigate properties of the melding approach in situation where the “truth” is known. Then it is critically tested on a ozone mapping application using ozone data from the air quality system (AQS) database and simulated data from the multiscale air quality simulation platform (MAQSIP) chemical transportation model for ozone. In all cases, the melding is testing against Kriging, a simpler and more traditional way of mapping spatial fields. Conclusions and recommendations for future work are provided. Copyright © 2011 John Wiley & Sons, Ltd.

Bayesian Maximum Entropy Integration of Ozone Observations and Model Predictions: An Application for Attainment Demonstration in North Carolina

States in the USA are required to demonstrate future compliance of criteria air pollutant standards by using both air quality monitors and model outputs. In the case of ozone, the demonstration tests aim at relying heavily on measured values, due to their perceived objectivity and enforceable quality. Weight given to numerical models is diminished by integrating them in the calculations only in a relative sense. For unmonitored locations, the EPA has suggested the use of a spatial interpolation technique to assign current values. We demonstrate that this approach may lead to erroneous assignments of nonattainment and may make it difficult for States to establish future compliance. We propose a method that combines different sources of information to map air pollution, using the Bayesian Maximum Entropy (BME) Framework. The approach gives precedence to measured values and integrates modeled data as a function of model performance. We demonstrate this approach in North Carolina, using the State's ozone monitoring network in combination with outputs from the Multiscale Air Quality Simulation Platform (MAQSIP) modeling system. We show that the BME data integration approach, compared to a spatial interpolation of measured data, improves the accuracy and the precision of ozone estimations across the state.

Multiscale Air Quality Simulation Platform (MAQSIP): Initial applications and performance for tropospheric ozone and particulate matter

The performance of the Multiscale Air Quality Simulation Platform (MAQSIP) in simulating the regional distributions of tropospheric ozone and particulate matter (PM) is evaluated through comparisons of model results from three‐dimensional simulations against available surface and aircraft measurements. These applications indicate that the model captures the dynamic range of observations and the spatial trends represented in measurements. Some discrepancies also exist, however, and they are discussed in the context of model formulation, input data specification and assumptions, and variability and bias in measurements. The daily normalized bias (within ±20%) and normalized gross errors (<25%) for predicted surface level O3 over an entire summer season are within the suggested performance criteria for management evaluation studies and are comparable to, if not smaller than, those reported previously for other regional O3 models. Comparisons of modeled PM composition with speciated fine particle concentration measurements show that the model is able to capture the spatial variability in fine PM mass as well as in the inorganic component fractions. Both measurements and model results show that in the summertime in the eastern U.S., SO42− is a relatively large component of fine PM mass; in contrast, NO3− is a significant fraction in the western U.S. in the wintertime case studied. The ability of the model to simulate the observed visibility indices (extinction coefficient and deciview) are evaluated through comparisons of model estimates using both a detailed Mie theory‐based calculation (based on predicted aerosol size and number distributions) and an empirical mass reconstruction algorithm. Both modeled and observed data show that among the various aerosol components, in the eastern U.S. SO42− contributes the largest fraction to the aerosol extinction (35–85%), while organic mass contributes up to 20–25%. In contrast, in the western U.S., SO42− and NO3− have comparable contributions (20–50%) to the observed aerosol extinction. Comparisons with limited observational aircraft data, however, show moderate to poor correlation with measurements in the free troposphere. While these discrepancies can be attributed in part to model initialization and lateral boundary conditions specification, there is a need for further evaluation of the representation of boundary layer‐free troposphere exchange mechanisms as well as the chemical mechanisms currently used in the model for representing chemistry in the free troposphere.

Observed and modeled VOC chemistry under high VOC/NO conditions in the Southeast United States national parks

In airsheds that contain high volatile organic compounds (VOCs) and low NO x (=NO+NO 2 ) concentrations, ozone (O 3 ) production may be significantly suppressed by NO x reactions that lead to the formation of organic nitrates. O 3 and its precursors (VOCs and NO x ) ambient levels simulated using a regional-scale photochemical model, called Multiscale Air Quality Simulation Platform, are analyzed and compared to observed data from three southeast United States national parks.

A Real-Time Eulerian Photochemical Model Forecast System: Overview and Initial Ozone Forecast Performance in the Northeast U.S. Corridor

This article reports on the first implementation of a real-time Eulerian photochemical model f o recast system in the United States. The forecast system consists of a tripartite set of one-way coupled models that run routinely on a parallel micro process or supercomputer. The component models are the fifth-generation Pennsylvania State University (PSU)–NCAR Mesoscale Model (MM5), the Sparse-Matrix Operator Kernel for Emissions (SMOKE) model, and the Multiscale Air Quality Simulation Platform—Real Time (MAQSIPRT) photochemical model. Though the system has been run in real time since the summer of 1998, forecast results obtained during August of 2001 at 15-km grid spacing over New England and the northern mid-Atlantic—conducted as part of an “early start” NOAA air quality forecasting initiative—are described in this article.The development and deployment of a real-time numerical air quality prediction (NAQP) system is technically challenging. MAQSIP-RT contains a full photochemical oxidant gas-phase chemical mechanism together with transport, dry deposition, and sophisticated cloud treatment. To enable the NAQP system to run fast enough to meet operational forecast deadlines, significant work was devoted to data flow design and software engineering of the models and control codes. The result is a turnkey system now in use by a number of agencies concerned with operational ozone forecasting.Results of the chosen episode are compared against three other models/modeling techniques: a traditional statistical model used routinely in the metropolitan Philadelphia, Pennsylvania, area, a set of publicly issued forecasts in the northeastern United States, and the operational Canadian Hemispheric and Regional Ozone and NOx System (CHRONOS) model. For the test period it is shown that the NAQP system performs as well or better than all of these operational approaches. Implications for the impending development of an operational U.S. ozone forecasting capability are discussed in light of these results.

Propagation of uncertainty in hourly utility NOx emissions through a photochemical grid air quality model: a case study for the Charlotte, NC, modeling domain.

One of the major hypothesized sources of uncertainties in air quality model inputs is the emission inventory. A probabilistic hourly NOx emission inventory for 32 units of nine coal-fired power plants in the Charlotte domain for the year 1995 was propagated through the Multiscale Air Quality Simulation Platform (MAQSIP). The inventory was developed using time series techniques. Time series for a 4-d episode were simulated and propagated through the air quality model 50 times in order to represent the ranges of uncertainty in hourly emissions and predicted ozone levels. Intra-unit autocorrelation in emissions and inter-unit dependence were accounted for. The range of uncertainty in predicted ozone was greater when inter-unit dependence was included as compared to when units were treated as statistically independent. Uncertainties in maximum ozone hourly or 8-h concentrations at a specific location could be attributed to a specific power plant based upon regression analysis. Out of 3969 grid cells in the modeling domain, there were 43 and 1654 grid cells with a probability greater than 0.9 of exceeding a 1-h 120 ppb standard and an 8-h 80 ppb standard, respectively. The time series of predicted ozone values had similar autocorrelation as compared to monitored data. The implications of these results for air quality management are addressed.

Application of empirical orthogonal functions to evaluate ozone simulations with regional and global models

Empirical orthogonal functions are used together with standard statistical metrics to evaluate the ability of models with different spatial resolutions to reproduce observed patterns of surface ozone (O3) in the eastern United States in the summer of 1995. We examine simulations with the regional Multiscale Air Quality Simulation Platform model (horizontal resolution of 36 km2) and the global GEOS‐CHEM model (2° × 2.5° and 4° × 5°). As the model resolution coarsens, the ability to resolve local O3 maxima (O3 ≥ 90 ppbv) is compromised, but the spatial correlation improves. This result shows that synoptic‐scale processes modulating O3 concentrations are easier to capture in models than processes occurring on smaller scales. Empirical orthogonal functions (EOFs) derived from the observed O3 fields reveal similar modes of variability when averaged onto the three model horizontal resolutions. The EOFs appear to represent (1) an east‐west pattern associated with frontal passages, (2) a midwest‐northeast pattern associated with migratory high‐pressure systems, and (3) a southeast stagnation pattern linked to westward extension of the Bermuda High. All models capture the east‐west and southeast EOFs, but the midwest‐northeast EOF is misplaced in GEOS‐CHEM. GEOS‐CHEM captures the principal components of the observational EOFs when the model fields are projected onto these EOFs, implying that it can resolve the contribution of the EOFs to the observed variance. We conclude that coarse‐resolution global models can successfully simulate the synoptic conditions leading to high‐O3 episodes in the eastern United States.

Moment‐based simulation of microphysical properties of sulfate aerosols in the eastern United States: Model description, evaluation, and regional analysis

A six‐moment microphysics module for sulfate aerosols based on the quadrature method of moments has been incorporated in a host 3‐D regional model, the Multiscale Air Quality Simulation Platform. Model performance was examined and evaluated by comparison with in situ observations over the eastern United States for a 40‐day period from 19 July to 28 August 1995. The model generally reproduces the spatial patterns (sulfate mixing ratios and wet deposition) over the eastern United States and time series variations of sulfate mass concentrations. The model successfully captured the observed size distribution in the accumulation mode (radius 0.1–0.5 μm), in which the sulfate is predominately located, while underestimating the nucleation and coarse modes on the basis of the size distributions retrieved from the modeled six moments at the Great Smoky Mountains (GSM). This is consistent with better model performance on the effective radius (ratio of third to second moment, important for light scattering) than on number‐mean and mass‐mean radii. However, the model did not predict some of the moments well, especially the higher moments and during the dust events. Aerosol components other than sulfate such as dust and organics appear to have contributed substantially to the observed aerosol loading at GSM. The model underpredicted sulfate mixing ratios by 13% with about 50% of observations simulated to within a factor of 2. One of the reasons for this underestimation may be overprediction of sulfate wet deposition. Sulfate mass concentrations and number concentrations were high in the source‐rich Ohio River valley, but number concentrations were also high over the mid‐Atlantic coast (New Jersey area). Most (77%) sulfate amount was below 2.6 km, whereas most sulfate number (>52%) was above 2.6 km except over Ohio River valley (41%). These results demonstrate the accuracy, utility, practicality, and efficiency of moment‐based methods for representing aerosol microphysical processes in large‐scale chemical transport models.

Modeling nitric oxide emissions from biosolid amended soils

Utilizing a state-of-the-art mobile laboratory in conjunction with a dynamic flow-through chamber system, nitric oxide concentrations [NO] were measured and NO fluxes were calculated during the summer, winter and spring of 1999/2000. The field site where these measurements were conducted was an agricultural soil amended with biosolids from a municipal wastewater treatment facility. These NO flux values were then used to assess the impact of including biosolid amended soils as a land-use class in an air quality model. The average NO flux from this biosolid amended soil was found to be exponentially dependent on soil temperature [NO Flux ( ng N m −2 s −1)=1.07 exp(0.14 T soil ) ; R 2=0.81—NO Flux=71.3 ng N m −2 s −1 at 30°C]. Comparing this relationship to results of the widely applied biogenic emissions inventory system (BEIS2) model revealed that for this field site, if the BEIS2 model was used, the NO emissions would have been underestimated by a factor of 26. Using this newly developed NO flux algorithm, combined with North Carolina Division of Water Quality statistics on how many biosolid amended acres are permitted per county, county-based NO inventories from these biosolid amended soils were calculated. Results from this study indicate that county-level biogenic NO emissions can increase by as much as 18% when biosolid amended soils are included as a land-use class. The multiscale air quality simulation platform (MAQSIP) was then used to determine differences in ozone (O 3) and odd-reactive nitrogen compounds (NO y ) between models run with and without the biosolid amended acreages included in the inventory. Results showed that during the daytime, when atmospheric mixing heights are typically at their greatest, any increase in O 3 or NO y concentrations predicted by the model were small (<3%). In some locations during late evening/early morning hours, ozone was found to be consumed by as much as 11%.

Coupling the Vertical Distribution of Ozone in the Atmospheric Boundary Layer

Vertical measurements of ozone were made on a 610 m tall tower located about 15 km southeast of Raleigh, NC, as part of an effort by the State of North Carolina to develop a State Implementation Plan (SIP) for ozone control in the Raleigh Metropolitan Statistical Area and other metropolitan areas in the state. Ozone was monitored at 10, 250, and 433 m height levels during the summer months of 1993−1995 and at 10, 76, 128, and 433 m height levels during the summer months of 1996−1997. A regional atmospheric chemistry/transport model, called Multiscale Air Quality Simulation Platform (MAQSIP), was also employed to simulate three-dimensional O3 distribution over the eastern United States for a 2-month period (June 1−July 31, 1995). Through complementary analyses of the 5-year data and the modeling results, this paper examines the vertical distribution of ozone concentrations in the atmospheric boundary layer with the emphasis on the contribution of residual ozone aloft to the ground level ozone enhancement d...

Simulation of carbon monoxide transport during April 1994

The Multiscale Air Quality Simulation Platform (MAQSIP) is used to simulate transport of carbon monoxide (CO) as a passive tracer over North America, Europe, and the North Atlantic during the April 1994 Measurement of Air Pollution from Satellites (MAPS) mission. MAQSIP is driven by meteorological fields generated by the Pennsylvania State University/National Center for Atmospheric Research fifth‐generation mesoscale model. Model CO surface emissions from biomass burning, fossil fuel combustion, nonmethane hydrocarbon oxidation, oceans, and soils are based on inventories from the Belgian Institute for Space Aeronomy and the Global Emissions Inventory Activity. Predicted CO mixing ratios are vertically weighted for comparison with MAPS observations. The spread in the mission‐averaged vertically weighted simulated CO mixing ratios (∼38 ppbv, compared to 60 ppbv in the MAPS data) suggests that CO surface emissions significantly affect MAPS observations on a weekly timescale. Good qualitative agreement is found between MAPS observations and model predictions on several temporal and spatial scales. Possible reasons for discrepancies are examined. A simulation without cumulus convection increases CO mixing ratios in the lower model layers and depletes CO above, resulting in a complex pattern of increases and decreases upon vertical weighted integration. Another simulation, which included a diurnal emissions variation, produced significant changes in instantaneous local CO mixing ratios, but had a minimal effect on the mission‐averaged MAPS comparisons.