6.2
Attainment demonstration uncertainty stemming from poor meteorological model performance

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 27 January 2011: 8:45 AM
Attainment demonstration uncertainty stemming from poor meteorological model performance
3A (Washington State Convention Center)
Saffet Tanrikulu, Bay Area Air Quality Management District, San Francisco, CA; and S. Beaver, A. N. Palazoglu, and A. Singh
Manuscript (735.0 kB)

Coupled meteorological models (MMs) and photochemical air quality models (PAQMs) provide valuable technical information to air quality planners and policy makers. Models can estimate the sensitivities of pollutant levels to emissions reductions and evaluate proposed or adapted emissions control scenarios to demonstrate attainment. But, adequate model performance must be achieved before PAQM results can be effectively utilized for regulatory purposes. MM shortcomings can result in degraded PAQM performance. We propose a framework to quantify the impact of MM shortcomings on PAQM results.

Many techniques can be applied to optimize MM performance. Examples include testing available physics options, adjusting PBL parameterizations, and implementing four-dimensional data assimilation (FDDA). Still, systematic MM biases may be propagated through a PAQM. MMs tend to have difficulty simulating weak synoptic forcing events under which episodes of poor air quality mostly occur. This problem is important because attainment demonstration is based on modeling of episodic periods. Typically, simulated winds are too strong, thus pollutant levels are underestimated. Underestimated pollutant levels can result in underestimated pollutant sensitivities to emissions reductions. As a workaround to this problem, the US EPA guidelines specify adjusting raw PAQM sensitivity estimates using a Relative Reduction Factor (RRF). The RRF is simply a ratio of simulated to observed pollutant levels. This linear adjustment may be accurate only when PAQM biases are small.

The impact of MM shortcomings to produce underestimated pollutant levels and sensitivities was quantified for winter-season PM2.5 modeling over Central California. First, meteorological analysis was conducted based on clustering and pattern matching as described by Beaver et al. (2010). Observations were clustered to determine representative weather patterns impacting regional 24-hr PM2.5 characteristics. A label was generated reflecting the actual weather pattern that occurred on each day. The simulated fields for each day were then matched against each of the observation-based weather patterns. For a given day, a match between the observation-based label and the simulation-based label implied that the simulated fields were representative of actual conditions. Mismatches were associated with days having poorly simulated (nonrepresentative) meteorological fields. For these mismatching days, pattern matching identified the closest-matching (most representative) meteorological fields simulated by the MM. Second, a numerical experiment was performed. The closest-matching meteorological field was substituted as PAQM input for any day exhibiting a mismatch between observed and simulated meteorology. This substitution of the closest-matching meteorological fields for days with poor MM performance dramatically improved PAQM performance. This modeling approach to substitute the closest-matching meteorological fields may be inapplicable for attainment demonstration; however, it can quantify the uncertainty of an attainment demonstration stemming from poor MM performance.

The Bay Area Air Quality Management District (BAAQMD) has conducted preliminary seasonal PM2.5 photochemical modeling over Central California for December-January of 2000-01. Meteorological fields were prepared using the Fifth-Generation National Center for Atmospheric Research/Pennsylvania State University Mesoscale Model (MM5). These fields were used as inputs to the Models-3 Community Multiscale Air Quality (CMAQ) model. CMAQ was implemented using the SAPRC99 chemical mechanism, the Models-3 AE3 aerosol module, and the RADM aqueous-phase chemistry model. Modeling details are provided by Tanrikulu et al. (2010).

The meteorological analysis indicated that MM5 performance was poorest for the weather pattern under which most PM2.5 exceedances occurred. These episodic conditions typically consisted of several days to weeks during which a large-scale aloft ridge of high pressure persisted over Central California. Simulated conditions exhibited stronger winds and more dispersion than observed, resulting in the domain-wide underestimation of PM2.5 levels.

A 12-day episode occurring over 26 December 2000 through 6 January 2001 exhibited especially poor CMAQ performance mostly stemming from poor MM5 performance. Simulated PM2.5 levels were underestimated by around 50% relative to the observations, suggesting a RRF of roughly 2. The clustering assigned each day of the 12-day episode into the same persistent, episodic weather pattern. The simulated fields matched the actual weather pattern only for a single day (5 January) during this 12-day episode. Moreover, the pattern matching indicated that the simulated field for 5 January most closely matched the observed weather pattern for each of the other 11 episodic days. Therefore, the MM5 outputs for 5 January were deemed reasonably representative of the persisting conditions during the 12-day episode.

A numerical experiment was performed by repeating the winter-season CMAQ simulation with a single modification. For all days during the above 12-day episode, the MM5-simulated meteorological fields for 5 January were substituted as CMAQ inputs. Simulating the same meteorology over 12 consecutive days produced simulated 24-hr PM2.5 levels that increased pseudo-exponentially over 3-4 days before leveling off to a nearly constant value. (Emissions were also roughly the same every day, although weekday/weekend differences were significant.) The PM2.5 levels produced by the experimental run of CMAQ (with substituted closest-matching meteorology) were in excellent agreement with observations.

Next, the impact of the MM5 shortcoming on the CMAQ-estimated PM2.5 sensitivities to emissions reductions was quantified. PM2.5 sensitivities were estimated using CMAQ to simulate 20% across-the-board emissions reductions over the San Francisco Bay Area portion of the modeling domain. Sensitivity runs were performed relative to both the “control” base case simulation (with mismatching meteorological fields) and the “experimental” base case simulation (with substituted closest-matching meteorological fields). Sensitivity estimates from the experimental run were often quite larger than those obtained by multiplying the control run results by a RRF of 2.

We conclude that the RRF workaround specified in the US EPA guidelines may be inapplicable when poor MM performance results in seriously degraded PAQM performance. RRF-adjusted PAQM sensitivity estimates may be significantly underestimated. Thereby, attainment may be unduly difficult to demonstrate by photochemical modeling. We conjecture that this problem may become increasingly severe as regional air pollution levels become higher.