Environmental Engineering Reference
In-Depth Information
indicator of air quality degradation at a given location. The National Oceanic and
Atmospheric Administration (NOAA), in partnership with the United States
Environmental Protection Agency (EPA), is operationally implementing an Air
Quality Forecast (AQF) system. This program, which couples NOAA's North
American Mesoscale (NAM) weather prediction model with EPA's Community
Multiscale Air Quality (CMAQ) model, has provided forecasts of ozone (O
3
)
mixing ratios since 2004. Developmental PM
2.5
forecasts were initiated in 2005
(Mathur et al., 2008). The modeling domain for both the operational and develop-
mental predictions currently covers the continental U.S. (CONUS).
Bias-adjustment techniques have been used to correct systematic biases in
surface O
3
predictions (Delle Monache et al., 2006; Wilczak et al., 2006; Kang
et al., 2008), and more recently have also been extended for PM
2.5
forecasts (Kang
et al., 2009). Among these techniques, Kalman Filter (KF) predictor forecast method
has shown the most improvement in forecast skill. To test the applicability of the
methods in an operational real-time setting, during 2008 the KF bias-adjustment
technique (Kang et al., 2008, 2009) was implemented in near real-time along with
the NAM/CMAQ AQF system to provide daily bias-adjusted O
3
and PM
2.5
forecasts
at all the locations where observations from EPA's AIRNOW network were available
within the CONUS domain. The bias-adjusted O
3
forecasts were performed from
the beginning of April to the middle of September covering the entire O
3
season,
and the PM
2.5
bias-adjusted forecasts were conducted through the whole
year. In
this paper, the preliminary performance evaluations of the KF bias-adjusted O
3
and PM
2.5
forecasts are presented. To facilitate performance evaluations for PM
2.5
,
the study period is divided into cold season (from January to April 20 and from
September to December) and warm season (from April 21 to August 31).
2. Implementation of the KF Bias-Adjustment Method
The KF predictor bias-adjustment algorithm is described in detail by Delle Monache
et al. (2006). The adaptation and implementation of the technique in our applications
is presented in Kang et al. (2008). Also in our previous study (Kang et al., 2008),
the error ratio, a key parameter in the KF approach which determines the relative
weighting of observed and forecast values, was investigated extensively with O
3
forecasts at over 1,000 monitoring locations. Even though the optimal error ratios
inherent in the KF algorithm implementation were found to vary across space, the
impact of using the optimal values on the resultant bias-adjusted predictions was
insignificant when compared with using a reasonable single fixed value of this
parameter for all the locations within the modeling domain. In this study, the same
single fixed error ratio value of 0.06 was used to all the locations for the real-time
bias-adjusted O
3
and PM
2.5
forecasts.
The KF bias-adjustment technique was implemented for O
3
and PM
2.5
forecasts
separately. First, the KF was initialized with the initial estimates of KF parameters
as outlined in Kang et al. (2008) and with 2 days of hourly observations and raw
