Environmental Engineering Reference
In-Depth Information
4.1 Cluster Analysis and Classification of Wind Fields
for Meteorological and Air Quality Model Validation
Scott Beaver
1
, Saffet Tanrikulu
1
, Douw Steyn
2
, Yiqin Jia
1
, Su-Tzai Soong
1
,
Cuong Tran
1
, Bruce Ainslie
2
, Ahmet Palazoglu
3
,
and Angadh Singh
3
1
Bay Area Air Quality Management District, San Francisco, CA, USA
2
The University of British Columbia, Vancouver, BC, Canada
3
University of California, Davis, CA, USA
Abstract
Clustering of observed winds and classification of simulated winds were
used for meteorological and air quality model evaluation. We simulated meteorology
with MM5 and particulate matter (PM) with CMAQ for December to January
2000-2001 in the San Francisco Bay Area (SFBA). EOFs were used to classify
simulated winds among the patterns identified by a previous clustering of obser-
vations. We investigated the match between the classification of the simulated winds
and the original clustering. Agreement between the clustering of observed winds and
the classification of simulated winds implies model validity. Disagreement serves as
a diagnostic tool, indicating how inaccurately modeled winds may explain degraded
air quality model performance. This novel framework complements traditional
model validation methods.
Keywords
Pattern matching, model evaluation, weather patterns, and wind field
analysis
1. Introduction and Goal of Study
Accurate meteorological simulations are important for air quality modeling.
Traditional model validation techniques include error and bias statistics calculated
between simulated and observed fields [1]. Surface wind field accuracy is critical,
especially for winter PM modeling. Here, a new validation technique is presented
based on cluster analysis of observed winds and classification of simulated winds.
Categorical evaluation is important for seasonal simulations, as the model must
discriminate among different classes of conditions.
Clustering [2] is applied to wind field observations to group days sharing air
flow patterns impacting air quality. A similar classification process is then applied
to modeled wind fields. Agreement between the observation- and simulation-based
groupings implies consistent spatial structures between the wind fields. Disagreement
may indicate potential mismatches between the observed and simulated wind fields.