20.04.2013 Views

Documentation of the Evaluation of CALPUFF and Other Long ...

Documentation of the Evaluation of CALPUFF and Other Long ...

Documentation of the Evaluation of CALPUFF and Other Long ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

6.4.3.2 Effect <strong>of</strong> PiG on Model Performance<br />

Whe<strong>the</strong>r better model performance is obtained using <strong>the</strong> PiG module or not frequently<br />

depends on <strong>the</strong> statistical metric being analyzed <strong>and</strong> <strong>the</strong> CAMx model configuration (vertical<br />

diffusion algorithm <strong>and</strong> horizontal advection solver). However, whe<strong>the</strong>r <strong>the</strong> PiG is used or not<br />

has very little difference on <strong>the</strong> rankings <strong>of</strong> <strong>the</strong> CAMx model performance using <strong>the</strong> alternative<br />

vertical mixing <strong>and</strong> horizontal advection approaches. In general, it appears that <strong>the</strong> CAMx<br />

model performance without <strong>the</strong> PiG is performing slightly better than its performance using <strong>the</strong><br />

PiG.<br />

The spatial performance statistics are sometimes improved <strong>and</strong> sometimes degraded when <strong>the</strong><br />

PiG module is invoked. For <strong>the</strong> global statistics, <strong>the</strong> PCC performance statistic is degraded by ‐<br />

11% to ‐37% (‐0.03 to ‐0.13 points) when <strong>the</strong> PiG module is invoked. Similarly, use <strong>of</strong> <strong>the</strong> PiG<br />

versus NoPiG module increases (degrades) <strong>the</strong> FB metric by 5 to 18 percent <strong>and</strong> also increases<br />

(degrades) <strong>the</strong> NMSE metrics for all model configurations.<br />

Table 6‐2 summarized <strong>the</strong> RANK model performance statistic for <strong>the</strong> different CAMx model<br />

configurations with <strong>and</strong> without <strong>the</strong> PiG module. For each model vertical diffusion/horizontal<br />

advection configuration, using <strong>the</strong> PiG module always results in slightly lower RANK statistics<br />

that are from ‐3.9% to ‐8.5% lower than when <strong>the</strong> PiG module is not used. The ranking <strong>of</strong> <strong>the</strong><br />

top four CAMx vertical diffusion/horizontal advection configurations remains unchanged<br />

whe<strong>the</strong>r <strong>the</strong> PiG module is used or not. And by far <strong>the</strong> most important parameter examined in<br />

regards to <strong>the</strong> RANK model performance statistics for <strong>the</strong> ETEX experiment in <strong>the</strong> CAMx<br />

sensitivity tests is <strong>the</strong> vertical mixing algorithm, with <strong>the</strong> CMAQ Kz parameterization producing<br />

<strong>the</strong> best four RANK model performance statistics out <strong>of</strong> <strong>the</strong> 16 sensitivity tests: (1)<br />

NoPiG/CMAQ/PPM; (2) NoPiG/CMAQ/Bott; (3) PiG/CMAQ/PPM; <strong>and</strong> (4) PiG/CMAQ/Bott.<br />

Table 6‐2. CAMx RANK model performance statistic <strong>and</strong> model rankings for different model<br />

configurations with <strong>and</strong> without using <strong>the</strong> PiG subgrid‐scale puff model.<br />

Without PiG Module With PiG Module PiG‐NoPiG<br />

Model<br />

Model<br />

Model<br />

Configuration RANK Ranking RANK Ranking ΔRANK Percent<br />

OB70/BOTT 1.60 7 a 1.53 6 a<br />

‐0.07 ‐4.4%<br />

OB70/PPM 1.66 4 1.55 4 ‐0.11 ‐6.6%<br />

TKE/BOTT 1.65 5 1.51 7 ‐0.14 ‐8.5%<br />

TKE/PPM 1.70 3 1.56 3 ‐0.14 ‐8.2%<br />

ACM2/BOTT 1.54 8 1.48 8 ‐0.06 ‐3.9%<br />

ACM2/PPM 1.60 6 a 1.53 5 a<br />

‐0.07 ‐4.4%<br />

CMAQ/BOTT 1.90 2 1.76 2 ‐0.14 ‐7.4%<br />

CMAQ/PPM<br />

a<br />

tied<br />

1.94 1 1.80 1 ‐0.14 ‐7.2%<br />

128

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!