15 th IUAPPA Clean Air Congress Vancouver, BC Optional Presentation - - PowerPoint PPT Presentation

15th IUAPPA Clean Air Congress Optional Presentation Title Vancouver, BC Photo Slide Progress In Dispersion Model Evaluation And Prediction For Low Wind Speed

Presenter’s Name

Conditions (Control #19)

September 16, 2010 September 16, 2010 p , p , Study Study conducted by: conducted by: Jeff Connors, Bob Paine Jeff Connors, Bob Paine and Carlos Szembek and Carlos Szembek (photo by James Shuepp, provided by Larry Mahrt) Steve Hanna, Steve Hanna, subcontractor subcontractor Study funded by API and UARG

September, 2010

Outline of Presentation

• Why are low wind speeds a concern?
• Review of current AERMOD formulation
• Evaluation study approach

y pp

• Meteorological evaluation results

T t d l ti lt

• Tracer study evaluation results

September 12-16, 2010 Page 2 15th IUAPPA World Clean Air Congress - Vancouver, BC

Concerns Noted by EPA and the Modeling Community

• Brode USEPA 2007 Modeling Workshop from AERMOD

Implementation Workgroup Highlights

“Mandatory Work: Light winds Revise AERMOD’s – “Mandatory Work: Light winds. Revise AERMOD’s treatment of light winds to avoid unrealistically high concentrations”

• Reported at USEPA’s 9th Modeling Conference - Air &

Waste Management Association Comments

– Many investigators report that the worst-case AERMOD Many investigators report that the worst case AERMOD impacts occur for very low wind speeds at night, especially for low-level sources AERMOD has limited evaluation for these conditions – AERMOD has limited evaluation for these conditions – AERMOD needs supplemental evaluation to assess the accuracy of the model for these conditions

September 12-16, 2010 Page 3 15th IUAPPA World Clean Air Congress - Vancouver, BC

y

Current AERMET/AERMOD Approach

• AERMET computes the friction velocity (u*), which is

an important parameter for nocturnal hour estimates of an important parameter for nocturnal hour estimates of mixing height, sigma-z, and sigma-y

• AERMOD approach in low winds is reasonably simple,

AERMOD approach in low winds is reasonably simple, and involves a combined solution of a coherent plume (traditional Gaussian shape) and a random (pancake) plume plume

• Weighting of the two solutions depends upon wind

speed and turbulence provided to AERMOD speed and turbulence provided to AERMOD

September 12-16, 2010 Page 4 15th IUAPPA World Clean Air Congress - Vancouver, BC

Approach for this Study

• We initiated a new evaluation study to understand

AERMOD’s performance under low wind speeds

• The evaluation study featured existing research-grade

meteorological and low wind speed tracer databases

• Current and alternate versions of AERMET/AERMOD

were tested in this study

• Collaboration with USEPA and AERMIC review was
• Collaboration with USEPA and AERMIC review was

important for this study

• At the present time, AERMIC has provided limited

p , p review and EPA attention has been diverted to the new NAAQS modeling issues

September 12-16, 2010 Page 5 15th IUAPPA World Clean Air Congress - Vancouver, BC

Meteorological Evaluation Study

• Requested by EPA; evaluation focused upon u*
• Acceptable databases were selected for low wind
• Acceptable databases were selected for low wind

speeds and sonic anemometer to get u*

• Evaluation focused upon nocturnal low wind
• Evaluation focused upon nocturnal, low wind

conditions

• Cardington (flat grassy site in the UK) was included
• Cardington (flat, grassy site in the UK) was included

in the evaluation

• Other met databases (USA) were:

Other met databases (USA) were:

1. Bull Run (mixed land use/terrain Tennessee site) 2. FLOSS II (Fluxes Over Snow Surfaces, Phase 2: flat open site in northern Colorado)

September 12-16, 2010 Page 6 15th IUAPPA World Clean Air Congress - Vancouver, BC

in northern Colorado)

Initial Meteorological Evaluation Results

• Single-level friction velocity predictions by AERMET

were found to be too low for low wind nocturnal hours were found to be too low for low wind, nocturnal hours

• An adjustment to the formulation was suggested by

the data, and appeared to greatly improve the AERMET the data, and appeared to greatly improve the AERMET single-layer performance

• This adjusted formulation was tested all three met

This adjusted formulation was tested all three met databases

September 12-16, 2010 Page 7 15th IUAPPA World Clean Air Congress - Vancouver, BC

u* (y axis) vs. u for Bull Run Nocturnal Stable Hours

0.35 0.4

AERMET formulation is AERMET formulation is

0.25 0.3 0.35

Curves adapted from Curves adapted from underpredicting friction velocity underpredicting friction velocity U*

0.15 0.2 u* (m s-1)

Curves adapted from Curves adapted from AERMET formulation AERMET formulation <--

• -Transition point where single layer

Transition point where single layer Observed values Observed values i li ht bl i li ht bl

0.05 0.1 z0 = 0.51

p g y p g y quadratic equation has real solution quadratic equation has real solution in light blue in light blue 1-

• layer AERMET in beige

layer AERMET in beige (transition point is arbitrary) (transition point is arbitrary)

0.5 1 1.5 2 2.5 3 3.5 4 u (m s-1)

L.I. Residential, Clear L.I. Residential, 50% Cloud Cover L.I. Residential; 100% Cloud Cover Single-Layer Model Observed

U

September 12-16, 2010 Page 8 15th IUAPPA World Clean Air Congress - Vancouver, BC

Recommended Improvement to Single-Layer Method

Agreement with Agreement with

• bs is excellent
• bs is excellent

New AERMET transition point New AERMET transition point for cloudy skies. This avoids for cloudy skies. This avoids th “di ” b ti th th “di ” b ti th Avoid “dip” by connecting Avoid “dip” by connecting the “dip” by connecting the the “dip” by connecting the

• rigin to 1.25 times the
• rigin to 1.25 times the

transition wind speed transition wind speed Agreement with observations Agreement with observations U* Avoid dip by connecting Avoid dip by connecting

• rigin to 1.25 times
• rigin to 1.25 times

transition wind speed transition wind speed Agreement with observations Agreement with observations is much better! is much better! current AERMET transition point for current AERMET transition point for p cloudy skies cloudy skies U

September 12-16, 2010 Page 9 15th IUAPPA World Clean Air Congress - Vancouver, BC

Comparisons for u* with Cardington data (low wind speed, stable hours)

Current AERMET Current AERMET Modified AERMET Modified AERMET Si l l Si l l

September 12-16, 2010 Page 10 15th IUAPPA World Clean Air Congress - Vancouver, BC

Single layer Single layer

Conclusions from Met Evaluation

• Current AERMET formulation will likely underpredict u*

in low wind speed, stable conditions

• This would be expected to result in higher predicted

concentrations (lower dilution speed and dispersion t ) rate)

• This happens for both the single-layer and 2-layer (Bulk

Ri) th d Ri) methods

• Met model performance with the suggested

improvements is better overall improvements is better overall

• These changes were carried forward into the tracer

evaluation phase of the study

September 12-16, 2010 Page 11 15th IUAPPA World Clean Air Congress - Vancouver, BC

evaluation phase of the study

Tracer Database Evaluation

• Study focused on 3 databases:

1. Bull Run, TN (tall stack, buoyant plume) 2. Idaho Falls, ID (low-level releases) 3. Oak Ridge, TN (low-level releases)

September 12-16, 2010 Page 12 15th IUAPPA World Clean Air Congress - Vancouver, BC

Tracer Model Evaluation Procedures

• Model the plume trajectory toward the sampler with the max
• bservation – avoid any misrepresentation of model performance

do to error in wind directions

• Compare arc-wise maximum predicted and observed

concentrations

• Model performance evaluated with a Quantile-Quantile plot

– For each arc of samples/receptor the concentrations from all the modeled hour are sorted from largest to smallest i d d t f th independant of one another.

• Good performing model +/- factor of 2 as compared to observed

concentrations

• Conservative results are better for regualtory model trying to

protect public health and welfare.

September 12-16, 2010 Page 13 15th IUAPPA World Clean Air Congress - Vancouver, BC

Candidate Models

• Candidate models based on changes to

AERMET/AERMOD AERMET/AERMOD

• Results presented for 3 cases:

1.Base AERMET 2.Modified u* formulation in AERMET 3.AERMET/AERMOD with minimum sigma-v = 0.4 m/s

– Current minimum sigma-v = 0.2 m/s

September 12-16, 2010 Page 14 15th IUAPPA World Clean Air Congress - Vancouver, BC

Why Adjustments to Minimum Sigma-v?

• After running AERMOD with current AERMET, we constructed

Excel spreadsheet to replicate AERMOD predictions during stable hours (Oak Ridge and Idaho Falls) w/ model debug stable hours (Oak Ridge and Idaho Falls) w/ model debug

• utput
• Found sigma-v becomes very important under low-wind speed

conditions when sigma-theta data is not available because it helps define:

– lateral dispersion (sigma-y) – fraction of the random plume used to calculate total concentration

• AERMOD is underestimating the lateral dispersion and fraction

f th d l

• f the random plume
• This was causing the model to overpredict significantly for

VERY light winds (less than 0 5 m/s)

September 12-16, 2010 Page 15 15th IUAPPA World Clean Air Congress - Vancouver, BC

VERY light winds (less than 0.5 m/s)

Why Adjustments to Minimum Sigma-v, cont.?

• Model debugging showed the following:

– Random plume fraction is too low for very low winds – Availability of observed sigma-theta helps to increase lateral dispersion – Without sigma-theta measurements, the minimum sigma-v Without sigma theta measurements, the minimum sigma v needs to be increased from the current value of 0.2 m/s

• Key databases showing model overpredictions were

y g p near-surface releases (Idaho Falls and Oak Ridge)

• Tall stack evaluation study for Bull Run showed

y acceptable model performance for convective conditions

September 12-16, 2010 Page 16 15th IUAPPA World Clean Air Congress - Vancouver, BC

Idaho Falls QQ-plot – 1 met level, no sigma-theta with current AERMET

550 600 Idaho Falls: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Base 1-Layer) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 450 500 550 200-m Arc 400-m Arc

Large overpredictions, especially at 100 Large overpredictions, especially at 100-

• m arc

m arc (factor of 6 roughly) (factor of 6 roughly)

300 350 400 d (μg/m3)

(factor of 6, roughly) (factor of 6, roughly)

150 200 250 Predicte 50 100

September 12-16, 2010 Page 17 15th IUAPPA World Clean Air Congress - Vancouver, BC

50 100 150 200 250 300 350 400 450 500 550 600 Observed (fitted) (μg/m3)

Idaho Falls QQ-plot – 1 met level, no sigma-theta with new AERMET processing (improved u*)

300 Idaho Falls: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Modified AERMET 1-Layer) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 250 200-m Arc 400-m Arc

Better performance but still Better performance but still

150 200 ed (μg/m3)

Better performance, but still Better performance, but still

• verpredicting
• verpredicting (factor of 3, roughly)

(factor of 3, roughly)

100 Predicte 50

September 12-16, 2010 Page 18 15th IUAPPA World Clean Air Congress - Vancouver, BC

50 100 150 200 250 300 Observed (fitted) (μg/m3)

Idaho Falls QQ-plot – 1 met level, no sigma-theta with new AERMET processing (improved u* and min sigma-v)

300 Idaho Falls: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Modified AERMET 1-Layer, 0.4 Min sigma-v) Predicted Arc-wise Max @ Multiple Downwind Arcs 100 A 250 100-m Arc 200-m Arc 400-m Arc 150 200 ed (μg/m3) 100 Predicte

Some improvement from modified u* case Some improvement from modified u* case

50

September 12-16, 2010 Page 19 15th IUAPPA World Clean Air Congress - Vancouver, BC

50 100 150 200 250 300 Observed (μg/m3)

Idaho Falls QQ-plot – 1 level, with sigma-theta with current AERMET

350 Idaho Falls: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Base 1-Layer) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 200 m Arc 250 300 200-m Arc 400-m Arc 150 200 ed (μg/m3)

still overpredictions, but not as much as previous case still overpredictions, but not as much as previous case that had no sigma that had no sigma-

• theta data

theta data

100 150 Predict 50

September 12-16, 2010 Page 20 15th IUAPPA World Clean Air Congress - Vancouver, BC

50 100 150 200 250 300 350 Observed (fitted) (μg/m3)

Idaho Falls QQ-plot – 1 level, with sigma-theta with new AERMET processing (improved u*)

300 Idaho Falls: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Modified AERMET 1-Layer) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 200 250 200-m Arc 400-m Arc 150 200 ed (μg/m3) 100 Predicte

still overpredictions, but not as much as previous case still overpredictions, but not as much as previous case

50

much better performance, still high at 100 m much better performance, still high at 100 m

September 12-16, 2010 Page 21 15th IUAPPA World Clean Air Congress - Vancouver, BC

50 100 150 200 250 300 Observed (fitted) (μg/m3)

Idaho Falls QQ-plot – 1 level, with sigma-theta with new AERMET processing (improved u* and min sigma-v)

300 Idaho Falls: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Modified AERMET 1-Layer, 0.4 Min sigma-v) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 200-m Arc 200 250 200 m Arc 400-m Arc 150 200 cted (μg/m3)

I d f ith i i i I d f ith i i i

100 Predic

Improved performance with minimum sigma Improved performance with minimum sigma-

• v

50

September 12-16, 2010 Page 22 15th IUAPPA World Clean Air Congress - Vancouver, BC

50 100 150 200 250 300

Observed (μg/m3)

Overall Results for Idaho Falls

• Overpredictions clearly evident at 100 m, better model

performance further out Si th t b ti d di ti (b tt

• Sigma-theta observations reduce overpredictions (better

depiction of lateral plume spreading)

• Use of better AERMET (higher u*) reduces overpredictions by

Use of better AERMET (higher u*) reduces overpredictions by about a factor of 2

– Higher effective dilution wind speed – Higher turbulence levels in vertical and horizontal g

• Biggest improvement to model performance reformulated u*

in AERMET when lacking sigma-theta

• Increased minimum sigma-v results in modest performance

improvements

September 12-16, 2010 Page 23 15th IUAPPA World Clean Air Congress - Vancouver, BC

• Single-level AERMET works as well as 2-level AERMET

Oak Ridge QQ-plot – 1 met level no sigma-theta with Current AERMET

1600 Oak Ridge: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Base 1-Layer) Predicted Arc-wise Max @ Multiple Downwind Arcs

Large overpredictions, especially at 100 Large overpredictions, especially at 100-m arc (factor of 20+), m arc (factor of 20+),

1200 1400 100-m Arc

Large overpredictions, especially at 100 Large overpredictions, especially at 100 m arc (factor of 20 ), m arc (factor of 20 ), dominated by stable hours dominated by stable hours

800 1000 ed (μg/m3) 200-m Arc 400-m Arc 400 600 Predicte 200

September 12-16, 2010 Page 24 15th IUAPPA World Clean Air Congress - Vancouver, BC

200 400 600 800 1000 1200 1400 1600 Observed (fitted) (μg/m3)

Oak Ridge QQ-plot – 1 met level, no sigma-theta with new AERMET processing (improved u*)

900 Oak Ridge: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Modified AERMET 1-Layer) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 700 800 200-m Arc 400-m Arc

large overpredictions, but reduced from base case large overpredictions, but reduced from base case mostly during stable hours (factor of 10+) mostly during stable hours (factor of 10+)

400 500 600 ed (μg/m3)

y g ( ) y g ( )

200 300 400 Predicte

better performance, but still better performance, but still overpredicting

• verpredicting

100 200

September 12-16, 2010 Page 25 15th IUAPPA World Clean Air Congress - Vancouver, BC

100 200 300 400 500 600 700 800 900

Observed (fitted) (μg/m3)

Oak Ridge QQ-plot – 1 met level, no sigma-theta with new AERMET processing (improved u* and min sigma-v)

225 Oak Ridge: Quantile-Quantile Plot - Observed (fitted) vs AERMOD (Modified AERMET 1-Layer, 0.4 Min sigma-v) Predicted Arc-wise Max @ Multiple Downwind Arcs 100-m Arc 175 200 100-m Arc 200-m Arc 400-m Arc

m ch better performance b t still m ch better performance b t still o erpredicting

• erpredicting

100 125 150 ed (μg/m3)

much better performance, but still much better performance, but still overpredicting

• verpredicting

by factor of about 2 by factor of about 2-

• 3

3

50 75 100 Predicte 25 50

September 12-16, 2010 Page 26 15th IUAPPA World Clean Air Congress - Vancouver, BC

25 50 75 100 125 150 175 200 225 Observed (μg/m3)

Overall Results for Oak Ridge

• Substantial overpredictions occur, especially at closest

distances

• Overpredictions mostly due to model’s poor performance

Overpredictions mostly due to model s poor performance during stable hours

• AERMOD does reasonably well for unstable conditions
• There is a need to predict a larger lateral spread of the

plume for stable conditions (no sigma-theta data available here)

• Use of enhanced AERMET (higher u*) reduces
• verpredictions

– Higher effective dilution wind speed Higher effective dilution wind speed – Higher turbulence levels in vertical and horizontal

• Minimum sigma-v of 0.4 m/s substantially improves model

f

September 12-16, 2010 Page 27 15th IUAPPA World Clean Air Congress - Vancouver, BC

performance

Next Steps and Recommendations

• Provide results, code, and modeler’s archive to EPA

for review - DONE for review - DONE

• Encourage EPA to add our code changes as a beta
• ption to an upcoming AERMET/AERMOD release:
• ption to an upcoming AERMET/AERMOD release:

1. Set minimum sigma-v = 0.4 m/s instead of 0.2 m/s 2. Use alternative u* formulation for both single-level d 2 l l h and 2-level approaches

September 12-16, 2010 Page 28 15th IUAPPA World Clean Air Congress - Vancouver, BC

Questions?

Jeff Connors AECOM Environment Westford MA Westford, MA jeffrey.connors@aecom.com 978.589.3744

15th IUAPPA World Clean Air Congress - Vancouver, BC Page 29