Evaluation of REMSAD- BRAVO Simulations Using Tracer Data and Synthesized Modeling Michael Barna Cooperative Institute for Research in the Atmosphere Colorado State University, Fort Collins, CO Bret Schichtel, Kristi Gebhart and William Malm Air Resources Division National Park Service, Fort Collins, CO PM Model Performance Workshop RTP, NC 10-11 February 2004
35
Embed
Evaluation of REMSAD- BRAVO Simulations Using Tracer Data and Synthesized Modeling Michael Barna Cooperative Institute for Research in the Atmosphere Colorado.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Evaluation of REMSAD-BRAVO Simulations Using Tracer Data
and Synthesized Modeling
Michael BarnaCooperative Institute for Research in the AtmosphereColorado State University, Fort Collins, CO
Bret Schichtel, Kristi Gebhart and William MalmAir Resources DivisionNational Park Service, Fort Collins, CO
PM Model Performance WorkshopRTP, NC
10-11 February 2004
Acknowledgements
• Assistance for the REMSAD simulations conducted at CIRA/CSU
– Betty Pun, Shiang-Yuh Wu and Christian Seigneur (AER): initial assistance with REMSAD and met data processing
– Hampden Kuhns (DRI) and Jeff Vukovich (MCNC): emissions inventory
– Eladio Knipping and Naresh Kumar (EPRI): sulfur concentrations from GOCART
– Nelson Seaman (PSU): MM5 simulations
– Sharon Douglas, Tom Myers (ICF) and Tom Braverman (EPA): useful discussions on model evaluation
BRAVO: a study designed to understand haze at Big Bend National Park• Big Bend NP is located in remote southwestern
Texas, along the Texas/Mexico border
• Haze has increased in recent years – a rarity for a western park
• BRAVO (Big Bend Regional Aerosol and Visibility Observational Study) investigates the pollution sources that are contributing to this haze
– Field program: July-October 1999
– Many participants:
EPA NPS NOAA
EPRI CSU DRI
TCEQ AER Et al.
Flight Over BBNP Area (5 November 2003)
Who is contributing sulfate to BBNP?• Sulfate is the main constituent of visibility-impairing PM at
BBNP
• Who is contributing?
– the Carbon I/II power plant just over the border?
– sources in eastern Texas?
– sources in the eastern US?
– how large is the influence of the boundary concentrations?
Big Bend, Bext Budget, BRAVO
01020
3040506070
8090
100
7/1 7/15 7/29 8/12 8/26 9/9 9/23 10/7 10/21
1/M
m
Rayleigh Sulfate Nitrate Organics LAC Fine Soil Coarse
BRAVO’s “weight of evidence” approach to determine sulfate attributions
• Don’t rely on one analytical method or model; rather, use “weight of evidence” approach:
Source-oriented models: Receptor-oriented models:
Hybrid models:
REMSAD TrMB “Synthesized REMSAD”
CMAQ FMB “Synthesized CMAQ”
This talk will look at three ways to evaluate the BRAVO air quality simulations
• Simulation of conserved tracers
– Important but somewhat dull (Barna)
• Simulation of sulfate with base emissions
– Important but somewhat dull (Barna)
• Identifying model biases using “synthesis inversion analysis”
– Exciting! (Schichtel)
Evaluating the REMSAD BRAVO sims
• Simulation of conserved tracer
– examine transport and dispersion of conservative tracers
– if model can’t simulate transport and dispersion there’s not point in continuing
• Simulation of sulfate with base emissions
– time series analysis of predicted sulfate against BRAVO and CASTNET monitors
– evaluate different periods to identify potential temporal biases
– evaluate different monitors to identify potential spatial biases
– evaluate at spatial patterns of interpolated observations and predictions – do the match?
Evaluating the REMSAD BRAVO sims (cont’d)
• Use “synthesized inversion modeling” to identify biases with respect to different source regions
– A hybrid approach that starts with attribution results from REMSAD (or CMAQ or any model)
– Use a statistical approach to identify multiplicative terms for each source region that would result in a best fit to the measurement data
– If REMSAD attributions for that source region are
• perfect: scaling coef = 1
• underestimated: scaling coef > 1 (i.e., need to increase)
• overestimated: scaling coef < 1 (i.e., need to decrease)
Simulation of conserved tracers
Predicting transport is the most important aspect of air quality modeling
• No other modeled process, e.g., emissions, deposition, chemical transformation, has as big an impact on model results as transport
• transport = advection + turbulent diffusion
• A tracer experiment is the most robust method for evaluating transport
– Halocarbon tracer is conserved – negligible transformation and deposition
– Detectable at very low concentrations
– We know release rates – can check skill of receptor models for determining attribution
– expensive
BRAVO tracer source and receptor sites
Tracer release sites:•Eagle Pass•San Antonio•Big Brown PP•Parish PP
Performance (or lack thereof?) statisticsEagle Pass NE Texas Houston
San Antonio
Average Observed (ppqV) 0.21 0.00 0.06 0.52
Average Predicted (ppqV) 0.39 0.02 0.03 0.33
R: 0.47 0.34 0.31 0.52
Normalized Gross Error: 412% 130% 74% 70%
Normalized Bias: 380% 65% -71% -24%
• What do we expect for “good performance”? Expecting perfection is naïve….
– Grid models aren’t ideal for simulating plumes – the “real” plumes likely have very strong concentration gradients that won’t be represented by model
– Complex terrain is complex…and will not be resolved at 36 km
Problems with this time series analysis
• Tracer concentrations at two of the four sites are too low for meaningful time series analysis (negative concentrations!), but there is still useful information here
• Looking at the preceding time series, your eye tells you that the model clearly has some skill (e.g., timing of Eagle Pass tracer), but this is not reflected in the bias or error statistics
Comparing interpolated spatial patterns• Need to move beyond simple time series analysis to something more comprehensivie