1. Time-Resolved & In-Depth Evaluation of PM and PM Precursors using CMAQ Robin L. Dennis Atmospheric Modeling Division U.S. EPA/ORD:NOAA/ARL PM Model Performance Workshop U.S. EPA/OAQPS February 10-11, 2004 Research Triangle Park, NC

2. Objectives of Diagnostic/In-Depth Model Evaluation Test the model to check –Reliability of the Predictions (Right Reason) Right answer for the right reason Wrong answer for the right reason or understandable reason –Right Response Reasonably accurate response (a major focus of the work) Separate sources of error –Discern among: Emissions input error Meteorological error Chemistry/aerosol physics and chemistry error Aid model developers in identifying and treating problem areas

3. We need to understand what is behind the comparisons, to help interpret them. Importantly, we have to understand how the models’ state aligns with the real world state. –The model as predictor. –The model as imperfect. This talk will focus on the inorganic PM system. Focus on urban areas, where people live. The complementary probing with PM box models is very important, but will not be discussed in this talk.

4. Overview of Talk Issue of model “structure”, specifically meteorology and K Z Two issues relating to emissions inputs Reminder of the issue of oxidized nitrogen chemistry (total- nitrate) Assess the inorganic system state of the model A conclusion: We lack critical, key measurements to evaluate the model system, leaving us partially blind in our examination of the model as predictor

5. Issue of Model Structure/Meteorology We see a persistent premature collapse of the boundary layer and a morning rise of the mixed layer that is too slow. Always been there. We see this with the conservative species. The premature collapse also exists in the rural areas.

6. There is a clear, rapid rise to overprediction in the evening as the PBL collapses around 17:00 EST. Atlanta, August 1999: EC

7. The pattern of overprediction in the evening and morning occurs day-in and day-out.

8. We see similar behavior for NO Y and CO, especially the evening over- prediction. The obs rise more than for EC.

9. EC, NO Y and CO have the same diurnal pattern

11. Suburban/ Rural NO Y We also see the pattern of overprediction at suburban/rural and rural sites

12. Suburban/ Rural CO We also see the pattern of overprediction at suburban/rural and rural sites

13. There is a systematic problem with MM5 that leads to a premature collapse of the boundary layer. We need to be aware of how this affects comparisons. For the nighttime concentrations we have a situation of compensating errors. I do not think one should arbitrarily change CMAQ’s default K Z to get better performance statistics for O 3 without a thorough analysis for the period being simulated with regard to the conservative species like EC, CO and NO Y. For CMAQ, concentrations during the daylight hours, when the atmosphere is well mixed, are the best for checking the model for issues such as bias.

14. Emission Input Issues EC –We are not discerning the bias with 24-hr averages. NH 3 –Our ignorance regarding ammonia’s diurnal profile is causing problems to model ammonia concentrations.

15. Daylight hour predicted EC concentrations are low, indicating the EC emissions are biased low Atlanta: EC

16. The daylight hour EC underprediction is true for almost every day of the month Atlanta: EC

17. While the synoptic-scale agreement is quite good, the agreement of the 24-hr averages is for the wrong reason. Emissions of EC are actually baised low, something not discernable from 24-hr averages. Atlanta: EC

18. We use inverse modeling to set the overall monthly level of ammonia –factor of 1.2 x’s 1999 NEI annual average parsed into monthly 12ths for month of August –Factor of 0.4 x’s 1999 NEI annual average parsed into monthly 12ths for month of January Where we can test it against NH X (= NH 3 + NH 4 + ) it works pretty well. Ammonia

19. The CMAQ NHX predictions are tracking the synoptic signal quite well, but they are not tracking the measured diurnal pattern Atlanta: NH X

21. While the NHX pattern is not as pronounced as the EC pattern, it is most likely also caused by the MM5 issue along with possible errors in the NHX diurnal profile in SMOKE. How to separate? Atlanta: NH X

22. While there is an issue with NH X, for sulfate the diurnal pattern is inverted, the range of variation is smaller, and model and measurements are in much better agreement. Atlanta: SO 4 2-

23. Diurnal biases in NH X show up as biases in aerosol nitrate, especially in the early morning. Atlanta: NO 3 -

24. The predicted NH X also has a more pronounced diurnal swing in winter, with the evening peak showing the largest deviation or bias. Pittsburgh, January 2002: NH X

25. >>Reminder<< Oxidized Nitrogen Chemistry: total-Nitrate Heterogeneous N 2 O 5 Reaction 2002 release of CMAQ –Reaction probability γ = 0.1 recommended by Dentener and Crutzen (JGR 1993) –Makes a lot of HNO 3 at night Recent studies show wide range of γ values –Dependence on humidity, temperature, chemical composition - sulfate, nitrate, and organic content 2003 Release of CMAQ –Reaction probability  γ  = 0.002 - 0.02 depending on NO 3 /(SO 4 +NO 3 ) according to Riemer et al. (JGR 2003) based on lab measurements of Mentel et al (PCCP 1999)

26. HNO 3 concentrations significantly reduced with updated CMAQ Must turn off all production from N 2 O 5 to get down to observed levels of HNO 3 Atlanta: HNO 3 (average diurnal cycle) UrbanSuburban

27. Suburban Atlanta: HNO 3 (average diurnal cycle ) Daytime over-production of HNO 3 is also an issue

28. Same behavior of HNO 3 overprediction is observed at Pittsburgh. The overprediction of HNO 3 appears relatively smaller in summer (no daytime issue) than in winter. Winter may have bigger issues. Pittsburgh: WinterAtlanta: Summer

29. Pittsburgh: total-NO3 January ‘02 Pittsburgh: NH X January ‘02 At Pittsburgh the wintertime relative overprediction of total-NO3 is larger than the relative overprediction of NH X.

30. Pittsburgh: total-NO3 January ‘02 Pittsburgh: NH X January ‘02 Also seen in 24-hr data: At Pittsburgh the wintertime relative overprediction of total-NO3 is larger than the relative overprediction of NH X.

31. Setup of the Inorganic System State of the Model What sort of problem do these biases appear to create in terms of setting the model up for predicting the PM response to emissions reductions? We will use the Gas Ratio suggested by Spyros Pandis to examine the system state. First, what do the time-resolved patterns look like relative to the average diurnal patterns. We will include model sensitivities to help us learn.

32. Sulfate tracks pretty well except for a few excursions

34. Total-nitrate is overpredicted. Zeroing the heterogeneous production brings total-nitrate very close to the observations.

36. The NH X predictions track fairly well, but with periods of overprediction. Not much difference between model versions.

38. Gas Ratio (per S. Pandis) Free Ammonia NH X - 2 * SO 4 2- GR = ---------------------- = ------------------------------ Total Nitrate HNO 3 (g) + NO 3 - (p) Calculated in Molar Units GR > 1 => HNO 3 limiting 0 NH 3 limiting GR NH 3 severely limiting (can’t form NH 4 NO 3 )

44. The CMAQ O3 Release looks best, even though it has clear biases

45. Observations Gas ratio analysis shows that the model will need a “right” combination of off-setting errors to come close to the control response state of the atmosphere. May require some bias. Uncertainty in the ammonia inventory is a serious issue. PM predictions are very sensitive to errors in the NH X. Get NH X. We need a top-down engineering exam using measurements. Other sources of error combine differently with the MM5 or meteorological error, so that the errors do not consistently affect the PM predictions across different sections of the diurnal cycle. Errors are not necessarily consistent across space (needs to be further tested). We have a dilemma. Do we want the model to look good; use official inputs? Or Do we want the model to be a good predictor?

46. Observations (cont.) High time resolution is necessary to check for bias. Agreement on 24-hr averages may be for the wrong reason. Comparisons must include and involve multiple species, including conservative tracers. It is important to assess the models’ state relative to emissions changes. Currently this is not possible because we are missing key gas species and the temporal coverage is inadequate. Without measurements of NH 3 and HNO 3 to go along with aerosol measurements (forgetting size for the moment) we are walking into the SIP process partially blind as to the quality of the models. –Need NH 3 and HNO 3 24-hr averages, minimum, preferably hourly. –Need measurements every day, all seasons.