1. Relationship of PM and Meteorology in Fairbanks: the Sequel Robert Crawford, Rincon Ranch Consulting Tucson, AZ Sierra Research, Sacramento CA May 26, 2010

2. Background At the 2009 Symposium, we presented the first results of the PCA methodology for understanding the nexus between PM concentrations and meteorology in Fairbanks. From this, we came away seeing needs to improve and extend the analysis:  Augment the meteorology database  Include wind characteristics aloft (at or immediately above mixed layer)  Better understand the extreme event of 12/29-30/2008  Assess representativeness of 2008-2009 compared to prior 2 winters (both weather and our analysis) 2

3. Updates to the Analysis Met data regenerated using NCDC hourly observations (instead of 3 hours) and U of WY sounding data. Process automated to support analysis of earlier years. New variables introduced for air flow aloft  Average wind speed in mixed layer (replaces surface wind speed)  Average wind speed in 30m “cap” layer immediately above mixed layer  Bulk Richardson Number (turbulence vs. laminar flow) across “cap” layer Winter period extended to include March (previously October to February) to cover all winter months in which exceedances have been recorded. 3

4. Airport Observations (hourly)  Air temperature (ºF)  Wind speed (mph)  Barometric Pressure (in Hg, station)  Ceiling (100s ft)  Visibility (miles) Airport Soundings (12 hours)  Air temperatures at pressure intervals aloft  Precipitable H 2 0 (mm)  Other sounding variables considered but not used Variables computed from soundings:  Temperature profile at fixed heights up to 500m  AM Mixing Height (m)  Temperature gradient across mixing layer (ºF/m)  Avg wind speed in mixed layer (replaces surface wind speed)  Avg wind speed and Bulk Richardson Number for 30m “cap” layer above mixed layer Meteorological Variables Used 4

5. Analysis Approach Dependent variable: 24-hr average PM 2.5 concentration at downtown BAM (after calibration to FRM) Independent variables:  Meteorological conditions (15 variables)  Emissions inventories not yet available Six month winter period (Oct-Mar) Winter 2008-2009 and two prior winters  Extension of meteorology to 10+ years planned 5

6. Principal Component Analysis (PCA) A multivariate technique useful in dealing with multiple, inter-related predictors Creates eigenvectors to represent the varied inter- relationships that are empirically present in the data In LRR form (latent root regression), it extracts relationships among variables that include a dependent variable (here, PM) and its predictors PCA can be used to:  guide selection of variables for conventional analysis  create vector variables for use as predictors 6

7. Conventional Analysis  XYZ Coordinates  Correlated variables (often)  Selection of variables  Model building PCA-Based Analysis  Vector Coordinates  Orthogonal variables  Interpretation of vectors  Model estimation 7

8. PM/Meteorology Vectors  Vector 1 – Temperature Inversion (strongest PM effect)  Vector 2 – High/Low Pressure Centers (modest PM effect)  Vector 3 – Approaching High/Low Pressure (no PM effect)  Vector 4 – Wind Conditions (strong PM effect)  Vector 5 – a small vector having no importance for PM  Vector 6 – a small vector with a strong PM effect 8 Winter 2008-09 Vector 1Vector 2Vector 3Vector 4Vector 5Vector 6 Eigenvalues 8.442.141.751.460.720.52 Pct of Total 53%13%11%9%4%3% Cumulative 53%66%77%86%91%94%

9. Vector 1 – Classic Temperature Inversion Strongest PM effect – 133% increase per 1 std dev A Design Day Inversion (V1=+4.5) is shown below 9

10. 12/29-30/2008 Extreme Event Area-wide meteorology predicts very high, but not extreme levels, and these days remain outliers that are not fully explained. Local effects may be important – anecdotes say PM can be patchy and may drift like clouds when dispersion is very low. There is also some evidence for an unidentified effect on these days – a large PM impact, but infrequent occurrence. May not be fully described by the available variables. 10

11. Days Near the Design Value (41 ug/m 3 ) in Winter 2008-2009 a/ Yellow = Favors High PM Blue = Disfavors High PM Date 24-Hr PM a/ (ug/m 3 ) Airport Temperature (ºF) Vector 1 Temperature Inversion Vector 2 High/Low Pressure Center Vector 4 Wind Conditions Vector 6 Unknown 01/10/200942-414.22.8-1.60.5 12/21/200841-133.4 0.5 1.61.1 11/09/200841 -66.2-2.7-0.2 11/06/2008413.4-0.7 1.70.2 12/27/200840-294.9-0.9-0.1 a/ Based on calibrated BAM data, may differ from FRM data used to establish the design value. Design conditions are defined by the strength of the atmospheric inversion (Vector 1), modulated by the other vectors. A design day can occur at -6ºF under a very strong inversion, at -41ºF under a moderate inversion, and at other temps with multiple influences. Inversion is necessary to exceed the standard, but there are multiple pathways to a design day (and to any given PM level). Conclusions from Winter 2008-2009 11

12. Extension of Analysis to Past 3 Winters 12

13. Comparison of Winter Averages 13 VariableUnits2006-20072007-20082008-20093-Winters DwtnBAM ug/m 3 20.317.819.319.1 DryF FF 0.75.81.02.5 AvgWSpd mph 5.85.76.56.0 BaroSt inch Hg 29.8529.7529.9229.84 Ceil 100s feet 158131 140 Visib miles 8.88.68.18.5 Precip Water mm H 2 0 4.85.34.95.0 MixHt meters 446425451441 MixLapse  F / m 0.0270.028 dT050 dT(  F) 50m vs surface 2.5 2.62.5 dT100 dT(  F) 100m vs surface 4.44.34.24.3 dT200 dT(  F) 200m vs surface 6.36.05.76.0 dT300 dT(  F) 300m vs surface 7.16.66.36.7 dT500 dT(  F) 500m vs surface 7.56.86.77.0 CapWSpd mph, above mixed layer 8.98.79.99.2 BrnMixCap BRN, above mixed layer 0.300.420.280.33 VentRate m 2 /sec 1,642 1,442 2,039 1,707

14. Vector 1 – Temperature Inversion Consistent meteorological patterns over time Increasing PM in past 3 winters 1-sigma changes shown below 14

15. Other vectors are consistent in terms of meteorology, but more variable in PM Effects 15 PM Effects of Vectors 2 through 6

16. Vector 6 – Unidentified Factor Dominated by the PM effect with inconsistent meteorology except for winds. A residual term? A non-linearity? Decoupling? Other? 16

17. Winter 2008-09 View of Vector Influences Data for Winter 2008-09 Vector 1Vector 2Vector 3Vector 4Vector 5Vector 6 Eigenvalues 8.442.141.751.460.720.52 Pct of Total 53%13%11%9%4%3% Cumulative 53%66%77%86%91%94% lnDwtnBAM 0.2700.0690.0040.2210.0190.534 Pct of PM62% 1% 0% 7% 0%15% Cumulative62%63% 70% 85% 17 Emphasized Vector 1 as measure of temperature inversion and Vector 4 as measure of wind conditions. Vector 2 had a PM effect, but Vectors 3 & 5 had none. Vector 6 not considered in initial effort.

18. Composite View of Past 3 Winters Pooled Data for Past 3 Winters Vector 1Vector 2Vector 3Vector 4Vector 5Vector 6 Eigenvalues8.022.331.911.230.740.58 Pct of Total50%15%12% 8% 5% 4% Cumulative50%65%77%84%89%93% lnDwtnBAM0.2730.0940.0800.1130.0110.581 Pct of PM60% 2% 1% 2% 0%20% Cumulative60%62%63%64% 84% 18 Data for past 3 winters combined (without inventory adjustment). PCA leads to the same six vectors. All (except V5) have statistically significant PM effects Variance analysis says V1 & V6 matter most for PM

19. Design conditions continue to a governed by Vector 1 with modulation from the other vector influences. True design days (PM 2.5 = 41 ug/m 3 ) are defined by Vector 1. Vector 6 is not an important factor on design days. Key Conclusions for Past 3 Winters 19 Days Near Design Value (41 ug/m 3 ) in Past 3 Winters a/ Yellow = Favors High PM Blue = Disfavors High PM Date 24-Hr PM a/ (ug/m 3 ) Airport Temperature (ºF) Vector 1Vector 2Vector 3Vector 4Vector 6 Temperature Inversion High/Low Pressure Centers High/Low Pressure Approaching Wind Conditions Unknown 11/18/200642-194.0 0.0 0.9 1.0-0.2 11/20/200642-222.6-0.1 1.5 0.9 0.5 12/27/200642-192.8-0.3 1.2 0.2 1.2 2/22/200742-292.4 2.1 0.8-0.1 1/10/200942-413.7 3.1 1.6-2.2 0.9 11/6/2008413.2-0.3-0.4 1.5 0.2 11/9/200841 -66.1-2.2-0.5 -0.2 12/21/200841-133.2 0.8 0.3 1.2 0.9 12/11/200640 -65.2-2.5-0.4-0.3 0.8 1/20/200740 -74.5-0.6 0.6 0.7 12/27/200840-294.7-0.6 1.4-0.2 0.0 a/ Based on calibrated BAM data, may differ from FRM data used to establish the design value.

20. Design Day Meteorological Conditions 20 Days Near Design Value (41 ug/m 3 ) in Past 3 Winters a/ Date 24-Hr PM a/ (ug/m 3 ) Airport Temperature (ºF) Avg WSpd (mph) Cap WSpd (mph) BRN (laminar flow) Mixing Height (m) Ventilation Rate (m 2 /sec) Maximum120 4138504.5 1,998 34,353 Average 19 0 6 90.4 403 1,499 Minimum 1 -43 0 10.0 24 7 11/18/2006 42 -19 2 30.2 73 54 11/20/2006 42 -22 4 80.6 108198 12/27/2006 42 -19 4 60.1 111184 2/22/2007 42 -29 2 30.4 178162 1/10/2009 42 -41 2 50.5 116112 11/6/2008 41 2 30.2 93 68 11/9/2008 41 -6 2 60.2 41 40 12/21/2008 41-13 1 30.2 97 54 12/11/2006 40 -6 3 60.1 52 74 1/20/2007 40 -7 1 30.2 77 44 12/27/2008 40 -29 2 60.1 69 70 a/ Based on calibrated BAM data, may differ from FRM data used to establish the design value.

21. Summary The challenge in PCA is understanding (and explaining) what meteorological conditions the vectors represent. But the picture is straightforward:  The combination of temperature and inversion strength largely defines design day conditions.  Variations in ventilation (mixing height, wind speed, laminar airflow) modulate the PM concentrations.  Vector 6 remains a puzzle, but not germane to design day. The results support and enlarge on our prior understanding:  There are multiple meteorological pathways to a design day and to any given PM 2.5 level  Design conditions are much less severe than extreme days. 21