Download presentation
Presentation is loading. Please wait.
Published byDuane Jennings Modified over 8 years ago
1
Causes of Haze Assessment (COHA) Update Jin Xu
2
Update Visibility trends analysis (under revision) Assess meteorological representativeness of 2002 (modeling base year) (in progress) PMF Modeling and case study (in progress) Evaluate winds used in back-trajectory analysis (to be done)
3
Trends Analysis Pages Are there any statistically significant multi-year trends in the haze levels and causes of haze? http://coha.dri.edu/web/general/tools_trendanaly.html National maps and tables Individual site analysis
6
Trends Analysis for Aerosol Light Extinction Coefficients (1/Mm) in 20% Worst Days
7
Trends Analysis for Nitrate Light Extinction Coefficients (1/Mm) in 20% Worst Days
10
Meteorological Representativeness of 2002 - Backtrajectory Analysis –Generate 8-day back-trajectories of all WRAP IMPROVE aerosol monitoring sites (every 3 hrs, from 3 starting heights) for 2003 and 2004 to give 5 years of trajectories –Produce residence time maps for 2002 and the 5-year period (2000 – 2004), plus maps of ratios and of differences of 2002 and the 5-year period for each site –Interpret the maps for each monitoring site and document on the COHA web site Differences in residence times between July-October 2002 and the five-year (1998-2002) July-October average at Big Bend. Blue colors denote greater back trajectory residence times in 1998-2002, and red colors denote greater residence times in 1999.
11
Meteorological Representativeness of 2002 - National Temperature and Precipitation Maps Maps of average temperature and precipitation total averaged by state and whether it was normal, much above, much below etc. summarized by month, season, or year.
12
Positive Matrix Factorization for Groups of Sites PMF is a statistical method that identifies a user specified number of source profiles (i.e. relative composition particle species for each source) and source strengths for each sample period that reduce the difference between measured and PMF fitted PM 2.5 mass concentration In matrix notation, X = GF + E where X is the matrix of measured composition for each sample period, F is the source profile, G is the source strength or factor scores for each sample period, and E is the residual or error matrix.
13
Model Description X = GF + E X (n * Sp) = a matrix of observed fine particulate species concentrations with the dimensions of number of observations by the number of species G (n * f) = a matrix of source contributions by observation day with the dimensions of number of observations by the number of factors F (f * Sp) = a matrix of source profiles with the dimensions of number of factors by the number of species E (n * Sp) = a matrix of random errors with the dimensions of number of observations by number of species
14
Model Description – Cont.
15
Robust Mode – the value of outlier threshold distance = 4.0 i.e. if the residue exceeds 4 times of the standard deviation, a measured value is considered outlier. The least squares formulation thus becomes: Error Mode (decides the standard deviation of the data S ij ): EM = -12 (based on observed value) S ij = T ij + C*X ij EM = -14 (based on observed and fitted value) S ij = T ij + C*max(X ij, FPEAK and FKEY Matrix (controls the rotation) – default: 0 (central), try different numbers PMF Running Parameters
16
PMF Inputs PM2.5 chemical speciation data from VIEWS web site. Data are screened to remove the days when either PM10 or PM2.5 mass concentration is missing. Data value and associated uncertainty (T) If data is missing Then data value = geometric mean of the measured values uncertainty = 4 * geometric mean of the measured values Else if data bellows detection limit data value = 1/2 * detection limit uncertainty = 5/6 * detection limit Else data value = measured data uncertainty = analytical uncertainty + 1/3 * detection limit
17
PMF Outputs Source factor profiles (ug/ug) Contribution of each source factor to aerosol mass and light extinction for each sampling day at each monitoring site (ug/m 3 )
18
How Many Source Factors? Regression coefficients for PM2.5 > 0 Scaled source profiles <1 Experience (arbitrary)
19
PMF for Group 1 – Washington State Class I Areas: MORA1, NOCA1, OLYM1, PASA1, SNPA1, SPOK1, and WHPA1
20
Urban/Diesel Aged sea salt Sulfate-rich secondary Smoke Dust Industrial/Incinerator Nitrate-rich secondary Smoke II ? Dust II
21
Two smoke factors are not correlated
22
Two dust factors (factor 5 and factor 8) are highly correlated – Maybe 8 factors is enough
23
Smoke Nitrate-rich secondary Sulfate-rich secondary Dust Urban/Diesel Aged sea salt Industrial/Incinerator Smoke II
24
Urban/Diesel Aged sea salt Sulfate-rich secondary Smoke Dust Smoke II Dust II Smoke Nitrate-rich secondary Sulfate-rich secondary Dust Urban/Diesel Aged sea salt Smoke II Industrial/Incinerator Nitrate-rich secondary 9 Factors 8 Factors
25
Two smoke factors from the 8 factor modeling correlated well with the two factors from the 9 factor modeling
26
Sulfate-rich secondary Mixture Nitrate-rich secondary Dust Urban/Diesel Aged sea salt Smoke How about 7 factors – only one smoke factor left, no industrial/incinerator, add a mixture factor (smoke, dust, and urban/power plant?)
27
Urban/Diesel Dust Nitrate-rich secondary Sulfate-rich secondary Aged sea salt Smoke Mixture Contributions to PM2.5 Mass (7 Factors)
28
The single smoke factor from 7 factor modeling is correlated to the sum of two smoke factors in 8 factor modeling The correlation between the single smoke factor from 7 factor modeling and any one of the two smoke factors in 8 factor modeling is not very high Have we identified different smoke factors?
29
Let’s try 6 factors – no mixture factor any more.
30
Percentage Contributions of PMF Factors to Major PM2.5 Components at Mt. Rainier
31
Percentage Contributions of Major PM2.5 Components to PMF Factors at Mt. Rainier
32
Factor Contributions to PM2.5 Mass at Mt. Rainier (3/1988-2/2004) – Compare With Keith Rose’s PMF Results
33
PMF application to Hawaii IMPROVE Particle Speciation Data All available PM 2.5 speciation data for both sites (>2 years each) are used together in the PMF to explain measured PM 2.5 mass Six factors seemed to separate reasonably explained source factors Multiple linear regression was used to explain coarse mass using the six PMF factors
34
Haleakula and Hawaii Volcano National Park Monitoring Sites
35
Six Source Profiles from Hawaii PMF Analysis #1, Sea salt #3, Dust #4, Smoke #5, Secondary Nitrate #6, Secondary Sulfate & Nitrate #2, Volcano sulfate
36
Contributions to PM 2.5 by Source Factors Haleakula Hawaii Volcano All DaysWorst 20% Haze Days Volcano Sea salt Dust Smoke Nitrate Sulfate & Nitrate Site
37
Contributions of Source Factors to PM2.5 in 20% Worst Days of 2003 Haleakula Hawaii Volcano At Haleakula, about half of worst haze days are associated with volcano emissions, while the others are associated with different factors (e.g. smoke, secondary sulfate and nitrate) At Hawaii Volcano, all worst haze days are dominated by the volcano sulfate factor. Note that October 24, 27, & 30 had trajectories from the volcano to Haleakula
38
PMF Work Plan PMF modeling for each group of sites (based on AOH report) using all the IMPROVE data available at the site. Case study for selected sites: PMF modeling for individual site using data from certain time period (e.g. 2000-2004). Compare PMF results for the selected sites based on group modeling and individual modeling. Combine PMF modeling results with the backtrajectories and emission inventories to investigate the major source regions of certain aerosol sources (e.g. smoke) for each site. Episode analysis based on PMF results
39
Backtrajectory analysis for PMF modeled factor 5 (BWS5) (Weighted – Unweighted). This serves to confirm that the factor 5 is in actual fact a “vegetative burn” factor from wildfires to the northwest of Boundary Waters Canoe Area IMPROVE site (Engelbrecht et al., 2004). Backtrajectory Analysis for PMF Factor - Example
40
PMF Modeling for Group 19 (BRCA1, CAPI1, ZICA1 and ZION1) Secondary Sulfate Smoke Secondary Nitrate Dust Mobile & Other Urban
41
Time Series of Factor 2 (Fire) Contributions 10/30/2003
42
Time Series of Factor 4 (Dust) Contributions 10/30/2003
43
10/30/2003 –The worst day at all three sites Intense wildfires burning around Los Angeles and San Diego, very windy Cedar City, Utah hourly wind speed on 10/30/2003, max gust 53mph
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.