using analysis over multiple time scales to assess air
TRANSCRIPT
Using Analysis over Multiple time Scales to Assess Air Quality Over
Delhi
Milind KandlikarArvind Saraswat
UBC
This Talk
• Overview of Air Quality in Delhi• Detection of Trends, Seasonal Cycles and
Oscillations using multiple approaches– Daily Data at a hotspot (ITO)
• Spectral Analysis (FFT, Singular Spectrum)• Weekend-Weekday differences
– Monthly Data at Multiple Locations• Trend detection
– Hourly Data at hotspot (ITO)
• Insights
Criteria Pollutants
?Lung DamageMortality?
0.08 ppm (8 hour)Ozone
40% Transport60% Industry
PM, O3, ToxicsPrecursor
0.053 ppm (100 µg/ m3)-ANOx
20% Transport
80% IndustryRespiratoryPM precursor
0.03 ppm (80 µg/ m3)-A0.14 ppm (364 µg/ m3)-D
SOx
65% Transport
10% Biomass
25% Industry
Acute effectsOzone formation
9 ppm/10 mg/ m3 (8 hr)35 ppm/40 mg/m3 (1 hr)
CO
20% Biomass
30% Transport fuel
50% Rest
Mortality (Heart disease, Lung Function, Stroke)
50 µg/ m3 (A)150 µg/ m3 (D)
PM-10
A- Annual; D- Daily
Delhi Air Quality
HEI (2001-2004)
Delhi Data at Multiple Time Scales
• Daily average data (2000-2006) from CPCB for PM-10, CO, NOx, SOx at a hotspot (ITO Crossing)
• Monthly Data at 7 locations (2000-2007)
• Hourly data at hotspot (2006-2007)
• Data at different scales provide us with different insights
Analysis of Daily Data
PM-10µ = 234 µg/ m3
σσσσ = 125 µg/ m3
Daily limit exceeded on 80% of days!
µ = 3.45 mg/ m3
σσσσ = 2.62 mg/ m3
CO
µ = 10.9 µg/ m3
σσσσ = 4.5 µg/ m3
µ = 80 µg/ m3
σσσσ = 23 µg/ m3
Spectral Analysis
• Trends and oscillations in the Data • Trends correspond to low frequency
components• Three Methods
– Power Spectral Density• Fourier Transform of the Auto-Correlation function of a
time series
– Singular Spectrum Analysis• Preserves phase information• Non-linear trends and oscillations
– LOWESS Regression
Power Spectral DensityCO and NOx
Power Spectral Density PM10 and SOx
FFT findings
• Low frequency variation, i.e., trends, dominate for CO, NOx, and SOx, but not PM-10.
• Annual/seasonal cycles for all pollutants• Filter noisy data and decompose into
trends and seasonal cycles
Trend and Seasonal Cycle Detection
• The Model
– X(t) = T(t) + ΣΣΣΣOi(t) + εεεε– T(t) = Trend
– Oi(t) = ith Oscillation /Seasonal Cycle
– & εεεε is zero mean noise
• Find Trend and Oscillations in the face of “Red Noise”
Singular Spectrum Analysis (Overview)
• A Method to Detect Oscillations – including long term ones (“trends”)
• Suited for environmental variables (climatology) because it allows for red noise
• An “optimal” spectral approach – unlike FFT, orthogonal functions emerge from data.
• Allows for non-linear modulations
• Very similar to Principal Components Analysis
PM10 with Component overlay
T +O1+O2 (47%)T +O1 (31%)T (7%)
Reconstructed Components
16%22%
11% 9%
F=1/yr
F=2/yr F=3/yr
25%
16%
11%
F=1/yr
F=2/yr
SOx Reconstructed Components
8%
F=1,0.4/yr
8% F=2/yr
32%
Net 48%
Daily data – Weekday/Weekend EffectsParameter Nox CO RSPM SOx
Intercept 27.88*1943.49*53.88* 7.35*Lagged-value 0.64* 0.75* 0.70* 0.63*Temperature -0.10*-8.70* 0.02 -0.01*Windspeed -1.56*-123.24*-2.60* -0.23*Max. Windspeed 0 0.35 0.02* 0Precipitation -1.30* 1.18 -3.96 0.08Fog -1.64 -111.429.38* -0.56*Weekday 6.80*348.72* 3.87 1.01*2001 2.89* 137.89 1.06 -1.21*2002 4.93*-466.78*21.71* -58292003 11.66*-603.01*18.58* -3.19*2004 9.87*-616.31* 9.38 -3.62*2005 8.33*-599.52*22.56* -3.34*2006 6.13*-652.43* 10.57 -2.06*
R-squared 0.65 0.72 0.58 0.67
Daily Data Findings
• Strong downward trends (50% reduction) in Sox (’01-’03) and CO (’01-'02)
• Strong upward trend in NOx (50% rise)• ~ 15% PM dip in (’04)• Strong annual cycles for all pollutants with
winter peaks (weather related)• Roughly 50% of variation in trends and
oscillations• No PM traffic signal in weekend-weekday
differences
Monthly Data
Hourly Data
Chowdhury, Z., M. Zheng, J. J. Schauer, R. J. Sheesley, L. G. Salmon, G. R. Cass, and A. G. Russell (2007), Speciation ofambient fine organic carbon particles and source apportionment of PM2.5 in Indian cities, J. Geophys. Res., 112, D15303, doi:10.1029/2007JD008386.
Wrapping up
• Increasing trends for NOx, decreasing trends for SOx and CO, PM10 ambiguous
• CNG - contributed to 10% drop in PM10….– but only at a very traffic sensitive location ?
• Need better characterization of in-use emissions
• Weather-air pollution interaction• Other sources contribute to poor air quality in
Delhi – Need more focus on those• Very little known about AQ in other cities