Improved MODIS Aerosol Retrieval using Modified VIS/MIR Surface Albedo Ratio Over Urban Scenes Min Min Oo, Matthias Jerg, Yonghua Wu Barry Gross, Fred Moshary, Sam Ahmed Optical Remote Sensing Lab City College of New York 1
Aerosol Retrieval from Satellites over Coastal Urban Areas Comparisons of MODIS AOD to our ground based Aeronet sky radiometer AOD retrievals reveal the aerosol optical depth derived from the MODIS sensor is significantly overestimated over New York City Current algorithms do not model the surface albedo for urban areas very well which is critical since the ground reflectance in the visible is determined as some correlation multiplied by the MIR (2.1um) reflectance A better understanding of this correlation between the visible and mid-infrared surface albedo is needed to tune the algorithm for urban areas 2
MODIS and Aeronet AOD Comparison (10km) using C005 Significant overestimate seen in operational product AERONET (CIMEL) dataset was the average of a 4 hour interval surrounding the MODIS observation and at least 10 data points must be available and stable within +/-10% 3
Higher spatial resolution retrievals To assess the effects of urban surfaces on aerosol retrieval at higher resolution, a reprocessing of the (500m) L1B product must be initiated. This includes the implementation of a cloud clearing algorithm that will provide outputs at higher resolution Normally, at 10km resolution, a certain fraction of cloud cleared high resolution pixels must appear in the low resolution cell for the 10km product to be considered cloud cleared In examining higher resolution, we break the data into 3 x 3 cells (like the standard algorithm) and mark the center pixel clear if all radiometric and smoothness tests are passed However, unlike the standard 10km mask that requires a certain percentage of cloud cleared pixels, we perform our processing directly on the high resolution data 4
High resolution cloud clearing Image (example) 2 4 6 8 10 12 14 0.0612 0.061 0.0608 0.0606 0.0604 0.0602 0.06 If the red cell represents the 1.5 km region around the CIMEL Radiometer, the Reflectance is taken to be the mean of all TOA reflectances In the region (even if only 1 500m pixel is chosen) 16 18 20 2 4 6 8 10 12 14 16 18 20 0.0598 0.0596 0.0594 This is done to obtain more Data for analysis without Too much error expected 5
Intercomparison at higher resolutions assuming C005 albedo τ MOD ( 550) 2 1.8 1.6 1.4 1.2 1 0.8 y = 1.2*x + 0.11 τ MOD ( 550) 1 0.9 0.8 0.7 0.6 0.5 0.6 10 Km 0 0 0.5 1 1.5 τ aer ( 550) 0.3 0.1 1.5 Km 0 0 0.1 0.3 0.5 0.6 0.7 0.8 0.9 1 τ aer ( 550) Our high resolution data was only collected when aerosols were fine mode and optical depth <.2 since this filtering is needed for calculating surface reflection directly (in 2120nm) 6
Observations Increasing the spatial resolution from 10km to 1.5 km leads to significant increase in bias Even small error in the ground reflectance model will lead to dramatic changes in AOD (if AOD is small) Necessary to refine the surface reflection model to remove the bias even for the 10km product as well as pushing the resolution higher 7
Direct surface using Hyperion Imagery including Supervised Classification RGB Image with Training regions Segmented Image Blue=Water Green=Vegetation Red=Urban Atmospheric correction Included using FLAASH software based on Visibility level of 120km (as seen on next slide) and the use of a standard urban aerosol model 8
Sensitivity to atmospheric correction ρ ( 470) / ρ (2160) ~.25 c(470) g g Collect 4 ρ ( 660) / ρ (2160) ~.50 c(660) g g Collect 4 Urban scenes have a significantly larger VIS/MIR coefficient. Since algorithm underestimates albedo, AOD is generally overestimated 9
Operational MODIS C005 Surface Reflection Ratios (assumption) Rho466/ Rho2120 0.36 0.34 0.32 0.3 8 6 4 2 0.1 0.3 0.5 0.6 0.7 0.8 0.9 0 100 200 MVI C004 Scattering Angle Rho644/ Rho2120 0.62 0.6 0.58 0.56 0.54 0.52 0.5 8 6 4 0.1 0.3 0.5 0.6 0.7 0.8 0.9 0 100200 MVI C004 Scattering Angle MVI ρ + ρ TOA TOA 1240 2130 = TOA TOA MVI (Modified Vegetation Index) ρ1240 ρ2130 10
Collection 5 approach Clearly, different surfaces have different correlations with clear differences between vegetation and urban scenes so Collection 4 algorithm was not reasonable The Collect 5 approach allows the VIS-MIR ground albedo correlation coefficients to be a function of surface type (urban/vegetation MVI) and observation angles (scattering angle). This is done by performing global matchups of aeronet aerosol to MODIS to pull out surface albedo. The C005 approach also tries to get some estimate of aerosol reflectance at 2130nm to account for errors when large particulates (ex. Desert Dust) are present. 11
Obtaining surface albedos using Combined MODIS Aeronet Data Aeronet Optical Depth + MODIS Aerosol Phase Function consistent with AOD [ 550 ] aer aeronet [ ] aer scat aer urban nonabs ( 550 ) τ P Θ,τ, λ Use Aeronet AOD to fix the MODIS Aerosol Phase function model From this, we can get all relevant atmospheric scattering parameters ρ s ( ) atm λ, θ v, θ i, Δφ = path reflectance d (, ), λ θ = ( λ ) = Atmospheric spherical albedo T u Upward and downward total transmission ρ TOA = ρ Once this is done, we can Isolate Lambertian albedo atm ρ gtdt + 1 sρ g u ρ g = T d T u ρ ρ TOA atm + s( ρ ρatm ) TOA 12
Correlation coefficients as function of scattering angle 10 km resolution Rho460/Rho2120 Rho660/Rho2120 Scattering angle Scattering angle Very small changes observed as function of scatter angle Lambertian assumption seems reasonable in first order 13
Correlation coefficients as function of scattering angle 1.5 km resolution Rho460/Rho2120 Rho660/Rho2120 Scattering angle Scattering angle ρ g ( λ ) = f ( Θ ) ρ ( 2120) i i i = 1: 2 s g Once new correlations are found, we can replace the COO5 Correlation procedures and assess retrieval of AOD (for all cases) 14
AOD Retrieval with refined model at different resolutions 15
Validation of Surface albedo model over vegetation (rural) 1.5 Km resolution 460/2120 660/2120 Scattering angle Scattering angle Good agreement to C004/C005 reference Due to surface albedo ratios of both urban and vegetated area weakly depend on scattering angle, the surface albedo ratios MAP can be considered (next slide) 16
Surface Reflectance Ratio MAP Google MAP surface surface ρ 460 / ρ 2120 surface surface ρ 660 / ρ 2120 surface surface ρ 460 / ρ 2120 VIS/MIR correlation coefficient ratios of 460nm/2120nm and 660nm/2120nm in nearby New York City area. VIS/MIR ratio is significantly higher in the urban area compared to the vegetated areas 17
NY City area MVI (Modified Vegetation Index) 41 1.4 40.95 40.9 40.85 40.8 0.5 5 0.35 0.3 5 Rho460/Rho2120 1.2 1 0.8 0.6 40.75 0.15 0-0.1 0 0.1 0.3 0.5 0.6 1.2 MVI 40.7 40.65-74.2-74.1-74 -73.9-73.8 MVI = ρ ρ TOA 1240 TOA 1240 + ρ ρ TOA 2130 TOA 2130 0.1 0.05 0 Rho 660/Rho2120 1.1 1 0.9 0.8 0.7 0.6 0.5 0.3-0.1 0 0.1 0.3 0.5 0.6 MVI 18
Observations No tuning is needed for vegetative areas since correlation coefficients are in good agreement with operational results Refined correlation coefficient model is needed for urban area and significantly improves the final AOD retrieval To assess whether high resolution AOD retrieval is possible, we need to examine the resolution of the underlying retrieved surface. 19
Mexico City urban area To verify with another heavy urban area, Mexico City is selected Unlike New York City, Mexico City is situated ~ 2.2 Km above sea level and most of the fine mode aerosol model is within Smoke Aerosol model (MODIS LUT) category surface reflectance ratios and AOD retrieval procedure are the same method as before in NY City case The results of VIS/MIR surface ratios are similar to NY City urban area s outcome (next slide) 20
Mexico City surface reflectance ratios (10x10 Km^2) 0.65 0.6 0.55 data 1 linear 0.8 0.78 0.76 y = 0.00094*x + 0.57 data 1 linear Rho470/Rho2120 0.5 5 0.35 0.3 y = 0.00085*x + 0.32 5 90 100 110 120 130 140 150 160 170 Scattering angle Rho670/Rho2120 0.74 0.72 0.7 0.68 0.66 0.64 0.62 90 100 110 120 130 140 150 160 170 Scattering angle Average rho470/rho2120~ 335 Average rho670/rho2120~ 0.6974 21
Mexico City surface reflectance ratios (3x3 Km^2) Rho470/Rho2120 0.7 0.65 0.6 0.55 0.5 5 0.35 y = 0.000205*x + 16 data 1 linear Rho670/Rho2120 0.8 0.78 0.76 0.74 0.72 0.7 0.68 data 1 linear 0.3 0.66 5 90 100 110 120 130 140 150 160 170 Scattering angle 0.64 y = 0.00039*x + 0.66 0.62 90 100 110 120 130 140 150 160 170 Scattering angle Average rho470/rho2120~ 424 Average rho670/rho2120~ 0.7121 22
AOD Retrieval with local (tune) VIS/MIR surface reflectance ratio MODIS C005 L2 AOD <550nm> 1.4 1.2 1 0.8 0.6 C005 0 0 0.6 0.8 1 1.2 1.4 AERONET ( AOD <550nm> ) Modified Vis/Swir AOD <550nm> 1.4 1.2 1 0.8 0.6 10Km 0 0 0.6 0.8 1 1.2 1.4 AERONET ( AOD <550nm> ) 1.5 Using local Vis/MIR surface reflectance ratios, significant improvement in retrieved AOD matchup with Mexico City s AERONET station is obtained Modified Vis/Swir AOD <550nm> 1 0.5 3Km 0 0 0.5 1 1.5 AERONET ( AOD <550nm> ) 23
Mexico City MVI Vs Correlation Coefficient Rho466/ Rho2120 1.4 1.2 1 0.8 0.6 0-0.1 0 0.1 0.3 0.5 0.6 MVI Rho644/ Rho2120 1.2 1.1 1 0.9 0.8 0.7 0.6 0.5 0.3-0.1 0 0.1 0.3 0.5 0.6 MVI NY City Area = Blue o Mexico City = red * 24
New York City regional AOD retrieval Example (10-3-2006) 41.4 41.3 41.2 41.1 41 40.9 0.9 0.8 0.7 0.6 0.5 0.3 41.4 41.3 41.2 41.1 41 40.9 0.9 0.8 0.7 0.6 0.5 0.3 A significant improvement can be observed as artificial hot spots in the AOD map are significantly reduced 40.8 40.8 40.7 0.1 40.7 0.1-74.2-74.15-74.1-74.05-74 -73.95-73.9-73.85-73.8-73.75-74.2-74.15-74.1-74.05-74 -73.95-73.9-73.85-73.8-73.75 Regional map of AOD (550nm) retrieval with modified VIS/SWIR ratio (left panel) and retrieval with Collection (5) algorithm (right panel) Date: 10-03-2006. 25
Histogram of retrieved AOD (10-03-2006) τ 550 5 0.35 0.3 5 AERONET MODIS Terra pass Time 1600 CCNY AERONET station derived AOD ~ 6 MODIS (C005) L2 data AOD = 0.52 MODIS ( modified VIS/MIR) AOD = 0.32 0.15 0.1 12 13 14 15 16 17 18 19 20 Coordinate Universal Time (UTC) 26
Conclusions This procedure shows that operational algorithms significantly underestimate the critical surface correlation coefficient ratios over urban areas. Overestimates in aerosol retrieval in heavy urban area can be corrected using refined surface reflection obtained from coincident AERONET radiometer / MODIS measurements 27