1. Under the direction of Rudolf Husar
Estimation of Daily Surface Reflectance Over the United States from the SeaWiFS Sensor
April 29, 2000, Day 120
July 18, 2000, Day 200
October 16, 2000, Day 290
Sean Raffuse
Under the direction of Rudolf Husar
Thesis presented to the Henry Edwin Sever Graduate School of Washington University in partial fulfillment of the requirements of the degree of Master of Science
May 23, 2003

2. Outline
Goal
Introduction
Approach
Methodology
Results
Discussion

3. Goal
Development of a procedure for the automated production of daily surface reflectances from SeaWiFS satellite data
Applications of surface reflectance data
Vegetation mapping
Aerosol retrieval
Radiative balance/climate
Domain of study
Continental United States
April – August 2000

4. Introduction – SeaWiFS Satellite Platform
SeaStar satellite maps the world daily in 24 polar swaths, carrying the Sea-viewing Wide Field-of-view Sensor (SeaWiFS)
The 8 channels of the sensor are in the transmission windows between the atmospheric gas absorption bands in the visible & near IR
Swath
2300 KM
Polar Orbit: ~ 1000 km, 99 min.
Equator Crossing: Local Noon

5. Radiation detected by satellites
Air scattering depends on geometry and can be calculated (Rayleigh scattering)
Clouds completely obscure the surface and have to be masked out
Aerosols redirect incoming radiation by scattering and also absorb a fraction
Surface reflectance is a property of the surface

6. Apparent Surface Reflectance, R
The surface reflectance R0 is obscured by aerosol scattering and absorption before it reaches the sensor
Aerosol acts as a filter of surface reflectance and as a reflector solar radiation
R = (R0 + (e-t – 1) P) e-t
Aerosol as Filter: Ta = e-t
Aerosol as Reflector: Ra = (e-t – 1) P
Surface reflectance R0
The apparent reflectance , R, detected by the sensor is: R = (R0 + Ra) Ta
Under cloud-free conditions, the sensor receives the reflected radiation from surface and aerosols
Both surface and aerosol signal varies independently in time and space
Challenge: Separate the total received radiation into surface and aerosol components

7. Spectra of surface reflectances
Surface reflectance R0 is dependent on wavelength, surface type, and scattering angle
Aerosol (haze) modifies sensed reflectance
Chlorophyll Absorption
Soil
Vegetation
Water

8. Scattering angle correction 1
Surface reflectance is dependent on sun-target-sensor angle (non-Lambertian)
Time series shows dependence

9. Scattering angle correction 2
Pixels are normalized to a scattering angle of 150°

10. Preprocessing Transform raw SeaWiFS data
Georeferencing – warping data to geographic lat/lon coordinates with a pixel resolution of ~ 1.6 km
Splicing – mosaic data from adjacent swaths to cover entire domain
Rayleigh correction – remove scattering by atmospheric gases and convert to reflectance units
Scattering angle correction – normalize all pixels to remove reflectance dependence on sun-target-sensor angles
Result is daily apparent reflectance, R for all 8 channels

11. Approach – Time Series Analysis 1
For any location (pixel), the sensor detects a “clean” day periodically
Aerosol scattering (haze) is near zero, thus R ≈ R0
Pixel must also be free of other interferences
Clouds
Cloud shadows
R = (R0 + (e-t – 1) P) e-t

12. Approach – Time Series Analysis 2

13. Methodology – Cloud shadows
Clouds are easily detected by their high reflectance values
Cloud shadows are found in the vicinity of clouds
We enlarge the cloud mask by a three-pixel ‘halo’ to remove cloud shadows
Cloud shadows reduce the apparent surface reflectance considerably in all channels

14. Methodology – Preliminary anchor days
Surface reflectance is retrieved for individual pixels from time series data (e.g. year)
The procedure first identifies a set of ‘preliminary clear anchor’ days in a 17-day moving window
The main interferences (clouds and haze) tend to increase the apparent surface reflectance, especially in the low wavelength channels
The anchor day is chosen as the day with the minimum sum of the lowest four channels

15. Methodology – Refinement/Interpolation
Anchor days are further refined using a jump filter on the channel 1 (blue) time series
Surface reflectance in channel 1 does not change rapidly
Channel 1 is strongly affected by haze
If an anchor day is over reflectance units greater than the previous good anchor day, it is assumed to be influenced by haze and is removed.
Values are interpolated between the refined anchor days to create the daily surface reflectances

16. Methodology – Residual haze reduction 1
In some regions, aerosol haze is persistent throughout over long periods e.g. Southeast in the summer
Anchor days chosen by the routine may still contain small amounts of haze, especially vegetation and water pixels
Spectral analysis is used to reduce the residual haze over these surfaces

17. Methodology – Residual haze reduction 2
Vegetated surfaces do not have a channel 1 reflectance greater than 0.03
Haze increases the apparent reflectance most in channel 1 and somewhat less at each subsequent band
Vegetation pixels with excess channel 1 reflectance are reduced to 0.03
All other channels are reduced proportionately

18. Methodology – Residual haze reduction 3
Hazy water pixels are reduced using the assumption that water is completely black (reflectance = 0) at channel 8 (near-IR)
The residual haze reduction does not completely eliminate haze, but provides a good estimate

19. Process Flow Diagram L1A Key Create Geometry Rayleigh Correction
Scattering Angle Corr.
A  B Conversion
Georeference L1B
Georef. geometry
Filter bad pixels
L1A
Geometry
File
Filtered
L1B
Warp
Points
Georeferenced
Daily Radiance
Image
Rayleigh Corrected
Reflectance
Splice, merge, crop
Daily Reflectance
Mask clouds
Calibration
File
Cloud
Mask
Cloud/Cloud
Shadow Mask
Enlarge cloud mask
Select anchor points
Initial
Anchor Points
Clean Surface
Reflectance
Refine anchor points
Haze Reduction
Refined
Anchor Points
Surface
Images
Interpolate
Key
Inputs multiple swaths from a single day
Outputs single file
Inputs all daily files from the time span
Inputs single file
Outputs daily file for each day in the time span
Elevation
Data

20. Results – Seasonal surface reflectance, Eastern US
April 29, 2000, Day 120
July 18, 2000, Day 200
October 16, 2000, Day 290

21. Results – Seasonal surface reflectance, Western US
April 29, 2000, Day 120
July 18, 2000, Day 200
October 16, 2000, Day 290

22. Results – Seasonal surface reflectance urban pixels

23. Results – Eight month animation

24. Results – Spatial variation: 9 pixel rectangle
Adjacent pixels show similar values in some areas, more variability in others

25. Results – Comparison with clean air mass
Surface reflectance estimates should be similar to apparent reflectance values for days with clean air
Daily reflectance
Ch1 difference
Surface reflectance
Channel 1 difference is near zero except at clouds and areas of rapid surface change (max difference ~ 0.02)

26. Discussion - Advantages
Resolution independent – adaptable to other datasets that operate at different resolutions that provide appropriate spectral coverage (available bands near 0.4, 0.6, and 0.85 mm)
Fully automated, requiring no user input once initiated
Spatial, spectral, and temporal resolution of the sensor data are maintained
Minimal need for a priori aerosol knowledge
Detects surface color change on the order of days/weeks when cloud free data exist

27. Discussion – Limitations
Requires 30 – 60 consecutive days of input data
Does not fully remove the haze influence from the surface reflectance
Currently tuned to SeaWiFS

28. Limitations – Cloud shadows
Some cloud shadows remain in the surface reflectance data
Could be removed in future studies with a final spike filter on the time series.
Daily Image
Surface Reflectance

29. Limitations – Georeferencing
Quality of results is dependent on accuracy of georeferencing
Process preferentially selects dark pixels, creating a spreading effect at sharply contrasting images
Daily Image
Surface reflectance

30. Future Work Test other regions and years
Compare year-to-year results
Improve cloud shadow filtering
Aerosol retrieval
Using surface reflectance data and aerosol model
Refined surface reflectance
Using retrieved aerosol

31. Acknowledgements
Fang Li
Eric Vermote
Rudolf Husar

32. Thank You!