Developing a Dust Retrieval Algorithm
Jeff Masseyaka
“El Jeffe”
Motivation
• Dust can cause the snowpack to melt out a month in advance causing many water management issues
• Need a better understanding of processes behind how dust plumes originate and where they originate from
Dust Events timing
Occur an average of 4 times a year
Most common in spring Most common in afternoon
Background• Dust detection uses the IR and visible bands• Dust can only be remotely detected during the
day (zenith angle < 80), when clouds aren’t present, and when there is no snow or ice on the ground
• There are different detection schemes over the ocean and land, this project is only concerned with land
• MODIS (36 channels, 6 used) and GOES (5 channels, 4 used) data was used
IR bands: Split window technique
• Dust has a higher spectral absorption at 11 microns than 12 microns• Opposite for clouds
• Brightness temperature differences can detect dust.
• Less pronounced in thick dust near the surface since transmission distinction is weaker
• Similarly, dust has higher absorption at 3.9 microns and lower absorption at 11 microns than clouds
Visible Light difference
• Dust becomes increasingly absorptive with decreasing visible wavelengths (absorbs more blue light)
• This method is most effective over water since land surface can look similar to dust
Utah Specific Dust detection limitation
• Optically thick dust near the surface produces small BT differences• Utah dust is from
nearby point sources that usually does not leave the boundary layer
SLC
Additional limitations• Algorithm may need tuning for different
seasons as brightness temperatures change• False positives tend to show up over cold
ground (mountains), or desert areas• Areas far away from nadir are more likely to
have false positives
Zhoa et al (2010) Algorithm
4/19/2008 at 19Z (1pm MDT)• Strong SW winds over Utah
and Nevada (v>25kts) ahead of land falling Pacific trough
• Clear skies over majority of area
• Dust plumes identifiable on visible image making algorithms easier to test
• Near solar noon so reflectivity adjustment errors should be low
• Multiple dust plumes over different regions make for an interesting event
Zhoa Algorithm
Looks like all this did was detect deserts and mountains.
What went wrong?• To get brightness temperature I inverted the
Planck function, thus assuming the earth is a blackbody at these wavelengths
• Wavelength differences:• Location differences
• They used Mexico to test their algorithm
• Different season?• Did they assume dust was above BL?
They Used
I used
.47 .470
.64 .659
.86 .8651.38 1.363.9 3.9611 11.0312 12.02
Adjustments after trial and error
Overall the Following Occurred:
(1) Brightness temperature differences relaxed(2) Reflectivity conditions were relaxed and simplified(3) Reflectivity and brightness temperature conditions were combined
Results for 4/19/2008
Comparison with AVHRR algorithm
Note: images are about an hour apart. MODIS is 18Z, AVHRR is 19z
Other events:
Top: non-dust eventUpper right: 3/22/2009Lower right: 3/21/2011
3/21/2011 compared to navy algorithm (only a couple of weeks archived)
Goes algorithmTheory: focus on BT differences where there aren’t clouds
Dust retrieval will be lower resolution
More false positives over “dusty” terrain since reflectivity constraints were removed
4/19/08 14:45Z to 4/20/08 01:15Z every 15 to 30 minutes
Conclusions• “All data is bad, but
some is useful”
• Data cannot be fully trusted, but GOES makes it easier to separate dust from false positives
• Important tool for researching dust event case studies
Questions?