Cloud Detection from Space using passive imagers

1
Cloud Detection from Space (using passive imagers)

• Steve Ackerman
• Cooperative Institute for Meteorological
  Satellite Studies
• CIMSS UW-Madison

2
Overview

• Clouds
• Passive Observations
• MODIS approach
• Cloud detection approach
• Validation
• Summary and Assignment

3
Some Cloud Climatologies

• International Satellite Cloud Climatology
  Project (ISCCP)
• NOAA University of Wisconsin HIRS (UWHIRS)
• Cloud climatologies exist for most missions
  (NIMBUS-7, SAGE, EOS)
• Other AVHRR climatologies includes several region
  studies (PATMOS, APP-x Polar Regions, SCANDIA
  - European)
• MODIS (UW - starting about 2000)

4
Comments
APPROACH EVALUATION

• Solar and Infrared spectral methods from
  imagers/sounders
• Temporal test or spatial consistency tests
• ISCCP standard along with AVHRR and HIRS CO2
  slicing

• Visual inspection
• Comparison with separate methods
• Comparison with lidar/radar analysis
• Comparison with aircraft observations

5
Some tests see cloud.
6
Some tests see cloud, some dont
7
What is a cloud?
8
MODIS

• The MODIS (Moderate Resolution Imaging
  Spectroradiometer) measures radiances at 36
  wavelengths including infrared and visible bands
  with spatial resolution 250 m to 1 km.
• MODIS cloud mask algorithm uses conceptual
  domains according to surface type and solar
  illumination including land, water, snow/ice,
  desert, and coast for both day and night.
• a series of threshold tests attempts to detect
  the presence of clouds or optically thick aerosol
  in the instrument field-of-view.

9
MODIS spectral tests

• No one test dominates
• Global means can differ by 2
• Reflectance at 0.86 micron over oceans and out of
  sun-glint holds potential for comparison with
  other satellite cloud records

10

Our approach to the MODIS Cloud Mask, is for each
pixel to provide a confidence flag that indicates
how certain we are that the pixel is clear.

• Restrictions
• Real time execution
• Computer storage (4.8 g bytes per day)
• Comprehension

11
Thresholds range from 0 to 1
12
Cloud Mask Confidence

• Confidence intervals are based on closeness to a
  threshold
• Confidence tests are combined to arrive at a
  Quality Flag (2 bits)

13
Quality Flags

• Each test returns a confidence (F ) ranging from
  0 to 1.
• Similar tests are grouped and minimum confidence
  selected min (Fi )
• Quality Flag is
• Four values 0, gt.66, gt.95 and gt.99

14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Aqua MODIS from February 3, 2003
MODIS Band 7 Hudson Bay
20
Gray snow/ice
Aqua Cloud Mask Surface Snow/Ice Test
21
Aqua MODIS from February 3, 2003
Very Dry
MODIS Band 27 Hudson Bay
22
Aqua MODIS from February 3, 2003
MODIS Band 26 Hudson Bay
23
Aqua Cloud Mask Band 26 Cloud Test without
Correction
24
Green conf. clear Cyan prob. clear Red
uncertain White cloudy
Aqua Cloud Mask Final Result without Correction
25
(No Transcript)
26
Aqua MODIS from February 3, 2003
MODIS Band 2 Sahara Desert
27
Aqua MODIS from February 3, 2003
MODIS Band 26 Sahara Desert
28
Aqua MODIS from February 3, 2003
MODIS Band 33 Sahara Desert
29
Aqua MODIS from February 3, 2003
MODIS Band 28 Sahara Desert
30
Aqua Cloud Mask Band 26 Cloud Test without
Correction
31
Green conf. clear Cyan prob. clear Red
uncertain White cloudy
Aqua Cloud Mask Final Result without Correction
32
Terra MODIS band 3, proposed smoke mask, and
cloud mask for 6 July, 2002, 1550 UTC.
33
Clockwise from top left Terra MODIS band 1,
band 20, proposed cloud mask, and smoke mask
for 6 July, 2002, 1545 UTC.
34
How to explain differences

• Is it algorithm differences?
• Is it instrument differences?
• Is it orbit differences?

Need to understand the characteristics of the
algorithm and its application.
35
Cloud detection depends on IFOV
Small pixel more clear
36
Sensitivity to Input Reflectance Biases and
Reflectance Thresholds Daytime Terra MODIS Data
April 1, 2003 60N to 60S
37
Cloud detection depends on threshold
Lower 0.86 reflectance threshold more cloud
38
Cloud detection depends on threshold
Lower 0.86 reflectance threshold more cloud
39
Cloud detection depends on threshold
Sensitive to view angle
Lower 0.86 reflectance threshold more cloud
40
Validation Approaches

• Image analysis
• Field experiments
• Aircraft missions
• Ground-based observations
• Consistency Checks
• Global Statistics
• Comparison with other satellite analysis

41
Global Cloud Detection Comparison
July 2002 cloud frequency from Terra MODIS
Collection 3 cloud mask and NOAA-16 CLAVR.
42
3-Hourly Cloud Changes Measured by Terra and Aqua
Shown at left are zonal values of daytime land
Terra and Aqua total high cloud frequency (top),
and high, opaque cloud frequency (bottom) from
August 24, 2002. The latter are mostly cold
convective towers. With a local observing time of
about 130 pm, roughly three hours later than
Terra, we expect the Aqua measurements to
indicate more convective activity and hence more
thick, high clouds and more high clouds in
general. This is clearly seen in the tropics and
northern hemisphere where solar heating is
greatest. For reference, the same data is
plotted for ocean surfaces (right) where we would
not expect to see changes in high clouds due to
solar heating. Differences between Terra and
Aqua are small as we expect, especially for high,
opaque clouds
43
Validation of MODIS Cloud Mask
Comparison of cloud heights from the Micropulse
Lidar/ Millimeter Cloud Radar (MPL/MMCR) at the
DOE ARM SGP CART site to MODIS cloud mask
results. The MODIS cloud mask algorithm and
MPL/MMCR agreed on the existence of clear or
probably clear 86 of the time (86 65/175) and
92 of the time that a cloud was present. An
uncertain result occurred in less than 3 of the
total comparisons.
44
CALIOP and MODIS make very different measurements
with different sampling characteristics. To
correctly compare, collocation must be done
carefully!
MODIS 250 m resolution image with the CALIOP
sampling represented by the red line. The finer
resolution of CALIOP makes careful collocation
important in an analysis of combined data streams.
45
Careful Collocation
46
Global Comparison
 
Comparison of MODIS cloud detection with
collocated observations from CALIPSO for the
entire month of August 2006. Over 5 million
observations went into the analysis. The results
are expressed as a percentage.
47
AUGUST 2006
The fractional agreement between the MODIS and
CALIPSOCALIOP cloud mask for clear FOV is
presented in the top image. The fraction
agreement calculated at 5-degree resolution in
the figure. A grid cell with perfect MODIS
agreement will have a fractional of 1 (red) while
regions of poor agreement are colored blue. The
bottom figure presents the number of MODIS FOV
for each grid cell used to generate the fraction
agreement.
48
AUGUST 2006
The fractional agreement between the MODIS and
CALIOP cloud mask for cloudy FOV is presented in
the top image. The fraction agreement calculated
at 5-degree resolution in the figure. A grid cell
with perfect MODIS agreement will have a
fractional agreement of 1 (red) while regions of
poor agreement are colored blue. The bottom
figure presents the number of MODIS FOV for each
grid cell used to generate the fraction
agreement.
49
MODIS capability for regional studies
Island effects on cloud fraction
50
Annual Cloud amount around Hawaiian Islands
Cloud fraction in 1 degree grids
Alliss
51
Lee side of South Georgia Island has reduced
cloud cover
Cloud fraction in 1 degree grids
52
Summary

• Cloud coverage varies with
• the spatial resolution of the instrument
• spectral resolution of the instrument
• viewing geometry and scene illumination.
• MODIS dependence has be quantified
• The dependence of cloud detection on these
  parameters and the need to monitor with changing
  instruments and satellites, will likely make it
  difficult to compare cloud amounts from different
  approaches and achieve the 1 accuracy needed for
  long-term monitoring of cloud amount.
• Reflectance of 0.86 over dark ocean may be a good
  approach to compare different methods.
• MODIS cloud detection optical depth threshold
  0.4
• Level-3 properties are accurately capturing small
  spatiotemporal scale variability.

53
Give it a try.

• Three data sets all collocated MODIS with
  CALIOP. A visible, a window IR and both.