The Basics

1
A Probabilistic Nighttime Fog/Low Stratus
Detection Algorithm Corey G Calvert and Michael J
Pavolonis Cooperative Institute for
Meteorological Satellite Studies, Madison,
Wisconsin NOAA/NESDIS/Center for Satellite
Applications and Research Advanced Satellite
Product Branch, Madison Wisconsin

Fog Detection Approach
Reducing Noise from the Final Product
Small-Scale Detection Capabilities

• The Basics
• The aviation-based level of cloud ceiling under
  which pilots must fly using instrument flight
  rules (IFR) is approximately 300m. Therefore, the
  goal for the GOES-R fog detection algorithm is to
  create a fog mask to detect liquid stratus clouds
  with bases lower than 300m.
• The GOES-R fog detection algorithm builds off the
  widely-used 3.9, 11?m channel combination for
  nighttime detection of fog.
• Strategy
• Nighttime fog is typically characterized by the
  following traits
• Cloud top is close to ground so the difference
  between cloud temperature and surface temperature
  is typically small
• Relatively high spectral emissivity signal
  (3.9,11?m) at night
• Fog detection is based on finding small
  differences between the radiometrically-derived
  and NWP surface temperature along with strong
  signals from the 3.9?m pseudo-emissivity.
• Rather than using specific thresholds, training
  data were used to create look-up tables (LUTs)
  that assign a probability a pixel returning
  certain spectral information is fog.
• Cloud objects are created to group neighboring
  pixels with similar radiometric signals. This
  allows pixels within an object that have a
  stronger signal (usually at the center) to
  represent the entire object, which is useful for
  small-scale fog events (see far right).
• Algorithm
• The GOES-R fog detection algorithm can be broken
  down into the following four steps

• Due to the spatial resolution (4km) of current
  GOES satellites, detecting small-scale fog events
  (e.g., within river valleys) is difficult.
• The use of cloud objects can enhance the
  detection ability by allowing pixels with a
  stronger radiometric signal to represent pixels
  within the object that may not have otherwise
  been classified as fog.
• This helps areas such as fog edges where the
  signal may not be strong enough by itself to be
  classified as fog, but has pixels in the same
  object with a stronger signal that is more
  representative.
• Using the cloud objects can help to restore
  detail that may be missed without increasing the
  spatial resolution of the instrument.
• The images below depict a valley fog event over
  Pennsylvania and upstate New York on September
  17, 2007. The upper left image is the first
  daylight image (1145 UTC) showing the fog within
  the valleys. The upper right image is the high
  resolution (1 km) MODIS fog/stratus product at
  738 UTC. The lower left image is the heritage
  fog algorithm at 745 UTC displaying fog where
  the BTD is below -2 K. The lower right image is
  the corresponding GOES-R fog product creating
  cloud objects using the 3.9?m pseudo-emissivity.
  The improved spatial resolution of the GOES-R ABI
  will greatly enhance the future GOES fog product.

MODIS Fog/Stratus Product

• In the upper left false color image, the
  white/red crosses represent surface observations
  meeting the no fog/fog criteria, respectively,
  for non ice clouds (given by GOES-R cloud type
  product). The light blue/magenta crosses
  represent surface observations meeting the same
  criteria respectively under multi-layer or ice
  clouds (areas excluded from the GOES-R fog
  detection algorithm). The upper right image is
  the GOES-R cloud type product.
• The heritage fog algorithm (bottom left image)
  flags pixels with a 3.9-11?m BTD less than -2 K.
  In the presence of convective clouds and non-fog
  water clouds this algorithm has the tendency to
  return noisy pixels that are usually false
  alarms.
• The GOES-R algorithm (bottom right image) screens
  these areas out using the cloud type information
  along with 3.9 ?m pseudo-emissivity data. It also
  uses cloud objects to group neighboring pixels
  with similar radiometric signals. Using a minimum
  object size of 3 pixels, it can significantly
  reduce noise in the final product.

• Look-up tables were created using surface
  temperature bias along with the 3.9?m
  pseudo-emissivity as predictors. Truth was
  gleaned from surface observations.
• The 3.9 ?m clear-sky surface emissivity was also
  used to separate pixels with different surface
  types (e.g., desert and forest) which might lead
  to unrepresentative fog probability calculations.
• Below is the fog LUT for the 3.9?m
  pseudo-emissivity and surface temperature bias
  for pixels with a clear-sky surface emissivity
  between 0.90 and 0.95.

• Pixels with small surface temperature biases and
  low 3.9 ?m pseudo-emissivity will be given a
  higher probability of containing fog.

For further information contact the author at
corey.calvert_at_ssec.wisc.edu