Hi Jacob,
Thanks for the explanation of how RTMA/URMA analyzes clouds. I have
been looking at the operational RTMA/URMA sky analysis grids (for
daytime) yesterday and today in AWIPS/GFE and I have to give them only
about a C-. I would imagine that the quality of the sky analysis is
also impacting the quality of the NBM output. The methodology you
describe sounds pretty good and it seems like with some work the sky
analysis could easily become an "A" especially utilizing
during the day over the water where there is high contrast between
cloud and water in the visible satellite imagery. Using the new GOES R
channels should make this even more reliable. Is there a way we can
see what the "GOES imager sky cover obs" look like? It seems
like the satellite observations of sky cover should be excellent and
therefore the Sky analysis should be excellent.
I am guessing that not as much time or effort have been put into
the Sky analysis as say the T Analysis. I would suggest we stop adding
new elements to the RTMA/URMA or NBM and concentrate on improving some
of the basic forecast elements such as Sky. By adding new elements we
are not moving toward our goal of utilizing NBM for first guess
because now there are too many elements with poor quality. In
addition, much time, discussion and effort are spent on temperature.
Yes, temperature is important, but everyone thinks in terms of the
quality of temperature grids as if that reflects the entire quality of
the RTMA/URMA and NBM which it does not. It seems that verification
statistics of NDFD temperature showing little if any improvement in
over NBM have somehow becomes an argument that we can stop spending
time on the entire grid forecast. On the other hand there has been
great improvement in the RTMA and NBM over the past few years and
forecasters don't always think in terms of today's performance, but
rather past performance or a single bad experience. I would encourage
everyone to keep working on these basic elements as I believe with
dedication and teamwork, utilizing machine learning and AI techniques,
the girded elements in RTMA/URMA and NBM can become excellent. And
thus, the quality of the forecast can be improved which is more
significant than reducing forecaster time spent on grid editing.
Here are a couple of thoughts/observations for improving sky
analysis in RTMA/URMA:
1. I did some crude work using the GFE and I think we could
develop a methodology that would detect the sky cover very well over
the water, which would be very beneficial to coastal marine offices.
This in turn should have a positive feedback on the NBM sky forecast.
As you said the sun angle plays a role in the reflectance count
returned from clouds over the water. For example, I found that, at
14z reflectance (or count value if that is the correct term) values
greater than 40 correspond to a "cloudy" area (say something
with clouds greater than 50%). As the sun rises reflectance of 40 is
too low to separate clouds from clear air. After a couple of hours
during the morning the whole image is over 40 so that number would
need adjustment for sun angle. I didn't do anything formal, but it
looks like around 40 at 14z and by 20z a reflectance value of 55
separates cloudy from partly cloudy and once above about 80 it is
likely cloudy. (See attached examples). This could vary from season
or location on the globe and there might be a slight dependence of
water temperature.
2. When the sun angle is low, using the visible imagery alone can
cause areas of "clear" sky to appear when there is higher
clouds (e.g. a cirrus band) that are casting a shadow on a lower
stratus cloud deck...resulting in a false negative. This might be
mitigated by employing other channels to detect the lower clouds or
even detecting the higher cloud tops and predicting what shadowing
effect they may have.
3. Over land using the visible imagery is more challenging. The
reflectance count alone can give a lot of false positives over areas
with bright surfaces and snow. For example Mt Rainer and Mount Hood
would be detected as clouds right now because they of snow cover...and
during the winter there would be lots of snow cover over broad areas.
4. It appears to me that the HRRR background is degrading the
analysis. I look at the HRRR grids where the sky is clear, then the
satellite image for the same time and location where there are clouds
and the URMA seems to be weighted heavily toward the incorrect HRRR
forecast. If the satellite detection of sky is perfected there may be
no reason to use background field for Sky grids. How can the model be
expected to improve upon what is actually observed at each grid cell
by the satellite? I can see that it makes sense to start with a model
background for temperature analysis because there are many grid cells
where there are no observations, or multiple observations that may
conflict. However, with sky there is actual high quality observed data
(from the satellite) for every grid cell. True, it will take some work
to leverage all the different satellite channels to provide an
excellent sky analysis grid, but I am confident that it can be done.
5. If a background model is going to be used the HRRR may not the
best. My observations indicate the the HIGHRESWarw and RAP both give a
better areal representation of clouds...although they seem to under
do the percent of cloud cover a bit. While the HRRR is better showing
a higher density of cloud cover it often is not as good or detailed
with the areal coverage of marine clouds.
6. The METAR observations have too much influence over too broad an
area and I can see large circles of cloud cover likely resulting from
METAR observations that are really only valid at a point (ASOS looking
straight up) or at most one 2.5 km grid box vs a 50km diameter
circle. We can easily have low stratus covering a valley where a
METAR observation is detecting OVC007 but a few km away it is clear.
I wonder if METAR observations are even worth including in the
analysis, but it is possible that utilizing some techniques the METARS
combined with the satellite data could be leveraged to improve the
analysis. .
7. At night the difference products for fog/stratus detection could
be use for sky cover detection.
Thanks,
Bill
