Noel Davidson

G'Day Folks,

My friend and colleague, Lawrie Rikus - our radiation guru - has an algorithm to create synthetic satellite images from
numerical model data. He has run his scheme off output from our hires TC system, TC-LAPS, on an experimental forecast of Chris,
which made landfall as a cat4/5 on the NW coast of Aust in Feb 2002.

Attached are gifs of the actual satellite imagery, and a 42 hour forecast of the satellite imagery, valid near to the time of landfall.
       
We think this is a truly remarkable comparison. The depiction of the near-core convection, the cloudbands at outer radii, the
cirrus shield  with embedded convection to the southeast, and even the strato-cu in the southerlies off the W. Aust coast is most fascinating.
The track error for this time is ~150km. Initial central pressure was ~985 hPa. Estimated and forecast
central pressures at 42 hours were/are 930 and 960 hPa.
{We are hoping some-one might do a Dvorak analysis on the forecast imagery for us, to see if it agrees??}
We actually initialise the model with a synthetic vortex and then build in the clouds and asymmetries using the satellite imagery,
so that the model knows something about the cloudiness at t=0.

Is this a fluke? More work needed but Lawrie has comparisons every 6 hours which show similar quality. He also has results using operational
forecasts for TY Utor (July 2001) as it passed to the north of the Philippines. These are also encouraging.

Besides the obvious forecast applications, wouldn't it be nice to (a) verify the cloud symmetries and asymmetries more
generally for many cases using this technique, and (b) try to understand the contributions of the evolving vortex, wind shear and environment
to the cloudband and cloudiness changes?

For more info about this great work by Lawrie -> l.rikus@bom.gov.au
Best wishes
Noel Davidson

Roger Edson
Noel,

I hate to burst your bubble....

1st) For starters, you can't do a Dvorak analysis without knowledge of
blackbody temperatures from the IR imagery (we use the NESDIS 'BD" curve).

2nd) The 'newly' created TC is all eye--without an eyewall (not quite what I
would imagine for an 'ideal' structure)

and 3rd), I can swear you have the 'outflow' along the northwest quadrant of
the TC running clockwise---this is 'ok' in the northern hemisphere, but not
quite as good in the southern hemisphere.

Good luck on your next try...

Roger
(always a pain-in-the-neck)

Mark Lander

Hi Noel,

    The model cloud presentation is capturing sufficient detail to make
the image almost look like real clouds.  A side-by-side comparison of the
TCs (real and model) (attached Fig) yields quite a different intensity
estimate.  First of all, since the imagery is IR, one can't do a proper
Dvorak intensity analysis on it, unless the temperatures of the eyewall
clouds are known.  But, a general evaluation of the TC structure could
yield a reasonable intensity:

    In the real image, there is a small well-defined eye.  The eyewall
ring looks to be quite cold and completely encircles the eye.  At a
minimum the real image could be given a T6 (115 kt) and probably higher.

    In the model image, the eye is large and ragged, with a break in the
eyewall on the NE side.  On EIR analysis, the prescence of the break in
the eyewall will really hammer the T number down.  But, just as a pattern
type, this TC could be given a T5 (90 Kt).

    So the real TC looks to be a solid CAT 4 hurricane (maybe even a CAT
5), while the model TC is sitting down in the CAT 2 range.

    Regards,  Mark Lander

Chris Velden

Hi Noel. This is an intriguing prospect/idea (forecasting a Dvorak intensity
from NWP cloud output). Would/could it be better than the actual model
MSLP/maxwind forecast? Is the assumption here that the model cloud forecasts
initialized from the parameterized convection might yield a better TC intensity
forecast (indrect) than the (direct) balanced model geopotential fields (MSLP)?

The previous responses by Roger and Mark are correct. The Dvorak method needs
brightness temps (Tbb). We would have to begin by "reverse-engineering" the
image to calibrate and convert the model-produced images from the digitized
(gray scale) cloud field to a simulated Tbb field. Then apply the ODT to the
simulated Tbb image. Any ideas? Would be happy to pursue this with you.

Chris

Lawrie Rikus

Chris,

The model imagery Noel sent to the discussion group is in fact brightness
temperature as seen by the GMS-5 IR1 sensor (although I could produce the
same imagery using the GOES sensor if it is of any use to you). It is
calculated from the model's cloud and temperature fields using a radiative
transfer code and is not a proxy field. The grey scale is the same for
both the synthetic and genuine satellite brightness temperatures. (The
synthetic visible images are also quite interesting because they give a
different weighting to the cloud fields and the surface albedo can be
'turned off' which further emphasizes the cloud structure).

Given that this is already bright temperature what else do we need for the
Dvorak technique?

P.S. The replies I have seen so far have shown that synthetic satellite
pictures are a useful product - the model cyclone properties have been
quickly assessed from only one picture!

Lawrie

Jeff Kepert

Noel

I'm highly impressed by your's and Lawrie's satellite image. Clearly the
model is doing a highly impressive job of getting a great many things
right, and (if nothing else) this is a potentially great diagnostic tool
for getting a quick feel as to whether the model is on or off track, on
a particular forecast.

But wasn't your Dvorak comment supposed to be just a little
tongue-in-cheek? I'm a little surprised its being taken seriously -
wouldn't the 15 km resolution mean the crucial coldest cloud tops are
unlikely to be sufficiently accurate?

But if not, how about once-and-for-all solving the calibration issues
with the Dvorak technique by using the model clouds to accurately and
definitively tune the CI => Vmax  relationship in the various basins,
and get rid of all those nasty uncertainties caused by using the "real"
data.

And yes, my tongue is most definitely planted, quite firmly, in my
cheek!

Jeff