Interpreting UW-CIMSS Advanced Microwave Sounding Unit (AMSU) Imagery/Products
Background:

The AMSU instrument detects earth/atmosphere emitted radiation in the microwave portion of the electromagnetic
spectrum. Microwaves, in comparison to more familiar infrared or visible radiation, are less energetic and have
longer wavelengths (distance between successive wave crests/troughs) on the order of centimeters (10-2 meters)
vs. micrometers (10-6 meters). Based on energy considerations and AMSU-A instrument performance, this dictates
that the sensor be placed in a lower earth-orbit (~810km above the earth's surface vs. ~36,000km for geostationary
satellites) to improve instrument signal-to-noise. The AMSU-A (temperature sounder), a 15 channel
passive radiometer, detects energy emitted by atmospheric molecular oxygen (a major atmospheric constituent)
and is largely unaffected by the presence of clouds--from the emission source, through the atmosphere, to the
sensor which resides on the NOAA-15, NOAA-16 and NOAA-17 polar orbiting satellites.

AMSU-A Sensor and Radiative Transfer Theory:

Contributions to the upwelling terrestrial radiation sensed by the AMSU-A (neglecting the effects of reflection
and scattering which themselves can behave as source/sink terms) are largely comprised of two terms--the earth's surface
and the overlying atmosphere. Individual AMSU-A channels (i.e., frequencies) are carefully chosen based on principles
of radiative transfer theory. Each channel (frequency) is radiatively selective in the sense that it detects microwave
radiation from discrete layers within the earth's atmosphere. Satellite meteorologists typically relate the radiation
sensed in individual AMSU-A channels/frequencies to specific atmospheric layers (characterised by the abundance
of molecular oxygen O2 and temperature) by use of a term called a weighting function:


Source: Kidder et al. CSU/CIRA, 1999


Simply put, the weighting function for AMSU-A Channel 7 (54.94 GHz) has a maximum amplitude (i.e., contribution to
upwelling microwave radiation sensed by the AMSU-A instrument) at approximately 250 hPa (~12km above the earth's
surface) whereas Channel 5 (53.6 GHz) has a maximum weighting function at approximately 550 hPa (~5km above
the earth's surface).

Exploitation of AMSU-A Channels 5-8:

As tropical storms develop into hurricanes, they're characterized by upper tropospheric warming (the troposphere being
defined as that portion of the earth's atmosphere extending from the earth's surface to approximately 15km altitude)
as a result of adiabatic warming/compression of air as it subsides (sinks) within the storm center. The following is an example of a cross-section image of Hurricane Floyd derived from AMSU-A data (contours are degrees C above storm environmental temperature):


AMSU-A warm anomaly: Hurricane Floyd


A similar, more familiar, warming occurs when inflating a bicycle tire with a hand held-pump. Within the sealed pump
cylinder, air is compressed and warms. The result is a bicycle pump that becomes warm to the touch. Subsidence within a tropical storm/hurricane warms the overlying troposphere, suppressing clouds and leads to the characteristic 'eye'
commonly observed in satellite pictures.

AMSU-A Channels 5-8, as visualized on the UW-CIMSS AMSU Homepage during tropical storm/hurricane events,
measure the storm-related warming at different elevations within the troposphere:

Channel 8 (55.5 GHz) ~100mb (~15km)
Channel 7 (54.94 GHz) ~200mb (~12km)
Channel 6 (54.46 GHz) ~350mb (~10km)
Channel 5 (53.6 GHz) ~550mb (~5km)

As storms mature and the circulation/associated convection become more organized, the amount of microwave radiation emitted by the atmosphere towards the AMSU-A instrument increases as tropospheric temperatures (again, as a result of storm-related subsidence/warm ing) increase. The exception occurs in cases where ice/liquid water droplets
reduces the upwelling radiation due to the effects of scattering (commonly seen in AMSU-A Channel 5).

The matrix of AMSU-A brightness temperatures (analogous to temperature) generated for each storm system shows the time evolution (from left to right) as well as the vertical distribution of storm-related tropospheric heating (top to bottom). In general as the storm becomes more intense the warm anomaly seen in the images will increase in magnitude. This is best seen in channels 7 and 8 which are high enough to not be affected by precipitation. In storms with very large eyes or storms which have lost much of their convection strong warming may be present in channels 6 and 5. An example of the imagery for Hurricane Karl (2004) is below. Karl was estimated to have winds of 100 knots with a pressure of 955 millibars at the time of these images :

Channel Latest Data

100 hPa

200 hPa

350 hPa

550 hPa
AMSU-B
89 GHz

Notice that a warm anomaly can be seen in all 4 channels extending from channel 8 down into channel 5. However in channel 5 the effects of precipitation in the form of cooler brightness temperatures (green shading) can be seen to the east of the warm anomaly (orange shading). This cooling corresponds well with convective banding on the east side of Karl as seen in the 89Ghz AMSU-B imagery in the last image at the bottom. The 89 Ghz images are included on the web page to help identify convective structure and aid the user in evaluation of regions where heavy convection may be decreasing the warm anomaly signal in the various AMSU-A channels.

The CIMSS AMSU Algorithm:

As stated earlier there is strong relationship between the brightness temperature anomalies as measured by the AMSU-A instrument and Tropical Cyclone (TC) intensity. This relationship can be seen in the following graph which relates channel 8 brightness temperature anomalies to observed TC minimum sea level pressures.

The CIMSS AMSU algorithm uses this relationship to estimate TC MSLP. In general during the early stages of TC development the associated warm core is located near channel 7 and that channel is used to produce an estimate. As the TC intensifies the warm core moves higher in the atmosphere closer to the mean location of channel 8. Experience indicates that once the TC reaches hurricane intensity channel 8 tends to be the better indicator of storm strength and the algorithm uses that channel. While warm anomalies often show up in the lower channels these channels tend to suffer from precipitation effects that reduces their effectiveness. Current research is addressing this issue (see below) and it is possible they can be used in the future.

In the above image it is apparent that there is increased scatter in the brightness temperature relationship as intensity increases. This scatter has several sources including sub-sampling by the AMSU instrument, precipitation contamination in the channel being used to produce the intensity estimate and positioning of the storm center in-between AMSU Fields of View. Each of these complications is addressed below:

Sub-Sampling
The AMSU instrument employs a cross-track scanning strategy. As such the resolution of each of the 30 Fields Of View (FOV), or scenes, used by the instrument decreases with increasing scan angle. A horizontal representation looks like the scan swath below:

Much of the warming associated with a TC is concentrated in the eye region. The eye of a TC may be much less than the 150 km resolution of the near limb region of the scan swath. For example the eye of Hurricane Charley in 2004 was only about 8 km across. In these cases the warm anomaly signal will be signficantly sub-sampled. To correct for this effect the cases used to develop the CIMSS AMSU algorithm were divided into 2 stratifications, well resolved and sub-sampled. In order to determine whether or not a case falls into either category information regarding the eye size was needed. All cases for the 1998-2003 data sample were evaluated for eye size using available microwave, recconaisance obervations and IR imagery. These eye sizes were then compared to the FOV used to produce the intensity estimate for each AMSU pass. Cases in which the storm was well- resolved were then used to develop regression coefficients relating the brightness temperature anomalies to observed MSLP. These coefficients were then used to estimate the MSLP of the sub-sampled cases. In nearly all the cases the estimate was substantially weak, in extreme cases by as much as 30 mb. These extreme cases were the result of a combination of very small eye size and poor scan geometry (FOV near the edge of the scan swath). A bias relationship was then developed for the sub-sampled cases. The eye size / FOV resolution comparison explains about 50% of the observed error. When a storm is determined to be poorly sampled this bias is subtracted from the initial estimate. The method rely's on accurate determination of the eye size. The operational algorithm gets the eye size from two sources. If IR imagery is available and a clear well-defined eye is present the eye size is pulled from an algorithm which relates IR imagery to wind structure (including the radius of maximum winds or RMW). Otherwise the RMW from the ATCF messages distributed by the operational TC forecast centers is used. There are times when the RMW does not accurately represent the storms core size. Storms undergoing Eyewall Replacement Cycles (ERC) often have more than one wind maxima. It is not known how much and at what point the outer convective ring becomes the dominant source of subsidence and when the inner ring no longer contributes. High level recconnaisance above 500 mb would be helpful in understanding this process better.

Precipitation Contamination
Channels 7 and 8 are often high enough in the atmosphere to avoid the effects of precipitation. However, occasionally the convective towers associated with the TC may reach high enough to decrease the magnitude of the warm core signal. This is especially problematic when the TC is weak and the associated warm core is weak. In extreme cases the cooling effect of precipitation may completely mask the warm core signal and the result is a weak intensity estimate or no estimate at all. The example below shows a case for Tropical Depression 01W on January 11, 2002 where the effects of precipitation extend well into channel 8. The cooling is located very near the storm center making an intensity estimate difficult.
Robert Wacker who completed his PhD at UW in 2005 developed a correction for this effect. Details on the implementation of the correction can be seen in this document (Word). AMSU algorithm 2007

An example of the impact of this correction can be seen below in Figures 2 and 3 from the linked document
Bracketing
Another source of error for the algorithm is associated with storm position relative to the FOV used for the estimate. The initial estimate of storm position is obtained from operational warnings and storm center fixes from the TC Forecast Centers. From the initial fix a search is performed to check adjacent FOV's for the warmest pixel. However, given the instrument resolution it is possible that the storm may fall in between FOV positions. If the storm center falls nearly equidistant from adjacent FOV positions then the FOV will only sample the edge of the warm core and a weak estimate will result. This problem is aggravated for cases when the eye is small. At the other extreme if the storm center is centered on the FOV position and the eye is large then the estimate may be a 5-10 mb too deep since there is very little or no convection within the FOV to decrease the signal. An objection measure of the amount of bracketing can be obtained by convolving the AMSU-B 89 Ghz Tb to the AMSU-FOV used for the intensity estimate. Cold AMSU-B Tb's indicate that the AMSU-A FOV is located at least partially within the TC eyewall while warm AMSU-B Tb's indicate the liklihood that the AMSU-A FOV is located within the eye. Details on this bracketing factor can be found in the most recent algorithm upgrade document (Word) here AMSU algorithm 2007
Example of a fully bracketed estimate. Storm center falls between AMSU-A FOV positions.

Extratropical Transition:
Under construction
Algorithm Performance:

The following results are from the 1998-2006 seasons. Validation consisted of recconnaisance observations within 3 hours of the AMSU estimate. Dvorak statistics were produced using an average of available Dvorak estimates from TAFB and SAB. Again details on the latest version of the algorihtm are here (Word doc) AMSU algorithm 2007


Atlantic/East Pacfic (N=727)

MSLP

CIMSS AMSU         Dvorak
Bias       -0.1 mb        -2.2 mb
AAE      5.3 mb         6.5 mb
RMSE   7.3 mb          9.1 mb


Max Sustained Winds

CIMSS AMSU          Dvorak
Bias      -0.1 kts         -2.1 kts
AAE      7.8 kts         7.2 kts
RMSE   9.9 kts          9.5 kts

Return to the UW-CIMSS AMSU Homepage