NOTES AND CORRESPONDENCE A J. S

advertisement
1800
JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY
VOLUME 24
NOTES AND CORRESPONDENCE
A Multispectral Technique for Detecting Low-Level Cloudiness near Sunrise
ANTHONY J. SCHREINER, STEVEN A. ACKERMAN,
AND
BRYAN A. BAUM
Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin—Madison, Madison, Wisconsin
ANDREW K. HEIDINGER
National Oceanic and Atmospheric Administration/National Environmental Satellite, Data, and Information Service, Center for
Satellite Applications and Research, Madison, Wisconsin
(Manuscript received 19 July 2006, in final form 6 February 2007)
ABSTRACT
A technique using the Geostationary Operational Environmental Satellite (GOES) sounder radiance
data has been developed to improve detection of low clouds and fog just after sunrise. The technique is
based on a simple difference method using the shortwave (3.7 ␮m) and longwave (11.0 ␮m) window bands
in the infrared range of the spectrum. The time period just after sunrise is noted for the difficulty in being
able to correctly identify low clouds and fog over land. For the GOES sounder cloud product this difficulty
is a result of the visible reflectance of the low clouds falling below the “cloud” threshold over land. By
requiring the difference between the 3.7- and the 11.0-␮m bands to be greater than 5.0 K, successful
discrimination of low clouds and fog is found 85% of the time for 21 cases from 14 September 2005 to 6
March 2006 over the GOES-12 sounder domain. For these 21 clear and cloudy cases the solar zenith angle
ranged from 87° to 77°; however, the range of solar zenith angles for cloudy cases was from 85° to 77°.
The success rate further improved to 95% (20 out of 21 cases) by including a difference threshold of 5.0
K between the 3.7- and 4.0-␮m bands, requiring that the 11.0-␮m band be greater than 260 K, and limiting
the test to fields of view where the surface elevation is below 999 m. These final three limitations were
needed to more successfully deal with cases involving snow cover and dead vegetation. To ensure that only
the time period immediately after sunrise is included the solar zenith angle threshold for application of these
tests is between 89° and 70°.
1. Introduction
A noted difficulty in cloud detection using remotely
sensed radiances occurs when attempting to detect low
clouds and fog just after sunrise, during the transition
from nighttime cloud-detection techniques to daytime
methods. In practice, the thermal difference between
the longwave window (11.0 ␮m) brightness temperature and the surface skin temperature is frequently
within the noise limitations of the observed brightness
temperatures when low clouds and fog are present. As
a result cloud detection errors near sunrise (this region
Corresponding author address: Anthony J. Schreiner, CIMSS,
University of Wisconsin—Madison, 1225 W. Dayton St., Madison,
WI 53706.
E-mail: tonys@ssec.wisc.edu
DOI: 10.1175/JTECH2092.1
© 2007 American Meteorological Society
JTECH2092
is also described as the day–night terminator) occur in
orbiting and geostationary remote sensing platforms. In
the case of geostationary satellites, as will be discussed
here and in particular the Geostationary Operational
Environmental Satellite (GOES) sounder (Menzel and
Purdom 1994; Menzel et al. 1998), the “disappearing
clouds syndrome” can at times be seen to “move” from
east to west over the course of 3 or 4 h.
For the GOES sounder cloud mask algorithm
(Schreiner et al. 2001), as the sun rises over a region,
the daytime series of tests become the primary means
for identifying clouds. Reasons for the transition from
“nighttime” to “daytime” are twofold. First, the techniques for detecting low clouds at night strongly depend
on the differences between the infrared (IR) 11.0-␮m
and shortwave window (3.7 ␮m) bands (Eyre 1984;
d’Entremont 1986; Saunders and Kriebel 1988;
OCTOBER 2007
NOTES AND CORRESPONDENCE
1801
FIG. 1. Composite of the GOES sounder cloud mask showing 1200–1400 UTC images for 6 Jun 2005.
Note the undetected and, then again, detected clouds occurring (area within the yellow oval) after local
sunrise. (lower-right-hand panel) The 1346 UTC sounder visible image.
Kleespies 1995; Lee et al. 1997; Ackerman et al. 1998).
At night for clear-sky scenes, the 3.7-␮m minus 11.0␮m brightness temperature difference over land ranges
from approximately ⫺4 to ⫹1 K, depending on surface
emissivity and the atmospheric water vapor distribution. However, the difference between these two bands
can be a strong indicator of low-level cloudiness and/or
fog. For the case of clouds having small effective radii
and high optical thicknesses, the 11.0-␮m value tends to
be greater than that of the 3.7-␮m value (Baum et al.
2003). As the sun rises above the horizon this difference
becomes positive because of the contribution of solar
reflection in the 3.7-␮m band.
A second reason for the “disappearance” of low
clouds just after sunrise is the failure of the visible reflectance tests for low-altitude clouds and fog. Visible
thresholds for cloud/no-cloud detection during the daytime are higher over land than over water, as well as a
function of terrain type. These terrain types are dependent on the time of year, elevation, type of vegetation,
snow, and roughness of the landscape. Just after sunrise
the visible reflectance of clouds, especially low clouds
and fog, is below the threshold for cloudiness over the
conterminous United States (CONUS), and thus are
incorrectly flagged as clear. Mid- and high-level clouds
are not as sensitive to this visible threshold, as they tend
to be brighter and can be detected by comparison of the
11.0-␮m brightness temperature to a skin temperature
or modified surface observed temperature. Also, as
with the current GOES imager, the visible band of the
sounder is not calibrated once the satellite achieves orbit. The degradation of the visible (imager) band has
been observed by Hillger et al. (2003) and Daniels et al.
(2001), since the launch of GOES-11 and GOES-12.
The net result of this cloud-detection shortcoming
can frequently be seen when a loop of derived imagery
is set in motion (e.g., Fig. 1). Prior to sunrise low clouds
are correctly depicted (1146 UTC), as noted by the area
highlighted within the oval in the southwestern portion
of the figure. For the 1246 UTC scan line start time
image, the first image after sunrise, nearly the entire
region of low cloudiness is not detected by the algorithm currently employed operationally. Then, once
again at 1346 UTC the cloud bank along the southern
portion of the CONUS is correctly depicted. In addition to incorrectly portraying cloudiness in the loops of
derived imagery these derived data may have a negative impact when used in the initialization step for numerical weather prediction models (Bayler et al. 2000).
By exploiting three of the IR bands of the GOES
sounder for a particular field of view (FOV), a technique for identifying these low clouds just after sunrise
1802
JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY
has been developed. The IR window bands for the
GOES sounder are the longwave window (11.0 ␮m),
shortwave window (3.7 ␮m), and a second shortwave
window (4.0 ␮m). In essence, the simple difference
(SIMDIF) technique looks at the difference between
the 3.7- and 11.0-␮m bands and the difference between
the 3.7- and 4.0-␮m bands. If the differences fall within
the predetermined thresholds and some additional criteria based on 11.0-␮m temperature, surface elevation,
and solar zenith angle (SZEN), the FOV is defined as
cloudy. The SIMDIF technique defines “just after sunrise” as the first time period of the GOES sounder data
at a particular location where the SZEN is 89° ⱖ SZEN
ⱖ 70°. The cutoff is set at 70° because at SZEN ⬍ 70°
certain difference thresholds begin to break down. This
will be demonstrated in the following section.
The purpose of this note is to detail the criteria
needed to satisfy the SIMDIF technique and to define
why the SIMDIF is successful. The background section
briefly describes the reasoning. Two case studies will be
examined in section 3 showing both the success and the
limitations of the SIMDIF technique. The summary
and future work will summarize the note and introduce
future goals.
2. Background
a. Theory
The low thermal contrast between clear skies and low
cloud and fog makes detection of these clouds at night
challenging (Ellrod 1995). Techniques that make use of
the shortwave infrared (i.e., 3.5–4-␮m region) and longwave infrared (i.e., 10–12 ␮m) observations exploit the
emissivity differences at these wavelength regions for
stratus and fog conditions. These clouds have a lower
emissivity in the shortwave infrared window than the
longwave infrared window, resulting in negative brightness temperature differences when using the following
relationship: BT3.7 ⫺ BT11. This negative difference is a
function of the cloud droplet size and the underlying
surface. Baum et al. (2003) provide some radiative
transfer calculations for both ice and water clouds at
night that provide insight into the behavior of this
brightness temperature difference as a function of particle size and optical thickness, but new calculations are
not provided herein. Over clear-sky vegetative surfaces
and water the differences are generally larger than ⫺0.5
K. Clear-sky desert scenes have brightness temperature
differences that range between approximately ⫺5 and
⫺1 K, making detection of optically thin low-level
clouds at night more difficult because of the emissivity
differences of soil. During the day the brightness temperature difference between BT3.7 and BT11 is large
VOLUME 24
because of the reflection of solar energy at 3.7 ␮m. This
brightness temperature difference technique is very
successful at detecting low-level water clouds during
the day. The approach is generally not applied over
deserts during daytime, as bright desert regions with
highly variable emissivities tend to be classified incorrectly as cloudy with this test. In general, the emissivity
differences of the same stratiform water cloud being
observed simultaneously by the 3.7- and 4.0-␮m bands
is small as is the emissivity differences with varying
surface types (Hunt 1973; Sutherland 1986; Ellrod
2006). Here we explore using the differences between
BT3.7 and BT4.0 for cloud detection, since over clouds
during the day, BT3.7 ⬎ BT4.0 because there is more
reflected solar energy at 3.7 ␮m.
The following approach has been developed to ascertain whether the observed FOV is either clear or
obscured by low clouds. The logic is applied to a given
FOV when the SZEN is within the following window:
89.0° ⬍ SZEN ⬍ 70.0°, but is limited to FOVs with a
surface elevation (EL) threshold less than 999 m.
The SZEN window described above was chosen because it roughly defines no more than one time period
following sunrise for a GOES sounder time period.
This is important because the defined threshold, listed
in Eq. (1) below, can be easily surpassed in clear-sky
situations for SZEN values less than 70°. Given these
conditions, an FOV is defined cloudy if the brightness
temperature differences all meet the following criteria:
BT3.7 ⫺ BT11 ⬎ 5.0 K,
BT3.7 ⫺ BT4.0 ⬎ 5.0 K,
BT11 ⬎ 260.0 K.
and
共1兲
共2兲
共3兲
The elevation check was added as a result of continued failure of this check in elevated terrain, especially
in the late fall and early spring when vegetation is quite
sparse over the mountainous regions. Thermal checks
(2) and (3) were included to minimize the effects of
snow, especially fresh snow cover at all elevations.
b. Data
Twenty-one different clear and cloudy cases were examined to determine whether the BT3.7 ⫺ BT11 brightness temperature difference [Note: This difference is
defined as the window difference (WNDF).] for a given
FOV at a clear versus cloudy scene just after sunrise
correctly satisfied the difference threshold (5 K) to
properly identify the FOV as clear or cloudy. For the 21
cases ranging from clear (i.e., 1) and low cloudy (i.e., 3)
conditions over water to clear (i.e., 10) and cloudy or
foggy conditions (i.e., 7) over land at various locations,
OCTOBER 2007
NOTES AND CORRESPONDENCE
FIG. 2. Comparison of the shortwave window (3.7 ␮m) minus
the longwave window (11.0 ␮m) in the ordinate vs solar zenith
angle in the abscissa. Included are 8 cloudy (solid dots) or foggy
sites (solid triangles), 3 clear snow-covered locations (open diamonds), and 10 additional clear sites (open squares). The thick
solid line at 5 K denotes the WNDF threshold cutoff between
clear (WNDF ⱕ 5K) and cloudy (WNDF ⬎ 5K).
the differences were tabulated. The numbers in parentheses indicate number of cases for each set of conditions. The cloudy cases were selected due to the inability of the old technique to correctly identify them as
such. “Clear” and “cloudy or foggy” were determined
in two ways. First, sites were verified by inspecting
hourly surface observations. Second, to ensure that locations were not affected by mid- and high clouds,
GOES visible and longwave window imagery were also
used. It should be noted that foggy locations were primarily determined by the Automated Surface Observation System (ASOS) hourly observation (National Oceanic and Atmospheric Administration/Department of
Defense/Federal Aviation Agency/U.S. Navy 1998). At
each case (or ASOS site) a 3 ⫻ 3 FOV box was defined
and a simple average of the nine FOVs for the three IR
bands (3.7, 4.0, and 11.0 ␮m) was determined.
A variety of locations were made ranging from the
Gulf of Mexico (“water”) and surface hourly sites along
the East Coast to as far west as Kansas, south to
Florida, and north to Minnesota. Spatially, these locations represent a sampling of the area the WNDF technique is expected to be the most effective, namely, the
eastern half of the United States. Finally, these cases
span from the beginning of September 2005–March
2006, a temporal range that goes from late Northern
Hemisphere summer to early spring.
Figure 2 shows the distribution of the differences for
the BT3.7 ⫺ BT11 bands at various solar zenith angles
between 70° and 89° for all 21 cases. It should be noted
that no cases were found near the solar zenith limits.
Given the predefined thresholds, it is very possible that
cloudy cases very near the solar zenith angle of 89° will
still be missed. The thick solid line at 5 K represents the
cutoff threshold for the WNDF. The range of local zenith angles for the 21 locations vary from 32° to 57° for
the clear comparisons and 37° to 57° for the cloudy
cases. Nine of the cases are plotted above the WNDF
1803
threshold value. Seven of the nine are cloudy or foggy
cases (solid circles or solid triangles) and two were clear
snow (hollow diamonds). Eleven of the cases fell below
the WNDF threshold. All but one was either clear (hollow squares) or clear snow cases. In total, Fig. 2 shows
that 18 of the 21 cases were correctly identified using
the WNDF.
Three of the 21 cases depicted in Fig. 2 were clear sky
with snow on the surface. They ranged from fresh snow
(1 day old) to 4- and 7-day-old snow. Two of the three
initial failures were snow cases (1- and 4-day-old snow).
Both snow cases were correctly flagged with the addition of the additional thermal checks (BT3.7 ⫺ BT4.0 ⬎
5.0 K and BT11 ⬎ 260.0 K) and the elevation check. The
third failure in Fig. 2 was a low cloud case over Texas
during November in which the WNDF value was less
than 4.0 K. No modifications could be made that would
uniquely capture this situation and not negatively affect
the other 20 cases, primarily the clear cases. This particular case exposes a weakness in the technique.
Of the 18 successful cases 11 were clear sites and 7
were low cloud or fog examples. A typical example of a
successful clear-versus-cloudy comparison is shown in
Fig. 3. This figure describes the 3.7 ⫺ 11.0 ␮m value at
different SZEN values based on GOES-12 sounder
data for two sites (one clear and one cloudy) over Texas
on 14 November 2005. Note the differential rate of increase for the 3.7 ⫺ 11.0-␮m band difference just after
sunrise for the clear versus the cloudy case. (SZEN
decreasing from 92.0° to 80.0°, corresponding to the
nominal 1200 and 1300 UTC time periods.) For the
eight cloudy locations the average 3.7 ⫺ 11.0 ␮m value
is 8.04 K, where the maximum difference is 16.6 K and
the minimum difference is 3.8 K. For all (10) clear cases
(not including clear where the ground is covered with
FIG. 3. Comparison of clear (dashed) vs low cloud or fog (solid)
difference for the 3.7- and the 11-␮m bands aboard the GOES-12
sounder over TX on 14 Nov 2005. Along the abscissa is the solar
zenith angle (deg) and the ordinate is the temperature difference
(K). At each location a 3 ⫻ 3 FOV box was defined and a simple
average of the nine FOVs for the two IR bands (3.7 and 11.0 ␮m)
was determined for each time period.
1804
JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY
FIG. 4. Same as in Fig. 3, but the differences refer to the 3.7-␮m
band minus the 4.0-␮m band aboard the GOES-12 sounder. The
sites are located in TX on 14 Nov 2005. At each location a 3 ⫻ 3
FOV box was defined and a simple average of the nine FOVs for
the two IR bands (3.7 and 4.0 ␮m) was determined for each time
period.
snow, i.e., clear snow), the average difference is 2.1 K.
For these 10 clear cases the range of difference is 3.3–
0.9 K. When the three “clear snow” sites are added, the
average difference increases to 2.8 K (6.6-K maximum
difference and 3.5-K minimum difference for clear
snow sites only). Using this distribution, a 3.7 ⫺ 11.0
␮m value of 5 K was defined as a threshold to delineate
clear from cloudy scenes as a first step in this technique.
It should be noted that the 3.7- and 11.0-␮m difference
(WNDF) was independent of local zenith angle.
With a threshold of 5 K for the WNDF 3 of the 21
cases or locations failed to be correctly identified as
either clear or cloudy. Therefore, a second test was
included. Figure 4 details the comparison between the
3.7–4.0-␮m band and SZEN for the same case as in Fig.
3. Similar to the WNDF, the 3.7–4.0-␮m bands difference is greater for a cloudy location (5.2 K for this site)
than for a clear example (2.4 K). For each of the cloudy
cases (except the one cloudy failure noted above, 2.4
K), both the WNDF and the 3.7- and 4.0-␮m bands
(SWNDF) difference were greater than 5.0 K for the
first GOES sounder time period after sunrise. The reverse was true for the clear cases with the exception for
the locations that included fresh snow (4 days old or
less). For these locations the WNDF was greater than
5.0 K, but the SWNDF was less than 5.0 K. By adding
this additional check, the remaining two clear snow
cases were correctly flagged.
It should be noted that, typically, for the 21 cases no
more than two sites were tested in the “zone of interest” (89.0° ⬍ SZEN ⬍ 70.0°). A more rigorous approach would be to investigate the thresholds at numerous locations or sites for varying SZENs and note the
threshold differences. Although not performed, it is assumed that as one more closely approaches SZEN ⫽
89° the threshold difference criteria for Eqs. (1) and (2)
will break down. But for all 21 cases, with the exception
VOLUME 24
of the one snow case noted above, the defined thresholds in this study for both the WNDF and BT3.7 ⫺ BT4.0
were satisfactory.
After the SIMDIF was included in the cloud detection subroutine for the GOES sounder cloud (Schreiner
et al. 2001) processing at the University of Wisconsin—
Madison Cooperative Institute for Meteorological Satellite Studies (CIMSS), it was found that an elevation/
terrain-type check was also required, as well as an additional thermal check [11.0 ␮m ⬎ 260.0 K, Eq. (3)
above]. The additional thermal check helped in detecting “old” snow (⬎1 day). The elevation/terrain-type
check assisted in discriminating between low cloud/fog
and clear-sky and arid scenes during the late fall and
early spring in mountainous regions. As was noted earlier when these tests were applied to the three failing
cases of the original 21 cases, 2 of the 3 failures were
correctly identified.
3. Case studies
a. 9 March 2006 (cloud mask only)
Figures 5a–d detail an example comparing the GOES
sounder cloud mask with (labeled “NEWMSK”) and
without (labeled “OLDMSK”) the SIMDIF for
4-hourly time periods (1246–1546 UTC) on 9 March
2006. This test was applied to GOES-12 sounder data
covering the eastern portion of the United States. Of
interest is the cloud cover at two locations (Fargo,
North Dakota, and Bemidji, Minnesota) corresponding
to the gray and white dots, respectively, within the
white and stippled ovals.
At 1246 UTC (Fig. 5a) both the NEWMSK and the
OLDMSK are correctly indicating cloud for both surface locations, as is indicated in Table 1. The SZEN at
this time is greater than 89°; therefore, the SIMDIF
does not affect the region within the stippled oval. The
1346 UTC time period shows a difference between the
NEWMSK and OLDMSK, with the NEWMSK detecting cloud at both Fargo and Bemidji, while OLDMSK
indicates cloud only at Bemidji, which upon examination of the longwave window image (top right) is determined to be high-level cloud. The visible image for
this time period denotes cloudy conditions for both locations and is supported by the surface reports for the
same time period. It appears the NEWMSK is better
able to identify clouds in northern Minnesota and
North Dakota than the OLDMSK.
At 1446 UTC once again the OLDMSK is apparently
only detecting the upper-level cloud (see the longwave
window image), while the NEWMSK is able to identify
both the high and low cloud deck in the region of Bemidji and Fargo. Surface observations for this time re-
OCTOBER 2007
NOTES AND CORRESPONDENCE
1805
FIG. 5. (a) A comparison of the cloud mask (lower left) with the new simple difference technique and (lower right) without for 1246
UTC 9 Mar 2006. (top left) The GOES-12 sounder visible band (0.65 ␮m) and (top right) is the longwave window band (11.0 ␮m). The
dark gray dot (located in the white and stippled ovals) is the location of Fargo, ND, and the white dot (also in the white and stippled
ovals) is the location of Bemidji, MN. The solid lines represent SZEN ⫽ 89° (white) and SZEN ⫽ 70° (orange). (b) Same as in (a), but
at 1346 UTC. (c) Same as in (a), but at 1446 UTC. (d) Same as in (a), but at 1546 UTC.
veal that there may be an extensive low cloud deck
passing over both surface sites in addition to high cloud
over Bemidji. At 1546 UTC (Fig. 5d) the NEWMSK
and OLDMSK are in agreement again. By 1546 UTC
the SZEN is less than 70° within the oval of interest and
the SIMDIF is not applicable. The time surface reports
indicate that Bemidji is still under low clouds and this
situation is captured by both cloud mask versions.
However, the cloud cover at Fargo has cleared off, although haze is reported. The cloud masks show that
Fargo is right on the edge of what is apparently a low
cloud deck. Closer investigation of the visible image
reveals that the highly reflective region over Fargo is in
fact snow. This feature is confirmed in subsequent time
periods (visible images not shown). For this particular
case the SIMDIF more accurately depicts the cloud
mask for the northern Minnesota and North Dakota
region using the GOES sounder radiance information.
Tables 2a,b refer to the average 3 ⫻ 3 brightness
temperatures based on the observed GOES-12 sounder
radiance data at Fargo and Bemidji, respectively. This
table details the differences (WNDF and BT3.7 ⫺ BT4.0),
the SZEN and BT11 used in the SIMDIF. At both sites
for both time periods (1346 and 1446 UTC) all four
tests, including the two difference thresholds, are seen
to indicate clouds, in this case low clouds.
One of the weaknesses of the technique is also observed in this case study. By 1446 UTC additional
“cloud” is observed over the extreme eastern Colorado/
western Nebraska region. In higher elevations solar reflectance in the 3.7- and 4.0-␮m bands is locally high in
clear scenes, even when snow is not present. An increase in solar reflectance is a result of barren vegetation at this elevation during the late winter early spring
time frame in this region. Dead vegetation, especially at
high elevations, tends to demonstrate similar thresholds
1806
JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY
VOLUME 24
TABLE 1. ASOS hourly observations for Fargo, ND (KFAR), and Bemidji, MN (KBJI), from 0955 to 1755 UTC (approximately 4:00
to 12:00 P.M. local time) 9 Mar 2006 (2006068). The remaining column labels are defined as follows. The times of the observation
are in UTC. WX, with respect to ASOS, can be defined as a visibility obstruction and is limited to clear (blank), fog (F), haze (H), and
freezing fog (IF). WX, again with respect to ASOS, can refer to certain weather types such as rain (R) and snow (S). Temperature (T)
and dewpoint (TD) in K. Visibility (VIS) in km. C refers to cloud cover, where 0 is clear, 1 is scattered, 2 is broken, and 3 is overcast.
CIGH and ZCL1 are the height of the ceiling and lowest cloud base, respectively, in km.
A.M.
ID
State
Day
UTC
WX
T
TD
VIS
C
CIGH
KFAR
KFAR
KFAR
KFAR
KFAR
KFAR
KFAR
KFAR
KFAR
ND
ND
ND
ND
ND
ND
ND
ND
ND
2006068
2006068
2006068
2006068
2006068
2006068
2006068
2006068
2006068
0953:00
1053:00
1153:00
1253:00
1353:00
1453:00
1553:00
1653:00
1753:00
F
F
F
F
F
F
H
271.5
272.0
271.5
270.9
271.5
272.0
273.7
275.9
277.0
270.4
270.4
269.8
269.8
270.4
270.4
270.9
270.4
270.4
4.8
6.4
8.0
8.0
4.8
6.4
9.7
12.9
14.5
3
3
3
3
3
3
0
0
0
0.64
0.70
0.70
0.70
0.70
0.70
0.00
0.00
0.00
ID
KBJI
KBJI
KBJI
KBJI
KBJI
KBJI
KBJI
KBJI
KBJI
State
MN
MN
MN
MN
MN
MN
MN
MN
MN
Day
2006068
2006068
2006068
2006068
2006068
2006068
2006068
2006068
2006068
UTC
0955:00
1055:00
1155:00
1255:00
1355:00
1455:00
1555:00
1655:00
1755:00
WX
T
272.0
272.0
272.0
272.0
272.0
273.2
273.2
273.2
273.2
TD
270.4
270.4
270.4
270.4
270.4
269.3
269.3
269.3
270.4
VIS
16.1
16.1
16.1
16.1
16.1
16.1
16.1
16.1
16.1
C
3
3
3
3
3
3
3
3
3
CIGH
0.58
0.52
0.52
0.52
0.52
0.46
0.40
0.40
0.40
for the WNDF and BT3.7–BT4.0 as low clouds at lower
elevations (Ackerman et al. 1998). The SIMDIF will
fail in these instances because the 5-K threshold is surpassed. The region of interest is below the elevation
threshold (999 m) used in the logic, and therefore cloud
is being incorrectly identified.
b. 9 May 2006 (cloud mask and cloud-top
pressure)
A second case over Texas and the southeastern
United States also demonstrates the effectiveness of the
WNDF technique. In this case the SIMDIF was incorporated in the GOES sounder Cloud Height Program
(Schreiner et al. 2001). This cloud height processing
C
ZCL1
C
ZCL1
algorithm is a two-step process. The first step defines a
cloud mask (where the SIMDIF is applied). The second
step then will determine a cloud-top pressure only if a
FOV has been determined to be cloudy. The effect of
the SIMDIF is shown for the nominal time periods
from 0946 to 1247 UTC 9 May 2006 using the GOES-12
sounder data in Figs. 6a–d, which are essentially the
cloud mask augmented with cloud-top pressures.
Figure 6a (0946 UTC) shows the cloud coverage approximately 1 h prior to sunrise along the east coast of
the United States with the exception of the extreme
northern portion of the east coast. The synoptic conditions include a low pressure system located over the
Central Plains generating some convective activity over
eastern Kansas, Missouri, and eastern portions of Iowa.
TABLE 2a. Hourly brightness temperatures based on GOES-12 sounder radiances for band 8 (11.0 ␮m, BT11), band 17 (4.0 ␮m,
BT4.0), and band 18 (3.7 ␮m, BT3.7) for six consecutive hourly time periods (time in UTC) on 9 Mar 2006 for Fargo, ND. The visible
reflectance (VIS in %), solar zenith angle (SZEN), and difference between the shortwave window and the longwave window (WNDF)
and the shortwave bands (BT3.7–BT4.0 in K) are shown in the rightmost four columns.
UTC
BT11
BT4.0
BT3.7
VIS
SZEN
WNDF
BT3.7 – BT4.0
1146:00
1246:00
1346:00
1446:00
1546:00
1646:00
265.88
265.82
266.58
269.30
270.77
273.84
262.41
261.56
270.17
278.46
272.28
277.09
261.67
261.03
283.15
292.92
276.85
281.78
0
0
6
15
24
25
102.25
91.97
81.89
72.40
64.01
57.26
⫺4.21
⫺4.79
16.57
23.62
6.08
7.94
⫺0.74
⫺0.53
12.98
14.46
4.57
4.69
OCTOBER 2007
1807
NOTES AND CORRESPONDENCE
TABLE 2b. Same as in Table 2a, but the location coincides with Bemidji, MN.
UTC
BT11
BT4.0
BT3.7
Vis
SZEN
WNDF
BT3.7 – BT4.0
1146:00
1246:00
1346:00
1446:00
1546:00
1646:00
263.61
253.08
262.48
263.91
263.79
263.99
260.42
254.55
263.05
279.22
285.64
292.26
260.39
255.05
269.37
295.35
304.32
313.99
0
1
6
20
29
38
101.01
90.84
80.92
71.64
63.51
57.12
⫺3.22
1.97
6.89
31.44
40.53
50.00
⫺0.03
0.50
6.32
16.13
18.68
21.73
There are also convective storms over the Florida Peninsula. The southeastern portion of the country is overcast due to low-level clouds and/or fog throughout the
time period 1000–1500 UTC. These conditions are indicated in Table 3.
At 1046 UTC (Fig. 6b), the sun is just above the
horizon for the eastern coast of the United States.
When this time period is compared to the previous image (Fig. 6a), there appears to be a clear region encompassing both North and South Carolina in the
FIG. 6. (a) A comparison of the derived CTP image based on GOES-12 sounder radiance data (bottom left) with and (bottom right)
without the new simple difference technique at 0946 UTC 9 May 2006. The color levels refer to various levels of CTP (hPa), and the
gray shades in the CTP images indicate cloud-free regions. (top left) The GOES-12 sounder visible band (0.65 ␮m) and (top right) the
longwave window band (11.0 ␮m). The four square dots (white) refer to the location of the surface observation sites (see Table 3) used
to verify the existence of low cloud. The surface observations sites are Jacksonville, NC; Albany, GA; Houston, TX; and Corsicana, TX.
The solid lines represent SZEN ⫽ 89° (white) and SZEN ⫽ 70° (orange) (b) Same as in (a), but at 1046 UTC. The white dot indicates
the location of Jacksonville. (c) Same as in (a), but at 1146 UTC. The white dot indicates the location of Albany. (d) Same as in (a),
but at 1247 UTC. The white dot indicates the location of Houston.
1808
JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY
VOLUME 24
TABLE 3. Surface hourly reports for the four locations on 9 May 2006 (day 2006129) noted in Figs. 6a–d. KNCA is the ID
for Jacksonville, NC; KABY is Albany, GA; and KHOU is Houston, TX. The remaining column labels are the same as defined in Table 1.
ID
State
Day
UTC
WX
T
TD
VIS
C
CIGH
C
ZCL1
KNCA
KNCA
KNCA
KNCA
KNCA
KNCA
NC
NC
NC
NC
NC
NC
2006129
2006129
2006129
2006129
2006129
2006129
0956:00
1056:00
1156:00
1256:00
1356:00
1456:00
F
F
F
285.9
285.9
286.5
287.0
288.7
289.3
284.3
284.3
284.3
283.7
283.7
283.7
8.0
9.7
9.7
11.3
11.3
11.3
2
2
2
2
2
2
0.30
0.30
0.30
0.30
0.61
0.61
3
2
2
1.22
0.61
0.91
2
3
3.05
0.91
ID
KABY
KABY
KABY
KABY
KABY
KABY
KABY
State
GA
GA
GA
GA
GA
GA
GA
Day
2006129
2006129
2006129
2006129
2006129
2006129
2006129
UTC
0953:00
1053:00
1153:00
1225:00
1253:00
1353:00
1453:00
WX
F
F
F
F
F
F
H
T
291.5
291.5
292.0
292.0
292.0
292.6
293.2
TD
290.9
290.4
290.4
290.9
290.4
290.4
290.4
VIS
4.8
4.0
4.0
4.8
4.8
4.8
6.4
C
3
3
3
3
3
3
3
CIGH
0.12
0.12
0.12
0.12
0.15
0.15
0.21
C
ZCL1
ID
KHOU
KHOU
KHOU
KHOU
KHOU
KHOU
State
TX
TX
TX
TX
TX
TX
Day
2006129
2006129
2006129
2006129
2006129
2006129
UTC
0953:00
1053:00
1153:00
1253:00
1353:00
1453:00
WX
F
T
296.5
297.0
297.0
298.2
299.3
300.4
TD
295.4
295.9
295.9
296.5
296.5
296.5
VIS
9.7
11.3
8.0
8.0
8.0
9.7
C
2
2
2
2
2
2
CIGH
0.46
0.61
0.34
0.43
0.49
0.70
C
1
1
2
2
1
1
ZCL1
0.24
0.34
0.70
0.61
0.30
0.46
F
F
H
H
“OLDCTP” image in the lower-left-hand corner. This
clear area is also observed in the “NEWCTP” (lowerright-hand corner), but the NEWCTP is not as void of
clouds as the OLDCTP. Both the visible imagery for
this time period (upper-right-hand corner) and the surface observation for Jacksonville, North Carolina (the
white dot in Fig. 6b), at 1100 UTC strongly contradict
the OLDCTP and also indicate that the NEWCTP is
not cloudy enough either.
The primary reasons for the lack of cloudiness in the
NEWCTP version of the cloud-top pressure (CTP) are
twofold. First, the difference between 3.7- and 11.0-␮m
bands falls below the criteria threshold (i.e., 5 K). A
second reason for the failure of the technique to correctly detect cloud is due to the terrain elevation limitation for the SIMDIF. Table 4a (similar to Table 2, but
that the locations are Jacksonville; Albany, Georgia;
Houston, Texas; and Corsicana, Texas; for 9 May 2006)
does show the SIMDIF correctly detecting fog at Jacksonville at 1046 UTC. By 1146 UTC (Fig. 6c) the sun is
far enough above the horizon that in the visible band,
clouds and clear regions are distinctly visible as far west
as Texas and Wisconsin. Despite these conditions, low
clouds over Georgia and Alabama are not properly detected in the OLDCTP. Based on comparisons to the
visible imagery for this time period and comparisons to
surface observations of clouds at Albany (see Table 3
for details), the NEWCTP correctly identifies low
clouds over this region. And, at 1247 UTC the trend of
the OLDCTP not detecting low clouds near the terminator continues to move westward (Fig. 6d). For this
time period all of eastern Texas is devoid of clouds for
the OLDCTP, while the NEWCTP more correctly defines the cloudiness over this region based on comparisons with the visible image for the same time period
and the surface observation at Houston (Table 3). In
TABLE 4a. Same as Table 2a, but for 9 May 2006. The surface location coincides with Jacksonville, NC.
UTC
BT11
BT4.0
BT3.7
VIS
SZEN
WNDF
BT3.7 – BT4.0
0946:00
1046:00
1146:00
1246:00
1346:00
1446:00
276.14
279.16
278.82
277.10
278.74
279.77
274.42
279.54
284.20
286.89
294.97
296.99
274.62
285.16
293.28
298.49
309.76
311.52
0
5
12
24
33
39
93.14
81.43
69.29
56.80
44.67
32.80
⫺1.52
6.00
14.46
21.39
31.02
31.75
0.20
5.62
9.08
11.60
14.79
14.53
OCTOBER 2007
1809
NOTES AND CORRESPONDENCE
TABLE 4b. Same as in Table 4a, but for Albany, GA.
UTC
BT11
BT4.0
BT3.7
Vis
SZEN
WNDF
BT3.7 – BT4.0
0946:00
1046:00
1146:00
1246:00
1346:00
1446:00
286.52
285.82
286.26
287.34
287.49
286.01
283.44
283.53
288.94
291.99
294.31
296.20
283.90
284.58
294.76
298.87
302.60
306.38
0
2
12
22
32
47
98.83
86.93
74.51
61.66
48.83
35.67
⫺2.62
⫺1.24
8.50
11.53
15.11
20.37
0.46
1.05
5.82
6.88
8.29
10.18
this case Table 4b does point out a weakness of the
SIMDIF. Note the 1046 UTC GOES-12 sounder observations and differences. At Albany with the SZEN ⫽
86.93°, the SIMDIF does not accurately detect the observed fog. This problem is rectified in the following
time period.
As in Table 4b, Table 4c once again points out a
weakness. For the first time period immediately after
sunrise (SZEN ⬎ 83°) at the Houston location the
SIMDIF fails to detect fog. Although for the following
time period (75° ⬎ SZEN ⬎ 70°) fog is correctly identified. Despite these very specific shortcomings, by incorporating the SIMDIF a more complete definition of
this low cloud is delineated (NEWCTP). Figure 6f is
included to show that 1) the two cloud product versions
once again converge on the same answer once the
SZEN is less than 70° and 2) the cloudiness for the
southeastern portion of the United States is continuous
for the six time periods included in this particular
case.
4. Summary and future work
A technique has been developed and tested to detect
low-level clouds and fog just after sunrise over the eastern portion of the United States, based on analysis of 21
cases from mid-September 2005 to early March 2006
using GOES sounder radiance data. Frequently these
clouds are not properly flagged as the nighttime cloud
mask is replaced with a daytime version of the mask at
large solar zenith angles. Low clouds and fog just after
sunrise are missed as they are not as reflective in the
visible band of the GOES sounder as deep convective
clouds.
The SIMDIF uses the brightness temperature differences between the longwave window (11.0 ␮m), the
shortwave window (3.7 ␮m), and also employs another
shortwave window (4.0 ␮m) for solar zenith angles between 89.0° ⬍ SZEN ⬍ 70.0°, and surface elevations
less than 999 m. The strengths in this technique lie in its
ability to identify low-level water clouds or fog over
low, flat terrain over the eastern portion of the conterminous United States (CONUS). Testing of the SIMDIF was not performed over the western sections of the
CONUS because of the terrain elevation limitations.
This technique has been shown to be effective during
the late fall, winter, and early spring seasons. A weakness of this method is mistakenly identifying low cloud
when there are clear skies over fresh snow and dead
vegetation, especially at high elevations.
In summary, for both the cloud mask and the CTP
case study the SIMDIF was able to more correctly identify low-level cloudiness and fog that up to now was not
identified. The SIMDIF is currently being applied to
the GOES sounder cloud product algorithm from
CIMSS. Routine hourly process incorporating this technique has shown a significant improvement in the identification of low cloud and fog. The goal is to apply this
technique to the GOES sounder temperature/moisture
retrieval algorithm. (Ma et al. 1999; Schmit et al. 2002).
(In fact, the latest iteration of this processing system
generates temperature/moisture products and cloud
products, CTP, and the effective cloud amount from
one processing system.)
TABLE 4c. Same as in Table 4a, but for Houston, TX.
UTC
BT11
BT4.0
BT3.7
Vis
SZEN
WNDF
BT3.7 – BT4.0
0946:00
1046:00
1146:00
1246:00
1346:00
1446:00
289.88
292.80
292.27
292.38
293.34
294.23
290.38
291.28
290.03
295.59
299.69
303.69
290.95
291.71
292.28
301.11
306.28
310.51
0
0
3
9
16
17
107.41
95.78
83.44
70.55
57.52
44.09
1.07
⫺1.09
0.01
8.73
12.94
16.28
0.57
0.43
2.25
5.52
6.59
6.82
1810
JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY
Although the SIMDIF was not tested for cases near
sunset (because none were found during the test period
of 14 September 2005 and 6 March 2006 over the eastern portion of the CONUS), there is no reason to believe the SIMDIF will not work for this time period as
well. By incorporating the SIMDIF into the operational
production of GOES sounder products, the effectiveness of this technique near sunset will also be determined. Some experimentation using this method in a
temporal mode has shown favorable results in addition
to being more robust in its application. Future work
along this line will continue.
Acknowledgments. The authors wish to acknowledge
Mr. Timothy J. Schmit, Mr. Richard A. Frey, Dr. Jun
Li, Dr. W. Paul Menzel, Ms. Leanne Avila, and Mr.
Thomas H. Achtor for their insightful comments and
helpful suggestions. In addition, the authors wish to
thank three anonymous reviewers who provided constructive criticism and helpful remarks. The Space Science Engineering Center supplied the data for this
study. Funding for this work was provided by NOAA
Grant NA06NES4400002.
REFERENCES
Ackerman, S. A., K. I. Strabala, W. P. Menzel, R. A. Frey, C. C.
Moeller, and L. E. Gumley, 1998: Discriminating clear-sky
from clouds with MODIS. J. Geophys. Res., 103, 32 141–
32 158.
Baum, B. A., R. A. Frey, G. G. Mace, M. K. Harkey, and P. Yang,
2003: Nighttime multilayered cloud detection using MODIS
and ARM data. J. Appl. Meteor., 42, 905–919.
Bayler, G. M., R. M. Aune, and W. H. Raymond, 2000: NWP
cloud initialization using GOES sounder data and improved
modeling of nonprecipitating clouds. Mon. Wea. Rev., 128,
3911–3920.
Daniels, J. M., T. J. Schmit, and D. W. Hillger, 2001: GOES-11
Science Test: GOES-11 imager and sounder radiance and
product validations. NOAA Tech. Rep. NESDIS 103, U.S.
Department of Commerce, Washington, DC, 49 pp.
d’Entremont, R. P., 1986: Low and midlevel cloud analysis using
nighttime multispectral imagery. J. Climate Appl. Meteor., 25,
1853–1869.
Ellrod, G. P., 1995: Advances in the detection and analysis of fog
at night using GOES multispectral infrared imagery. Wea.
Forecasting, 10, 606–619.
VOLUME 24
——, 2006: Evaluation of Moderate-Resolution Imaging Spectroradiometer (MODIS) shortwave infrared bands for optimum
nighttime fog detection. Preprints, 14th Conf. on Satellite Meteorology and Oceanography, Atlanta, GA, Amer. Meteor.
Soc., CD-ROM, P3.24.
Eyre, J. R., 1984: Detection of fog at night using Advanced Very
High Resolution Radiometer (AVHRR) imagery. Meteor.
Mag., 113, 266–271.
Hillger, D. W., T. J. Schmit, and J. M. Daniels, 2003: Imager and
sounder radiance and product validations for the GOES-12
science test. NOAA Tech. Rep. 115, U.S. Department of
Commerce, Washington, DC, 70 pp.
Hunt, G. E., 1973: Radiative properties of terrestrial clouds at
visible and infrared thermal window wavelengths. Quart. J.
Roy. Meteor. Soc., 99, 346–369.
Kleespies, T. J., 1995: The retrieval of marine stratiform cloud
properties from multiple observations in the 3.9-␮m window
under conditions of varying solar illumination. J. Appl. Meteor., 34, 1512–1524.
Lee, T. F., F. J. Turk, and K. Richardson, 1997: Stratus and fog
products using GOES-8–9 3.9-␮m data. Wea. Forecasting, 12,
664–677.
Ma, X. L., T. J. Schmit, and W. L. Smith, 1999: A nonlinear physical retrieval algorithm—Its application to the GOES-8/9
sounder. J. Appl. Meteor., 38, 501–513.
Menzel, W. P., and J. F. W. Purdom, 1994: Introducing GOES-I:
The first of a new generation of Geostationary Operational
Environmental Satellites. Bull. Amer. Meteor. Soc., 75, 757–
782.
——, F. C. Holt, T. J. Schmit, R. M. Aune, A. J. Schreiner, G. S.
Wade, and D. G. Gray, 1998: Application of GOES-8/9
soundings to weather forecasting and nowcasting. Bull.
Amer. Meteor. Soc., 79, 2059–2077.
National Oceanic and Atmospheric Administration/Department
of Defense/Federal Aviation Administration/U.S. Navy,
1998: ASOS user’s guide. NOAA/DOD/FAA/U.S. Navy, 72
pp.
Saunders, R. W., and K. T. Kriebel, 1988: An improved method
for detecting clear sky and cloudy radiances from AVHRR
data. Int. J. Remote Sens., 9, 123–150.
Schmit, T. J., W. F. Feltz, W. P. Menzel, J. A. Jung, A. P. Noel,
J. N. Heil, J. P. Nelson III, and G. S. Wade, 2002: Validation
and use of GOES sounder moisture information. Wea. Forecasting, 17, 139–154.
Schreiner, A. J., T. J. Schmit, and W. P. Menzel, 2001: Trends and
observations of clouds based on GOES sounder data. J. Geophys. Res., 106, 20 349–20 363.
Sutherland, R. A., 1986: Broadband and spectral emissivities (2–
18 ␮m) of some natural soils and vegetation. J. Atmos. Oceanic Technol., 3, 199–202.
Download