estimating tropical cyclone intensity from infrared …ritchie/refs/pineros_etal_2011.pdfestimating...
TRANSCRIPT
Estimating Tropical Cyclone Intensity from Infrared Image Data
MIGUEL F. PINEROS
College of Optical Sciences, The University of Arizona, Tucson, Arizona
ELIZABETH A. RITCHIE
Department of Atmospheric Sciences, The University of Arizona, Tucson, Arizona
J. SCOTT TYO
College of Optical Sciences, The University of Arizona, Tucson, Arizona
(Manuscript received 20 December 2010, in final form 28 February 2011)
ABSTRACT
This paper describes results from a near-real-time objective technique for estimating the intensity of
tropical cyclones from satellite infrared imagery in the North Atlantic Ocean basin. The technique quantifies
the level of organization or axisymmetry of the infrared cloud signature of a tropical cyclone as an indirect
measurement of its maximum wind speed. The final maximum wind speed calculated by the technique is an
independent estimate of tropical cyclone intensity. Seventy-eight tropical cyclones from the 2004–09 seasons
are used both to train and to test independently the intensity estimation technique. Two independent tests are
performed to test the ability of the technique to estimate tropical cyclone intensity accurately. The best results
from these tests have a root-mean-square intensity error of between 13 and 15 kt (where 1 kt ’ 0.5 m s21) for
the two test sets.
1. Introduction
Tropical cyclones (TC) form over the warm waters of
the tropical oceans where direct measurements of their
intensity (among other factors) are scarce (Gray 1979;
McBride 1995). In general, the primary sources of ob-
servations for these intense vortical weather systems are
from satelliteborne instruments (e.g., Ritchie et al. 2003;
Velden et al. 2006b). Although these instruments provide
many observations, including winds at various levels of
the atmosphere and temperature and humidity sound-
ings, among others, none of these include direct measure-
ments of the maximum wind speed or minimum sea level
pressure intensity of a tropical cyclone.
Because of the lack of direct in situ measurements of
tropical cyclone intensity, several techniques have been
developed to estimate the intensity based on indirect
factors. The most-used technique in operation to
estimate the intensity of tropical cyclones was developed
by V. Dvorak in the 1970s during the early years of
satellites (Dvorak 1975). In this technique, an analyst
classifies the cloud scene types in visible and infrared
satellite imagery and applies a set of rules to calculate
the intensity estimate. The original Dvorak technique is
subjective, is time intensive, and relies on the expertise
of the analyst, but it is still used as the primary intensity
forecasting tool in many tropical cyclone forecasting
centers around the world (e.g., Velden et al. 1998, 2006b;
Knaff et al. 2010). Velden et al. (1998) introduced the
difference of temperature between 1) the warmest pixel
temperature near the eye of the tropical cyclone and 2)
the coldest of the warmest pixel temperatures found on
concentric rings around the center. This modification is
known as the objective Dvorak technique, and, although
the intensity is objectively calculated, the location of the
eye of the tropical cyclone must still be determined by an
expert or by using external sources. Olander and Velden
(2007) developed the advanced Dvorak technique (ADT),
which introduces new procedures in making an intensity
estimate from satellite-based imagery rather than sim-
ulating the original Dvorak technique. One of the most
Corresponding author address: Miguel F. Pineros, PAS Bldg.,
Rm. 542, P.O. Box 210081, The University of Arizona, Tucson, AZ
85721-0081.
E-mail: [email protected]
690 W E A T H E R A N D F O R E C A S T I N G VOLUME 26
DOI: 10.1175/WAF-D-10-05062.1
� 2011 American Meteorological Society
important improvements of the ADT consists of the in-
troduction of regression equations to estimate the tropical
cyclone intensity. Kossin et al. (2007) recently described
a new satellite-based technique in which the radius of
maximum wind, the critical wind radii, and the two-
dimensional surface wind field are estimated from in-
frared (IR) imagery. This technique uses 12-h mean IR
imagery and best-track position data to estimate the
two-dimensional wind fields, which are compared with
aircraft wind profiles. In addition to visible and infrared
imagery, techniques for estimating the intensity of a tropi-
cal cyclone have also been developed on the basis of mea-
surements from the Advanced Microwave Sounding Unit
(AMSU; Spencer and Braswell 2001; Demuth et al.
2004). Some of these techniques have been combined to
enhance the TC intensity estimation (e.g., Velden et al.
2006a).
A different approach for characterizing the dynamics
of tropical cyclones was described in Pineros et al.
(2008). In that study, a method to quantify the axisym-
metry of a tropical cyclone from remote-sensing data
was introduced. Using 30-min-resolution geostationary
infrared imagery, the gradient of the brightness tem-
peratures was calculated, and the departure of that
gradient from a perfectly axisymmetric hurricane was
determined. A single value that quantified that depar-
ture from asymmetry was calculated, and a time series
was built and correlated with the best-track intensity
estimates from the National Hurricane Center (NHC).
The technique proved to be quite successful because the
organization of the clouds about the vortex, including
the cirrus shield, is directly tied to the kinematic orga-
nization of the vortex, including the organization of the
eyewall, rainbands, and tangential winds.
In this paper, an improvement of the tropical cyclone
intensity estimation technique described in Pineros et al.
(2008) is presented. In the next section, a brief review of
the method is presented and the improvement of the
technique is introduced. Results are shown in section 3.
Conclusions are discussed in section 4.
2. Method
The study incorporates the 2004–09 North Atlantic
Ocean hurricane seasons (Franklin et al. 2006; Beven
et al. 2008; Franklin and Brown 2008; Brennan et al.
2009; Brown et al. 2010). The data used in this study are
longwave (10.7 mm) IR satellite imagery from the Geo-
stationary Operational Environmental Satellite-12
(GOES-12). The data are available at ;4-km spatial
resolution, but we found previously that reducing the
resolution does not particularly influence the results but
does decrease the computational time considerably.
These images are cropped to cover an area from 48 to
348N and from 1058 to 288W over the northern Atlantic
basin and are resampled to a spatial resolution of 10 km
per pixel. Although the period of interest is from 2004 to
2009, tropical cyclones that had the majority of their
trajectory outside the footprint of the cropped satel-
lite image were excluded from the study. This included
Hurricane Vincent (2005) and Tropical Storms Beryl
(2006), Chantal (2007), Ten (2008), and Grace (2009).
As a result, a total of 15 147 half-hourly images from
2004 to 2009 were analyzed, covering the life cycle of 36
tropical storms and 42 hurricanes.
All samples that were located over land (center
passed over continents and large islands) were removed
from the database for consistency. Observations show
that tropical cyclones that make landfall rapidly decay at
a rate that is inconsistent for overocean tropical cy-
clones. Thus, a different set of parametric curves will be
required for landfalling TCs and is a topic of future
work. For now, all overland samples are simply removed
from the training set.
The original technique to determine the axisymmetry
of a cloud cluster using the deviation angle is illustrated in
Fig. 1 (Pineros et al. 2008). First the gradient of the IR
image at every pixel (in vector form) is calculated. Figure
1a shows the pseudo-IR image for an idealized hurricane.
The associated IR gradient field is shown in Fig. 1b. Next,
choosing a reference or center pixel, the deviation of the
IR gradient vector in a pixel from a radial extending from
the center pixel is determined and stored. This calculation
of the deviation angle is repeated for every pixel within
350-km radius of the center pixel. Next, the distribution
of the deviation angles is plotted (Fig. 1c) and the vari-
ance of that distribution (the deviation-angle variance or
DAV) is determined. The higher the variance of the an-
gle distribution is, the more disorganized is the cloud.
The lower the variance is, the closer to pure axisymme-
try is the cloud pattern. Figures 1d and 1e show the same
sequence as in Figs. 1a–c but for a single snapshot of
Hurricane Rita (2005). The calculation is repeated using
every pixel in turn as the reference center. The variance
values are then plotted back into the reference pixel lo-
cation to create a ‘‘map of DAVs’’ (Pineros et al. 2010)
that corresponds to the original IR image. In Pineros et al.
(2010), the map of variances was used to detect tropical
cyclogenesis. In this study, the map of variances is used to
estimate the tropical cyclone intensity by developing
a parameterized curve fitting that relates the DAV values
with a parameterized function.
The original DAV technique used a fixed 350-km ra-
dius for calculation. Here, we improve the system by
using eight different maps for each image in the training
set at radii varying from 150 to 500 km in steps of 50 km.
OCTOBER 2011 P I N E R O S E T A L . 691
Figure 2 shows an example of three images and their
400-km maps for Hurricane Dean (2007). For each
analysis radius, a time series of the minimum DAV in
the sequence of maps associated with a given tropical
cyclone over its life cycle was constructed (the DAV
signal). A single-pole low-pass filter (impulse response:
e2kt) with a cutoff frequency of 0.01p radians per sample
(filter time constant of 100 h) was applied to smooth the
signal and provide a better correlation with the best-
track intensity estimates, which are only available every
6 h but are interpolated to 30 min.
Next, the filtered DAV signals were mapped to the
NHC best-track intensity records to obtain a parametric
curve between the variance and maximum wind speed
for each of eight radii of analysis. Because the filtered
DAV signal is created using 30-min imagery but the
best-track intensity estimates are available only every
6 h, there is considerably more structure in the DAV
signal. The oscillations present in the filtered DAV
signal include diurnal and semidiurnal frequencies, as
well as some smaller-scale components, and they make it
difficult to relate the DAV to the best-track intensity.
Thus, one intensity value in the best track could have
several associated DAV values. To overcome this prob-
lem, the median of all DAV values associated with a
single best-track intensity estimate was used to create
the data scatterplot. A sigmoid was then fit to the rough
DAV–intensity scatterplot so that the final parametric
curve was described by a continuous equation:
f (s2) 5160
1 1 exp[a(s2 1 b)]1 25 (kt), (1)
where a and b are two parameters to fit from the input
data and s2 is the filtered DAV value. Note that the es-
timated wind speed f(s2) is bounded between 25 and
FIG. 1. (a) Brightness temperature image of an ideal vortex. (b) Gradient vectors of the central section in (a). (c) Distribution of
deviation angles for the ideal vortex in (a). (d) Hurricane Rita, 1415 UTC 21 Sep 2005 (intensity: 130 kt and 932 hPa). (e) Distribution of
deviation angles in (d), with variance 5 593 deg2.
692 W E A T H E R A N D F O R E C A S T I N G VOLUME 26
185 kt (where 1 kt ’ 0.5 m s21) in Eq. (1). Although we
considered several different polynomial functions, the
sigmoid was chosen for this application because it is
bounded at both ends of the intensity range, thus avoiding
the possibility of obtaining unrealistically high or low
intensity estimates. Last, the parametric curve for the ra-
dius with the minimum sum of squares of error over the
training set was chosen as the final intensity estimator
parametric curve. The specific optimum radius depends
on the training data used, and we typically see values be-
tween 300 and 400 km. Once the system is trained, DAV
maps are computed for the testing storms with this single
optimum value. This process was repeated with two dif-
ferent training sets, and the resulting two testing data-
sets, to measure its effectiveness as an intensity estimate.
3. Results
For the first test, the intensity estimation parametric
curve was calculated using a training set of 50 tropical
cyclones (70% of the available data) randomly chosen
from the period 2004–08. Only samples with intensities
above tropical storm strength were considered because of
uncertainty in the best-track database at lower intensities
(D. Brown, NHC, 2010, personal communication). The
remaining tropical cyclones were used as an independent
test set. This set included five tropical cyclones from
2004, seven from 2005, two from 2006, and six from
2008. Figure 3 shows the two-dimensional histogram of
the filtered DAV samples and best-track intensity es-
timates for this random set. The best-fit sigmoid curve
for this set is shown as a solid line and was obtained
FIG. 2. Map of the variances for Hurricane Dean (2007) with a 400-km radius: (a) 1215 UTC 14 Aug—35 kt, 1004 hPa, and a map
minimum value (MMV) of 1727 deg2; (b) 0015 UTC 16 Aug—60 kt, 991 hPa, and an MMV of 1548 deg2; (c) at 0015 UTC 18 Oct—125 kt,
944 hPa, and an MMV of 1250 deg2.
FIG. 3. Two-dimensional histogram of the 300-km filtered DAV
samples and best-track intensity estimates using 20 deg2 3 5-kt bins
for 70% of the tropical cyclones randomly chosen from the period
2004–08. The curved line corresponds to the best-fit sigmoid curve
for the median of the samples.
OCTOBER 2011 P I N E R O S E T A L . 693
using a radius of 300 km. The root-mean-square error
(RMSE) for the testing set of tropical cyclones was
14.7 kt (Fig. 4).
The second test consisted of training the intensity es-
timator with all tropical cyclones from the years 2004–08
and then using the eight tropical cyclones in the dataset
from 2009 as the independent test set. Figure 5 shows the
two-dimensional histogram and the best-fit parametric
curve, which was obtained using a radius of calculation
for the variance of 350 km. The total RMSE for the 2009
test set was 24.8 kt. The increase in RMSE was entirely
due to just two cases: Tropical Storms Ana and Erika.
Although these tropical cyclones were only weak trop-
ical storms, the cloud structure associated with each
showed high levels of axisymmetry (Fig. 6), resulting in
a DAV that was very low and thus an estimated intensity
that was too high. For these two cases the technique over-
estimated the intensity by more than 250% (Fig. 7). The
likely cause for this overestimate is the dislocation of the
systems’ centers from the very circular cloud masses by
environmental vertical shear. Work is on going to reduce
the overestimates of intensity caused by these kinds of
very circular, but displaced, systems. Until this is accom-
plished in an automated way, it will be necessary to su-
pervise the results of the intensity estimator to avoid this
particular kind of error. The RMSE for the six tropical
cyclones of 2009 excluding Tropical Storms Ana and
Erika is 12.9 kt (see Fig. 8). Table 1 summarizes the curve
parameters and the radius selected for these results.
4. Discussion
a. DAV time series
Although the filtered DAV signals are negatively cor-
related with the best-track intensity estimates (Pineros
et al. 2008), the oscillations present in the signals produce
some dispersion of DAV–wind speed samples, shown in
Figs. 3 and 5. This dispersion is unavoidable because the
best-track intensity estimates from the NHC that are
used as the reference are available only every 6 h as com-
pared with the 30-min resolution of the DAV estimates.
For example, Fig. 9 shows the 350-km DAV signal and the
wind speed for Hurricane Jeanne (2004); the open circles
indicate a diurnal oscillation (23.5 h apart). These fluc-
tuations in the DAV signal in comparison with the lower-
temporal-resolution best-track estimates decrease the
correlation with the best-track intensity. Figure 10 shows
the dispersion produced by the DAV–wind speed samples
of Jeanne in the two-dimensional histogram of Fig. 5. The
open circles shown in Fig. 9 are also plotted in Fig. 10.
Although we have mitigated this problem to some degree
by smoothing the DAV signal and fitting the sigmoid curve
to the data to produce our final DAV–intensity relation-
ship, this mismatch in the temporal resolution between
the DAV signal and the best-track intensity estimate will
always be a limiting factor on the agreement between the
DAV and the best track.
FIG. 4. Intensity estimates and best-track intensities for 20 tropical cyclones (30% of the dataset) randomly chosen from 2004 to 2008, and
the remaining 70% used to obtain a and b. The RMSE is 14.7 kt.
FIG. 5. Two-dimensional histogram of the 350-km filtered DAV
samples and best-track intensity estimates using 20 deg2 3 5-kt bins
for tropical cyclones from the period 2004–08. The curved line cor-
responds to the best-fit sigmoid curve for the median of the samples.
694 W E A T H E R A N D F O R E C A S T I N G VOLUME 26
b. Intraseasonal and interannual variability
Previous work by our group (Demirci et al. 2007) has
demonstrated that interannual and intraseasonal vari-
ability of the atmospheric circulation patterns can also
be a limitation on how well an automated technique
can estimate or predict tropical cyclone behavior. To
investigate whether there is sensitivity to either seasonal
or annual variations, the DAV–intensity curves were
recalculated based on training by year and training by
month over the 5-yr period of 2004–08. The fitted curves
for individual years and for individual months over
the 5-yr period are very similar to the overall training
curve for the entire period, suggesting that seasonal and
FIG. 6. Two weak tropical cyclones in 2009 that were removed from the analysis because of their high level of
axisymmetry: (a) Tropical Storm Ana, 0615 UTC 12 Aug (intensity: 30 kt and 1006 hPa), and (b) Tropical Storm
Erika, 0615 UTC 1 Sep (intensity not reported).
FIG. 7. Intensity estimates of two weak tropical cyclones in 2009 that were removed from the analysis: Tropical
Storms (a) Ana and (b) Erika.
OCTOBER 2011 P I N E R O S E T A L . 695
interannual variations in the North Atlantic basin do not
appear to affect the DAV intensity curve. The greatest
departure from the general intensity curve for any of the
annual curves is less than 10 kt (data not shown). A sim-
ilar result was obtained for the months of July–October,
where there are enough samples for the system to be stable
(data not shown). Thus, we conclude that there is no sig-
nificant value added in developing different parametric
curves for individual seasons or for different months.
c. Physical reasons for the overall robustnessof the method
Similar to the Dvorak technique that has proven to be
so successful for more than 30 years, this technique has
a physical foundation for its success. This foundation is
based on the premise that the cloud patterns and their
similarity to a perfectly annular pattern are directly re-
lated to the organization of the secondary circulation,
which includes the eyewall and rainband patterns. The
organization (and strength) of the secondary circulation
is then directly related to the size and intensity of the
tropical cyclone primary wind field. Thus, as a general
rule, the symmetric organization of the observed cloud
patterns is an indirect indicator of the intensity of the
primary wind circulation.
d. Rapid intensification
The cutoff frequency of the low-pass filter applied to
smooth the DAV signal determines its transient response,
which in turn increases as the bandwidth decreases
(Priemer 1991). Thus, decreasing the cutoff frequency of
the filter to reduce the fluctuations of the DAV signal so
as to obtain DAV–wind speed samples that are more con-
centrated around the curved line in Fig. 5 simultaneously
increases the technique’s error for rapid-intensification
tropical cyclones. The rapid intensification of Hurricane
Wilma (2005) is an example of how this trade-off can de-
crease the DAV performance. In the case of Wilma, the
low-pass filter that is applied to smooth the DAV signal
results in a response in the DAV signal that is behind the
actual intensification of Wilma (Fig. 11). The rate of in-
tensification and final intensity of Wilma are actually very
well modeled by the DAV parametric curve. However,
the starting time of the rapid-intensification phase is late,
and the time difference between the maximum values of
both signals is around 35 h. Implementing another fil-
ter configuration such as a finite impulse response, which
typically has lower transient responses, or utilizing a higher
cutoff frequency for rapid intensification cases might solve
this problem. This will be a subject for future study.
e. Real-time implementation
The first step in the process is to convert the imagery
from a natural Earth coordinate system to a Cartesian
FIG. 8. Intensity estimates and best-track intensities for 2009, using tropical cyclones from the period 2004–08 to obtain a and b. The
RMSE is 12.9 kt.
TABLE 1. Estimator parameters calculated for two training sets.
Training set
50 tropical cyclones (70%)
randomly chosen from
2004 to 2008
70 tropical cyclones
from 2004 to 2008
a (deg22) 0.002 655 0.002 697
b (deg2) 1008 1110
Radius (km) 300 350FIG. 9. Best-track intensity and 350-km DAV signal for Hurricane
Jeanne (2004). The two open circles are 23.5 h apart.
696 W E A T H E R A N D F O R E C A S T I N G VOLUME 26
projection that removes the differences in pixel resolution
that are due to Earth’s curvature. A standard software
package developed by the University of Wisconsin—
Madison, known as the Man–Computer Interactive Data
Access System (McIDAS; see online at http://www.ssec.
wisc.edu/mcidas/software/about_mcidas.html) was used
to compute the projection change. Processing the DAV
maps of an IR image takes less than 1 min on a 2.3-GHz
Intel Core i7 computer with 8 gigabytes of memory using
the Linux CentOS 5.5 operating system and running the
technique’s program with the software package Matlab
7.10 (in no-display mode).
Once the parameterized wind speed estimator is ob-
tained, a shell script can be executed every hour to com-
pute the projection change and calculate the DAV values
of the image using the radius chosen in the development
of the estimator. The process takes less than 4 min for
a single image. The DAV signal is obtained by adding one
sample every time that one image is processed. Although
these tasks can be automatically executed at a specific time
by programming a Linux script, the user should manually
start and stop the job.
5. Conclusions
This paper describes improvements to a completely
objective technique developed in Pineros et al. (2008,
2010) to characterize the intensity of a tropical cyclone.
The technique quantifies the axisymmetry of a tropical
cyclone with the deviation-angle-variance metric as an
indirect estimate of the intensity from infrared imagery
alone. In this paper, a set of parametric curves relating
DAV to maximum wind speed that are the result of the
technique were tested on two independent sets of trop-
ical cyclones: one set randomly selected from the 2004–
08 period to be used as the testing set and the other set
comprising eight tropical cyclones from 2009. These
tests produce an RMSE that is between 13 and 15 kt
after two obviously bad cases are removed from the
testing set. Although part of the remaining 13–15-kt
error is probably due to the DAV signal oscillations that
do not occur in the smoothed best-track intensity esti-
mates, other factors that may help to reduce the overall
FIG. 10. Histogram of Fig. 4. The red points are the 350-km DAV best-track samples for
Hurricane Jeanne (2004). The two open circles from Fig. 9 are plotted to pinpoint the dis-
persion produced from the DAV oscillations.
FIG. 11. Estimated intensity results for Hurricane Wilma (2005).
The RMSE is 31 kt.
OCTOBER 2011 P I N E R O S E T A L . 697
RMSE include binning cases by environmental factors
such as the environmental vertical wind shear and sea
surface temperatures; these are the topics of future work.
The potential of the DAV technique to quantify the
axisymmetry of a tropical cyclone (and thus to charac-
terize its dynamics) has been demonstrated in this and
previous papers. This characterization of the tropical cy-
clone dynamics with a single parameter is what makes it
possible to estimate robustly the intensity of the tropical
cyclone. The testing results presented in this paper sug-
gest that the intensity estimates produced by this tech-
nique are reasonably accurate, and, in its current version
as an independent estimate of tropical cyclone intensity,
the DAV technique is a complement to other estimates of
tropical cyclone intensity. Future work includes running
real-case simulations of the life cycle of tropical cyclones
using a full-physics, high-resolution mesoscale model to
calculate the DAV signal and compare it with the syn-
thetic maximum surface wind data from the same sim-
ulations at the same temporal resolution. From these
models the technique can be calibrated to estimate in-
tensity without the need to filter the signal to improve
the correlation with the lower-resolution best-track re-
cords. In addition, although initial results indicate that
there is little interannual or intraseasonal variability in
the DAV–intensity parametric curves, additional analysis
of these potential variations that might reduce the RMSE
will be undertaken. Furthermore, the physical processes
behind the robustness of the DAV–intensity relationship
will be studied using the high-resolution simulation out-
put to understand better, and thus improve, the DAV–
intensity relationship for intensity estimation.
Acknowledgments. Tropical cyclone best-track data
were obtained from NOAA’s National Hurricane Cen-
ter Internet site (http://www.nhc.noaa.gov). This study
has been supported by the Office of Naval Research
NOPP program under Grant N00014-10-1-0146.
REFERENCES
Beven, J. L., and Coauthors, 2008: Atlantic hurricane season of
2005. Mon. Wea. Rev., 136, 1109–1173.
Brennan, M. J., R. D. Knabb, M. Mainelli, and T. B. Kimberlain,
2009: Atlantic hurricane season of 2007. Mon. Wea. Rev., 137,4061–4088.
Brown, D. P., J. L. Beven, J. L. Franklin, and E. S. Blake, 2010:
Atlantic hurricane season of 2008. Mon. Wea. Rev., 138, 1975–
2001.
Demirci, O., J. S. Tyo, and E. A. Ritchie, 2007: Spatial and spa-
tiotemporal projection pursuit techniques to predict the ex-
tratropical transition of tropical cyclones. IEEE Trans.
Geosci. Remote Sens., 45, 418–425.
Demuth, J. L., M. DeMaria, J. A. Knaff, and T. H. Vonder Haar,
2004: Evaluation of Advanced Microwave Sounding Unit
tropical cyclone intensity and size estimation algorithms.
J. Appl. Meteor., 43, 282–296.
Dvorak, V. F., 1975: Tropical cyclone intensity analysis and
forecasting from satellite imagery. Mon. Wea. Rev., 103,
420–430.
Franklin, J. L., and D. P. Brown, 2008: Atlantic hurricane season of
2006. Mon. Wea. Rev., 136, 1174–1200.
——, R. J. Pasch, L. A. Avila, J. L. Beven, M. B. Lawrence, S. R.
Stewart, and E. S. Blake, 2006: Atlantic hurricane season of
2004. Mon. Wea. Rev., 134, 981–1025.
Gray, W. M., 1979: Hurricanes: Their formation, structure and
likely role in the tropical circulation. Meteorology over
Tropical Oceans, D. B. Shaw, Ed., Royal Meteorological So-
ciety, 155–218.
Knaff, J. A., D. P. Brown, J. Courtney, G. M. Gallina, and J. L. Beven
II, 2010: An evaluation of Dvorak technique–based tropical
cyclone intensity estimates. Wea. Forecasting, 25, 1362–1379.
Kossin, J. P., J. A. Knaff, H. I. Berger, D. C. Herndon, T. A. Cram,
C. S. Velden, R. J. Murnane, and J. D. Hawkins, 2007: Esti-
mating hurricane wind structure in the absence of aircraft
reconnaissance. Wea. Forecasting, 22, 89–101.
McBride, J., 1995: Tropical cyclone formation. A Global View
of Tropical Cyclones, R. Elsberry, Ed., WMO (Tech. Rep.
TCP-38), 63–105.
Olander, T. L., and C. S. Velden, 2007: The advanced Dvorak
technique: Continued development of an objective scheme to
estimate tropical cyclone intensity using geostationary in-
frared satellite imagery. Wea. Forecasting, 22, 287–298.
Pineros, M. F., E. A. Ritchie, and J. S. Tyo, 2008: Objective mea-
sures of tropical cyclone structure and intensity change from
remotely sensed infrared image data. IEEE Trans. Geosci.
Remote Sens., 46, 3574–3580.
——, ——, and ——, 2010: Detecting tropical cyclone genesis from
remotely-sensed infrared image data. IEEE Geosci. Remote
Sens. Lett., 7, 826–830.
Priemer, R., 1991: Introductory Signal Processing. World Scientific,
574 pp.
Ritchie, E. A., J. Simpson, T. Liu, J. Halverson, C. Velden,
K. Brueske, and H. Pierce, 2003: Present-day satellite tech-
nology for hurricane research: A closer look at formation and
intensification. Hurricane! Coping with Disaster, R. Simpson,
Ed., Amer. Geophys. Union, 249–891.
Spencer, R., and W. D. Braswell, 2001: Atlantic tropical cyclone
monitoring with AMSU-A: Estimation of maximum sustained
wind speeds. Mon. Wea. Rev., 129, 1518–1532.
Velden, C., T. Olander, and R. Zehr, 1998: Development of an
objective scheme to estimate tropical cyclone intensity from
digital geostationary satellite infrared imagery. Wea. Fore-
casting, 13, 172–186.
——, D. C. Herndon, J. Kossin, J. Hawkins, and M. DeMaria,
2006a: Consensus estimates of tropical cyclone intensity using
integrated multispectral (IR and MW) satellite observations.
Preprints, 27th Conf. on Hurricanes and Tropical Meteorol-
ogy, Monterey, CA, Amer. Meteor. Soc., P4.1. [Available
online at http://ams.confex.com/ams/pdfpapers/107408.pdf.]
——, and Coauthors, 2006b: The Dvorak tropical cyclone in-
tensity estimation technique: A satellite-based method that
has endured for over 30 years. Bull. Amer. Meteor. Soc., 87,1195–1210.
698 W E A T H E R A N D F O R E C A S T I N G VOLUME 26