Vous êtes sur la page 1sur 5

254 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 12, NO.

2, FEBRUARY 2015

Automatic Tracking of Pregenesis Tropical


Disturbances Within the Deviation
Angle Variance System
Oscar G. Rodríguez-Herrera, Kimberly M. Wood, Klaus P. Dolling, Wiley T. Black,
Elizabeth A. Ritchie, and J. Scott Tyo, Fellow, IEEE

Abstract—The deviation angle variance (DAV) method is an been developed to monitor their progress. The deviation angle
objective tool for estimating the intensity of tropical cyclones (TCs) variance (DAV) technique was introduced by Piñeros et al. as an
using geostationary infrared (IR) brightness temperature data. objective method to estimate the intensity of a TC from infrared
At early stages in TC development, the DAV signal can be also (IR) satellite imagery [1]–[4]. First, the gradient of IR bright-
a robust predictor of tropical cyclogenesis. However, one of the
problems with using the DAV method at these early stages is ness temperatures in the satellite image is computed at every
that the operator has to subjectively track potentially developing pixel within a given distance from a selected point that is being
cloud systems, sometimes before they are clearly identifiable. Here, assessed as a potential center of circulation of a developing or a
we present a method that allows us to automatically track the mature TC. Next, the angular deviation between the computed
evolution of cloud clusters using only the raw IR imagery and gradient and a radial from the selected center is calculated
the resulting DAV maps. We have compared our objective method for every pixel. The variance of this angle within the same
with results manually obtained on a limited data set spanning a
radius—the deviation angle variance or DAV—is computed to
12-day period during the 2010 hurricane season in the western
North Pacific and tuned the performance of the method to the obtain a single quantity that characterizes the symmetry of the
manual results. The performance of the method was then tested TC for the selected center. DAV values are negatively correlated
by comparing the results with best track and invest files produced with TC intensity. Thus, by using a training set of historical
by the Joint Typhoon Warning Center for the four-year period TCs, a parametric curve can be calculated that directly relates
2009–2012. The long-term results agree well with the best track the DAV to TC intensity for a particular basin that can be then
and invest files for the disturbances analyzed in terms of start used to estimate the intensity of current TCs [2], [3]. The DAV
time, end time, and locations of disturbances. The automatic
tracking method presented in this letter may be used to reduce the
methodology is discussed in detail elsewhere [1]–[4].
dependence of tropical cyclogenesis DAV analyses on the expertise The DAV method has been also extended to assist in iden-
and ability of the operator. tifying tropical cyclogenesis [5]. For this latter application, a
map of DAV values is created using every pixel in the satellite
Index Terms—Automatic storm tracking, deviation angle vari-
ance (DAV), remote sensing.
image, one at a time, as the center for the calculation and
then plotting the DAV value at that pixel’s location in the
resulting map. The map is then analyzed to find regions with
I. I NTRODUCTION DAV values below a predetermined threshold, which indicate
regions of high symmetry where a TC may be developing. The
A tropical cyclone (TC) can be defined as a warm-core non-
frontal synoptic-scale cyclone originating over tropical or
subtropical waters, with organized deep convection and a closed
application of this method to the North Atlantic [3] basin has
produced excellent results with median TC detection times just
surface wind circulation about a well-defined center (http:// before (0.6–3 h) the first indication of a tropical depression
www.nhc.noaa.gov/aboutgloss.shtml#t). Because these intense (TD) in the National Hurricane Center’s retrospective best track
convective systems form in regions where it is challenging archives [5].
to obtain direct measurements, satellite-based methods have To enable the transition of the DAV method to a real-time
operational forecast environment, it is necessary to keep track
of potentially developing cloud clusters in IR satellite images
Manuscript received July 24, 2013; revised May 23, 2014; accepted
June 18, 2014. Date of publication July 21, 2014; date of current version August
between synoptic times when the operational fixes are not
21, 2014. This work was supported in part by the National Oceanographic available. Objective methods such as centroid locations and
Partnership Program under Grant N00014-10-1-0146 and in part by the Office location of minimum DAV may be used to obtain a center fix
of Naval Research Marine Meteorology Program under Grant N00014-13-1- estimate for a given cloud cluster in an IR image. Unfortunately,
0365.
O. G. Rodríguez-Herrera, W. T. Black, and J. S. Tyo are with the College these objective methods have not proven to provide a robust
of Optical Sciences, The University of Arizona, Tucson, AZ 85719 USA way to keep track of a given disturbance, particularly early in
(e-mail: orh@optics.arizona.edu; wblack@optics.arizona.edu; tyo@optics. its development [6].
arizona.edu).
K. M. Wood, K. P. Dolling, and E. A. Ritchie are with the Department
Automatic storm tracking methods found in the literature
of Atmospheric Sciences, The University of Arizona, Tucson, AZ 85721 may be divided into three categories: methods based on pat-
USA (e-mail: wood@atmo.arizona.edu; dolling@atmo.arizona.edu; ritchie@ tern recognition [7], [8], cross correlation [8]–[11], and shape
atmo.arizona.edu). analysis [12]–[14]. In this letter, we present a shape-analysis-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. based objective automatic tracking method that uses the DAV
Digital Object Identifier 10.1109/LGRS.2014.2334561 information to track the position and evolution of potentially

1545-598X © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
RODRÍGUEZ-HERRERA et al.: TRACKING OF PREGENESIS TROPICAL DISTURBANCES WITHIN THE DAV SYSTEM 255

developing cloud clusters within the DAV system. This method


provides the framework for transforming the DAV technique
into a fully automated real-time forecaster aid for identifying
tropical cyclogenesis. The main objective of the work presented
herein is to keep track of potentially developing cloud clusters
and create a DAV cloud cluster evolution history before the
actual genesis of a TC, simplifying the use of the method in
operational centers.

II. DATA AND AUTOMATIC S TORM


T RACKING M ETHODOLOGY
This study focuses on western North Pacific TCs during
the four-year period 2009–2012. The data are derived from
longwave (10.7 μm) IR satellite imagery from the Japanese
Multifunctional Transport Satellite (MTSAT) of the western
North Pacific basin. The images are resampled to a spatial
resolution of 10 km per pixel and encompass an area from
0◦ to 40◦ N and from 100◦ to 180◦ E. This study consists
of a total of 19 558 unique hourly images over the four-year
period and includes 82 developing systems and 194 nonde-
veloping systems that were labeled as invests by the JTWC
but never achieved TC intensity. The results of the tracking
technique are compared against the JTWC best track database,
which can be obtained from http://www.usno.navy.mil/NOOC/
nmfc-ph/RSS/jtwc/best_tracks/ and the JTWC invest database
(M. Kucas, personal communication).
The objective DAV system automatically identifies locations
that meet user-defined thresholds as potential TCs. The operator
must then decide if a given detection belongs to a new potential Fig. 1. Flowchart with the main steps of the objective automatic tracking
storm or if it is part of a previously identified storm that the method. Details on a number of validations have been omitted to simplify the
chart. The dashed rectangles enclose the detections that are labeled as either
operator is already tracking. This part of the DAV analysis was ghost, positive, or true detections.
previously manually accomplished, was subjective, and placed
an additional burden on the operator, who was already work-
ing in a time-constrained environment. The tracking system ness temperature within a region with radius Rcloud centered
relieves the analyst from having to manually track pregenesis on the minimum DAV point and compares it with a user-
disturbances, allowing the analysis of the current state of evo- defined threshold THBmin . THBmin was chosen to be the 8-bit
lution of a disturbance without having to analyze previous IR equivalent of 273.15 K (∼147) as obtained from the MTSAT-2
imagery. calibration table [15]. The typical radius of a TC as reported
The tracking method presented herein uses information from in various studies (e.g., [16]) was used as the value of Rcloud
two fields: the brightness temperature IR satellite image and (250 km). Although TCs vary considerably in size, Rcloud
its corresponding DAV map. Fig. 1 shows a flowchart with need only encompass the main cloudy region associated with
the main steps in the method. The satellite IR images used the TC. Regions with a minimum DAV below THmax and
in this analysis were converted to an 8-bit digital format to an average brightness temperature greater than THBmin are
facilitate the computation of the DAV maps. Using this set of labeled as detections, assigned a number, and included in a list
data, the tracking system first looks for regions on the DAV that contains the minimum DAV, the latitude and longitude of
map with a variance below a given threshold THmax . The the point, and the detection time.
choice of THmax was obtained from the receiver operating char- If either of the two thresholds is not reached, the point
acteristic (ROC) curve computed in [4], which plots positive of minimum DAV is labeled as a ghost detection, and the
detection of pregenesis cloud clusters versus a false-alarm rate automatic tracking system checks the detections table for a
for particular DAV values. According to those results, a value previous detection within radius Rtrack of the ghost detection
of THmax = 2000 deg2 identifies 100% of the cloud clusters within a time period Ttrack . Rtrack = 500 km is consistent
that might develop into TCs, albeit with high false alarm rates. with the distance a cloud cluster would have propagated on
Next, the tracking system locates pixels with locally minimum average within the time Ttrack (initially set to 24 h). Ghost
DAV values and compares those pixels to the corresponding detections are used to follow cloud clusters that are of interest
IR image to verify that clouds likely associated with a weather in the event that no actual detection is found over an extended
disturbance of interest are present. This step is necessary as the period of time Ttrack . If no previous detection satisfying both
DAV map solely contains symmetry information and can find conditions is found in the detections table, the ghost detection
high symmetry in clear sky regions. The automatic tracking is discarded. Otherwise, the tracking system upgrades the ghost
system defines a cloud cluster by computing the average bright- to a full detection and includes the cluster information for that
256 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 12, NO. 2, FEBRUARY 2015

is a function of the DAV value for each detection: large dots


correspond to low DAV values, and small dots correspond to
high DAV values. Two TCs shown in Fig. 2, namely, Lionrock
(19.4◦ N, 118.1◦ E, current intensity = 50 kt, dark brown dots)
and Kompasu (28.4◦ N, 126.6◦ E, current intensity = 103 kt,
light brown dots), were included in the 2010 best track database
from the JTWC.1

III. P ERFORMANCE OF THE M ETHOD


Fig. 2. (Color online) Snapshot of the automatic tracking system with two The automatic tracking method was tuned by comparing
dissipating TCs that were previously tracked and six identified cloud clus-
ters being tracked in the western North Pacific basin on August 31, 2010, with consensus results manually obtained from a subjective
at 2132 UTC. blind test by two operators. Each operator analyzed a set of
262 hourly IR brightness temperature images of the western
time in the detections table. This is an important step because North Pacific basin between August 25, 2010, at 0032 UTC and
a cloud cluster being tracked often loses symmetry during its September 5, 2010, at 2332 UTC. A total of 26 nonconsecutive
early evolution, but later reorganizes and continues developing. hourly IR images was missing from the data set. The operators
If either no detection associated with a previous detection is in this case were postdoctoral researchers in our group with ex-
found within a time period Ttrack or the average brightness tensive experience in TC analysis. Cloud clusters were tracked,
temperature falls below THBmin , the cloud cluster is considered and the location and minimum DAV values were recorded. Be-
to have dissipated. Any new detection in the vicinity is labeled cause there were minor differences between the resulting anal-
as a new cloud cluster and tracked as described above. yses, a subjective consensus between the operators was formed.
The THmax DAV value used to identify detections produces The results from the subjective consensus were compared with
realistic TC durations in comparison with those reported in the the results from the automatic system, and some parameters
best track database. However, using this value as the threshold (THmax , THtrue , Rcloud , THBmin , Rtrack , and Ttrack ) were
for true detections, i.e., cloud clusters that have the potential to calibrated to ensure that the automatic results match the manual
develop into TCs, produces a large number of false positives results within a radius of 5◦ and a time of detection of 6 h. The
[5]. Therefore, a lower DAV value, i.e., THtrue , is used to filter majority of parameters had little or no effect on the performance
out false detections and generate a second table that contains of the method when varied by less than 10% of their initial
the corresponding cluster information. The value THtrue = value. The only exception was THBmin , for which a small
1750 deg2 was chosen for this study from the ROC curve in variation enabled a better match with the subjective results. The
[4] because it yielded a true positive rate greater than 95% adopted value of THBmin is 145.
with a false positive rate (FPR) of less than 25% in the earlier Table I shows the start and finish dates/times reported by
work. In addition, this value yields a median time of detection the automatic tracking using the THmax threshold. The time
approximately equal to the designation of a TC as a TD in difference Δτ1 is the duration of the storm from the first time
the best track archive in both the North Atlantic [3] and the that the DAV value falls below THmax until the last time that it
western North Pacific [4]. A strategy to further reduce the FPR is below that value. The differential Δτ2 is the time difference
was implemented in the tracking system using the cumulative between the first and last times that the DAV falls below THtrue ,
time that a potentially developing storm spends below THtrue . i.e., the duration obtained by the automatic tracking system that
The minimum cumulative time to consider a cloud cluster as corresponds to the duration reported by the manual test. The
an actual true positive, i.e., Ttrue , was chosen to be a full comparison between the automatic tracking and the manual test
diurnal cycle. Furthermore, a land mask was applied to initial reported latitude, longitude, start time, and finish time showed
detections, since cloud clusters originating over land rarely that these were exactly the same for most storms, with the
develop into TCs. However, subsequent detections over land of exception of the finish time reported for cloud clusters 2 and
an already identified cloud cluster of interest are retained. The 3 discussed below. The perfect agreement between the latitude
combination of these two criteria resulted in no false positives and longitude of the cloud clusters is to be expected, since they
detected in this analysis. In real-time genesis analyses, the value correspond to those of the point with minimum DAV within a
of Ttrue can be reduced to only a few hours (e.g., 3–5 h) in order circular region of radius Rtrack . The differential Δτ12 is the
to avoid delaying the detection time by a full diurnal cycle. time difference between the first time that the DAV value falls
However, the reduction in the detection time delay comes at below THmax and the first time that it falls below THtrue . The
the expense of a higher FPR. TC names reported in the table are the names assigned to true
Fig. 2 shows a snapshot of the automatic tracking system positives by the JTWC.
with six identified cloud clusters being tracked and two dis- All of the manually detected cloud clusters have a cor-
sipating TCs that are still being tracked. The circles in the responding cloud cluster detected by the automatic tracking
image are centered on the current detections and color-coded to system. The assigned numbers do not coincide between the
distinguish between new detections (blue), previously identified
1 This letter has supplementary downloadable material available at http://
detections (green), and track detections (yellow). The eight sets
of dots represent the current and previous locations of each ieeexplore.ieee.org, provided by the authors. This includes a README file and
a movie file showing the performance of the tracking system for the western
cloud cluster with a time latency Ttrack . Different colors have North Pacific basin between August 25, 2010, at 0032 UTC and September 5,
been used to mark each identified cloud cluster. The size of each 2010, at 2332 UTC. This material is 7 MB in size.
RODRÍGUEZ-HERRERA et al.: TRACKING OF PREGENESIS TROPICAL DISTURBANCES WITHIN THE DAV SYSTEM 257

TABLE I
L IST OF T RUE P OSITIVE D ETECTIONS O BTAINED F ROM THE B LIND T EST A FTER C ONSENSUS AND THE O BJECTIVE AUTOMATIC T RACKING S YSTEM .
T HE S TORM N UMBERS L ISTED IN E ACH C OLUMN C ORRESPOND TO THE N UMBERS A SSIGNED BY E ITHER THE
M ANUAL T RACKING OR THE AUTOMATIC T RACKING S YSTEM

TABLE II TABLE III


C OMPARISON B ETWEEN THE S TART AND F INISH DATES /T IMES OF THE N UMBER OF D EVELOPING AND N ONDEVELOPING D ISTURBANCES
T RUE P OSITIVE D ETECTIONS O BTAINED F ROM THE B LIND T EST A FTER R EPORTED BY THE JTWC AND D ETECTED BY THE O BJECTIVE
C ONSENSUS AND THE O BJECTIVE AUTOMATIC T RACKING S YSTEM AUTOMATIC T RACKING M ETHOD D URING E ACH S EASON OF THE
F OUR -Y EAR P ERIOD 2009–2012

manual and automatic tracking, except for the first part of


the first cloud cluster and the second cloud cluster, because representative of the performance of our method. The numbers
the total number of cloud clusters detected in each case is of disturbances reported by the JTWC and our tracking method,
different. In most cases, the start and finish dates/times used including best track and invests, for which enough IR imagery
to compute Δτ2 coincide perfectly between the manual and is available in our archive to enable the tracking analysis are
automatic tracking, as shown in Table II. Two exceptions are shown in Table III. Most TCs in the best track archive for which
cloud clusters 2 and 3 (Lionrock and Kompasu, respectively) we have data available were detected and successfully tracked
from the automatic tracking system, which detected the last by the tracking system, with a maximum location difference
time the DAV value was below THtrue up to 17 h after the end between best track entries and tracking system detections of 5◦
of the cloud cluster recorded on the manual test. This shows the in latitude and longitude within a time difference of 6 h, i.e., the
high sensitivity of the automatic system to track even vanishing time intervals at which the best track and invests are recorded.
cloud clusters. Furthermore, cloud cluster 2 on the automatic The TCs that the system missed either did not meet the THtrue
results was identified by the automatic tracking system as a threshold or met it for a time shorter than Ttrue . That is, the
potentially developing TC 6 h in advance of the manual test missing TCs were not identified as true positives. The system
(see Table II). achieved a correct detection rate of 91.4%, which is close to the
An important issue observed in the results of the automatic rate of 95% for the same threshold reported in [4] during the
tracking is the split of the storm labeled by the system as cloud development of the genesis predictor. The mean durations of
cluster 1, which was identified on the manual test as cloud automatic tracking for developed TCs and the detected invests
clusters 1, 3, and 5. This tracking mismatch is due to the com- that matched JTWC invests were 1.27 and 0.93, respectively,
plexity of the cloud cluster that developed into TS Namtheun. when normalized by the duration recorded in the best track and
Its average brightness temperature falls just below THBmin and invest databases, meaning that, on average, the system tracked
its DAV value rises just above THmax at intermediate stages true TCs for longer than the JTWC and tracked nondeveloping
of its development. This is a tricky case that was difficult to iden- systems for a slightly shorter time period. The standard devi-
tify as a single cloud cluster on the test because that particular ation for the entire population was ∼1.2. The tracking system
cloud cluster was surrounded by other complex cloud clusters identified the disturbances an average 26 h before they were
that later developed into TS Lionrock and Typhoon Kompasu. designated as TD in the best track. In the case of invests, most
disturbances for which we have data available were success-
fully identified and tracked by the automatic system. The rest
IV. P ERFORMANCE IN L ONG -T ERM T RACKING
of them were very weak and did not produce a DAV signal
Once the automatic tracking system was tuned to produce low enough to reach the THmax threshold. This property of the
results comparable with the manual analysis for the 12-day automatic tracking method can be used to discriminate invests
period, the performance was tested using four years of best with potential to develop from those that are highly unlikely
track and invest data from the JTWC in the western North to develop. A number of disturbances detected by the tracking
Pacific encompassing 2009–2012. The test was limited to 52% method using the THmax threshold do not have a counterpart in
of the disturbances in the JTWC archives because of the lack the invest archive, but do not reach the THtrue threshold or do
of availability of some IR imagery in our database. The miss- it for a period shorter than Ttrue . Overall, the system identifies
ing intervals were randomly distributed through the four-year many more clusters that cross the THmax threshold, but the vast
period and did not introduce any temporal bias into the test. majority of these are automatically discarded.
Other basins were not included in this study. Nevertheless, Approximately 30% of the disturbances in the JTWC’s best
due to the high complexity of the disturbances in the western track and invest files were detected as two or three differ-
North Pacific, we expect the results obtained in this basin to be ent disturbances by the automatic tracking system. We have
258 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 12, NO. 2, FEBRUARY 2015

identified three classes of disturbances that present this issue. those cloud clusters with potential to continue developing, and
The first class is composed of disturbances that form in the report those filtered cloud clusters continuously to a database,
South China Sea. These disturbances often have large cloud relieving the forecaster of considerable manual work. Current
bands far from, yet associated with, the central circulation. and future work include the following: 1) testing the automatic
However, the automated tracking system has difficulty con- tracking technique in the eastern North Pacific basin in con-
necting them, and the close proximity of land complicates junction with the genesis DAV technique; 2) further improve
the process. The second class is constituted by systems that the technique for those identified cases where the technique
undergo extratropical transition. Because these systems rapidly does not perform well; and 3) using the built-in capabilities of
accelerate, they require different internal settings of Rtrack and the automated system to develop the probabilistic genesis DAV
Ttrack than the main set of cases and are beyond the scope technique.
of this work. Finally, the third class is constituted by invests
with low-level circulation centers but little deep convection or
high cirrus shields that are tracked by the JTWC using visible ACKNOWLEDGMENT
imagery. Because the automated tracking system utilizes only The authors would like to thank the JTWC for providing the
IR imagery, such systems are currently problematic. satellite imagery and invest files used in this study. The best
Because of the large amount of data produced for the four- track data were obtained from the JTWC’s public website.
year comparison between the objective automatic tracking
method and JTWC’s tracking files, it is impractical to present R EFERENCES
the results of the comparison in this letter. However, the data
[1] M. F. Piñeros, E. A. Ritchie, and J. S. Tyo, “Objective measures of tropical
set is available at http://ieeexplore.ieee.org as single-year tables cyclone structure and intensity change from remotely sensed infrared
with a side-by-side comparison of the detection date, time, and image data,” IEEE Trans. Geosci. Remote Sens., vol. 46, no. 11, pp. 3574–
location recorded in the best track or invest files with those 3580, Nov. 2008.
reported by the objective automatic tracking method for each [2] M. F. Piñeros, E. A. Ritchie, and J. S. Tyo, “Estimating tropical cyclone
intensity from infrared image data,” Weather Forecast., vol. 26, no. 5,
disturbance identified by the latter. In general, the coincidence pp. 690–698, Oct. 2011.
between the JTWC’s tracking files and the results of the auto- [3] E. A. Ritchie, G. Valliere-Kelley, M. F. Piñeros, and J. S. Tyo, “Tropical
matic tracking method is extremely close, although not perfect. cyclone intensity estimation in the North Atlantic basin using an improved
deviation angle variance technique,” Weather Forecast., vol. 27, no. 5,
One issue we particularly noted during the four-year test was pp. 1264–1277, Oct. 2012.
that, in the event of a temporary satellite shutdown, or other [4] M. F. Piñeros et al., “Tropical cyclone intensity estimation and formation
reason for loss of images, human intervention is necessary to detection using the deviation-angle variance technique,” presented at the
realign the tracked cloud clusters. Proceedings 30th Conference Hurricanes Tropical Meteorology, Ponte
Vedra Beach, FL, USA, Apr., 2012, Paper 6C.3.
[5] M. F. Piñeros, E. A. Ritchie, and J. S. Tyo, “Detecting tropical cyclone
genesis from remotely sensed infrared image data,” IEEE Geosci. Remote
V. D ISCUSSION AND C ONCLUSION Sens. Lett., vol. 7, no. 4, pp. 826–830, Oct. 2010.
[6] M. F. Piñeros, “Objective measures of tropical cyclone intensity and for-
An automated cloud cluster tracking system based on the mation from satellite infrared imagery,” Ph.D. dissertation, Univ. Arizona,
DAV has been presented, which is designed to ease the burden Tucson, AZ, USA, 2009.
of keeping track of pregenesis disturbances identified by the [7] D. E. Wolf, D. J. Hall, and R. M. Endlich, “Experiments in automatic
cloud tracking using SMS-GOES data,” J. Appl. Meteorol., vol. 16, no. 11,
DAV system. The method provides an objective estimation pp. 1219–1230, Nov. 1977.
of the location of a cloud cluster that can be used to create [8] R. M. Endlich and D. E. Wolf, “Automatic cloud tracking applied to
a pregenesis history of both developing and nondeveloping GOES and METEOSAT observations,” J. Appl. Meteorol., vol. 20, no. 3,
storms. This history may be used to identify features of devel- pp. 309–319, Mar. 1981.
[9] J. Schmetz et al., “Operational cloud-motion from Meteosat infrared
oping tropical disturbances during their early stages of devel- images,” J. Appl. Meteorol., vol. 32, no. 7, pp. 1206–1225, Jul. 1993.
opment and separate them from cloud clusters that are clearly [10] A. V. Gambheer and G. S. Bhat, “Life cycle characteristics of deep cloud
nondeveloping. systems over the Indian region using INSAT-1B pixel data,” Mon. Weather
To demonstrate the effectiveness of the technique, first, a Rev., vol. 128, no. 12, pp. 4071–4083, Dec. 2000.
[11] A. N. Evans, “Cloud motion analysis using multichannel correlation-
12-day comparison with a manual analysis was used to tune the relaxation labeling,” IEEE Geosci. Remote Sens. Lett., vol. 3, no. 3,
automatic technique. After tuning, the technique was compared pp. 392–396, Jul. 2006.
with four years of invest and best track data from the JTWC [12] P. Gamba, “Meteorological structures shape description and tracking by
means of BI-RME matching,” IEEE Trans. Geosci. Remote Sens., vol. 37,
in the western North Pacific. The results presented in this letter no. 2, pp. 1151–1161, Mar. 1999.
show that the objective automatic tracking method is capable [13] D. P. Mukherjee and S. T. Acton, “Cloud tracking by scale space
of tracking cloud clusters in the western North Pacific, and its classification,” IEEE Trans. Geosci. Remote Sens., vol. 40, no. 2, pp. 405–
results are in good agreement with the JTWC best track and 415, Feb. 2002.
[14] C. Thomas, T. Corpetti, and E. Mémin, “Data assimilation for convective-
invest databases using a four-year period of comparison. cell tracking on meteorological image sequences,” IEEE Trans. Geosci.
Although this is a limited sized data set, preliminary work in Remote Sens., vol. 48, no. 8, pp. 3162–3177, Aug. 2010.
other basins does appear to corroborate the findings presented [15] Internet: Calibration Table, MTSAT-2 (HRIT), Sep. 10, 2012. [Online].
Available: http://mscweb.kishou.go.jp/operation/calibration/mt2/HRIT/
here. While the automatic tracking system has been designed to mt2_hrit_20100315.htm
work with the DAV genesis technique [5], it can be also used [16] R. T. Merrill, “A comparison of large and small tropical cyclones,” Mon.
to initially locate all existing cloud clusters, filter them for only Weather Rev., vol. 112, no. 7, pp. 1408–1418, Jul. 1984.

Vous aimerez peut-être aussi