Abstract

A technique to identify and quantify intense convection in tropical cyclones (TCs) using bispectral, geostationary satellite imagery is explored. This technique involves differencing the water vapor (WV) and infrared window (IRW) channel brightness temperature values, which are available on all current operational geostationary weather satellites. Both the derived IRW minus WV (IRWV) imagery and the raw data values can be used in a variety of methods to provide TC forecasters with important information about current and future intensity trends, a component within the operational TC forecasting arena that has shown little improvement during the past few decades.

In this paper several possible uses for this bispectral technique, both qualitative and quantitative, are explored and outlined. Qualitative monitoring of intense convection can be used as a proxy for passive microwave (MW) imager data obtained from polar-orbiting satellite platforms when not available. In addition, the derived imagery may aid in the TC storm center identification process, both manually and objectively, especially in difficult situations where the IRW imagery alone cannot be used such as when the storm circulation center and/or eye features are obscured by a cirrus canopy. Quantitative methods discussed involve the predictive quality of the IRWV data in terms of TC intensity changes, primarily during TC intensification. Strong correlations exist between storm intensity changes and IRWV values at varying 6-h forecast interval periods, peaking between the 12- and 24-h time periods. Implications for the use of the IRWV data on such objective satellite intensity estimate algorithms as the University of Wisconsin—Madison (UW) Cooperative Institute for Meteorological Satellite Studies (CIMSS) advanced Dvorak technique (ADT) are also discussed.

1. Introduction

Identifying and forecasting tropical cyclone intensity changes have been longstanding issues for operational topical cyclone (TC) forecasters and modelers. Recent improvements to track forecasts have been realized, but intensity forecast accuracy has stayed relatively constant over the same time period. Identification of intense convection and how it relates to TC intensity changes has been studied for years (Zehr 1989; Steranka et al. 1986; Dvorak 1984), but without direct measurement of the convective strength and magnitude, it is difficult to predict how and when these intensity changes will occur. In the last 10 years or so, passive microwave (MW) imagery from sensors on polar-orbiting platforms have become more available to TC analysts for viewing beneath the clouds to help identify the areas of strongest TC convection (Hawkins et al. 2008; Lee et al. 2002; Hawkins et al. 2001). Unfortunately, the polar satellite overpasses of a target TC can have large temporal gaps, thereby missing convective structure changes that may be relevant to TC intensity changes.

A concept to address this issue using geostationary satellite data was presented by Velden and Olander (1998), but until recently this potential method has not been explored further. In this study, a technique is developed that builds on the Velden and Olander (1998) concept and highlights the spectral response differences between geostationary infrared window (IRW) and water vapor (WV) channel data in regions of intense TC convection. As shown in the Geostationary Operational Environmental Satellite-13 (GOES-13) standard tropical atmosphere weighting functions in Fig. 1, the WV will typically be colder than the IRW during tropospheric clear-sky conditions. The WV spectral response peak is about 350 mb while the IRW peak is at or near the surface. However, in opaque cloud conditions associated with intense, active convection penetrating the tropopause, the sign of the measured difference between the two channels can reverse due to the reemitted absorbed radiation from upper-tropospheric–lower-stratospheric (UTLS) water vapor (Schmetz et al. 1997). Observations of UTLS water vapor have been documented since the 1940s (Brewer 1949), with measurements in the tropics cited in several studies (Kley et al. 1982; Fritz and Laszlo 1993; Ottenbacher and Schmetz 1994). Cloud model simulations (Grosvenor et al. 2007) and field experiment in situ data collected during the Cirrus Regional Study of Tropical Anvils and Cirrus Layers-Florida Area Cirrus Experiment (CRYSTAL-FACE) over Florida during 2002 (Rosenlof 2003) have further verified the convective processes involved in this phenomenon.

Fig. 1.

Standard tropical atmosphere weighting functions for GOES-13 (left) channel 3/WV and (right) channel 4/IRW.

Fig. 1.

Standard tropical atmosphere weighting functions for GOES-13 (left) channel 3/WV and (right) channel 4/IRW.

Typically, the most intense convection associated with a TC is within the storm eyewall where the surface convergence and corresponding vertical convective updrafts are strongest. Figure 2 displays a cross section of GOES-12 brightness temperatures (Tbs) during Hurricane Wilma (2005) near the time of its maximum intensity at 0915 UTC 19 October. The west to east cross section transverses the eyewall as well as the outer TC rainbands of the storm. In the outer regions of the TC, away from the eyewall region, the clouds are typically less opaque or missing entirely, resulting in brightness temperature measurements in line with those expected; WV Tbs are colder than the corresponding IRW values. In the inner eyewall region, however, this relationship reverses as water vapor above the convection is forced through the tropopause and into the stratosphere where it is reemitted in the WV at higher temperatures. In the Tb cross-section plot in Fig. 2 within the pixel number region from approximately 120–330 the IRW minus WV (IRWV) difference values are near zero or negative, indicating the presence of vigorous convection “overshooting” into the stratosphere. The most negative values are concentrated in the approximate 150–195 and 215–295 pixel count regions of the cross section within the TC eyewall. Outside of the eyewall regions, in the outer regions of the TC, the IRWV values are near zero or positive, indicating less vigorous convection.

Fig. 2.

Cross section of GOES-12 WV (red line) and IRW (blue line) cloud-top temperature values during Hurricane Wilma (2005). Satellite inset displays the location of the cross section (yellow line) transversing the eye region.

Fig. 2.

Cross section of GOES-12 WV (red line) and IRW (blue line) cloud-top temperature values during Hurricane Wilma (2005). Satellite inset displays the location of the cross section (yellow line) transversing the eye region.

This signal is used as the basis for our study and is explored further by examining several TC cases. Both qualitative and quantitative applications are investigated.

2. Product derivation

Generation of the IRWV-derived data and imagery is relatively straightforward. Coincident geostationary satellite IRW and WV brightness temperatures are differenced. Since all current operational IR sensors possess the same spatial resolutions for the IRW and WV imagery (4 km for the current GOES imagery), the process is a simple differencing of corresponding pixels in the respective channels. Both the analytical derivation and graphical display of the derived temperature difference field are performed using the Man computer Interactive Data Access System (McIDAS) from the Space Science and Engineering Center (SSEC) at the University of Wisconsin—Madison, but the process is not restricted to the McIDAS system, as it could easily be duplicated in most satellite display and analysis systems.

To conduct the various statistical analyses performed for this study, the temperature differences between the two geostationary satellite channels must be derived. This is done by subtracting the WV channel temperatures from the IRW channel temperatures. As expected, most of the differences will be positive except in the regions of vigorous convection. For this study, the areal analysis range is limited to a radius of 136 km from the selected storm center position.

The absolute temperature difference values are sorted into individual 0.5°C bins for values between +10° and −10°C, as well as two additional bins for values above and below +10° and −10°C, respectively. It is from this dataset that individual bins, combinations of bins, and other selected cloud-top region parameters are regressed against aircraft reconnaissance measurements to determine the correlation and statistical significance with regard to corresponding MSLP measurements (discussed further in the next section).

Once the differencing is completed, the output data can be converted for qualitative analysis from IRWV values to a corresponding “stretched” brightness value range for graphical display and enhancement of the difference field. The enhancement stretches/highlights the more negative IRWV temperature value range using the following conversion equation: Bexp = exp[1 + (−1 × IRWV)], where the Bexp brightness value range is from 1 to 255. The IRWV values below approximately −4.5 are capped at Bexp of 255. The exponential Bexp brightness value range, corresponding to negative IRWV temperature values, is then recombined with the positive IRWV temperature values to produce the final IRWV enhancement (IRWV-exp).

A two-dimensional low-pass filter is then applied to the differenced IRWV field for values less than 0°C using line and element coefficient values of 0.4. The low-pass filter removes much of the noise element from the derived image (especially when utilizing GOES imagery, because of oversampling and striping issues) while retaining coherent features/structures.

3. Data analysis and application

This technique has several potential quantitative and qualitative uses for tropical cyclone intensity analysis, including as an aid in storm positioning and predicting storm intensification onset.

a. Qualitative analysis

1) Storm center determination

A potential qualitative use of the IRWV imagery could be in aiding forecasters in the determination of TC center positions during periods when the storm is covered by a uniform cloud shield, otherwise known as a central dense overcast (CDO). In these situations, TC forecasters typically rely on non-geostationary-satellite-based observations, such as MW imagery from polar satellites, and/or have to rely on the extrapolation of previously determined storm center locations. Passive MW imagery can be extremely useful in these situations; however, due to the infrequent temporal sampling and limited areal viewing swath of the MW imagery obtained from polar-orbiting satellites, this method can only offer help periodically. In situ observations may provide some help, but are typically (outside of Atlantic reconnaissance missions) not near enough to the storm center to provide exact vortex position measurements.

An objective method for determining TC intensity, the advanced Dvorak technique (ADT; Olander and Velden 2007), relies on the automated interrogation of infrared satellite imagery to deduce a TC’s position before it can estimate a current TC intensity. The ADT relies upon a two-step process to automatically determine a TC’s position from a single IRW image. It first performs a “spiral centering” (SC) analysis upon the entire IR field of Tb encompassing the TC, focusing on the storm-scale structure/organization to estimate a center position. A second finer-scale “ring fit” (RF) analysis is then performed, which searches for high-gradient temperature fields corresponding to the eye/eyewall region of the TC. These methods are reliable during situations when an eye has developed or when spiral banding features are apparent. However, in situations when the cloud-top region is covered with a uniform cirrus shield, masking the TC eye or spiral band features, determination of the TC storm center can prove to be difficult using automated methods. Interrogation of the IRWV fields using the combined SC–RF automated analysis described above, especially during the CDO-type conditions, can provide additional information to aid the automated ADT storm center determination process.

Figure 3 illustrates an example of how a derived IRWV image could be used to aid TC forecasters in either automated or manual selection of a TC center position. The example shows Hurricane Katrina on 26 August 2005, located over the eastern Gulf of Mexico and southern Florida prior to its rapid intensification to maximum intensity. The images show that the storm is becoming more organized and steadily intensifying. Locating the TC storm center using traditional IR imagery is very difficult at this time, [shown with a stretched IRW black-and-white enhancement, with the standard and commonly employed Dvorak “BD curve” enhancement (IRBD), and with a maximum-stretched, colorized, linear enhancement displaying only the IRW pixels matching the IRWV negative pixels (IRW-max)], and passive MW imagery ±3 h from this time was either not available or did not provide a nadir view of the storm from the polar-orbiter overpasses. Displaying the three IR enhancements along with the IRWV-exp enhancement in Fig. 3 highlights the additional information provided by the IRW and WV channel differencing over a simple contrast stretch or enhancement of the IRW imagery alone. This is particularly highlighted by comparison of the IRW-max enhanced IRW image versus the IRWV-derived image. The convective signatures displayed in the IRWV-exp image are not apparent in the IRW-max enhanced image, or any of the additional IR enhanced–stretched images, highlighting the information content provided by the differencing technique. It must be emphasized that the IR-max enhancement is not an existing product or method for examining convection within a TC. It is only used here to highlight the pixels in the IR imagery that exactly correspond to the pixels identified in the IRWV technique (and IRWV-exp enhancement) to further illustrate the limitations of examining the cloud-top regions in the IR imagery alone.

Fig. 3.

Comparison between a (top left) GOES IRW image, (middle left) operational “BD curve” enhanced IRW image, (top right) IRW image with the IRW-Max enhancement, (middle right) derived IRWV channel image with exponential color enhancement, and (bottom right) NOAA/NWS NEXRAD composite reflectivity radar image from Key West, FL, for Hurricane Katrina at 0915 UTC 26 Aug 2005. Official NHC final BT (magenta), interpolated NHC real-time forecast (cyan), and ADT automated storm center position derived from the IRW image (blue) are shown by colored squares. Storm center position resulting from the ADT automated storm center determination scheme using the derived IR-WV image is shown by the red arrow. (Radar image courtesy of the National Climatic Data Center, Washington, DC.)

Fig. 3.

Comparison between a (top left) GOES IRW image, (middle left) operational “BD curve” enhanced IRW image, (top right) IRW image with the IRW-Max enhancement, (middle right) derived IRWV channel image with exponential color enhancement, and (bottom right) NOAA/NWS NEXRAD composite reflectivity radar image from Key West, FL, for Hurricane Katrina at 0915 UTC 26 Aug 2005. Official NHC final BT (magenta), interpolated NHC real-time forecast (cyan), and ADT automated storm center position derived from the IRW image (blue) are shown by colored squares. Storm center position resulting from the ADT automated storm center determination scheme using the derived IR-WV image is shown by the red arrow. (Radar image courtesy of the National Climatic Data Center, Washington, DC.)

To further verify the reliability of the convective signatures noted by the IRWV technique, a corresponding 0915 UTC National Oceanic and Atmospheric Administration/National Weather Service (NOAA/NWS) Next-Generation Doppler Radar (NEXRAD) level III, 124-n mi range, composite reflectivity radar image from Key West, Florida, is shown beneath the IRWV-max enhancement image in Fig. 3. Much greater similarities are noted between the IRWV image and the radar image, than in the IRW only, especially with regard to the partial eyewall region around the storm center, and the large convective area to the southeast of the center over the Florida Keys.

Use of the ADT automated storm determination scheme would not be of much help in determining the TC center for this IRW image, with the resultant automated position located to the southeast of the National Hurricane Center (NHC) forecast. The derived IRWV image using the IRWV-exp enhancement, however, was able to clearly highlight the region of intense convection and possible forming eyewall not apparent in the three IRW images provided. This location was also identified as the storm center by the ADT automated storm center determination algorithm after interrogating the derived IRWV image, further supporting the selection of this point as the storm center. The IRWV’s automatically derived position corresponds nicely with the NHC interpolated final best-track position of Hurricane Katrina at that time.

In summary, the derived IRWV imagery can be used by operational TC forecasters in some cases to help identify the more vigorous convection regions not apparent in the IR imagery alone and to better estimate TC center positions. The derived IRWV imagery can also provide an objective storm center determination scheme (such as the SCRF techniques used in the ADT) with a more defined signal in the cloud-top temperature region to be interrogated, resulting in more precise storm center positions in CDO-type situations. Once the ADT is calibrated to use the IRWV information correctly, improvements in storm center determinations will result in more accurate ADT scene-type determinations and thus more accurate intensity estimates.

2) Comparison with passive microwave imagery

The unique ability of passive microwave imager data in the 85–89-GHz channel range to identify and measure active TC convection located under a cirrus canopy (obscured within typical geostationary IRW imagery) has been a huge benefit to operational TC forecasters, providing storm center positions and structure (organization) analyses. This is particularly true at night, when visible (VIS) imagery cannot be used to aid the IRW analysis. The main drawback to the availability of the MW imagery, however, is that the sensors are located on polar-orbiting satellites, which may only overpass a particular TC a few times a day, if at all. This infrequent data sampling can result in long stretches of time where the storm structure and exact position are not known to the degree desired by TC forecasters.

The IRWV-derived imagery can qualitatively be used as a proxy for the MW image data during these long data gaps. Figure 4 illustrates the similarity that can exist between the IRWV-derived imagery as compared to MW 85–89-GHz channel imagery, for a case during Hurricane Wilma (2005). The IRWV image is enhanced using the exponential enhancement IRWV-exp described earlier. This enhancement will highlight the heaviest convection (the largest negative difference values) in the derived IRWV imagery and help TC forecasters discern the locations and organizational characteristics of the strongest convective regions. Notable similarities exist between the MW imagery and the IRWV imagery, especially in the areas of heaviest convection (yellow and red regions in each image) as well as regions where convection is absent (due to a possible eyewall replacement cycle in this example). Corresponding IRW imagery enhanced with the operational BD enhancement (IRBD) imagery is also provided to show that the strong convective regions and structure may not be apparent in the routinely used IRW (and/or WV) imagery alone.

Fig. 4.

Comparison of (right) derived IR-WV channel image using exponential enhancement with (left) Tropical Rainfall Measuring Mission (TRMM) 85-GHz passive MW image for Hurricane Wilma on 19 October 2005. Insets show corresponding IRW images using (top right) BD-curve enhancement and (top left) matched IR image with IRW-Max enhancement. TRMM overpass was at 1740 UTC; IR-WV and IRW images were at 1745 UTC. [The MW image was obtained from the Naval Research Laboratory at Monterey, CA (NRL—Monterey), tropical cyclone Web site archive.] Units of IR-WV and two IRW image enhancements are in °C.

Fig. 4.

Comparison of (right) derived IR-WV channel image using exponential enhancement with (left) Tropical Rainfall Measuring Mission (TRMM) 85-GHz passive MW image for Hurricane Wilma on 19 October 2005. Insets show corresponding IRW images using (top right) BD-curve enhancement and (top left) matched IR image with IRW-Max enhancement. TRMM overpass was at 1740 UTC; IR-WV and IRW images were at 1745 UTC. [The MW image was obtained from the Naval Research Laboratory at Monterey, CA (NRL—Monterey), tropical cyclone Web site archive.] Units of IR-WV and two IRW image enhancements are in °C.

While the IRWV imagery cannot “see” beneath the clouds, as with the MW imagery, it highlights the areas of intense convection that can enable the TC forecaster to monitor their evolution over time, especially during longer MW data gaps. The IRBD displays a more uniform CDO cloud-top region, while a double-eyewall structure is noted in both the MW and the IRWV imagery. In fact, putting this imagery into motion (sequential animations) can reveal organization trends in convective structures such as eyewall replacement cycles that can infer intensity changes.

b. Quantitative analysis: TC intensity estimation and prediction

As well as improving the TC position estimation process currently employed in the ADT, the IRWV digital information can be used to test a new predictive intensity element by introducing IRWV parameter(s) to the different linear regression-based schemes that ADT employs for TC intensity determination.

An extensive linear regression analysis was performed on a sample of derived IRWV data to identify any parameters that possess a significant correlation with concurrent in situ aircraft reconnaissance measurements of TC intensity. To assess the intensity forecast potential of the derived IRWV data, 6-h time lag increments between 0 and 24 h were also examined.

The data sample for this study consists of a set of 50 Atlantic basin TCs of varying intensity and duration from between 1999 and 2006. Examination of all derived IRWV images for each storm, at the current analysis time, as well as the four additional time lag increments, was conducted within an approximate 1.25° (136 km) radius from the ADT automated storm center position valid at the time of the image. The analysis was conducted over the entire life cycle of each storm from formation to land interaction and/or dissipation. The image time (plus the lag time, if applicable) was constrained to be within 1 h of the aircraft reconnaissance time to be considered for a validation match. The 0.5° Tb bins of the IR minus WV temperature differences between −10° and +10°C were tallied. IRWV-derived parameters such as the average IRWV difference, the total region and “cloud only” histogram bin counts, and several multi-Tb bin groupings were investigated to identify the most significant correlations to the TC intensity measurements.

A final investigation was conducted to assess the IRWV and TC intensity correlation during instances when the ADT scene type was classified as a CDO or embedded center to ascertain the potential improvement to the ADT intensity estimation process during these especially difficult situations.

Table 1 presents the derived correlation coefficients of the most statistically relevant analysis parameters for each of the five individual 6-h time lag periods investigated over the entire life cycle of the TC. Table 2 repeats Table 1 but includes only ADT CDO and embedded center scene types during the formation and mature stages of the TC life cycle.

Table 1.

Correlation coefficients between the satellite image temperature counts/values and coincident aircraft reconnaissance MSLP measurements for the average IRW cloud-top brightness temperature measurement (first row) and combined cloud-top brightness temperature and cloud symmetry value (second row) between 0 and 136 km from the storm center position, the derived pixel count of the IR-WV channel brightness temperature difference <0°C (third row), and for all three components combined through linear regression (fourth row) for the five 6-h lag periods between 0 and 24 h, inclusive.

Correlation coefficients between the satellite image temperature counts/values and coincident aircraft reconnaissance MSLP measurements for the average IRW cloud-top brightness temperature measurement (first row) and combined cloud-top brightness temperature and cloud symmetry value (second row) between 0 and 136 km from the storm center position, the derived pixel count of the IR-WV channel brightness temperature difference <0°C (third row), and for all three components combined through linear regression (fourth row) for the five 6-h lag periods between 0 and 24 h, inclusive.
Correlation coefficients between the satellite image temperature counts/values and coincident aircraft reconnaissance MSLP measurements for the average IRW cloud-top brightness temperature measurement (first row) and combined cloud-top brightness temperature and cloud symmetry value (second row) between 0 and 136 km from the storm center position, the derived pixel count of the IR-WV channel brightness temperature difference <0°C (third row), and for all three components combined through linear regression (fourth row) for the five 6-h lag periods between 0 and 24 h, inclusive.
Table 2.

Same parameters as in Table 1 but only for ADT CDO–embedded center scene types during the formation and mature stages (up to and including the maximum intensity) of the TC life cycle.

Same parameters as in Table 1 but only for ADT CDO–embedded center scene types during the formation and mature stages (up to and including the maximum intensity) of the TC life cycle.
Same parameters as in Table 1 but only for ADT CDO–embedded center scene types during the formation and mature stages (up to and including the maximum intensity) of the TC life cycle.

The first two rows in Tables 1 and 2 contain elements relevant to the current operation of the ADT when various regression equations are utilized to derive the current TC intensity estimate. The first row features the correlation coefficients for the average cloud region IRW channel Tb (aveTbIR) value over each of the 6-h lag periods. The aveTbIR value closely mimics the environmental temperature parameter historically used within both the subjective Dvorak technique and objective satellite-based TC intensity algorithms, including the ADT. The second row contains the correlation coefficient values using the aveTbIR value and the ADT cloud region “cloud symmetry” (SymCloud) parameter, which measures the IR temperature difference between opposing sections within the ADT cloud-top analysis region. The aveTbIR and SymCloud values are currently utilized in the ADT intensity estimation regression equations for eye and CDO scene types.

The third row contains the correlation coefficient values for the total pixel count of all negative IRWV differences (TbIR-WV), which possess the highest correlation values for all of the potential IRWV parameters noted earlier. Finally, the fourth row displays the correlation coefficients for the combined linear regression analysis of the aveTbIR, SymCloud, and the TbIR-WV values to assess the potential impacts of the IRWV field on the current ADT regression analysis for the entire TC life cycle, and the subset analysis period containing the CDO–embedded center scene types during the development and mature stages only.

As shown in Tables 1 and 2, the TbIR-WV displays increased correlation with aircraft reconnaissance measurements over most time lag periods as compared to the individual aveTbIR values and combined aveTbIR + SymCloud values, especially in the CDO scene type cases. This demonstrates the skill of the IRWV methodology compared with the more traditional and current methodologies for relating satellite-measured environmental parameters to TC intensity and the potential benefit to adding the IRWV information to the current analysis parameters used in the ADT regression equations.

The most interesting aspects to note in Tables 1 and 2 are the correlation increases of the TbIR-WV parameter with aircraft reconnaissance as the lag period increases, while the more traditional aveTbIR + SymCloud values possess no corresponding correlation coefficient increase over the same lag periods. The highest correlation coefficient values occur for the TbIR-WV parameter in the 12- and/or 24-h lag periods. These results echo similar findings by Cecil and Zipser (1999) in their study of TC intensity and satellite-based indicators on inner convection using 85-GHz brightness temperature values.

The IRWV technique will locate and measure the intensity and areal coverage of convective overshooting tops. As the convective intensity decreases, the magnitude and coverage of the IRWV signal will correspondingly decrease while the IRW signature may remain after the convection has dissipated. The generated cirrus, either from the convective anvil or longer-lived thin tropopause cirrus (Garrett et al. 2004, 2005), will remain beyond the life of the convective burst and disperse around the central region of the TC, as observed in the IRW imagery as a CDO. Following that rationale, a peak in the IRWV correlation coefficient values would be expected due to the finite nature of the convective burst as opposed to the more persistent nature of the residual cirrus. The IRWV data provide a more decisive predictive element for TC intensity forecasting than is achieved with the aveTb value alone. A plot of the negative IRWV counts versus the aircraft reconnaissance measured MSLP is provided in Fig. 5 to illustrate the strongly linear correlation between the two parameters for the 12-h lag case.

Fig. 5.

Scatterplot showing the correlation between aircraft reconnaissance MSLP measurements and all negative IR-WV channel difference counts from images 12 h prior to reconnaissance measurements for over 3200 individual comparisons between 1999 and 2006 during the Atlantic tropical cyclone seasons. Red line indicates the linear regression best fit. Total counts are within 136 km of the storm center.

Fig. 5.

Scatterplot showing the correlation between aircraft reconnaissance MSLP measurements and all negative IR-WV channel difference counts from images 12 h prior to reconnaissance measurements for over 3200 individual comparisons between 1999 and 2006 during the Atlantic tropical cyclone seasons. Red line indicates the linear regression best fit. Total counts are within 136 km of the storm center.

Figure 6 displays time series plots of the negative IRWV (red lines) and aircraft reconnaissance MSLP measurements (black lines) for four Atlantic basin hurricanes used in the case study set. The count values were smoothed using a five-point running average to reduce the noise signal in the time series plots. In general, a decrease in pressure is preceded by about 12–24 h by a corresponding increase in the negative IRWV count values. The increases can be subtle, as with Ivan (2004), or quite dramatic, as with Keith (2000) and Wilma (2005). Large decreases in the bin count values, such as with Keith near 1800 UTC 3 October or Wilma near 2100 UTC 21 October can be attributed to land interaction and a corresponding decrease in storm organization. Other marked TC intensity changes in the earlier stages of the storm life cycle can be directly attributed to diurnal TC variations or large-scale environmental effects (trough interactions or shear). Corresponding changes in the negative IRWV count values are noted in the graphs during these early stages, but are typically less pronounced and shorter in duration than those observed during the more convective storm-scale events preceding the drastic increases–decreases in MSLP (Katrina on 27–28 August, Keith on 30 September–1 October, and Wilma on 18–19 October). In these cases the IRWV signal clearly marked the onset of a significant intensification period. The predictive quality of the IRWV signal could provide additional satellite-based guidance to operational TC forecasters when predicting short-term intensity changes.

Fig. 6.

Time series plots of all negative IR-WV difference counts (red line) vs aircraft reconnaissance MSLP measurements (black line) for four Atlantic hurricanes. Red lines have been smoothed using a five-point running-average scheme.

Fig. 6.

Time series plots of all negative IR-WV difference counts (red line) vs aircraft reconnaissance MSLP measurements (black line) for four Atlantic hurricanes. Red lines have been smoothed using a five-point running-average scheme.

This study focused on demonstrating the derived IRWV product using GOES satellite data. This was done for two primary reasons. First and most obvious, the GOES viewing domain contains a significant number of reconnaissance aircraft measurements of TC intensity necessary for method validation (e.g., the North Atlantic and, very infrequently, the eastern North Pacific).

The second reason stems from the shortcomings of certain currently operational geostationary weather satellites regarding the calibration Tb values obtained from the WV and IRW channels at colder Tb ranges. Satellite Tb values are typically derived using a conversion from imager counts (value from 0 to 1023) to a radiance and finally temperature (Tb). The counts to Tb conversions are not linear, typically having larger Tb resolutions at higher count values (lower Tb) for most current geostationary satellite infrared channel calibration tables/equations. The GOES satellites are reversed, however, with lower count values corresponding to lower Tb values. For example, at −70°C (count value of about 65), the GOES-12 IRW and WV resolutions are both approximately 0.5°C. The Japanese Multifunctional Transport Satellite-1R (MTSAT-1R), like the GOES satellites, possesses an IRW channel Tb resolution of approximately 0.3°C at −70°C (count value about 998; see Table 3). The WV channel temperature resolution, however, is significantly larger than the IRW channel near −70°C, around 8°C, since the coldest WV temperature values are at the extreme upper edge of the count to brightness temperature conversion table (count value near 1022) in that Tb range, as shown in Table 3. Derived IRWV temperature magnitudes using MTSAT-1R WV channel values colder than about −72°C will be suspect since the Tb resolution between counts 1022 and 1023 is about 80°C, thus impacting the precision of the computed IRWV; however, there will still be valid qualitative information offered from the data. Other current operational satellites, such as the Meteosat satellites, also possess such a limitation in the WV channel calibration. GOES Imager calibration information and conversion lookup tables are available from the NOAA Office of Satellite Operations Web site (http://www.oso.noaa.gov/goes/goes-calibration/gvar-conversion.htm). MTSAT calibration information and conversion lookup tables are available from the Japan Meteorological Agency (JMA) MTSAT-1R operational information Web site (http://mscweb.kishou.go.jp/operation). Meteosat calibration information can be found at the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) calibration Web site (http://www.eumetsat.int/home/Main/Access_to_Data/Meteosat_Meteorological_Products/Calibration/index.htm).

Table 3.

Data calibration table for MTSAT-1R satellite. Digital number–temperature relationship only shown for the top nine digital number values for the IRW and WV channels. (Table data obtained from the JMA’s Meteorological Satellite Center (MSC) Web site; see the “operational information” section: http://mscweb.kishou.go.jp/operation/calibration/mt1r/HRIT/mt1r_hrit.htm)

Data calibration table for MTSAT-1R satellite. Digital number–temperature relationship only shown for the top nine digital number values for the IRW and WV channels. (Table data obtained from the JMA’s Meteorological Satellite Center (MSC) Web site; see the “operational information” section: http://mscweb.kishou.go.jp/operation/calibration/mt1r/HRIT/mt1r_hrit.htm)
Data calibration table for MTSAT-1R satellite. Digital number–temperature relationship only shown for the top nine digital number values for the IRW and WV channels. (Table data obtained from the JMA’s Meteorological Satellite Center (MSC) Web site; see the “operational information” section: http://mscweb.kishou.go.jp/operation/calibration/mt1r/HRIT/mt1r_hrit.htm)

Therefore, care must be taken in universally applying the IRWV technique. Figures 7 and 8 exhibit the derived IRWV fields for two west Pacific storms. It should be noted that the MTSAT IRWV differences are typically much larger than those obtained with the GOES satellites, due to the deficiencies with the WV calibration discussed previously. The derived IRWV field using MTSAT-1R data still possesses a striking similarity to passive MW imagery, when properly enhanced. Figure 7 shows a derived IRWV field for Super Typhoon 02W (Yutu) during May 2007 as compared to a TerraAqua 89-GHz MW image. Figure 8 compares the derived IRWV image with an enhanced IRW image from MTSAT-1R during Super Typhoon 24W (Durian) during November 2006. The IR image exhibits a CDO, which obscures the storm center under the cirrus shield. The derived IRWV image, however, displays the storm center surrounded by a partial ring of convection. The position obtained by the ADT automated center determination scheme, using the IRWV derived image (the automated storm center determination scheme was unable to identify a storm center using the IRW image for this case), corresponds very well with the Joint Typhoon Warning Center’s (JTWC) final best-track position. With proper adjustment, the derived MTSAT IRWV fields can still be utilized by TC forecasters or the ADT to assist in the determination of the storm center position.

Fig. 7.

As in Fig. 4 but for a comparison between (left) a TerraAqua 89-Ghz MW image (1625 UTC) and (right) a MTSAT-1R-derived IR-WV image (1600 UTC) from Super Typhoon 02W (Yutu) on 17 May 2007. (The MW image was obtained from the NRL—Monterey tropical cyclone Web site archive.)

Fig. 7.

As in Fig. 4 but for a comparison between (left) a TerraAqua 89-Ghz MW image (1625 UTC) and (right) a MTSAT-1R-derived IR-WV image (1600 UTC) from Super Typhoon 02W (Yutu) on 17 May 2007. (The MW image was obtained from the NRL—Monterey tropical cyclone Web site archive.)

Fig. 8.

As in Fig. 3 (without radar image) but for an image of Typhoon 24W (Durian) at 2233 UTC 30 Nov 2006. Official JTWC final BT position (magenta), interpolated JTWC real-time forecast position (cyan), and IR-WV image (blue) location using the ADT autocentering algorithm are shown using colored squares.

Fig. 8.

As in Fig. 3 (without radar image) but for an image of Typhoon 24W (Durian) at 2233 UTC 30 Nov 2006. Official JTWC final BT position (magenta), interpolated JTWC real-time forecast position (cyan), and IR-WV image (blue) location using the ADT autocentering algorithm are shown using colored squares.

4. Potential impact on the advanced Dvorak technique

In the previous section, examples were used to illustrate the potential use of the IRWV information in objectively determined storm analysis schemes. Specific use of the IRWV data within the ADT algorithm is only briefly hypothesized in this section and will be further explored in future research efforts.

As has been shown previously, the correlation improvement realized with the inclusion of the IRWV information along with the specified satellite variables currently used by the ADT to estimate TC intensity during CDO situations is significant. The CDO and embedded center scene types are particularly difficult situations for the ADT to analyze using IR imagery alone. The potential impact of the IRWV represents a significant upgrade in the methodology used by the ADT to determine the TC intensities for these situations.

Another potential impact of the IRWV data is on the ability to provide predictive TC intensity change information for short-term outlooks. As discussed previously, a significant increase in correlation between the IRWV information and TC intensity has been identified for lag periods greater than 12 h. This information could be used by the ADT to foreshadow possible significant rapid intensification periods, which are of great importance to TC forecasters.

In addition to improving the ADT intensity estimations, the derived IRWV imagery could perhaps most plausibly be used to improve the storm center determination process within the ADT. The following discussion will focus specifically on the potential integration of the derived IRWV digital information into the ADT to augment the current methodology.

As mentioned previously, the automated storm determination scheme utilizes a spiral-centering and ring-fitting analysis of the IRW cloud-top temperature field. This technique has proven to be quite useful and reliable in situations where a defined spiral-banding feature and/or an eye-type central feature exist. When the storm is lacking one of these characteristics, the ADT autocentering technique will employ the initial-guess forecast interpolation as the final location for the ADT intensity estimation process. Forecast interpolation can provide an adequate position estimate in certain situations; however, it can quickly become less reliable in situations where the storm is accelerating or decelerating and/or moving in a nonlinear path.

To determine whether the use of the IRWV information could positively impact the storm center determination process within the ADT in the CDO situations, ADT autocentered position estimates, using either the IRW imagery or the IRWV imagery, were derived and compared against interpolated NHC and JTWC final best-track (BT) positions for the Atlantic and west Pacific basins, respectively. The BT positions were interpolated from the 6-h temporal resolution provided in the BT files to 15-min resolution using a polynomial interpolation algorithm.

Two separate tests were conducted. The first, shown in Table 4, compared all available position estimates during ADT objectively determined CDO scene types when both the IRW and IRWV imagery deduced a center position using the SCRF routines. The second test, shown in Table 5, only compared storm positions when the ADT determined a SCRF-derived position in the IRWV imagery but was unable to derive such a position using the IRW imagery (thus utilizing the interpolated forecast position). The final ADT storm center positions are also compared to the real-time Operational Forecast Center (OFC) forecast positions (provided by NHC or JTWC), which are also interpolated to the image time using the same polynomial interpolation scheme, for both the Atlantic and west Pacific basins.

Table 4.

Latitude and longitude distance (in °) error statistics for derived North Atlantic and west Pacific automated storm center positions (using the ADT autocentering technique) vs interpolated BT positions for ADT CDO scene type cases only. Statistics are presented for positions derived using IR-WV and IRW imagery when both positions differed from the initial OFC interpolated forecast position first guess (ADT default). AAE is the average absolute error.

Latitude and longitude distance (in °) error statistics for derived North Atlantic and west Pacific automated storm center positions (using the ADT autocentering technique) vs interpolated BT positions for ADT CDO scene type cases only. Statistics are presented for positions derived using IR-WV and IRW imagery when both positions differed from the initial OFC interpolated forecast position first guess (ADT default). AAE is the average absolute error.
Latitude and longitude distance (in °) error statistics for derived North Atlantic and west Pacific automated storm center positions (using the ADT autocentering technique) vs interpolated BT positions for ADT CDO scene type cases only. Statistics are presented for positions derived using IR-WV and IRW imagery when both positions differed from the initial OFC interpolated forecast position first guess (ADT default). AAE is the average absolute error.
Table 5.

Latitude and longitude error statistics for derived North Atlantic and west Pacific automated storm center positions (using the ADT autocentering technique) vs interpolated BT positions for ADT CDO scene type cases only. Statistics are presented for positions derived using only the derived IR-WV when only the IR-WV position differed from the initial OFC interpolated forecast position first guess (the autocentering using the IRW image resulted in the default OFC position).

Latitude and longitude error statistics for derived North Atlantic and west Pacific automated storm center positions (using the ADT autocentering technique) vs interpolated BT positions for ADT CDO scene type cases only. Statistics are presented for positions derived using only the derived IR-WV when only the IR-WV position differed from the initial OFC interpolated forecast position first guess (the autocentering using the IRW image resulted in the default OFC position).
Latitude and longitude error statistics for derived North Atlantic and west Pacific automated storm center positions (using the ADT autocentering technique) vs interpolated BT positions for ADT CDO scene type cases only. Statistics are presented for positions derived using only the derived IR-WV when only the IR-WV position differed from the initial OFC interpolated forecast position first guess (the autocentering using the IRW image resulted in the default OFC position).

As shown in Table 4, the automated position errors using the IRWV imagery are superior to using the IRW imagery alone in both the Atlantic and west Pacific basins for ADT objectively determined CDO scene types when both the IRWV and IRW automated storm center determination methods successfully derived a position using the SC and/or RF methods. It is interesting to note that the computed statistical errors for the IRWV and the OFC position estimates were comparable, highlighting the difficulty in determining a TC storm center for both automated, objective computer-based algorithms and manual selections by experienced TC forecasters during CDO situations.

In those situations where the ADT objective auto-centering algorithm was able to discern a center position in the IRWV imagery using the SC and/or RF methodology, but was unable to do so using IRW imagery alone, the IRWV storm center positions were found to be about 0.05 to 0.10 degrees inferior to using the OFC forecast alone (Table 5). While use of the OFC forecasts provides the best statistical center position estimate overall, use of the IRWV imagery can certainly provide valuable information for the auto-centering methodology utilized in the ADT during certain situations, as discussed previously. Further investigation of how to identify these situations objectively will be explored to optimize the use of IRWV data within the ADT algorithm.

5. Conclusions

In this study we have demonstrated the potential utilization of a satellite-based, multispectral differencing scheme in tropical cyclone applications. The approach employs differenced IR and WV Tbs to depict intense (overshooting) convection, and shows promise for use in TC position and intensity monitoring. The IRWV imagery is easily derived and available from all operational meteorological satellites. The IRWV-derived imagery is being produced experimentally at the University of Wisconsin—Madison (UW) Cooperative Institute for Meteorological Satellite Studies (CIMSS) and can be viewed in real time, along with time series plots of the IRWV counts, at the CIMSS Tropical Cyclone Web site (http://cimss.ssec.wisc.edu/tropic2).

Methods for applying the IRWV information quantitatively into automated TC intensity estimation algorithms, such as the ADT, are currently under investigation. Continued research into exploiting the IRWV data for use in TC intensity forecasting is encouraged, since very little improvement in this field has occurred over the past 20 years.

Acknowledgments

The authors wish to thank the two anonymous reviewers and the following people for their invaluable discussions during the development of this study: Timothy Schmit (NOAA/STAR), Dave Santek (UW/SSEC), Kris Bedka (UW/CIMSS), and Jeff Hawkins (NRL—Monterey).

REFERENCES

REFERENCES
Brewer
,
A. W.
,
1949
:
Evidence for a world circulation provided by the measurements of helium and water vapor distribution in the stratosphere.
Quart. J. Roy. Meteor. Soc.
,
75
,
351
363
.
Cecil
,
D. J.
, and
E. J.
Zipser
,
1999
:
Relationships between tropical cyclone intensity and satellite-based indicators of inner-core convection: 85-Ghz ice-scattering signature and lightning.
Mon. Wea. Rev.
,
127
,
103
123
.
Dvorak
,
V.
,
1984
:
Tropical cyclone intensity analysis using satellite data.
NOAA Tech. Rep. NESDIS11, 47 pp. [Available from NOAA/NESDIS, 5200 Auth Rd., Washington, DC 20233]
.
Fritz
,
S.
, and
I.
Laszlo
,
1993
:
Detection of water vapor in the stratosphere over very high clouds in the tropics.
J. Geophys. Res.
,
98
, (
D12
).
22959
22967
.
Garrett
,
T. J.
,
A. J.
Hymsfield
,
M. J.
McGill
,
B. A.
Ridley
,
D. G.
Baumgardner
,
T. P.
Bui
, and
C. R.
Webster
,
2004
:
Convective generation of cirrus near the tropopause.
J. Geophys. Res.
,
109
,
D21203
.
doi:10.1029/2004JD004952
.
Garrett
,
T. J.
, and
Coauthors
,
2005
:
Evolution of a Florida cirrus anvil.
J. Geophys. Res.
,
62
,
2352
2372
.
Grosvenor
,
D. P.
,
R. W.
Choularton
,
H.
Coe
, and
G.
Held
,
2007
:
A study of the effect of overshooting deep convection on the water content of the TLL and lower stratosphere from cloud resolving model simulations.
Atmos. Chem. Phys. Discuss.
,
7
,
7277
7346
.
Hawkins
,
J. D.
,
T. F.
Lee
,
K.
Richardson
,
C.
Sampson
,
F. J.
Turk
, and
J. E.
Kent
,
2001
:
Satellite multisensor tropical cyclone structure monitoring.
Bull. Amer. Meteor. Soc.
,
82
,
567
578
.
Hawkins
,
J. D.
,
F. J.
Turk
,
T. F.
Lee
, and
K.
Richardson
,
2008
:
Observations of tropical cyclones with the SSMIS.
IEEE Trans. Geosci. Remote Sens.
,
46
,
901
912
.
Kley
,
D.
,
A.
Schmeltekopf
,
K.
Kelly
,
R.
Winkler
,
T.
Thompson
, and
M.
McFarland
,
1982
:
Transport of water vapour through the tropical tropopause.
Geophys. Res. Lett.
,
9
,
617
662
.
Lee
,
T. F.
,
F. J.
Turk
,
J.
Hawkins
, and
K.
Richardson
,
2002
:
Interpretation of TRMM TMI images of tropical cyclones.
Earth Interactions
,
6
.
[Available online at http://EarthInteractions.org]
.
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 infrared satellite imagery.
Wea. Forecasting
,
22
,
287
298
.
Ottenbacher
,
A.
, and
J.
Schmetz
,
1994
:
Relationship of collocated cloud radiances in the METEOSAT IR and WV channel.
Passive Infrared Remote Sensing of Clouds and the Atmosphere II, D. K. Lynch, Ed., International Society for Optical Engineering (SPIE Proceedings, Vol. 2309), 45–51
.
Rosenlof
,
K. H.
,
2003
:
How water enters the stratosphere.
Science
,
302
,
1691
1692
.
Schmetz
,
J.
,
S. A.
Tjemkes
,
M.
Gube
, and
L.
van de Berg
,
1997
:
Monitoring deep convection and convective overshooting with METEOSAT.
Adv. Space Res.
,
19
,
433
441
.
Steranka
,
J.
,
E. B.
Rodgers
, and
R. C.
Gentry
,
1986
:
The relationship between satellite measured convective bursts and tropical cyclone intensification.
Mon. Wea. Rev.
,
114
,
1539
1546
.
Velden
,
C. S.
, and
T. L.
Olander
,
1998
:
Bispectral satellite technique for delineating intense convection: Applications to tropical cyclones.
Preprints, Ninth Conf. on Satellite Meteorology and Oceanography, Paris, France, Amer. Meteor. Soc., 458–461
.
Zehr
,
R.
,
1989
:
Improving objective satellite estimates of tropical cyclone intensity.
Extended Abstracts, 18th Conf. on Hurricanes and Tropical Meteorology, San Diego, CA, Amer. Meteor. Soc., J25–J28
.

Footnotes

Corresponding author address: Tim Olander, Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin—Madison, 1225 West Dayton St., Madison, WI 53706. Email: timo@ssec.wisc.edu