The Impact of Multispectral GOES-8 Wind Information on Atlantic Tropical Cyclone Track Forecasts in 1995. Part I: Dataset Methodology, Description, and Case Analysis

Christopher S. Velden Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin—Madison, Madison, Wisconsin

Search for other papers by Christopher S. Velden in
Current site
Google Scholar
PubMed
Close
,
Timothy L. Olander Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin—Madison, Madison, Wisconsin

Search for other papers by Timothy L. Olander in
Current site
Google Scholar
PubMed
Close
, and
Steve Wanzong Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin—Madison, Madison, Wisconsin

Search for other papers by Steve Wanzong in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Satellite-based remote sensing has long been recognized as an important method to reconnoiter oceanic tropical cyclones due to the scarcity of in situ observations. Beyond the standard qualitative applications offered by imagery, algorithms are being developed to process the information-wealthy imagery into quantitative parameters necessary to positively impact objective analyses on which numerical track predictions are initialized. Techniques developed at the University of Wisconsin Cooperative Institute for Meteorological Satellite Studies enable the automated extraction of displacement vectors from animated imagery featuring sequential geostationary satellite multispectral observations of clouds and water vapor. Recent upgrades to these algorithms and a focused processing strategy directed toward optimizing the retrieved wind vector coverage are discussed. In combination with advanced sensing technology afforded by the National Oceanic and Atmospheric Administration’s latest generation of geostationary meteorological satellites, GOES-8, superior vector yield and quality are being realized.

In this set of two papers, datasets produced during the 1995 Atlantic hurricane season are examined for their impact on tropical cyclone analyses and numerical track forecasts. In Part I, the wind retrieval methodology and data characteristics are described, along with a brief discussion of the tropical cyclones selected for study. Part II addresses the input of the GOES-8 wind information into a global data assimilation system, and the resultant impact on numerical track predictions.

Corresponding author address: Christopher Velden, UW CIMSS, 1225 West Dayton St., Madison, WI 53706.

Email: chrisv@ssec.wisc.edu

Abstract

Satellite-based remote sensing has long been recognized as an important method to reconnoiter oceanic tropical cyclones due to the scarcity of in situ observations. Beyond the standard qualitative applications offered by imagery, algorithms are being developed to process the information-wealthy imagery into quantitative parameters necessary to positively impact objective analyses on which numerical track predictions are initialized. Techniques developed at the University of Wisconsin Cooperative Institute for Meteorological Satellite Studies enable the automated extraction of displacement vectors from animated imagery featuring sequential geostationary satellite multispectral observations of clouds and water vapor. Recent upgrades to these algorithms and a focused processing strategy directed toward optimizing the retrieved wind vector coverage are discussed. In combination with advanced sensing technology afforded by the National Oceanic and Atmospheric Administration’s latest generation of geostationary meteorological satellites, GOES-8, superior vector yield and quality are being realized.

In this set of two papers, datasets produced during the 1995 Atlantic hurricane season are examined for their impact on tropical cyclone analyses and numerical track forecasts. In Part I, the wind retrieval methodology and data characteristics are described, along with a brief discussion of the tropical cyclones selected for study. Part II addresses the input of the GOES-8 wind information into a global data assimilation system, and the resultant impact on numerical track predictions.

Corresponding author address: Christopher Velden, UW CIMSS, 1225 West Dayton St., Madison, WI 53706.

Email: chrisv@ssec.wisc.edu

1. Introduction

Recent advances in numerical weather prediction (NWP) have led to increasing reliance on model guidance products by the forecast community. In particular, official tropical cyclone (TC) track forecasts issued by the National Centers for Environmental Prediction (NCEP) Tropical Prediction Center (TPC) are influenced by model forecast information from a myriad of numerical prediction systems. While NWP has advanced through increasing attention to model resolution, physics, and parameterizations, the fact remains that the precise specification of initial conditions, principally the wind field, is essential to accurate numerical TC track forecasts.

Given the oceanic nature of TCs, in situ tropospheric observations are usually scarce. In the Atlantic, air force reconnaissance aircraft flight-level information, and occasional dropwindsonde missions (Franklin and DeMaria 1992) conducted by the National Oceanic and Atmospheric Administration (NOAA) Hurricane Research Division (HRD) are the only conventional means currently available for providing a detailed description of the environmental wind field. While these are important sources of information, they are spatially limited. The reconnaissance data is chiefly restricted to a single-level in the near-storm environment, and to date the NOAA HRD missions rarely provide observations above 350 mb. Therefore, an opportunity exists for remotely sensed observations from space-borne meteorological platforms in geostationary orbit to augment the existing sources of wind information available for analysis.

The prolific Atlantic hurricane season of 1995 offered an opportunity to examine measurements from the newest generation of NOAA’s Geostationary Operational Environmental Satellites, GOES-8, which was launched in 1994 and positioned to cover the western Atlantic region (Menzel and Purdom 1994). This generation represents a significant improvement in observing and sensing capabilities over previous GOES platforms (e.g., upgraded instrument performance, radiometers with higher spatial resolution and radiometric sensitivity, and increased sampling frequency and capability from independent imager and sounder components). Demonstrational datasets consisting of wind vectors derived from sequential GOES-8 multispectral imagery were produced in quasi-real time by the University of Wisconsin Cooperative Institute for Meteorological Satellite Studies (CIMSS) during Atlantic TC events in 1995. These datasets were provided to TPC for their qualitative analysis, and delivered to numerical modeling centers for assimilation and forecast impact evaluation. Part I of this paper describes the datasets and their applications. Part II (Goerss et al. 1998) assesses the value of the satellite-derived wind information on numerical TC track forecasts through a series of data assimilation experiments.

2. Background

Wind extraction algorithms developed at CIMSS have progressed to the stage of operational implementation by NOAA (Merrill et al. 1991; Nieman et al. 1996). These algorithms are fully automated and calculate vector displacements from cloud and water vapor motions in sequential satellite imagery at high spatial resolution. Attributable to the advances in computing power, algorithm automation, and communications, high-density vector fields derived from multiple spectral channels are achievable in a time frame and frequency commensurate with operational requirements.

In addition to traditional cloud-motion vectors processed from infrared (IR) imagery, upper-tropospheric wind information can be derived from water vapor imagery to fill in data voids between cloud systems. GOES-8 contains three quasi-independent water vapor absorption bands; the 6.7-μm channel is housed on the imager, while the 7.0- and 7.3-μm channels are housed on the sounder (Menzel and Purdom 1994). These three channels can be employed collectively to produce water vapor wind vectors (WVWV) that represent the mid- and upper-tropospheric flow (200–600 mb) in cloud-free regions. Additional information on the coverage, characteristics, and accuracy of these measurements can be found in Velden et al. (1997) (WVWV) and Nieman et al. (1997) (IR vectors).

A final component of the multispectral dataset derives from tracking cloud elements in high-resolution visible imagery. Specifically, cumuliform cloud tracers provide a good approximation to the lower-tropospheric flow around TCs. While the thick storm cirrus canopy often prohibits lower-level cloud tracking in the inner vortex, good coverage is usually achievable in the cirrus-free sectors of the outer vortex and near environment. Research is under way to objectively identify and track cumulus elements visible under thin cirrus.

An example of the GOES-8 vector coverage attainable over a hurricane environment employing the high-density multispectral approach is illustrated in Fig. 1. Winds derived from each of the spectral bands mentioned above are plotted with respect to Hurricane Felix (August 1995). This example illustrates the spatial coverage that can be obtained, and the general height distribution of the wind vectors processed from the multiple spectral channels (Fig. 1f). The majority of the upper-level winds are derived from the 6.7-μm WVWV. The middle-level observations are mainly dual-channel sounder WVWV, and the lower-level winds are predominantly derived from clouds tracked in the visible channel. During selected Atlantic TCs in 1995, the high-density datasets such as shown in Fig. 1 were produced by CIMSS over a domain covering a large portion of the North Atlantic on a 6-h basis. The datasets were processed to demonstrate operational feasibility, and therefore were accomplished in a time frame commensurate with real-time operational forecast and NWP data cutoff criteria (within 2 h of synoptic time).

The GOES-8 wind vector plots have proven useful to subjective analyses of storm events and environmental interactions (Velden et al. 1997). The plots were disseminated to NCEP TPC in the latter part of the 1995 season, and also during the 1996 season for subjective evaluation. The response has been positive (TPC 1996, personal communication). Similar satellite-derived wind fields processed in real time by CIMSS over the western North Pacific from the Japanese geostationary satellite (GMS-5) for the Joint Typhoon Warning Center (JTWC) at Guam have also been very well received. Features such as TC outflow characteristics, environmental circulations and shear zones, and outer vortex wind radii (lower levels) are frequently well depicted in the satellite datasets and are being employed to augment conventional observations and model-derived fields in subjective evaluations by the operational analysts. The Australian Bureau of Meteorology is also demonstrating forecast impact using winds derived from GMS-5 (LeMarshall et al. 1996).

Qualitatively, the three-dimensional coverage and spatial coherence achievable from the multispectral vector field is impressive (Fig. 1), especially given the relative scarcity of otherwise available wind observations over these regimes. However, it is imperative that the vector field be of sufficient quality to improve the representation of the TC environmental flow field in subsequent objective analyses in order to realize positive impact on NWP track forecasts. A comprehensive statistical evaluation of vector quality versus collocated rawinsonde reports is given in Velden et al. (1997) and Nieman et al. (1997). Generally, the WVWV and upper-tropospheric IR cloud-tracked winds were found to be of equal quality (root-mean-square differences). WVWV degrade slightly (∼1 m s−1) relative to IR vectors in the middle troposphere, and are more difficult to obtain in dry air masses (Velden et al. 1997). In regards to data impact on NWP, vector quality must be considered not only in an absolute sense, but in relation to model background fields that have improved considerably in recent years. Such is the focus of this two-part paper.

Prior studies have examined the impact of the multispectral vector datasets derived from previous GOES platforms on numerical TC track forecasts (Velden and Goldenberg 1987; Velden et al. 1992). In general, the impacts were modestly positive. However, the numerical analysis and forecast systems employed in those studies have been replaced with more sophisticated, fully three-dimensional analysis and forecast systems that have proven to be superior in terms of TC track forecast guidance. The results presented in Part II of this paper derive from a state-of-the-art assimilation system chosen to take full advantage of the high-density satellite information and maximize the potential forecast impact. This prediction system was developed at the Naval Research Laboratory (NRL) for operational use at the Navy’s Fleet Numerical Meteorology and Oceanography Center (FNMOC), and has shown significant skill in tropical cyclone track forecasting. The NRL–FNMOC model (NOGAPS) is a global assimilation system that contains special tropical cyclone bogussing procedures (Goerss and Jeffries 1994). This system was one of the top performers in terms of operational objective hurricane track predictions during the prolific 1995 Atlantic season (Gross and Lawrence 1996). Further details are provided in Part II.

The NWP impact results presented in Part II are encouraging and are to be considered a benchmark. Evolving and improving assimilation strategies and parameterizations are constantly emerging. For example, the direct assimilation of polar-orbiting satellite radiances has significantly impacted the NCEP global analysis and as a result has improved the model forecast skill scores (Derber and Wu 1996). These upgrades are placing increasing demands on observation quality, and research aimed at refining and advancing satellite data extraction techniques is an ongoing mission of CIMSS.

3. Data processing methodology

The basic processing (wind extraction) philosophy strives toward maximizing the potential from GOES-8 observing capabilities. In this regard, selected multichannel imagery with unique qualities are utilized at the highest possible horizontal resolution to extract tropospheric motions for deriving high density vector fields. The complementary nature of the vectors derived from each individual spectral band allows for a spatially coherent wind field that can benefit both qualitative analysis and data assimilation. Although the dataset vectors are predominant in the upper half of the troposphere, mid- and lower-tropospheric information is also attainable on a more limited basis (Fig. 1). The processing strategy not only focuses on observation density and vertical distribution but also by necessity on the attention to vector quality. This is addressed through the optimization of vector height attribution methodology and quality control procedures as part of dataset post-processing discussed below.

To establish the operational feasibility of the GOES-8 dataset production cycle, the extraction algorithms have been fully automated. During the 1995 Atlantic TC season, datasets were produced under simulated real-time conditions in order to demonstrate the expected operational quality. In 1996, datasets were produced in actual real time to test the capability to meet operational data cutoff constraints (this exercise was highly successful).

a. Basic tracking algorithm description

Over the past several years, CIMSS has upgraded several aspects of the wind extraction algorithm originally reported on by Merrill et al. (1991). The fully automated algorithm is housed within McIDAS-X (UNIX) architecture (Santek et al. 1991). New automated procedures have been introduced that correct image registration inconsistencies within the tracking sequence (Nieman et al. 1997). Tracer selection methodology has been expanded to include a more rigorous filtering of undesirable targets, and optional targeting of bidimensional gradients rather than brightness maxima exclusively. Semitransparent cloud tracers (thin cirrus) derive their heights from an H2O-intercept method, replacing the CO2 slicing technique (Menzel et al. 1983). A statistically based vector speed bias adjustment has been implemented (described later in the text). Increased computing resources have enabled more elaborate pattern matching routines and higher vector density. All of these features are highlighted in more detail in the following text, in terms of their application to wind vector extraction from specific spectral channels.

1) Cloud-tracked winds from the infrared window channel

Target selection begins with a software routine that segments the first of three sequential IR images. Within each segment (typically 60-km cells), bidirectional brightness temperature gradients surrounding each pixel (4-km resolution) are computed. Gradients that exceed empirically determined thresholds undergo a pixel brightness check as a first step to ensure cloud edges are being targeted. All prospective targets also are subject to a spatial-coherence analysis (Coakley and Bretherton 1982). In this scheme, potential targets within the segment are filtered by a two-dimensional clustering algorithm, which is designed to isolate targets that may represent undesirable features such as coastlines or multideck scenes (difficult height diagnosis). This filter economizes the vector extraction algorithm by eliminating these features as prospective targets. This CPU time savings can be used to increase the target attempts within an image segment, and thereby increase the density of acceptable targets that will serve as potential tracers.

Following target selection, the three sequential images are subject to a careful registration check. Registration is a measure of navigation consistency between images, and accurate registration is crucial to eliminate artificial tracking errors. The automated registration quality control employs pattern matching of land features. Initial landmarks are identified by the spatial coherence analysis, and sought in subsequent images using extremely tight correlation thresholds. Mean deviations are computed and if they exceed empirically determined tolerances, registration corrections are invoked (Nieman et al. 1996).

At this point, an initial height determination is calculated for each target. Satellite radiances in the target area are first converted to equivalent blackbody temperatures (Tbb) through the relationship
i1520-0493-126-5-1202-e1
where F is the radiometer spectral response function, B is the Planck blackbody intensity for temperature Te, and I is the intensity of emitted energy at wavelength λ received by the radiometer. To the extent that Tbb represent actual cloud temperatures, they can be matched with a collocated model forecast temperature profile to obtain a crude height assignment for the target. The mean of the coldest 20% of the pixels in the target area is used as the representative Tbb. The assumption of clouds as radiative blackbodies is valid for opaque clouds, but not for semitransparent clouds such as thin cirrus. Direct use of IR window channel Tbb in thin cirrus will result in height assignments that are systematically too low. In a vertically sheared troposphere, this will lead to vector speed biases, even if the features are being correctly tracked.

The above conditions have necessitated a multispectral height assignment approach. This algorithm takes advantage of the 6.7-μm water vapor channel on GOES-8. The H2O-intercept method (Schmetz et al. 1993) is predicated on the fact that the radiances from a single cloud deck for two spectral bands vary linearly with cloud amount. Radiances from the IR and H2O channels are measured and compared to calculated Planck blackbody radiances as a function of cloud-top pressure. Model guess profiles are employed for the necessary radiative transfer calculations. Measured and calculated radiances will agree for clear sky and opaque cloud conditions. The cloud-top height is inferred from the linear extrapolation of radiances onto the calculated curve of opaque cloud radiances [see Fig. 1 of Nieman et al. (1993)], and assigned as the target height. Nieman et al. (1993) found this height assignment methodology to be an adequate replacement for the proven CO2 slicing technique (Menzel et al. 1983), and is employed since the current GOES series does not contain a CO2 absorption channel.

To generate displacement vectors from the initial target image, the automated tracking algorithm examines successive images (normally two) for coherent pattern matches. The tracking metric searches for the minimum in the Tbb sum of squares difference between target and search pixel arrays:
Frd
where F is the tracking function and r(d) is the Tbb difference between target and potential match at displacement distance d. The solution to (2) can then be converted to a velocity. Here, F can be any function with a relative minimum. The function currently employed is
FrdrTdrd
where T is the transpose and the function represents the sum of squares of the Tbb differences. A correlation failure will result if an empirically derived minimum threshold is not achieved, in such instance no displacement vector is derived.

Employing the three successive images, normally 30 min apart for IR tracking, two vector displacements are attempted from each initial target. A successful pattern match between the first and second image becomes the target for the second attempt to find a displacement between images two and three. Both attempts must meet the match criteria above or a correlation failure will result. Successful displacement vector pairs are then subject to consistency checks to eliminate accelerations that exceed empirically determined tolerances.

A final step in the tracking process is an adjustment to the vector speed. The well-documented slow bias in upper-tropospheric cloud-tracked winds is mitigated by incrementing each vector speed by 8%. This adjustment magnitude was determined from ECMWF statistical analyses. The slow bias is primarily confined to winds above 400 mb, and the adjustment is applied only to those vectors.

2) Water vapor–tracked winds

The procedure for deriving WVWV (water vapor wind vector) parallels the basic technique employed for cloud tracking. In the targeting scheme, the gradient thresholds are relaxed to take into account the reduced horizontal resolution of the imagery (8 km) and the inherent weaker gradients in water vapor relative to IR scenes. In addition, in cloud-free regions, the mean Tbb of the target area is taken as the target Tbb to be employed for initial height assignment (in clouds, the coldest 20% is used as with the IR cloud tracking). This is the only means of height assignment based on the radiance information since the H2O-intercept technique cannot be applied due to the water vapor absorption characteristics, which do not allow clear-sky Planck Tbb calculations.

A long-standing concern regarding WVWV is the height assignment. Particularly in cloud-free regions, the radiometric signal from pure water vapor structures is a result of emittance over a finite layer. This can be complicated further by radiance contributions from multiple moist layers (Weldon and Holmes 1991). The challenge is to assign a height level that best represents the motion of the feature (layer) being tracked. The strategy employed in the CIMSS algorithm for WVWV height assignment is a two-step approach.

The first step employs the channel Tbb as noted above. It is important to stress that the water vapor absorption bands available from GOES-8 sense radiation from slightly different parts of the 5–8-μm water vapor absorption band. Sampling the center of the absorption band (around 6.3 μm) yields radiation from the upper levels of the atmosphere (e.g., radiation from below has already been absorbed by the water vapor). Sampling away from the center of the absorption band yields radiation from lower levels of the atmosphere. Figure 2 shows typical clear-sky weighting functions (which denote energy contribution profiles that factor into the determination of pixel Tbb) for the GOES-8 imager and sounder water vapor sensitive bands. The GOES-8 sounder 7.3-μm band senses deeper into the atmosphere than the other bands, as it is farther away from the water vapor absorption band center.

As indicated in Fig. 2, trackable water vapor features are attainable from the three quasi-independent channels within the tropospheric range of 200–600 mb. Peak radiance information emits from a rather shallow layer in most cases when adequate moisture is available. However, the emitting layer can broaden in drier air masses. The 7.3-μm channel is the most difficult to attribute single-level heights due to the broadness of the weighting (atmospheric contribution) function. The second height assignment step attempts to mitigate these difficulties; the initial heights assigned from the Tbb are reassessed as part of an objective analysis outlined in section 3b.

Future research will be aimed at more accurately describing the altitude of WVWV motion in terms of the tropospheric layer being tracked. Incorporation of information from the energy contribution profiles should provide an improved height attribution, and this may be optimally accomplished within numerical assimilation

As with IR cloud tracking, sequential 30-min interval imagery is used to trace features in the 6.7-μm observations. The 7.0- and 7.3-μm images from the sounder, with a horizontal resolution of 10 km, are nominally available at 60-min intervals. See Velden et al. (1997) for further details on WVWV.

3) Cloud-tracked winds from the visible channel

During daylight hours, visible channel data can be utilized to track cloud motions. The advantages of the GOES-8 visible channel are the high horizontal resolution (1 km) and the frequent image sampling (15–30 min normally; higher in special rapid scan schedules). Of particular importance to TC scenarios, the visible channel can depict lower-tropospheric cumuliform tracers in the outer storm vortex region (in areas not covered by opaque cirrus). The outer vortex structure and wind profile are important considerations in storm motion (Fiorino and Elsberry 1989). In our study, datasets produced from the visible channel focus on tracking cloud elements below 600 mb, and within a 15° radius of the storm center.

The automated wind vector extraction processing from the visible imagery closely follows the IR cloud-tracking methodology. Target density parameters are adjusted to allow for the increased resolution of the data (45-km segments). Somewhat tighter constraints are imposed on acceleration checks due to the shorter sampling interval. Brightness temperature thresholds (on the warm end) are relaxed to allow tracers in low light conditions. Initial heights are assigned based on collocated (subsampled) IR Tbb.

b. Postprocessing: Vector height reassessment and quality control

The automated algorithms allow for high-density winds to be processed from multispectral channels in a reasonable time frame. The resulting datasets typically consist of vectors totaling around 20000 per time period. Automated quality control (QC) procedures based on an objective analysis were developed in order to accommodate these data volumes. In addition, the same objective analysis figures in the reassessment of each vector height; the rationale being that the height assignment is the greatest source of vector error. The tracers unquestionably indicate motion in the troposphere, which may not represent motion of air parcels at a given level as we assign it. Rather, especially in regards to water vapor tracked winds, the tracers represent a layer-averaged motion. In reassigning the vector height via the objective analysis, we seek to assign the vector to the level where it best fits other information, including neighboring vectors, rawinsondes, aircraft reports, and numerical model forecast data. In this sense the QC strategy has an element of data assimilation (Hayden and Velden 1991).

In regards to the incorporation of numerical model forecast information into the QC procedure, it is recognized that a trade-off exists in the choice of the allowable influence. If only a small deviation from the model forecast is allowed, potential vector information is reduced. If large deviations are permitted, potential information is increased but so are potential errors. In addition, use of a particular model forecast may bias the height errors to that model, making it difficult to assign observation errors if assimilated by another modeling center. The obvious goal is a QC procedure as independent of model forecasts as possible, and this is the subject of ongoing research. In the meantime, the particular philosophy applied to TC situations is discussed further in section 3c.

The purpose of the postprocessing step is threefold: 1) search for the tropospheric level that best represents the vector motion being traced, 2) edit out vectors that are in obvious error, and 3) provide end users with discretionary vector quality flags. The general method can be summarized as follows (and illustrated in Fig. 3): 1) the multispectral satellite-derived wind vectors with initial assigned heights are objectively analyzed collectively in conjunction with ancillary data and a background field from a numerical forecast model (the CIMSS algorithm currently employs the 12-h forecast fields from the NCEP Aviation model). The three-dimensional objective analysis employed is a recursive filter method described in the appendix. Each vector has the opportunity to influence the analysis and is appended with a quality flag as an initial indicator of its “fit” to the analysis. 2) Vector heights are reassessed based on a variational penalty function (described below) that seeks to minimize vector difference with ancillary information. 3) The vectors with reassigned heights are reanalyzed using the output of the first analysis as the background or first guess. The output of this second analysis provides the final quality flags. Vectors that do not attain an empirically determined quality threshold are edited from the dataset and are not disseminated to end users.

The QC process involves a reassessment of the pressure altitude assigned initially to each vector. The heights are reevaluated and adjusted by minimizing a simple variational penalty function (4), which uses the initial RF analysis and numerical model temperature forecast fields (Fig. 3):
i1520-0493-126-5-1202-e4
where V is velocity, T is temperature, P is pressure, dd is direction, and s is speed. Subscript m refers to a vector measurement, i and j are horizontal dimensions in the analysis, and k is the vertical level. The denominators, F, are weights defined by the operator that may be varied to emphasize the vector velocity, temperature (target Tbb), pressure, direction, and speed terms in the penalty function. Default values for F are 2, 10, 100, 1000, 1000, respectively. The default values have been empirically optimized over several years of application. However, as with the RF tolerance and threshold parameters, the optimal settings can be situation dependent and these weights can be varied by the operator (section 3c). The vertical search for a best fit (and hence, the maximum allowable reassignment) is limited to ±150 mb from the initial assignment. In practice, experience has shown the average reassignment is on the order of 30 mb. The height reassignment may fail if no minimum is found or if the minimum exceeds empirically determined thresholds. In these instances the vector is rejected.

While a particular discrete tropospheric level will not precisely represent a layer-mean motion, our analysis of vector height assignment quality compared with collocated rawinsondes after the above methodology is employed generally indicates placement confidence within 50 mb in most cases. While this represents an improvement in vector height location over traditional methods, further study is clearly needed in this area. For example, it has been discovered that in very dry mid- and upper-tropospheric air masses above low-level (∼700 mb) clouds, the sounder WVWV may actually represent the motion of the cloud tops while being assigned pressure heights much higher based on the energy contributed by the small amounts of water vapor above the clouds. In high vertical wind shear situations, this can lead to a significant slow speed bias. While this scenario is not common in tropical applications, the speed bias is unacceptable in midlatitude applications. A multispectral cloud-mask routine is being developed to identify these situations. In the meantime, checks have been added in the postprocessing algorithm to screen for these situations.

c. Special tuning for TC dataset processing

TCs are highly anomalous meteorological events over data-sparse regions, and their circulation features are often not well defined in global numerical analyses or forecasts (e.g., strong upper-level outflow jets). As discussed above, these forecasts serve as pseudo-observations for the background field in the RF analysis, which determines the QC of the multispectral satellite wind field. Employing the empirically determined default settings for the constraints, weights and thresholds in the RF and variational penalty function, many vectors will be rejected based on large deviations from the background field. This is undesirable, since the objective is to successfully modify the numerical model fields in order to positively impact the subsequent forecasts. Therefore, an effort has been undertaken to reexamine several of the tunable parameters in TC situations in order to optimize the settings in these highly anomalous situations and retain vector information to be passed on to the assimilation systems.

The instrument precision and increased horizontal resolution of imagery from GOES-8 allows higher-density vector fields of higher quality. Considering the multispectral vectors at high-density collectively, the RF analysis benefits through increased influence of observations and neighbor checking, and lessened influence of the numerical forecast. In this regard, data saturation is desirable as long as the data are spatially coherent and reliable.

To take full advantage of this data coverage, the vector field postprocessing algorithm is tuned in the following ways: 1) the objective analysis horizontal resolution (lat/long) is increased to 1° to capture finer-scale features (default is 2°), 2) the model background pseudo-observations are downweighted relative to the satellite measurements by a factor of 2, 3) a tighter fit of each vector to the analysis is demanded due to the abundance of “neighbor checking,” 4) increased allowance (deviation) is given to the velocity term in the height reassignment penalty function (Fυ = 5), 5) vertical coupling constraints in the three-dimensional RF are adjusted to be less strict (more independence for data at individual levels), and 6) the rejection thresholds based on the final quality indicators are adjusted based on vector type and assigned height. The RF tolerance settings (To and n) are not adjusted directly; however, the data fit and gross error allowances are implicitly regulated by the tuning of the above parameters.

The result of this postprocessing strategy is a quality-controlled vector field regulated by an objective analysis with pressure heights assigned to discrete levels based on a blend of radiometric properties and assimilation with ancillary information. Specific analysis operating parameters have been tuned and QC settings have been relaxed to increase the retention of vectors in regions of the TC and its environment that may be deviant from the background guess fields, but in spatial coherence with neighboring reports. This strategy is programmed to operate within a 15° radius of a target TC. The resulting satellite-derived information can lead to an improved depiction of the TC wind environment, and superior numerical track forecasts (see Part II).

4. Case studies

Ultimately, the significance of the processing strategy and tuned data extraction methodology outlined above must be measured in terms of the impact on objective analyses and numerical prognoses of TC tendencies. Previous studies have provided a glimpse into the potential impact these data can have on NWP (Velden et al. 1992; Velden 1996; Velden et al. 1997). In these studies, it was found (for a limited sample) that the satellite-derived wind information can effectively reduce the error of objective track forecasts. These encouraging results provided the stimulation for a more thorough investigation.

The 1995 Atlantic tropical cyclone season was one of the most active in recorded history. The TC cases that were selected during this period for further examination include Tropical Storm Chantal, and Hurricanes Humberto, Iris, and Luis (referred to collectively as CHIL). Datasets were processed at CIMSS in a simulated operational setting and were disseminated to the Naval Research Laboratory in Monterey for model impact evaluation. The results of that impact test are given in Part II. In this section, the TC cases are briefly described, along with the general characteristics of the satellite datasets.

a. Tropical Storm Chantal

Chantal originated from a tropical wave that emerged off of the African coast on 5 July 1995. The system reached tropical storm strength on 14 July as it was moving west-northwest, passing just to the north of Puerto Rico (Fig. 4). The storm then quickly recurved around the western periphery of the Atlantic subtropical ridge before racing to the northeast. The forecast dilemma with Chantal was the recurvature. The official forecasts from the NCEP TPC exhibited a left bias and prompted tropical storm warnings for the Bahamas. However, recurvature was more dramatic than forecast.

Multispectral satellite datasets were derived at 12-h intervals during Chantal from 14–19 July. Included in these datasets were high-density IR cloud-tracked winds, WVWV (imager and sounder), and VIS cloud-tracked winds below 600 mb covering the outer vortex region. An example of the data coverage is shown in Fig. 5. It is important to note at this point that the spatial coverage attainable from the sounder WVWV and low-level VIS vectors is limited. The sounder scans are constrained by other operational considerations to mainly cover the continental United States and adjacent coastal waters of the western Atlantic Ocean. A small sector scan was added to cover an additional segment of the western Atlantic. This coverage extends north to near 50°N, south to near 25°N, and east to near 55°W. Chantal’s track was close to, or within, this coverage during most of the period.

Evident in Fig. 5 is the mid- to upper-level flow immediately north of Chantal, which defines the extent of the westerlies impinging on the outer circulation of Chantal and suggests imminent recurvature to the northeast. Indeed, incorporation of these observations into the NOGAPS resulted in a more rapid (and accurate) recurvature in the numerical forecast tracks relative to the control forecasts run without the satellite data (Part II).

b. Hurricane Humberto

Humberto developed from one of several strong tropical waves that moved off of the coast of Africa in August of 1995. As the disturbance encountered a favorable upper-level wind environment on 22 and 23 August, it quickly intensified to hurricane status. The track of Humberto (Fig. 4) was mainly influenced by a strong midlevel trough over the central Atlantic. This trough steered Humberto northward. A more vigorous midlatitude trough then picked up Humberto and turned it to the northeast. Humberto was briefly affected by outflow from nearby Hurricane Iris, which created a strong upper-level shear environment over the TC and resulted in deintensification. However, the effects of any binary vortex interaction on the motion of Humberto was minimal and track deviations were not readily apparent.

Multispectral satellite datasets were processed beginning on 24 August, and continued through 31 August. The datasets were produced at 6-h intervals, and consisted of IR and water vapor motion vectors (imager and sounder). No winds were derived from the visible channel since the high resolution imagery at 15-min intervals is not available over the mid–Atlantic Ocean from GOES-8. An example of the multispectral vector coverage is shown in Fig. 6. It is important to note that the sounder water vapor motion vector coverage remains upstream of Hurricane Humberto during the period.

During this event, a height assignment problem was discovered in regards to the sounder WVWV. Figure 6 shows vectors in the dry (dark in water vapor image) air mass to the southwest of the midlatitude trough. Several of these vectors were assigned heights of around 500 mb based on the methodology outlined above. However, the NCEP Environmental Modeling Center (EMC) noted a pronounced slow bias in this region upon inspection of analyses that incorporated these vectors. Upon further examination, it became evident that low-level cloud tops near 700 mb were actually being tracked. The extremely dry mid- to upper-tropospheric air mass allowed radiation from these cloud tops to be sensed by the sounder water vapor channels that exhibit energy contribution profiles (Fig. 2) capable of peaking at lower-tropospheric layers under these conditions. Since the cloud tops were relatively bright, the radiometric response was to assume this was midlevel moisture. Therefore the initial heights were assigned near 400 mb. The objective height reassignment did succeed in lowering the heights, but only to near 500 mb. Due to the correlated error [several vectors in the region were (incorrectly) in agreement], these vectors passed the RF analysis constraints despite the strong deviation from the background model field. Since there was relatively strong vertical shear in this instance (wind speed increasing with height), the vector height misplacement resulted in a slow bias. This may have contributed to the mixed forecast results from NOGAPS for Humberto (Part II). While this situation is rare in the Tropics, it is of concern toward the routine use of the sounder water vapor wind vectors in NWP as a general practice. Solutions to this problem are being investigated.

c. Hurricane Iris

Like its predecessor Humberto, Iris also formed from a strong easterly wave. It reached tropical storm strength on 22 August as it moved westward. On 23 August, it was classified as a hurricane. Iris tracked slightly south of west for a few days possibly due to a slight binary interaction with Humberto (Fig. 4). Steering currents ahead of a trough to the northwest of the TC turned Iris abruptly to the north on 27 August. The track became erratic for a few days as Iris interacted strongly with Tropical Storm Karen, which had moved into the region from the southeast. Iris eventually absorbed Karen’s circulation on 3 September. The TC then continued its northward path and eventually was caught up in the westerlies and steered into western Europe as a powerful extratropical system.

Iris was a difficult system to forecast due to the environmental interactions. This was reflected in the NWP guidance. The key environmental factors were the TC–trough interaction, and the influence of the other TCs in the region. The satellite-derived wind vector field shown in Fig. 7 illustrates the complicated tropospheric pattern over the central Atlantic during this period.

Multispectral satellite datasets were derived on a 6-h basis for the entire lifecycle of Iris. However, low-level VIS winds were somewhat limited due to the TC crossing the extreme eastern edge of the high-resolution, 15-min GOES-8 coverage. The storm also traversed the extreme eastern edge of the GOES-8 sounder coverage. Nonetheless, the satellite winds produced significant reductions in track forecast errors from NOGAPS upon assimilation (Part II). Presumably, this was a result of an improved definition of the environmental features and circulations that played a role in Iris’s track.

d. Hurricane Luis

Luis strengthened from a depression into a tropical storm on 29 August and gained hurricane strength late on 30 August. Intensification continued and Luis reached category 4 on the Saffir–Simpson scale on 3 September. The track of Luis was generally smooth and predictable, and this was reflected by low track forecast errors in both the TPC and NWP forecasts. The steering currents were well defined by the large-scale circulation, and the track gradually recurved around the subtropical Atlantic high pressure system (Fig. 4). There were no strong interactions with adjacent environmental circulations, as with Iris. As a result, the operational NOGAPS model performance for Luis was one of the best on record, and for that reason the satellite-derived data did not improve the track forecasts (Part II).

Multispectral satellite observations were focused on Luis beginning on 1 September, and continued through 10 September at 6-h intervals. VIS low-level winds were produced from 5 to 10 September and for the most part covered the outer vortex region quite well. Luis moved into the sounder WVWV coverage around 7 September. An example dataset is shown in Fig. 8. Note the tremendous outflow from Luis during this time as captured in the multispectral observations.

5. Discussion

The focus of Part I of this article involved the description of initial attempts to optimize the extraction and processing of information from GOES-8 into high-density and quality, spatially coherent wind vector fields around tropical cyclones. It has been demonstrated that multispectral datasets covering the North Atlantic can be reliably processed via automated methodologies at densities to meet varying meteorological requirements, and in a time frame commensurate with operational demands. The vector fields are passed through a rigorous postprocessing procedure that includes the reevaluation of vector height assignments based on variational methods. The vector field quality has been evaluated by statistical analysis (Velden et al. 1997; Nieman et al. 1997), subjective interpretation and applications [e.g., NCEP TPC, Naval Atlantic Meteorology and Oceanographic Center (NLMOC)], and through impact studies on numerical track forecasts (Part II).

It will be shown in Part II of this paper that the GOES multispectral wind information had a significant positive impact on the NOGAPS forecasts of the CHIL ensemble. These results support previous studies and are encouraging, and should be considered a benchmark. Further experiments with other numerical modeling centers including NCEP EMC and the Geophysical Fluids Dynamics Lab are under way. In the meantime, the positive impact on NOGAPS TC track forecasts prompted the initiation of operational assimilation of the GOES-8 multispectral winds produced by CIMSS into NOGAPS beginning in late July of 1996.

The experience with the CHIL datasets prompted upgrades to the dataset collection and processing routines during the 1996 TC season. Increased computer resources allowed an expanded coverage of IR, WVWV (imager), and low-level VIS winds. The National Weather Service created a special programmable GOES-8 sounder hurricane scan to allow coverage over selected regions of the Atlantic, which yielded superior WVWV coverage from the 7.0- and 7.3-μm channels. Finally, experience with the 1995 datasets instigated several changes and upgrades to the processing algorithms. For example, the scenario leading to a slow bias in the sounder WVWV has been identified and the problem eliminated. Also, a two-stage objective editing strategy has been invoked. Tighter constraints are applied well away from the target TC, while relaxed values are allowed in the storm environment to increase the potency of the satellite information.

The 1996 datasets were disseminated in real time on a 6-h basis to the meteorological community. This included NCEP TPC, who found the data useful to their qualitative analyses (TPC 1996, personal communication). The Naval base at Norfolk, Virginia, now includes the product as a routine part of daily hurricane briefings. The data weigh heavily in decision-making processes such as fleet evacuations and ship sorties (NLMOC 1996, personal communication). In a similar response from the JTWC at Guam, analysts are indicating the GMS-5 WVWV being routinely produced by CIMSS are having a positive affect on depicting upper-level features and conditions pertinent to TC development and motion. These datasets are also being operationally assimilated into NOGAPS.

As touched upon in Velden et al. (1997), several research initiatives aimed at a better understanding of the processes that can affect tropical cyclones are benefiting from the information contributed by the satellite-derived datasets. Future work will focus on data assimilation studies. The objective assimilation of the high-density satellite winds is currently done in a less than optimal manner. Assimilation of datasets at higher temporal frequencies (hourly) is possible and should lead to greater data impact (LeMarshall et al. 1996). The issue of an effective height assignment for the layer-mean motion vectors (WVWV) remains, and may be best addressed within the assimilation process itself. Clearly, further work is needed in this area. In addition, the impact of the multispectral satellite-derived winds on numerical prediction of TC genesis and intensity change will be investigated.

Acknowledgments

The authors are grateful to the following individuals for their contributions: Paul Menzel and Kit Hayden (since retired) of NOAA/NESDIS, Steve Nieman and Bob Merrill of UW/CIMSS, Jeff Hawkins and Jim Goerss of NRL-Monterey, Steve Lord of NCEP/EMC, and Don Gray of NOAA/NESDIS. The reviews by Graeme Kelley, Lance Leslie, and Jenni Evans improved the content of the paper. This research was supported primarily by NOAA Grant 50WCNE-306075, with partial funding by NRL Contract N00014-95-C-6017 (Space and Naval Warfare Systems Command PE 0603207N and Office of Naval Research PE 0602435N).

REFERENCES

  • Coakley, J., and F. Bretherton, 1982: Cloud cover from high resolution scanner data: Detecting and allowing for partially filled fields of view. J. Geophys. Res.,87, 4917–4932.

  • Derber, J., and W.-S. Wu, 1996: Use of cloud-cleared radiances in the NCEP SSI analysis system. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 236–237.

  • Fiorino, M., and R. L. Elsberry, 1989: Some aspects of vortex structure in tropical cyclone motion. J. Atmos. Sci.,46, 979–990.

  • Franklin, J. L., and M. DeMaria, 1992: The impact of Omega dropwindsonde observations on barotropic hurricane track forecasts. Mon. Wea. Rev.,120, 381–391.

  • Goerss, J. S., and R. A. Jeffries, 1994: Assimilation of synthetic tropical cyclone observations into the Navy Operational Global Atmospheric Prediction System. Wea. Forecasting,9, 557–576.

  • ——, C. S. Velden, and J. D. Hawkins, 1998: The impact of multispectral GOES-8 wind information on Atlantic tropical cyclone track forecasts in 1995. Part II: NOGAPS forecasts. Mon. Wea. Rev.,126, 1219–1227.

  • Gross, J. M., and M. B. Lawrence, 1996: 1995 National Hurricane Center forecast verification. Proc. 50th Interdepartmental Hurricane Conference, Miami, FL, NOAA OFCM, B10–B28.

  • Hayden, C. M., and R. J. Purser, 1986: Applications of a recursive filter, objective analysis in the processing and presentation of VAS data. Preprints, Second Conf. on Satellite Meteorology, Williamsburg, VA, Amer. Meteor. Soc., 82–87.

  • ——, and C. S. Velden, 1991: Quality control and assimilation experiments with satellite derived wind estimates. Preprints, Ninth Conf. on Numerical Weather Prediction, Denver, CO, Amer. Meteor. Soc., 19–23.

  • ——, and R. J. Purser, 1995: Recursive filter objective analysis of meteorological fields: Applications to NESDIS operational processing. J. Appl. Meteor.,34, 3–15.

  • LeMarshall, J., L. M. Leslie, and A. F. Bennett, 1996: Tropical Cyclone Beti—An example of the benefits of assimilating hourly satellite data. Aust. Meteor. Mag.,45, 275–279.

  • Menzel, W. P., and J. F. W. Purdom, 1994: Introducing GOES-I: The first of a new generation of geostationary operational environmental satellites. Bull. Amer. Meteor. Soc.,75, 757–780.

  • ——, W. L. Smith, and T. R. Stewart, 1983: Improved cloud motion vector and altitude assignment using VAS. J. Climate Appl. Meteor.,22, 377–384.

  • Merrill, R. T., W. P. Menzel, W. Baker, J. Lynch, and E. Legg, 1991:A report on the recent demonstration of NOAA’s upgraded capability to derive cloud motion satellite winds. Bull. Amer. Meteor. Soc.,72, 372–376.

  • Nieman, S. J., J. Schmetz, and W. P. Menzel, 1993: A comparison of several techniques to assign heights to cloud tracers. J. Appl. Meteor.,32, 1559–1568.

  • ——, W. P. Menzel, C. M. Hayden, S. Wanzong, and C. S. Velden, 1996: Upgrades to the NOAA/NESDIS automated cloud-motion vector system. Preprints, Eighth Conf. on Satellite Meteorology and Oceanography, Atlanta, GA, Amer. Meteor. Soc., 1–4.

  • ——, ——, ——, D. Gray, S. Wanzong, C. S. Velden, and J. Daniels, 1997: Fully automated cloud-drift winds in NESDIS operations. Bull. Amer. Meteor. Soc.,78, 1121–1134.

  • Purser, R. J., and R. McQuigg, 1982: A successive correction analysis scheme using recursive numerical filters. Met. O 11 Tech. Note 154, British Meteor. Service, 17 pp.

  • Santek, D., T. Whittaker, J. T. Young, and W. Hibbard, 1991: The implementation plan for McIDAS-AIX. Preprints, Seventh International Conf. on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, New Orleans, LA, Amer. Meteor. Soc., 177–179.

  • Schmetz, J., K. Holmlund, J. Hoffman, and B. Strauss, 1993: Operational cloud-motion winds from Meteosat images. J. Appl. Meteor.,32, 1207–1225.

  • Velden, C. S., 1996: Winds derived from geostationary satellite moisture channel observations: Applications and impact on numerical weather prediction. Meteor. Atmos. Physics,60, 37–46.

  • ——, and S. Goldenberg, 1987: The inclusion of high density satellite wind information in a barotropic hurricane-track forecast model. Preprints, 17th Conf. on Hurricanes and Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 90–93.

  • ——, C. M. Hayden, W. P. Menzel, J. L. Franklin, and J. Lynch, 1992: The impact of satellite-derived winds on numerical hurricane track forecasting. Wea. Forecasting,7, 107–118.

  • ——, ——, S. J. Nieman, W. P. Menzel, S. Wanzong, and J. S. Goerss, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc.,78, 173–195.

  • Weldon, R. B., and S. J. Holmes, 1991: Water vapor imagery: Interpretation and applications to weather analysis and forecasting. NOAA Tech. Rep. NESDIS 67, 213 pp. [Available from NOAA/NESDIS, 5200 Auth Rd., Washington, DC 20233.].

APPENDIX

Recursive Filter Objective Analysis

The recursive filter (RF) objective analysis method is a successive approximation, empirical linear interpolation method. It is relatively fast and economical and is well suited for applications with smaller-sized computers or limited time constraints. Despite the simplicity, it is appropriate for large volume datasets with a high degree of spatial inhomogeneity (due to the feature of locally varying scaling) such as the multispectral vectors discussed here. As mentioned above, the method offers a confidence value associated with each vector to be used in quality control, or as a weighting factor in assimilation applications. The RF was developed at the British Meteorological Office (Purser and McQuigg 1982), and first applications to satellite data were reported on by Hayden and Purser (1986).

The basic algorithm, in one dimension, is described by
AixAi−1xAix
applied to a line of input values Ai, where i is the integer index of grid points along this line. The output values Ai are “forward” smoothed in this example, with i increasing sequentially. The smoothing factor (1 − x) controls the spatial scale of the filter, and the coefficient x will vary in space in response to the changing effective density of the observations.
Hayden and Purser (1995) describe the RF in terms of the relationship between the smoothing factor and the intrinsic distance scale, or “characteristic spatial scale” R
i1520-0493-126-5-1202-ea2
where L is the number of iteration of the filter, and δ is the analysis grid spacing. As described in appendix B of Hayden and Purser (1995), R is prescribed for a given analysis pass and the smoothing factor (1 − x) is determined from an inversion of (A2). Extension to three dimensions involves equivalent operations (iterations) over all horizontal rows and vertical columns of data. A horizontal/vertical distance scaling ratio is prescribed (typically about 100). This procedure is repeated for the L iterations of an analysis pass. The characteristic scale, and therefore the smoothing parameter, vary from one grid point to the next depending on the quality and density of the surrounding data.
On a given analysis pass (m), the quality of an observation (k) is defined as ranging from zero to unity according to
i1520-0493-126-5-1202-ea3
where Ô is the observation, Â is the background value, and T is a pass-dependent tolerance defined as
TmTSmT0T
The constant s defines the rate of change in nominal characteristic scale or tolerance. Here, T0 and T are values for the initial and asymptotic (m − ∞) passes, respectively. The initial value T0 is the largest deviation from the background  that will be tolerated, and is decreased with each pass anticipating that the data fit relative to the background should improve if the data quality is good. The exponent n determines the magnitude of the permitted excursion beyond the tolerance. Higher values of n lead to tighter constraints. The relationship between the quality fit parameter as a function of n is shown in Fig. 1 of Hayden and Purser (1995). Details for the choice of T0 and n are given in appendix C of Hayden and Purser (1995). Note that an observational discrepancy that exceeds the tolerance is not rejected outright, but rather it is given an observational weight that reflects the deviation. Supportive environmental observations may improve the quality rating in subsequent iterations.
Final quality indicators for each vector are appended at the completion of the analysis. This indicator reflects both the fit of the observation and the quality of the neighboring analysis:
qkkŴk1/2
where k was defined in (A3), and Ŵk is the observation weight, which is the product of the current estimate of quality and a predetermined reliability both of which are defined to lie between zero and unity. The predetermined reliability is based on prior estimates of probable accuracy and plays the role of the observation or background error covariance included in optimal analysis.

The RF’s rejection of vectors is nominally based on a threshold value of 0.5 for the quality indicator given in (A5). This threshold was determined from a statistical analysis involving matches with a collocated sample of rawinsonde winds. However, this threshold may vary with vector type, and can be situation dependent in which case adjustments and optimal tuning are warranted. The options for tuning the performance of the quality control system are accomplished by keyword entries in the initiation of the program. Several options that are employed specific to TC dataset processing are geared toward retaining vectors in situations where the background guess field has difficulties. Examples include the adjustments to 1) the background weight in the analysis, 2) observation gross error limits, 3) the horizontal increment of the analysis, and 4) vertical coupling constraints in the analysis. These adjustments applied to TC situations are examined in section 3c.

Fig. 1.
Fig. 1.

Wind vectors derived from multispectral GOES-8 imagery during Atlantic Hurricane Felix on 14 August 1995. All data are plotted over a 6.7-μm water vapor image: (a) 6.7-μm water vapor motion winds, (b) infrared cloud-tracked winds, (c) 7.0-μm water vapor motion winds, (d) 7.3-μm water vapor motion winds, (e) low-level cloud-tracked winds from the visible channel, and (f) collective five-channel dataset illustrated by vector height (color) as shown in caption.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 1.
Fig. 1.
Fig. 2.
Fig. 2.

Weighting (energy contribution) functions associated with the GOES-8 water vapor channels for a typical subtropical atmosphere. Numbers in parentheses refer to channel central wavelengths. Peaks in the functions correspond to tropospheric layers of greatest energy (irradiance) contribution.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 3.
Fig. 3.

Schematic diagram of the objective postprocessing and quality control procedure employed in the CIMSS automated satellite wind vector extraction algorithm. Details of the procedure are provided in the text. WVWV represents input satellite wind data, in this case water vapor wind vectors.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 4.
Fig. 4.

Tracks of 1995 tropical cyclones Chantal (12–22 July), Humberto (22–31 August), Iris (22 August–4 September), and Luis (29 August–12 September). Each tick mark represents a 6-h position.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 5.
Fig. 5.

Typical spatial coverage of multispectral wind vectors derived at 12-h intervals from GOES-8 during Tropical Storm Chantal (30°N, 70°W). The image (water vapor) and dataset corresponds to 1200 UTC 17 July 1995. The vectors are color coded to show those that fall within the 100–300 mb (yellow), 301–700 mb (beige), and 701–950 mb (white) tropospheric layers. The upper layer contains mainly the 6.7-μm water vapor motion vectors and IR cloud-tracked winds; the middle layer is dominated by the 7.0- and 7.3-μm water vapor motion winds; the lower layer is predominantly cloud-tracked winds from the visible channel.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 6.
Fig. 6.

Same as Fig. 5 except for Hurricane Humberto (located near 20°N, 47°W) at 1200 UTC 26 August. There were no low-level visible channel cloud-tracked winds available during Humberto.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 7.
Fig. 7.

Same as Fig. 5 except for Hurricane Iris (located near 27.5°N, 59°W) at 1200 UTC 31 August. Tropical Storm Karen (19°N, 52°W), remnants of Hurricane Humberto (upper right), and approaching Hurricane Luis (lower right) are also depicted. There were no low-level visible channel cloud-tracked winds available at this time.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Fig. 8.
Fig. 8.

Same as Fig. 5 except for Hurricane Luis at 1200 UTC 7 September.

Citation: Monthly Weather Review 126, 5; 10.1175/1520-0493(1998)126<1202:TIOMGW>2.0.CO;2

Save
  • Coakley, J., and F. Bretherton, 1982: Cloud cover from high resolution scanner data: Detecting and allowing for partially filled fields of view. J. Geophys. Res.,87, 4917–4932.

  • Derber, J., and W.-S. Wu, 1996: Use of cloud-cleared radiances in the NCEP SSI analysis system. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 236–237.

  • Fiorino, M., and R. L. Elsberry, 1989: Some aspects of vortex structure in tropical cyclone motion. J. Atmos. Sci.,46, 979–990.

  • Franklin, J. L., and M. DeMaria, 1992: The impact of Omega dropwindsonde observations on barotropic hurricane track forecasts. Mon. Wea. Rev.,120, 381–391.

  • Goerss, J. S., and R. A. Jeffries, 1994: Assimilation of synthetic tropical cyclone observations into the Navy Operational Global Atmospheric Prediction System. Wea. Forecasting,9, 557–576.

  • ——, C. S. Velden, and J. D. Hawkins, 1998: The impact of multispectral GOES-8 wind information on Atlantic tropical cyclone track forecasts in 1995. Part II: NOGAPS forecasts. Mon. Wea. Rev.,126, 1219–1227.

  • Gross, J. M., and M. B. Lawrence, 1996: 1995 National Hurricane Center forecast verification. Proc. 50th Interdepartmental Hurricane Conference, Miami, FL, NOAA OFCM, B10–B28.

  • Hayden, C. M., and R. J. Purser, 1986: Applications of a recursive filter, objective analysis in the processing and presentation of VAS data. Preprints, Second Conf. on Satellite Meteorology, Williamsburg, VA, Amer. Meteor. Soc., 82–87.

  • ——, and C. S. Velden, 1991: Quality control and assimilation experiments with satellite derived wind estimates. Preprints, Ninth Conf. on Numerical Weather Prediction, Denver, CO, Amer. Meteor. Soc., 19–23.

  • ——, and R. J. Purser, 1995: Recursive filter objective analysis of meteorological fields: Applications to NESDIS operational processing. J. Appl. Meteor.,34, 3–15.

  • LeMarshall, J., L. M. Leslie, and A. F. Bennett, 1996: Tropical Cyclone Beti—An example of the benefits of assimilating hourly satellite data. Aust. Meteor. Mag.,45, 275–279.

  • Menzel, W. P., and J. F. W. Purdom, 1994: Introducing GOES-I: The first of a new generation of geostationary operational environmental satellites. Bull. Amer. Meteor. Soc.,75, 757–780.

  • ——, W. L. Smith, and T. R. Stewart, 1983: Improved cloud motion vector and altitude assignment using VAS. J. Climate Appl. Meteor.,22, 377–384.

  • Merrill, R. T., W. P. Menzel, W. Baker, J. Lynch, and E. Legg, 1991:A report on the recent demonstration of NOAA’s upgraded capability to derive cloud motion satellite winds. Bull. Amer. Meteor. Soc.,72, 372–376.

  • Nieman, S. J., J. Schmetz, and W. P. Menzel, 1993: A comparison of several techniques to assign heights to cloud tracers. J. Appl. Meteor.,32, 1559–1568.

  • ——, W. P. Menzel, C. M. Hayden, S. Wanzong, and C. S. Velden, 1996: Upgrades to the NOAA/NESDIS automated cloud-motion vector system. Preprints, Eighth Conf. on Satellite Meteorology and Oceanography, Atlanta, GA, Amer. Meteor. Soc., 1–4.

  • ——, ——, ——, D. Gray, S. Wanzong, C. S. Velden, and J. Daniels, 1997: Fully automated cloud-drift winds in NESDIS operations. Bull. Amer. Meteor. Soc.,78, 1121–1134.

  • Purser, R. J., and R. McQuigg, 1982: A successive correction analysis scheme using recursive numerical filters. Met. O 11 Tech. Note 154, British Meteor. Service, 17 pp.

  • Santek, D., T. Whittaker, J. T. Young, and W. Hibbard, 1991: The implementation plan for McIDAS-AIX. Preprints, Seventh International Conf. on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, New Orleans, LA, Amer. Meteor. Soc., 177–179.

  • Schmetz, J., K. Holmlund, J. Hoffman, and B. Strauss, 1993: Operational cloud-motion winds from Meteosat images. J. Appl. Meteor.,32, 1207–1225.

  • Velden, C. S., 1996: Winds derived from geostationary satellite moisture channel observations: Applications and impact on numerical weather prediction. Meteor. Atmos. Physics,60, 37–46.

  • ——, and S. Goldenberg, 1987: The inclusion of high density satellite wind information in a barotropic hurricane-track forecast model. Preprints, 17th Conf. on Hurricanes and Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 90–93.

  • ——, C. M. Hayden, W. P. Menzel, J. L. Franklin, and J. Lynch, 1992: The impact of satellite-derived winds on numerical hurricane track forecasting. Wea. Forecasting,7, 107–118.

  • ——, ——, S. J. Nieman, W. P. Menzel, S. Wanzong, and J. S. Goerss, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc.,78, 173–195.

  • Weldon, R. B., and S. J. Holmes, 1991: Water vapor imagery: Interpretation and applications to weather analysis and forecasting. NOAA Tech. Rep. NESDIS 67, 213 pp. [Available from NOAA/NESDIS, 5200 Auth Rd., Washington, DC 20233.].

  • Fig. 1.

    Wind vectors derived from multispectral GOES-8 imagery during Atlantic Hurricane Felix on 14 August 1995. All data are plotted over a 6.7-μm water vapor image: (a) 6.7-μm water vapor motion winds, (b) infrared cloud-tracked winds, (c) 7.0-μm water vapor motion winds, (d) 7.3-μm water vapor motion winds, (e) low-level cloud-tracked winds from the visible channel, and (f) collective five-channel dataset illustrated by vector height (color) as shown in caption.

  • Fig. 1.

    (Continued)

  • Fig. 1.

    (Continued)

  • Fig. 2.

    Weighting (energy contribution) functions associated with the GOES-8 water vapor channels for a typical subtropical atmosphere. Numbers in parentheses refer to channel central wavelengths. Peaks in the functions correspond to tropospheric layers of greatest energy (irradiance) contribution.

  • Fig. 3.

    Schematic diagram of the objective postprocessing and quality control procedure employed in the CIMSS automated satellite wind vector extraction algorithm. Details of the procedure are provided in the text. WVWV represents input satellite wind data, in this case water vapor wind vectors.

  • Fig. 4.

    Tracks of 1995 tropical cyclones Chantal (12–22 July), Humberto (22–31 August), Iris (22 August–4 September), and Luis (29 August–12 September). Each tick mark represents a 6-h position.

  • Fig. 5.

    Typical spatial coverage of multispectral wind vectors derived at 12-h intervals from GOES-8 during Tropical Storm Chantal (30°N, 70°W). The image (water vapor) and dataset corresponds to 1200 UTC 17 July 1995. The vectors are color coded to show those that fall within the 100–300 mb (yellow), 301–700 mb (beige), and 701–950 mb (white) tropospheric layers. The upper layer contains mainly the 6.7-μm water vapor motion vectors and IR cloud-tracked winds; the middle layer is dominated by the 7.0- and 7.3-μm water vapor motion winds; the lower layer is predominantly cloud-tracked winds from the visible channel.

  • Fig. 6.

    Same as Fig. 5 except for Hurricane Humberto (located near 20°N, 47°W) at 1200 UTC 26 August. There were no low-level visible channel cloud-tracked winds available during Humberto.

  • Fig. 7.

    Same as Fig. 5 except for Hurricane Iris (located near 27.5°N, 59°W) at 1200 UTC 31 August. Tropical Storm Karen (19°N, 52°W), remnants of Hurricane Humberto (upper right), and approaching Hurricane Luis (lower right) are also depicted. There were no low-level visible channel cloud-tracked winds available at this time.

  • Fig. 8.

    Same as Fig. 5 except for Hurricane Luis at 1200 UTC 7 September.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 533 134 4
PDF Downloads 323 86 5