Cloud Tracking with Satellite Imagery: From the Pioneering Work of Ted Fujita to the Present

Cloud Tracking with Satellite Imagery: From the Pioneering Work of Ted Fujita to the Present W. Paul Menzel NOAA/NESDIS/ORA, Madison, Wisconsin ABSTRACT Tetsuya (Ted) Fujita was a pioneer in remote sensing of atmospheric motion. When meteorological satellites were introduced, he developed techniques for precise analysis of satellite measurements (sequences of images from polar orbiting platforms first and then from geostationary platforms). Soon after his initial work, the ability to track clouds and relate them to flow patterns in the atmosphere was transferred into routine operations at the national forecast centers. Cloud motion vectors derived from geostationary satellite imagery have evolved into an important data source of meteorological information, especially over the oceans. The current National Environmental Satellite, Data, and Information Service operational production of Geostationary Operational Environmental Satellite cloud and water vapor motion winds continues to perform well; rms differences with respect to raob s are found to be 6.5 7.5 m s 1. 1. Early work on detection of atmospheric motion by remote sensing Tetsuya Theodore (Ted) Fujita was a pioneer in remote sensing of atmospheric motion. He introduced methods for gridding satellite images that made accurate inferences of winds possible with early Television Infrared Observation Satellite (TIROS) pictures. He developed laboratory techniques that facilitated cloud motion analysis. Together with his staff and students, he made numerous investigations of clouds (motion, growth or decay, and height) using observations from the ground, aircraft, and space; these contributed to our understanding as to how cloud motion measurements relate to the winds in the cloud environment. Finally, his use of geostationary satellites to study cloud motions associated with storms and mesoscale systems provided both research and operational meteorologists with new insights into atmospheric circulations on many scales. Corresponding author address: Paul Menzel, NOAA/NESDIS/ ORA, 1225 W. Dayton St., Madison, WI 53706. E-mail: paul.menzel@ssec.wisc.edu In final form 22 June 2000. 2001 American Meteorological Society Cloud pictures from the first TIROS polar orbiting satellites in 1960 showed the potential, when viewed in time sequence, for inferring atmospheric motions. Fujita developed the necessary rectification (Fujita 1961) and analysis (Fujita 1963, 1964) techniques to make those satellite photographs useful for estimating the velocity of both low- and high-level winds. In the study of a 1960 South Pacific tropical storm, Fujita and Uchijima (1963) analyzed clouds to provide information about the direction of low-level winds and the vertical wind shear between 700 and 200 Mb. Fujita also showed how cloud shadows in these early satellite pictures could be used to quantitatively determine cloud-top height. Geostationary remote sensing started in December 1966 with the first Applications Technology Satellite (ATS- 1). The ATS-1 spin-scan cloud camera (Suomi and Parent 1968) pro- James Purdom (speaking on behalf of W. Paul Menzel) 33

FIG. 1. An example of a stereo pair of whole-sky pictures taken within 2 s of each other. The degrees on the concentric circles in each picture denote the elevation angles. Selected cloud elements for velocity computation are identified by the numbers 1 6. vided full-disk visible images of the earth and its cloud cover every 20 min. Suomi (1969) noted that now the clouds move not the satellite. Within the first month of the availability of ATS-1 imagery, Fujita and Bohan (1967) produced a movie showing enlarged views of mesoscale cloud patterns in motion. This film was widely shown and demonstrated to the meteorological community how animated satellite images reveal atmospheric motion and their potential use for research and operations. Fujita also collaborated with Suomi and Parent to produce the first color movie of the Planet Earth using the three different simultaneous spectral images provided by ATS-3; this movie introduced the sequence animation technique that became the standard for operational geostationary image movie loop production (Fujita 1992). Fujita applied cloud motion analysis on ATS data (Fujita 1968) to investigate the formation and structure of atmospheric circulations on all scales. He analyzed the mesoscale structure 34 of a subtropical jet stream using cloud motion vectors to compute the divergence and vorticity in the region of the jet and illustrated the effects within the jet clouds in a cloud-relative (Lagrangian) frame of reference (Fujita 1969a). He generated a wealth of information about thunderstorms and tornadoes [examples include Fujita and Bradbury (1969) and Fujita (1970c)]. He used cloud tracking to show how jet stream flow is modified by large convective storms (Fujita 1969b) and the interaction between a jet stream and outflows from a hurricane and large rain areas (Fujita 1969c). A number of hurricane and tropical cloud studies incorporated satellite cloud winds. Hurricane studies included Brenda (Fujita 1969d), Gladys (Gentry et al. 1970), Camille (Fujita 1972b), Ginger (Tecson et al. 1973), and Anna (Fujita and Tecson 1974). He also FIG. 2. A series of ATS-1 pictures on 15 Mar 1967 started at 1417, 1440, 1503, 1526, 1549, and 1612 Hawaii standard time (HST). These pictures were enlarged from 8 10 in. high-resolution negatives provided by V. Suomi, University of Wisconsin. Vol. 82, No. 1, January 2001

wrote on the use of ATS pictures in hurricane modification (Fujita 1972a). Tropical Pacific studies included the outflow from a tropical cloud cluster (Fujita 1969e) and an equatorial anticyclone (Fujita et al. 1969). In the tropical Atlantic, Fujita tracked Saharan dust (Fujita 1970b) and computed cloud motions over the Global Atmospheric Research Program (GARP) Atlantic Tropical Experiment area (Tecson and Fujita 1975). Fujita pursued an understanding of the relationship between cloud motions and the wind with a series of cloud truth experiments during most of his career. The first was the Haleakala experiment in 1967. The Satellite and Mesometeorology Research Project at the University of Chicago, under Fujita s direction, established a camera network in Hawaii to gather photographic cloud data during the Line Island Experiment in March April 1967. He developed techniques using ground-based cameras to stereoscopically track cloud features (Bradbury and Fujita 1968); he inferred velocity and height from dualcamera whole-sky time-lapse picture sequences (Fig. 1). The purpose of Fujita s experiment was to validate the contention that ATS could greatly enhance the efforts to detect atmospheric motions on synoptic scales. Fujita refined the techniques for tracking clouds and inferring (a) (b) winds using image sequences (Fujita et al. 1968) and validated those satellite cloud motion vectors with the ground cameras; his careful analysis of mesoscale cloud motions from a combination of ATS-1 (Fig. 2) and terrestrial photographs (Fig. 1) produced the confirmation that atmospheric motions could be correctly determined from properly navigated geostationary images (Fig. 3a). Merging information derived from FIG. 3. (a) Computed velocities of cloud elements selected as good tracers. The velocities were computed from the cloud displacements during the period between 1223 and 1354 HST on 15 Mar 1967. (b) Low cloud velocities inferred from ATS images on 15 Mar 1967 superimposed on the 850-hPa analysis; the large convective cloud in the ATS image is clearly located on a convergence line. satellite images and analyses of conventional data, Fujita was able to infer greater detail in the synoptic situation; the data clearly showed that the large convective cloud was on a convergence line (Fig. 3b). The combination of Suomi s spin-scan camera and Fujita s cloud tracking techniques opened the door to a new field of inferring atmospheric motions from remotely sensed image sequences. This early work (with se- 35

quences of images from a geostationary platform) still sets the standard. Fujita conducted three more cloud truth experiments with ground cameras (Fujita 1992). In Barbados (1967), Cape Hatteras (1988), and Key West (1989), he provided independent measurements of cloud height and motion that could be used for validation of both satellite winds and radiosonde data. By the early 1970s ATS imagery was being used in operational forecast centers, with the first movie loops being used at the National Severe Storm Forecast Center in the spring of 1972 and at the National Hurricane Center later that year. To display the satellite images for useful and timely analysis, Fujita developed the Fujita wheel, which consisted of a turntable with registration pins to hold and display pictures. The pictures were registered using punched holes on landmarks and illuminated by a strobe light. When properly adjusted, the analyst saw the clouds in motion; the image sequence was updated by adding the latest and dropping the oldest picture. Shortly thereafter, Fujita and his staff at Chicago developed the Meteorologists Tracking Computer, an interactive computer system for manual gridding of the pictures and estimating atmospheric velocities from cloud displacements in a sequence of navigated timed satellite images (Chang et al. 1973; Tecson and Fujita 1975). Many of the techniques and procedures developed by Fujita were later used to make movie loops and to estimate satellite winds operationally (Hubert and Whitney 1971). The Man computer Interactive Data Access System refined the techniques for computerassisted movie loop generation and winds production (Suomi et al. 1983) and enabled subsequent global operational activities. Fujita and his collaborators produced a number of other studies using ground, aircraft, astronaut, and satellite observations (Fujita and Bradbury 1964; Fujita 1970a; Fujita et al. 1975; Fujita 1991) that dealt with cloud size and persistence; these were helpful for identifying proper tracking targets for producing satellite winds. Figure 4 shows an example; a drifting anvil off the coast of Miami is found to have higher velocities around the anvil edge and much slower velocities in the cloud center where there is still considerable vertical motion deep inside the anvil cloud. Fujita correctly concluded that the best cloud tracers are in the downwind edge of a drifting anvil cloud (Fujita 1991). In the 1970s, Fujita used his experience to advise the National Aeronautics and Space Administration (NASA) on future sensor design characteristics; his recommendations covered horizontal resolution, temporal resolution or picture intervals, and radiance enhancement (Tecson et al. 1977). In particular, he was concerned that tracers associated with cumulus clouds over land demonstrated a trackable life of only 5 10 min, making rapid-scan pictures necessary for computing these winds (Fujita et al. 1975). The evolution of ATS into the Geostationary Operational Environmental Satellite (GOES) made possible the routine tracking of clouds at high resolution, with infrared imagery available for cloud height assignment and nighttime tracking. Cloud motion vectors (CMVs) derived from geostationary satellite imagery became an important data source of meteorological information, especially over the oceans. The early satellite CMV production peaked during the First GARP Global Experiment (FGGE) in 1979. Global fields of motion vectors derived from five geostationary satellites were generated twice daily for the whole year to assist the study of global atmospheric circulations. The advent of water vapor imagery with the first Meteosat in 1978 suggested a further multitude of tracers, some associated with water vapor gradients and not clouds (Eigenwillig and Fischer 1982). Water vapor images often reveal mid- to upper-tropospheric structure that is not apparent in infrared or visible images and water vapor motion vectors (WVMVs) were soon derived by tracking features in water vapor imagery (Fig. 5). These water vapor winds, along with the low-level cloud drift winds de- FIG. 4. Twenty cloud motion winds obtained by tracking various parts of a thunderstorm anvil off the coast of Miami, FL, in Sep 1989. Fujita found that, in general, the leading edge of a drifting anvil cloud moves faster than the central region of the cloud. 36 Vol. 82, No. 1, January 2001

rived from animated visible and infrared window images, have been used to help define atmospheric steering currents that influence weather trajectories, especially in hurricane situations (Velden et al. 1992, 1997; Holmlund 1993; Langland et al. 1999). Both cloud and water vapor motion vectors are now produced operationally. When the two-goes system became operational, stereo imaging was possible. In 1979 during a project known as the Severe Environmental Storm And Mesoscale Experiment, the two GOES satellites were synchronized to produce 3-min-interval rapid-scan imagery to study storm development. Fujita (1982) used these data to produce very accurate cloud height assignments using stereographic techniques. Results from this and earlier research prompted 5-min-interval imagery to become a routine part of satellite operations during severe storm outbreaks. 2. Current operational satellite detection of winds The early satellite CMV production peaked during FGGE in 1979. Global fields of motion vectors derived from five geostationary satellites were generated twice daily for the whole year to assist the study of global atmospheric circulations. More recently geostationary satellites from China, Europe, Japan, Russia, and the United States are enabling the generation of FGGElike datasets four times daily. Operational winds from GOES are derived from a sequence of three navigated and earth-located images taken 30 min apart. The winds are calculated by a three-step objective procedure. The initial step selects targets, the second step assigns pressure altitude, and the third step derives motion. Altitude is assigned based on a temperature/pressure derived from radiative transfer calculations in the environment of the target. Motion is derived by a pattern recognition algorithm that matches a feature within the target area in one image within a search area in the second image. For each target two winds are produced representing the motion from the first to the second, and from the second to the third image. An objective editing scheme is then employed to perform quality control: the first-guess motion, the consistency of the two winds, the precision of the cloud-height assignment, and the vector fit to an analysis are all used to assign a quality flag to the vector (which is actually the average of the two vectors). FIG. 5. Water vapor winds derived from a sequence of halfhourly Meteosat 6.7- m images; from a dataset developed by V. Suomi. WVMVs (imager band at 6.7 m, which sees the upper troposphere, and sounder bands at 7.0 and 7.5 m, which see deeper into the troposphere) are derived by the same methods used with CMVs. Heights are assigned from the water vapor brightness temperature in clear sky conditions and from radiative transfer techniques in cloudy regions. In 1998, the National Environmental Satellite, Data, and Information Service (NESDIS) operational GOES-8/9/10 CMV and WVMV production increased to every 3 h [every other wind set is transmitted over the Global Telecommunications System in the new BUFR (Binary Universal Format for the Representation of Meteorological Data) format] with high spatial density tracers derived from visible, infrared window, and water vapor images. The number of motion vectors has increased dramatically to over 10 000 vectors for each winds dataset. The quality of the wind product is being reported monthly in accordance with the Coordination Group for Meteorological Satellites (CGMS) reporting procedures (Schmetz et al. 1997, 1999); this involves direct comparisons of collocated computed cloud motions and radiosonde observations. It reveals GOES cloud motion wind rms differences to be within 6.5 7 m s 1 with respect to raob s, with a slow bias of about 0.5 m s 1 ; water vapor motion rms differences are within 7.0 7.5 m s 1 (Nieman et al. 1997). The NESDIS operational winds inferred from infrared window and water vapor images continue to perform well. Figure 6 presents the summary from a recent 12-month period. The next few sections describe the current operational procedures. 37

a. Tracer selection In the 1990s, tracer selection for GOES winds was improved. In the old tracer selection algorithm, the highest pixel brightness values within each target domain were found, local gradients were computed around those locations, and adequately large gradients were assigned as target locations. In the new tracer selection algorithm, maximum gradients are subjected to a spatial-coherence analysis. Too much coherence indicates such features as coastlines and thus is undesirable. The presence of more than two coherent scenes often indicates mixed level clouds; such cases are screened. The resulting vectors from the new scheme (Fig. 7) show a much higher density of tracers in desirable locations. b. Height assignment techniques Semitransparent or subpixel clouds are often the best tracers, because they show good radiance gradients that can readily be tracked and they are likely to be passive tracers of the flow at a single level. Unfortunately their height assignments are especially difficult. Since the emissivity of the cloud is less than unity by an unknown and variable amount, its brightness temperature in the infrared window is an overestimate of its actual temperature. Thus, heights for thin clouds inferred directly from the observed brightness temperature and an available temperature profile are consistently low. Presently heights are assigned by any of three techniques when the appropriate spectral radiance measurements are available (Nieman et al. 1993). In opaque clouds, infrared window (IRW) brightness temperatures are compared to forecast temperature profiles to infer the level of best agreement, which is taken to be the level of the cloud. In semitransparent clouds or subpixel clouds, since the observed radiance contains contributions from below the cloud, this IRW technique assigns the cloud to too low a level. Corrections for the semitransparency of the cloud are possible with the carbon dioxide (CO 2 ) slicing technique (Menzel et al. 1983) where radiances from different layers of the atmosphere are ratioed to infer the correct height. A similar concept is used in the water vapor (H 2 O) intercept technique (Szejwach 1982), where the fact that radiances influenced by uppertropospheric moisture (H 2 O) and IRW radiances exhibit a linear relationship as a function of cloud amount is used to extrapolate the correct height. An IRW estimate of the cloud height is made by averaging the infrared window brightness temperatures of the coldest 25% of the pixels and interpolating to a pressure from a forecast guess sounding (Merrill et al. 1991). In the CO 2 slicing technique, a cloud height is assigned with the ratio of the deviations in observed radiances (which include clouds) from the corresponding clear air radiances for the infrared window and the CO 2 (13.3 m) channel. The clear and cloudy radiance differences are determined from observations with GOES and radiative transfer calculations. Assuming the emissivities of the two channels are roughly the same, the ratio of the clear and cloudy radiance differences yields an expression by which the cloud-top pressure of the cloud within the field of view can be specified. The observed differences are compared to a series of radiative transfer calculations with possible cloud pressures, and the tracer is assigned the pressure (a) (b) FIG. 6. CGMS statistics (bias and root-mean square) for GOES-8 (GOES-E) and GOES-10 (GOES-W) cloud drift (CD) and water vapor motion (WV) winds for Jul 1998 Jun 1999. 38 Vol. 82, No. 1, January 2001

FIG. 7. Target distribution resulting from the (top) old and (bottom) new tracer selection algorithm. Wind vectors shown represent the improvements in the vector density realized in the past 10 years. The images are from 25 Aug 1998 during the lifetime of Hurricane Bonnie. Heights of wind vectors (see section 2b) are indicated by color: yellow between 150 and 250 hpa, orange between 251 and 350 hpa, and white between 351 and 600 hpa. that best satisfies the observations. The operational implementation is described in Merill et al. (1991). Fujita participated in the validation of the CO 2 heights (Fujita 1993). GOES rapid-scan images (every 5 min) were used to track thin cirrus and CO 2 slicing heights were assigned; two whole-sky cameras in Chicago were taking stereoscopic pictures to infer winds and heights independently. One validation case study involved three distinct clouds, named as only Fujita could: the pelican cloud, the fish cloud, and the bait cloud (Fig. 8a); the satellite wind speeds (Co 2 heights) were found to be within 1.0 m s 1 (0.5 km) of the ground stereo determinations (Fig. 8b). The careful work by the University of Chicago team helped to validate the operational techniques for wind derivation and height assignment. The H 2 O intercept height assignment is predicated on the fact that the radiances for two spectral bands vary linearly with cloud amount. Thus a plot of H 2 O (6.5 m) radiances versus IRW (11.0 m) radiances in a field of varying cloud amount will be nearly linear. These data are used in conjunction with forward calculations of radiance for both spectral channels for opaque clouds at different levels in a given atmosphere specified by a numerical weather prediction of temperature and humidity. The intersection of measured and calculated radiances will occur at clear-sky radiances and cloud radiances. The cloud-top temperature is extracted from the cloud radiance intersection (Schmetz et al. 1993). Satellite stereo height estimation has been used to validate H 2 O intercept height assignments. The technique is based upon finding the same cloud patch in several images. For cloud motion, the cloud needs to change slowly relative to the image frequency. For stereo heights, the cloud needs to be distinct and ap- 39

(a) (b) FIG. 8. (a) GOES-7 visible picture at 1701 UTC on 27 Aug 1993 on the pelican, fish, and bait clouds. Superimposed are the geodetic latitudes and longitudes elevated to the mean height of the clouds. (b) High cloud winds (direction speed) computed from the GOES infrared images. Ground stereo determinations agreed within 1 m s 1 and 2. Cloud heights (not shown) of the pelican belly from CO 2 slicing and cloud shadow geometry (estimated to be 10.5 km) agreed with ground stereo camera determinations within 0.5 km. c. Objective editing Automated procedures for deriving cloud motion vectors from a series of geostationary infrared window images first became operational in the National Oceanic and Atmospheric Administration in 1993 (Merrill et al. 1991). NESDIS has been producing GOES-8/9/10 cloud motion vectors without manual intervention. Suitable tracers are automatically selected within the first of a sequence of images (see section 2a) and heights are assigned using the H 2 O intercept method (see section 2b). Tracking features through the subsequent imagery is automated using a covariance minimization technique (Merrill et al. 1991) and an automated quality control algorithm (Hayden and Nieman 1996) is applied. Editing the CMVs through analyses with respect to a first-guess wind and temperature profile field involves speed adjustment, height adjustment, and quality assessment (Fig. 9). This procedure, with some modifications, is also used to infer water vapor motion vectors; tracer selection is based on gradients within the target area and vector heights are inferred from the water vapor brightness temperatures. To mitigate the slow bias found in upper-level GOES CMW in extratropical regions, each vector above 300 hpa is incremented by 8% of the vector speed. There are indications that the slow bias can be attributed to (a) tracking winds from sequences of images separated by too much time, (b) estimating atmospheric motion vectors from inappropriate tracers, and (c) height assignment difficulties. Some also suggest that cloud motions will not indicate full atmospheric motions in most situations. pear nearly the same from the two viewpoints (after remapping to the same projection). Campbell (1998) built upon earlier work of Fujita and others (Fujita 1982) to develop a method that adjusts for the motion of the cloud so that simultaneity is not required for the stereo height estimate. A test analysis was performed with Meteosat-5/7 data; stereo heights and H 2 O intercept heights agreed within 50 hpa. As more geostationary and polar orbiting satellites remain in operation, the prospects for geometric stereo height validations of the operational IRW, H 2 O intercept, and CO 2 slicing heights become very promising. FIG. 9. Schematic of the procedure for editing cloud motion vectors. 40 Vol. 82, No. 1, January 2001

A height adjustment is accomplished through a first-pass analysis of the satellite-derived CMV at their initially assigned pressure height and data from a coincident National Centers for Environmental Prediction (NCEP) short-term Aviation Model forecast. The analysis is a three-dimensional objective analysis (Hayden and Purser 1995) of the wind field using background information from the numerical forecast. The pressure altitudes of the CMVs are adjusted by minimizing a penalty function given by B mk, V = m dd + V F m v dd F T + s + T F s F P + P F ijk,, m ijk,, m ijk,, dd 2 2 2 ijk,, m ijk,, t 2 2 where V = velocity, T = temperature, P = pressure, dd = direction, and s = speed. Subscript m refers to a measurement; i and j are horizontal dimensions in the analysis, and k is the vertical level. The F are weighting factors given to velocity, temperature, pressure, direction, and speed; default values are 2 m s 1, 10 C, 100 hpa, 1000, and 1000 m s 1 respectively. Increasing a value of F downweights that component. As currently selected, neither speed nor direction factor into the computation of the penalty. Note that these default selections give equal worth to a 2 m s 1 discrepancy, a 10 temperature discrepancy, or a 100-hPa discrepancy. Further details can be found in Velden et al. (1998). The pressure levels for height adjustment in hpa are 925, 850, 775, 700, 600, 500, 400, 350, 300, 250, 200, 150, and 100. The pressure or height reassignment is constrained to 100 hpa. A tropopause test looks for lapse rates of less than 0.5 K (25 hpa) 1 above 300 hpa and prohibits reassignment to stratospheric heights. s, p altitudes and by inspecting the local quality of the analysis and the fit of the observation to that analysis. Thresholds are given for rejecting the data. Accepted data and the associated quality estimate, denoted by RFF for recursive filter flag, are passed to the user (Hayden and Purser 1995). Several options are available for regulating the analysis, the penalty function, and the final quality estimates. These have been optimized, over several years of application, for the operational GOES CMVs. However the optimization may be situation dependent; what works best with GOES CMVs may not be optimal for WVMVs or winds generated at higher density, or winds generated with an improved background forecast, etc. Research on optimal tuning of this system in its various applications continues (Velden et al. 1998). The European Organization for the Exploitation for Meteorological Satellites (EUMETSAT) has developed a quality indicator (QI) for use with Meteosat data that checks for direction, speed, and vector consistency in the vector pairs (derived from the three images in the wind-determining sequence). In addition, consistency with nearest neighbors vectors and with respect to a forecast model are factored in. A weighted average of these five consistency checks becomes the QI. Depending on the synoptic situation between 10% and 33% of the processed tracers are rejected in the cloud vector inspection and quality flag assignment. A combination of the EUMETSAT QI and the NESDIS RFF wind quality indicators has been shown to enhance utilization of the high-density CMVs in numerical weather prediction models and is in the process of being implemented (Holmlund et al. 2000). TABLE 1. Recent forecast results using GOES high-density winds in the GFDL hurricane model from selected Atlantic tropical cyclones in 1996 98 (this includes 10 storms and 103 cases). Forecast trajectory errors are estimated for 12, 24, 36, and 72 h. Improvements are indicated in percentage of the mean forecast error as well as the percentage of the forecasts. Forecast Mean forecast Forecast error improvement Forecast improved out to (h) error (km) with GOES winds (%) with GOES winds (%) 12 70 4.5 50 d. Quality flags A quality assessment for each vector is accomplished by a second analysis using the CMVs at the reassigned pressure 24 125 10 60 36 160 12.5 60 72 350 9 60 41

for consistency. The dependency on a numerical weather prediction model first guess was diminished. 3) A dual-pass autoeditor was put in place by summer 1998 that relaxes rejection criteria for winds around a feature of interest and uses normal procedures elsewhere. This enables better retention of the tighter circulation features associated with tropical cyclones and other severe weather. 4) Water vapor winds were designated as being determined in clear skies or over clouds; the clear sky WVMVs are representative of layer mean motion while cloudy sky WVMVs are cloud-top motion. 5) A quality flag that combines QI and RFF has been attached to each wind vector indicating the level of confidence resulting from the postprocessing. FIG. 10. The ITBB is usually straddling the equator as indicated in the monthly mean water vapor wind field for Dec 1981; 12 months before the El Niño onset the ITBB dips south of the equator as indicated in the monthly mean water vapor wind field Jan 1982. Systematic investigation of this phenomena was suggested by Fujita. e. Recent upgrades The major operational changes in the past years are summarized as follows. 1) Winds inferred from visible image loops as well as sounder midlevel moisture sensitive bands were added to operations by summer 1998. 2) This enabled ensemble autoediting, where the combined wind sets from visible, infrared window, and three water vapor sensitive bands are compared 3. Example applications a. Hurricane trajectory forecasting The plethora of motion vectors (CMVs and WVMVs) is being studied in various forecast models; the best approaches toward assimilation remain under evaluation. Hurricane trajectory forecasting is an example of promising positive impact. A study was conducted at the Geophysical Fluid Dynamics Laboratory (GFDL). Highdensity, multispectral GOES-8 winds (from three water vapor, the infrared window, and the visible bands) were assimilated into the GFDL hurricane forecast model for over 100 cases from Atlantic tropical cyclone situations in 1996 98. The winds were directly assimilated using optimal interpolation and vertical blending schemes. GOES data reduced the 24 72-h hurricane track forecast errors by ~10% on average (see Table 1). In summary, the impressive GFDL control runs 42 Vol. 82, No. 1, January 2001

(relative to other models) without the GOES data were improved even more with the GOES winds (Soden et al. 2000). A similar improvement in performance by the Navy Operational Global Atmospheric Prediction System model with high-density GOES winds has been noted in prior hurricane studies (Velden et al. 1998; Goerss et al. 1998). Further research is planned with NCEP to achieve optimum operational utilization of the high-density GOES winds. b. Forecasting the onset of an El Niño event One of the last initiatives proposed by Fujita was to investigate the feasibility of predicting El Niño by monitoring a shift in the intertropical breeze belt (ITBB) that consists of a band of winds less than 10 m s 1 across the Pacific Ocean along the equator. Fujita noticed that 10 12 months before the peak of the sea surface temperature associated with El Niño, the ITBB shifts from along-equatorial alignment and concaves down below the equator. His investigation of the 1972 73 and 1982 83 El Niños suggested that this could be used as an important trigger in predicting El Niño (see Fig. 10 for an example of ITBB preceding the 1982 83 El Niño). Using CMVs and WVMVs, he hoped that this behavior could be detected. This work is being continued with results pending. 4. Research areas TABLE 2. Preliminary evaluation of GOES-10 full-resolution infrared and visible winds performance at 5, 10, 15, and 30 min compared to rawinsonde balloon determinations. Aviation Model 12-h forecast (which served as first guess) wind estimates are also compared. IR Raob satellite (m s 1 ) Raob guess (m s 1 ) 30 min (144 raob matches) bias 0.81 0.89 rms 6.38 5.98 15 min (251 raob matches) bias 0.97 1.37 rms 6.76 6.05 10 min (290 raob matches) bias 0.55 0.97 rms 7.11 6.61 5 min (396 raob matches) bias 1.74 1.27 rms 8.01 7.21 Visible Raob satellite (m s 1 ) Raob guess (m s 1 ) 30 min (167 raob matches) bias 1.09 2.11 rms 5.69 4.74 15 min (199 raob matches) bias 1.54 2.25 rms 4.55 4.68 10 min (298 raob matches) bias 1.19 2.13 rms 4.73 4.86 5 min (429 raob matches) bias 1.16 2.08 rms 5.11 5.32 a. Winds from rapid scan images GOES-10 in April 1998 made 5-min-interval observations routinely with the imager. These datasets were used to study the balance of image resolution and time interval between images for tracking features (Velden et al. 2000). Motion vectors were derived from GOES-10 full-resolution water vapor (8-km resolution), infrared window (4-km resolution), and visible (1-km resolution) images from 8 April 1998 near 2300 UTC using automated procedures described earlier. Three images spaced at 5-, 10-, 15-, and 30-min intervals were used, with the shorter intervals nested inside longer loops for best time matching. Results (based on quantity, quality, and comparison with radiosonde observations) are summarized in Table 2; these indicate that the optimal time lapse between images for visible low-level winds is 5 min, for infrared cloud motion winds it is 10 min, and for water vapor layer drift in clear skies it is 30 min. Operational processing currently uses 30-min sampling for all bands. Synoptic analysis of the wind fields over the Alabama region shows much improved depiction of the upperlevel jet streak and divergence in the optimized (10-min loop) IR winds. The short-wave circulation to the northwest of the severe weather outbreak is best described by the clear-sky WVMVs at 30-min intervals. Finally, the 5-min loop visible winds clearly show a strong low-level inflow into the developing 43

FIG. 11. GOES-10 visible winds derived from sequences of three images separated by 5, 10, 15, and 30 min. Shorter time sequence image loops are nested inside longer loops for best time matching. cells that was not depicted well at 30-min image frequency (see Fig. 11). These preliminary results and others (e.g., Hasler et al. 1998) indicate that the optimum balance of spatial and temporal resolution is not currently being achieved in the half-hourly GOES schedule. Fujita s experience with rapid-scan images in the 1970s had also suggested that 5-min loops produced superior results to 30-min loops (Fujita et al. 1975). Further trade-off studies are planned and schedule adjustments will be investigated. 5. Summary Operational GOES clouddrift and water vapor wind production within NESDIS has evolved considerably since the days of Ted Fujita s first involvement. Winds are generated from purely automated procedures at 3-hourly intervals over the full-disk coverage, and distributed at high spatial density. User feedback and verification statistics over the past several years show that product quality has steadily improved. Effective utilization of these satellite-derived winds shows great promise for increasing forecast skill. The creative and careful work of Ted Fujita paved the way for routine determination of atmospheric motions from sequences of satellite images. Some of his early techniques are still fundamental to present-day tracer identification and displacement determination. His work touched upon many areas. Dr. Jim Purdom, director of the NESDIS Office of Research and Applications (ORA), cited Fujita s unique role in his remarks upon hearing of his death: b. Winds from 3.9- m image sequences Examination of nighttime 3.9- m winds is under way. The 3.9- m band is more sensitive to lower-tropospheric radiation than the 10.7- m counterpart. Initial results show a marked increase in vector coverage at night over low-level cloudy regions and show promise in helping to fill in data-void regions (see example in Fig. 12). FIG. 12. Comparison of lower-level winds produced by tracking features in half-hourly (left) 4- vs (right) 11- m images. 44 Vol. 82, No. 1, January 2001

Scientists in ORA today, as well as a number of ORA alumni, had the good fortune to collaborate with Ted on many of those research efforts. His SMRP (Satellite Mesometeorology Research Project) placed in print a number of landmark journal articles, and well over 100 SMRP Blue Books that many of us still refer to on a routine basis: science from SMRP was always sound and on the cutting edge.... When one looks back on the rich heritage that Ted leaves behind it is truly incredible mesoscale meteorological analysis can be traced to his brilliant work in the 1950 s, tornado damage classification, original work on satellite derived cloud drift winds, theory and explanation of overshooting thunderstorm tops and severe weather; the downburst was his alone. This list could extend for pages, so perhaps it is enough to point out that in 1967 Ted received the AMS Meisinger Award for pioneering research on mesometeorological analysis and broad contributions to the use of meteorological satellites. In 1985, when the 25th anniversary of weather satellites was underway, at a National Space Club reception at the Smithsonian Institution s National Air and Space Museum, Ted, along with a small group of others, received special awards for their contributions that led to the success of the U.S. weather satellite program. At that event, Ted was cited for creative scientific leadership as an enthusiastic pioneer in the use of satellite imagery to analyze and predict mesoscale weather phenomena and to understand severe thunderstorms, tornadoes, and hurricanes. Acknowledgments. This paper relied on generous contributions from my colleagues Ralph Anderson, Ken Holmlund, Tim Olander, Jim Partacz, Jim Purdom, Brian Soden, and Chris Velden. Two reviewers also provided valuable suggestions. I thank them. References Bradbury, D. L., and T. Fujita, 1968: Computation of height and velocity of clouds from dual, whole sky, time lapse picture sequences. Satellite and Mesometeorology Research Project Paper 70, 17 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.] Campbell, G. G., 1998: Applications of synchronous stereo height and motion analysis. Proceedings of the Fourth International Winds Workshop, EUMETSAT Publ. 24, J. Schmetz and K. Holmlund, Eds., 271 278. Chang, C., J. Tecson and T. Fujita, 1973: METRACOM system of cloud-velocity determination from geostationary satellite pictures. Satellite and Mesometeorology Research Project Paper 110, 29 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.] Eigenwillig, N., and H. Fischer, 1982: Determination of mid-tropospheric wind vectors by tracking pure water vapor structures in Meteosat water vapor image sequences. Bull. Amer. Meteor. Soc., 63, 44 58. Fujita, T., 1961: Outline of a technique for precise rectification of satellite cloud photographs. Mesoscale Research Project Paper 3, 25 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], 1963: A technique for the precise analysis of satellite photographs. Mesoscale Research Project Paper 17, 50 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637], 1964: Evaluation of errors in the graphical rectification of satellite photographs. Mesoscale Research Project Paper 30, 29 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], 1968: Present status of cloud velocity computations from ATS-1 and ATS-3. COSPAR Space Res., 9, 557 570., 1969a: Mesostructure of a subtropical jetstream. Satellite Meteorology: Proceedings of the Inter-regional Seminar on the Interpretation of Meteorological Satellite Data, Melbourne, Australia, Bureau of Meteorology, 67., 1969b: Modification of jetstream by large convective storms. Satellite Meteorology: Proceedings of the Inter-regional Seminar on the Interpretation of Meteorological Satellite Data, Melbourne, Australia, Bureau of Meteorology, 71., 1969c: Interaction between a jetstream and outflows from a hurricane and large rain areas. Satellite Meteorology: Proceedings of the Inter-regional Seminar on the Interpretation of Meteorological Satellite Data, Melbourne, Australia, Bureau of Meteorology, 77., 1969d: Kinematic analysis of Hurricane Brenda, Satellite Meteorology: Proceedings of the Inter-regional Seminar on the Interpretation of Meteorological Satellite Data, Melbourne, Australia, Bureau of Meteorology, 75., 1969e: Outflow from a large tropical cloud mass. Satellite Meteorology: Proceedings of the Inter-regional Seminar on the Interpretation of Meteorological Satellite Data, Melbourne, Australia, Bureau of Meteorology, 73., 1970a: Basic problems on cloud identification related to the design of SMS-GOES spin scan radiometers. Satellite and Mesometeorology Research Project Paper 84, 33 pp. [Available from Wind Engineering Research Center, Texas Tech University, P.O. Box 41023, Lubbock, TX, 79409-1023.], 1970b: Application of ATS-3 photographs for determination of dust and cloud velocities over the northern tropical Atlantic. Satellite and Mesometeorology Research Project Paper 90, 17 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], 1970c: The Lubbock tornadoes: A study of suction spots. Weatherwise, 23, 160 173. 45

, 1972a: Use of ATS pictures in hurricane modification. Satellite and Mesometeorology Research Project Paper 106, 31 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], 1972b: Fotografias y movemiento de nubes de la Haracane Camille. Paper presented at Joint UN/WMO Panel and Training Seminar on the Use of Meteorological Satellite Data, Mexico City, Mexico, 28 November 8 December 1972. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], 1982: Infrared, stereo, cloud motion, and radar-echo analysis of SESAME-day thunderstorms. Proc. 12th Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 213 216., 1991: Interpretation of cloud winds. Proceedings of the First International Winds Workshop, EUMETSAT Publ., 10, 99 104., 1992: The mystery of severe storms. Winds Research Laboratory Paper 239, Dept. of Geophysical Sciences, University of Chicago, 298 pp. [NTIS PB-182021.], 1993: Ground-Truth Experiment of Cloud Drift Winds. Proceedings of the Second International Winds Workshop, EUMETSAT Publ. 14, 21 28., and Uchijima, 1963: Use of TIROS pictures for the internal structure of tropical storms. Mesoscale Research Project Paper 25, 25 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], and D. L. Bradbury, 1964: A study of cumulus clouds over the Flagstaff research network with the use of U-2 photographs. Satellite and Mesometeorology Research Project Paper 33, 25 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], and W. A. Bohan, 1967: Detailed Views of Mesoscale Cloud Patterns Filmed from ATS-1 Pictures. 16 mm, 9 min. Walter A. Bohan Co. [Available from Walter A. Bohan Co., P.O. Box 736, Park Ridge, IL 60068-0736.], and D. L. Bradbury, 1969: Determination of mass outflow from a thunderstorm complex using ATS-3 pictures. Preprints, Sixth Conf. on Severe Local Storms, Chicago, IL, Amer. Meteor. Soc., 38 43., and J. Tecson, 1974: A kinematic analysis of a tropical storm based on ATS cloud motions. Satellite and Mesometeorology Research Project Paper 125, 20 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], D. L. Bradbury, C. Murino, and L. Hull, 1968: A study of mesoscale cloud motions computed from ATS-1 and terrestrial photographs. Satellite and Mesometeorology Research Project Paper 71, 25 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], K. Watanabe, and T. Izawa, 1969: Formation and structure of equatorial anticyclones caused by large-scale cross-equatorial flows determined by ATS-1 photographs. J. Appl. Meteor., 8, 649 667., E. Pearl, and W. Shenk, 1975: Satellite-tracked cumulus velocities. J. Appl. Meteor., 14, 407 413. Gentry, R., T. Fujita, and R. Sheets, 1970: Aircraft, spacecraft, satellite, and radar observations of Hurricane Gladys. J. Appl. Meteor., 9, 837 860. Goerss, J. S., 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, 1202 1218. Hasler, A. F., K. Palaniappan, C. Kambhammetu, P. Black, E. Uhlhorn, and D. Chesters, 1998: High-resolution wind fields within the inner core and eye of a mature tropical cyclone from GOES 1-min images. Bull. Amer. Meteor. Soc., 79, 2483 2496. Hayden, C. M., and R. J. Purser, 1995: Recursive filter objective analysis of meteorological fields Applications to NESDIS operational processing. J. Appl. Meteor., 34, 3 15., and S. J. Nieman, 1996: A primer for tuning the automated quality control system and for verifying satellite-measured drift winds. NOAA Tech. Memo. NESDIS 43. 27 pp. [Available from National Technical Information Service, 5285 Port Royal Road, Springfield, VA 22161.] Holmlund, K., 1993: Operational water vapor wind vectors from Meteosat imagery data. Proc. Second Int. Wind Workshop, Tokyo, Japan, EUMETSAT Publ. 14, 77 84., C. S. Velden, and M. Rohn, 2000: Enhanced automated quality control applied to high-density GOES winds derived during the North Pacific Experiment (NORPEX). Mon. Wea. Rev., in press. Hubert, L. F. and L. F. Whitney Jr., 1971: Wind estimation from geostationary-satellite pictures. Mon. Wea. Rev., 99, 665 672. Langland, R., and Coauthors, 1999: The North Pacific Experiment (NORPEX-98): Targeted observations for improved North American weather forecasts. Bull. Amer. Meteor. Soc., 80, 1363 1384. Menzel, W. P., W. L. Smith, and T. R. Stewart, 1983: Improved cloud motion wind 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, D. Gray, S. T. Wanzong, C. S. Velden, and J. Daniels, 1997: Fully automated cloud drift winds in NESDIS operations. Bull. Amer. Meteor. Soc., 78, 1121 1133. Schmetz, J., K. Holmlund, J. Hoffman, B. Strauss, B. Mason, V. Gaertner, A. Koch, and L. van de Berg, 1993: Operational cloud motion winds from METEOSAT infrared images. J. Appl. Meteor., 32, 1206 1225., H. P. Roesli, and W. P. Menzel, 1997: Summary of the Third International Winds Workshop. Bull. Amer. Meteor. Soc., 78, 893 896., D. Hinsman, and W. P. Menzel, 1999: Summary of the Fourth International Winds Workshop. Bull. Amer. Meteor. Soc., 80, 893 899. Soden, B. J., C. S. Velden, and R. E. Tuleya, 2000: The impact of satellite winds on GFDL hurricane model forecasts. Mon. Wea. Rev., in press. Suomi, V. E., 1969: Recent developments in satellite techniques for observing and sensing the atmosphere. The Global Circulation of the Atmosphere, G. A. Corby, Ed., Royal Meteorological Society, 222 234. 46 Vol. 82, No. 1, January 2001

, and R. Parent, 1968: A color view of Planet Earth. Bull. Amer. Meteor. Soc., 49, 74 75., R. Fox, S. S. Limaye, and W. L. Smith, 1983: McIDAS III: A Modern Interactive Data Access and Analysis System. J. Climate Appl. Meteor., 22, 766 778. Szejwach, G., 1982: Determination of semi-transparent cirrus cloud temperatures from infrared radiances: Application to Meteosat. J. Appl. Meteor., 21, 384 393. Tecson, J., and T. Fujita, 1975: Cloud-motion vectors over the GATE area computed by McIDAS and METRACOM methods. Satellite and Mesometeorology Research Project Paper 136, 23 pp. [Available from Department of Geophysical Sciences, The University of Chicago, 5734 S. Ellis Ave., Chicago, IL 60637.], C. Chang, and T. Fujita, 1973: Cloud motion field of hurricane Ginger during the seeding period as determined by the METRACOM system. Eighth Tech. Conf. on Hurricanes and Tropical Meteorology, Key Biscayne, FL, Amer. Meteor. Soc., 46 51., F. Umenhofer, and T. Fujita, 1977: Thunderstorm associated cloud motions as computed from 5-minute SMS pictures. Preprints, 10th Conf. on Severe Local Storms, Omaha, NE, Amer. Meteor. Soc., 22 28. Velden, C. S., C. M. Hayden, W. P. Menzel, J. L. Franklin, and J. S. Lynch, 1992: The impact of satellite-derived winds on numerical hurricane track forecasting. Wea. Forecasting, 7,107 118.,, S. J. Nieman, W. P. Menzel, and S. Wanzong, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc., 78, 173 195., T. L. Olander, and S. Wanzong, 1998: The impact of multispectral GOES-8 wind information on Atlantic tropical cyclone track forecasts in 1995. Part I: Dataset methodology, description, and case analysis. Mon. Wea. Rev., 126, 1202 1218., D. Stettner, and J. Daniels, 2000: Wind vector fields derived from GOES rapid-scan imagery. Preprints, 10th Conf. on Satellite Meteorology, Long Beach, CA, Amer. Meteor. Soc., 20 23. 47