Aerospace Photos Capture Launch Clouds

Launch of a Titan IVB rocket from Cape Canaveral Air Force Station

Launch of a Titan IVA rocket from Cape Canaveral Air Force Station. (USAF)

Aerospace Photos Capture Launch Clouds

Robert Abernathy

A new and improved method of measuring launch-vehicle ground clouds leads to fewer launch delays and reduces costs.

The spectacular display of billowing smoke that envelops the launchpad during a rocket launch has become synonymous with the launch itself. Rockets release immense amounts of exhaust. Titan IVB solid-rocket motors emit 118,000 pounds of exhaust during the first 10 seconds of firing. The hot exhaust accumulates at the launchpad to form what is called the "ground cloud." As the rocket ascends, it also leaves behind a continuous stream of exhaust, known as the launch column.

During countdown to the launch, the U.S. Air Force Range Safety uses an onsite computer model known as the Rocket Exhaust Effluent Dispersion Model (REEDM) to predict the rise and dispersion of the expected ground cloud. REEDM relies on meteorological data such as wind, cloud cover, solar angle, and weather-front conditions to predict the extent of the toxic hazard corridor, the downwind area where ground concentrations of chemicals may exceed allowable public exposure limits. When REEDM predicts the launch may cause exposure to unsafe levels of toxic gases, the launch is delayed until meteorological conditions improve.

When REEDM was being developed during the 1960s and 1970s, several cloud-transport parameters were not well known, so the model was deliberately designed to be conservative. Over the years, the level of toxic compounds considered acceptable for public exposure has been lowered, increasing the likelihood of a launch delay. By 1994, toxic hazard corridor predictions were beginning to reduce launch availability at both Air Force launch ranges, the Eastern Range at Cape Canaveral Air Force Station and the Western Range at Vandenberg Air Force Base.

The high cost of launch delays, up to $1 million a day, and the continued concern for public welfare prompted development of the Air Force Atmospheric Dispersion Model Validation Program (MVP) to test and improve REEDM's accuracy. The Aerospace Corporation developed this validation program and provides technical management.

An Aerospace method of monitoring exhaust clouds using photographic imagery showed that REEDM consistently underestimated the ground-cloud stabilization height and overestimated the extent of the toxic hazard corridor. With subsequent Aerospace modifications, REEDM predictions have improved launch-range availability, preventing unnecessary launch holds and saving the government millions of dollars while protecting the public safety.

The Aerospace Corporation's imagery crew tracking a Titan IV ground cloud at the Eastern Range at Cape Canaveral Air Force Station. The vehicle assembly building on the left and the mobile service tower next to the ground cloud served as useful calibration landmarks. The rocket's launch column extends above the ground cloud.

Monitoring Ground Clouds

Ground clouds are difficult to monitor. Within the first few minutes after a launch, they grow to dimensions of 1–2 kilometers and rise to similar heights. Also, although the optimal wind for a launch carries the ground cloud out to sea, such a wind direction does not allow for the use of ground-based launch-cloud sampling systems.

Aircraft instrumented to sample and measure ground-cloud concentrations of toxic compounds have been used during launches of the space shuttle, but aircraft sampling is expensive and doesn't provide an instantaneous three-dimensional extent for the cloud. Additionally, this method of measurement typically relies on the pilot's ability to fly into the center of the visible cloud, which precludes accurate nighttime sampling. Without visible feedback, the pilot doesn't know the aircraft's location relative to the center of the cloud, so concentration measurements are extracted from unknown regions.

Aerospace proposed an alternative approach to monitoring the rise and expansion of exhaust clouds, and in 1994 it developed the technology to track ground clouds using multiple cameras that capture images simultaneously from various locations surrounding the launch pad. Because a ground cloud rises and stabilizes quickly, the cloud needs to be tracked for only a few minutes following launch. The images captured during the tracking complement measurements garnered from aircraft sampling.

Day and night images of the launch cloud were needed to test REEDM under all launch conditions. Visible and infrared imagery would be captured, using visible charge-coupled-device cameras and thermal infrared scanners. In daytime, the visible cameras "see" the scattering of sunlight caused by aerosols from the solid-rocket motors. Throughout the day and night, the infrared scanners observe the temperature difference between the warm launch cloud (vapors) and the cooler background sky.

Cameras provide better-resolution images than infrared scanners, but the quality of camera imagery is subject to adequate lighting, which depends upon the relative positions of the sun, cloud, and camera. The camera can also provide clearer cloud-edge-detection imagery of a low-elevation cloud on a hot sticky day because atmospheric humidity and cloud elevation affect the quality of the infrared imagery.

Because the two imaging systems complement each other, Aerospace designed and built four visible and infrared imaging systems (VIRIS) in which the visible camera and the infrared scanner are mounted on a single tripod. The camera and infrared scanner are "coaligned," meaning the center pixel of each camera simultaneously views the same distant object. The four systems were shipped alternately to Cape Canaveral and Vandenberg for use during Titan IV launches from both launch ranges.

PLMTRACK's analysis of blimp imagery from two sites. The red pixel (+) is projected as a blue ray, and the blue pixel (+) as a red ray, in the sister images. The Good Year blimp was used to test the accuracy of both PLMTRACK and PLMVOL.

Calibrating VIRIS

Aerospace designed custom tripod heads that accurately encode, or digitize, the viewing direction (azimuth and elevation) as VIRIS tracks a cloud. Optimal camera locations depend upon wind direction, so they cannot be selected until a few hours before launch. The challenge is to quickly calibrate the systems, aligning a known pixel with true north for zero azimuth and level for zero elevation (see sidebar, Calibrating Amateur Abort Cloud Imagery).

Calibrating the angle encoders of the tripods requires the camera crew to identify large landmarks that can be observed by both the cameras and the infrared scanners. Before the imagery systems are deployed, a survey provides accurate position information on the observable landmarks, using either a differentially corrected Global Positioning System (GPS) receiver or accurate maps of the launch range. Once the camera is set up, the GPS receiver provides the camera-site location.

The camera crew calculates the azimuth and elevation from the camera to the landmark. Once the center pixel is aligned with the landmark, it is a simple matter to set the correct azimuth and elevation readout from the tripod. The field of view is calibrated by scanning the landmark horizontally and vertically while recording the change in encoded azimuth and elevation.

Typically, camera crews can set up and calibrate to 0.1 degree of accuracy within 45 minutes. Each calibrated imagery system provides a real-time display of the azimuth and elevation to the ground cloud from each camera's perspective. Hence, the ground cloud's approximate position can be triangulated in real time using the pointing angles to the cloud from all sites.

PLMTRACK-derived Cartesian extent and position from analysis of multiperspective imagery.

Triangulating Cloud Position

Aerospace developed PLMTRACK software to triangulate the position and extent of a ground cloud with imagery captured simultaneously from any two sites. Once an image is calibrated, each pixel represents a ray into space from the camera's position. PLMTRACK converts a selected pixel in one image (the top of the cloud, for example) into an azimuth and elevation from that image's camera location. It then projects that ray across the simultaneous image from the other site. The analyst identifies the same feature (top of the cloud) in the sister image using the projected ray for perspective, and the same feature is thereby seen from two perspectives. PLMTRACK converts this information into a ray from each site that passes through the same feature and calculates the closest approach of the two rays, which represents the position of the selected feature in three-dimensional space. This approach works well when an object or feature can be observed from two perspectives.

Triangulating the top and bottom of a cloud at low elevations is easy because both top and bottom are observable from multiple perspectives. But how is the horizontal extent of the cloud determined when the same sides cannot be seen from both perspectives? Where is the "middle" of the cloud? PLMTRACK allows the analyst to use a rectangle to define the top, bottom, left, and right extremes of the cloud from each camera's perspective, and the projection of these rays provides an estimate of the extent for the cloud. The middle of the rectangle provides the ray through the middle of the cloud from each site, and the nearest approach of these middle rays represents the position of the cloud.

Validating Multicamera Imagery

Neither the camera nor the infrared scanner directly detects the toxic hydrochloric acid in rocket exhaust. For this reason, it is important to document not only that the cameras and infrared scanners see the same extent (angular size) of the ground cloud, but also that the observable (seen by VIRIS) extent contains the toxic acid that might pose a hazard.

Aerospace images obtained from the coaligned camera and infrared scanner have consistently shown the same angular extent for the Titan IV ground clouds, verifying that both image-capturing methods are similarly useful when applied to tracking ground clouds. However, these observations do not prove that hazardous levels of toxic chemicals do not extend beyond the observable extent of the cloud.

Aircraft sampling of hydrochloric acid within four Titan IV exhaust clouds provided the complementary data needed to validate the coaligned multicamera imagery. An aircraft was fitted with a Geomet hydrochloric-acid monitor to obtain concentration profiles during four launches between May 1995 and December 1996—two each from Cape Canaveral and Vandenberg. Comparison of the aircraft data and imagery showed that the observable extent contained the measurable acid (in the form of both vapor and aerosol). These observations are consistent with the mechanism of atmospheric dispersion: Atmospheric eddies mix vapors (seen in the infrared) and aerosols (seen in the visible) equally well. These results show that VIRIS provided the cloud's extent for Titan IV day and night launches, and that the extent includes the hazardous levels of hydrochloric acid.

The ground cloud and the abort cloud are produced by a failed launch. If the abort occurs several hundred feet above ground, REEDM may predict a larger toxic hazard corridor for the unburned oxidizer than for the ground cloud. (USAF)

Predicting Ground-Cloud Stabilization Height

Because it is initially warmer than the surrounding air, the ground cloud rises. As it does, it entrains the ambient air, which causes it to cool and lose buoyancy. Within three to four minutes, the cloud reaches its "stabilization height," where it attains thermal equilibrium with the surrounding air and stops rising. The 1995 version of REEDM was used to predict the height of the ground cloud prior to the May 14, 1995, launch of a Titan IV. The REEDM prediction underestimated the stabilization height of the cloud by half, which corresponds to overestimating the ground-level concentration by a factor of eight. During the next three years, the Aerospace Titan IV ground-cloud imagery consistently showed a difference between the observed and predicted stabilization heights for 13 launches from both launch ranges.

Because the stabilization heights derived from the images consistently remained much higher than REEDM's predictions, the REEDM code was reviewed, revealing several errors. Yet even after these errors were corrected, predictions of stabilization height remained too low. Speculating that the values of two volumetric parameters in the cloud-rise algorithm might be wrong, Aerospace focused its image-analysis efforts on the accurate measurement of the cloud's volume immediately after launch and during the cloud's rise. The desired volumetric parameters, which are simply the initial radius of the ground cloud and the rate of increase in radius with altitude (the air entrainment coefficient), would come directly from these measurements.

Reconstructing the Cloud

Aerospace developed PLMVOL, a software application based on a second analysis algorithm, which reconstructs the three-dimensional cloud from the two-dimensional imagery collected simultaneously at multiple locations. First, the simultaneous imagery from all available locations is digitized and imported with the calibration information into PLMVOL. Then the analyst traces the outline of the exhaust cloud within each image.

Next, PLMVOL converts the pixels within these outlines into rays projected into space from each camera's location. The exhaust cloud is located at the intersection of these rays. To derive the points where the rays intercept, PLMVOL divides space into small cubes and marks them as occupied by the ground cloud only when intercepted by rays from all available perspectives. It maps the three-dimensional extent of the ground cloud as the Cartesian locations (x,y,z) of all the occupied volume elements. Since the volume elements are adjacent (stacked cubes), summing all occupied volume elements yields the imagery-derived volume of the ground cloud.

PLMVOL's cloud volume (purple shape) is mapped by projection of pixels (rays) into volume elements. The rays are identified as outside (dotted) or inside (solid) the ground cloud's outline (red line) from each perspective. Volume elements are occupied by the ground cloud only when intercepted by "inside" rays from all available perspectives.

Finally, the sphere-equivalent radius, used by REEDM, is calculated from the imagery-derived cloud volume by determining the radius of a sphere with an equivalent volume. Typically, the Titan IV ground cloud is not spherical in shape, but the sphere-equivalent radius is a convenient unit for comparison. The accuracy of PLMVOL estimates depends upon the relative position of the camera sites to the ground cloud's position. PLMVOL can't provide an accurate cloud volume if cameras do not see the ground cloud from complementary perspectives, for example along-wind and crosswind perspectives, simultaneously.

Aerospace used PLMVOL to extract cloud-volume data for only 6 of 13 imaged Titan IV ground clouds. Several factors led to this low yield of volumetric data. For example, the cloud didn't always travel in the predicted direction, Vandenberg restricted access to camera sites to the east and south of the launch pad, and low atmospheric clouds blocked visibility from one or more of the sites. In sum, the available camera locations and visibility did not always provide the complementary perspectives needed to accurately map the cloud volume.

Analyzing Amateur Imagery

REEDM also predicts toxic exposure from a low-altitude launch-vehicle abort. In the event of an abort, the launch vehicle would be destroyed in an explosion that releases a hypergolic mixture of liquid fuel and oxidizer. If the abort occurs several hundred feet above the ground, REEDM may predict a larger toxic hazard corridor for the unburned oxidizer than for the ground cloud.

Aerospace proved that normal launch-cloud data could be applied to the abort-cloud scenario. It measured the behavior (rise, growth, and stabilization) of normal launch clouds during launches from 1994 and 1997. Abort clouds weren't measured because no Titan IVs failed during deployment. Without abort-cloud data, the accuracy of REEDM predictions for an abort situation could not be validated nor could the ground cloud's entrainment data be shown to apply to the abort cloud.

The few options for obtaining the necessary data were considered in 1997. One possibility was to use an explosive release of oxidizer to simulate the abort cloud because the most toxic component of the abort cloud is unburned oxidizer. Safety concerns, however, limited test sites to desert locations that did not match the terrain or the meteorological conditions of the launch ranges. Two other options were to continue imaging Titan IV launches on the chance of a failure or search for imagery of earlier aborts hoping to interpret that imagery quantitatively. We chose the latter.

The REEDM-predicted and Aerospace-imagery-derived cloud-height curves for a May 14, 1995, Titan IV launch. The imagery-derived rise curve for the ground cloud revealed a factor-of-two discrepancy between measured and predicted stabilization height. Subsequent MVP deployments documented that REEDM systematically underestimated the stabilization heights for all 13 Titan IV ground clouds at both ranges under a variety of launch conditions.

A worst-case abort scenario was inadvertently tested on April 18, 1986, when a defect in a Titan 34D-9 solid rocket motor caused the rocket to explode at an elevation of 830 feet at Vandenberg. Aerospace reviewed the videotapes from the three range-tracking cameras. Unfortunately, the camera operators at two locations did not keep the abort cloud completely within the field of view. To obtain a cloud's volume, which involves analyzing the abort-cloud imagery from the third range camera, a second, complementary perspective of the abort cloud had to be available. Without abort-cloud imagery from a second camera, the abort-cloud imagery from the third range camera could not be interpreted quantitatively (as cloud position and volume).

Luckily, an amateur photographer videotaped the launch and its abort cloud with a handheld camcorder. This photography provided the necessary second, nearly perpendicular, perspective. Interpretation of the abort-cloud imagery was complicated because both cameras were panned and zoomed several times during the three minutes the images were captured. Neither camera was mounted on an angle-encoding tripod, nobody intentionally calibrated the field of view of the cameras, and the camera operators did not realize that abort clouds would later be of more interest than burning ground debris. Fortunately, Aerospace was able to calibrate much of the abort-cloud imagery and used PLMVOL to quantify the position and volume of the abort cloud during its rise.

Improving Model Predictions

Aerospace imagery-derived air entrainment rates for the Titan 34D-9 abort cloud measured at Vandenberg and for normal Titan IV ground clouds measured at both Cape Canaveral and Vandenberg were substantially lower than the default value used in REEDM calculations during the past 30 years. These values indicate that REEDM-based predictions of ground-cloud stabilization height have been consistently too low and toxic-hazard predictions too high. The imagery-derived results also show that the air entrainment coefficient and the initial cloud size are constants for the Titan IV normal launch cloud and have the same value for both launch ranges and for both sets of solid rocket motors. The coefficient is the same for the 34D-9 abort cloud, which indicates similar behavior for both normal and abort clouds.

Aerospace imagery-derived air-entrainment rates for normal Titan IV ground clouds and for the Titan 34D-9 abort cloud. The values are substantially lower than the default value used in REEDM calculations during the past 30 years. Legend: A—solid rocket motor; B—upgraded solid rocket motor; C—Cape Canaveral Air Force Station; V—Vandenberg Air Force Base.

The current version of REEDM (7.09) provides improved stabilization height predictions through the use of Aerospace imagery-derived values for both the air entrainment coefficient and the initial radius. The new REEDM predictions, which are in closer agreement with the observed launch-cloud stabilization heights, have improved launch-range availability by preventing unnecessary launch holds.

In addition, a Titan IV database now establishes the margin of safety for current and future dispersion models. The MVP database includes quantitative analysis of imagery from nine Titan IV launches at Cape Canaveral and four at Vandenberg between 1994 and 1997. MVP deployments involved collecting meteorological data, necessary for running REEDM or improved future dispersion models.

Aircraft samples were taken from two Cape Canaveral and two Vandenberg Titan IV launches during MVP. Aerospace analysis of these aircraft data revealed that both the visible and the infrared imagery "see" the full extent of the cloud containing hydrochloric acid during the first few minutes after launch. This means that the observable aerosol and vapor disperse at the same rate as the unobservable acid, which is consistent with the behavior of aerosols and the mechanism of turbulent dispersion in the atmosphere.

Tracking Tracer Gas

The Aerospace imagery of the Titan IV launches provided useful cloud rise and stabilization data under favorable meteorological conditions, that is, when winds carried the ground cloud out to sea or over unpopulated areas. Aerospace imagery crews at Cape Canaveral and Vandenberg supported four two-week-long elevated-tracer-gas releases that provided complementary dispersion data, including winds that carried the innocuous tracer toward populated areas. During these MVP efforts, Aerospace established the usefulness of quantitative imagery for measuring the near-field (2–5 kilometers) dispersion of tracer gases.

The Aerospace surveillance technology crew in front of a mobile laboratory. These mobile laboratories, equipped with visible and infrared imagery systems, support remote detection and tracking of chemicals, such as those in launch abort clouds, bomb detonations, and tracer release experiments. They are deployed at launch and test ranges throughout the continental United States. Shown in the photo from left to right, beginning with the back row (in doorway): Bruce A. Rockie, Luis J. Ortega, Michael A. Rocha; left center row: Gary N. Harper, Brian P. Kasper, Karl R. Westberg, Jess T. Valero; right center row: Robert N. Abernathy, Kenneth C. Herr, Jeffrey L. Hall, Donald K. Stone; front row: Mark L. Polak, Andrew D. Shearon, J. Thomas Knudtson, Naomi J. Rose, George J. Scherer, Roberta S. Precious, Karen L. Foster.

A blimp released an invisible inert tracer gas at various heights when the wind was blowing inland. This allowed for dispersion measurements over the complex inland terrain of both ranges. Analysis of the infrared imagery provided the crosswind and along-wind expansion rates in the near field at the release altitude. During these elevated-tracer-release experiments, aircraft and van sampling provided trajectory and dispersion information further afield. These tracer data complement the Titan IV launch-cloud data.

Predicting Ground Clouds in the Future

The ability of Aerospace to capture and process quantitative imagery of Titan ground clouds has provided, at a low cost to the consumer, the rise and dispersion data necessary to tune REEDM for more accurate prediction of ground-cloud toxic hazard corridors. Such accurate prediction also reduces the launch costs because it leads to fewer launch holds. A similar measurement program could be used to tune current and future dispersion models for the other heavy launch vehicles, such as the space shuttle today and the Evolved Expendable Launch Vehicle in the future. In addition, routine imagery of launch clouds could provide real-time range-safety information, not only for normal launch clouds but also for the more toxic abort cloud.