Atmospheric Science at NASA: A History

• 3 Constructing a Global Meteorology
• Chapter
• Johns Hopkins University Press
• pp. 64-93
• View Citation
• Additional Information

CHAPTER THREE

Constructing a Global Meteorology

In early 1961, President John F. Kennedy’s science advisor, MIT physicist Jerome Wiesner, had asked the National Academy of Sciences’ Committee on Atmospheric Science to propose a ten-year program for the profession. The report, drafted by meteorologist Sverre Petterssen, called for establishment of a set of new international institutions to further expand meteorology’s reach.¹ An International Atmospheric Science Program should carry out scientific research on the global atmosphere, an International Meteorological Service Program should provide global-scale forecasting, and a World Weather Watch should sustain a global atmospheric observation system. These recommendations formed one basis of UN Resolution 1721, “International Co-Operation in the Peaceful Uses of Outer Space.”²

In 1950, the United Nations had established a World Meteorological Organization (WMO), and this was the entity tasked with implementing the meteorological portions of Resolution 1721.³ The WMO staff generated a report recommending the establishment of the World Weather Watch’s observation capability within the WMO, while arguing that the atmospheric research program was properly the domain of the International Council of Scientific Unions (ICSU). Resolution 1802, adopted in December 1962, had accepted this arrangement.

The principal architects of the WMO’s report were the Soviet Union’s academician V. A. Bugaev and the Weather Bureau’s chief of research, Harry Wexler. Wexler had been personally involved in postwar meteorology’s first great advancement, the construction of workable numerical weather prediction models during the late 1940s and through the 1950s, and was an important promoter of scientific internationalism via his International Geophysical Year (IGY) efforts.⁴ He was also, of course, a supporter of satellite-based meteorology, funding Vern Suomi’s first satellite instrument for the IGY, establishing the Weather Bureau’s satellite center, and, until his untimely death in 1962, serving as a highly respected advocate of satellite meteorology. He had believed that the union of these two great new technologies of numerical prediction and satellite data could produce truly global forecasts that might eventually be accurate to periods of a month.

NASA’s role in the Global Atmospheric Research Program (GARP) that evolved during the late 1960s was as a provider of new technologies and support for large-scale field experiments. GARP’s principal purpose was to provide quality-controlled global meteorological datasets for use in improving future numerical prediction models, and therefore the goals of the program were unobtainable without space-based measurements and telecommunications. But GARP required field experiments, too, to provide ground truth for the space-based measurements and to collect data that could not yet be gained from satellites. In keeping with the longstanding geophysical tradition of field science, these expeditions were international in organization and very large in scope. GARP’s final field experiment, the First GARP Global Experiment (FGGE) of 1979, finally achieved Charney’s dream of a global, quality-controlled, extensive meteorological dataset.

FORMING GARP

Science writer James Gleick, in his history of chaos science, remarked that GARP was founded in “years of unreal optimism about weather forecasts.” There was, he continued, “an idea that human society would free itself from weather’s turmoil and become its master instead of its victim.”⁵ The linkage of satellite observation to numerical forecasting would, in this view, permit not only month-long forecasts but eventually weather control. John von Neumann had believed that weather control could eventually result from this research area, and he was hardly alone. Robert M. White, head of the Weather Bureau and eventually the National Oceanic and Atmospheric Administration (NOAA), then believed weather control was within reach. Nobel Laureate Irving Langmuir did too, spending many years on cloud seeding research. NASA’s Homer Newell was routinely asked about the relevance of NASA’s meteorological research for weather control during his annual budget testimony before Congress, suggesting the importance of the issue to the agency’s funders.

The program that became GARP started as an American initiative. Acting in its capacity as U.S. representative to the ICSU, the National Academy of Sciences had asked Jule Charney to serve as the organizer of the American proposal for WMO’s research program. Charney’s plan was based on three principles: the atmosphere was a single system such that disturbances in one area propagated around the world in four to five days; a new approach to observational techniques based on both satellite-derived quantitative data and satellite-relayed in situ data was necessary to improve prediction; finally, high-speed digital computers were capable of coping with the torrent of data such satellites could provide.⁶ Its observing system was based on Vincent Lally’s constant-level satellite-balloon system, at this point still an unknown quantity.

In March 1964, Morris Tepper, Jule Charney, and Philip Thompson visited colleagues in London, Paris, Geneva, and Brussels to explain Charney’s proposal and seek their reactions. Here they found great enthusiasm for Charney’s plan, with the only expressed concern being the balloon subprogram. The French government’s space research establishment, Centre National d’Études Spatiale (CNES), had already gotten approval for a satellite-balloon tracking experiment that was originally to be launched in 1967 from Algeria.⁷ During negotiations over who would pay for the eventual operational global observing system, the French government had agreed to fund the satellite-balloon subsystem, and EOLE was the result.

The balloon program, however, raised two potential challenges. First, the European scientists all emphasized that the potential impact of the balloons on aircraft needed to be investigated. Jet aircraft operated at altitudes near those the balloons needed to be at to provide data on the desired wind velocities, and the balloon payloads needed to be designed so that they would not damage aircraft. The second was that the balloons probably could not fly over the Soviet Union. During the 1950s, the United States had flown intelligence cameras and radiation detectors over the USSR, causing a diplomatic fiasco that resulted in complete ban on balloon overflights. This meant that the Global Horizontal Sounding Technique (GHOST) and EOLE balloon flights would have to be confined to the Southern Hemisphere. This restriction would significantly impair the utility of the balloon system for the operational global observing system unless the USSR could be recruited into the effort.

Three months after the three American delegates’ visit to Europe, ICSU agreed to form a Committee on Atmospheric Sciences to plan the global experiment. At its first meeting in Geneva, held during February 1965, this committee agreed that the research program should be directed at understanding the general circulation of the troposphere and lower stratosphere and should contain two elements.⁸ In a theoretical element, the program should develop dynamical, that is, numerical, models of the general circulation of the atmosphere that included radiation, momentum, and moisture movement on local and regional scales as well as the global scale. Second, the program should specify the observational needs of global atmospheric research, including the technological capabilities of its sensors and its telecommunications system, and carry out full-scale observation programs over time-limited periods.

Prior to the second meeting of the ICSU’s Committee on Atmospheric Sciences in April 1966, Charney had assembled the formal American proposal for the research program. Titled The Feasibility of a Global Observation and Analysis Experiment, Charney’s proposal became known as the Blue Book for the color of its cover.⁹ It also divided the research program into two problem areas: exploitation of space and data processing technologies to provide global observations, and improvement of scientific understanding of turbulent transport of matter and energy in the atmosphere. ICSU’s committee, in turn, proposed carrying out the global experiment in 1972. The year would be “designated as a twelve-month period for an intensive, international, observational study and analysis of the global circulation in the troposphere and lower stratosphere.” In preparation, researchers would carry out a series of other investigations. Tropical circulation was poorly understood, and as the tropics were where most of the Sun’s radiation reached Earth, a tropical subprogram was essential. Energy exchange between the atmosphere and land and ocean surfaces was also a poorly understood process, but one essential to accurate long-range weather prediction, and an observational program to determine the dynamics of these energy flows was vital. Finally, design studies of a global observing system that could meet the scientific needs of the research program had to be carried out.

By the end of the year, however, it was already clear to the committee’s members that their chosen date was highly unrealistic. Neither the satellite-based temperature profile instrument nor the balloon-tracking system would reach space before 1969 due to the loss of Nimbus B; even if they worked as expected, their project scientists would need several more years to understand their capabilities and limitations. Subsequent, improved instruments would not be available until 1974 or 1975. No one expected that the first pair of operational geosynchronous satellites would be available until those later dates either—and it was absurd to believe in 1966, when satellites routinely failed in a few months, that the spin-scan cameras on the two Advanced Technology Satellite (ATS) satellites would still be sending back pictures in 1972. Furthermore, the subprograms themselves were going to require a good deal of effort. Design studies for the global observing system required numerical simulation on large computers whose time was expensive and often difficult to acquire. The tropical subprogram would involve an international flotilla of ships that had to be loaned by national governments, a complex, time-consuming process. For all these reasons, the global experiment had to be postponed to 1976.

In early March 1967, at the third meeting of the Committee on Atmospheric Sciences, the scientist-delegates began to discuss the details of what they now called GARP. NASA’s Morris Tepper, chairman of the Committee on Space Research’s meteorological subcommittee, had established three panels to look at different aspects of the future observing system at the previous meeting, and at this meeting the chairmen of these panels presented their findings. UCLA meteorologist Yale Mintz, who specialized in numerical modeling, told the committee that what modelers needed was a set of global observations of the atmosphere extending over a few months, up to a year. Such a dataset would provide a detailed, global snapshot of the atmosphere that model researchers could use to initialize global prediction models and a set of real-world results to compare to the models’ output forecasts. This was the only way the models could be improved. William Nordberg presented the status of temperature profile instruments for satellites. And J. E. Blamont, from France’s Centre National de la Recherché Scientifique, presented the status of the satellite-balloon research. Preliminary experiments with the National Center for Atmospheric Research’s (NCAR) GHOST and the French EOLE balloon systems, had shown mixed results. Flights at high altitudes had gone relatively well, with some of the balloons surviving more than two hundred days. But lower-altitude balloons (500 millibars and below) tended to ice up. This panel concluded that “there [was] little likelihood of the availability of a global balloon-satellite observing system by 1972.”¹⁰

The outcome of this meeting was a set of recommendations on the structure, timing, and contents of the proposed global program. Completion dates for the major field experiment should be moved to 1972–73 and for the final global experiment to 1975–76. The group asserted that a large-scale tropical observation subprogram should be the primary field experiment, to be carried out in the 1972–73 period. Finally, they recommended that the somewhat unwieldy committee structure be replaced by a special joint scientific committee that could provide a unified front for GARP and that could carry it out relatively unhindered by the three organizations that supported it (ICSU, WMO, and the International Union of Geodesy and Geophysics [IUGG]).¹¹ Their recommendations were accepted by the three sponsoring organizations later that year, and the new committee became the GARP Joint Organizing Committee.¹²

This third meeting left the details of the global experiment unplanned, however, and the Committee on Atmospheric Science’s chairman, Bert Bolin of the Stockholm Meteorological Institute, another veteran of Charney’s numerical group at Princeton, arranged for a Study Conference to be held in Stockholm in early July 1967 to complete them. He invited specialists in all the different subfields of meteorology that the global research program had to address—boundary layer flux, air-sea interaction, convective processes, meso-scale phenomena, atmospheric radiation—and have them work with the numerical modelers to define the program.¹³ At this conference, the global program took its (mostly) final form. The conference ratified the importance of the tropical subprogram and pushed its date back to 1974, when they hoped better satellite instruments and the satellite-balloon system would finally be available, although the group retained the 1975–76 date for the global experiment.

The two-year American budget formulation cycle ensured that nothing much happened to get GARP going until 1969, however. In the words of NOAA’s Robert White, NOAA’s GARP office had to “mark time” while waiting for funds to come through. This had the fortunate result that GARP offices at NASA and NOAA received their go-aheads just as one element of the future observing system, the infrared sounder, got its first successful space-borne test. The successful retrieval of tropospheric temperatures by the Nimbus 3’s Satellite InfraRed Sounder (SIRS) instrument team served as an additional stimulus to American GARP efforts. Morris Tepper took the retrievals up to a meeting with Robert Jastrow, Jule Charney, and Milton Halem in Jastrow’s office at the Goddard Institute for Space Studies (GISS) in New York, where the data convinced the three men that it was finally time to start carrying out the detailed design studies that would result in the eventual GARP experimental observation system.¹⁴ The Nimbus 5 and 6 launches with improved temperature sounders and balloon tracking systems were already in the development pipeline, and these were scheduled to be in orbit by the time all the rest of the infrastructure necessary to carry out the research program was in place.

After Tepper’s meeting at GISS, he established a planning committee to map out a GARP strategy for NASA. He obtained permission to establish a GARP project office at Goddard Space Flight Center in Maryland, with Harry Press as the project manager and Robert Jastrow the project scientist. At the 1969 meeting of the Joint Organizing Committee, the structure of GARP was finalized, and NASA became responsible for specific tasks within it. GARP would consist of the large tropical experiment known as GARP Atlantic Tropical Experiment (GATE); a Data Systems Test (DST) that would carry out an evaluation of the global observing system; and the FGGE, which would produce its first global datasets. In the United States, the National Academy of Sciences was tasked with handling the planning and Academy president Philip Handler appointed Jule Charney to chair the U.S. Committee for GARP. NOAA became lead agency, with NASA responsible for the hardware development for the global observing system and for the DST. The space agency was also responsible for carrying out simulation studies necessary to support the detailed planning for the DST and the FGGE.¹⁵ Finally, NOAA contracted the planning of the tropical field experiment to NCAR in Colorado.

SIMULATION STUDIES

A crucial component of GARP planning were simulation studies carried out by GISS, the Geophysical Fluid Dynamics Laboratory (GFDL) at Princeton, and NCAR. Using numerical prediction models, these studies addressed two important questions: the optimum configuration of the future global observing system, and the realistic time horizon of predictions using it. The first question would affect the technologies chosen for the global observing system, and how much building and operating it would cost. The second was aimed at understanding what GARP actually had the potential to achieve. Despite the enthusiasm for month-long forecasts, it was not at all clear by late in the decade that this was even theoretically possible.

After seeing the temperature soundings from Nimbus 3’s SIRS instrument in April 1969, Jule Charney had asked Robert Jastrow and Milton Halem at GISS to collaborate on a study to determine whether in situ wind measurements were really necessary for the proposed global observation system. GISS had been founded by Jastrow at Columbia University in New York in May 1961. GISS served as a center for theoretical modeling and data analysis studies, which Jastrow had believed NASA needed for its science program. The university location would foster better links with the scientific community, and much of GISS’s early work had been in the development of atmosphere models of Venus and Mars.

By this time, Charney had grown disenchanted with the constant-level balloon system. The balloons’ short lives at low altitudes made a balloon-based observing system expensive to maintain at all the different altitudes the numerical models needed. He had also had a thought that the models might not actually need wind measurements in any case. At a numerical simulation conference in Tokyo in 1968, Charney had postulated that since wind in the real atmosphere derived from temperature differences, one might be able to simulate this process in the model by continuously inserting temperatures while the model was running.¹⁶ This historical temperature data would, he thought, permit the model to generate wind fields in the lower atmosphere accurately without any need for an actual wind measurement. Winds from the upper atmosphere, necessary to provide a check on the calculations, could be obtained from either the constant-level drifting balloon system that both the United States and France were working on or from another of Vern Suomi’s ideas, wind vectors derived by tracking clouds using geosynchronous satellite cloud imagery.

This was the thesis that Charney wanted GISS to evaluate. GISS’s Halem obtained a copy of the Mintz-Arakawa two-level general circulation model from Yale Mintz at UCLA to run the experiments with. Using GISS’s IBM 360-95 computer, he, Jastrow, and Charney performed simulation experiments to investigate Charney’s idea. In a first set of experiments, they sought to determine an optimum period between temperature insertions. Insertion too frequently created spurious gravity waves in the model atmosphere, and they established twelve hours as the optimum period. Then they simulated the results that two potential observing systems might give. Temperature profiles generated by a barebones observing system consisting of two Nimbus satellites orbiting twelve hours apart, and nothing else, produced winds of useable accuracy, but only if their temperature errors were 0.25 degrees C or less. This was far better than what the SIRS sounder obtained. Simulations of a more robust observing system consisting of the two Nimbuses, upper troposphere and stratospheric winds from satellite-tracked balloons, and surface pressures from satellite-monitored buoys provided much more satisfactory results.¹⁷

In their resulting article, the three men merely concluded that their simplified model had only shown the possibility that historical temperature data insertion could result in accurate wind fields. Other researchers needed to do much more experimentation with more sophisticated models to check and refine this conclusion. Halem recalls that the paper was nonetheless greeted with a great deal of skepticism.¹⁸ A lot of researchers were surprised that one could insert temperature data at all during the model run without causing spurious oscillations. The numerical prediction models were initial state models. Operators fed them observational data at the beginning of a run and then left the model alone to calculate the desired length of time; the models were not designed to be updated. In fact, the tendency of global circulation models to destabilize when fed new data, or sometimes simply reject the real-world data and continue using their internally calculated results, turned out to be a very difficult challenge for researchers in numerical modeling. Eventually, Charney’s “direct insertion” methodology fell out of favor and was replaced by a more complex, but more effective, methodology called “four-dimensional assimilation.”

The article also served as a preliminary study of what a global observing system would have to consist of to produce the desired outcome of GARP, thirty-day global weather predictions. GISS undertook more studies during the second half of 1969 to further help define the GARP observing system, leading to considerable unease in the profession about the achievability of their goals. Halem and Jastrow found that with the error limits set by GARP planners of 3 meters per second for winds, 1 degree C for temperatures, and 3 millibars for surface pressure, they could achieve skillful predictions of only three to four days. Reducing the wind error alone to 1 meter per second could increase predictability to eight days, but to reach two weeks, the upper altitude wind observations had to have errors of less than 0.5 meter per second, the temperature soundings less than 0.5 degree C, and the surface pressure 0.5 millibar.¹⁹ These errors were far beyond the state of the technical art. As Halem put it thirty-three years later, “with GARP error limits, we wouldn’t be able to make monthly forecasts. And that disturbed people.”²⁰ Indeed, these numerical experiments showed that two-week forecasts were impossible within either the proposed GARP error limits or those imposed by the state of observation technologies.

Turning their attention to studies of what might be achievable within the limitations of near-term satellite technology, Jastrow, Halem, and their team at GISS found that the GARP Observing System probably would not need wind information from balloons or surface pressures from buoys to meet its requirements, except in the tropics. Indeed, through further simulations with the Mintz-Arakawa model, they determined that GARP error limits for winds and surface pressures were too generous and actually destructive of forecast accuracy. Based on the 1 to 2 degrees C error that the Nimbus 3 SIRS instrument was achieving, Jastrow and Halem reported, insertion of historical temperature profiles during the model runs produced wind and pressure fields that were more accurate than GARP error specifications for winds and pressures. Adding wind and pressure data at GARP error limitations produced worse forecasts than using the temperature data alone. They concluded that GARP wind and pressure error specifications should be tightened to 1.5 meters per second and 2 millibars, respectively, and that if observations confirmed these simulation results, the global observing system would not need ongoing measurement of wind velocities by the satellite-balloon system.²¹

The GISS team also carried out simulations directed at other aspects of the observing system. At Suomi’s request, they analyzed the potential utility of vertical temperature profile instruments like those on the polar orbiters on geosynchronous satellites.²² First, they examined the impact of geosynchronous sounding, without corresponding polar-orbiting satellites. Jastow and Halem reported that the geosynchronous sounders alone would result in inferior forecasts. This was due to their inability to provide soundings above 60 degrees latitude, which the satellites could not see from their equatorial orbits. The poles were crucial to wind determination, and without polar soundings the wind errors grew very rapidly. When added to the soundings provided by two polar orbiters, however, geosynchronous satellite soundings resulted in a substantial reduction in wind error. More important, the simulation studies showed that the geosynchronous sounding data could substitute for the loss of one polar-orbiting satellite’s sounder, preventing a reduction in forecast skill and providing a valuable backup capacity. Hence, temperature sounders on geosynchronous satellites would be a useful addition to the global observing system.

Finally, the GISS team studied the need for a side-scanning capability in the satellite sounder. The SIRS instrument on Nimbus 3 was fixed and only observed the atmosphere directly below the satellite (nadir viewing), which meant that it did not provide complete global coverage in each orbit. Instead, it sounded a relatively narrow swath of the Earth and only repeated each swath every few days. This, Jastrow and Halem reported, was insufficient for a two-polar orbiter observing system. It would not provide temperatures of a large enough portion of the atmosphere to accurately control the wind and pressure fields in the circulation model. The satellite instrument needed to scan at least 30 degrees to either side of the satellite’s track to ensure nearly complete coverage of the Earth on every orbit. They also found that four polar-orbiting satellites without side-scanning instruments would achieve the same results, but with obviously greater costs.²³ In planning for GARP, a side-scanning capability had already been assumed for the polar orbiters, however, and this result ratified GARP planners’ intentions.

Other simulation studies were also carried out for GARP. Akira Kasahara at NCAR and Joseph Smagorinsky, director of GFDL at Princeton, conducted simulation experiments aimed at determining the overall predictability of the atmosphere.²⁴ This was a controversial subject, as earlier experiments by MIT meteorologist Edward N. Lorenz had indicated that GARP’s goal of long-range weather forecasting was impossible. He had found that using the same initial data, his simplified model would output the same results for the first few days of a forecast and then start to diverge. Eventually, the forecasts from successive runs bore no relationship to each other at all. Early on, he recognized that when he had entered the initial state data into the model, he had rounded the numbers to fewer decimal places than the computer used, introducing a small error—much smaller than the measurement errors of real instruments. This small error then grew as the computer worked through its iterative calculations. The growth and propagation of error had been enough to produce the forecast divergence he had witnessed. Lorenz expanded this result in a 1963 paper to argue that his results suggested that the weather was so sensitive to initial conditions that the meteorological profession’s dreams of monthly forecasts would never be realized.²⁵ One could never measure the initial state of the atmosphere precisely enough to accomplish it.

Lorenz’s paper was well-known, but highly controversial, in the small community of numerical prediction researchers. While he had explained something that they had seen happen with their own models for more than a decade, his explanation was one that challenged the foundation of their beliefs. Numerical researchers like Charney, Smagorinsky, and Philip Thomas had been trained in the new physics-based meteorology, and the physics community prided itself on its ability to achieve prediction. Physicists were all trained to believe that once one understood the mathematics underlying a phenomenon, one could predict its future states accurately, forever. Yet this belief was based upon an assumption that errors would remain small throughout a calculation.

From a big-picture perspective, Lorenz’s argument was that this would not be true for nonlinear phenomena. Errors would inevitably grow as the calculations progressed, eventually overwhelming the original data. Hence, Lorenz’s argument was one that physicists, and even most meteorologists, did not accept easily. It was a denial of a central tenet of their science. The simulation experiments carried out between 1969 and 1971 at GISS, NCAR, and GFDL, however, served as further confirmation that Lorenz was correct. The inability of the GARP observing system to achieve prediction lengths beyond a week reflected the inherent sensitivity of the model to the data it was fed. Simulation studies of predictability continued for several more years, but the principal remaining question was whether these simulation studies were adequate reflections of reality. This could not be known until the GARP observing system was built and its data used to confirm this disturbing limit to predictability.

Finally, the observing system simulation experiments led to the downgrading of the constant-level balloon system’s priority. By the Joint Organizing Committee’s 1971 meeting in Toronto, it was already clear that a satellite-balloon system covering the entire Southern Hemisphere was unnecessary. The ability to use temperatures in place of winds at most altitudes had reduced the need for a wind-finding system to a single altitude that was referred to as a reference level, which would provide a check on the model calculations. But simulation studies had demonstrated that a balloon system would not produce the data meteorologists wanted. In the simulations, the balloons tended to cluster in certain areas, leaving other areas without coverage. This clustering tendency reduced the usefulness of the system to operational forecasting models, which needed the data to be relatively uniform in spatial distribution. EOLE therefore became an experiment to determine whether the simulations were accurate representations of the (still relatively unknown) Southern Hemisphere’s general circulation when it was finally carried out in 1973, and the Southern Hemisphere reference level need was filled by a drifting buoy system.

The remaining need for a balloon system seemed to be for determination of winds in the tropics. The substitution of temperatures for winds in the tropics did not work because the equations that inferred the winds from temperatures, in Vince Lally’s words, “blew up.” They required a non-zero Coriolis force acting on the air masses being measured, and at the equator the Coriolis force was zero. Hence, prediction models routinely had large errors in their tropical winds that propagated into the mid-latitudes, and an equatorial observing system in addition to the satellite sounders seemed necessary. In three sets of studies, NCAR, GISS, and GFDL researchers examined the question of what kind of system was really necessary.

There were two possibilities for a tropical wind system: a variation of the constant-level balloon system Lally named Mother GHOST (formally the Carrier Balloon System), and the use of winds derived from tracking clouds via imagery from the geosynchronous satellites. The Mother GHOST was a large GHOST balloon that remained at 200 millibars while releasing dropsondes on command relayed to it via geosynchronous satellite. The dropsondes would provide a vertical wind profile while allowing the Mother GHOST to stay at an altitude high enough for a relatively long life. The alternative, cloud-tracked winds, depended upon successful development of a way to produce them quickly and inexpensively. Suomi’s Space Science and Engineering Center (SSEC) was working on a system that replaced the film-loop-based method with a computer-based semiautomated one known as WINDCO. It would only provide winds at two altitudes, however, simply because the people doing the altitude assignment could only usefully distinguish between high- and low-altitude clouds.

The studies carried out by the three simulation study centers produced mixed results at first, with GISS finding that there was no improvement in wind errors through use of either system based on studies with the two-level Mintz-Arakawa model. Using Smagorinsky’s nine-level model, the GFDL staff had found that the data did improve forecasting. Further studies by all three organizations indicated that the difference resulted from the vertical resolutions of the models, with the more realistic nine-level models consistently showing improved forecasts from use of the tropical wind data.²⁶ The Mother GHOST system was the only one that resulted in tropical wind errors within the error limits that the Joint Organizing Committee had specified; the cloud-tracked winds improved the wind errors at all latitudes, but still resulted in wind errors in the tropics that were greater than desired. Hence, Mother GHOST was approved for the FGGE. For the earlier tropical experiment, a constant-level balloon system proposed jointly by NCAR and SSEC, the Tropical Wind, Energetics, and Reference Level Experiment (TWERLE) was selected to provide a check on the cloud-tracked winds, as Mother GHOST would not be available in time.

By 1973, then, the global observing system for GARP consisted of a permanent system composed of two polar-orbiting satellites provided by the United States and five geosynchronous satellites providing complete, and to some degree overlapping, coverage of the Earth between 60N and 60S latitudes. Two of these were to be provided by the United States, with one additional satellite each from Japan, the Soviet Union, and the European Space Agency. These satellites would form the space portion of the World Weather Watch system during and after GARP. A series of special observing systems would complement the permanent system during the global experiment’s intensive observation periods: a Southern Hemisphere drifting buoy system to define the surface reference level that in the Northern Hemisphere was provided by ground stations and weather ships, and Mother GHOST to provide tropical wind profiles. These studies had also shown that neither monthly nor two-week forecasts would result from the technologies planned for the 1970s. Instead, they would probably achieve forecasts of four to five days—the same forecast length already available from the conventional surface data.

At GISS, Jastrow had responded to all the interest in extending forecast length by hiring meteorologist Richard Somerville to lead an effort to improve the Mintz-Arakawa model. The Mintz-Arakawa model had an extremely efficient scheme for calculating the motion of air around the world and Somerville kept that dynamical core, but his group began reworking most of the rest of the model processes. They expanded the number of atmosphere levels from two to nine, to improve the vertical fidelity of the model, and wrote a new radiative transfer code. Jastrow was still interested in longer-range forecasts, and pushed them to make a model that could make “farmer’s forecasts,” as he called them. These would be seasonal forecasts, aimed at helping farmers choose the best crops for the next growing season. The new model was completed in 1974, and the group first used it to explore some old questions, including whether day-to-day solar variability affected the weather.²⁷

GATE AND THE DSTS

The first major GARP experiment was the tropical experiment that had been part of the original GARP proposal. Originally named TROMEX, for Tropical Meteorological Experiment, and intended for the equatorial Pacific, it evolved through several iterations into GATE. Carried out simultaneously was a set of DSTs intended to identify problems with the observing system’s hardware and software. Planners for both GATE and the DSTs understood that properly handling the data stream from the observing systems was the largest challenge facing them. The scientific purpose of GARP experiments was collection of high-quality data-sets for use in future research; the experiments would fail if the data was unusable. Similarly, if the data could not be processed in near real time, it would not be useful to the global operational forecasting system GARP was to demonstrate for the future World Weather Watch. The GATE and DST experiences proved enlightening.

One of the central questions in synoptic-scale meteorology during the 1960s was how energy moved from the tropics, which received the majority of the Earth’s overall solar input (insolation), into the mid-latitudes and then to the poles. At its highest level, the process was well understood. Solar radiation passed through the atmosphere and was absorbed by the oceans, heating the surface. Evaporation from the surface then carried the energy into the atmosphere. When this water vapor condensed into rain, this energy, technically called the latent heat of evaporation, was released, heating the surrounding air. Because the atmosphere is mostly transparent to the incoming sunlight, this is the primary mechanism for transfer of solar energy into the atmosphere. What meteorologists did not know were the details of how this happened—how much energy was received, on average, by each kilometer of ocean, how much water evaporated and recondensed, and most important for their global forecasting ambitions, how the small-scale convective systems that resulted from this process affected the synoptic-scale motions of the atmosphere they wanted to predict. This last question was the central scientific objective of GATE.

The first Television-Infrared Observations Satellite (TIROS) satellite had presented meteorologists with some intriguing data from the remote tropical oceans. Its cloud photographs had revealed the existence of very short-lived, small (i.e., 30 to 100 kilometers), but intense convective systems. They had been labeled cloud clusters for their appearance in these images. The clusters formed and dissipated within twelve hours, usually appearing and disappearing between two orbits of the satellite. These cloud clusters were an obvious mechanism for energy transfer, and they had caught the interest of Jule Charney and Vern Suomi, among many others. Because study of them offered the potential to provide important insights into the dynamics of the energy transfer process, they became GATE’s central phenomenon.

In late October 1968, Suomi hosted a study group in Madison to speculate about the potential linkage of these cloud clusters to larger-scale phenomena, prepare a preliminary tropical cloud climatology from the prior year’s satellite imagery, and formulate recommendations on how to construct a ship-based observation network that would permit study of the full life-cycle of the clusters. They produced a recommendation that the experiment be conducted around the Marshall Islands using an overlapping set of ship networks, with a meso-scale array to study the clusters themselves and a second array covering ten times the first’s area to permit linking the meso-scale data with synoptic-scale phenomenon.²⁸ GATE would also require the satellite-based global observing system in order to complete the series of linkages from the small scale to the global. In particular, the experiment depended upon the geosynchronous imaging satellites, whose ability to produce images of the same area in relatively rapid succession would be crucial, and it needed a tropical balloon system to better define winds. These recommendations became the basis of the formal National Academy of Science’s proposal to the U.S. government regarding GATE.²⁹

The study group at Madison had chosen the Marshall Islands because the cloud clusters were quite common, occurring every four to five days, and because the region was logistically feasible. The chain had a large number of small, uninhabited islands that could be used as observing stations, thus reducing the number of ships required, and there were airfields available at Kwajalein and Eniwetok islands for logistical support and to host aircraft-based experimenters. However, these were military installations and that led to the shift of the tropical experiment to the Atlantic. The tropical experiment, while principally planned by the United States, was an international experiment. There would be a substantial number of Soviet ships involved—as it turned out, the Soviet Union provided more ships to GATE than any other nation—and hosting them at the American naval station at Kwajalein for the experiment period of nearly a year was infeasible. The experiment area was therefore replanned into the equatorial Atlantic, with an enlarged ship array as a substitute for fewer island-based observing stations.

The Nixon administration’s March 1970 approval of the National Academy of Science’s plan for American participation in GARP permitted detailed international negotiations over funding as well as ship, aircraft, and satellite availability to go ahead.³⁰ By late 1971, the experiment had taken on its final form. There were to be three special observing periods during the summer of 1974, each three weeks long. A network of thirty-eight ships, organized into three nested arrays straddling the equator, would stretch from the east coast of South America to the west coast of Africa.³¹ The ship array was be supplemented by thirteen aircraft, whose missions would be planned daily based upon the next day’s forecast and adjusted on the fly through use of imagery from a geosynchronous imaging satellite. Initially, the experiment was to be run from the British Meteorological Office’s facilities in Bracknell, England, but after a diplomatic row caused by the expulsion of Soviet diplomats on charges of espionage, Bracknell became politically unacceptable.³² Senegal offered use of a new facility at Dakar’s airport for the experiment’s headquarters, and this became home to the experiment’s Scientific Management Group. The Dakar site had the advantage of putting the management group in the same location as the aircraft, facilitating mission planning, at the cost of requiring construction of a ground station for receipt of the satellite imagery.

NASA’s primary contribution to GATE was a set of satellites for the global observing system and experiments related to those satellites. NASA Goddard had scheduled the launch of the prototype of NOAA’s Geosynchronous Operational Environmental Satellite (GOES), SMS 1, for mid-1973, and this would be parked over the experiment area to provide the overhead imagery the planners considered crucial. It would also provide the experiment data link support and NOAA’s Weather FAX service to permit rapid distribution of forecast maps. Goddard was also responsible for providing Nimbus 6 to the experiment. In many respects, this was a functional prototype for NOAA’s next generation of polar-orbiting weather satellites. It carried the prototypes of Bill Smith’s High Resolution Infrared Spectrometer (HIRS), David Staelin’s Scanning Microwave Spectrometer (SCAMS), and John T. Houghton’s Pressure Modulated Spectrometer. These three instruments were to be functionally unified into the TIROS Operational Vertical Sounder on TIROS N, the actual hardware prototype of the new operational satellite due for launch in the late 1970s. Nimbus 6 also carried the tracking and communications equipment for the TWERLE experiment, which was an evolution of the Interrogation, Recording, and Location System (IRLS) system demonstrated on earlier satellites. NASA also provided other support to GATE, including its tracking ship USNS Vanguard, Ames Research Center’s Convair 990, and data processing facilities at both Goddard and GISS. But the satellites were the centerpiece of NASA’s GATE effort.

GATE ship array (center), with GATE land-based observation locations marked on continental areas. From Joachim P. Kuettner, “General Description and Central Program of GATE,” Bulletin of the American Meteorological Society, 55:7 (July 1974), p. 713.

The satellites also turned out to be its primary challenge. SMS 1 suffered development problems related to Goddard’s choice of a contractor that lacked the necessary expertise and resources to complete it on schedule, and its launch date slid from mid-1973 to June 1974. It reached orbit barely in time for the experiment, whose ships left port 15 June for the first observing period. Its delay caused a great deal of concern on the U.S. Committee for GARP, because it was the only geosynchronous satellite to have infrared imaging capability and the Scientific Management Group intended to use the nighttime cloud imagery in its aircraft mission planning. The agency was less fortunate with Nimbus F, which did not launch until well after GATE. Its delay was a product of instrument development problems. The contractor for the HIRS instrument had not been able to deliver it in time. Hence, the satellite observing system that ultimately supported GATE consisted of the still partly operational Nimbus 5, SMS 1, the still-functioning ATS 3, and two operational NOAA polar-orbiting satellites, NOAA 2 and NOAA 3. These carried the earlier version of Smith’s temperature profile instrument, the Vertical Temperature Profile Radiometer (VTPR), and the Advanced Very High Resolution Radiometer (AVHRR).

In addition to these changes in the satellite observing system, there were other changes in NASA’s support for GATE during the three years between acceptance of the 1971 plan and its conduct in 1974. NCAR’s Paul Julian and Robert Steinberg from the Lewis Research Center in Cleveland had conceived a way to acquire wind data from commercial airliners equipped with inertial navigation systems.³³ These sensed wind motions automatically, and Julian and Steinberg’s idea was to equip the navigation systems with recorders and pick up the resulting tapes when the aircraft landed. These would only provide wind data for a very narrow set of altitudes, and only along airline flight routes. But several routes overflew the GATE array, and NASA and NOAA arranged contracts with several airlines flying those routes to install recorders and collect the data tapes. They also equipped a U.S. Air Force C-141 that had a daily route over the array for the experiment.³⁴

GATE’s three field phases took place between June and September 1974. By this time, GATE had evolved to include five hundred different experiments organized into a Central Program and five subprograms: a synoptic-scale program, a convective-scale subprogram, a boundary-layer subprogram, a radiation subprogram, and an oceanographic subprogram.³⁵ The Central Program’s objectives were to examine the interaction between smaller-scale tropical weather phenomena and the general circulation and to improve numerical modeling and prediction methods. Each of the subprograms supported the Central Program in some way. The synoptic-scale subprogram supported it through description of synoptic-scale disturbances within the experiment region and by providing the datasets for numerical models. The convection subprogram included the cloud cluster investigation that had been important to GATE’s foundation as well as a budget experiment vital to understanding scale interaction. The boundary-layer subprogram included surface flux measurements needed for the convective studies and for efforts to parameterize convective processes. The radiation program focused on radiative heating and cooling rates and processes, also necessary quantities for parameterization efforts, while the oceanographic subprogram was aimed at ocean-atmosphere forcings.³⁶

In his early comments after GATE field phase, experiment director Joachim Kuettner, a veteran of the NASA Mercury program and a meteorologist who had specialized in mountain-induced waves in the atmosphere, wrote that “it [was] common experience that no field project achieves 100% of its goal.” Atmospheric scientists had to work in the “laboratory” of the Earth’s atmosphere, and between its vagaries and those of the machinery of the observing systems, they were generally lucky to get a majority of the observations they sought. In GATE’s case, the availability of real-time and near-real-time data relayed via satellite had allowed the Scientific Management Group to identify parts of the observation system that were not performing as expected and either fix them or compensate for them—repositioning ships with unreliable wind-finding equipment from more to less important areas, replacing ships with mechanical difficulties, or reassigning aircraft missions. He estimated that GATE accomplished about 80 percent of the observations intended for it, with the most disappointing results coming from the conventional surface stations of the World Weather Watch.³⁷

Originally, GATE and the DSTs had been scheduled to coincide. GATE’s requirement for data from conventional, special, and space-borne observing systems had made it an obvious opportunity for verification of the data transmission and processing system that was to be the basis of a future American global forecasting system. The DST project office at NASA Goddard had generated a set of four tests to be carried out on individual pieces of the observing system, such as processing of geosynchronous imagery to extract wind velocities from cloud motions, between 1972 and the beginning of GATE. During GATE, a fifth test encompassing all of the observing systems would be carried out, followed by a sixth, and final, test during the 1974–75 winter—one could not be certain that the observing systems functioned well in all conditions without testing them in the best and worst seasons, after all. The delayed launches of SMS 1 and Nimbus F, both intended to be functional prototypes of the operational GOES and TIROS N series satellites, however, forced the postponement of the DST series. Hence the two full-up tests were carried out for sixty-day periods in August–October 1975 and January–March 1976.³⁸

The results of the final two DSTs were somewhat disturbing. Portions of the tests went very well. The TWERLE experiment, for example, carried out launches of 393 instrumented, constant-level balloons to 150 millibars altitude in the tropical Southern Hemisphere between October 1975 and January 1976. The RAMS tracking system performed essentially as expected, giving a location accurate to within about 3 kilometers, and demonstrating a form of clustering that had not been seen with the mid-latitude EOLE experiment. A substantial number of balloons clustered in the Gulf of Guinea region, which project scientist Paul Julian interpreted as verifying analyses from conventional data that suggested large-scale, long-duration divergence. TWERLE’s RAMS system also successfully tracked and received data from drifting buoys, an important part of the future global observation system.³⁹

But the DSTs also demonstrated that any future global observation system needed to have much greater quality control in its data processing path. The quality control challenge derived from the use of cloud-tracking to derive wind velocities and from the well-known cloud clearance difficulties with the infrared temperature profile instruments. During the early 1970s, Suomi had developed a method of partially automating the cloud-tracking process that allowed derivation of wind velocities from geosynchronous satellite imagery. Initially called WINDCO, and later McIdas, the system utilized a midi-computer as a workstation, a remote mainframe for processing, and a television-quality monitor that was linked via a lookup table to the much-higher-resolution image data. This permitted operators to work with the high-resolution data without actually having to display it—video displays with sufficient resolution to display the 120-megabyte satellite images did not exist.

One of the principal difficulties of tracking clouds in successive images had been that the satellite’s own motion was imperfect, and it had to be subtracted from the overall cloud motion. WINDCO did this using two corrections, one based on the satellite’s known motion from the satellite tracking system, and one based on registration of the images to an obvious landmark on Earth. The WINDCO operator chose a landmark visible in successive images with a light pen, and the computer made the necessary corrections to its model of the satellite’s motion. The operator could then select clouds for tracking and produce the wind set. Suomi liked to call this method of combining human intelligence with automation man-interactive computing, and this was the genesis of the system’s final name, the Man-Computer Interactive Data Access System (McIdas). McIdas, demonstrated for NASA and NOAA officials in April 1972, became the means of production of the cloud-tracked wind sets for GATE and for the DSTs in 1974–75.⁴⁰ SSEC was assigned responsibility for receiving and archiving the satellite imagery during these experiments and producing four sets of wind data each day.

But when GISS and the National Meteorological Center tried to use the wind sets in their experimental global models during 1976 and 1977, they found the data had damaging errors. The principal flaw was in the operator’s assignment of altitude to the clouds being tracked, and thus to the resulting wind vector. The datasets were supposed to contain wind vectors at two levels, and the operators had not been able to reliably distinguish between upper- and lower-level clouds. Although the number of errors was actually small in relation to the size of the dataset, the erroneous winds had large impacts on the resulting forecasts.

Similarly, the temperature profiles derived by the National Environmental Satellite Service from Smith’s infrared temperature sounder on Nimbus 6 contained substantial errors. Some of the errors derived from the cloud clearance problem, while others occurred under specific meteorological conditions. The automated inversion method that Smith and Hal Woolf had developed proved not to give accurate results under all conditions. The erroneous temperatures caused forecast errors just as the wind errors had. Writing for the record in November 1977 about the temperature sounding challenge, one member of the U.S. Committee for GARP commented that the committee had concluded that the data processing methods used during the FGGE should not be the same as those used during the DSTs. A special effort to check the quality of the sounding retrievals during the experiment was going to be necessary.⁴¹

Most disturbing of all to GARP participants, however, was the National Meteorological Center’s assessment that the satellite temperature profiles did not significantly improve forecast skill in the Northern Hemisphere.⁴² This determination was independent of the problem of erroneous soundings; even when they were weeded out and discarded, the National Meteorological Center’s forecast model produced essentially the same results as it did when given only the conventional radiosonde data. In the Southern Hemisphere, where almost no conventional data existed, the satellite data improved forecasts significantly, bringing Southern Hemisphere forecasts to nearly the same level of skill as those in the Northern Hemisphere. This ratified the results of the Observing Systems Simulation Experiments, which had suggested that this generation of satellite sounders would only produce skillful four- to five-day forecasts. But GARP planners, most of whom represented Northern Hemisphere nations, had expected that satellite data would extend Northern Hemisphere forecasts beyond the four to five days possible with conventional data. Instead, the DSTs had suggested that the satellite soundings produced no benefit to their nations at all.

The results of the DSTs were troubling, but their purpose, after all, had been to evaluate the functioning of the data processing system prior to GARP’s primary goal, the FGGE (also known internationally as the Global Weather Experiment). In that sense, the DSTs had been very successful. While GARP planners could not address the disappointing predictability outcome prior to the FGGE—this, they understood, required a new generation of satellite sensors, new prediction models, or both—they could fix the quality control problem. In January 1978, the First GARP Global Experiment Advisory Committee met in Madison to work out how NASA, NOAA, and SSEC would deal with it. At this meeting, NOAA’s Bill Smith argued that the satellite sensors already provided most of the data necessary to produce better outcomes.⁴³ Assignment of altitudes to the clouds tracked to produce winds could be done more accurately if the McIdas operators had access to the cloud-top temperature data provided by the window channel since cloud temperature was directly related to altitude. Similarly, McIdas operators could cross-check temperature soundings with the National Meteorological Center’s analysis charts, and with cloud imagery to evaluate them if these were made available to the operator. In this way, trained meteorologists’ subjective judgment could be used to check the performance of the automated cloud clearance process.

While having to use humans to inspect the results of the automated sounding results slowed the process somewhat, only a small fraction of the twenty thousand soundings generated by TIROS N’s HIRS instrument each day would require human intervention. A set of automated filters that compared soundings to nearby radiosondes already threw out obviously bad measurements. Another set of filters that compared the soundings from the infrared instrument to those from the microwave instrument was under development at the National Meteorological Center. Because the two instruments had significantly different resolution, instead of automatically discarding soundings that differed, these filters would flag them for inspection. Under certain meteorological conditions, one would expect the large-area microwave sounding to differ from the higher-resolution infrared result without either of them being incorrect. Meteorologists could identify these and validate or reject the sounding. This was another area McIdas’s ability to display multiple data sources graphically would help speed the process.

As a result of this meeting, the committee recommended that SSEC develop and implement McIdas software to enable its use for these quality control measures. In a “special effort,” the FGGE datasets would be checked by meteorologists using McIdas; later in the year, SSEC proposed for and won this task. It also built a McIdas system for NOAA’s National Environmental Satellite Service for operational use. As a result of these recommendations, the Goddard GARP project office restructured the data flow path for the FGGE, linking SSEC’s McIdas terminal to a mainframe computer at GISS for the data processing and to the National Meteorological Center so that forecast analyses could be imported directly into McIdas.⁴⁴

Finally, recognition that full automation of the retrieval process would not produce data of sufficient quality caused the National Environmental Satellite Service’s director, David Johnson, to adopt Suomi’s man-interactive data processing concept for the operational system post-FGGE. McIdas overcame one of the central problems of the early space age: the ability to produce overwhelming amounts of data without a corresponding capacity to analyze it all. McIdas’s graphical display of large datasets maximized its human operators’ capabilities while preserving the digital nature of the data. While it did not reduce the data torrent, McIdas put the data in a form more meaningful to humans, allowing its operators to apply knowledge and judgment in evaluating the data.

THE FIRST (AND LAST) GARP GLOBAL EXPERIMENT

The FGGE, renamed the Global Weather Experiment (GWE) as it became clear to the GARP Joint Organizing Committee that it would not be followed by further global experiments, was finally carried out in two special observing periods, January–March 1979 and May–July 1979. Initially planned to encompass an entire year of detailed observation, the GWE was reduced to two sixty-day intensive observation periods by a combination of lack of funds and lack of interest. The purpose of the experiment was the production of datasets for use in improving numerical forecast models; originally, GARP planners’ beliefs in the possibility of month-long forecast had generated the need for larger, longer-term datasets. By 1979, GARP’s founders no longer believed that thirty-day forecasts were possible, and they could not justify the cost of a full year’s detailed, quality-controlled data-sets. Their apparent inability to provide a great leap in forecast length or skill had reduced politicians’ interest in the program, with a consequent reduction in financial support. The two sixty-day periods would provide enough meteorological diversity for model improvement.

The GWE also served as an operational test of the prototype of the new American operational polar-orbiting satellite, TIROS N, and this was the source of GWE’s delay from its originally planned year of 1976 to 1979. NASA’s TIROS N effort had started in 1971 but had soon run into troubles. In this case, the troubles were not primarily technological. In 1972, the Office of Management and Budget (OMB) had embarked on one of its occasional “streamline the government” initiatives and put TIROS N on hold while it investigated whether the nation should continue to maintain two separate polar-orbiting weather satellite programs, the Defense Department’s Defense Meteorological Satellite Program (DMSP) series and the NOAA series.

OMB’s preference was to eliminate the civilian program. This had sparked a meeting of the U.S. Committee for GARP to discuss the issue, and the committee decided to advocate in favor of the civilian program.⁴⁵ Vern Suomi and Richard Reed recruited National Academy of Sciences president Philip Handler into the effort, and Handler raised the subject with the director of the White House Office of Science and Technology Policy, Russell Drew, resulting in a meeting on 15 October 1973. Suomi followed that meeting up with a letter delineating the scientific requirements imposed on the satellite sensors by the GWE experiments: the sounder’s accuracy needed to approach 1 degree C, which required use of both infrared and microwave sounders. Because France was providing the balloon and drifting buoy tracking and data system for the GWE, the United States also had a commitment to provide space and an appropriate interface on the satellite for it. These were the vital requirements for the GWE, Suomi had argued, and whatever system OMB chose needed to support them.⁴⁶

To a degree, OMB relented after the scientists’ intercession. The two polar orbiter programs remained separate, but NASA was required to use the Defense Department’s satellite and modify it to take the instruments it and NOAA had developed over the previous decade. The name TIROS N remained attached to the project, however, and the delay pushed its launch back to October 1978. The first NOAA-funded operational version of the satellite, NOAA 6, followed it into orbit in June 1979.

One more satellite was supposed to join the GWE constellation, Seasat. Developed at the Jet Propulsion Laboratory (JPL), Seasat carried a radar altimeter to precisely measure the height of the ocean surface. Sea surface height varied with wind, current, and temperature, and this measurement was of interest to physical oceanographers. Seasat also carried a scatterometer, permitting it to indirectly measure wind velocity at the ocean surface. If this experiment worked out, the sea surface wind measurement would provide a replacement for the lower-altitude variants of the constant-level balloon system that had been too short-lived to be of use.⁴⁷ This would be particularly useful for tropical forecasting, where winds could not be inferred from temperature histories by the prediction models accurately.

Seasat was launched on time, but failed in orbit after 106 days. It returned enough data to demonstrate that the scatterometer and surface height functions produced good results, but did not function long enough to be used during the GWE. It took NASA many years to repeat the experiment because it could not reach agreement with the U.S. Navy to help fund it. Instead, NASA eventually arranged a joint effort with France known as TOPEX/Poseidon to replace the surface height measurement, and with Japan to replace the scatterometer. Neither of these flew before 1990, however. Seasat’s loss reduced the completeness of GARP datasets, meaning, for example, that surface wind measurements in the Monsoon Experiment’s area would be less than desired. Further, of course, its data would not be available for operational forecasting, which NASA had hoped to achieve.

During the GWE, the satellite network was supplemented by all the special observing systems developed during the preceding decade: dropsondes, the Southern Hemisphere drifting buoy system, and an equatorial constant-level balloon system much like TWERLE. There had been a great deal of concern about these during the preceding few years, because the experiment’s funding nations did not provide enough money to carry them all out. Vincent Lally’s Carrier Balloon System had been intended as the source of vertical wind profiles for comparison to model-produced wind fields, but its high cost, combined with doubts about the utility of its data—the constant-level balloons’ tendency to cluster meant large data gaps—had resulted in its cancellation in favor of aircraft-based dropsondes. France cancelled its funding of the ARGOS tracking system, and the system was salvaged by a donation from the Shah of Iran. The Soviet Union was unable to meet its commitment of a geosynchronous meteorological satellite, and NASA, NOAA, and the National Science Foundation (NSF) had to scrounge for the funds necessary to revive the now-mothballed ATS 3 for the duration of the experiment, and constructed a ground station in Europe to serve it.

Despite all of these departures from the original plans, the observing systems performed as expected during the two intensive observing periods, as did the revised data processing procedures. NOAA P-3 and C-130, and USAF WC-135 and C-141 aircraft flew dropsonde missions from Hickam Air Force Base and Acapulco International Airport, Howard Air Force Base, Ascension Island, and Diego Garcia to provide tropical wind measurements. Boeing 747 airliners equipped with the Lewis Research Center’s automatic data reporting system submitted 240,000 observations via the geosynchronous satellites. NOAA provided and operated sixty-four drifting buoys in the Southern Hemisphere for the reference-level experiment, supplementing a larger international flotilla.

Hence, all of the preparation that had gone into getting ready for FGGE paid off in an experiment that was essentially anticlimactic. The datasets were prepared and archived between 1979 and 1981 and became the basis of future research on prediction models. Because the special observing periods were carried out in conjunction with another large-scale field experiment in the Indian Ocean, the Monsoon Experiment, the data was also in high demand for research into the lifecycle of these annual events. This was precisely what GARP’s founders had hoped for, and the smoothness of the FGGE and the quality of its data after all the delays and disappointments they had experienced reflected both their hard work and their ability to adapt when things did not quite work out the way they had hoped.

WHAT HATH GARP (AND NASA) WROUGHT?

GARP had been founded to advance two sets of technologies, numerical models and satellite instruments, and to use them to enhance human knowledge of atmospheric processes and to increase forecast lengths. It’s fair, then, to ask what it actually achieved. Writing in 1991, the European Center For Medium Range Forecasting’s Lennart Bengtsson credited the post-GWE global observing system, the development of data assimilation schemes for non-synoptic data, and improved numerical models with having achieved five-day forecasts in the mid-latitudes with the same level of skill as one-day Northern Hemisphere forecasts had had in the early 1950s.⁴⁸ While this was far from the two-week forecasts Charney and many others had hoped for in the mid-1960s—and the monthly forecasts predicted earlier in the decade—it was consonant with the results of the Observing System Simulation Experiments and with the DST results. The convergence of these different experiments on the same number produced, over time, a great deal of confidence within the meteorological profession that the models represented the atmosphere’s large-scale processes relatively accurately.

Bengtsson also addressed a highly controversial subject: whether better models or the addition of satellite data had resulted in these gains. Several examinations of this subject had resulted in the conclusion that improvements in the models had produced most of the gains.⁴⁹ When run using only the FGGE satellite data, the prediction models’ forecasts degraded about a day faster than when run using only the conventional data. The model researchers believed that because the satellite measurements were volumetric averages and tended to wash out the fine vertical structure of the atmosphere, they resulted in analyses that consistently underestimated the energy available in the atmosphere, leading to earlier forecast degradation. Satellite advocates believed that the assimilation schemes for the data were at fault, because they treated the satellite data as if it were radiosonde data delivered at non-synoptic times. Instead, the satellite data was a fundamentally different kind of measurement, and assimilation schemes designed for it would show better results.⁵⁰

In giving the Royal Meteorological Society’s Symons Memorial Lecture in 1990, NOAA’s Bill Smith contended that both of these arguments had validity. But pointing to a surprising recent study that showed rapid improvement in forecast skill between the end of the GWE and 1986, followed by a plateau of skill in subsequent years, Smith argued that improvements in model physics and assimilation schemes following the GWE had achieved all that was possible with the current, late 1970s generation of satellite instruments. He conceded that their poor vertical resolution was responsible for the lack of significant positive impact the satellite data had on Northern Hemisphere forecasting. Future improvements in forecast length and skill required a new generation of satellite instruments with much greater vertical resolution.⁵¹

Smith believed that future instruments could produce radiosonde-like data, and in 1979 had proposed an interferometer-based instrument known as HRIS—the High-Resolution Interferometer Spectrometer.⁵² A Michelson interferometer like Hanel’s InfraRed Inferometer Spectrometer (IRIS), but higher in spectral resolution, this technique would not produce the volumetric averages that Smith’s spectrometer-type earlier instruments did. Instead, it would produce a continuous atmospheric profile, exactly like a radiosonde. His proposal was funded by NASA and NOAA, and began to fly on NASA’s ER 2 in 1986. It took a second form as a satellite instrument called CrIS, the Crosstrack Infrared Sounder, which was designed to supplant the infrared sounding unit on the TIROS N series in the late 1980s.

But no new instruments flew before the turn of the century, reflecting GARP’s principal shortcoming—it did not live long enough to carry out its full program. The FGGE had not been intended to be the last GARP global experiment; instead, it was to have provided the data necessary to design better models (which did happen) and, aided by those new models, to build and fly improved satellite sensors (which did not). The reason GARP did not complete its program, Smith had not bothered to tell his audience of practitioners, was that NASA had restructured its satellite instrument development programs in ways that unintentionally led to a two-decade-long hiatus in new instruments for the polar orbiters.

The first generation of NASA’s leadership retired during the late 1970s, and the new leadership did not think well of the approach the old Meteorological Programs Office had taken toward instrument development. Shelby Tilford, who became head of NASA’s Upper Atmosphere Research Program (UARP) in 1977, and later head of its overall Earth sciences program, recalls that the conflict had been over the relationship between instrument developers and model developers. Within NOAA, the National Environmental Satellite Data and Information Service (NESDIS) designed instruments and operated the satellites, while the National Weather Service developed models. The conflict had come to a point when NOAA/NESDIS had sent over to NASA requirements for a next-generation sensor and the model developers at the National Weather Service had refused to verify them. Indeed, they took a position of rejecting the value of satellite data entirely. Because the satellite data did not produce better forecasts than the radiosondes, the Weather Service only employed the satellite data from the Southern Hemisphere and used the radiosonde data in the Northern Hemisphere. Tilford saw little sense in continuing to spend money on a program to develop sensors whose data would not be used. So NASA and NOAA leaders agreed to end the Operational Satellite Improvement Program in 1982.⁵³

They did not, however, intend meteorological satellite instrumentation development to end. Instead, the two agencies agreed to incorporate new instrument development into NASA’s atmospheric sciences program, with the agency supporting both the instruments and the models needed to use them. This would prove the value of satellite remote sensing to science as GARP studies had not, and perhaps eventually permit overcoming the resistance to satellite data that had grown in the National Weather Service. The new instruments could be transitioned to NOAA after NASA had demonstrated their capabilities. Yet the election of Ronald Reagan in 1980 undermined this plan. Reagan’s budget officials believed that the government should not provide operational functions, such as the meteorological satellite systems, and sought to privatize them. While this effort failed, in the process, the administration cut both NASA and NOAA budgets substantially. This left NOAA without the ability to finance even incremental improvements to the weather satellite series, and the instrument generation of 1978, with only minor updates, continued to fly through the end of the century.⁵⁴

In pursuit of GARP, and driven by their own ambitions to remake meteorology into a global science, NASA, NOAA, NCAR, and SSEC produced a technological legacy of sophisticated global models and instruments to feed them that had been dreams when they had started in 1960, and in a few cases that had not been thought of at the time. They made mainstream the use of simulation studies, graphical display of data, and remote sensing. Yet they did not achieve what they had set out to do: provide a revolutionary increase in forecast length.

Instead, what GARP and its myriad supporting studies accomplished was a powerful demonstration of the unpredictable nature of the atmosphere. In the process, it also provided a large, expensive case study of the limits of the belief system postwar scientists had gained from physics. The ability to predict phenomena, the ultimate test of a physical theorem’s correctness, was not fully applicable to meteorology. Instead, regardless of the quality or completeness of meteorologists’ understanding of atmospheric processes, one could not predict the weather into an indefinite future. The weather, and the atmospheric processes that produced it, had a strong element of chaos at its root.

This issue of limits to predictability, and the directly related idea that tiny, even immeasurable, changes can have global effects, was profoundly disturbing. It was quickly adopted in the public sphere, achieving a cultural resonance as the “butterfly effect,” the name given it by Ed Lorenz in a 1972 lecture—a butterfly flapping its wings on one side of the world could change the weather on the other.⁵⁵ But within the scientific community, this exploding of the belief that once one fully understood a phenomenon one could make accurate long-range predictions left in its wake a community that had to rebuild itself around the notion of uncertainty. One could make probabilistic forecasts with sufficient understanding of a complex phenomenon, but the deterministic, long-range predictions the postwar meteorology community had thought was in its reach was impossible.

The Joint Organizing Committee of GARP voted itself out of existence in 1980. Or, rather, voted to transform itself into the Joint Scientific Committee of the World Climate Research Program. GARP’s goal had been twofold, to improve weather forecasting and to investigate the physical processes of climate; having done what seemed possible with the weather, its leaders turned to Earth’s climate. Interest in climate research had grown throughout the 1970s, in part due to NASA’s planetary studies and in part due to increasing evidence that humans had attained the ability to change climate by altering the chemistry of Earth’s atmosphere. NASA would never play as large a role in the World Climate Research Program (WCRP) as it had in GARP. But its role in climate science would far surpass its role in weather forecasting. One of Robert Jastrow’s young hires at GISS, radiative transfer specialist James E. Hansen, would become one of the foremost climate specialists in the world during the 1980s, and one of the most controversial.