A brief history of the workshop

Previous workshops


Some history of the workshops

The first adjoint workshop was organized by Ronald Errico from NCAR, Philippe Courtier from ECMWF, and a few of their colleagues. One of its motivations was to encourage the application of adjoint models, not just in 4DVAR but also elsewhere. The fact that adjoints efficiently produce quantitative estimates of sensitivity for arbitrary perturbations for a wide range of metrics indicates that they can be applied to many central questions of interest pertaining to model dynamics and analysis. That these are only linearized estimates presents important limitations but their utility has nonetheless been amply demonstrated.

The first workshop was formally called "The workshop on adjoint applications in dynamic meteorology." The delimiter "dynamic meteorology" was purposefully included to limit the number of attendees attracted because at that time there were also several oceanographic groups attempting to employ adjoints, particularly in forms of what we now call 4DVAR. In practice however, the workshop has welcomed any interested attendees and presenters. In fact, cross pollination with other disciplines has been encouraged. Many oceanographic talks are now especially enjoyed. Several talks concerned with nuclear or aerospace engineering or with oil exploration have been presented.

The workshop's format was partially motivated by the 1982 Stanstead Seminar on data analysis and initialization for NWP, held in Lennoxville Quebec (the same location as the 3rd Adjoint Workshop) and by a meeting on mesoscale modeling held in 1989 in Norway by the HIRLAM organization. Both of these meetings were located in relatively remote places, such that almost all participants boarded, dined, and evening-socialized at the same facility. This acted to maximize the contact time between all participants. The HIRLAM meeting included an afternoon mid-week break allowing some refreshing sightseeing and physical activity. This later became a hallmark of all the adjoint workshops.

The workshop format has always included four or five types of presentations. For the early workshops, each session began with a longer assigned talk introducing the session's topic. These were intended to help novices understand the topic better and to obviate the need for presenters of subsequent shorter talks to expend some of their limited time on common fundamentals. One or more poster sessions replaced oral sessions so that their allotted times were ample. Typically, the choice of oral or poster presentation was made by the presenter, so that unlike at some meetings, posters were never treated as second class. On the evenings of the mid-week afternoon breaks, invited long talks were presented. These were intended partly to lure attendees back together after a tiring afternoon apart and partly to encourage a non-technical, free-ranging, perhaps speculative, and hopefully entertaining or exciting presentation. At the early workshops, a pair of introductory session talks covered the topics of "What is an adjoint model" and "What is 4DVAR" but at later workshops these were instead moved to a pre-workshop day of tutorials intended for students or other new attendees. This allowed these most fundamental lectures to be more comprehensive and less rushed, with more time for questioning, and for more available time slots on other days.

Before the workshop begins, presenters are specifically asked not to present a summary of all the relevant work that they have recently conducted, effectively as an advertisement. Instead, they are instructed to focus on their most interesting results that can be adequately covered in their short allotted time. They are warned that enough detail should be included so at least several listeners can evaluate the work's quality and discern its broader implications. The audience is told that the reason for our gathering is not actually to simply listen to the presentations but to hear the questions or comments by others who can help the audience place the work in its proper perspective. Otherwise we can all remain comfortably and inexpensively at home, reading extended abstracts or watching YouTube videos. Time for questions, responses, and follow-up questions therefore has been sacrosanct at all the workshops, since they are the reason for our gathering. For this reason, speakers are warned to strictly adhere to their allocated times. Remarkably, almost all do! Also, the session chairpersons are instructed to refrain from the too usual "We only have time for 2 (or no) questions."

Unlike many small meetings this workshop series endeavors to allow all attendees to present. This of course a poses a potentially great problem if the number of speakers is too large, but usually the number has been between 55 and 75. Once there were over 100 attendees (luckily not all presenting) and some concurrent sessions were then required. Consequently, some attendees were very unhappy that they had to choose between which to attend. Shortening the times allocated for all the types of talks remains unacceptable because presenters must have reasonable time to present the required details for discerning a work's quality and the audience must have ample time for questions and comments. 

Another constraint self-imposed on most of the Adjoint Workshops stemmed from a desire to keep the participant costs very low. There are many competing meetings, and attendance at some are almost considered compulsory, such as those where projects can be advertised to funding managers. As long as the Adjoint Workshop cost could be kept low, attendees could attend both it and the others. Philippe Courier managed to secure a site in France, within a short bicycle ride to the beach, for $40 per night per person, inclusive (even wine at dinner)! In Italy. Carla Cardinali secured a site built into a cliff on the Mediterranean Sea for $100 per person per night inclusive (with a 4 course dinner, and wine at both lunch and dinner)! These presented very high bars for the later organizers and eventually became impossible to attempt. That did not prevent them from trying. Finding a suitable venue that was remote but accessible, inexpensive but nice, intimate but accommodating for at least 70, attractive for some break activities but without distractions, etc., has been the most time-consuming organizational task.

There was once a complaint about the format not actually being a "workshop" in that there were no working groups preparing reports. There have been some published post-workshop reports, notably in BAMS when it encouraged such submissions. So, "Adjoint Worksop" is perhaps a misnomer, but that is what it has been called for more than 30 years!

One anecdote regarding the first adjoint workshop concerns its very first presentation, by Ron Errico (then at NCAR) entitled "What is an Adjoint Model." This special topic was motivated by his previous attendance at an instructional meeting at NCAR, where after a full week of long seminars on adjoint models by several illustrious speakers, one student asked "Yes, I follow all this math, but what is an adjoint?" Ron totally sympathized with the questioner and felt he could explain adjoints better. If he had to be invited to give such a talk, he would organize his own meeting to do just that. More pointedly, however, Ron was motived by his strong desire to encourage adjoint applications to questions in dynamic and synoptic meteorology. That is one reason why almost all the workshops have begun with presentations on just such topics. Even so, there continue to be too few such presentations, reflecting both the continued ignorance of the power of adjoint models and the general unavailability of suitable model adjoints to employ.

Another interesting anecdote regarding the first workshop concerns the use of the term "singular vector" in dynamic meteorology. Before then, different investigators used different words to denote these "optimal structures." During an evening gathering in a small room packed with people, mostly sitting on the floor, with the passing of lots of wine, several thought the terminology should be made uniform. Someone suggested "singular vectors" so this became the norm. The next day, one attendee regretted this because the label is so generic, without connoting any specifics or significance regarding the meteorological applications presented. He had been too tired and wine imbibed to protest the night before! So, today we all use the uninformative label "singular vectors" or "SVs."

Over the years many attendees have attended several workshops. To some, it has even become like a science-family reunion. The quality of talks have been quite high. At the 11th workshop there were typically 10 public questions or audience comments after each talk. Organizers of some other meetings have used it as their own model. So, the Adjoint Workshop series should be called especially successful.

Notes:

Photos from previous workshops