Wednesday, November 14, 2007

More Wayback Machine

In early 1975, I went to work for a company called Systems Applications Inc. The job literally fell into my lap.

Henry and I were staying with Douglas in Berkeley, looking for jobs and sleeping on the floor. I'd had a part-time job in the fall of '74 writing practice test questions for a small firm offering a course in passing the FCC 1st Class Radiotelephone License exam, but in January of '75 I signed onto the quality assurance group at the Nuclear Submarine Fueling Station at Mare Island Shipyards in Vallejo. I wasn't fond of the job; it was winter and I had to be there before dawn. I'm not a morning person.

One evening, after I got back from work, Henry came in and dropped a 3 x 5 index card in my lap. "I talked to these guys on the phone," he told me. "It's not a good fit for me, but it looks like it would be right up your alley."

See, when I say "literally dropped into my lap," I mean literally.

SAI had begun its life as a consulting firm for telecommunication policy, primarily for the Office of Telecommunication Policy (which no longer exists, so enough of this "government bureaus never die" crap). Later, some disgruntled folks from Shell Oil, along with some Caltech brainpower, signed on to bid on a "seed money" contract to the USEPA, to develop a simulation model for urban smog, to ultimately be used in devising State Implementation Plans (SIPs) for urban smog abatement. There were three such contracts originally; it was essentially a competition, with the best initial design getting a much larger follow-on project to continue model development.

Okay, quick nurd stuff. There are two ways of modeling fluid mechanics, Lagrangian and Eulerian, named after 18th Century mathematicians. Eulerian modeling is probably the easiest to describe and understand: just divide the volume holding the fluid into a lot of small, connected boxes, and calculated the flows among the boxes. For incompressible fluids (and on the urban scale, air can be considered incompressible, though you sometimes have to make adjustments for altitude), you can take advantage of those nice conservation of mass laws.

In Lagrangian mechanics, you select a bit of the fluid and you follow it around, like watching snowflakes in the wind. Over time, you can follow the trajectory of an individual snowflake and that tells you how the wind got from point A to point B. Lagrangian models are sometimes called trajectory models, for the obvious reason.

Two of the three companies developing smog models chose the Lagrangian approach, creating what are called "trajectory models," because you are following the trajectory of an air parcel over land. Such models are much computationally cheaper than Eulerian models (often called "grid models"), and are also cheaper to develop. You don't need to worry about the fluid flow equations, for one thing. The trajectory can simply follow the estimates of wind speed and direction, getting those estimates from the nearest wind stations. So if you wanted to model the observed ozone peak at an individual monitoring station, you just "back calculated" the air parcel trajectory to some starting point, like sunrise, then ran it over the emissions field and calculated how the chemistry behaved, until it ran into the monitoring station.

Computer time was expensive in 1975. A trajectory model might have 3-5 stacked boxes in the air parcel, and you might have to run the thing 10-12 times to get a full prediction at a single monitoring station. But a grid model would typically have at least 25 cells in both horizontal directions, plus the same number of stacked boxes that a trajectory model would have. You can run a lot of trajectory simulations when the difference in computing cost per simulation is almost 3 orders of magnitude. There were some concerns as to whether or not a grid model could be made to work at all.

SAI, however, did manage to produce a photochemical grid model, in some measure thanks to the CDC 7600 and a lot of prior academic research on computing fluid flow etc. So SAI won the follow-on contract, and hired several new people. I was one of them. My job? To develop a trajectory model.

Okay, yeah, that's a bit funny. They won because they'd developed a grid model and one of the first things they did was develop a trajectory model. But it did make sense. As I say, trajectory models are much cheaper to run. They are also easier to diagnose and debug, because they are simpler. And there were a lot of bells and whistles that were slated for inclusion in the final product, things like surface deposition (assessing how rapidly smog ozone is destroyed by ground surfaces), changes in light through the air column (smog is hazy and haze redistributes light), microscale effects (does chemistry that takes place at small scales have a big impact on a 5 mile x 5 mile grid?), and so forth. There was also an ongoing development contract to research smog chemistry, so it was useful to have a cheap version of what would go into the grid model, in order to test that against other, more sophisticated chemical kinetics solvers.

So, after the usual water-up-your-nose that occur when you jump feet first into a new pool, I got the trajectory model running. I also got to know the new guy running the atmospheric chemistry show, Gary Whitten (the previous guy left for a lucrative career at Chevron), and learned how Whitten's new ideas about smog chemistry worked. This was called the "Carbon Bond Mechanism" (which gets about 350 hits on Google scholar). I learned it from the guts out, as I had to hard code every single reaction into the chemistry solver shared by the trajectory and grid models.

During the next couple of years, I also worked out the method for estimating pollutant depositions on surfaces, got estimates for how the photolysis of nitrogen dioxide and other important species varied with solar zenith angle, and provided a quick and dirty (so to speak) method for calculating aerosol haze formation in smog, along with how the haze affected the photolysis of the important chemical species. I revised the emissions inventories that we had, because those came in the form of simply "reactive hydrocarbon" (RHC) or "non-methane hydrocarbon" (NMHC), and we needed the hydrocarbons split into different reactive species.

I also coded a new vertical dispersion algorithm into the model. I had no input into this particular piece of work; the guy in charge of it tended to treat anyone else, as one of the programmers put it, "as just another pair of hands." I'm pretty sure he actually got it wrong, because his implementation used calculations at a point for diffusivity, but the algorithm he was using varied considerably over the bottom grid cell. Diffusivity is rather like conductivity; you can't use averages for conductivity and get meaningful results. You have to use its inverse: resistance. It's also called resistance in diffusion calculations as well, and getting that right was critical for the surface deposition calculations.

Later, I worked out some new methods for calculating wind fields that reduced some modeling artifacts caused by a spurious convergence that is created when winds turn. I think I had a handle on how to do really right when everyone switched over to what are called "prognostic models" for winds (basically using a full fluid flow model for your wind field), so I never got to see that idea in action.

And if all this sounds very productive, realize that I haven't even mentioned the work that I was doing with Whitten on the basic photochemistry of aromatic hydrocarbons, isoprene and other biogenics, and peroxyacetyl nitrate (PAN).

So there I was, fresh out of school with a newly minted Master's Degree, which meant that I was cheaper than any PhD. And within a short period of time I was doing major development and scientific work that is being cited to this day. Within a year of my hiring I was technical lead on the Denver modeling project that was the first application of the new, realistic chemistry that Whitten had developed. I did literature reviews on atmospheric sources of nitrogen oxides, including a pretty comprehensive review of nitrogen oxide production from internal combustion engines. At one point one of the senior scientists said that I was "essential" to any urban airshed modeling project that SAI wanted to undertake.

I was also, apparently, so fundamentally obnoxious that years later, Whitten told me that at a management meeting in late 1975, he was the only manager who was willing to supervise me. He followed that with, "I never saw what was so hard about it. A project would come up. I'd talk to you about it for a bit. You'd usually be pretty negative and pessimistic at first, then something would catch your interest. Then all I had to do was wait for you to come back and report on what you'd done, which you'd do every couple of days."

I'm reasonably sure I'm more easygoing and likeable these days. But that's still pretty much the way to manage me. Some managers are fine with it. Others, it drives up the wall. Sorry.

2 comments:

black dog barking said...

I was also, apparently, so fundamentally obnoxious that …

That was a good time to be a young whippersnapper because technology was leveling the field. Given the choices of learning to map your own experiences to FORTRAN or tolerating someone that could, I can see (an extremely!) reluctant tolerance winning out. Can't imagine getting too chummy with someone that a) lacks credentials or experience and b) gets the job done.

Middle management is where youth goes to die.

James Killus said...

Well, eventually I got the experience, and credentials too, if publishing papers counts. I used to joke that they'd put up with anything I did, so long as I also produced at least two miracles a year. Then I got sick and the miracles stopped, and it took a while to come back from that. But then it took rather a long while to stop being sick, and that was the more maturing experience, I think.