I travel a lot for my work in particle physics, but it’s usually the same places over and over again — Fermilab, CERN, sometimes Washington to meet with our gracious supporters from the funding agencies. It’s much more interesting to go someplace new, and especially somewhere that has some science going on that isn’t particle physics. I always find myself trying to make connections between other people’s work and mine.
This week I went to a meeting of the Council of the Open Science Grid that was hosted by the Oklahoma University Supercomputing Center for Education and Research in Norman, OK. It was already interesting that I got to visit Oklahoma, where I had never been before. (I think I’m up to 37 states now.) But we held our meeting in the building that hosts the National Weather Center, which gave me an opportunity to take a tour of the center and learn a bit more about how research in meteorology and weather forecasting is done.
OU is the home of the largest meteorology department in the country, and the center hosts a forecast office of the National Weather Service (which produces forecasts for central and western Oklahoma and northern Texas, at the granularity of one hour and one kilometer) and the National Severe Storms Laboratory (which generates storm watches and warnings for the entire country — I saw the actual desk where the decisions get made!). So how is the science of the weather like and not like the science that we do at the LHC?
(In what follows, I offer my sincere apologies to meteorologists in case I misinterpreted what I learned on my tour!)
Both are fields that can generate significant amounts of data that need to be interpreted to obtain a scientific result. As has been discussed many times on the blog, each LHC experiment records petabytes of data each year. Meteorology research is performed by much smaller teams of observers, which makes it hard to estimate their total data volume, but the graduate student who led our tour told us that he is studying a mere three weather events, but he has more than a terabyte of data to contend with — small compared to what a student on the LHC might have to handle, but still significant.
But where the two fields differ is what limits the rate at which the data can be understood. At the LHC, it’s all about the processing power needed to reconstruct the raw data by performing the algorithms that turn the voltages read out from millions of amplifiers into the energies and momenta of individual elementary particles. We know what the algorithms for this are, we know how to code them; we just have to run them a lot. In meteorology, the challenge is getting to the point where you can even make the data interpretable in a scientific sense. Things like radar readings still need to be massaged by humans to become sensible. It is a very labor-intensive process, akin to the work done by the “scanner girls” of the particle physics days of yore, who carefully studied film emulsions by eye to identify particle tracks. I do wonder what the prospects are in meteorology for automating this process so that it can be handed off to humans instead. (Clearly this has to apply more towards forefront research in the field about how tornadoes form and the like, rather than to the daily weather predictions that just tell you the likelihood of tornado-forming conditions.)
Weather forecasting data is generally public information, accessible by anyone. The National Weather Service publishes it in a form that has already had some processing done on it so that it can be straightforwardly ingested by others. Indeed, there is a significant private weather-forecasting industry that makes use of this, and sells products with value added to the NWS data. (For instance, you could buy a forecast much more granular than that provided by the NWS, e.g. for the weather at your house in ten-minute intervals.) Many of these companies rent space in buildings within a block of the National Weather Center. The field of particle physics is still struggling with how to make our data publicly available (which puts us well behind many astronomy projects which make all of their data public within a few years of the original observations). There are concerns about how to provide the data in a form that will allow people who are not experts to learn something from the data without making mistakes. But there has been quite a lot of progress in this in recent years, especially as it has been recognized that each particle physics experiment creates a unique dataset that will probably never be replicated in the future. We can expect an increasing number of public data releases in the next few years. (On that note, let me point out the NSF-funded Data and Software Preservation for Open Science (DASPOS) project that I am associated with on its very outer edges, which is working on some aspects of the problem.) However, I’d be surprised if anyone starts up a company that will sell new interpretations of LHC data!
Finally, here’s one thing that the weather and the LHC has in common — they’re both always on! Or, at least we try to run the LHC for every minute possible when the accelerator is operational. (Remember, we are currently down for upgrades and will start up again this coming spring.) The LHC experiments have physicists on on duty 24 hours a day, monitoring data quality and ready to make repairs to the detectors should they be needed. Weather forecasters are also on shift at the forecasting center and the severe-storm center around the clock. They are busy looking at data being gathered by their own instruments, but also from other sources. For instance, when there are reports of tornadoes near Oklahoma City, the local TV news stations often send helicopters out to go take a look. The forecasters watch the TV news to get additional perspectives on the storm.
Now, if only the weather forecasters on shift could make repairs to the weather just like our shifters can fix the detector!