I have been posting my forecasts on Facebook & Twitter, honing my social media skills. I hope to post here more often as severe weather season ramps up to a July peak, and then the tropics enter their peak months of Aug. through Oct. For now, look on social media first as links to blog entries & videos (usually using Google Hangouts On Air) will be posted there. With that out of the way, I now take you back in time to the last two weeks of October 2012.
The date is October 19th, 2012. A disturbance in the Caribbean has just garnered the National Hurricane Center's (NHC) attention. The Canadian-run Model (CMC) is the first to develop it into a named storm, as it heads north. Thirty-six hours after the NHC starts watching it (at noon ET on 10/22), they classify it as Tropical Depression 18. At 5 PM, it becomes the 18th named storm of the season... Sandy. The European model forecasts Sandy to turn to the NW in about nine days. Meteorologists shrug it off... for now. At 11 AM Wed., the NHC classifies Sandy as a Hurricane as it nears Jamaica. A few hours later, the 12z run of the Euro comes in: this is the sixth consecutive run that it forecasts a left-hand turn past the SE coast. Meteorologists analyze the synoptic pattern and agree that the disturbance coming in from the Pacific NW and a blocking high developing over the Atlantic will turn Sandy to the left. However, the American-run Global Forecast System (GFS) & its ensembles (GEFS), along with just under half of the European models' ensembles, say that it will go out to sea. To better predict its track, the NWS, for the first or second time ever, orders an increase in the frequency of launches from all weather balloon stations in the entire US: from every 12 hours to every 6 hours. On the 25th, as Sandy emerges from Cuba, the GFS & GEFS start to show the left hook. After some initial confusion as to the forecast from the NHC, on Saturday the 27th, NYC orders evacuations. Many others municipalities in the vicinity of NY, NJ, & New England order evacuations as well. On October 29th, at approximately 8 PM EDT (00z 10/30), Sandy makes landfall near Atlantic City, NJ, as a hybrid storm (a disturbance with both tropical & extra-tropical characteristics). The NHC - having issued advisories every three hours but no hurricane & tropical storm alerts north of VA - transfers advisories to the HPC/ WPC. The recovery process begins over the next few days as Sandy's remnants headed into Canada.
So, why did the European-run Model predict that left hook more than a week in advance while the GFS took an additional four days to come around? To answer that, we must ask what is a weather model? They are the solution to a set of partial differential physics-based equations run on Supercomputers. The global models run by the Europeans, Canadians, and us are called global spectral models, which solve these equations mathematically by assuming a wave-like solution (using Taylor/ MacLauren series expansion of trigonometric functions). Our regional models & the United Kingdom's global model (UKMET) are solved at evenly-spaced grid points, and are called - creatively enough - grid-point models. Now, we can answer the first question: how did the Europeans solve their model correctly sooner than us?
Well, the European Center for Medium-Range Weather Forecasts (ECMWF) in Reading, England, have powerful supercomputers that are devoted to running their high-resolution global spectral model every 12 hours. The National Weather Service had, in Oct. 2012, just enough supercomputers to run a suite of models every six hours, limiting the GFS's ability to spend more time to initialize (the Euro spends at least six hours doing so), a key part in the process. The low computing power has also forced the GFS's resolution to 27 km as opposed to the ECMWF's 16.7-km resolution. A budget-fight-weary Congress, under "extreme" public pressure, then allocated some of the relief funds to the NWS for purchasing new Supercomputers (one was already ordered before Sandy hit & that was turned on in July 2013), and the Environmental Modeling Center, which runs the GFS, put in the order to IBM, then an American-based company. NCEP EMC subsequently "announced" that once the Sandy-funded computers come online, the GFS & GEFS would be upgraded & then the regional grid-point models (NAM, SREF, NMM, RAP, ARW), which are nested within the GFS & GEFS, would then receive their upgrade.
However, according to Cliff Mass (Univ. of Washington), Lenovo (a Chinese-owned company) has purchased a majority share of IBM, putting the order at risk. I asked the EMC, via Facebook message, about this and they said that they're only allowed to order from American Companies. At the time of the order, IBM was owned by Americans, but was probably negotiating with Lenovo at the time. This could be grounds for invalidating the contract with IBM for the new supercomputer. While that's being sorted out, the GFS & GEFS are 4th globally in forecasting ability... having trouble predicting even day-to-day forecasts, and it gets worse. The European Center has ordered a new American-made Supercomputer to run a 10-km model. RESOLVED: Cray technologies, one of Lenovo's/ IBM's subsidiaries, has taken on NOAA's order.
(In my opinion, however, current research & performance suggests that running a 10-km global spectral model is a bad idea. I'd run a series of nested continental grid models instead, with resolutions (in km) of 16, 12, 4, and 4/3.)
So, when is the NWS getting the Supercomputer? At this time, it is still scheduled for a delivery within the next year or so, but that could change depending on the status of the NWS's contract with IBM. Let's hope there are no (extremely long) delays.
No comments:
Post a Comment