Why An Index Is Important...Added : Thursday at 18:30 'Twas a lazy Wednesday morning here at the Towers and everything on the servers were humming away nicely and we were looking forward to covering the tropical systems and then all of a sudden the overnight GFS ► Ensembles started shuddering to a halt...
Usually, when this occurs we either find out that NCEP are having problems or perhaps we are having network problems? A few minutes later we had a message from NCEP to say that they were experiencing network issues once again... at least it wasn't our servers anyway.
An hour or so later and there were still no signs of the GFS ► ensembles which we use to calculate RainRisk, SnowRisk etc. At around 9.30am we pulled the plug on the run and hoped that the 6z GFS ► would come through on time.
Meanwhile, NCEP found that a dodgy router which had failed when a vendor upgraded it was the problem and they set out to fix it, but at the same time they moved all their super computer facilities from the main site in Boulder, CO to Arizona.
At that point things sprung into life and all of sudden we were getting the GRIB files (Gridded Binary) which is how weather data is packed and also the IDX files which are the index files.
Here is a look inside one :-
They aren't very pretty, but what they do is allow us to cherry pick which variables we want to decode and turn into the data you see on the site or the images which we create.
An index file is vital... The first column tells us the record number, the second is where in bytes it starts in the GRIB file and other columns tell us the variable and also the timestamp etc.
So, yesterday the files came back online from Arizona, but all our models began failing once again as they decoded the GRIB files.
After a few minutes of head scratching and a cup of coffee we decided that the index files must be wrong. So, we downloaded a GRIB file and created our own index and indeed it was different.
The reason, was when NCEP switched to Arizona, it restarted many of the runs like the GFS ► and GFS ► Ensembles and overwrote the GRIB files but didn't update the index files as they were already there!
It's a bit like releasing a cookery book, then adding a few more delicious recipes like blueberry muffins etc and then ripping out the old index and sticking it in the new book... the index was wrong. So if you were looking up blueberry muffins, you might end up staring at a chicken tikka masala instead on page 58!
That would obviously annoy you, but weather decoders don't get annoyed... they just break... We're the ones who get annoyed...
So, a delay to the data yesterday and also the 180 day CFS forecast broke today as it was reading files released by NCEP yesterday.
This error is pretty unique and nothing we have ever encountered before. We let NCEP know this issue and they will work on correcting this if it happens again. Meanwhile, we have also taken steps to ensure that if a main server switches at NCEP then we will check the index files first and if wrong create our own.
All is now working. But we just wanted to explain all the processes behind the scenes which turn your forecasts into lovely colours and icons whilst you nibble away on a blueberry muffin...
METEOROLOGIST : CHICKENTIKKAMASALA |