How The Advance Weather Forecast Got Good (npr.org) 80
NPR notes today's "supercomputer-driven" weather modelling can crunch huge amounts of data to accurately forecast the weather a week in advance -- pointing out that "a six-day weather forecast today is as good as a two-day forecast was in the 1970s."
Here's some highlights from their interview with Andrew Blum, author of The Weather Machine: A Journey Inside the Forecast : One of the things that's happened as the scale in the system has shifted to the computers is that it's no longer bound by past experience. It's no longer, the meteorologists say, "Well, this happened in the past, we can expect it to happen again." We're more ready for these new extremes because we're not held down by past expectations...
The models are really a kind of ongoing concern. ... They run ahead in time, and then every six hours or every 12 hours, they compare their own forecast with the latest observations. And so the models in reality are ... sort of dancing together, where the model makes a forecast and it's corrected slightly by the observations that are coming in...
It's definitely run by individual nations -- but individual nations with their systems tied together... It's a 150-year-old system of governments collaborating with each other as a global public good... The positive example from last month was with Cyclone Fani in India. And this was a very similar storm to one 20 years ago, that tens of thousands of people had died. This time around, the forecast came far enough in advance and with enough confidence that the Indian government was able to move a million people out of the way.
Here's some highlights from their interview with Andrew Blum, author of The Weather Machine: A Journey Inside the Forecast : One of the things that's happened as the scale in the system has shifted to the computers is that it's no longer bound by past experience. It's no longer, the meteorologists say, "Well, this happened in the past, we can expect it to happen again." We're more ready for these new extremes because we're not held down by past expectations...
The models are really a kind of ongoing concern. ... They run ahead in time, and then every six hours or every 12 hours, they compare their own forecast with the latest observations. And so the models in reality are ... sort of dancing together, where the model makes a forecast and it's corrected slightly by the observations that are coming in...
It's definitely run by individual nations -- but individual nations with their systems tied together... It's a 150-year-old system of governments collaborating with each other as a global public good... The positive example from last month was with Cyclone Fani in India. And this was a very similar storm to one 20 years ago, that tens of thousands of people had died. This time around, the forecast came far enough in advance and with enough confidence that the Indian government was able to move a million people out of the way.
Re: "Got Good" (Score:2)
Git gud! Or quit playing,newb!
Re: (Score:2)
Re:Data or calculation? (Score:5, Informative)
The increases in resolution and sampling rate have helped a LOT, especially with detecting tornadoes as they form. Recognizing a tornado from low-resolution weather radar that updates every 5 minutes is HARD. Recognizing a tornado forming with high-resolution phased array radar that performs a complete quadrant volume scan every few seconds is almost child's play by comparison.
For a textbook-perfect example, consider the 2013 El Reno tornado, which was captured using NOAA's experimental phased-array radar. You can LITERALLY see the hook echo form in near-realtime starting ~14 seconds into the video -- https://www.youtube.com/watch?... [youtube.com]
With rapidly-updating high-res data like you'd get from PAR, you barely even NEED to do much more than render it to a bitmap and throw image-recognition software at the problem. In contrast, if you look at the conventional WSR88-D radar data for the same tornado, you can still detect the tornado by analyzing additional data, but until someone confirms the sighting of a tornado on the ground and its approximate location, your confidence will be fairly low. In contrast, if you see a hook echo like that form on PAR, there's no hand-wringing... it's almost certainly, beyond doubt, a tornado.
Unfortunately, PAR is still way too expensive to deploy as a generalized replacement for WSR88-D radar. By several orders of magnitude. Even in Oklahoma, they only had enough money to deploy enough of an array to scan 90 degrees at a time & had to physically rotate the array itself to point it in the direction of the weather of interest (which you can see them doing as the storm moves from east to west). Still, it shows what's possible when you're able to replace sparse, infrequent data with high-res frequently-updating "big data".
Re: (Score:2)
Holy moly, that is gorgeous radar footage.
With the old NWS doppler radar in my area, I have been able to see hook echos happen. They have turned into EF0 or EF1 at most, depending on where it happens. Luckily, most of them funneled and dissipated.
This is all based on their website, and radar images. I wish they'd update from Shockwave Flash to HTML5 for the radar loop/zoom on the website, but again. Money.
Re:Data or calculation? (Score:4, Interesting)
Personally, I think NOAA should explore the idea (possibly getting Congress and the FCC to do some arm-twisting, if necessary) of deploying tens of thousands of relatively low-cost short-range digital radar systems onto cell phone towers (involving the FCC to make it a condition of license approval going forward from some date 5-10 years in the future).
Suppose that instead of trying to send out a megawatt-pulse with 480 nautical mile range with a rotating antenna, you did the equivalent of mount 16 directional antennas with the approximate gain of Yagis (just to use a simple example), and instead of transmitting on-off pulses, you took advantage of a lower-power signal to digitally modulate it using FHSS and CDMA (so that every "pulse" could be uniquely matched to a specific tower and sequence number of pulses transmitted by that tower... say, some incrementing/rolling-over value between 0..16 for sequence, along with a 4-6 bit ID unique among towers within double the signal's range. Since you'd then be able to match a "pulse" with the tower that sent it and sequence number, you'd eliminate the range-folding problem, gain the ability to use transmitting radar sites as illuminators for OTHER nearby radar sites (so you wouldn't just see the echoes that bounced back 180 degrees... a neighboring tower site might see an echo that bounced off at a 30-degree angle, for example).
Going back to the 16-yagi idea. Suppose you had the 16 yagi antennas arranged in a circle 22.5 degrees apart, inclined upwards at something like 1-5 degrees with a range of 10km max. In an urban area with cell towers every mile or two, you wouldn't NEED to worry about tilting the antenna, because you'd have enough tower density to just use one tower for your "low" tilt, use the adjacent towers for the next tilt up, use the next-adjacent towers for the third tilt, and so on until you hit the limit of the signal range. The spatial resolution would be poor... but instead of being able to do a volume scan every ~15 seconds (which I believe is what the PAR system in Oklahoma did), you could sustain literally a 16-direction 360-degree read per second... or 10 per second... or more. In effect, by aggregating the data from hundreds or thousands of such systems across a large urban area, you'd have a real-time view of weather near the ground.
For tornadoes, the system would probably be far superior to WSR-88D, because it would be concerned almost ENTIRELY with things going on below ~2000 feet... precisely the area where WSR88D (and to a lesser extent, TDWR) has the poorest coverage, and where most of the interesting things going on with tornadoes tends to occur. Sure... coverage out in rural areas wouldn't be great. But then again, if an EF5 digs a trench through a forest, it doesn't directly affect very many people. In contrast, an EF3 blowing apart a neighborhood of townhomes in the middle of a big city is an EXTRAORDINARILY big deal. The system wouldn't REPLACE WSR88-D... it would mostly just fill in the present-day big coverage gap in places where it makes the biggest difference: urban areas.
Another reason to locate the systems on cell phone towers: there's ALREADY fiber or microwave backhaul there. And at a tower site that serves multiple carriers, you'd have multiple redundant backhaul paths to get the data to NOAA's ingest servers.
Or... we could do what El Salvador did, and scatter lots of regional X-band radars networked together across the country. Seriously, I can't find the link to it now, but I was BLOWN AWAY by the density of weather radar sites in El Salvador. Pretty much any given square block of a big city is going to have coverage from a regional radar site, plus at least one... probably two or three... local radar sites with overlapping coverage. I wasn't able to find a whole lot of other information about it, but AFAIK it's kind of like a real-life experiment to demonstrate the benefits of having a dense network of lower-powered overlapping radar sites.
----
On an unrelated note... can anybody ex
Re: Data or calculation? (Score:2)
I agree about signal propagation being useful. If towers transmit periodic beacons & indicate their broadcast power, you could use neighboring sites to log the observed signal strength. Once you came up with a calibration technique (perhaps just observing them over time to establish max & min attenuation, then correlating it to observed rainfall), you'd have a handy way to detect neighborhood rainfall, even if it had no doppler capabilities.
I've actually hunted for LTE documentation to explore this
Re: (Score:2)
Wouldn't this only work where there are cell towers thought? I.e. it wouldn't be much good out at sea, which is where a lot of the interesting weather happens because you generally want to predict this stuff before it hits land?
Re: Data or calculation? (Score:2)
"at sea" is important, but few weather phenomena occurring "at sea" more than a few miles offshore that can't be detected by existing surveillance (or radar on a ship) really present an immediate danger to thousands or millions of people the way a tornado in or near an urban area does.
At satellite broadband (like Starlink) becomes available & affordable, it might be worthwhile for NOAA to come up with a way for radar on ships to upload their data in near-realtime as well (filling in more offshore gaps).
International collaboration (Score:4, Informative)
It's not as simple as saying that governments work together. The US freely disseminates a very large amount of weather data and the source code to the models. There's GFS FV3 source code [github.com] on Github. WRF-ARW [ucar.edu] (the core used in the RAP and HRRR models) and WRF-NMM [dtcenter.org] (the core used in the NAM and NAM NEST models) are freely available, too. And when these models are run on NCEP supercomputers, the data are freely distributed on servers, limited only by the available bandwidth. Anyone can modify these models, and many people contribute new code to WRF. The open nature of these models allows anyone to run the models, to find and fix bugs, and to contribute improvements. An example is how researchers determined that radiative processes are really important to tropical cyclone forecasting [ametsoc.org] and eventually discovered there were significant bugs in operational models.
Not everyone is so willing to share data and code, unfortunately. The gold standard in global models is the ECMWF model, developed by the European Centre for Mid-Range Weather Forecasting in Reading, UK. Unlike with the US models, they don't freely give their data or source code. Obtaining ECMWF data is actually quite expensive, which is unfortunate. While aspects of their model are discussed in scientific papers, one cannot readily access much of their data or obtain their source code.
Although the US openly collaborates, sharing source code and data, not everyone is so willing to do so. While National Weather Service forecasters certainly can look at ECMWF data, a researcher in the US cannot so readily do so without paying quite a bit of money for it.
Re: (Score:1)
Is it also true that part of what makes ECMWF so good is that they simply have more raw computing power than the US models have available to them?
Re: (Score:2)
Re: (Score:1)
Wait, you're complaining because they can predict thunderstorms but not tornadoes?
Re: (Score:2)
Re: (Score:1)
There's no such thing as a two day tornado prediction. You can't even make a one day prediction. All meteorologists can report is whether conditions are right for storms that could produce tornadoes during that day, but not actually predict when or where they form.
Re: (Score:1)
There are limits on our ability to produce accurate and skillful forecasts, in large part because the atmosphere is a chaotic system. A recent paper [ametsoc.org] suggested we may have reached the limit in accurately predicting tropical cyclone tracks. Generally speaking, skillful forecasts can be made at longer lead times for larger phenomena. A thunderstorm is a smaller scale process and, therefore, is more difficult to predict at longer lead times. Nonetheless, new models like the HRRR [noaa.gov] do a remarkably good job of
Re: (Score:2)
Do you understand what the number mean? seriously? 10% chance of rain doesn't mean rain on you.
Got good (Score:2)
Re: (Score:2)
A penny saved is a penny got.
Eh, still not always accurate.:P (Score:2)
Even if it is minutes away like rain! :P Also, I noticed various forecasters show different results. :/
Re: (Score:2)
You look adorable. sitting there falling for the nirvana fallacy.
Re: (Score:2)
How so?
Yeah right! (Score:2)
Re: (Score:2)
Being on the North Coast usually means the weather report is worthless as well. Supposedly they recently began doing some additional offshore modeling (and improved the quality of water vapor modeling) but they're still getting it pretty seriously wrong.
Re: (Score:2)
"so called scientist"
are you stupid?
"Once, I was heading north from Branson Missouri after a service call, and it was 53 degrees, but in Springfield Missouri, 30 miles north, it was freezing rain."
yep.
Maybe actually learn about meteorology, resolution, the difference between where you ar at, and the area the forecast is for?
The reason you d't get the is money. No one is pang a meteorologist to just do those areas, that are part of a wider area forecasted for.
In Japan flipping a coin is more accurate (Score:2)
Accurate? (Score:2)
Storms (Score:1)
Re: (Score:2)
Just not true (Score:2)
They may be more precise, but they're not more accurate - or at least not that much more. A six-day forecast is pretty worthless at the moment.
Until... (Score:3)
Until 5g fucks it up...
https://physicsworld.com/a/deb... [physicsworld.com]
. Farmers' Almanac . (Score:2)
The old reliable standby. Forecasts up to a year in advance. Not just data points: wind speed, relative humidity, etc; but soul. How will you *feel* in New England when the humidity rises and the insects are buzzing all around you? And practical: When should you plant those azaleas or grimpin poggles?
It's lovely that the billions spent on satellites and super computers can give us more detailed data, but the Farmers' Almanac gives us the Big Picture at a far more reasonable cost. And it's (almost) always ri
No penalty for getting it wrong (Score:2)
"Just as good..." Isn't there a TV ad campaign with that theme?
Bottom line: there is no penalty for getting it wrong. If there were, they wouldn't be making predictions 7 days out. Also notice that the predictions change every day.
Re: (Score:2)
yes, there is a penalty, actually.
The problem is, there isn't a penalty for idiots posting on sites about things they don't understand. As an example see: Your post.
total rubbish (Score:2)
"a six-day weather forecast today is as good as a two-day forecast was in the 1970s."
you mean, total rubbish?
in the 70's you couldn't trust the next day forecast, let alone two days.
these days the next two days are pretty accurate, however 7 days is just not very reliable.
Not in the Midwest (Score:2)
Really? This is the best they can say? (Score:1)
three comments (Score:2)
2. The wx forecast didn't got good (sheesh) in my part of the country. Forecasters around here would have trouble forecasting yesterday's wx.
3. I never believe anything NPR says, too much bias, too many lies and distortions.