From the Cliff Mass Weather Blog
Cliff Mass
Numerical weather prediction has improved dramatically over the past decades, providing potent warnings for extreme weather, such as hurricanes.
There are few better examples than the prediction of Hurricane Lee, which will make landfall near the Maine/New Brunswick border late Saturday.
The U.S. global model, the GFS, has been spectacularly skillful in predicting this storm, well more than a week ahead.
The latest forecast run shows the storm making landfall near the international border around 5 PM PDT on Sunday. That is a 54-h hour prediction This is so close enough in time…and so consistent with other model forecasts… that you can be assured that this is close to what will happen.
But how did extended forecasts do?
The 72-h prediction is pretty much the same.
The 126 h prediction is nearly identical in position:
The 198 hr (8.25 day) forecast has a strong hurricane in pretty much the same location.
Folks, this is a stunningly good forecast for over a week ahead.
Professor Brian Tang of the University of Albany has a wonderful website that verifies the hurricane track (position) forecasts of major modeling/forecasting systems. The results for Hurricane Lee are shown below for forecasts of 120 hours (5 days) or less.
In general, the track accuracy gets better for shorter forecasts…which makes sense. But let’s compare the American model (blue color, AVNO), the European Center model (red color), and the UKMET office model (green color). The human (official) forecast is shown in black.
Wow. The American model is STUNNINGLY accurate at all projections in time.
It is FAR better than the nominally top two global modeling systems in the world: the European Center and UKMET. The forecast error is under 100 km (60 miles) for all projections shown.
Extraordinary.
The model forecasts are better than the official Hurricane Center forecasts….I suspect that humans are probably hedging their bets with the European Center model solution.😅
This was a truly excellent forecast and not the only success for the American model this season. Hopefully, this extraordinary performance will be persistent for future storms, perhaps reflecting recent improvements in the U.S. global modeling system.
Finally, I should note there are real policy implications of the rapidly advancing weather prediction skill now available to decision-makers. Excellent forecasts can help protect people and economic assets from extreme weather.
Better forecasts are the first line of defense against severe weather.
Better forecasts have great potential for reducing the negative impacts of global warming.
One of the reasons I have spent some time trying to calm down some who are panicking over global warming and extreme weather.
Climate-related deaths are down…and I mean WAY down. Better adaptation and a richer world have contributed, but so have better forecasts.
Importantly, we have only begun taking advantage of improved forecast skill.
The winds on Maui were nearly perfectly predicted on August 7-8 of this year, yet 115 people died and nearly 10 billion dollars in damage was done. We could have easily stopped the carnage, by shutting off the power and effectively evacuating the population.
Most major wildfires are related to strong winds and such winds are often forecast with great skill. Few should be a surprise.
In summary, coastal New England has had nearly a week to prepare for strong winds and heavy precipitation (over northern Maine)– and we can be proud of the technological advances and investments in NOAA and in other government agencies that made such forecasting prowess possible.
How satisfying to read of something going right. And from NOAA. But to avoid misinterpretation by others I would have liked to see these accurate limited-purpose weather models clearly distinguished from the hopelessly inaccurate climate models.
I think managing fuels in Maui would have helped much more than any forecasting. Some goats or cattle eating the grass before it got so high would have been better.
The thousands of flood deaths in Derna, Libya will be blamed on “climate change”.
Consistently getting the forecast track error at 5 days to under 100km would have an enormous economic benefit as it allow for a much more focused preparation for the storm. Kudos to the group working on the GFS model for the stunningly accurate forecast.
Wait a minute, wasn’t it predicted to hit Southern Florida for a while? One day it was gone from the headlines as it stayed at sea.
I am glad GFS was so accurate. But this case was fairly simple: the macro steering elements were both well defined and fairly long lived. The big question is what did ECMWF do to be worse when it is usually better?
Put differently, sometimes you get lucky.
Nice.
California is doing it again.
https://www.foxbusiness.com/politics/california-goes-war-oil-gas-giants-climate-change-deception
story tip
So here’s the puzzle for us all – where does Meteoblue (MB) get their data from?
Because what’s happened here is exactly what MB have been saying I first checked on Lee when it was way way south and (seemed to be) headed for Florida.
MB said ‘no’ – its’ gonna stay out at sea and run aground at the day/time on the Canada/US border = exactly where/when it has done now
Meteoblue also very correctly predicted what Hilary would do, at the day/time that Hilary did what she did and I copied it into here nearly 4 or 5 days in advance
BUT, when I did that, The News, The Forecasts, MSM and youtube pundits were in Total Panic Mode – talking of a hideous hurricane shaped disaster looming for LA and places north of there
(If you just enter meteoblue-dot-com – this is where you land)
It all came soooo true – we all saw what 0.75mm (over 3 days) of torrential rain made of Black Rock Desert in Nevada. Thanks Hils
(Girls are sometimes like that – making total fools out of boys)
Meanwhile. here’s an MB screenshot for now (midnight BST Saturday 16th)
The colours are not = Temperature, the colours are indicating wind speed where Red = strong wind and Blue = weak wind
Isn’t it so lovely how it shows where all the bare ground is after the combine harvesters are nearly all finished this year’s work
See how North America and all of Europe are under heat domes = high barometric pressure and low winds.
Temps now (midnight BST) across the SE tip of England and all of Europe are about the same at ~ 20°C
On that map it’s hard to tell them apart from The Sahara. (desert)
Don’t you wonder why that might be…
Cliff Mass (and some commenters) wax lyrical about the accuracy of US predicted hurricane landings.
I’m bemused that little information is given about the accuracy of the predicted STRENGTH of the hurricanes.
I do hope that the US predictions of strength are just as impressive as those of location.
Certainly, accurate prediction of both would be a huge life saver.
Much to the disappointment of our shroud waving climate apocalypse chums, of course.
In recent experience, the UK MET prophecies of storm strength or heat waves are pretty much a bad joke.
These model forecast about the weather are NOT climate model forecast.
Short term. Fewer variables. If a value for something is off a couple of decimal points? Short term no great effect in the forecast. Observations are “quick”. Success for the forecast is quickly evident.
Climate models have been around for a few decades now. They have yet to be even close.
I will agree from start to finish the GFS did pretty well with this one. But before we get carried away, this time of year in the Northern Hemisphere weather systems, fronts and winds are fairly predictable. As the seasons moves on the weather gets more chaotic and so do the forecasts.
Story Tip — William Happer live now on Outsiders – Sky News Australia. 10:30HRS est
I disagree. The UKMET model at Weather Underground is a better model. GFS tends to vary too much before finally setting onto the track show by UKMET. The UKMET had Ian hitting Ft Myers a full week in advance. It also had Idalia well off shore near me the entire time. It had Lee making the turn well before the GFS model setttled in. Because of my location, on the central west Coast of Florida, the decisions concerning my life and property depends on me trusting the forecast. For that I prefer the UKMET model.
NOAA has had significant upgrades to their computing power over the last few years.
https://www.noaa.gov/news-release/us-supercomputers-for-weather-and-climate-forecasts-get-major-bump
The U.S. global model, the GFS, has been spectacularly skillful in predicting this storm, well more than a week ahead.
Even a blind hog finds an acorn every once in a while. From what I have observed the European wins the majority of the time.
But projecting tracks has generally gotten better with time. Wish I could say the same for climate models.
My real issue with NOAA is that they will not adopt a better way of classifying storms in order to have a real reference of the power of a given storm tto better forewarn those in the potential path and to be able to do a far more accurate reanalysis of storms of the past.
The Saffir Simpson is only about sustained windspeed at 10 meters. In tropical cyclones the size of the wind field matters a great deal when it comes to the actual impact at landfall. It’s not just the area of the wind fields, it is also about storm surge. All other variables being equal but size, the larger the storm the more severe the surge.
Lee was a monster of a storm in size relative to the majority of the Atlantic basin Hurricanes we have been seeing the last few years.
The skillfulness of computer models (of all types) is dependent on
🗹 1(a). how well the processes which they model are understood, AND
🗹 1(b). how faithfully those processes are simulated in the computer code, AND
🗹 2. whether the results can be repeatedly tested so that the models can be refined.
If you can check ALL those boxes then you can certainly build a useful computer model.
If you can check EITHER boxes 1(a) & (b) or box 2, but not both, then you still have a fighting chance of building something useful, but it will be very challenging
There are an average of about 86 named storms per year, so the models for predicting their development and tracks get tested over and over, every year. That’s why they’re improving.
But the current crop of GCMs (climate models) are all untested, and they are effectively untestable, because the predictions which they make are for the distant future.
The predictions of some very early GCMs can now be compared against reality, but (contrary to claims by some climate alarmists) they failed miserably. I wrote about the most famous of them, here:
https://www.sealevel.info/hansen1988_retrospective.html
Weather models and climate models have one problem in common: we don’t really understand what we’re trying to model very well. But climate models also have a second problem: their predictions are for so far into the future that they cannot be checked.
Climate modelers sometimes cite “hindcasting” as evidence that their models are accurate. That’s wrong, because that’s the same data the models are tuned to. It’s not evidence that they’re any good at forecasting the future, because, in the words of the late Larry Niven, “Any damn fool can predict the past.”
For instance, a model predicated on the assumption that rates of cheese consumption govern deaths by bedsheet entanglement can do a good job of hindcasting actual deaths by bedsheet entanglement. But it’s still 100% worthless for forecasting future deaths by bedsheet entanglement.
Short-term weather forecasts are often wrong, yet they’re noticeably better than they used to be. They’ve been gradually improving for many years. but those improvements are only possible because 10-day weather forecasts are continually tested against reality. How good do you think weather models would be if their forecasts were untestable, like climate models are?
By the time a GCM’s predictions can be compared to reality, it is long obsolete (and the source code has probably been lost).
Most climate modelers at NASA and elsewhere think that that their sophisticated (but untestable) computer models of the Earth’s climate make reliable predictions. That belief is based on nothing but wishful thinking.
Do you remember NASA’s “sophisticated computer model of the sun’s inner dynamo,” and its prediction for Solar Cycle 24? Those modelers were confident in that model, too.
https://science.nasa.gov/science-news/science-at-nasa/2006/21dec_cycle24/#:~:text=a%20sophisticated%20computer%20model%20of%20the%20sun%E2%80%99s%20inner%20dynamo%20to%20produce%20a%20physics%2Dbased%20prediction
Excerpt:
It turned out to be the weakest in a century.
(The only thing consistent from one solar cycle to the next is that NASA’s “sophisticated computer model of the sun’s inner dynamo” is apparently always wrong. They predicted that Solar Cycle 25 would be another very weak one, but it is shaping up to be much stronger than SC24 was.)
Ask anyone in the weather biz how good they think weather models would be if they’d NEVER been tested at all, and NEVER corrected when they made inaccurate forecasts. That’s the fundamental problem with climate models: they’re basically untestable.
The fact that CMIP6 models vary wildly in their baked-in assumptions about even the most basic & heavily studied climate parameters proves that the Earth’s climate is not well-enough understood to be modellable. If anyone ever says that GCMs do a good job of modeling reality, point out to them that the CMIP6 models differ from one another by a factor of three (!!!) in their baked-in estimates of ECS climate sensitivity. If the modelers knew what they were doing, that’d obviously be impossible.
Better forecast are a good thing but not the only good thing. Experience has show that adequately constructed structures survive hurricanes with minimal damage, yet the $ cost of hurricane damage continues to grow in many instances, along with the value of the human infrastructure in hurricane prone areas. What would be very interesting is a comparison between the repair and replacement costs (always computed after a storm) with what it would have cost to have built adequately in the first place. I know that many people cannot afford to build adequately, it is a societal level problem, but reasonable numbers would be interesting and might be useful.
Let’s just hope it doesn’t cause the death and destruction of the Newfoundland Hurricane of 1775 which was Canada’s all time worst natural disaster, killing over 4,000 people back in the later part of the Little Ice Age.
The tropical storm will again bypass Hawaii from the south, which means an easterly circulation continues in the central Pacific.
I am just guessing that the storm predictions have zero sensitivity to CO2 in the simulation.
Great to highlight but then again, even a blind squirrel catches a nut every now and then.
From the article: “Better forecasts have great potential for reducing the negative impacts of global warming.”
Well, I have to ask the question: What negative impacts?
I don’t see any negative impacts coming from CO2. I’m assuming “global warming” means human-caused/CO2-caused global warming.
There is no evidence that CO2 is causing global warming, at least not enough warming to be detectable, therefore, no evidence, and certainly not enough to have negative impacts, or change how the Earth’s weather unfolds. No evidence.
From the above article, referring to the second-to-last graph (the bar graph) of positional error versus forecast lead time:
“Wow. The American model is STUNNINGLY accurate at all projections in time.
“It is FAR better than the nominally top two global modeling systems in the world: the European Center and UKMET. The forecast error is under 100 km (60 miles) for all projections shown.
“Extraordinary.”
My take is a bit more reserved. I don’t view a 120 hour-in-advance predicted landfall positional error of about 80 km, nor a 24 hour-in-advance predicted landfall positional error of about 40 km, as anything exceptional.
And this is for the single example of Hurricane Lee . . . what are the similar positional errors vs. forecast lead time for other hurricanes having more erratic tracks over, say, the last three years? . . . much worse I daresay. Without consideration of such, this article can be considered to be nothing more than “cherry picking” to front an unfounded assertion.
After all:
“The first hurricane forecast models (dynamical and statistical) were developed during the 1950’s in response to two major technological advancements: aircraft reconnaissance of hurricanes starting in the mid-1940’s, which provided accurate estimates of a hurricane’s current position and intensity, and the development of computer technology in the mid-1950’s.”
— http://www.hurricanescience.org/science/forecast/models/modelshistory/
This is the best we can do after some 70 years of modeling hurricane tracks, especially given the additional tracking and weather forcing inputs obtained from orbiting weather satellites starting in the mid-1960’s?
Even a blind pig finds a acorn once a while. When they get the tracks right over 50% or the time then we may have something.