TLDR: the input data for the model now also includes real-time observations (satellite and weather stations) on top of the typical (re)analysis data, improving model resolution, run frequency and timestep frequency.
The Netherlands might not have the most extensive or fast rail network, but it's frequencies are pretty much unbeatable. In some places it runs more as a metro network than a heavy rail one. This visualisation shows that nicely!
The Monaco number is a little unintuitive - it has one line running through it, but the principality's tiny physical footprint makes it look railway dense on that chart
And as far as delays go, I believe it is (or was, last I looked) nr 3 in the world, after Japan and Switzerland.
Of course, Japan's trains are world-class, but at least those are widely recognised as such. The Dutch rail network is much denser than Switzerlands, but catches a lot of flak.
(Which is not to say that it can't or shouldn't be improved, but it's also good to recognise what's going well.)
Wasn't my experience, frequencies are fine but I don't think it's unbeatable at all. The default simulation is 50x speed by the way, perhaps that gave you the impression.
Based on... What exactly? The Dutch rail network is already considered the busiest network in Europe. So I'm really interested what knowledge you bring that says that it could be busier if many experts say that the busiest trajectories (look at Amsterdam - Schiphol and Amsterdam - Utrecht, for example) are at max capacity already.
Upgrading to full ERTMS can increase capacity on the Utrecht Amsterdam line. Longer platforms with longer trains. Reducing accessibility in the trains* for more seating. Less seats. Removing first class. Remove train crossing. Buy out NIMBIES. There is plenty you can do to increase passenger capacity on that line.
*) If you want to absolutely maximize passengers in a train you can reduce the number of toilets to one or switch to the peebag system the old trains had. And reduce the amount of wheelchair spots to the absolutely lawfully required amount. I’m not saying this is a good thing, just spinning some ideas to increase passenger capacity.
Japan beats Dutch train frequencies, hence it's not unbeatable.
And if you include metro rail there are plenty of more frequent rail systems including again Japan but also China, France, UK, Hong Kong, Singapore, South Korea.
It's not comprehensive. More than 75% of the passenger lines are missing in Belgium. I imagine this is the same for many other countries, especially the ones with a lot of passenger service.
Everything in AI seems to be focused on LLMs lately. But in my opinion, powerful problem-specific models like this are even more interesting. The SOTA AI models used in weather forecasting are already outperforming the classic NWP models while being orders of magnitude more efficient (inference). Most are based on multi scale (hierarchical) Graph Neural Networks, an architecture which is not often talked about. The original Graphcast paper is worth a read if you think this is interesting: https://arxiv.org/abs/2212.12794
I can assure you that anyone who touches numerical simulations of any kind (physicists, engineers, chemists, biophysicts...etc) has tried their hand at ML based surrogate models in the last 5 years, so it's not like they aren't being tested. From my experience, they aren't very robust. Weather modeling is actually one of the very few areas where it seems to work half decently.
Why does it work for weather at all? Is there something that the mathematical models are over-simulating? Is weather easier to predict than we thought? Just curious what the intuition is to regarding the success of ML weather modelling...
There is just A LOT of data available- usually an order of magnitude more than in any other related problem.
And general weather forecasts are not that hard - we have semi useful forecasts for more than 50 years. It’s when you want to do something special: long range, nowcasting of convective storm, other extreme weather etc. that is hard. And even then it’s as much a problem of input data accuracy than the models themselves.
Traditional physics based weather models also rely heavily on physical parameterization for sub grid scale processes (think clouds, microphysics of rain sleet snow, etc) so even the deterministic physics models are learned approximations from data.
traditional physics-based weather models also rely heavily on humans looking at the output and the evaluation of the output to discard wacky runs. Let's not pretend that existing physical models of the atmosphere stay on the rails all the time.
When you say runs that is an ensemble model forecast which is different than a deterministic model forecast. Any type of weather model sensitive to initial input state errors will have outlier model runs.
I suspect it's the massively rich, detailed, accurate, multimodal data.
For example, you have satellite images, coupled with real on the ground measurements of windspeed, sunlight, air pressure and precipitation going back decades.
For what it's worth, that paper is a spiritual successor to Keisler (2022) which was the first published work that took this approach: https://arxiv.org/abs/2202.07575
You say this as if you don't need he MWP models to train the AI models? The accuracy of the AI Prediction depends entirely on the quality of the training dataset...
Pretty much all of the AI weather prediction models are trained on ECMWF ERA5, which is kinda like a numerical weather prediction model run to forecast at t=0. ERA5 is historical weather data, but it’s a “reanalysis” of it.
Indeed, you can imagine this as some sort of advanced physics-based interpolation of various measurements (land stations, satellite data, ...) to fill in every cell in a latitude-longitude grid. This is not only used for ERA5 (training data for the models), but also to determine the initial conditions for every grid cell which are used to roll out the forecast. So AI weather models depend greatly on the NWP/physics used in for reanalysis and initial conditions. That being said, there is also research being conducted in training models straight from the raw data (weather stations, satellite, ...), thus bypassing the "interpolation" step.
Yeah. I can’t remember names off the top of my head, but there are a few companies, and I think many researchers, working on AI “data assimilation” for this.
ECMWF has an experimental AIFS direct observational prediction model (AIFS-DOP) that has become competitive with their physics based IFS model on certain metrics just in the past year.
I'm interested in understanding wheater prediction models because accurate wind forecasts make a big difference to my personal life (sports).
Is there a good overview to learn about the current models, which all just seem like cryptic acronyms to me? in apps like Windy etc.
WRF, TRRM, IK-HRRR-3km, ECMWF-9km,...
I understand by now that small grid cells are better for local prediction and that thermic winds are mostly missing from them all.
Ask your favorite AI to give you a crash course, but to start the main models you need to know are the GFS and the ECMWF. In the US where available in high res, the HRRR is excellent, but doesn’t forecast very far out. The PWG/PWE 1km PredictWind models are also very good at picking up land based features and other more precise patterns. If you are in the US everything else is probably not super relevant.
Any advice will really depend on where you live and the dominant source of local error. These are all good options for the US and if the one main source of error is near surface winds around complex terrain then you can also look into WindNinja from the National Forrest Service.
Historical weather data is discrete. You need continuous state for weather modelling which is currently achieved through conventional reforecasts using those historical observations.
A more interesting question is...does differential equations based models like mamba/state space models perform better on this sort of physics problem than pure transformer LLMs?
Insightful paper, thanks for sharing. Two things stand out to me.
First, it reinforces that you want methods that get better with more data. It emphasizes that the current approach cannot improve based on historic data - that’s the opportunity that ML based approaches exploit.
Second, it highlights that mature legacy solutions are tough competitors. They benefit from extensive tuning and real world feedback. Even when you have a genuinely better approach, it will take meaningful time & effort to achieve the current standard.
Patience and solid long term strategy are needed to make progress in these situations. You need confidence that your approach will win long term, backed by enough money & time to prove yourself correct.
LLM’s are an evolutionary deadend with the power and resource demands to make and run them for diminishing returns, but that may be okay as their own reasoning capabilities are big enough, and spur investment into the supporting infrastructure for them
everything in AI is not focused on LLM, if you think so then that's because you are in LLM bubble. The big idea with LLM is that it's generative AI, the generative could be anything! Not just large languages, we have seen break through in image generation, video, audio, but guess what. Anything that you have enough data and given data you can predict what comes next can have gen AI applied, so we are seeing it with physical actions so robots get trained to generate the next move, and I think the same thing applies to weather forecast. It's predictable too given enough data
One of my professors is referenced in the Wikipedia page of graph neural networks. It's funny that he explained them in the worse way possible and I eventually understood them better with another professor
Everything in the western world isn't focused on LLM. The top western players are heavily focused on AGI.
Meanwhile the Chinese are using LLMs and other non-AGI AI tech at the edge wherever they think to put it for task-specific productivity or optimization. They don't really care about AGI, or more accurately: they're working on getting their society more efficient and decarbonized, and then they'll be free to work on AGI with far fewer resources.
OpenAI, Anthropic, et al are working toward someday having AGI, and if they ever do, when they do, the Chinese will be hopelessly far ahead of us on energy, manufacturing, logistics (especially low/zero carbon transport of goods and people) and so on.
Once the Chinese figure out how to train an AI for ULEV lithography, especially once they figure out how to train it for semiconductor design or validation - it's game over for the semiconductor industry, and the big AI players will follow, because they won't possibly be able to compete against a Chinese version of NVIDIA with TSMC-like capabilities, or Chinese AI companies running on those much cheaper chips, with cheap, zero carbon power.
Well if you count every minor service outage which maybe 0.1% of the users are non-critically affected by, you quickly get to 0.6%. So, this doesn't really tell you anything.
New book by Eric Riess (writer of the Lean Startup) came out today. Haven't had the chance to read it yet, but certainly sounds interesting.
Excerpt from the Amazon page:
"In Incorruptible, Ries reveals the hidden forces that cause even great organizations to drift from their values and shows how to design businesses that can withstand that pressure. Drawing on two decades of work with founders, CEOs, and investors around the world, offers the blueprint for “mission-locked” organizations that can grow, prosper, and endure without losing their soul."
Very interesting, but sounds like an extra complication to use binary search on time to find the right position of the sun, when you can just directly calculate this, eg. using the formulas in here: https://gml.noaa.gov/grad/solcalc/solareqns.PDF
I've worked a lot with weather data in the past (and I still am), and I have to appreciate all the work that went into this. Weather model data is notoriously messy with many different formats and standards, and then I'm not even talking about radar data, etc. Probably when you've got this all abstracted away behind an API it is easier to build such a powerful application as this.
Depends on which model. Only really the ECMWF weather model is not fully free. The German, French, Dutch, ... models are all free (regional and global models). Of course, these global models are generally less accurate than ECMWF, still ECMWF has a lot of free data available too. US models are also freely available, and quite easy to work with (as opposed to some European ones).
TLDR: the input data for the model now also includes real-time observations (satellite and weather stations) on top of the typical (re)analysis data, improving model resolution, run frequency and timestep frequency.
reply