Outreach via cancellation
-
The weather service Dark Sky has been acquired by Apple:
(the blog has been trimmed to just that one post)They say
We’re thrilled to have the opportunity to reach far more people, with far more impact, than we ever could alone.
How are they reaching more people? The
GoogleApple Way™!Android and Wear OS App
The app will no longer be available for download.
Weather forecasts, maps, and embeds will continue until July 1, 2020.
The API will continue to function through the end of 2021.
And the help centre reiterates, in less friendly terms:
The Android and Wear OS apps are no longer available for download and will be shutdown on July 1, 2020.
Maps, Forecast, Sports and existing Embeds will be shutdown on July 1, 2020.
“reach far more people” indeed. Time to find a new website, apparently.
-
@kazitor said in Outreach via cancellation:
“reach far more people” indeed.
The obvious solution is to buy .
-
This annoyed me, as I rather liked Dark Sky. I wonder if their website version will continue to work or if it'll become app-only for iDevices.
-
@e4tmyl33t said in Outreach via cancellation:
their website version will continue to work
Says the website, it says:
Weather forecasts, maps, and embeds will continue until July 1, 2020.
Not very long.
-
@Tsaukpaetra said in Outreach via cancellation:
@e4tmyl33t said in Outreach via cancellation:
their website version will continue to work
Says the website, it says:
Weather forecasts, maps, and embeds will continue until July 1, 2020.
Not very long.
I figured that was for people who already had the Android app
-
@e4tmyl33t said in Outreach via cancellation:
I wonder if their website version will continue to work or if it'll become app-only for iDevices.
It appears that they are going iOS App only:
There will be no changes to Dark Sky for iOS at this time. It will continue to be available for purchase in the App Store.
-
@e4tmyl33t said in Outreach via cancellation:
@Tsaukpaetra said in Outreach via cancellation:
@e4tmyl33t said in Outreach via cancellation:
their website version will continue to work
Says the website, it says:
Weather forecasts, maps, and embeds will continue until July 1, 2020.
Not very long.
I figured that was for people who already had the Android app
Considering there's tabs for both Android and Website (and the text changes for both), I'm inclined to believe
They're separate but have the same date of cancellation.
-
I would be cross about this… except I don't actually care as it is a service that I don't use at all.
-
@dkf I've used it and will miss it a bit, but I'm sure someone else will make a similar near-term weather forecast app/site with radar and whatever else
-
@El_Heffe said in Outreach via cancellation:
It appears that they are going iOS App only:
There will be no changes to Dark Sky for iOS at this time. It will continue to be available for purchase in the App Store.
I'm left wondering how long "at this time" will last, and what will happen when it ends.
-
@kazitor
sounds like they need to change their blog title...Disappointing. Oh well.
-
I used the android app for a while. It always seemed pretty egregiously inaccurate compared to almost any other prediction
-
@Jaloopa I used the website, and while its definition of "light rain" was usually a bit different from the actual rain that was coming down, I found it to be quite good
-
@Jaloopa said in Outreach via cancellation:
It always seemed pretty egregiously inaccurate
It's a weather forecast; "egregiously inaccurate" is bullet point #1 of the product definition.
-
@HardwareGeek said in Outreach via cancellation:
It's a weather forecast; "egregiously inaccurate" is bullet point #1 of the product definition.
I do wonder about that (I mean, not the joke, that's pretty obvious).
I have two weather apps (websites really but who cares) that I use for my home location. They provide similar predictions (well that's reassuring!) but not quite, in particular I've noticed that one of the two tends to predict rain arriving earlier than the other does (one also tends to predict tiny amounts of rain when the other doesn't but that one is probably simply due to how they interpret the same raw data and translate it into pictograms, so that's not the issue here -- though that is the reason why I've got two apps, I like some aspects of how each of the two present (what should be) the same data and since I can't get both aspects on the same page, I keep the two...).
It's sometimes just off by 1 or 2 h, which again is likely just a slightly different way to present the same data, but often it's off by as much as 6 or even 12 h (for 2-3 days predictions), which seems a bit too much to be just a presentation difference.
Now to get different predictions of the same event on the same area, that must mean they are looking at two different models, right? But I thought that, at the fine-grained level of local locations, only the national weather office of that place was running very fine scale predictions, and that they were using coarser scales for areas outside of their country (because no weather office has the computing power to run a fine scale model of a much larger area, and also because if there is another weather office running that simulation in that other area, why would you waste computing power doing it as well when you can instead focus on getting finer results on your main country of interest?), and that end-user providers of weather just mesh together all those predictions.
But that doesn't seem to match my observations with those two apps. If there is only one fine scale model, then hours-by-hours predictions should both come out of this model and should therefore match, but they don't. So in the end, I have no idea how weather models are run and how general public websites translate them to what's shown.
-
@remi As I understand it, the national office doesn't produce one model. It produces a series of predictions which were run with slightly different input parameters. As such, the output is not "there is 10% chance of rain at 11:00". Rather, it is "5 out of 50 prediction runs show rain at 11:00". And over a longer interval you might get "5 out of 50 prediction runs show rain all day", or you might get "40 out of 50 prediction runs show an hour of day, but they vary wildly in the time at which they predict it".
The app I use offers both predictions per hour and predictions per 6 hours for the next 24 hours, and usually a all-day prediction that overlaps that range as well. These are often at first glance inconsistent with each other, even though they are definitely based on the same data.
I also use a separate app which does 2 hours of rain forecast purely based on weather radar data. That one is reasonably reliable in NL, but it's utterly useless in mountainous terrain.
-
@PleegWat
Additionally, they'll give you those predictions for some general regions e.g. North and South of the country, but those apps extrapolate that to exact small city in the middle of nowhere.
-
@PleegWat said in Outreach via cancellation:
As I understand it, the national office doesn't produce one model. It produces a series of predictions which were run with slightly different input parameters. As such, the output is not "there is 10% chance of rain at 11:00". Rather, it is "5 out of 50 prediction runs show rain at 11:00". And over a longer interval you might get "5 out of 50 prediction runs show rain all day", or you might get "40 out of 50 prediction runs show an hour of day, but they vary wildly in the time at which they predict it".
Ah, that makes sense. I was thinking they were delivering mostly the first one, i.e. probabilities of such and such weather (probably in some more refined form than just "rain", but probabilities nonetheless), but it actually makes a lot more sense to deliver different simulations. And I should kick myself for not having thought about that myself, because this is almost 100% the same thing than what I am doing professionally (in a different field), so I know very well why it is this way and why it is a much better approach than anything else and so on...
After that, I agree that it's just a matter of how individual websites interpret the data to produce a single output to the end user (which absolutely makes sense as interpreting multiple outputs, or even just probabilities, requires much more data interpretation knowledge than most people have!).
I assume it's not publicly-available data (or rather, not available for free!), and likely also not in an easily visualisable form (i.e. bunch of tables of numbers), but out of curiosity once I'd like to see what the actual raw outputs from met offices look like...
-
@Luhmann said in Outreach via cancellation:
@PleegWat
Additionally, they'll give you those predictions for some general regions e.g. North and South of the country, but those apps extrapolate that to exact small city in the middle of nowhere.Obviously there is no weather station, nor prediction grid point, for every single village in the middle of nowhere, but AFAIK the grid on which the simulation runs is fairly high-res, something like 5 km or so (that being just an average number, an intelligent model will likely use adaptive meshing, or maybe even a meshing based on some sort of parametric function to get a better resolution where needed, or maybe also several meshes for different aspects of the model -- usually some physical properties are much smoother than other and can be modelled on a coarser grid). So there isn't that much interpolation to do in most places.
(except of course some very specific areas such as mountains, where you at least need to correct by the altitude to get the rain/snow limit correctly, and also where changing the mountain side can dramatically change the weather in much less than 5 km)
-
@remi said in Outreach via cancellation:
I'd like to see what the actual raw outputs from met offices look like...
There are whole earth models that use a 3D (triangular? hexagonal?) grid and compute what the bulk air movements are, where air is rising and descending, what's going on with frontal systems, that sort of thing. Then there's a whole different set of models that run in much finer detail on a rectangular grid and which produce the detailed predictions, taking into account the details of local hills, the heat island effects of small settlements, etc. The data moved between them is… large (and describes a time series of the state of lots of variables at each grid point for each simulation). Even 15 years ago or so, we're talking about moving lots of gigabytes for a dataset, and I cannot imagine that being smaller now! I do remember that running those local simulations was itself a significant supercomputer time consumer at the time (which was what WTF-U was actually commercially interested in, as they ran a supercomputer centre at that point).
I've done programming projects on all sorts of scientific things over the years.
-
@dkf: So... basically, you were doing cloud computing years before everybody else?
...I'll get my coat...
-
-
@Zerosquare said in Outreach via cancellation:
So... basically, you were doing cloud computing years before everybody else?
Apart from the joke, yes, I was.
-
@dkf said in Outreach via cancellation:
The data moved between them is… large (and describes a time series of the state of lots of variables at each grid point for each simulation). Even 15 years ago or so, we're talking about moving lots of gigabytes for a dataset, and I cannot imagine that being smaller now!
Was that per simulation? Because I understand the practice now is to run an ensemble of half a dozen or so, varied by filling gaps between recorded data points with different plausible estimates drawn from the meteorologists' experience, who then compare the differing outcomes before making the "official" forecast, which could in turn differ from any or all of the generated ones in detail.
-
@Watson said in Outreach via cancellation:
Was that per simulation?
Don't know. This was at around the time when ensemble forecasting was becoming practical, but I was working on systems for planning where to put the simulations and how to move the data, and not on what the details of the simulations and data actually were.
-
The first of July came and went, with Dark Sky (now Apple, Inc.) wimping out and postponing:
Update: Service to existing users and subscribers of the Android app will now continue until August 1, 2020, at which point the app will be shut down. Subscribers who are active at that time will receive a full refund. Weather forecasts, maps, and embeds available on the Dark Sky website will also shut down on August 1, 2020.
Now the first of August has come and ended almost everywhere, with half the planned outreach taking effect:
Update: The Android and Wear OS apps shut down on August 1, 2020. Subscribers who were active at that time will be receiving a full refund.
But it seems they no longer intend to reach and impact as many more people as initially planned:
Weather forecasts and maps on the Dark Sky website remain active.
Looks like the website will function forever more? That’s cool. Though for some reason, I haven’t needed weather forecasts so much in recent months
I don’t have any before-and-after shots, but it also seems like the website is a touch less overtly “we are now owned by Apple!” now.
No news on the API. It still doesn’t accept new signups, so until otherwise specified I guess it will be reaching far more people once it’s axed after the end of 2021.
-
@kazitor said in Outreach via cancellation:
Though for some reason, I haven’t needed weather forecasts so much in recent months
LOL had the same thought the other day. I used to check out hourly weather multiple times a day - I even change User Agent in my phone's Firefox to iOS to stop Google from using absolutely illegal but somehow still allowed tricks that prevented me from having that straight on the results page. Some time ago, I lost all my extensions except uBlock Origin, but I don't care enough to reinstall them.
-
@kazitor said in Outreach via cancellation:
Though for some reason, I haven’t needed weather forecasts so much in recent months
Dunno about that. I still like to know if the weather is predicted to be unraisinably hot (seeing as how my appartement doesn't have aircon), so forecasts are good during the heights of summer. On Friday, Lille had a high of 39 °C.(0) For the refuseniks in the audience, that's 102 °F. Normal summer temperatures in this area are 15-20 degrees C lower than that. Conclusion: at the end of the day, I went down into town to eat somewhere air-conditioned.(1)
(0) That's OK. The toppest temperature recorded in Lille was this time last year, officially noted as 41,5 °C, but that's (as usual) somewhere "out of town", in the nearby aeroport. In town, it probably touched or exceeded 42 °C, which is 107 °F. Farblin' hot for somewhere that's north of 50°N latitude.
(1) For ... raisins,
runs on a time zone that's an hour east of where the country is, and that makes "mean solar noon" happen in Lille at about 13h48 (almost 2pm) in summer. The hottest part of a summer day tends to be around 15h00 to 16h00, all else being equal, which puts it just before 18h00 CEST in Lille. So on a punkin' hot day like that, the temperature is still way up high at dinner time.
-
@Steve_The_Cynic said in Outreach via cancellation:
(1) For ... raisins,
runs on a time zone that's an hour east of where the country isThe real reason is of course continental European trade and travel, but the
reason is that no way would want to run on the same time zone as 
. UTC being centered around Greenwich is bad enough as is.
-
@topspin said in Outreach via cancellation:
@Steve_The_Cynic said in Outreach via cancellation:
(1) For ... raisins,
runs on a time zone that's an hour east of where the country isThe real reason is of course continental European trade and travel, but the
reason is that no way would want to run on the same time zone as . UTC being centered around Greenwich is bad enough as is.History suggests otherwise. France started using CET in 1940, but not by choice. Prior to that, it used GMT (since 1911). Or so they say. https://en.wikipedia.org/wiki/Central_European_Time
It's worth noting that France didn't use CEST until 1976.
