I'm not sure what happened but Dark Sky went from awesome to awful in recent months after the Apple acquisition. Is it just me? Going from "no rain all day" to actively raining with a huge storm 2 hours later. The app does update with this info, but very close to the actual storm.

As an Android user, it used to be a fun party trick at the bar, to show everyone the alert on my phone from dark sky that it was going to start raining in X minutes or that it was going to stop raining in Y minutes and then everyone checking later and it being basically down to the second accurate.

As a former paid user of Dark Sky that lost that information when Apple bought them I feel no sympathy for iOS users that Dark Sky has gone to crap

I am sad such an amazing tool for the world was destroyed by a corporation trying to deprive another corporation so they could maybe make a little more money.

The integration of dark sky was definitely rocky but apple is running dark sky as a data api now, this android app is apparently using it as it’s weather data source now:

https://www.boondogglelabs.com/projects/forecaster.html

The Dark Sky API is getting shut down in a few months: https://blog.darksky.net/

I too have noticed this. I believe the Dark Sky website and app will continue to become less accurate... I think the explanation is (somebody who knows more please correct me): Dark Sky's technology was acquired by Apple and integrated into the Apple Weather built-in app... and the decision was made that any improvements/updates/maintenance on the Dark Sky service itself are accessible exclusively through that Apple Weather app now. As a user of the Dark Sky website, I feel sad.

Well, there is a red banner on the site saying

> Support for the Dark Sky iOS app will end on December 31st, 2022, and support for the Dark Sky API will end on March 31st, 2023.

So basically them buying then closing the service to make their own seem better

They made their own actually better if that matters to you. The iOS weather app now has many of DarkSky's features.

It's not just you, but this started well before the Apple acquisition. I "sold" many copies of the Dark Sky app to family and friends just by showing its magic but that magic died awhile ago. I don't think it's been very accurate for 3-4 years now.

This has been happening across nearly every weather source I've been using. My hypothesis is either:

* Climate change is causing problems with forecasting models

* Forecasting models changed, which has introduce unexpected edge cases

This study claims that there was a decrease in weather prediction accuracy caused by the reduction in commercial flights due to Covid. Maybe there are still some lingering effects from that?

https://www.sciencedaily.com/releases/2020/07/200717101026.h...

Wow! This is extremely interesting!

While this is confirmation bias, this reflects almost identically with the timeframe in which the forecasts became significantly worse in my area.

Could 5G have something to do with this? https://backreaction.blogspot.com/2022/09/the-trouble-with-5...

The former is going to be literally true on the grounds that climate change is increasing chaos in the system, hence unpredictability increases strikingly.

I'm not sure this happens on the time scale of 'app got acquired', though. It's absolutely going to increase year by year, and if you'd got used to how it was three years ago and didn't notice some of the outlier events, you might be shocked at the new normal.

Climate change absolutely will cause problems with any conceivable forecasting model because you can't encapsulate chaos, only map its expected potential outcomes: and the nasty truth is that energy is fed into the system, those potential outcomes blow up exponentially. Not even slightly linear. So that's what we're seeing.

I've noticed that the "rain starting in your area" alerts are close to correct - and spot-on for the local airport.

So I'd not be surprised if they somehow switched their data source to be not as granular as it once was.

> Going from "no rain all day" to actively raining with a huge storm 2 hours later.

Honestly, that's about right here in Florida. :)

Those watches measure the temperature at the underside of the device, i.e. the temperature of the wearer not the air. And that data they claim stays on your phone, not in their cloud.

They called out in the presentation that it's also got an ambient temperature sensor just underneath the display, to help calibrate body-temp vs room-temp for the overnight "difference from baseline" temperature measurements.

Agreed on the local-storage-for-health-data bit, but I'm sure they could argue that the ambient sensor isn't privacy-relevant.

The temperature inside a winter coat though, not outside in the freezing air right?

It’s also very easily anonymizable

They claimed that data is encrypted on device and Apple never sees it.

why would THEY pay to store YOUR biometric data when YOU could store it for THEM!

Aren’t those thermometers pointing the wrong way (towards the wrist)?

There are two, one inward and one outward facing. This was highlighted when talking about taking water temp readings when swimming or diving.

The new Apple Watches will actually have two thermometers, one facing inward and one facing outwards.

No, they have two thermometers, one for ambient temperature and one for body temp.

I would bet that the internal sensor is used in a model with the skin sensor to remove the effects of the internal temperature of the watch and get a more accurate skin temp reading.

Not so sure about this. There's going to be a ton of noise in this data from an outsider POV. At an absolute minimum, for this to be usable as weather data they will have to determine when it's measuring the actual outside weather, as opposed to the indoor, or in-car/-bus/-train temperature.

> At an absolute minimum, for this to be usable as weather data they will have to determine when it's measuring the actual outside weather,

Does Apple Watch have a light sensor for an auto dimming display? If so, that can be used as a proxy for indoor/outdoor. It isn't great, shade and such. Throw barometer in there, and you can do a decent job.

Microsoft Band had a UV sensor, so we were able to get pretty accurate detection of indoor/outdoor when combined with all the other sensors, at least during the day. I'm sad UV sensors didn't take off on smart watches in general, it is a super useful sensor to have, but they are rather large and even the lens material to go over them is different than what is otherwise required for a wearable.

I don't see this working effectively. A UV sensor is great during daytime, but what about at night?

Then, you still have to account for the fact that your thermometer is located right next to a variable ~100W heater, vs. normal weather stations that are normally located away from heat sources. Also, the thermometer and heater are occasionally placed inside a tight-fitting insulated container.

Good point! During times when AC or heat is running a lot, there might be a bimodal distribution of reported temperatures. I wonder if they could use the ambient light sensor to determine if the person is inside or outside, or if they could use location data to figure out that 'this Apple Watch is in the middle of a large field, so he must be outside'. Then they could just ignore other data in that microclimate area, and create a weather mapping based on watches located in known open areas.

> I wonder if they could use the ambient light sensor to determine if the person is inside or outside

Maybe they are they next to a window or under a sky light.

> or if they could use location data to figure out that 'this Apple Watch is in the middle of a large field, so he must be outside'.

Could be in a tent or other temporary structure.

Yep, there will always be exceptions. But taking the median of a decently-filtered set could work. And Apple obviously has a team of folks working on this to come up with something smarter than some random HNer (me) who's just shooting from the hip).

> And Apple obviously has a team of folks working on this to come up with something smarter than some random HNer (me) who's just shooting from the hip).

Don't talk like that, its a self defeating mindset. You are likely as competent as most Apple engineers (the 10x benefit from less inhibitions and tons of energy.

The people working at apple are like most of us here, curious and driven. Apple has a lot of money to throw at people who bring in experience whereas you and I are individuals who do not work in the field of "figuring out where things are using computers." So we obviously are not experts in the field and not familiar with existing and emerging solutions for the problem.

You said a thousand words worth with "bimodal".

Everything old is new again: PressureNET

https://en.wikipedia.org/wiki/PressureNET

The barometers combined with GPS and temp is an incredibly powerful sensor network.

Its a shame that it is locked up inside of apple.

It's not that much better than the normal network of AMS sensors. For one, it doesn't give you wind speeds at various levels (critical for predicting weather).

Second, it doesn't give you any additional readings over oceans, which is where the data is lacking in the current network of data.

It does give you more robust surface temperature readings in cloudy areas but that doesn't really help you predict the weather significantly better.

Source: degree in atmospheric sciences

Yeah, but maybe Apple will take all of this private data, and then build their own GCM, and solve data assimilation for their new model … yeah right.

Let’s pretend that data from user portable devices isn’t garbage (it is): then what do you even do with it? In cities you’ll have thousands of readings per voxel none of which are more valuable. Then you have to clean your data, excluding data from inside or vehicles or hot pavement or heated patios and so on. Apple won’t go to the expense of running their own GCM, and I doubt even mesoscale models. All I could see is some sort of correlative AI, doing microscale adjustments to government sponsored models. So all of this work, and maybe you get one degree better at forecasts and a bit of nowcasting. And because it isn’t physics-based, you might occasionally get something very wrong happening.

The oceans are a data hole (technical term, really) as you mentioned, but the real missing data is above the surface. It’s too bad about Project Loon, because that would have been much more helpful to forecasting than a million monkeys wearing watches.

I glide quite a bit, so I'm VERY interested in low level micro climate.

Is there a really no use in having higher resolution for modeling the boundary layer?

It seems like having a mountain covered in tiny barometric and temp sensors could at least validate a lot of what we think we know about the boundary layer.

I personally would love to be able to see what the temperature at the bottom and top of a given topographic feature is, as well as how atmospheric pressure is deviating from pure altitude differences.

I build instruments for gliders so I know that the sensors in our phones are capable of incredibly fine resolution. Is there really no use for billions of high resolution data points in climate science?

Wouldn't they only be able to make a detailed mapping of the temperatures of people's wrists?

> Wouldn't they only be able to make a detailed mapping of the temperatures of people's wrists?

Unless you're sick, most people's body temperature is within a fairly tight range.

So they could compensate for that ...

Is there any correlation between your body temperature and the outdoor temperature?

Depends on if you're a lizard or a bird :)

It would be hilarious: the temperature worldwide is roughly 98.6° F, except in NYC where the marathon is being run and it is somewhat warmer. /s

> except in NYC

Or, ahem.... "cold winter evenings". ;-)

And, well, are ouside at the time. Also need to consider location of the watch, easy to heat it just by sun (in summer) or by wearing long sleeves (when it's colder)

Air pressure on the other hand...

there is a lot of money [1] in harvesting location data from weather apps. apple wants that revenue [1]https://themarkup.org/privacy/2021/09/30/theres-a-multibilli...

Apple doesn’t need a weather app to obtain location data from users. If they were going to track and sell location data they would just do it invisibly through iOS itself. This (and other tracking opportunities) is the reason Google gives Android away for “free”.

Please tell me it is called "Genysis"!

thats not forecasting though is it? its measuring. and only with total coverage its predicting too?

How does the watch differentiate ambient temp from wearer's temp?

Calibration and ML. Like an infrared forehead thermometer.

Maybe, but the watch can be enclosed up a sleeve, in a car, in an office, house, direct sunlight, in the shade... Good if it takes all that into account.