I can’t imagine why this project exists. We have thousands of weather stations across the world already. Disentangling the temperature of your pocket from the actual temperature seems like an exercise in futility to me. Even the authors claim they can only get within 2.7 degrees Fahrenheit, so what is the point of having this inaccurate data?
From the American Geophysical Union
Crowdsourcing weather using smartphone batteries
WASHINGTON, DC — Smartphones are a great way to check in on the latest weather predictions, but new research aims to use the batteries in those same smartphones to predict the weather.
A group of smartphone app developers and weather experts discovered a way to use the temperature sensors built into smartphone batteries to crowdsource weather information. These tiny thermometers usually prevent smartphones from dangerously overheating, but the researchers discovered the battery temperatures tell a story about the environment around them.
Crowdsourcing hundreds of thousands of smartphone temperature readings from phones running the popular OpenSignal Android app, the team estimated daily average temperatures for eight major cities around the world. After calibration, the team calculated air temperatures within an average of 1.5 degrees Celsius (2.7 degrees Fahrenheit) of the actual value, which should improve as more users join the system.
While each of the cities already has established weather stations, according to the new method’s creators it could one day make predictions possible at a much finer scale of time and space than is currently feasible. Whereas today, weather reports typically provide one temperature for an entire city and a handful of readings expected throughout a day, the technique could lead to continuously updated weather predictions at a city block resolution.
“The ultimate end is to be able to do things we’ve never been able to do before in meteorology and give those really short-term and localized predictions,” said James Robinson, co-founder of London-based app developer OpenSignal that discovered the method. “In London you can go from bright and sunny to cloudy in just a matter of minutes. We’d hope someone would be able to decide when to leave their office to get the best weather for their lunch break.”
The work was published today in Geophysical Research Letters, a journal of the American Geophysical Union.
Smartphone sensors
Robinson’s OpenSignal app collects information voluntarily sent from users’ phones to build accurate maps of cellphone coverage and Wi-Fi access points. The app boasts about 700,000 active users according to Robinson, about 90 percent of which opt in to providing statistics collected by their phones.
Robinson originally wondered whether smartphones running on newer, 4G networks ran hotter than those running on older networks. When no difference showed up, he looked for other potential uses of the temperature information available on Android-powered devices.
“Just sort of for fun we started looking to see if there was a correlation with anything else,” said Robinson. “We got some London weather data for comparison and found the two sets of temperatures were offset, but they had the same sort of shape.”
While OpenSignal is available to iPhone and iPad users, the temperature readings on those devices are not accessible like on their Android counterparts.
Cellphone thermometers
After finding the correlation between smartphone and air temperatures in London, Robinson and his fellow developers assembled temperature data from other major cities where they had a large number of users. Comparing data from Los Angeles, Paris, Mexico City, Moscow, Rome, San Paulo and Buenos Aires, Argentina, they saw the same link between the two sets of temperatures they saw in London.
“It was amazing how easily the correlation sort of popped out,” said Robinson. “We didn’t do any handpicking of data—it sort of just emerged.”
A smartphone’s environment affects its temperature, according to Robinson. On a sweltering day, a cell phone tucked in a pocket will be hotter than the same cell phone on an icy day. Weather experts helped Robinson develop a way to calculate outdoor temperatures from smartphone battery temperatures, the latter of which are typically hotter.
However, other factors unrelated to the outdoor weather can play a role. A phone outdoors running the latest 3-D game could run at 46 degrees Celsius (115 degrees Fahrenheit) while the same phone idling in an air-conditioned building nearby could be only running at 27 degrees Celsius (80 degrees Fahrenheit).
To avoid fluctuations in temperature unrelated to the real outdoor temperature, Robinson needed large amounts of data. While an individual phone might not provide an accurate representation of the weather, combining the readings from hundreds or thousands of phones together gives a more truthful overall picture. Currently Robinson collects over half a million temperature readings each day from users of his OpenWeather app. He said he plans to make the data freely available to academic researchers.
“There’s the wider promise when logging all this information that there will be something really interesting you can understand,” said Robinson. “The most obvious application is climate and weather tracking.”
Personal weather predictions
Currently weather tracking primarily takes place at weather stations, such as those at airports. However, weather stations provide only one point of reference and are rare outside of densely populated areas, forcing weather forecasters to fill in the gaps when making their predictions, reducing both accuracy and how specific an area they can make predictions for.
While Robinson says his multitude of mobile phones can provide large amounts of data, individual areas still need to be fine-tuned using existing weather stations before the incoming information can be usable for weather prediction.
“The challenge is whether we can take this technique and use it in places where we don’t already have reliable weather information to retune the model,” said Robinson. “That’s something we’re still working on.”
Robinson says some recent smartphones come with built-in sensors specifically built to measure the environment around them such as air temperature, humidity and pressure. To take advantage of these features, Robinson and his fellow developers released WeatherSignal, an app built around mobile weather watching.
As these features become commonplace in the smartphone market, Robinson foresees smartphones becoming an important tool in weather monitoring.
Notes for Journalists
Journalists and public information officers (PIOs) of educational and scientific institutions who have registered with AGU can download a PDF copy of this early view article by clicking on this link: http://onlinelibrary.wiley.com/doi/10.1002/grl.50786/abstract
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
“Even the authors claim they can only get within 2.7 degrees Fahrenheit, so what is the point of having this inaccurate data?”
Anthony, after years of writing this blog and dealing with the warmers, I’m surprised you still think they actually care about accuracy.
Gail Combs says:
August 13, 2013 at 1:56 pm
RE: “sorry this comment cannot be posted”.
Try hitting the refresh button (It worked for me)
Thanks for the suggestion Gail, I originally thought it was a browser problem a date/time issue, but after looking into the problem I think it’s due to family visiting for the summer, which means nephews and nieces with their wireless Xboxes, Ipads/ipods, smart-phones and laptops fighting over the signal and my router is disconnecting idle connections such as mine.
I basically wrote a comment explaining that I tried this with a program I wrote (about 10 years ago) that monitored the temperature of processor cores and logged the data which I could access remotely, it is possible to work out a fairly accurate environment temperature as long as it is calibrated with an actual temperature of the environment the computer was in, it looks like they have taken a decades old impractical idea and ran with it on a large scale.
I looked at every comment, and of course the image that will stay in my head all day is the phrase, “except when stuffed into a thong at the beach.” Gak!
1) Glad that everyone understood the error-reduction effect inherent in multiple data readings. This is done all the time in industry.
2) Sorry that one poor soul didn’t pick up the point of the whole idea, which is to have vastly more than “one” reading for a given geographic region (like millions in a large city).
3) Sorry that other poor souls didn’t take the time to read the fact that the experimenters are proceeding with this because they discovered a good statistical correlation between the temperatures they obtain with this telemetry and the official temperatures. So, it’s not a matter of theory, it’s a matter of empirical evidence.
4) All the carping about sensor placement and circumstances fails to consider the most general likelihood that all these effects will largely cancel out (random variances). Anthony’s most perfect contribution to this debate (in my opinion) is his “outing” of the imperfection of the very temperature data that everyone asserts is God’s Truth. Get over it. Either this approach will be useful, or it will not. Sometimes a grade-school ruler is as handy as, or handier than, a millimeter scale. Particularly if it gives you the ability to perform in-depth regional contouring and time-variance studies. (Hint: this is not model-based science. It is MEASUREMENT.)
Had a chance to scan more of the dialogue. Apparently, there are those out there who do not understand how error is reduced by sample size. Sorry, the fact that they do not understand means only that they don’t understand a point that comes up early in any class on industrial statistical control. There is a very humorous exercise that is sometimes done to illustrate this point. An arbitrary length is drawn on the blackboard and each member of the class is asked to estimate what it is. All the estimates are taken together, the mean is taken, and—mirable dictu—it is surprsingly close to the actual length of the line (like within a few percentage points), far more accurate than any individual guess.
As for the regional effects of Big Events, like floods, earthquakes, tornadoes, intense hailstorms…I guess conventional thermometers would be impervious to these, right?
Alas, what mostly seems to shine through is (1) ignorance of the methodology, and (2) prejudice about its utility based on the assumed ideological mendacity of the experimenters. Pretty disgusting, folks.
MJ Dunn, did I actually just see you defend this idea, and even do some name calling toward the detractors?
So explain to me how crowd-sourcing the average temperature of where our cell phones are, which is primarily in climate controlled conditions or our pockets, can possibly be useful. Because even though I’ve tried, I can’t even imagine how that could be a useful metric.
Maybe it’s because I live in a winter climate, where it can reach -40 during winter and +40C in summer, but the advantage of measuring indoor spaces is lost on me.
Michael J. Dunn says:
August 14, 2013 at 12:36 pm
Your comment makes no sense: To use YOUR example: If my task was to measure the size of the parking lot outside the building, what does estimating the size of a line on the blackboard by having a class make guesses mean?
Now, if you said everybody WALKED OUTSIDE and took measurements independently of the same parking lot, and THEN estimated the length and width of the parking lot, you’d have a point. Not a very good point, but at least you’d have a point. But, what good does measuring the temperature of a person’s pocket indoors in a city have to do with the average AIR temperature in the country around that city?
kadaka (KD Knoebel) says:
August 13, 2013 at 11:35 pm …
I’ve used several generations of phones with lithium ion batteries. As the phones age, battery life shortens. While charging, phones are warmer than when not charging. I mean seriously, stop me when I am wrong here… If battery life goes down, and charging time goes up, then this will induce a bias in the phone’s temperature reading. And we’re not talking just a few degrees. Every cellphone I’ve ever owned was significantly warmer while it was charging. More time charging = more warmth as the phones get older. It’s really all battery technology though… not just lithium…
The batteries emit heat when either when they charge or discharge. And it can be a lot of heat. I seriously don’t think it is possible to correct for this type of bias.
How about this:
Take a brand new phone. How long does the average phone not get some software installed by the end user? Well, according to this website:
http://www.slashgear.com/angry-birds-reaches-one-billion-downloads-09227363/
One billion copies of Angry birds were downloaded between December 2009 and May 2012. Running angry birds increases power usage on your phone. I would posit that running Angry Birds on your phone will both cause the phone’s temperature to be higher after its installed, and also cause it to need to be charged more frequently. Since your phone probably doesn’t come with Angry birds, the day you buy the phone, its going to be cooler than the day after you install Angry Birds.
Who cares about Angry birds?!? Well, of course, I really dont – but there is almost no way to know how much any given application that is added to a phone will bias its reading towards heating. There are at least 10’s of thousands of applications available for smartphones. Who could possibly ever tell how much each app was influencing power usage on a given phone. You don’t think that running an app on a phone has no impact, do you?!? If nothing else, just the user holding onto the phone (vs. having it whereever a user normally stores a phone) will significantly affect the temperature.
In my experience, the radios in the phones also consume more power when they have weaker signal. I suppose as cell phone companies add towers, this would induce an apparent cooling trend… assuming of course that cell phones don’t play the cellphone version of the “cocktail conversation” game where in a crowded room, the output level increases to try to overtalk nearby cellphones…
It’s ironic. Anthony’s site was started to document issues with weather station siting (sites that show warming from other sources due to location issues). Only a complete brain-dead idiot (or a complete charlatan) would even remotely think that putting temperature sensors in people’s pockets would produce anythng but complete and total nonsense. Yes, lets get more and more sites with extremely bad siting… that will fix the problem. Clearly no effort should be put forth on this fool’s errand until problems with the “real sites” have been addressed.
I’ve had small poorly-written applications on my cell phone that caused the processor to run at unexpected times. This resulted in a warm-to-the-touch cell phone and short battery life.
Let me guess what this study will demonstrate.
“Apparently, there are those out there who do not understand how error is reduced by sample size.” [Dunn the App Salesman]
Apparently, there is a Mr. Dunn here who does not understand that:
“Garbage x 1,000,000 = garbage.” [Jorge Kafkazar 8/13 9:20PM]
It would seem to me that the law of large numbers needed to distinguish between transient effects (the outdoor game-player vs the indoor idler) will completely defeat the stated goal of greater resolution.
I am also skeptical that they can ever compensate for human factors such as the fact that we generally come indoors when it’s raining, put on coats when it’s cold (usually covering the phone), spend less time outdoors when it’s really cold – or really hot, set the thermostat lower at night and so on. Too many situations leading to too many different types of systemic bias.
I suppose averaging the readings between [device in arse-pocket in sweaty elevator] against [device in pannier of bike next to pre-frozen bottle of water] might approximate to [ghastly ‘climate’ ( aka ongoing efforts of planet to destroy all life, via “weather” in Scotland, where I live)]. I so looked forward to “global warming”. Seems it’s not to be.
Mr. Dunn wrote;
“Had a chance to scan more of the dialogue. Apparently, there are those out there who do not understand how error is reduced by sample size.”
With all due respect, error is NOT reduced by sample size. NOISE is reduced by sample size. When multiple samples from a single sensor are averaged together the noise in the averaged sample is less (subject to numerous caveats like the probability distribution shape and stationary). The accuracy (difference between the readings and reality) is not increased.
If I take a temperature sample (measurement) with a thermometer that is wrong by 1 degree, or I take a bazillion readings and average them together the result is still wrong by 1 degree.
Your “line on a blackboard” is not an example of increasing accuracy with multiple samples, Rather, it is an example of many multiple estimates being distributed in a gaussian distribution function with the mean approaching the “consensus” length. Assuming the observers are “pretty good” at estimating the length the answer will converge to something close to reality. If the observers are all near sighted the answer will be wrong.
In control systems the use of averaging to reduce noise is well known to reduce the “update frequency” or bandwidth. For example, if I’m extruding pasta and want the length to be 250 mm +/- 1mm I could
1) make an accurate measuring device to tell me when to cut the pasta to length
or
2) make an inaccurate measuring device and average a million readings and then cut the pasta to length
Option 1 will give me lots of pasta that is within my specifications
Option 2 will give me a little bit of pasta that is all exactly the same wrong size
I suggest that before you throw the “ignorance flag” and express your “disgust”, you might want to crack a textbook or two about measurement accuracy and signal-to-noise ratio. These are two totally different concepts.
You are totally welcome for the free education.
Cheers, Kevin
CodeTech
It’s not merely an idea, it’s a technique; they have actually tried it out. If calling someone a “poor soul” is “namecalling,” then you are welcome to it, but in my view of life, the truly ignorant are indeed poor souls.
As for how it can be useful? I quote: “…the team estimated daily average temperatures for eight major cities around the world. After calibration, the team calculated air temperatures within an average of 1.5 degrees Celsius (2.7 degrees Fahrenheit) of the actual value…it could one day make predictions possible at a much finer scale of time and space than is currently feasible. Whereas today, weather reports typically provide one temperature for an entire city and a handful of readings expected throughout a day, the technique could lead to continuously updated weather predictions at a city block resolution.” It helps to read before scoffing.
As for indoor-outdoor readings, one of their problems will be to establish a seasonal calibration. Note that calibration is one of their early concerns. It may prove to be a limitation on the utility of the technique.
RAookPE1978
If I wanted to measure the size of the parking lot with this technique, I would indeed take the class outside and ask for guesses as to the length of its sides. I’m talking about lines on the blackboard, you want to talk about parking lots. Is that all this comes to?
As to the correlation with air temperature, refer back to the quotation above. It all has to do with calibration against STANDARDIZED MEASUREMENTS.
Janice Moore
Talk to Mr. K (below), or read a text on statistical methods.
KevinK
Thank you that someone else out there knows what I am talking about. We need to correlate our terminology in order to avoid talking past each other. I am used to the error of a distribution function being separated into the bias error (the mean or systematic error, or accuracy) and the random error (standard deviation, or precision). All my comments have pertained to the random error, if anyone will go back and see I stipulated that large ensembles will do nothing to reduce systematic or bias error. I think we are on the same page.
And what if there is a 1-degree bias error? From a data collection standpoint, I say “So what?” Are you trying to tell me that anything we can deduce from temperature data is so infamously sensitive that a 1-degree accuracy error will upset the applecart? That’s exactly the conceit that the Warmists invoke with all their incredible projections of climate sensitivity. In the course of a year, when temperatures swing through maybe 50-60 Fahrenheit degrees, a few degrees error is going to make a difference? The implication of having a chaotic feedback system is that we can’t predict when things will change anyway, and until they change, they will stay pretty much the same. Therefore, the error is probably insignificant. Meanwhile, this is a wonderful asset to perform spatial and temporal studies.
The blackboard example is a commonplace, and it is used not only to show the effect of ensembles on random error reduction, but also to show that the mean error is surprisingly small (we are not so feeble a measuring instrument as we are claimed to be).
The pasta example does not prove your point, which only means your point is wrong or misunderstood. If I made a moderately accurate sensor (no one bothers to make an inaccurate sensor), probably the size of a pinhead, and put a thousand of them in a row, and sampled them a thousand times a second for each measurement, I would probably obtain “lots of pasta that is within my specifications.” It is only statistical process control. If there is a bias error, I can find that out with the first cut and fill in a calibration correction for every succeeding cut—just as I would have to do with the “more accurate” sensor. The bias is the part that does not move around with time. If it did, it would merely be part of the random error.
Since I do this for a living (sensor architecture error fusion), I don’t understand how I have learned anything I didn’t already know correctly.
The Lesson
As wonderful as this site is, you, the denizens, are not always very educated in the science you purport to defend. Read more. Opine less. Learn.
And as much as it is very clear that the Warmists are mendacious to a fault, do not fill in your own ignorance with prejudice. There is no point to it, and you just cheapen yourselves as human beings. Not to mention sully Anthony’s reputation. Is this how you reward him, by behaving like rabble? You can do better. Of course, many if not most are gentlemen and ladies, but why not all?
Peace be with you and a warm hearth in the coming ice age!
MJ Dunn:
I wrote before:
Although you danced around it, you have failed to make a case.
We’ve seen quite enough of “calibrations” over the last few decades. Heck, they “calibrated” the 1930s right out of the record.
There is no possible way to make this metric useful in any way. None. Apparently you think otherwise, but that just makes you wrong. You’ll still be measuring the climate-controlled areas, not outside. Measuring indoors and in pockets and in cars is hardly useful no matter what “calibration” technique is used. As an “index”, it might have some value. As a measurement, it has exactly zero value.
It helps to have a clue before attempting to educate people who know more than you do.
I share scepticism about this – apart from anything else, most smartphones will be inside buildings, let alone pockets, during most of the day.
However, as a general point, the resolution of weather forecast models is continually increasing, and meteorologists are struggling to find sufficiently well-resolved observational data to initialise them with. If your surface weather stations are hundreds of kilometres apart that isn’t much help when your weather model has a grid spacing of 300m.
There has been some good work on using GPS satellite signals to retrieve atmospheric data, and I believe there is ongoing work on extracting humidity data from mobile phone base station data.
It might be that if you can collect enough data from enough smartphones that you can extract a useful signal from it.
I want to clarify one thing before this topic drops out the bottom:
MJ Dunn is arguing a different thing than everyone else is. Yes, having lots of measurements can increase the accuracy. The whole topic, and even its title, isn’t about the benefits of lots of sensors. The point is that what these sensors are measuring is not a useful thing to measure.
It’s equivalent to building a climate monitoring network and putting the sensors inside climate controlled housings to ensure their safety and longevity…
CodeTech
I quote again: “…the team estimated daily average temperatures for eight major cities around the world. After calibration, the team calculated air temperatures within an average of 1.5 degrees Celsius (2.7 degrees Fahrenheit) of the actual value.” It’s not modeling, it is measurement by a new technique, and they came close to the measurements of accepted sensors. The proof of the pudding is in the eating. They told Robert Goddard that rockets wouldn’t work in vacuum, either. I guess you don’t like thermometers that are accurate to only a few degrees. Can you explain to anyone why more accuracy is essential to the purpose pursued by these experimenters, and why anything less must be disdained, scorned, and cursed? Your laboratory must be an exciting place to work! Take it easy, relax, and see how it turns out. If it turns out well, you have an interesting crow to swallow. If it doesn’t, that’s only what I am waiting to find out.
Michael J. Dunn says:
August 15, 2013 at 4:29 pm
After calibration, the team calculated air temperatures within an average of 1.5 degrees Celsius (2.7 degrees Fahrenheit) of the actual value.”
And how many measurements did they have to make to get the actual value?
The problem here is that although these measurements may not be intended to demonstrate man’s influence on the climate, they will in fact be used to demonstrate man’s influence on the climate (at some point) – just like all the previous measurements have been used to do – including cases where previous measurements have been adjusted to amplify recent warming.
I have been arguing that with my experience with cell phones, I would expect a warming *trend* to come out of these measurements. Of course, the trend could be filtered out (there could be software on the phones to report back just how long the battery lasts between charges, and to somehow comphensate). At some point though, besides being tremendously invasive (the data about where you at all times would be in a database somewhere). I honestly wouldn’t trust a warmista to accurately filter out any data that didn’t support their idealogical pet.
Finding a way to be accurate to within 5.5 degrees doens’t mean they are filtering out spurious trends, and for most of the uses of this data, I would think trend would be the most important information, wouldn’t you?
I have found that it is impossible to get within two degrees even using the same brand and type of weather station. Nest door moving a pot plant caused a greater change than the claimed accuracy of the weather data used to promote climate change. I note that the manufacturer of a paint which allowed the wood to breath like the old fashioned paints used in the screens made nearly a degree temperature difference if the previous day had been raining.