A couple of years ago, our ginger CTO (yes, that’s how he likes to introduce himself!) was having a look at battery temperature data collected by OpenSignal users when he was struck by an unexpected discovery…insert suspense music here. What James realised was that air temperature could be retrieved from smartphones’ battery temperature by means of a simple mathematical transformation. The idea was further developed in collaboration with a team of researchers including Aart Overeem from Wageningen University and the Royal Netherlands Meteorological Institute (KNMI), and Berthold Horn from MIT. Their findings can be read in the Geophysical Research Letters.
Two years on, Jan Jaap Pape, one of Aart’s students, has picked up the torch and produced an awesome bachelor’s thesis, where he put the heat transformation model to the test. For this he studied battery temperature readings from OpenSignal users in São Paulo and Buenos Aires. Although his research hasn’t yet been published, Jan Jaap has kindly allowed us to share with our readers his promising results. So here they are, in a nutshell:
- For a city with the climatic characteristics of São Paulo or Buenos Aires, it takes around 250 battery temperature readings to get an accurate estimate of daily-averaged temperatures.
- The model was also tried for the retrieval of hourly-averaged temperatures. Although these estimates are not as reliable as the daily ones, Jan Jaap notes that a “signal” or correspondence with the weather station data “is nonetheless present”.
- A first attempt was made to detect spatial variability for daily-averaged temperatures, which would allow to address phenomena such as the urban heat island effect (UHI). It was observed that as the city of São Paulo was further divided in smaller parts, the more the correlation between daily-averaged temperatures from batteries and weather station data decreased. However, more weather station data would be needed to reach reliable conclusions.
- Jan Jaap tried as well new ways of calibrating and validating the smartphone data. In fact, for daily-averaged temperature estimates in São Paulo, better results were obtained when calibration was done using data for the entire year of 2013 and validation using data for the whole of 2014 (as opposed to taking the odd days for calibration and the even ones for validation).
- Last but not least, Jan Jaap was able to determine that estimating air temperatures using one specific type of smartphone model can be significantly more accurate compared to using data from all smartphone models. The graph below shows that the smartphone temperature curve is closely correlated to that of weather station datas when a unique series of smartphones is used (in this particular case, the Samsung GTI-series).
As the team working to bring OpenSignal to your smartphone or tablet, we glow with pride when we read a paper or thesis like Jan Jaap’s. But really, it’s you as our user who should give yourself a hearty pat on the back. Thanks to you, our crowdsourced database on wireless networks grows every day: your contribution has helped improve OpenSignal and made possible WeatherSignal, CrisisSignal and WiFiMapper; and it is thanks to you that we can collaborate with researchers such as Aart, Berthold and Jan Jaap. So here here to you!