I make a platform that automatically collects the atmospheric pressure sensor reading on a schedule, and sends it to scientists for analysis. The idea is to build the best weather forecast (higher accuracy, hyperlocal) by using this dramatically denser data source as inputs to models.
Do you use other data to correlate locations that might have different offsets at different times? (EG; slightly pressurized buildings with forced air circulation or vehicles in motion?)
There's actually a wide variety of noise in the data, but the one that really matters is altitude (the barometer can measure the altitude difference between your feet and your head)! Other sources of error are the biases in the sensors themselves (place two phones side by side, get different readings), pressure-controlled buildings, sensor drift, etc.
We have a number of mechanisms to filter out most of the noise, but it's not easy.
Intriguing! How frequently does the SDK sample the barometer's readings? Does having the phone in one's pocket affect the perceived atmospheric pressure?
Our SDK default is 10-minute intervals, but some apps choose 1-hour to better fit normal schedules. The pocket problem isn't a big deal, but altitude changes do add noise to our system. The barometers are very elevation-sensitive (see other poster, commenting about flying!), which means that the reading will be different every time you move your phone.
There are a number of ways to solve this noise problem - first, merely detecting trends in the data over 6 hours is quite easy and tells us a lot about the atmosphere (is the pressure rising or falling? easy to find out from the noise). Second, getting enough density means we can use statistics to filter out most of the noisy data.
