Understanding the potential of human–machine crowdsourcing for weather data

Abstract

Reliable weather estimation traditionally requires a dense network of meteorological measurement stations. The concept of participatory sensing promises to alleviate this requirement by crowdsourcing weather data from an ideally very large set of participating users instead. Participation may involve nothing more than downloading a corresponding app to enable the collection of such data, given that modern smartphones contain a plethora of weather-related sensors. To understand the potential of participatory sensing for weather estimation, and how humans can be put in the loop to further improve such sensing, we created Atmos a crowdsourcing weather app that not only periodically samples smartphones sensors for weather measurements, but also allows users to enter their own estimates of both current and future weather conditions. We present the results of a 32-month public deployment of Atmos on the Google Play Store, showing that a combination of both types of sensing results in accurate temperature estimates, featuring an average error rate of 2.7C, whereas when using only user inputs, the average error rate drops to 1.86C. Sensor readings revealed significant variations during users commuting times.User inputs were more accurate in estimating actual temperature than sensor inputs.Bagged decision trees with user reported temperature achieved the lowest error rate.A 32-month public deployment on Google Play Store.