Live E! Sensor Network: Correlations in Time and Space

The Live E! project is a Japanese research consortium among industry and academia to develop a platform collecting and sharing digital information related with the earth. This platform consists of a large number of spatially distributed weather stations measuring different environmental quantities such as temperature, humidity, pressure, etc. In this paper, we conduct the first analysis on this huge data set. To that end, we perform a synchronous-average based decomposition of the collected time series, and we explore the structures of correlations both in time and space on the observed data. 1 Motivation and context Recent natural disasters, such as heat island effect or hurricane, have highlighted the need for concerted research efforts to better understand the causes and consequences of such natural phenomena whose impacts on social life and business activity is significant. In this context, the live E! project [1], developed by the WIDE consortium and the IPv6 Promotion Council Japan, deploys a global infrastructure aiming at collecting and sharing environmental information. Live E! provides a multi-domain sensor networking platform, which is composed of several weather stations deployed across several countries of Asia, including Japan, Thailand, etc. Each weather station is operated in autonomous and distributed manner and embeds several sensors measuring standard environmental quantities, such as temperature, pressure, humidity, wind speed, etc. As depicted in Fig. 1, each weather station is managed by an operational-unit such as an university or a company which has sensors for their own activities. An operationalunit has a server, which collects data from weather stations basically in real-time via the Internet. Those servers collaboratively exchange the data among them using the Live E! standard protocol over the Internet, providing multi-attribute sensor search and data retrieval interface for users. In February 2009, the Live E! platform has 106 weather stations deployed across 13 countries and 11 servers across 9 organizations. Readers are referred to [1, 2] for further information on the live E! platform. Fig. 1: Live E! network architecture. In this paper, we provide a first analysis of the space and time correlations of the collected live E! measurements. We propose an original synchronous average-based model to decompose each time series in order to separate dependencies stemming from expected cyclic trends, 10 0 10 2 10 4 10 6 10 8 −1 −0.5 0 0.5 1 Temperature Log( lag(mn) ) 1 year 1 day 10 0 10 2 10 4 10 6 10 8 −1 −0.5 0 0.5 1 Humidity Log( lag(mn) ) 1 year 1 day 10 0 10 2 10 4 10 6 10 8 −1 −0.5 0 0.5 1 Pressure Log( lag(mn) ) 1 year 1 day Fig. 2: Autocorrelation functions (ACF) on Temperature, Humidity and Pressure. such as seasonal (year) and daily effects from dependencies actually observed on residual fluctuations. Then, we investigate the autoand crosscorrelations of both the measurements observed at a given weather station and between measurements collected at adjacent stations. The remainder of this paper is organized as follows. Section 2 describes the data set. In Section 3, we propose an original synchronous average-based decomposition of time series. Then, we investigate, in Section 4, the space and time correlations both for the entire data and the fluctuations. Finally, we conclude in Section 5 and we give some prospects of future work in the area.