Validation of data from a wireless sensors network: application to frost alert

Abstract

Recent researches have shown the ability of Wireless Sensor Networks (WSN) to collect real-time data acquired by sensors distributed throughout a vineyard or a watershed area. Due to the emergence of new technologies, these researches focused on the technical and practical aspects of deployed solutions. They mainly focused on i) networks architecture, ii) communication protocols, iii) energy savings and iv) the ability to produce relevant real-time information. Few studies have focused on the quality of data collected by WSN. However, considering the cost of installation and the required maintenance for such devices, the deployment of a new WSN on a vineyard raises the question of data consistency and/or data redundancy. The objective of this study is to propose and test two methods of data processing coming from a WSN recently deployed over a commercial vineyard. The first method aims at testing data consistency in order to implement possible corrective actions on the sensors. The second method aims at identifying one or more representative sites of a more or less extensive area. Both methods have been implemented on temperature data acquired through a WSN set up in 2012 on a vineyard in the Bordeaux region. The WSN consisted of 54 temperature sensors. 51 sensors were positioned on a group of six contiguous fields with a total area of 11 ha. 3 sensors were positioned on three independent fields located at a distance of about 1 km from the main WSN. Each sensor records temperature every 10 min. A period of measurement stretching from April 4^th to May 15^th was considered. This period is interesting because frost can cause significant yield losses. This is also a time when preventive actions (water spray) can be quickly implemented. The study of WSN information aimed at i) analyzing the consistency of measured temperature by comparing it with a reference weather station (national french network) and ii) analyzing the covariance of 54 temperature measurements over time. The results showed that it is possible i) to identify and eliminate outliers corresponding to the logistical constraints of implementation during the installation of sensors, ii) to find different sets of fields corresponding to different climatic conditions iii) to identify homogeneous zones of reference for triggering a possible anti-frost alert.