1 Introduction2 System2.1 Application Background2.2 Architecture2.3 Implementation2.3.1 Burrow and Weather Motes2.3.2 Mote-based Networks2.3.3 Verification Network2.3.4 WAN and Base Station2.4 Media Access and Routing3 Analysis3.1 Lifetime3.2 Multi-hop network structure3.3 Routing Stability3.4 Packet Delivery Effectiveness4 Discussion4.1 Node Reclamation4.2 Physical Design4.3 Mote Software4.4 External Tools5 Related Work6 Conclusions7 REFERENCES -9ptAn Analysis of a Large Scale Habitat MonitoringApplicationRobert Szewczyk†, Alan Mainwaring?, Joseph Polastre†, John Anderson‡and David Culler††EECS Department?Intel Research Berkeley‡College of the AtlanticUniversity of California, Berkeley 2150 Shattuck Avenue 105 Eden St.Berkeley, California 94720 Berkeley, California 94704 Bar Harbor, ME 04609ABSTRACTHabitat and environmental monitoring is a driving application forwireless sensor networks. We present an analysis of data from asecond generation sensor networks deployed during the summerand autumn of 2003. During a 4 month deployment, these net-works, consisting of 150 devices, produced unique datasets forboth systems and biological analysis. This paper focuses on nodaland network performance, with an emphasis on lifetime, reliabil-ity, and the the static and dynamic aspects of single and multi-hopnetworks. We compare the results collected to expectations set dur-ing the design phase: we were able to accurately predict lifetime ofthe single-hop network, but we underestimated the impact of multi-hop traffic overhearing and the nuances of power source selection.While initial packet loss data was commensurate with lab experi-ments, over the duration of the deployment, reliability of the back-end infrastructure and the transit network had a dominant impacton overall network performance. Finally, we evaluate the physicaldesign of the sensor node based on deployment experience and apost mortem analysis. The results shed light on a number of de-sign issues from network deployment, through selection of powersources to optimizations of routing decisions.Categories and Subject DescriptorsC.2.1 [Computer-Communication Networks:]: Network Archi-tecture and DesignWireless Communications; C.3 [Special-PurposeAnd Application-Based Systems]: Real-Time and embedded sys-tems; C.4 [Performance of Systems]: Design StudiesGeneral TermsPerformance, Design, ImplementationKeywordsSensor Networks, Habitat Monitoring, Microclimate Monitoring,Network Architecture, Long-Lived Systems, Application Analysis1. INTRODUCTIONPermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.SenSys’04, November 3–5, 2004, Baltimore, Maryland, USA.Copyright 2004 ACM 1-58113-879-2/04/0011 ...$5.00.A broad class of applications are within the reach of contempo-rary wireless sensor networks (WSNs). These applications sharea common structure, where fields of sensors are tasked to take pe-riodic readings, and report results and derived values to a centralrepository. There are both scientific and commercial applications,for example: microclimate monitoring, plant physiology, animalbehavior [16], precision agriculture [2, 4], structural monitoring[5] and condition-based maintenance. These sense-and-send appli-cations have widely-varying sampling rates and network bandwidthdemands.In the context of habitat and environmental monitoring, WSNsoffer significant advantages. Individual devices can be made suf-ficiently numerous to take measurements at many locations of in-terest, and mitigate errors arising from the interpolation and ex-trapolation from coarser-grained samples. They can be sufficientlysmall to be co-located with phenomena of interest without alteringthe parameters to be measured. And they can be unobtrusively em-bedded in the environment without creating conspicuous landmarksthat change the behaviors of its inhabitants.Long-term unattended operation enables measurement at spatialand temporal scales impractical with human observers or sparselydeployed instruments. The lifetimes made possible with contempo-rary low-power microelectronics can prolong the duration of exper-imental observations. At the same time, automation improves thedata quality and uniformity of measurement, while reducing datacollection costs as compared with traditional human-centric meth-ods. Devices can operate for prolonged periods in habitats that areinhospitable, challenging or ecologically too sensitive for humanvisitation. Unobtrusive observation is key for studying natural phe-nomena.WSNs offer more capabilities than standalone dataloggers andwired instrumentation. Wireless telemetry is valuable because itminimizes observer effects, study site intrusions and environmentalalterations. For example, visits to study areas to monitor and down-load loggers are no longer necessary, while health and status of in-strumentation can be monitored remotely. More general network-ing offers great benefits, such as continuously updated databases ofsensor readings accessible through the web, access to live readingsfrom individual sensors, and is key to distributed in-network pro-cessing. These capabilities may yield new experimental designs,and paradigms for data publication, dissemination, and scientificcollaboration.We have incrementally deployed several sensor networks of in-creasing scale and physical extent in a wildlife preserve. Whileamassing a novel dataset for biological analysis, the annotated dataare interesting from a systems perspective. The packet logs froma single-hop and multi-hop network reveal insight on lifetimes,packet yields, network structure and routing. For example, someFigure 1: Geospatial distribution of petrels obtained by direct human observation (left) and a particular feature of the habitat(average temperature at midnight in the burrows (center) and on the surface (right) collected from out sensor network)nodes ran for nearly four months but some for just a few days.Analysis reveals changes in network structure and performance overthe lifetime of the deployment. Though the application was sim-ple, it exhibited interesting and unexpected behaviors after its ini-tial setup.