Low-cost, low-power wireless sensor networks (mote networks) have the potential to revolutionize data collection methods in hydrology. They promise the ability to monitor catchments at very high spatial and temporal resolution with flexible sampling schemes, real time data processing and high levels of quality control. We operated an experimental network of 41 motes monitoring seven different parameters each at 15 min intervals for 10 months in a small forested catchment in southwestern British Columbia, Canada, to determine if this emerging technology is suitable for use by hydrologists in its current form. Our particular interests were ease of setup, sampling reliability, power consumption, and hardware resilience. We found that while motes gave the ability to monitor a catchment at resolution levels that were previously impossible, they still need to evolve into an easier to use, more reliable platform before they can replace traditional data collection methods.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 A dense network of inexpensive logging stations can provide as much or more useful data than a few expensive stations [Delin, 2002; Mainwaring et al., 2002]. Most current catchment experiments require that researchers make a choice between high spatial resolution, where many measurements are taken manually at whatever time intervals are possible for the researcher [Western et al., 1999] or high temporal resolution, where a few expensive logging stations are deployed at strategic locations and record information over time. With an inexpensive and reliable method for sensing and data collection, it would not be necessary to choose between spatial and temporal resolution. The emerging technology of low-cost, low-power wireless sensor networks (known as mote networks) offers the potential for much denser instrumentation and hence more detailed data collection. At current prices of $200 U.S. and dropping for an individual mote (the core of each measurement station in a mote network) it is possible that the cost of the sensor platform could become a small component of the price of the entire data collection network. Total cost of a network would be largely dictated by the cost of the sensors that are employed.
 An area in hydrology where motes have the potential to make a large impact is for input into physically based distributed models. As opposed to lumped models that consider a catchment as one unit, distributed physical hydrological models take into consideration the spatial variability of hydrologic parameters within a catchment [Refsgaard, 1997]. While it is desirable to use a model based on physical characteristics, physically based hydrological models present many difficulties that stem from our inability to accurately describe the spatial and temporal variability of parameters in a catchment [Seyfried and Wilcox, 1995]. Overcoming this inability requires a cheaper and simpler method to obtain rich spatiotemporal data, such as a mote network.
 The low cost of a mote network also allows for the possibility of an abundance of measurement stations to cover an area and self organize into a network where information is processed in place [Hill and Culler, 2002] and then communicated directly to an existing network server in real time. As an individual mote can wirelessly communicate and send data through other motes so that it reaches a base station (as opposed to having to send data directly to a base station) there is no limit on the ability to expand the network size as long as each mote is within range of at least one other. This type of data transfer is known as mesh networking, where each node (in this case a mote) has the ability to choose the best path to send data to the base station. A mesh network will result in a very adaptable environmental monitoring system that does not require complete redesign when experimental goals change; instead it requires activation of different sensors that are already deployed in the field. This switch from instrumentation deployed for specific purposes to instrumentation deployed for monitoring, with no specific experimental goal in mind is a step toward proposed terrestrial observatories, where different monitoring networks can be linked together and used for a wide range of purposes [Bogena et al., 2006].
 Real time data collection will greatly help to speed along the process of collecting hydrological data, and hopefully proceed toward integrating the steps of collecting and processing data. Also, real time collection will allow for online data quality control and on demand maintenance as it has already been shown that anomalies in sensor network readings can reliably predict sensor failures [Szewczyk et al., 2004]. This will lead to less data loss and better data quality as the malfunctioning instruments can be repaired right away instead of left running in the field because they are assumed to still be functioning.
1.2. Current State of Development
 The possibility of inexpensive and simple high-resolution hydrologic monitoring is getting closer to reality. Crossbow Technology, Inc.™ began selling the world's first open architecture wireless sensing platform (mote) in 2002 (S. Lee, Crossbow Technology, Inc. quick facts, 2007, http://www.xbow.com/General_info/Info_pdf_files/Crossbow_Quick_Facts.pdf). Their sensor platform, called the MICA platform, is a flexible and inexpensive device made of off the shelf components that is capable of sensing, data routing, communicating with other devices, storing data and on site processing [Hill and Culler, 2002].
 Field experimentation with the Crossbow motes is necessary to evaluate if they can deliver on their promises of a flexible and inexpensive wireless environmental monitoring system in situations more challenging than the sterile lab setting. One of the first field experiments with the Crossbow motes was the Great Duck Island experiment in Maine, United States. This experiment focused on nonintrusive monitoring of the nesting habits of the Storm Petrel by using light sensors to determine when the birds are inside their burrows [Szewczyk et al., 2004]. In this experiment, which included 43 nodes, the initial 4 months of data had an average daily success rate ranging from 70 to 95% (excluding an approximately 2 week interval where the entire system was down. Considering that it would have been much more difficult to meet the goal of nonintrusive monitoring with traditional logging it can be considered successful; however, they did experience many node failures for unexplained reasons.
 Outside of the natural environment, there have also been industrial experiments with motes to determine their readiness for use beyond the realm of scientific research. Glaser  investigated the applicability of motes, in their current development form to specific industrial monitoring applications. This experiment focused on how ready the TinyOS operating system and wireless sensor nodes are for real world applications. On paper, the motes offered by Crossbow seem like an ideal candidate for identifying structural damage in buildings, as “dense, inexpensive instrumentation is needed to identify structural damage and prognosticate future behaviour” [Glaser, 2004]. However, with a less than 50% success rate, Glaser concluded that “the hardware is not electronically robust,” and the software is not user friendly enough to be used by most engineers in an industrial scenario yet.
 The objective of this study was to test the ability of motes to reliably collect and store data in a complex environment (forested watershed). Other groups are testing the mesh network capabilities of the Crossbow motes, but this capability is not useful to hydrologists unless they are first known to be able to reliably data collection devices. We were also interested in determining if the potential of greatly reduced data collection time and effort is worth the increased complexity of initial setup for scientists who do not have extensive knowledge of computer science [Glaser, 2004].
2.1. Study Site
 Field experiments were carried out in a 7 ha rain dominated forested catchment in Malcolm Knapp Research Forest, a research forest operated by the University of British Columbia in Maple Ridge, BC, Canada. The catchment is located in the Coastal Western Hemlock biogeoclimatic zone (B. Egan, The ecology of the coastal western hemlock zone, 1999, http://www.for.gov.bc.ca/hfd/pubs/docs/Bro/bro31.pdf), has an annual average temperature of 9.6°C and receives an average of 2200 mm of precipitation per year [Environment Canada, 2006–2007], almost all in the form of rain. The digital elevation model shown in Figure 1 was supplied from a previous study in the catchment [Moore and Thompson, 1996]. The study catchment included 41 mote-based measurement stations. Measurement station locations were chosen with the desire to cover the different range of topographic features present in the catchment.
2.2. Hardware Description
 Each measurement station in this study catchment included a powered data box with a Crossbow Technologies, Inc. MDA300 data acquisition board, and MICA2 processor board. Each parameter was measured and recorded every 15 min and stored in flash memory on the MICA2 processor boards. When fully operational, the MICA2 processors (motes) will communicate with each other over a radio frequency of 433 MHz to form a wireless sensor network. However, our first priority was to develop a well functioning instrument cluster that can store the measured data reliably and communicate only directly with a base station to download the data, and as such must be individually polled by a wireless base station attached to a laptop computer. The motes run the open source operating system TinyOS, which uses the NesC programming language [Gay et al., 2003].
 Each measurement station included the following measurements: air temperature, humidity, soil temperature, rainfall intensity, soil moisture, and groundwater head. Sixteen of the 41 locations also included a sensor to measure overland flow. Air temperature and humidity were measured using Humirel HTM 2500 transducers (Humirel, Relative humidity/temperature module technical data, 2002, http://www.humirel.com/product/fichier/HTM2500.pdf); soil temperature was measured using a thermistor encased in a thin layer of epoxy and inserted 15 cm into the ground. Throughfall was measured using Rainwise, Inc. Rainew tipping buckets (Rainwise, Inc., Wired rain gauge, 2007, http://www.rainwise.com/products/detail.php?ID=6697&Category=Rain_Gauges:Wired&pageNum_cart=/products/index.php). The tipping buckets were factory calibrated to tip every 0.25 mm of rainfall and mounted on posts 1 m above the forest floor. Soil moisture was measured using Decagon Devices, Inc. ECH2O dielectric aquameters (Decagon Devices, Inc., ECH2O soil moisture probe operator's manual, 2006, http://www.decagon.com/manuals/echomanual.pdf). ECH20 aquameters have an advertised accuracy of ±4%. Bogena et al. , however, found that the probes were somewhat influenced by soil temperature and conductivity. Groundwater level was measured at each site in a PVC piezometer inserted into the ground down to bedrock (usually 1 to 1.5 m) with a slotted section in the bottom 10 cm of length. The water level was recorded with 4 m SensorTechnics pressure transducers with a range of 0–3 psig and a sensitivity of 0.1% of the total length (SensorTechnics, OEM stainless steel submersible pressure transducers, 2007, http://www.sensortechnics.com/download/cte-ctu9000cs-594.pdf). Overland flow was recorded using small (5 cm) V notch weirs inserted into the ground perpendicular to the fall line and recorded in binary (yes/no) using a simple float switch [Srinivasan et al., 2000]. See Table 1 for a summary of instrumentation at each measurement location. An enclosure to house the Mica2 motes, MDA300 environmental data acquisition boards and power supply (two D-cell alkaline batteries) was constructed from GSI Industries™ lexan plastic boxes and is shown in Figure 2.
Table 1. Instrument Cluster Functions
% by volume
Decagon Devices, Inc. ECH2O dielectric aquameter
Rainwise, Inc. Rainew tipping bucket
Sensor Technics pressure transducer
Humirel HTM 2500 transducer
Humirel HTM 2500 transducer
Custom Weir with binary float switch
Internal battery power
Internal to MDA300
Internal to MDA300
Internal to MDA300
2.3. Software Description
 The framework of our software system consisted of the following three components: (1) data collecting, (2) data logging, and (3) data polling. The data collecting component gathered data from the sensors, and the data logging component recorded all the information locally to the 512 kb of on board flash memory. The polling component downloaded data from each mote's flash memory to the polling node that was attached to a field laptop via the Crossbow MIB520 USB mote base station. Raw data was converted to a usable form using a conversion program written in IDL (Interactive Data Language) by RSI, Inc. version 6.2 (D. Stern, IDL student edition software, Research Systems, Inc. and ITT Visual Information Solutions, Boulder, Colo., 2003).
 Implementation of the Crossbow mote network went through multiple development stages. Initial TinyOS software development for the mote platform was indoor, lab style testing from June 2005 to January 2006. In January 2006, a pilot field study of 10 motes was deployed to Malcolm Knapp Research Forest. Of the first 10 motes deployed for 2 weeks, 8 functioned somewhat successfully, and for unclear reasons two did not collect any data. During this time each mote had to be hardwired and reprogrammed to enable downloading, which proved to be much more time consuming than anticipated. In July 2006, we moved TinyOS development from a Windows XP system to an Ubuntu Linux operating system. At this time the experiment was expanded to 41 measurement locations and wireless downloading was enabled. Data collection continued in this manner until 20 April 2007.
Figure 3 shows an example of the data collected from one logging station during a rain event on 6–7 November 2006. Soil moisture (measured every 15 min) and hourly throughfall is shown for the entire 48 h period. Comparison of the soil moisture data with throughfall illustrates very little moisture storage in the unsaturated zone, as soil moisture is constantly decreasing without water input from throughfall. Also, there is evidence of a small amount of data noise from the ECH2O soil moisture probes.
4.1. Power Usage
 Each mote was powered with two D-cell alkaline batteries (14,000 mAh each). Batteries were replaced approximately every 30 days throughout the course of the experiment. During one interval during the winter of 2006–2007, however, batteries were left in the motes until they were completely discharged because of our inability to access the catchment. This gave us a chance to evaluate battery performance during the harshest conditions encountered (very wet conditions with temperatures generally between −2 and 5°C). Results of this test are presented in Table 2. Though Crossbow motes have an advertised operating range of 2.7 to 3.3 V, our experiment showed reliable operation at as low as 2.55 V and some operation as low as 2.23 V.
AVG, average; STDEV, standard deviation; MIN, minimum; MAX, maximum.
Value at failure
 It would be possible to greatly increase the run time of the motes to well over 1 year with the same batteries with the use of the “sleep” function, where motes are only running for a few seconds at every measurement time [Anastasi et al., 2004]. Putting the motes to sleep was not feasible in this experiment because of the use of tipping buckets, because each mote had to be awake at all times to record when tips happened and sum the total amount of tips that occurred in each 15 min interval. The “idle” function available for the motes, however, could provide an intermediate level of processor functionality where tips could be recorded while using less than full power, thus increasing the battery life.
 As with any field experiment, uncontrollable environmental issues affected this project. Windfall debris from the forest canopy, snowfall, and wildlife all affected the operation of the monitoring system throughout the project. These things could be expected from almost any type of monitoring system deployed in the coastal mountains of British Columbia.
 Unexpected reliability problems occurred when the temperature probes produced erroneous readings when exposed to temperatures near or below freezing temperatures. Because the Humirel HTM 2500 has an operating range of −30 to 70°C (Humirel, Relative humidity/temperature module technical data, 2002, http://www.humirel.com/product/fichier/HTM2500.pdf), we suspect that this error resulted from the either the MDA300 or the MICA2 rather than errors from the temperature probes themselves. Different motes had this error at different times, and it affected six motes in total.
 A major disappointment in this experiment was with the groundwater head measurement. Pressure transducers had problems throughout the entire experiment. During the developmental stages of the TinyOS software, pressure transducers were tested with the factory calibration information using a Campbell Scientific CR10 logger. Results from these tests indicated a very high degree of linearity between output voltage and water depth (r2 = 0.999). When the identical program for operating the SensorTechnics pressure transducers was written for TinyOS the transducers seemed to function normally in the lab, but in the field we could not get the pressure transducers to operate at an acceptable level. Of the 41 operational instrument clusters, four had normally operating pressure transducers, 25 gave unreliable data and twelve had no response whatsoever.
 Aside from the data reliability problems described above, there were periods when individual sensors ceased operation for unknown reasons. Table 3 shows the amount of usable data given by each analog sensor as a percent of the total number of readings taken by each mote. Usable data was considered to be a sensor reading that was within the range of realistic values for each measurement. (i.e., a temperature reading that was between −10 and 40 C or a soil moisture percentage between 0% and 100%). Large changes between consecutive data points were also counted as errors on a case by case basis (i.e., a temperature reading of −5° followed 15 min later by a reading of 25°C).
Table 3. Percentage of Usable Data During the Main Sampling Period of 280 Daysa
AVG, average; MED, median; STDEV, standard deviation; MIN, minimum.
Internal battery power
 As illustrated in Table 3, measurements that were internal to the MDA300 had very good reliability. They were functional for almost the entire time that the MICA2 was functional. With a few exceptions, the soil temperature readings and readings from the Humirel probes (air temperature and humidity) gave very good results as well. Almost all errors with the Humirel probes occurred at the rare times where the temperature was below freezing (as described previously). Soil moisture readings were somewhat less reliable, but the median of 100.00 illustrates that more than half of the sensors had 100% reliability when the mote was active. The large standard deviation shows that there were a few numerous motes with very poor reliability results for soil moisture. Because of previously described problems with the pressure transducers, groundwater level was not included in this chart, as this measurement was almost completely unreliable. Instruments run on the digital channels of the MDA300 also were not included, as it was impossible to know for sure if a reading of 0 simply meant that there was no activity, or if there was a malfunction with the instrument. However, having 41 different locations in operation did allow for manual cross checking of throughfall measurements for data quality. There were, for example, large storm events where almost all motes were recording large amounts of throughfall, and just a few were recording nothing. With this information we could deduce a malfunction had occurred. When reviewing field notes we could determine if this malfunction was due to simple tipping bucket blockages or due to the hardware itself.
 While motes certainly have a large potential for hydrologic measurement, results of this experiment have shown that they may not yet be reliable or user-friendly enough for widespread use by hydrologists. Logging stations in this experiment often gave erroneous readings for unknown reasons. Because the software and hardware setup was identical for each mote, inconsistencies in the function of the MICA2 and/or MDA300 are the only places left to look for reasons for the unknown data failure. We hope and expect that future generations of the MICA platform will be more robust and consistent in this area.
 Aside from some inconsistency in the hardware, we feel that a major factor preventing motes from revolutionizing environmental data collection right now is software. Currently, an in-depth knowledge of computer programming in the NesC language is necessary for operation of motes in TinyOS. They are much more difficult to configure for operation than a traditional data logger and are still in the research stage instead of the operational phase. This fact makes motes less appealing for most hydrologists, who are interested in spending time understanding the data they have collected instead of configuring a sensor network. This will likely change in the near future, however, as software is being developed to make motes easier to use, such as MoteWorks 2.0, a software developing environment containing many tools that adds many tools to TinyOS to allow for easier initialization and operation of environmental monitoring systems using motes. MoteWorks 2.0 also promises better stability and much greater data transmission reliability than stand-alone TinyOS (Crossbow Technologies, Inc., MoteWorks 2.0 Software Platform, 2007, http://www.xbow.com/Products/Product_pdf_files/Wireless_pdf/MoteWorks_OEM_Edition.pdf).
 Another attribute of mote technology that needs to be considered is the extremely rapid advancements being made in the field. Multiple new versions of hardware and software have been available, and each has been an improvement on the last. This rapid advancement in technology is good for development, as problems can be fixed with new hardware, but it may be a detriment to widespread use of motes in hydrology or other environmental sensing. These types of data collection projects require stability in hardware and the ability to replace equipment if it becomes damaged by uncontrollable environmental factors. Even through the course of this project, the MICA2 processor boards that we used have already been replaced with newer hardware. While newer hardware will likely solve some problems that we had during this study, this means that replacing any equipment that breaks in the field will be difficult. It is not feasible for hydrologists to purchase new hardware every few years for long-term measurement projects, and technology development will have to stabilize before mote networks become a viable option for large-scale environmental sensing projects.
 We see a great potential for motes to allow hydrologists to perform experiments that were previously logistically impossible. They will be able to supply rich data in both space and time at a price that is significantly lower than current data collection hardware. This potential has not quite been reached yet; hydrologists are interested in reliability and ease of use for data collection. So far in this project we have found that these qualities have not yet reached a point that warrants widespread use of motes in hydrology; the hardware needs to become more reliable, and the software needs to become easier to use before they can compete with the current methods of data collection and recording. As mote technology and software is advancing at a rapid pace, this potential should become reality very soon.
 NSERC (National Sciences and Engineering Research Council of Canada) and the GEOIDE Sensor Web Automation Network (SWAN).