Computers & Geosciences 74 (2015) 50–59

Contents lists available at ScienceDirect

Computers & Geosciences

journal homepage: www.elsevier.com/locate/cageo

Niphargus: A silicon band-gap sensor temperature logger for high-

precision environmental monitoring
C. Burlet n, Y. Vanbrabant, K. Piessens, K. Welkenhuysen, S. Verheyden
Royal Belgian Institute of Natural Sciences, Geological Survey of Belgium, Rue Jenner 13, B-1000 Brussels, Belgium

art ic l e i nf o a b s t r a c t

Article history: A temperature logger, named “Niphargus”, was developed at the Geological Survey of Belgium to monitor
Received 17 November 2013 temperature of local natural processes. It has a sensitivity of the order of a few hundredths of degrees on
Received in revised form temperature variability in open air, caves, soils and river environment. The newly developed instrument
8 September 2014
uses a state-of-the-art band-gap silicon temperature sensor with integrated digital output. This sensor
Accepted 23 October 2014
Available online 29 October 2014
reduces the risk of drift associated with thermistor-based sensing devices, especially in humid en-
vironments. The Niphargus is designed to be highly reliable, low-cost and powered by a single lithium
Keywords: cell with up to several years autonomy, depending on the sampling rate and environmental conditions. A
Band-gap batch of Niphargus loggers was also compared to a precision thermistor to assess absolute temperature
Cave
accuracy. Further characterization came from two field case studies in Belgium: monitoring of a mi-
Monitoring
neralized water stream near the town of Spa and air temperature monitoring inside Han-sur-Lesse cave.
Sensor
Temperature & Elsevier Ltd. All rights reserved.
Water stream

1. Introduction
Stable and reliable temperature monitoring instruments with
high accuracy and/or precision are often required in geosciences
applications such as field microclimate models and geophysical
studies (e.g. Pfender and Villinger, 2002, Mentes, 2012). Likewise, a
precise temperature measurement is often a key element to
guarantee stable and calibrated outputs of many other sensors
across a defined temperature range. This is why temperature
sensors are one of the most ubiquitous devices in modern elec-
tronic monitoring equipment (e.g. Ambrosetti et al., 2012; Dedrick
et al., 2000; Han et al., 2012).
1.1. Comparison of common temperature sensor technologies
will determine whether or not the device will satisfy the re-
quirements of an application. In Table 1, for example, one can see
that the thermocouple has a wide temperature range, poor line-
arity and relatively low temperature sensitivity which make it
usable for high temperature applications where accuracy is not
decisive. On the other end, RTD and thermistor have good signal
linearity that allows more accurate measurements, but with
smaller temperature ranges. RTD has a good stability over time but
low sensitivity while thermistor has a better sensivity but is more
sensible to drift over time with its poor resistance to vibrations
and mechanical stress (Baker, 1999, 2004).
1.2. Silicon band-gap sensor interest in long-term environmental
monitoring.

Among the most popular temperature sensors used today are
the thermocouple, the resistive temperature device (RTD), the
thermistor, and the integrated silicon based sensors (Baker, 1999;
Benedict, 1969). There are also less used thermal sensing tech-
nologies, such as infrared (pyrometers) and thermopiles.
Table 1 resumes the most popular temperature sensing tech-
nologies and their main characteristics. None of these technologies
are ideal for all type of environments and applications. Char-
acteristics such as temperature range, sensitivity or robustness
n
Corresponding author.
E-mail address: christian.burlet@naturalsciences.be (C. Burlet).
Among temperature technologies, the silicon band-gap sensor,
based on a silicon diode with a temperature dependent forward
voltage, underwent a large development over the last decade. A
review of existing temperature measurement technologies and
recent developments in silicon band-gap temperature sensors
shows that the sensitivity of those sensors improved by a factor of
10–50 in the last 10 years, thanks to the miniaturization of precise
voltage references and sensible delta sigma Analog-to-Digital
Converter (Baker, 1999, 2004; Dong-Ok et al., 2008; Hagart-Alex-
ander, 2010). Moreover, the improvement in silicon wafer dicing
technologies and CMOS technology generalization enables more
precise factory calibration of the silicon diode, improving initial

http://dx.doi.org/10.1016/j.cageo.2014.10.009
0098-3004/& Elsevier Ltd. All rights reserved.
 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59 51

Table 1
Typical characteristics of the most common temperature sensing technologies (Baker 1999, 2004, Wang, 2005, Analog Devices, 2012).

Thermocouple RTD Thermistor Silicon band-gap

Measurement range
270 to 1800 °C
250 to 900 °C
100 to 450 °C
55 to 150 °C

Sensitivity Low Low High High
Typical absolute 0.5 °C 7 0.01 °C 7 0.1 °C 0.25 °C
accuracy
Linearity Low Good Medium Excellent
Long term drift Need regular cold junction Need regular calibration Need regular calibration Very low drift over time
calibration
Type of output Voltage Resistance Resistance Voltage
Robustness Usually rugged. Depend on the RTDs are susceptible to damage as a re- Very sensible to mechanical Rugged as any integrated circuit
casing (wire gage, steel rod). sult of vibration, humidity also affect stress and humidity housed in a plastic package.
resistance.

absolute accuracy of the sensors (Wang, 2005; Analog Devices,

2012).
Nowadays, silicon band-gap sensors are often included in
monolithic plastic or ceramic packages (integrated chips) with an
embedded voltage reference and an analog to digital conversion
device. Due to their monolithic design, the sensors are very robust
and less sensitive to environmental conditions (Wang, 2005;
Spevak and Forstner, 2009). Their small size allows them to be
plotted in resin or coated for use in aggressive chemical environ-
ments. They present a low temperature drift over time. For ex-
ample, the ADT7420 from Analog Devices has been rated with a
drift of less than 0.0073 °C over the estimated operating life of the
sensor (150 °C heating test during 500 h following JEDEC standard
22A108D) (Analog Devices, 2012; JEDEC, 2010). This very low drift
is crucial during long-term monitoring where no regular calibra-
tion is possible.
The ADC circuitry (16 bits conversion), integrated in the chip,
allows a good temperature resolution acquisition (0.0078 °C). This
kind of state-of-the-art temperature sensor represents a good al-
ternative for precision temperature measurements, and a com-
promise between the stability of the RTD and the sensivity of the
thermistor.
One drawback of the monolithic design of band-gap sensors is
the relatively small size of the sensing diode associated with the
varying size and shape of the plastic housing of the chip. The
design tends to reduce its thermal response comparatively to ex- Fig. 1. A Niphargus logger in its standard size (ADT7410 model). SMD components
posed thermistor devices. This makes the sensors less suitable in side. Photo credit: W. Miseur (RBINS).
fast processes (see laboratory results for an evaluation).
Table 2
Individual components used in the Niphargus Temperature logger.
2. Methodology and principle of operation
Parts Model Provider

2.1. Hardware design Integrated circuitsPIC16LF1825 microcontroller Microchip

24AA1025 EEPROM Microchip
The design of the Niphargus logger (Fig. 1) is aimed at a low ADT7410 (ADT7420) Analog Device
RTC DS1337C# Maxim IC
assembling cost and reliable operation in high humidity environ-
FT230XQ FTDI
ments. The first intended application was the monitoring of cave Passive components Resistors SMD (1  5.1 Kohm, Vishay
air temperature in Belgium, with a requirement for continuous 1  2 Kohm, 5  10 kohm)
operation during months to years. Those environments frequently Capacitors SMD (5  100 nF) AVX
present high humidity levels, close or at condensation point Electric double layer capacitor Panasonic
(1  0.015 F)
(4 95% R.H.). In order to increase reliability, the number of dis- Diodes LED SMD 120 mcd, 590 nm Wurth
crete components on the circuit was minimized, hence reducing Elektronik
the number of soldered joints potentially affected by oxidation. General purpose Schottky diode NXP
The components used in the device are listed in Table 2. The 1PS76SB40
Battery LiMnO2 CR123 type (typical 1400 mA h Panasonic
logger is based on an 8 bit microcontroller (Microchip
capacity)
PIC16LF1825). This microcontroller presents features particularly CR123 battery holder Keystone
well-fitted for this application: among them an operating voltage Connector Mini USB Molex
of 1.8–5.5 V and 14 kB of program memory. The microcontroller PCB Double layer FR4 170 Tg/290 Td (lead- OSHPark
needs only a few external parts to perform and has been design for free), 1.6 mm thick. 1 ounce copper on
both sides
low power applications (current consumption during sleep run- ENIG finish (gold)
ning is about 20 nA: Microchip Technology Inc., 2011a). The
52 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59
Fig. 2. Electrical schematic of the Niphargus logger (ADT7410 model).
measurements are stored on a 1024  8 bits EEPROM (Microchip
24AA1025) and the applied temperature sensors are silicon band-
gap type ADT7410 (Analog Devices, 2011) or ADT7420 (Analog
Devices, 2012). A small low-current LED was fitted on the circuit to
show activity and the 2.048 V voltage reference integrated in the
microcontroller is internally wired to its 10 bits analog-to-digital
converter for battery feedback. Configuration of the logger and
data retrieval are made using a USB to serial (TTL) interface based
on a FTDI FT230XQ chip (FTDI, 2013) and a Mini USB type B con-
nector mounted on the board. This chip emulates a Communica-
tion “COM” port on the host computer.
Time keeping is done using an integrated real-time-clock (RTC),
the DS1337C (Maxim Integrated, 2011), with an integrated crystal
oscillator of 32.768 kHz (see Section 2.3). The EEPROM, RTC and
temperature sensor are interfaced with the microcontroller using
I²C communication protocol (Philips semiconductors, 1995). See
Fig. 2 for a detailed electrical schematic. A “trickle-charge” circuit
is used to maintain time-keeping operation during a power failure
or battery replacement. This circuit is composed of a Schottky
diode connected in series with a 2 kΩ resistor in order to maintain
the charge of a 0.015 F Electric Double Layer Capacitor (real-time-
clock backup power of about 14 h).
The circuit is intended to be powered by a 3 V lithium non-
rechargeable cell (LiMnO2 type), CR123 size, with a nominal ca-
pacity of about 1400 mA h (see Section 2.4 Power budgeting). No
linear regulator is used in the circuit but as the ADT7410 and
ADT7420 temperature sensors have their own build-in regulator,
no drift in temperature measurement is expected throughout the
life of the battery.
With this setup, the Niphargus is able to record temperatures
between
55 and 125 °C (maximum tested range), with a theo-
retical resolution of 0.0076 °C per sample and initial absolute ac-
curacy of 0.5 °C for the ADT7410 sensor or 0.25 °C for the ADT7420
sensor. The cost price of the Niphargus is around 20€. This price
does not include soldering and coating consumables, nor the Ni-
phargus casing.
2.2. Humidity-effects mitigation
As the sensor is intended to be left in humid conditions, the
following measures have been taken to protect the electronics.
– A single cell lithium manganese oxide (LiMnO2) was chosen to
power the Niphargus logger. This battery chemistry was chosen
for its long shelf life ( 410 years), low internal equivalent re-
sistance, high power density, and wide temperature operating
range. A CR123 size battery was used to maximize the distance
between battery poles reducing the risk of a short circuit due to
condensation that affected for instance coin cell battery pow-
ering temperature sensors in previous cave monitoring
experiments.
– A conformal coating (Electrolube acrylic toluene free (TFA)
type) was applied on the soldered PCB (except the USB con-
nector and battery holder) to further reduce the risk of circuit
performance degradation or shortcuts between soldered parts.
The coating has a typical thickness of 25 mm and contains an
active UV trace to control proper application. The coating is
 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59
stable between
65 and 125 °C and has a low dielectric con-
stant of 2.5 (Electrolube, 2012).
– The circuit board of the Niphargus is available in two layouts, to
enable fitting into two different containers. The first layout is
designed to fit in a 35 mm film box and is used for standard
field measurements. The other is designed to enter in a sealed
32 mm PVC tube with screw top, for submersed locations.
These containers are described in the presentation of the case
studies below.
2.3. Integrated crystal real-time-clock (RTC) and time-keeping
precision
During the early development of the sensor, a microcontroller
with an integrated real-time-clock was used (PIC18F24J50) to
timestamp the recorded data. This microcontroller, like most of
real-time-clock devices, uses a Pierce Oscillator circuit (Bible and
De Boni, 2007; Spevak and Forstner, 2009) based on a low fre-
quency external tuning fork quartz crystal to drive its timing
register. The layout and use of those crystals in miniature devices
are often a delicate part of the design, due to the sensitivity of
those components to temperature variations and EMI (electro-
magnetic interference) (Sridharan and Melear, 2005). While this
setup worked in dry conditions, in high-humidity conditions cur-
rent leaks due to condensation on the circuit board (increased by
solder flux residues on SMD parts) occurred and prevented the
oscillator from starting or affected its stability over time.
The initial design was replaced by an RTC with an oscillator
integrated directly in the plastic package of the chip to overcome
these issues and especially improve reliability in high-humidity
environments. Initial frequency accuracy of integrated the crystal
is 715 ppm at 25 °C (¼maximum deviation of 8 min/year). For
budget reasons, the crystal is not temperature compensated and
has typical ∼1 ppm/°C frequency dependence in the
10 °C to
30 °C range. This type of time-keeping accuracy is sufficient for our
current applications (monitoring of slow environmental processes)
especially as actual time-keeping accuracy was observed to be
much better than datasheet figures (∼1 min deviation in 6 months
in the Han-sur-Lesse cave).
Fig. 3. Block diagram of the Niphargus datalogger working principle and associated “Shrimp” PC software.
53
2.4. Microcontroller software
Fig. 3 presents a block diagram of the Niphargus logger and
associated software. The PIC microcontroller was programmed
using JAL programming language (sourceforge.net/projects/jal), a
Pascal-like language developed for the 8 bits microcontroller of
the Microchip PIC 12, 16 and 18 models. The software is designed
to handle temperature measurements and logging functions with
a serial ASCII-based communication protocol implemented for PC
communication. So it is possible to reprogram and debug the
processor.
Each software measurement loop consists of 10 consecutive
temperature samples (250 ms apart). The final measurement is the
arithmetic mean of the 10 samples values, which is stored in the
EEPROM in a 15-bit þ1 sign bit format along with the time of
measurement (encoded in Binary Coded Decimal – BDC format).
This configuration enables saving up to 13,000 temperature results
on a single 1024  8 bits EEPROM. Temperature measurements can
be recorded with any interval set between 4 s (near continuous
measurements) and 18 h.
Between temperature acquisitions, the microcontroller stays in
sleeping mode, like the temperature sensor and the EEPROM. The
RTC wakes up the microcontroller every second and a software
interval counter is incremented. Every 10 s the LED is briefly lit
(50 ms) to show activity or measurement (2 flashes).
2.5. Power budgeting and thermal power dissipation
Since the Niphargus is developed for long-term monitoring (i.e.
years), components were selected to ensure the longest autonomy
in combination with the use of a single non-rechargeable lithium
battery cell. To estimate the battery life of the logger, the current
consumption must be known and therefore a power budget must
be established. The total electric discharge (current  time) for
each mode of operation (Table 3) is obtained by multiplying each
mode current consumption by its running time in mode. Typical
current consumption values of individual components were ob-
tained from respective datasheets or directly measured on a test
circuit. The battery self-discharge current is that of a typical
LiMnO2 battery (Microchip Technology Inc., 2011a, 2011b). The
average current consumed by the sensor is the sum of the electric
charges of each mode of operation divided by the total duration of
the software measurement loop.
54 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59

Table 3
Power budget calculations for a Niphargus measurement cycle of 600 s (Microchip Technology Inc., 2011a, 2011b; Panasonic, 2002).

Measurements cycle Duration Current Electric charge (Current  Duration)

Individual device Total

Sleep
18LF1825 sleep 590.8 s 0.08 mAa 1.58 mA 0.259 mA h
Sensor OFF 0 mA
EEprom OFF 0 mA
RTCC standby 1.5 mAa
LED OFF 0 mA
Initialize (every 1 s)
18LF1825@4 MHz 3 s (0.050  60) 378 mAa 379.5 mA 0.316 mA h
Sensor OFF 0 mA
EEprom OFF 0 mA
RTCC standby 1.5 mAa
LED OFF 0 mA
Temperature sample
18LF1825@4 MHz 3.1 s 378 mAa 579.5 mA 0.499 mA h
Sensor ON 200 mAb
EEprom OFF 0 mA
RTCC standby 1.5 mAa
LED OFF 0 mA
Storage
18LF1825@4 MHz 0.050 s 378 mAa 653 mA 0.009 mA h
Sensor OFF 0 mA
EEprom ON 125 mAb
RTCC ON 150 mAb
LED OFF 0 mA
LED activity (every 10 s)
18LF1825@4 MHz 3.050 s (60  0.050 s) þ(1  0.050 s) 378 mAa 1019 mA 0.863 mA h
Sensor OFF 0 mA
EEprom OFF 0 mA
RTCC standby 1.5 mAa
LED ON 640 mAb
Battery 600 s 0.778 mAc 0.197 mA h
Self-discharge (25 °C)

a
Value from datasheet.
b
Measured values.
c
All the battery capacity and self-discharge values are given for use at 25 °C.

As an example and for a temperature measurements interval of
10 minutes, we calculate the total electric charge for one mea-
surement cycle: 2.143 mA h; the average current consumption:
12.858 mA; and finally the autonomy based on 700 mA h battery
capacity at 25 °C (battery capacity/average current): 54,440 h or
6.2 years. This exercise can be repeated for other measurements
intervals (ex: 2 years autonomy at 1 min interval, 8 years at 1 h
interval).
Note that the nominal capacity of the battery, i.e. the value
where its nominal voltage drops from 3.1 V to 2 V, is voluntary
reduced from 1400 mA h to 700 mA h where the final voltage is
about 2.8 V according to the datasheet (Panasonic, 2002). This
enhances the reliability of the calculated autonomy since it en-
sures that all individual components are still powered up at their
capacity rating (the temperature sensor has the highest minimum
operating voltage at 2.7 V).
For temperature loggers, low power consumption is also of
importance because the recorded temperature may be influenced
by heat produced by the electronics. Table 3 shows that the elec-
tric current peaks during each measurement/storage operation to
around 580 mA during 3 s. Given the 3 V setup, this equates to
1.7 mW output during 3 s. This amount of power would allow to
heat an equivalent of about 10 g of water or 40 g of air with about
0.00004 °C, which is significantly lower than the absolute accuracy
and also the theoretical resolution of the sensor. One Niphargus
weighs about 30 g (including battery), the heat produced by one
measurement will therefore not increase the temperature of the
microcontroller, or the embedded temperature sensor.
2.6. PC based “Shrimp” application
The Shrimp software provides a friendly-user communication
interface between a Niphargus datalogger and a computer. The
software runs under Windows(R) and was developed with Visual
Cþ þ 2010(R) Express. The software features an automatic detec-
tion of the virtual COM port and communication speed used by
any Niphargus plugged in a computer USB port, data downloading
to the computer (.csv file) with direct visualization of the mea-
surements on a plot and the setup of a new temperature mon-
itoring period. Several sessions of Shrimp can be run at the same
time, so as many loggers as USB ports available on the computer
can be downloaded simultaneously.
Any measurement interval can be specified between 4 s and
18 h, by 1 s steps. The software automatically calculates the date at
which the memory will be full, and display of the logger status,
including the battery voltage, current temperature reading, the
time of the RTC (that is automatically synchronized with the PC
clock).
3. Laboratory results: temperature measurements reliability
Absolute temperature calibration of 9 Niphargus loggers was
done against a thermistor (Betacurve NTC 100 kohm) of which the
resistance was measured using an Agilent 34401A multimeter. The
precision of this absolute reference is 70.1 °C for the experi-
mental temperature interval, and depends mainly on the accuracy
 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59 55

of the instrument specific parameters of the Steinhart–Hart
equation (Steinhart and Hart, 1968).
Measurements were conducted in a thermostatic cabinet at
4 different temperatures (around þ7.5 °C, þ14.5 °C, 21 °C and
30 °C). Direct measurements inside the cabinet showed that the
temperature fluctuates by about 1 °C in cycles of between 5 and
10 min. In order to buffer these short term fluctuations, the actual
measurements were done in a box of 30 by 40 by 20 cm, placed
inside the thermostatic chamber, and which was filled with in-
sulation material and a slab of granite in the most central part. In
contact with this slab a heat conducting surface (metal and silicon
pad) was placed to which the thermistor and 9 Niphargus loggers
were fixed to ensure that all sensors were at the same tempera-
ture. This setup successfully buffered the heat cycles of about 1 °C
to better than 0.05 °C.
During the calibration experiment the temperature was al-
lowed to stabilize for at least 24 h to cancel out any effects of
thermal inertia of the silica band gap detectors. Readings were
collected where the average temperature changes were below
0.05 °C per hour. The Agilent was set to record at 1 s intervals,
while the Niphargus loggers recorded at 2 min intervals.
The average differences in temperatures measured by the Ni-
phargus loggers and the thermistor are shown in Fig. 4. The
measurements deviation increases from lower to higher tem-
peratures over the calibration interval in a linear way. All readings
are well within the technical specifications of the silica band gap
detectors (∼0.5 °C) and that of the thermistor. In fact, the data are
highly systematic. Applying a linear best fit calibration reduces the
difference between the measured and the absolute reference
temperature to less than 0.03 °C. This indicates that the Niphargus
loggers can be calibrated to a precision better than 0.1 °C provided
a more accurate temperature reference is used.
3.1. Temperature inertia
The main reservation for using silicon band-gap sensors is their
slower temperature response compared to most other tempera-
ture sensors. In order to estimate the temperature responsiveness
of a complete Niphargus, several were slightly cooled down to
15 °C and allowed to equilibrate to room temperature. The draft-
free, non-heated room provided a fairly stable temperature en-
vironment at about 21 °C with temperature variations below
0.7 °C. This simple and straightforward test allowed for an easy
assessment of the required equilibration time. The test results al-
low to conclude that for temperature differences between 1 and
10 °C, 90% of the actual temperature is reached after 30–35 min
equilibration with air. In outdoor field conditions in concluded
spaces or water bodies, temperature variations usually stay well
within 1 °C per hour. Given such conditions, the test results in-
dicate a temperature equilibration that is sufficiently fast to

Fig. 4. The difference between the temperature measured by the uncalibrated Niphargus loggers and the reference thermistor. Consequently, it determines the temperature
correction to apply to each of the loggers at four different temperatures in the 5–30 °C range. The connection lines between the measured points show that the use of a linear
trend through those points will allow a temperature calibration for each logger with accuracy better than 0.1 °C.
56 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59
guarantee a precision of 0.1 °C or better. For faster response, a
thermal shunt (ex: copper plate) could be placed between the
sensor chip surface and the measured media.
4. Monitoring results in two field environments – case studies
4.1. Case study #1: temperature logging in a turbulent stream with
water from mixed origin
4.1.1. Methodology
From June 21 to July 18 2013, two Niphargus temperature
loggers were installed for continuous monitoring of the tempera-
ture of a small, turbulent stream in the vicinity of Spa, Belgium.
The stream is fed by shallow groundwater, for a large part origi-
nating from a peat bog further upstream, and from a few localized
mineralized and CO2-bearing springs. In order to measure the
water temperature, the Niphargus loggers have to be fully sub-
merged, requiring watertight containers. At the same time, they
need to be inconspicuous and robustly anchored in order to pre-
vent detection by tourists and accidental turning over by animals
or playing children.
The stream is considered as a natural laboratory and is mon-
itored on a monthly basis in order to quantify how much geogenic
CO2 is released from the wells feeding into the river, with the ul-
timate goal to understand the role of the migrating CO2 in the
overall hydrogeological cycle (Welkenhuysen et al., 2012). A con-
tinuous monitoring network can be used to construct a model of
the stream that makes it possible to determine the flow rates
through time.
In order to blend into the field setting, two larger boulders from
the river were chosen in which a 7 cm deep hole was drilled with a
diameter of 48 mm. In these holes, the Niphargus loggers where
fitted in small waterproof PVC containers. The boulders, weighing
9 and 14.5 kg, were placed with the containers facing down, in the
river on some smaller boulders, allowing for the water to run
underneath them, in direct contact with the whole Niphargus
container. The loggers were programmed to take one measure-
ment every 10 min. Each of the containers was filled with silica gel
(with colored chemical indicator) in order to prevent condensa-
tion, and mainly to check for water infiltration.
4.1.2. Discussion of the results
After nearly one month, the boulders did not show any in-
dication of being disturbed, and the loggers had successfully
Fig. 5. Plot showing the registered temperature of the two dataloggers in red and blue, that were deployed adjacent to each other. The readings deviate especially on days
with a large difference between maximum and minimum temperature. The green curve (values on right vertical axis) shows the difference in temperature between the two
loggers. The difference is closest to zero during the coolest period, and is most likely caused by the different position of the two temperature loggers in the water stream. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
registered the daily temperature cycles (Fig. 5, left scale). A 24 h
moving average is used to separate daily variations from trends
spanning multiple days. This shows that on a typical day, the
average daily temperature is reached around midnight and gently
decreases until a minimum temperature is reached at 8:00 (Cen-
tral European Time, summer time). The temperature then in-
creases more rapidly to reach a maximum at 18:00. The pattern is
asymmetric with the maximum temperature almost 1 °C above,
and the minimum less than 0.8 °C below the daily average. The
95% confidence interval shows that the largest temperature var-
iations occur around the minimum and maximum temperature
intervals. This uncertainty is dominated by the day-to-day varia-
tions, not by instrumental variations, as can be confirmed by
comparing the data from the two loggers.
The systematic readings confirm that the data loggers are
capable of correctly registering the relative temperature variations.
More detailed inspection reveals that the temperature signal re-
corded by the two loggers is not identical. When the daily maxima
and minima are reached, and when shifts to warmer or colder
weather occur, the difference in readings can reach 0.8 °C (Fig. 5,
right scale). Lab tests were conducted to confirm that the Ni-
phargus data loggers that were used respond nearly identical
during heating and cooling cycles (including tests while mounted
in the boulders). The most probable cause of those temperatures
differences are thus the positions of the measurement points in
the river bed, the closest to the shore being buffered by the river
bed.
The most critical issue proved to be the slow temperature
equilibration of the container with the water in the stream. A next
generation of containers could be designed in order to adequately
measure the temperature of the river with accuracy better than
0.1 °C. These are equipped with metal caps at one side to allow for
proper temperature equilibration with water. In addition, the
temperature loggers will each be calibrated using a reference
thermistor (as in Section 3.1), which together with the very low
drift over time, will improve the absolute accuracy.
4.2. Case study #2: spectral analysis of temperature measurements
in the Han-Sur-Lesse cave
4.2.1. Methodology
A Niphargus datalogger was installed for a period of 27 days in
the Han cave (Han-sur-Lesse, Belgium) between the 17th of May
2013 and the 13th of June 2013. The temperature was recorded
each 15 minutes which represents 2595 measurements. In Fig. 6,
 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59 57

11.26
Data: Temperature (°C)

11.22
11.18
11.14

18/05/2013 25/05/2013 01/06/2013 08/06/2013

Trend
11.24
Trend and Trend+Daily (°C)

Trend + daily
11.20
11.16

18/05/2013 25/05/2013 01/06/2013 08/06/2013

0.002
Daily (°C)

0.000
-0.002

18/05/2013 25/05/2013 01/06/2013 08/06/2013

0.02
0.01
Remainder (°C)

0.00
-0.01
-0.02

18/05/2013 25/05/2013 01/06/2013 08/06/2013

Fig. 6. Decomposition of the temperature evolution by frequency domains in the Han-sur-Lesse cave between 17/05 and 13/06/2013. (a) Recorded signal; (b) trend or multi-
day contribution to the signal (see continuous curve). Dashed line is the sum of the multi-day and diurnal contribution; (c) diurnal signal and (d) remainder or high-
frequency signal. The process is detailed in the text. The dash-dot line on each plot corresponds to the mean value. (For interpretation of the references to color in this figure,
the reader is referred to the web version of this article.)
58 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59
the resulting signal has been processed and subdivided into a very
low-frequency signal, a periodic signal, and a high-frequency re-
mainder. The function ‘fft’ of package ‘stats’ of R-language (www.
r-project.org/) was employed for both Fast and Inverse Discrete
Fourier Transform steps.
4.2.2. Results and interpretation
During the measurement period, the mean and median tem-
perature in the cave were both equal to 11.20 °C with a minimum
of 11.14 °C observed on the 31st of May and a maximum of 11.27 °C
on the 10th of June. The total temperature range thus represents
only 0.13 °C (see Fig. 6a).
Despite the small signal amplitude, clear variations, within one
day as well as over several days, are observed and a regular diurnal
fluctuation have been detected in the temperature signal. After
separation with the Fast and Inverse Discrete Fourier Transform
(fft function from ‘stat’ package of R-language), the low frequency
component (temperature changes over several days) is obtained
by filtering out all frequencies including and over 1.16e
5 Hz
(24 h) (Fig. 6b). The contribution of the diurnal periodic signal is
extracted by plotting only the 24 h frequency component. The
resulting signal is clearly a periodic signal (Fig. 6c) with very low
amplitude (∼0.0025 °C) or in other words the daily variations are
about 0.005 °C.
The contribution of daily variations to the signal is therefore
very small, but was clearly detected by Niphargus. The maximum
of the daily contribution is observed around 6 o'clock in the
afternoon, while the minimum occurs at 6 o'clock in the morning.
It must be noted that these maximum and minimum contributions
do not necessarily correspond to the maximum and minimum
temperatures of the day, which are mainly controlled by the multi-
day trend. Hence, the maximum temperature was sometimes ob-
served in the morning during periods of general temperature de-
creases, such as for instance between the 3rd and 7th of June 2013.
The contribution of the daily variation can be further evaluated by
comparing the trend curve (see continuous curve in Fig. 6b) with
the combination of the trend and daily contribution (see blue
dashed curve in Fig. 6b).
The application of a high band-pass filtering for values above
1.16e
5 Hz (24 h) isolates the remaining noise (Fig. 6d) and ex-
hibits variations of 7 0.02 °C. Interestingly, the remaining high
frequency signal (short term temperature variations þsensor
noise) amplitude is of one order of magnitude higher than the
daily contribution.
5. Conclusions
The Niphargus is a low-cost reliable, accurate and high-re-
solution temperature logger for long-term temperature monitor-
ing in humid environments. The device is based on a monolithic
silicon band-gap temperature sensor in a miniaturized package
(ADT710 and ADT7420 from Analog Devices).
This temperature sensing technology is renowned for the very
low drift of the sensor readings over time (less than 0.007 °C over
10 years) and represents a good a compromise between the sta-
bility of the RTD and the sensivity of the thermistor. In our field
cases, we estimated that the response speed of the sensor was
sufficient for environmental monitoring of slow changing pro-
cesses (∼1 °C/h). The low-drift advantage associated with silicon
diode based sensors is far more important than the moderate
temperature response speed for this kind of measurements.
The factory calibrated temperature accuracy of the ADT7410
was calibrated against a reference thermistor to improve its initial
absolute accuracy from 0.5 °C to about 0.1 °C between 5 and 30 °C.
Temperature sampling resolution is 0.0078 °C, measurement
intervals can be set anywhere from 4 s to 18 h. The total capacity of
the memory is 13,000 measurements. The battery life reaches
several years. Communication and transfer of data between the PC
and Niphargus loggers relies on USB protocol.
Spectral analysis of the temperature signal in the Han caves
demonstrated that it is possible to detect and isolate a 0.005 °C
amplitude diurnal periodic signal in the temperature curve.
The small size and low power consumption of the logger allow
it to be used in difficult accessible environments, e.g. caves and
space-constrained applications, inside a rock in a water stream. In
both cases, the loggers have proven to be reliable and sufficiently
accurate devices for our field measurements.
Aknowledgments
This work was supported by the Royal Belgian Institute of
Natural Sciences (RBINS), internal funding, www.naturalsciences.
be, and by the BiSpEem Project, financed by the Belgian Federal
Science Policy Office (Belspo), grant number: A1 project MO/36/
028, www.naturalsciences.be/geology/research/environment/bis
peem. The authors thank the Grottes de Han. S.A. and Spadel S.A.
for their kind cooperation; the Royal Observatory of Belgium and
in particular M. Ping Zhu; and M. Tim Vermeiren and M. Remy
Charavel, from the company ON Semiconductor.
Appendix A. Supplementary materials
Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.cageo.2014.10.009.
References
Ambrosetti, R., Matteoli, E., Ricci, D., 2012. Note: a versatile, stable, high resolution
readout system for RTSD and thermistor sensors. Rev. Sci. Instrum. 83, 096101.
Analog Devices, 2011. ADT74107 0.5 °C Accurate, 16-Bit Digital I2C Temperature
Sensor Technical Datasheet Revision A. 24pp.
Analog Devices, 2012. ADT7420 7 0.25 °C Accurate, 16-Bit Digital I2C Temperature
Sensor Technical Datasheet Revision 0. 24pp.
Baker, B., 1999. Application Note AN679: Temperature Sensing Technologies. Mi-
crochip Technology Inc., USA, 13pp.
Baker, B., 2004. Application Note AN685: Thermistors in Single-Supply Tempera-
ture-Sensing Circuits. Microchip Technology Inc., USA, 13pp.
Benedict, R.P., 1969. Fundamentals of Temperature, Pressure, and Flow Measure-
ments. Wiley, New York, NY, pp. 53–126.
Bible, S., De Boni, S., 2007. Microchip Technology Inc., USA, 8pp.
Dedrick, R.R., USA, D., McKinney, D., Brooks, A., 2000. An inexpensive, micro-
processor-based, data logging system. Comput. Geosci. 26, 1059–1066.
Dong-Ok, H., Yong-Il, K., Tah-Joon, P. P., Heon-Chul, P., 2008. A CMOS temperature
sensor with calibration function using band gap voltage reference. In: Pro-
ceedings of the 3rd International Conference Sensing Technology ICST 2008.
pp. 496–499.
Electrolube, 2012. Toluene Free Acrylic Coating Technical Datasheet. 3pp.
FTDI, 2013. FT230X USB to Basic UART IC Technical Datasheet v1.2, 44pp.
Hagart-Alexander, C., 2010. Temperature measurement. In: Boyes, W. (Ed.), In-
strumentation Reference Book, 4th ed. Butterworth-Heinemann/Elsevier,
Boston, pp. 269–326.
Han, J., Wang, X., Yan, T., Li, Y., Song, M., 2012. A novel method of temperature
compensation for piezoresistive microcantilever-based sensors. Rev. Sci. In-
strum. 83, 035002.
JEDEC Solid State Technology Association, 2010. Standard JESD22-A108D: Tem-
perature, Bias, and Operating Life. 14pp.
Maxim Integrated Products, 2011. DS1337 I2C Serial Real Time Clock Datasheet,
19-5770, Rev 4/11. 16pp.
Microchip Technology Inc., 2011a. PIC16(L)F1825/1829 14/20-Pin Flash Micro-
controllers with nanoWatt XLP Technology Preliminary Datasheet. 478pp.
Microchip Technology Inc., 2011b. NanoWatt XLP Battery Life Estimator v1.1.
〈http://www.microchip.com/ble〉 (last accessed 05.11.13).
Mentes, G., 2012. A new borehole wire extensometer with high accuracy and sta-
bility for observation of local geodynamic processes. Rev. Sci. Instrum. 83,
015109.
Panasonic, 2002. Manganese Dioxide Lithium Cylindrical Batteries: Individual
Specifications, 1pp.
 C. Burlet et al. / Computers & Geosciences 74 (2015) 50–59
Pfender, M., Villinger, H., 2002. Miniaturized data logger for deep sea sediment
temperature gradient measurements. Mar. Geol. 186, pp. 557–570.
Philips semiconductors, 1995. I2C-Bus and How to Use it (including specifications).
24pp.
Spevak, P., Forstner, P., 2009. Application report SLAA322: MSP430 32 kHz crystal
oscillators. Texas Instrum., 10pp.
Sridharan, M., Melear, C., 2005. Application note AN2606: practical considerations
for working with low frequency oscillators, Revision 1. Freescale Semicond., 12.
Steinhart, S., Hart, S., 1968. Calibration curves for thermistors. Deep Sea Res.
Oceanogr. Abstr. 15, 497–503.
59
Wang, G., 2005. CMOS Bandgap References and Temperature Sensors and Their
Applications. Master of Sciences in Electronic dissertation, Nankai University,
China, 174pp.
Welkenhuysen, K., Schlömer, S., Möller, I., Janssens, R. & Piessens, K., 2012. Natural
CO2 releases of geogenic origin: Cross-border comparison (Belgium–Germany)
for selected emission locations. In: Proceedings of the 4th International Geo-
logica Belgica Meeting 2012, Brussels, Belgium, 11-14/09/2012. p. 256.