Low-Cost Electronic Sensors For Environmental Research: Pitfalls and Opportunities

Main Article
Progress in Physical Geography

1–34
Low-cost electronic sensors ª The Author(s) 2020
for environmental research: Article reuse guidelines:

sagepub.com/journals-permissions
Pitfalls and opportunities DOI: 10.1177/0309133320956567

journals.sagepub.com/home/ppg
Kristofer Chan Francis T O’Shea

King’s College London, UK King’s College London, UK
Daniel N Schillereff Reagan Pearce
Andreas CW Baas Thomas EL Smith
King’s College London, UK London School of Economics and Political Science, UK
Michael A Chadwick Arnout van Soesbergen
Bruce Main Emma Tebbs
Mark Mulligan Joseph Thompson
King’s College London, UK King’s College London, UK; Lichfields, London, UK
Abstract
Repeat observations underpin our understanding of environmental processes, but financial constraints often
limit scientists’ ability to deploy dense networks of conventional commercial instrumentation. Rapid growth in
the Internet-Of-Things (IoT) and the maker movement is paving the way for low-cost electronic sensors to
transform global environmental monitoring. Accessible and inexpensive sensor construction is also fostering
exciting opportunities for citizen science and participatory research. Drawing on 6 years of developmental work
with Arduino-based open-source hardware and software, extensive laboratory and field testing, and incor-
poration of such technology into active research programmes, we outline a series of successes, failures and
lessons learned in designing and deploying environmental sensors. Six case studies are presented: a water table
depth probe, air and water quality sensors, multi-parameter weather stations, a time-sequencing lake sediment
trap, and a sonic anemometer for monitoring sand transport. Schematics, code and purchasing guidance to
reproduce our sensors are described in the paper, with detailed build instructions hosted on our King’s College
London Geography Environmental Sensors Github repository and the FreeStation project website. We show in
each case study that manual design and construction can produce research-grade scientific instrumentation
(mean bias error for calibrated sensors –0.04 to 23%) for a fraction of the conventional cost, provided rigorous,
sensor-specific calibration and field testing is conducted. In sharing our collective experiences with build-it-
yourself environmental monitoring, we intend for this paper to act as a catalyst for physical geographers and the
Corresponding author:
Daniel N Schillereff, Department of Geography, King’s College London, Bush House North East Wing, 30 Aldwych,
London WC2B 4BG, UK.
Emails: daniel.schillereff@kcl.ac.uk
2 Progress in Physical Geography XX(X)
wider environmental science community to begin incorporating low-cost sensor development into their
research activities. The capacity to deploy denser sensor networks should ultimately lead to superior envi-
ronmental monitoring at the local to global scales.
Keywords
Arduino, build-it-yourself, data-logging, environmental monitoring, FreeStation, low-cost electronic sensors,
microcontrollers, open-source, open-source hardware
I Introduction shown to hamper ground-truthing efforts (Lorenz

and Kunstmann, 2012). This trend is concerning
1.1 Hurdles to environmental monitoring
since satellite remote sensing is not without lim-
Environmental science is rooted in observation. itations, including the mismatch in spatial and
Long-term measurements of ecological, meteor- temporal scales between satellite observations
ological and hydrological variables provide the and environmental phenomena. For example, the
foundation for understanding trends, establishing coarse spatial resolution of current satellite soil-
benchmarks and informing policy (Mishra and moisture products does not adequately capture
Coulibaly, 2009; Tetzlaff et al., 2017). Such data fine-scale variability (Larson et al., 2008; Tebbs
are critical for estimating magnitudes of change et al., 2019). In addition, several parameters can-
in natural systems induced by human activity, not be directly measured from satellite remote
such as projected climate warming (Hannah sensing (e.g. sub-surface soil moisture or dis-
et al., 2011). Although the ever-increasing solved oxygen; DO). In situ measurements are
sophistication of remote sensing tools and com- therefore essential for obtaining a more complete
putational models is delivering major advances understanding of environmental processes, vali-
in environmental science (McCabe et al., 2017), dating satellite products and improving model
gaps in on-the-ground observations are con- projections (e.g. Urban et al., 2016).
straining predictive models (Urban et al., Time and financial expense are major barriers
2016). Indeed, whilst data gathering by satellites to collecting ground-based environmental data
expands, there has been a concurrent shrinking of (Muller et al., 2015; Tauro et al., 2018). Sophisti-
conventional ground-based monitoring and mea- cated instrumentation brings high maintenance
surement for many environmental systems. costs and a continual need for skilled staff. Increas-
Appetite amongst funding agencies to support ing the spatial and temporal resolution of repeat
long-term environmental monitoring networks measurements demands proportionally greater
is diminishing (Tetzlaff et al., 2017) and experi- allocation of resources. Many legally mandated
mental and field research in the hydrological international monitoring schemes however suffer
sciences, for example, are in decline (Burt and logistical constraints. The EU Water Framework
McDonnell, 2015). The global density of river Directive, for example, requires member states to
gauging stations has decreased since the 1980s measure river water quality four times per year, but
(Global Runoff Data Centre, 2018; Hannah et al., summed annual loadings are almost certainly
2011) and similar rates of closure of hydrome- underestimates with wide margins of error (Skar-
teorological stations, especially in Africa and bøvik et al., 2012). Public health concerns are pro-
Latin America (Overeem et al., 2013; World moting investment in air quality monitoring
Meteorological Organization [WMO], 2009, networks in some regions (Environmental Defense
cited in van de Giesen et al., 2014), have been Fund, 2018; United Nations Environment
Chan et al. 3
Programme, 2020) but networks of high-cost fixed black box’ and understand how programmes
sensors may lack the granularity to pinpoint emis- and equipment work. The ultimate aim of this
sion hotspots or their sources, assess the influence is customisation for specific applications. Cus-
of localised meteorology or track pollution plumes tomisability heavily depends on the degree to
(Castell et al., 2017; Morawska et al., 2018; Rai which commercial manufacturers allow their
et al., 2017; Thompson, 2016). Though modelling software and hardware be altered by external
can go some way to filling the void, models such as parties. Open-source describes an alternative
atmospheric dispersion models are computation- approach by which any interested person or
ally heavy and limited in their predictive capabil- team can contribute to software or hardware
ities (Kumar et al., 2015). development (Wu and Lin, 2001). Whilst pro-
Alternative monitoring approaches using minent freely accessible software emerged in
innovative technology are gaining momentum the 1980s (GNU) and 1990s (Linux), there has
across the environmental sciences (Bramer been a proliferation of open-source software
et al., 2018; Muller et al., 2015; Tauro et al., (OSS) and open-source hardware (OSH/
2018). The use of bioacoustic monitoring for wild- OSHW) in the last several years (Boisseau
life research and biodiversity conservation has et al., 2018). OSH consists of physical technol-
surged in popularity and is showing great promise, ogy that can be freely replicated (e.g. circuit
for example (e.g. Browning et al., 2017; Teixeira boards) or assembled using openly available
et al., 2019). Low-cost, build-it-yourself instru- drawings, schematics and/or circuit board lay-
mentation is also becoming increasingly popular outs. OSS meanwhile consists of source code or
(Kumar et al., 2015; Mao et al., 2019; Mickley code-snippets, again, publicly available.
et al., 2019), enabling much finer-scale environ- The rise of OSH has been, to a considerable
mental data to be collected both spatially and tem- extent, attributable to the rise of the Arduino and
porally (Horsburgh et al., 2019). Low-cost sensor Raspberry Pi hardware and software platforms.
networks have been shown to improve the spatial Ferdoush and Li (2014) provide an overview of
coverage of ground-truthing data for validating environmental monitoring applications using
satellite products (Tebbs et al., 2019) and some Raspberry Pi microcomputers (www.raspberry
large-scale hydrometeorological monitoring net- pi.org). Our efforts focus predominantly on
works have been launched, including the FreeSta- Arduino boards, a brand of open-source micro-
tion initiative (www.freestation.org) and the controllers that are used to assemble environ-
Trans-African HydroMeteorological Observatory mental sensors and data loggers, alongside a
(www.tahmo.org; van de Giesen et al., 2014). plethora of other applications (see www.ardui
Alongside these research avenues, open-source no.cc for a sample of practical applications).
development communities such as the Gathering Commonly referred to as I/O devices due to
for Open Science Hardware organisation (GOSH) their ability to simultaneously act as input
continue to grow in prominence (Boisseau et al., devices (i.e. receiving, detecting or measuring
2018). This paper is therefore timely and will serve electronic signals or voltage levels) and output
to further encourage the expansion of low-cost devices (i.e. sending electronic signals or vary-
monitoring into environmental research. ing output voltage levels), microcontrollers
typically consist of the components outlined in
Figure 1.
1.2 Open-source hardware and the Arduino Though the development of microcontrollers
platform started many years prior to the emergence of
The open-source movement developed in Arduino (notable precursors include PIC and
response to the desire for users to ‘break the Parallax microcontrollers), the development of
Component Description and function

Ports to interface with a computer via a USB to serial (TTL) adapter.
This is both for uploading (or ‘flashing’) the program and
Programming Header
communicating results to the computer on the serial monitor of
Arduino IDE.
Embedded chip comprising the processing unit and hardware
Microprocessor
controller.
Connection for the positive terminal of an unregulated power supply
RAW
(e.g. battery pack).
Regulated 3.3V output to supplies up to 200mA (depending on model)
3V3
to periphery devices via an onboard voltage regulator.
Analogue Channels Measure voltage outputs from peripheral analogue devices
Digital I/O Channels Pins for interfacing with peripherical devices
Dedicated Channels
Pins with additional features for connecting periphery devices
(below)
A4, A5 SDA and SCL pins for I2C communication with peripheral devices
MOSI, MISO, and CLK data pins for Serial Peripheral Interface (SPI)
11, 12, 13 communication with peripheral devices such as a microSD card or
real-time clock
Transmit and receive (Tx, Rx) pins for Universal Asynchronous
0, 1
Receive/Transmit (UART) serial communication
Interrupt pins (IN0, IN1) programmable to identify if a change of
2, 3 voltage is detected, even when the microprocessor is in a low power
(‘sleeping’) state
Figure 1. Pro Mini anatomy, detailing the components and port functions common to many microcon-
trollers. Note: port numbers are specific to the ATmega328P microprocessor.
Chan et al. 5
Wiring and Arduino was pivotal in transferring impetus for each project varied: some were
microcontroller programming from the hands of geared towards research advancements while
specialised engineers to wider audiences. Prior others were initially developed as teaching
to Arduino, most microcontroller prototyping activities. In each case, numerous unforeseen
tools were prohibitively expensive and required challenges have helped us develop smooth,
steep learning curves (D’Ausilio, 2012; Kusher, effective workflows and reliable, research-
2011). Under the supervision of professors from grade data are now rapidly emerging. The fol-
the Interaction Design Institute Ivrea, Wiring lowing sections explain the core components of
(the predecessor of the Arduino programming an Arduino environmental data logger before
language) was developed as a simplified coding presenting six case studies: (a) a water table
language for programming Atmel microproces- depth probe; (b) an air quality monitor; (c) an
sors. Thus, the combination of the simplified, aquatic water quality multiprobe; (d) a customi-
OSS (the Arduino Integrated Development sable multi-parameter weather station; (e) a
Environment – hereafter, Arduino IDE) and time-sequencing sediment trap and (d) a sonic
widespread availability of low-cost hardware anemometer for monitoring wind-blown sand
(Arduino boards – based on Atmel’s ATmega dynamics. Lastly, we highlight frequently
processors – in addition to other OSH microcon- encountered pitfalls and outline best practice
trollers programmed through the Arduino IDE, methods to maximise success rates for research-
for example, the Adafruit Feather, ESP8266, ers new to build-it-yourself hardware construc-
RedBear and Particle Photons or Electrons) led tion and programming.
to the widespread adoption of microprocessors We intend this paper to act as a transformative
by hobbyists and the public (Furber, 2017). platform and the key reference for physical geo-
Through the release of tutorials, troubleshooting graphers (and environmental scientists more
forums and continual software and hardware broadly) looking to embed low-cost environmen-
development, the open-source nature of Ardu- tal monitoring into their research activities. Full
ino and other OSH microcontrollers has dimin- schematics, code, component costs and purchas-
ished the learning curve and created a growing ing guidance to reproduce our designs are avail-
user community of beginners and experts. The able on our KCL Geography Environmental
non-technical and inexpensive characteristics of Sensors Github repository (https://github.com/
OSH such as Arduino has also transformed citi- KCLGeography/environmental-monitoring) and
zen science (research conducted by amateur FreeStation project website (www.freestatio
scientists; Stevens et al., 2014) and presents n.org). Costs reported in this paper are based on
opportunities for innovative teaching and edu- purchases made by the KCL Geography Envi-
cation (e.g. Giocomassi Luciano et al., 2019; ronmental Sensor group during the 2019/20
Lim et al., 2020; Plaza et al., 2018). financial year. In our view, careful deployment
of open-source, low-cost environmental sensors
will deliver step-change improvements to global
1.3 Purpose of the paper environmental monitoring and management.
This paper aims to catalyse and accelerate the
deployment of low-cost sensors for environ- II Environmental sensors
mental research. It draws on 6 years’ experience
developing, testing and deploying a range of
2.1 Background to Arduino open-source
Arduino-based, low-cost environmental sensors hardware and software
by staff and students of the Department of Geo- Arduino is a brand of microcontroller boards
graphy at King’s College London (KCL). The with their own processor and memory that uses
Table 1. Key components for an environmental data logger. Unit cost ranges reflect purchases made by the
KCL Geography Environmental Sensor group during the 2019/20 financial year. Low costs are the single unit
price from bulk purchases (>5 units) from suppliers in Asia, whilst the upper range indicates UK-sourced
purchases. Note that connecting wires and other consumables (e.g. solder) are not included below.
Unit cost
range
Component Function Version (£ GBP)
Arduino board Microcontroller to interface sensors, Pro Mini or Nano 1.60–6.00
clock and memory
Real-time clock High-accuracy timekeeping DS3234 4.00–6.00
MicroSD card Portable data storage Class 4, 1–8 GB 0.40–3.00
MicroSD card shield To interface with the microSD card Generic SD card breakout shield 0.40–1.00
Battery holder and Power supply Safest with unidirectional 1.50–4.00
connector connector (e.g. Futaba or JST)
Breadboard or Breadboard for prototyping or 400 point 1.00–5.00
solderboard solderless circuits. Soldered
breadboards (solderboards)
recommended for improved
robustness with longer deployment
code to control and communicate with electronic 2.2 Water table depth probes
devices (Karvinen and Karvinen, 2011). The Tropical peatlands are one of the most carbon
Arduino programming language is based on
(C) dense ecosystems in the world, storing 3%
C/Cþþ that is accessed via a user-friendly IDE.
of global soil carbon on 0.25% of the total land
An enormous online community provides techni-
area. Peatlands in Southeast Asia have been
cal support as well as an extensive list of existing
widely deforested and degraded over the past
libraries (collections of code that provide bespoke
few decades, mainly by employing fire and drai-
functionality for individual sensors). Despite
nage, resulting in their conversion to a net car-
their versatility, microcontrollers are on their own
bon source (Evers et al., 2017). These
inadequate for formal scientific environmental
disturbances result in loss of vegetation struc-
monitoring. Core components required for most
sensors are described in Table 1. Peripherals such ture and enhance peat oxidative decomposition.
as microSD card shields and real-time clocks can To assess the impact of land-use change on peat-
be connected via breadboards and jumper wires land CO2 emissions (e.g. through gas chamber
(soldered or unsoldered), while bespoke printed experiments), flux measurements must be sup-
circuit boards (PCBs) can be used to simplify plemented with water table and soil temperature
construction and minimise connectivity issues monitoring. An ongoing CO2 and CH4 flux
such as loose wiring or poor soldering. Compo- monitoring site in the tropical peatlands of
nents can be easily acquired from UK or overseas Belait District, Brunei, consists of 10 sampling
sellers. Costs (Table 1) are usually lower from sites with bored water table monitoring wells.
suppliers in China, for example, at the expense Conventional water depth sensors (e.g. Van
of extended delivery times and delays due to cus- Essen Diver) are relatively expensive at approx-
toms duties. UK, European or US sources can be imately £600 (GBP) per unit (Table 2). To boost
an order of magnitude higher. monitoring capacity, our group developed a
Chan et al. 7
Table 2. Cost and accuracy comparisons between two commercial water depth loggers and the second
iteration of the low-cost sensor used in this case study. Operating conditions and technical specifications are
those reported on manufacturer datasheets. Unit costs are accurate as of April 2020.
Approx. unit
cost Range Accuracy at Resolution Stability Response
Sensor/instrument (£ GBP) (bar) 25 C (mbar) (mbar) (mbar/year) time (s)
Designed water table logger 40 (16) 0.1–1.1 +1.5 0.13 +1 0.8
(sensor: MS 5803-02ba)
Van Essen DI601 600 0–1 +1.0 0.20 NR 0.5
Hobo u20-001-04 600 0–1.45 +0.35 0.14 NR <1.0
Note: NR ¼ not reported.
low-cost water table depth sensor (*£40) that

can be easily integrated with water and/or soil
temperature sensors.
Our first model combined a differential pres-
sure sensor (NXP MPX5010DP) with an
analogue-to-digital converter (ADC) and the
standard Arduino logging components outlined
in Table 1, housed inside a weatherproof junc-
tion box. The differential pressure sensor has
two tube connections: one is submerged in the
well, sensing water head pressure and atmo-
spheric pressure, and the other is exposed to
atmospheric pressure only. There were a num-
ber of advantages to this setup: only the tube Figure 2. A set of water table depth loggers
needed to be submerged so all electronics were designed following the second, more successful
approach using a single pressure sensor (Note: Li-ion
housed above ground; the differential nature of battery not yet connected). See text for details.
the measurement provides high-precision data;
and there is no need for separate measurements
of atmospheric pressure. The major disadvan- with screw-threaded caps at each end, which sits
tage was that leaving the above-ground tube inside the well. The space constraints of the tube
exposed for atmospheric pressure measure- necessitated direct-soldering of short wires onto
ments made it overly susceptible to humidity components and the small lithium ion battery
and insects, resulting in failure or heavy degra- (there was insufficient space for a battery
dation of all units. holder). This approach meant atmospheric pres-
Our second, more successful approach used a sure had to be measured separately and assem-
single pressure sensor (TE Connectivity bly was more challenging. The sensor is delicate
MS5803-02ba). Here we use a 2-bar sensor and must be directly exposed to water pressure
(up to *10 m water depth; variants of the while electronic connections are kept water-
MS5803 module are available for either deeper, proof. Our solution was to attach a small seg-
or better precision for shallower, deployments). ment of rubber tubing to the protruding pressure
The sensor and Arduino components are housed sensor; this tubing was fed through a drilled hole
inside a waterproof aluminium tube (Figure 2) in one end of the tube housing. An epoxy
Figure 3. Comparison of water head pressure measured by the Arduino-based sensor (£16; grey and black
lines) and a commercial Van Essen Diver (£600; orange line). The Arduino sensor shows strong performance
(black line; RMSE ¼ 1.88 cm; Table 3) after applying a linear calibration (y ¼ 0.7873x þ 280) to the raw
measurements (grey line). Both loggers were deployed in the same well for 2 months. Values have not been
corrected for atmospheric pressure or water temperature.
Table 3. Assessment of Arduino-based sensor performance against the commercial Van Essen Diver using
data from a 2-month deployment in a peatland monitoring well in Brunei. These estimates use the calibrated
measurements (see Figure 3).
Parameter r2 Slope RMSE (cm) MAE (cm) MBE (%)
Water head pressure 0.94 1.02 1.88 1.46 –7.11
Note: r2 ¼ coefficient of determination; RMSE ¼ root mean square error; MAE ¼ mean absolute error; MBE ¼ mean bias
error.
coating was applied to the tube cap, fixing the of determination (r2), root mean square error
sensor and rubber tubing in place and isolating (RMSE; cm), mean absolute error – a good mea-
the exposed sensor from the Arduino micropro- sure of average performance – and mean bias
cessor and core peripherals. Exposed connec- error (MBE), which provides a useful evalua-
tions required taping to minimise the risk of tion of bias dimension (Figure 3; Table 3; Con-
electrical shorting, and fitting the direct- cas et al., 2020). As a further test, data presented
soldered wires into the tube also introduced a in Figure 3 have not been corrected for atmo-
risk of ripped wires. Thread seal tape was used spheric pressure or water temperature (how
on the screw threads for the housing to ensure comparative commercial sensors also function).
waterproofing. The Arduino-based sensor captured diurnal var-
An Arduino-based pressure sensor was iations in pressure as well as warming and cool-
deployed alongside a Van Essen Diver for a ing of the water column to a high degree
2-month period at one of the 10 wells in Brunei. of sensitivity, but with a systematic offset of
Sensor performance was assessed by coefficient 262 cm. After applying the linear calibration,
Chan et al. 9
the Arduino-based sensor showed marginal bias 5003 sensor (Table 4), an optical laser-scattering
(MBE ¼ –7.11%) and a low RMSE of 1.88 cm, sensor that achieves high accuracy (98%
well within the necessary accuracy for peat counting efficiency of PM 0.5 mm; Plantower,
GHG flux estimates (Lupascu et al., 2020) or 2016). Its rapid measurement response time
multi-year water table monitoring (Mew et al., (10 s) also allows reliable measurements to
1997). be made in transit. Tests of cheaper Sharp
GP2Y1010AU0F sensors showed inferior per-
formance and the need for continuous calibra-
2.3 Air quality loggers tion – a difficult undertaking for particulate
The pervasive threat of poor air quality in urban matter requiring equipment in the order of tens
settings and its acute physiological effects are of thousands of pounds. Electrochemical gas
becoming clear (Atkinson et al., 2016; Kelly sensors manufactured by Alphasense have been
and Fussell, 2019; Sundell et al., 2011). Gas- used to measure NO, NO2 and O3, with reported
eous and particulate emissions have been linked accuracies below 1, 0.5 and 0.5 ppm, respec-
to greater risk of child obesity (Kim et al., tively (Alphasense, 2017). We adapted two
2018), adverse effects on foetal growth (Smith designs: one incorporates a pump, air flow cir-
et al., 2017), decreased educational perfor- cuitry and filter system that keeps separate the
mance (Mendell et al., 2013; Wargocki et al., PMS-5003 inlet and outlet before removing
2020) and more frequent incidences of dementia particulates prior to entering the gas chamber
in London (Carey et al., 2018), for example. In (Figure 4). The second, simpler configuration
urban areas, the particulate component is pre- fixes the particulate matter and gas sensor inlets
dominantly derived from vehicles through com- and outlets to separate holes drilled through the
bustion emission, braking and tyre abrasion as housing (Figure 5). The Alphasense sensor out-
well as domestic wood burners, while the gas- put is between 300 and 400 mV so a 16-bit ADC
eous fraction is primarily released during fossil was used (the ADS1115 module), which has far
fuel combustion (Vicente et al., 2015). Urban superior voltage range and resolution compared
air quality is typically monitored using fixed, to ATmega328P’s 10-bit ADC (65,536 individ-
ground-based stations. Costs of such configura- ual detectable levels compared to 1024 respec-
tions run to many thousands of pounds per tively). We also ensured sampling intervals
instrument (Mead et al., 2013) and stationary were programmed to mimic the instrument’s
infrastructure is less suitable for pinpointing inhalation and exhalation cycle. We usually
emission point-sources and assessing personal mount a Bosch BME280 that measures tem-
exposure and localised risks. Low-cost air pol- perature, relative humidity and barometric
lution sensors offer valuable granularity and pressure or a BME680 (which also samples
portability with initial studies showing promise volatile organic compounds) alongside the
(Bulot et al., 2019; Mead et al., 2013; Munir PMS-5003.
et al., 2019; Piedrahita et al., 2014). We have We assessed the performance of the Arduino-
developed Arduino-based sensors to measure based air quality sensors against the LAQN flag-
particulate matter (PM1, PM2.5 and PM10) and ship kerbside monitoring station on Marylebone
trace gases (NO2, O 3 and NO) for *£410, Road, a major arterial route through west Lon-
which show good performance when tested don. Our sensors were positioned on its roof
against the London Air Quality Network (2 m above ground level) for 5 days in February
(LAQN). 2018. The Marylebone LAQN station measures
For particulate matter, we determined the PM2.5 on a Filter Dynamics Measurement Sys-
most effective sensor to be the Plantower PMS- tem and NO 2 via chemiluminescence. The
10
Table 4. Cost and accuracy comparisons for a range of commercial air quality loggers and low-cost sensors. Operating conditions and technical
specifications are those reported by manufacturer datasheets. Unit costs are accurate for UK online sellers as of April 2020. Unit cost prices for
the Plantower and Sharp are based on those paid by the KCL Geography Environmental Sensor group during the 2019/20 financial year.
Measurement
Unit cost Temperature Reported accuracy range
PM10 & PM2.5 instrument/sensor (£ GBP) range ( C) (mg m-3) (mg m-3) Resolution Response time (s)
-3
Plantower PMS-5003 (sensor only) 13–20 –10 to 60 +10 0–500 1 mg m <1
Sharp GP2Y1010AU0F (sensor 4–11 –10 to 65 Uncalibrated results 0–600 0.5+0.15 V per NR
only) 100 mgm-3
Sidepak AM520 3500 –40 to 85 0.2 1–100,000 1 mg m-3 1
AQMesh Combo 5500 –20 to 40 20 0–500 NR NR
Grimm 11-D 11,900 4 to 40 +3% 0–100,000 NR 6
Unit cost Temperature Typical reported accuracy Measurement Resolution Response time (s)
£ GBP)
( range ( C) range
NO2 instrument/sensor (ppb)
Alphasense NO2-A43F 300 –30 to 40 0.5 ppm (as low as 0.1 0–20,000 NR <80
(sensor only) in ideal conditions)
AQMesh Combo 5500 –20 to 40 +0.01 ppm 0–4000 NR NR
Teledyne T200 14,200 5 to 40 0.5% of readings >50ppb 0–20,000 NR <80 to 95%
Note: NR ¼ not reported.
Chan et al. 11
Plantower PMS3005 and Alphasense NO2-

A42F showed strong performance overall (Fig-
ure 6, Table 5). Our sensor returned higher
maximum PM2.5 readings and the RMSE indi-
cated the low-cost sensors appeared to be mar-
ginally oversensitive when measuring the
highest concentrations of PM 2.5 , with the
exception of an over-estimated peak in PM2.5
on 4 February. Mean concentrations were sim-
ilar for PM 2.5 and NO 2 and the low MBE
(Table 5) indicated limited systematic over-
or underestimation, however. Importantly, the
low-cost sensors accurately captured the tem-
poral picture, including rush-hour peaks and
variation between weekday and weekend traf-
fic density (Figure 6).
Figure 4. Air quality sensor array comprising
a Plantower PMS5003 and a set of Alphasense gas The growing commercial market for personal
sensors. This configuration encompasses a EPA filter air quality monitors has raised concerns about
and pump for particularly dusty environments, or data quality (Lewis and Edwards, 2016). Our
where faster reading stabilisation is required (e.g. bias estimates compare favourably to perfor-
handheld applications). mance testing of a Plantower PMS5003 by Bulot
Figure 5. Air quality sensor array comprising a Plantower PMS-5003 and a set of Alphasense chemical gas
sensors. This second design is simpler but assumes deployment at a fixed location.
Figure 6. Data comparison of (a) PM2.5 and (b) NO2 measurements made by an Arduino-based sensor (black
lines) and the London Air Quality Network’s Marylebone monitoring station over a 5-day period in February
2018. The Arduino-based sensor sits on top of the station to ensure measurements are made at similar
heights.
Table 5. Assessment of low-cost sensor performance based on a 5-day deployment at the LAQN flagship
Marylebone monitoring station.
Difference in means RMSE MAE
Parameter Instrument r2 Slope (mg m-3) (mg m-3) (mg m-3) MBE (%)
PM2.5 Plantower PMS5003 0.75 0.56 þ1.23 4.29 3.14 23.00
NO2 Alphasense NO2-A42F 0.88 0.88 þ1.06 12.63 9.20 –19.69
error.
et al. (2019; RMSE for PM2.5 of 6.5–7 mg m-3) concentration) on the market; these should evi-
and Sayahi et al. (2019; RMSE for PM2.5 of 5.5– dently not be clumped into a homogenous class
9.6 mg m-3). Although other researchers (e.g. of ‘low-cost sensors’. The Plantower performs
Feenstra et al., 2019; Johnson et al., 2018) report well in most studies although Kelly et al.
mixed performance between different models of (2017) showed deterioration in accuracy when
low-cost air quality sensor. There are numerous particulate concentrations exceeded 40 mg m-3.
models for measuring a specific environmental This underscores the importance of recognising
parameter (e.g. airborne particulate sensor choice should be guided by deployment
Chan et al. 13
Table 6. Cost and accuracy comparisons between a commercial DO logger and the low-cost Atlas probe
used in this case study. Operating conditions and technical specifications are those reported on manufacturer
datasheets. Unit costs are accurate as of April 2020.
Unit cost Temperature Range Reported accuracy Resolution Response time
Instrument/sensor (£ GBP) range ( C) (mg/L) (mg/L) (mg/L) (mg/L/s)
Atlas Scientific DO kit 270 1 to 60 0–100 +0.05 NR 0.3
HOBO U26-001 1600 –5 to 40 0–30 0.2–0.5 0.02 NR
Note: DO ¼ dissolved oxygen; NR ¼ not reported.
setting coupled with comprehensive field testing limited availability of real-time, on-the-ground
and sensor-specific calibration. Meteorological data (Behmel et al., 2016). Alongside warnings
factors, especially temperature and relative of ‘data-rich but information-poor’ scenarios
humidity, can also have large effects on sensor around water quality monitoring networks
readings (e.g. Feenstra et al. 2019; Jayaratne (Ward et al., 1986), the temporal and spatial
et al., 2018). We posit that build-it-yourself sen- scale of water quality testing is largely deter-
sors can help overcome such limitations by mined by finance and logistics, particularly due
increasing capacity for paired deployments of air to the expense of commercially available mon-
quality and meteorological sensors rather than itoring systems. Here we present our efforts to
depending on, for example, regional weather sta- develop an Arduino-based multi-parameter
tion data (e.g. Bulot et al., 2019). probe for water quality monitoring.
Our sensors produce promising research- Following global monitoring efforts (World
grade data and are enabling important research Health Organization [WHO], 1996; WHO,
questions to be explored at the individual or 2004), we chose to focus on temperature, con-
community level. For example, one deployment ductivity, and DO due to overall cost and like-
showed the installation of an ivy green screen at lihood of producing accurate readings (Wagner
a primary school in central London decreased et al., 2006). Temperature influences most
NO2 concentrations during peak traffic conges- water quality parameters (WHO, 1996). Not
tion by 35%, whilst another confirmed that only do temperatures in water bodies vary over
choosing an optimal form of public transport 24 h, but their daily averages change throughout
to minimise personal exposure in London pre- the year (Brümmer et al., 2003). DO, an indicator
sents a predicament: particulate matter was of aquatic biological health, is related to the photo-
higher when walking, cycling, or on the Under- synthetic and metabolic activity of aquatic organ-
ground, but time inside buses and cabs isms. Given DO is affected by temperature and
increased exposure to NOX.
there are noticeable diurnal and seasonal varia-
tions, temperature and DO are monitored simul-
2.4 Water quality loggers taneously (Kannel et al., 2007). We chose the
Threats to water quality and aquatic biodiversity Atlas Scientific DO kit, as galvanic cell-type sen-
from human activities are a global issue. sors have short response times and appropriate
Despite widespread acknowledgement that pol- robustness for outdoor deployments (Wei et al.,
lution is a major threat to the sustainable man- 2019) at a cost of £270 (Atlas Scientific, 2019;
agement of aquatic environments (Rockström Table 6). The associated shield allowed more
et al., 2014; Vorosmarty et al., 2010), local- and straightforward calibration and programming
regional-scale initiatives are constrained by the because it directly calculates actual DO values
effectively (r 2 ¼ 0.97, RSME ¼ 0.29 C;

Table 7). In situ DO measurements showed
satisfactory correlation with the Hobo logger
(r2 ¼ 0.87) but an offset is evident, with the
Hobo logger giving readings *20% higher.
Individual sensor calibration of the thermistors
was straightforward but acquiring accurate DO
readings was more complicated. It is difficult to
determine whether the lower mean readings are
associated with the inbuilt Atlas Scientific cali-
bration, electrical interference introduced by sen-
sor integration or issues with the data transfer
(Siragusa and Galton, 2000). Nevertheless,
applying a post-deployment linear calibration
function leads to excellent accuracy, with an
Figure 7. Deployment configuration of the water
quality multiprobe. MBE well below 1% (Figure 8; Table 7).
Arduino-based DO sensors should produce reli-
able readings provided an initial site-specific
from the voltage reading. Conductivity is a com-
calibration is performed. Conductivity measure-
monly measured water quality parameter
ments were consistently divergent, leading us to
(Wagner et al., 2006) and long-term monitoring
suspect issues with probe accuracy. Data are not
can be useful for tracking pollution sources
(Morrison et al., 2001). We used the DFRobot reported here and difficulties with conductivity
electrical conductivity probe and shield (£60) to probe calibration persist despite extensive lab
achieve an optimal balance between cost and testing.
accuracy. The glass design protects the sensitive A major pitfall in the construction of these
electrode, providing additional durability probes was underestimating the time requirements
(DFRobot, 2017). for troubleshooting. Non-compatibility between
The River Brent in London (UK) has a long sensors, especially during attempts to eliminate
history of poor water quality, and river restoration electrical interference, was an unexpected but
efforts are ongoing (Thames21, 2019). We major technical challenge. Achieving a watertight
deployed Arduino-based loggers (Figure 7) in enclosure was a foreseeable challenge but produc-
two locations along the River Brent – a river ing a design using ‘low-cost’ materials with ade-
restoration and an unrestored site – for 1 week quate ruggedness for aquatic deployment was
in February 2018 (Lavelle et al., 2019). We also enormously time-consuming. The number of pro-
deployed a commercial logger (HOBO U26-001) totype probes gives some insight into the time
measuring temperature and DO at the unrestored commitment. Four models were iteratively pro-
site to facilitate performance evaluation. All log- duced, each the result of six documented field tests
gers were placed in the middle of the river on and numerous laboratory trials. We have devel-
wooden stakes, hammered 15-cm deep into the oped two fully functional multiprobes but five oth-
riverbed and protected with rocks for security. ers were tested and failed due to calibration
The Arduino-based DO and temperature inaccuracy, water intrusion, faulty parts or short-
time-series closely follow diurnal fluctuations circuiting. Repeated replacement of parts and cali-
as measured by the Hobo logger (Figure 8). bration chemicals is a hidden labour and financial
Temperature was measured particularly cost.
Chan et al. 15
Figure 8. Comparison of (a) dissolved oxygen and (b) water temperature measurements at the unrestored
and restored sites. Data were measured 01–06 February 2018 by an Arduino-based sensor (£380; black and
grey lines) at both sites and a HOBO U26-001 (£1,600; orange lines) at the unrestored site to evaluate
performance. The Arduino-based DO sensor deployed at the unrestored site shows strong performance
(black line; RMSE ¼ 0.31 mg L-1; Table 7) after applying a linear calibration (y ¼ 1.184x þ 0.256) to the raw
measurements (light grey line).
Table 7. Assessment of Arduino-based sensor performance against the commercial HOBO U26-001 using
data from a 7-day deployment in an urban stream in London, UK. These estimates use the calibrated
measurements (see Figure 3).
Parameter Units r2 Slope RMSE MAE MBE (%)

Temperature C 0.97 0.98 0.29 0.26 –25.0
Calibrated DO mg L-1 0.87 0.87 0.31 0.21 –0.04
error; DO ¼ dissolved oxygen.
Reproducing our functional model does rep- phosphorous, nitrate, colour or chlorophyll sen-
resent an economically competitive alternative to sors to monitor eutrophication in freshwater sys-
commercial equipment, with scope for further tems (Ferreira et al., 2013). In situ water quality
improvements to the external housing. One of monitoring is complicated because a complete
the most exciting aspects is that the technology and precise assessment cannot be reached unless
allows for multiprobes to be customised for spe- several interacting parameters are measured
cific studies, such as the inclusion of nitrogen, simultaneously. While careful calibration will
Table 8. Cost and accuracy comparisons between two commercial weather sensors and the low-cost Bosch
BME280 used in several of our case studies. Operating conditions and technical specifications are those
reported on manufacturer datasheets. Unit costs are accurate as of April 2020.
T RH
reported reported Resolution
Approximate Temperature accuracy accuracy (T ¼ C, Drift Response
Instrument/sensor unit cost (£) range ( C) ( C) (%) RH ¼ %) (%/year) time (s)
Bosch BME280 3.70 –40 to 85 +1 +3 0.008% 0.5 1
(sensor only) 0.01 C
Kestrel 3000 145 –10 to 55 +1 +3 0.1% 0.1 C 1 1
Campbell Scientific 199 –40 to 60 +0.6 +3–7 NR NR 1
HMP60 (sensor
only)
FreeStation Meso 120 –40 to 85 +1 +3 0.008% 0.5 1
Automatic 0.01 C
Weather Station
Davis Vantage 950 –40 to 65 +0.3 +2 1% 0.1 C <0.25 NR
Pro2 Plus
Automatic
Weather Station
Note: T ¼ temperature; RH ¼ relative humidity; NR ¼ not reported.
be required for each sensor added to a multip- surface energy, water balances and horizontal
robe, low-cost, continuous water quality loggers fluxes of air. Automatic weather stations can
offer a valuable method for broadening the den- measure sub-hourly but usually aggregate data
sity of routine monitoring. These networks could to hourly or daily averages or totals. The cost of
go a long way to establishing long-term records commercial weathers stations increases with the
of baseline conditions and identifying specific number of measurable variables (Table 8). Mul-
environmental pollution sources. tivariate stations can be priced in the thousands
of pounds before specialist installation and
maintenance is factored in.
2.5 Automated weather stations Grid-connected weather stations are particu-
Meteorological data are fundamental to cli- larly sparse in low-income countries and their
matic, hydrological, ecological and geomorpho- state of maintenance can be poor (WMO, 2019).
logical research. Multivariable weather stations Climate change is likely to have disproportio-
are the standard system for monitoring meteor- nately large impacts in these regions so long-
ology, with >47,000 locations globally offi- itudinal data collection at local scales is
cially recording precipitation and >24,000 critical. The FreeStation project (http://
recording mean monthly temperature (Hijmans www.freestation.org/) is working to redress this
et al., 2005), though many more unofficial issue by expanding meteorological monitoring
(amateur) weather stations now exist. A weather capacity using open-source and low-cost
station normally measures air temperature, instrumentation. Since 2014, FreeStation has
atmospheric humidity and pressure, precipita- developed open-source designs for a range of
tion, solar radiation, and wind speed and direc- low-cost instrumentation and loggers. These
tion. These variables allow an assessment of include standalone and web-connected
Chan et al. 17
Figure 9. (a) The FreeStation Meso Automatic Weather Station (AWS) and (b) the FreeStation MesoLive
AWS.
automatic weather stations (AWS) based on well as by schools, NGOs and some govern-
Arduino and Particle microprocessors. FreeSta- mental authorities.
tion AWS have a component cost 3–13% the As our most established research programme
cost of a commercial station and require 2–4 h (since 2014), FreeStation sheds valuable light
of unskilled labour to build using the detailed on long-term sensor robustness and performance.
build instructions at www.freestation.org/ Promisingly, there have been zero sensor failures
building. The stations are designed to be easily in field deployment. Occasional data loss has
built from accessible components as well as occurred due to faulty SD cards, loss of power
accurate, robust and easy to transport and install. or external interference from animals (rabbits in
The FreeStation Meso station includes precipi- the UK; crocodiles in Burkina Faso!) or extreme
tation, temperature, humidity, pressure, wind weather. This shows comparable or superior per-
speed and direction and solar radiation formance to commercial data loggers, which
(Figure 9(a)). It reads instruments every 10 min report failure rates of 7–27% (Mickley et al.,
and writes hourly summaries to an on-board 2018). Moreover, failure in a commercial device
microSD card. The Meso can use an Arduino Pro is usually permanent because they are shipped as
Mini, a Particle Photon, or RedBear micropro- a sealed product. This means one faulty internal
cessor. The MesoLive (Figure 9(b)) has the same part can render the device inoperable, whereas
instrumentation on a smaller footprint with cel- individual components can be easily and cheaply
lular connectivity and access to data via a simple replaced in low-cost designs. This is an enor-
web application programming interface (API). mous benefit for the unimpeded collection of
More than 219 stations are currently collecting long-term time-series data. This also minimises
data at 43 sites in 15 countries and the design issues around sensor drift, for example. We have
has evolved significantly over time, guided by observed sensor degradation in very humid
deployments in a range of environments. Free- environments, such as cloud forests, but swap-
Stations are currently in use by research proj- ping the meteorological sensors on an annual or
ects in deserts, temperate and tropical forests as semi-annual basis has avoided this issue.
Figure 10. An example of the live data stream from a UK FreeStation weather station, including forecast
data (FreeStation, 2019).
FreeStations are built around the FreeStation deployments. Data streams are managed
PCBs and FreeStation firmware, which allow through a web platform and API (Figure 10),
‘plug and play’ connectivity of a variety of sen- which is capable of quality control, combining
sors through standard RJ45 and RJ12 cables incoming data streams with forecasts, early
(commonly known as ethernet and phone warning and direct connection to web-based
cables). FreeStations are designed to be build- modelling and policy support tools such as
able by students without prior electronics WaterWorld and Eco: Actuary (www.policysup
knowledge or interest in microprocessors, and port.org). This kind of integration of real-time
we work with students, extension workers and data streams with web-based models has signif-
government technicians to develop local capac- icant potential in environmental forecasting and
ity. Shipping build materials and components management.
overseas has been a challenge, however. Most The FreeStation project has now moved
FreeStation components are sourced from the beyond weather monitoring. As part of the Path-
web and direct postage has been problematic, ways out of Poverty for Reservoir-dependent
with lengthy delays at customs. Bringing parts Communities in Burkina Faso (POP-BF) project
and stations from the UK as personal baggage (www.sites.google.com/view/pop-bf), for
during research visits is easier but is not a long- example, a range of FreeStations have been
term option. The power and programmable installed that monitor local weather, water lev-
memory of the Arduino platform has also els in reservoirs using sonar, and soil moisture.
imposed technical constraints on integrating These stations are connected to the WaterWorld
multiple meteorological sensors and managing policy support system to deliver nowcasts and
the data streams emerging from multiple short-term forecasts (communicated via on-
Chan et al. 19
board switches and lights) on reservoir volume

and soil moisture to advise irrigation and har-
vest planning. The simplicity of the technology
and output has created a locally owned reservoir
monitoring system that will continue beyond the
lifetime of the project.
2.6 Time-sequencing lake sediment traps

Sediment traps installed in lakes capture parti-
cles settling through the water column. Long-
term, high-frequency monitoring offers insight
into the biogeochemical functioning, sedimen-
Figure 11. (a) Internal hardware components and
tation regime and seasonal changes in biodiver- (b) post-deployment, highlighting concerns around
sity that cannot be replicated in laboratory biofouling. An improved version will incorporate an
experiments (Bonk et al., 2015; Chmiel et al., enclosure around the rotating carousel.
2016; Schillereff et al., 2016). Static trap
deployment is common but requires manual
retrieval, severely restricting sampling fre- fixed in carousel holes with epoxy resin and
quency, especially at remote sites. Time- holes bored equivalent to the funnel diameter.
sequenced instruments that open separate The downpipe (h ¼ 75 cm, outer diameter ¼
containers at preprogrammed intervals provide 110 mm) aspect ratio of 6.8:1 follows the rec-
valuable temporal resolution. Commercial ver- ommendations of Bloesch and Burns (1980) to
sions are costly (>£10,000). Build-it-yourself ensure representative sediment capture in small
designs exist (e.g. Muzzi and Eadie, 2002) but lakes. The unipolar 28BYJ-48 stepper motor is
require greater expertise in mechanical and cost-effective (*£2.50) while offering high-
electrical engineering. A reliable, low-cost precision rotation at low speeds. Although the
sequencing sampler will therefore transform stepper motor draws 5V, testing confirmed a 3.3
limnological research, especially in light of V Pro Mini provided adequate power for 30-day
funding pressures on long-term lake monitoring rotation. Different time-steps are easily pro-
programmes. In total, our current design comes grammed for alternative applications.
to £80. A 12-month test deployment in Crose Mere,
Initially an undergraduate project, we swiftly Shropshire (52.86 N, 2.84 W) successfully
appreciated the research potential of an recovered sediment each month. Trap installa-
Arduino-based sequencing sediment trap. Our tion involved fixing the downpipe using
original design comprised three main compo- D-clasps to 5 mm wire held between a basal
nents: (a) two 3D-printed carousels (d ¼ 187 20-kg weight and buoys: one larger suspended
mm) holding twelve 60 mL NalgeneTM poly- below the annual minimum lake level to main-
ethylene bottles, fixed by threaded rod to a step- tain taut deployment and a small, coloured float
per motor (Figure 11(a)); (b) cylindrical PVC at the surface. The design was operationally
downpiping that feeds a funnel sitting over the effective but water seepage into the housing was
carousel hole (d ¼ 33 mm); and (c) an IP68- a concern, most likely through the cable gland
rated enclosure housing the stepper motor and during axle rotation. Our second version uses
Arduino-based electronics. Bottle lids were shaft seals to minimise water ingress and we are
trialling open-source underwater remotely oper- system. A Gill R3-50 sonic anemometer pro-
ated vehicle tricks of filling the housing with vides 3D wind vector measurements at 50 Hz,
wax. Trap recovery highlighted two further output via an RS232 serial ASCII data stream,
issues that are easily rectified by using improved while a Wenglor laser-counter detects sand
enclosures: biofouling (Figure 11(b)) and abra- grains flying through a narrow laser beam, out-
sion of bottle labels. The volumes of trapped putting a 100 ms voltage pulse for each interrup-
sediment dispelled concerns that 60 mL contain- tion (Davidson-Arnott et al., 2009). Traditional
ers are too small, at least in eutrophic, productive dataloggers struggle with these data output and
lakes. recording requirements; simple and low-cost
loggers exist for pulse signals, but typically do
not possess RS232 input capabilities and are
2.7 High-frequency measurement often restricted in temporal resolution to log-
of wind-blown sand ging at 1 Hz or less. High-end dataloggers
Research on sand transport by wind includes a (e.g. Campbell Scientific CR1000X at
rich variety of electronic sensors for measuring *£1550) on the other hand can handle RS232
and recording physical processes and flows at input, but have only a few dedicated pulse
relatively high frequencies (Hugenholtz and counter input channels and can be cumbersome
Barchyn, 2011; Sherman et al., 2013). Typical to transport. Our Arduino-based solution (*£45)
field instrumentation includes sonic anem- uses a Due microcontroller board, which operates an
ometers for recording wind vectors, electroni- 84 MHz processor and can accommodate several
cally weighing sand traps, sand-grain impact dozen count channels as well as RS232 input via
sensors, and laser interference instruments for an RS232-to-TTL (transistor-transistor logic) adap-
detecting saltating sand transport rates, and tor. The ASCII stream from the sonic anem-
additional equipment such as continuous soil- ometer is read and stored into an accruing string
moisture probes and further meteorological in the memory, one character at a time. Pulses
sensors. The acquisition and data storage of from the Wenglor are counted via an external
high-frequency time series of wind and sand interrupt routine. The anemometer sends a termi-
transport measurements are crucial to investi- nation character after each output of a wind vector
gating the relationship between turbulence in measurement stream (every 0.02 s) and receipt of
the airflow and the spatio-temporal variability this termination character at the Due triggers sev-
of sand transport, displayed particularly by the eral appendments to the memory string: the current
ubiquitous presence of streamers (also known as total pulse count, a time-stamp, and a carriage
sand snakes) in wind-blown sand (Baas, 2008; return (or ‘new line’ break), while the pulse coun-
Baas and Sherman, 2005). Sensors are posi- ter is reset to zero. After storing 200 lines (i.e. 4 s
tioned in close proximity to each other but data worth) of data, the memory is then written to a
outputs of different types are required to be microSD card attached to the Due. The temporary
stored synchronously as well as at the original on-board memory storage is crucial because the
high measurement frequencies. This poses sig- SD writing process is comparatively slow, render-
nificant challenges to traditional data loggers ing writing directly to the SD card at the ‘raw’ data
but provides opportunities for the custom-built rate of 50 Hz unfeasible. The code for the running
and low-cost Arduino-based data acquisition loop in the Arduino processor is very short and
system (DAS). efficient, minimising the processor overhead.
Our latest research combines sonic anemo- This Arduino-based DAS has been lab tested
metry with laser-counter sensors, which have using a custom-built rotating disc with a series
been integrated with an Arduino-based logger of perforations in its rim, mounted on a multi-
Chan et al. 21
Table 9. Number of pulse counts per second as

predicted from the reference track at different
drill-speeds, compared with the measured rates
from the test tracks.
Pulse counts (Hz)
Predicted Measured Measurement bias (%)
342 342 0.00
685 686 0.15
888 887 –0.11
1777 1774 –0.17
2271 2275 0.18
4543 4556 0.29
Figure 12. Rotating disc with three perforation
tracks, mounted in a multi-speed bench drill, with
sideways mounted Wenglor laser-counter fork-
sensor (red laser beam reflection visible) used
for calibration.
speed, belt-pulley driven, bench drill while

running the sonic anemometer in front of an
ordinary air fan. The purpose of the testing was
to verify whether the Arduino-based DAS – the
external interrupt routine in particular – was
capable of correctly recording the pulses from
the Wenglor, even at very high pulse-rates,
while also continuously processing the ASCII
data stream from the sonic anemometer. The Figure 13. The data acquisition system used for
rotating disc had three different tracks of per- synchronous recording of Gill sonic and Wenglor
forations (Figure 12): 72 holes along the rim, 36 laser-counter measurements, housed in a portable
holes on the inside track, and a reference track enclosure. Components: (a) Arduino Due microcon-
troller board, (b) voltage divider to reduce the
of only four holes covering one disc rotation
*11VDC output from the Wenglor to <3VDC input
further toward the centre of the disc. to the microcontroller board, (c) RS232 input from
The reference track was used to measure the the Gill Sonic communication unit, (d) mini-SD card
actual revolutions per minute (RPM)of the ‘shield’ for storing the data, (e) Gill sonic communi-
bench drill at different speeds (since the nom- cation unit, and (f) 12 VDC battery power supply.
inal spindle speeds from the belt-pulley ratios
are not very precise), followed by the two test highest RPM test, the results show that the sys-
tracks during the same drill-run. The test results tem can easily measure and record pulse rates of
reported in Table 9 show that the number of at least 4500 counts per second (as well as the 3D
pulses per second counted on the two test tracks airflow data) at the required 50 Hz. This exceeds
correctly match the predicted number of pulses the tested capabilities of commercial logger com-
per second, within the prediction accuracy. Dur- binations (Bauer et al., 2018). A pilot field
ing all these tests the ASCII data stream was deployment has demonstrated the success and
recorded with no transcription errors. At the portability of the system (Figure 13) and its
compatibility with large-scale particle image wind direction sensors all draw 2 V or less. The
velocimetry equipment (Baas and Van den Berg, Alphasense NO2 sensor output, on the other
2018). The duration for which this DAS can be hand, is between 300 and 400 mV so an
deployed is only defined by the battery capacity ADS1115 module was used to improve reading
and SD card storage limit, running to several sensitivity, both through focused voltage ranges
days in the setup used here. and 16-bit resolution. This module has far
superior resolution (detecting 65,536 voltage
III Common pitfalls, lessons ‘steps’ compared to 1024) and voltage range
(with a full range as low as +0.256 V where
learned and best practice workflow one step equates to 7.81 pV, compared to 1.08
3.1 Major advances and successes mV when using the higher resolution 1.1 V ref-
Our cumulative experience has highlighted the erence voltage on the 3.3 V Pro Mini). Our
following key considerations from which we instruments that use digital sensors (tempera-
have developed a set of best practice guidelines. ture and relative humidity modules, DO, Plan-
tower PM sensors, 3D sonic anemometer) have
3.1.1 Standardised and bespoke circuit boards. inbuilt ADC convertors and communicate the
While Arduino offers near-limitless adaptabil- calibrated readings.
ity, a key aspect of our streamlined workflow is The scope to integrate multiple sensors, each
having core design frameworks. For example, measuring a different environmental parameter,
we now have a standard circuit board design is a major advantage of the build-it-yourself
for ultra-low-power loggers (important for approach but we repeatedly encountered prob-
long-term monitoring) that can be readily lems of compatibility. This was exacerbated by
adapted to most sensors. Similarly, we have sensors sourced from new manufacturers that
developed replicable methods of incorporating may draw different voltages or conflicting code
a solar panel onto designs where possible. While libraries. The increasing number of clone
we regularly use solderless breadboards for pro- microcontroller boards on the market may well
totyping and as teaching aids, soldered wires are exacerbate these issues. We therefore use hard-
near-essential to minimise the possibility of loose ware specifically designed for Arduino hard-
wires and short-circuits. Poor or incorrect wiring ware with pre-existing Arduino libraries for
is the most common malfunction, in our experi- most designs.
ence. We are increasingly making use of bespoke
PCBs, led by the FreeStation project. Designing 3.1.2 Documentation. Developing low-cost envi-
PCBs in conjunction with OSHPark is cost- ronmental sensors does not require prior exper-
effective, simplifies the electrical assembly and tise with electronics or programming, though
minimises wiring faults while maximising custo- experience in the latter is beneficial. What is
misability for multi-sensor applications. They crucial, however, is documenting every stage
can also accommodate web-integrated cellular of the design and testing process. During the
boards such as the Particle Electron and can be design stage we share build notes, schematics
designed to be swapped for the cheaper, Wi-Fi and ‘sketches’ on a shared web folder before
only Particle Photon (https://docs.particle.io/elec tidy versions are moved to our Github or Free-
tron/). Station repository. The requisition log is also
For data transfer, the suitability of the stan- shared, facilitating rapid price comparisons and
dard ADC on board the ATmega328P micro- bulk orders, minimising excess purchasing and
controller will depend on sensor and research highlighting reliable suppliers. The FreeStation
requirements. Our thermistor, conductivity and website fully documents the build steps and
Chan et al. 23
component list and displays live data, for exam- suppliers (Table 1), and there is a trade-off
ple. When writing code, best practice including between delivery time and cost, especially when
version control, and in-line commenting is ordering from China. This can be problematic
strongly recommended (Goodliffe, 2007). Shar- when a failed prototype requires one component
ing designs widely is at the core of the Arduino to be replaced.
open-source platform. This has the benefit of
effectively gaining free testing, troubleshooting 3.1.4 Workflow recommendations. Our stream-
and development of designs. lined workflow is presented in Figure 14.
Designing a reliable sensor is a highly iterative
3.1.3 Cost. Vastly reduced component costs process from sketch to successful deployment.
compared to conventional commercial instru- Log and photograph each wiring configuration
ments is a key benefit of Arduino technology. and housing assembly; it will assist in recrea-
Open-source medical technology is estimated to tion, troubleshooting and may be useful for a
provide a return on investment for funders future project. Think carefully from the outset
reaching hundreds or thousands of percent about research priorities: which components are
(Pearce, 2015). Build-it-yourself sensor net- essential? Each addition heightens risks of hard-
works have particular value in light of current ware or software issues. Testing must replicate
funding pressures in science (Tetzlaff et al., real-world deployment conditions as closely as
2017), with initiatives such FreeStation (Section possible, both in terms of environmental condi-
2.5) expanding capacity amongst government tions (sufficient solar power supply, for exam-
authorities with limited environmental monitor- ple) and length of deployment. We strongly
ing infrastructure. There are hidden costs to recommend verifying data quality after a short
acknowledge in terms of labour and failed pro- deployment phase, but keep in mind that not all
totyping, both of which are amplified when libraries are designed to automatically restart if
developing new sensors. Reproduction rather the SD card is removed.
than reinvention will reduce both of these costs.
The microprocessor and core peripheral com-
ponents in an environmental data logger are 3.2 Common pitfalls
very low (Table 1); sensors and enclosures rep- Though we outline our recommended workflow
resent the majority of expenditure for every above, it is essential to bear in mind the follow-
project. A wide variety of low-cost sensors for ing common pitfalls for successful deployment.
measuring specific environmental variables
exist on the market. Our testing of multiple par- 3.2.1 Testing and calibration. Unsurprisingly, test-
ticulate sensors, for example, showed the ing and calibration are crucial. Our experiences
importance of performing a cost-benefit analy- show that testing must follow the deployment
sis. The Sharp model is typically less than half protocol as closely as possible. This has implica-
the price of the Plantower PMS series (Table 4) tions during the build and programming phases.
but is significantly more sensitive to tempera- For example, a sensor that successfully logs at
ture fluctuations and requires manual calibra- 1-min intervals during lab testing offers no guar-
tion. In most cases, the additional outlay for antee that switching to, say, 30-min intervals
sensors that incorporate more reliable internal upon deployment will be faultless. Some Ardu-
calibration is advisable. Sensor quality versus ino libraries helpfully supply one line of code to
cost should also be guided by data-quality set measurement intervals, but we found more
requirements (Terando et al., 2017). Component substantive edits were often required when com-
costs can also vary by up to 50% between bining sensors, particularly when power-down
Figure 14. A schematic visualisation of our workflow for developing low-cost environmental monitoring
devices and key considerations at each stage.
commands are invoked. In some cases, elaborate conditions are highly recommended (Rai
apparatus needed to be constructed in a labora- et al., 2017) but may be logistically problematic.
tory to mimic real-world conditions (e.g. the Despite our geographical proximity, for example,
wind-blown sand laser-counter; Section 2.7). tight regulations mean testing in the River
Calibration checks under final deployment Thames is non-trivial. On the other hand, we
Chan et al. 25
are fortunate that LAQN allow our Arduino Khanfar et al., 2017; Metzger et al., 2018).
sensors to be tested and calibrated at their flag- These guides rarely comment on the time com-
ship Marylebone station. mitment, however. While this will depend on
Calibration, additionally, should not be con- the level of technical competence and experi-
sidered a one-time job. Componentry and sensor ence of the designer, the trial and error nature
materials are subject to degradation as they age, of designing new instruments exacerbates this
introducing drift in reported results and poten- issue. The water quality sensor (Section 2.4)
tially hampering accurate measurements development process illustrates this pitfall: four
(Artursson et al., 2000; Bourgeois et al., different models of the sensor were produced,
2003). Post-processing can be implemented to involving numerous field and lab tests, of which
correct for sensor drift through measuring stan- two probes were successfully deployed.
dard quantities or cross-calibration against other There is a crucial distinction to be made
more recently calibrated and/or accurate sen- between developing a new sensor and reprodu-
sors. Moreover, cross-calibration becomes cing an existing design. Labour is expensive so
increasingly powerful as a greater density of bespoke development is a serious commitment,
sensors are deployed (El-Jabi and Caissie, but reproduction vastly streamlines the time
2019). Lastly, whilst data quality from low- burden. A fully functioning FreeStation AWS
cost sensors depends on careful calibration of can be constructed in 4 h, for example, but the
individual sensors, the same practice applies FreeStation instructions, now online, are the
when deploying conventional commercial result of years of testing and refining.
instruments. Our ability to replace individual
sensors in self-build designs further mitigates 3.2.3 Sensor housing. Robust external housing is
against sensor drift (see Section 2.5). critical. The deployment environment will dic-
We recommend testing also be carried out at tate the sealing effectiveness and appropriate
the component level – that is, prior to sensor Ingress Protection rating required for a casing,
assembly – as visual and electronic inspection but preventing water ingress is a challenge that
can reveal flaws in purchased components. Wires we underestimated repeatedly. Diagnosing the
are often mounted differently to supplied sensor source of a leak is particularly challenging.
schematics, for example, potentially short- Moreover, constructing watertight enclosures
circuiting the Arduino board and, at worst, posing using materials in keeping with a ‘low-cost’
a fire hazard. Arduino components can usually be project adds an additional obstacle. Housing the
replaced – certainly more easily than commercial microprocessor and associated peripherals (SD
loggers – but early testing saves on time, expense card, clock) separately from the data-collecting
and frustration. device may minimise leak points, even if super-
ficially via sealing exposed components with
3.2.2 Time. Our experience shows clearly that epoxy, as implemented in our water depth log-
every stage in a new project takes longer than ger. Liberal application of resin, silicon sealant
pre-existing papers, instructions or even simple or silicone grease is warranted, often in con-
replication would have us believe. Repeated junction with cable glands or thread seal tape.
builds also bring an unexpected cost element. Housing dimensions also need to accommodate
Publications showcasing the ‘build-it-yourself’ appropriate battery options tailored to sensor
approach often present the methodology and power draw. Battery packs constitute up to
schematics for a functioning sensor, followed two-thirds of the space requirements for some
by a brief reflection on accuracy and future of our sensors, so a belated realisation that more
applications (e.g. Beddows and Mallon, 2018; power is necessary could necessitate a wholly
new housing. We increasingly manufacture Shaded deployment sites along riverbanks and
3D-printed containers to optimise protection obtaining adequate exposure in built-up areas
and streamline the design process, especially for have proved difficult. Integrating components
housing smaller components. Filling gaps in that draw 3.3 V and 5 V is another complication,
commercial casings with epoxy is extraordina- particularly when considering digital communi-
rily time-consuming, for example. Loose wires cation lines may run different voltages from
are a common malfunction; we advocate soldered sensor power voltages. Conversely, testing
wire connections, PCBs and the plug-and-play showed a 3.3 V Arduino Nano could drive the
approach of the FreeStation to maximise 5 V stepper motor on the sediment trap, which
durability. aided compatibility. There have also been nota-
Concerns around data quality have been ble developments around power saving in recent
raised regarding ad hoc housing. Terando years across the Arduino community, involving
et al. (2017), for example, identified discrepan- new hardware and scripts (Beddows and Mal-
cies of up to 3 C when testing build-it-yourself lon, 2018; Rocket Scream, 2020). Lastly,
Stevenson screens. Whilst this highlights a removing obsolete LEDs from the Arduino and
potential pitfall in the use of low-cost sensors, connected shields using a hot soldering iron or
commercial data loggers also report substantial carefully slicing tracks to superfluous compo-
variance (e.g. Whittier et al., 2020). Indeed, the nents with a sharp blade can reduce power draw
need for rigorous sensor-specific calibration substantially.
follows best practice in environmental monitor-
ing and is not a hurdle unique to build-it- 3.2.5 Sensor and library compatibility. Progressing
yourself sensors. Furthermore, the capacity to from a complete assemblage of sensors, board
deploy a higher number of sensors for equiva- and wires to an operating, reliable instrument is
lent financial outlay brings significant benefits easily underestimated. One of the biggest hur-
through detailed cross-calibration (El-Jabi and dles we repeatedly encounter is a lack of com-
Caissie, 2019). At the same time, there is clearly patibility between sensors and Arduino libraries
scope to promote reproduction rather than rein- when designing multiprobes. Each additional
vention. The open-source approach of Arduino component introduces a non-linear degree of
and Internet-of-Things technology could, in added complexity, with conflicting libraries a
fact, lead to greater methodological consistency common occurrence. For our aquatic multiprobe,
where, for example, a 3D-printer design for a individual sensors were accurately calibrated but
Stevenson shield is shared widely amongst the daily means did differ when integrated into a
research community. single instrument. We attributed these issues to
electrical interference, which requires targeted
3.2.4 Power. We have grappled at length with compensation (Siragusa and Galton, 2000) and
ensuring adequate power supply and maximis- significantly longer build and testing times.
ing longevity. Think carefully about minimum Similarly, whilst most PM sensors use laser scat-
measurement intervals, which will be guided by tering, internal differences between manufactur-
research objectives. Will a 30-min or 60-min ers produce unique biases. These are rarely clear
wake-up interval provide appropriate data? in supplied documentation.
What is the minimum period a sensor needs for
readings to stabilise? We now have a standard 3.2.6 Deployment considerations. We also empha-
core design for ultra-low-power sleeping log- sise that deployment protocol is a non-trivial
gers and increasingly incorporate solar- aspect that is rarely afforded due consideration.
powered, rechargeable lithium ion batteries. After the more arduous task of designing,
Chan et al. 27
building and calibrating low-cost environmen- generally included in methodological sections

tal loggers, deployment seems the simple and of journal articles, alternative methods may be
exciting job. This is a particular issue when sen- required for storing computer-aided design
sors are handed from makers – who may know (CAD) files, board designs, source codes, and
the particularities of the logger and sensor setup build or calibration instructions.
– to fieldworkers. Without adequate consider- We would encourage academic authors to
ation of the deployment criteria of specific sen- host build instructions and materials on widely
sors (e.g. under what conditions does the sensor used public-facing open-source sites wherever
accurately measure? What periodic mainte- possible. Helpfully, a range of suitable online
nance is required? Where specifically should repositories now exist, the most common being
the sensor be mounted?) results may pay a dis- Github (a more software-focused online reposi-
service to the effort expended in design and tory), PublicLab (focused on technologies or
development. This reinforces the need to share methodologies of measuring environmental
understanding of the sensors, loggers and field- quality parameters), Thingiverse (hosting CAD
work conditions between makers/electronic files) and the Open Hardware Repository
engineers and fieldworkers. General good- (focused on electronics hardware). Many of
practice guidance for attaining accurate mea- these repositories embrace aspects of open or
surements of the particular environmental ‘remix culture’ through enabling original source
parameter should also be adhered to. material to be ‘forked’ or ‘remixed’ – when
original source material is built upon indepen-
dently by developers. These improvements can
IV Attribution and intellectual then be ‘pushed’ (merged) to improve the orig-
properties inal code. Github in particular has become ubi-
The open-source revolution greatly enables cus- quitous for the sharing of source codes and
tomisation, enhancement and collaborative designs, possessing 50 million users and over
efforts within technological development 100 million repositories (Github, 2020).
through making software and hardware designs The above repositories enable collaborative
accessible and implementable. Those of us in efforts through the very principals of transpar-
the academic sphere, however, necessarily ency, encouragement of modification, and pro-
require attribution to ensure we as researchers motion of community contribution. Here, a
are recognised for our contribution to encourage conflict arises between the ability to modify
the conceptual and theoretical development of designs, around which the aforementioned repo-
research whilst ensuring this development can sitories exist, and the reproducibility (and
still be logically tracked. Though a number of necessarily explicit version control) expected
OSH journals have recently been released – for of the scientific field. Helpfully, digital object
example, Sensors (launched 2001), HardwareX identifiers (DOIs) provide a mechanism for both
(launched 2017), Journal of Open Hardware recognition of sources and direction to specific
(launched 2017) – journals focused on the more versions of digital material. A DOI is a unique
traditional scientific disciplines remain the pre- alphanumeric string assigned by the Interna-
ferred publishing destination for many users of tional DOI Foundation and associated registra-
build-it-yourself hardware. The majority of sci- tion agencies (e.g. Crossref – the DOI registrar
entific journals in geographical and environ- used by the majority of academic publishers).
mental fields however are clearly not geared Persistence is a key tenet of DOIs (International
towards technological design or hardware; thus, DOI Foundation, 2015), meaning very little, if
while instructions or design descriptions are any, modification is permitted to material
assigned a DOI. This makes DOIs more optimal Where no specification of a licence is given,
for scientific citation than adopting the web licensed usage should not be assumed.
addresses of earlier repositories. Common Choosing a licence under which to release
archives that are explicitly geared towards DOI your own codes and schematics is also a com-
creation are the Open Science Framework, plex topic, requiring consideration of what
Zenodo and Figshare. We particularly promote exactly is being licensed (software, hardware
Zenodo, which has integration with Github to and/or schematics), permissions for future use
allow archiving of specific versions of Github of your work (e.g. non-commercial applications
repositories, thus benefitting from the vast user- only), whether attribution is required, and pro-
base and exposure that Github offers. Note addi- tection of your future rights, in addition to abid-
tionally, however, that many research councils ing by the original licensing rules of any
(UK and abroad) now require that funded proj- material that you incorporate into your designs.
ect data are uploaded to their data repositories, Hundreds of licences now exist, and the nuances
which can often be assigned a DOI. of these licences clearly exceed the scope of this
Some awareness of licences should be con- paper. We, however, recommend three particu-
sidered essential in sensor development. The larly useful resources: Software Licenses in
open hardware and software community have Plain English (tldrlegal.com), choosealicense.
grown to embrace this aspect, but navigating the com and The Legal Side of Open Source (open-
options can be puzzling. It is important devel- source.guide/legal).
opers understand that just because codes and/or
designs are available online, this does not make
them free to use. From a hardware perspective, V Summary
Arduino has adopted the Creative Commons In this paper we have showcased the ability of
Attribution-ShareAlike (CC BY-SA), in brief low-cost sensors to transform environmental
meaning anyone can recreate the hardware, monitoring of aquatic, terrestrial and atmo-
though Arduino need to be credited and deriva- spheric systems around the world. By providing
tive hardware designs must be made available full design schematics, code and guidance on
under the same licence. From a software per- purchasing the components on our Github repo-
spective, use of the standard Arduino IDE and sitory (https://github.com/KCLGeography/envi
Arduino libraries is covered by the GNU Lesser ronmental-monitoring), we intend this paper to
General Public License (LGPL), meaning firm- act as a catalyst for geographers and environmen-
ware designed with non-modified versions of tal scientists to embed low-cost, build-it-yourself
these does not require sharing if the firmware sensors into their research programmes. Deriving
is not designed to relink to newer versions of insight from six case studies, including the global
either Arduino core or libraries. Modification of FreeStation hydrometeorological network
the Arduino IDE is required to be shared under (www.freestation.org), we have demonstrated
the General Public License of the IDE, whilst the potential for low-cost sensors powered by
modification of the Arduino environment (i.e. Arduino across a wide range of disciplines
the initial firmware uploaded to the microcon- including atmospheric science, ecology, geomor-
troller) or Arduino libraries is required to be phology and hydrology. By drawing on 6 years’
shared under the LGPL. It is important to note experience, we have also highlighted potential
that third-party libraries or environments will pitfalls in design and construction, recommenda-
likely have separate licensing agreements that tions for best practice have been proposed, and a
must be individually consulted. Similarly, licen- workflow for developing new sensors and over-
sing rules differ for commercial applications. coming technical challenges has been presented.
Chan et al. 29
In this paper, we have also evaluated the perfor- particular, Alex Blair, Nicole Cowell, Abbey Wong
mance of our Arduino sensors and found strong and Harriet Wilson.
performance in each case, reporting mean bias Thanks to Dr Massimo Lupascu and Hasan Akhtar
errors below 20%. This confirms that electronic (National University of Singapore) for their assis-
sensors designed and constructed for a fraction of tance with deploying and maintaining the water table
depth probes.
the conventional commercial cost can deliver
research-grade data, particularly where greater Author contribution
granularity is required. Data quality depends on
Daniel Schillereff and Kristofer Chan contributed
careful calibration that must be carried out on a equally to leading the writing of this paper.
sensor-specific basis; this equally follows best
practice when using conventional commercial Declaration of conflicting interests
instrumentation. Given global funding pressures The author(s) declared no potential conflicts of inter-
in science, low-cost sensor networks have the est with respect to the research, authorship, and/or
potential to deliver important benefits through publication of this article.
improved representation of spatial and temporal
variability as well as customisability – that is, the Funding
opportunity to develop sensors tailored to a par- The author(s) disclosed receipt of the following
ticular research need or physical setting. Our financial support for the research, authorship, and/
experience has demonstrated that the Arduino or publication of this article: the Natural Environ-
and Internet-of-Things technology and support- mental Research Council and the Arts & Humanities
ing communities are sufficiently developed to Research Council (grant number NE/R017999/1),
allow geographers and environmental scientists the Economic and Social Research Council (grant
number ES/R002126/1), the King’s College London
with no background in electronics and limited
(KCL) Undergraduate Research Fellow programme,
coding experience to build new sensors. The
the KCL Faculty of Social Science and Public Policy
potential for sensor development is essentially for a Research Grant, a KCL Education Grant, and a
limited only by imagination, as examples of KCL Widening Participation Grant, and the London
open-source Geiger counters and Arduino- School of Economics for a Learning Technology and
based CubeSat satellites demonstrate (Geeroms Innovation Ignite! grant and a Pro-Director for Edu-
et al., 2015; SeedStudio, 2011). Our workflow, cation Vision Fund grant, AmbioTEK Community
schematics, code and tools for web integration Interest Company (www.policysupport.org), and the
(e.g. FreeStation) presented in this paper estab- Nature Insurance value: Assessment and Demonstra-
lish a framework for enhancing environmental tion (NAIAD) project of the European Commis-
monitoring and management from the local to sion’s H2020 programme (grant agreement:
global scales. 730497).
ORCID iDs
Acknowledgements Daniel N. Schillereff https://orcid.org/0000-0002-
We thank the two anonymous reviewers for their 4928-6068
careful, comprehensive and constructive reviews, Francis T. O’Shea https://orcid.org/0000-0002-
which have allowed us to greatly improve the paper. 0808-7441
We are grateful to a number of funders. Numerous Emma Tebbs https://orcid.org/0000-0003-0575-
KCL Geography and London School of Economics 1236
Department of Geography and Environment under-
graduate and postgraduate students have supported, References
inspired and made vital contributions to our low-cost Alphasense (2017) OX-A431 Oxidising Gas Sensor 4-
sensor development. For this paper we thank, in Electrode Ozone þ Nitrogen Dioxide. Available at:
http://www.alphasense.com/index.php/products/ Bramer I, Anderson BJ, Bennie J, et al. (2018) Advances in

ozone-2/ (accessed 4 June 2019). monitoring and modelling climate at ecologically rel-
Artursson T, Eklöv T, Lundström I, et al. (2000) Drift evant scales. Advances in Ecological Research 58:
correction for gas sensors using multivariate methods. 101–161.
Journal of Chemometrics 14(5-6): 711–723. Browning E, Gibb R, Glover-Kapfer P, et al (2017) Pas-
Atkinson RW, Analitis A, Samoli E, et al. (2016) Short- sive acoustic monitoring in ecology and conservation.
term exposure to traffic-related air pollution and daily Woking: WWF
mortality in London, UK. Journal of Exposure Science Brümmer IHM, Felske A and Wagner-Döbler I (2003)
and Environmental Epidemiology 26(2): 125–132. Diversity and seasonal variability of b- proteo-
Atlas Scientific (2019) Dissolved Oxygen Kit. Available bacteria in biofilms of polluted rivers: Analysis by
at: https://www.atlas-scientific.com/product_pages/kits/ temperature gradient gel electrophoresis and clon-
do_kit.html (accessed 6 June 2019). ing. Applied and Environmental Microbiology 69(8):
Baas ACW (2008) Challenges in aeolian geomorphology: 4463–4473.
Investigating aeolian streamers. Geomorphology 93(1): Bulot FM, Johnston SJ, Basford PJ, et al. (2019) Long-
3–16. term field comparison of multiple low-cost particu-
Baas ACW and Sherman DJ (2005) The formation and late matter sensors in an outdoor urban environment.
behavior of aeolian streamers. Journal of Geophysical Scientific Reports 9(1): 1–13.
Research 110: F03011. Burt TP and McDonnell JJ (2015) Whither field hydrol-
Baas ACW and Van den Berg F (2018) Large-Scale Par- ogy? The need for discovery science and outrageous
ticle Image Velocimetry (LSPIV) of aeolian sand hydrological hypotheses. Water Resources Research
transport patterns. Aeolian Research 34: 1–17. 51(8): 5919–5928.
Bauer BO, Davidson-Arnott RG, Hilton MJ, et al. (2018) Carey IM, Anderson HR, Atkinson RW, et al. (2018) Are
On the frequency response of a Wenglor particle- noise and air pollution related to the incidence of
counting system for aeolian transport measurements. dementia? A cohort study in London, England. BMJ
Aeolian Research 32: 133–140. Open 8: e022404.
Beddows PA and Mallon EK (2018) Cave Pearl Data Castell N, Dauge FR, Schneider P, et al. (2017) Can
Logger: A flexible Arduino-based logging platform for commercial low-cost sensor platforms contribute to air
long-term monitoring in harsh environments. Sensors quality monitoring and exposure estimates? Environ-
18(2): 530. ment International 99: 293–302.
Behmel S, Damour M, Ludwig R, et al. (2016) Water Chmiel HE, Kokic J, Denfeld BA, et al. (2016) The role of
quality monitoring strategies: A review and future sediments in the carbon budget of a small boreal lake.
perspectives. Science of the Total Environment 571: Limnology and Oceanography 61(5): 1814–1825.
1312–1329. Concas F, Mineraud J, Lagerspetz E, et al. (2020) Low-
Bloesch J and Burns NM (1980) A critical review of Cost Outdoor Air Quality Monitoring and In-Field
sedimentation trap technique. Schweizerische Zeits- Sensor Calibration. Available at: https://arxiv.org/
chrift für Hydrologie 42(1): 15–55. abs/1912.06384 (accessed 28 February 2020).
Boisseau É, Omhover JF and Bouchard C (2018) Open- D’Ausilio A (2012) Arduino: A low-cost multipurpose lab
design: A state of the art review. Design Science 4: e3. equipment. Behavior Research Methods 44(2):
Bonk A, Tylmann W, Amann BJF, et al. (2015) Modern 305–313.
limnology, sediment accumulation and varve forma- Davidson-Arnott R, Bauer BO, Walker IJ, et al. (2009)
_
tion processes in Lake Zabińskie, northeastern Poland: Instantaneous and mean aeolian sediment transport rate
Comprehensive process studies as a key to understand on beaches: An intercomparison of measurements from
the sediment record. Journal of Limnology 74(2): two sensor types. Journal of Coastal Research 56:
358–370. 297–301.
Bourgeois W, Romain AC, Nicolas J, et al. (2003) The use DFRobot (2017) Analog EC Meter SKU: DFR0300.
of sensor arrays for environmental monitoring: Inter- DFRobot. Available at: https://www.dfrobot.com/wiki/
ests and limitations. Journal of Environmental Moni- index.php/Analog_EC_Meter_SKU:DFR0300 (acces-
toring 5(6): 852–860. sed 30 April 2019).
Chan et al. 31
El-Jabi N and Caissie D (2019) Characterization of https://www.bafg.de/GRDC/EN/01_GRDC/13_dtbse/

natural and environmental flows in New Brunswick, database_node.html (accessed 30 April 2019).
Canada. River Research and Applications 35(1): Goodliffe P (2007) Code Craft: The Practice of Writing
14–24. Excellent Code. San Francisco: No Starch Press.
Environmental Defense Fund (2018) Breathe London: Hannah DM, Demuth S, van Lanen HA, et al. (2011) Large-
Mapping air pollution at the street level. Available at: scale river flow archives: Importance, current status and
https://www.breathelondon.org/about/ (accessed 15 future needs. Hydrological Processes 25(7): 1191–1200.
April 2020). Hijmans RJ, Cameron SE, Parra JL, et al. (2005) Very
Evers S, Yule CM, Padfield R, et al. (2017) Keep wetlands high resolution interpolated climate surfaces for
wet: the myth of sustainable development of tropical global land areas. International Journal of Clima-
peatlands – implications for policies and management. tology 25: 1965–1978.
Global Change Biology 23(2): 534–549. Horsburgh JS, Caraballo J, Ramı́rez M, et al. (2019) Low-
Feenstra B, Papapostolou V, Hasheminassab S, et al. cost, open-source, and low-power: But what to do with
(2019) Performance evaluation of twelve low-cost the data? Frontiers in Earth Science 7: 67.
PM2.5 sensors at an ambient air monitoring site. Hugenholtz CH and Barchyn TE (2011) Laboratory and
Atmospheric Environment 216: 116946. field performance of a laser particle counter for
Ferdoush S and Li X (2014) Wireless sensor network measuring aeolian sand transport. Journal of Geo-
system design using Raspberry Pi and Arduino for envi- physical Research: Earth Surface 116(F1): F01010.
ronmental monitoring applications. Procedia Computer International DOI Foundation (2015) DOI Handbook.
Science 34: 103–110. Available at: www.doi.org/hb.html (accessed 20 April
Ferreira JG, Andersen JH, Borja A, et al. (2011) Overview 2020).
of eutrophication indicators to assess environmental Jayaratne R, Liu X, Thai P, et al. (2018) The influence
status within the European Marine Strategy Framework of humidity on the performance of a low-cost air
Directive. Estuarine, Coastal and Shelf Science 93(2): particle mass sensor and the effect of atmospheric
117–131. fog. Atmospheric Measurement Techniques 11(8):
FreeStation (2019) FreeStation introduction. Available at: 4883–4890.
http://www.freestation.org/ (accessed 6 June 2019). Johnson KK, Bergin MH, Russell AG, et al. (2018) Field
Furber S (2017) Microprocessors: The engines of the test of several low-cost particulate matter sensors in
digital age. Proceedings of the Royal Society A: 473(2199): high and low concentration urban environments.
20160893. Aerosol Air Quality Research 18(3): 565–578.
Geeroms D, Bertho S, De Roeve M, et al. (2015) Kannel PR, Lee S, Lee YS, et al. (2007) Application of
ARDUSAT, an Arduino-based Cubesat providing water quality indices and dissolved oxygen as indica-
students with the opportunity to create their own tors for river water classification and urban impact
satellite experiment and collect real-world space assessment. Environmental Monitoring and Assess-
data. In: Proceedings of the 22nd ESA symposium ment 132(1–3): 93–110.
on European rocket and balloon programmes and Karvinen T and Karvinen K (2011) Make a Mind-
related research (ed L Ouwehand), Tromso, Nor- Controlled Arduino Robot: Use Your Brain as a
way, 7–12 June 2015, xx–xx. Noordwijk, The Remote. Sebastopol: Maker Media, Inc.
Netherlands: ESA Communications. Kelly FJ and Fussell JC (2019) Improving indoor air
Giocomassi Luciano A, Fusinato PA, Gomes LC, et al. quality, health and performance within environments
(2019) The educational robotics and Arduino platform: where people live, travel, learn and work. Atmospheric
Constructionist learning strategies to the teaching of Environment 200: 90–109.
physics. Journal of Physics: Conference Series 1286(1): Kelly KE, Whitaker J, Petty A, et al. (2017) Ambient and
012044. laboratory evaluation of a low-cost particulate matter
Github (2020) The state of the Octoverse. Available at: sensor. Environmental Pollution 221: 491–500.
https://octoverse.github.com/ (accessed 25 May 2020). Khanfar MF, Al-Faqheri W and Al-Halhouli A (2017) Low
Global Runoff Data Centre (GRDC) (2018) Global Runoff cost lab on chip for the colorimetric detection of nitrate
Data Base: Status, development, use. Available at: in mineral water products. Sensors 17(10): 2345.
Kim JS, Alderete TL, Chen Z, et al. (2018) Longitudinal Mew HE, Jr MM, Heath RC, et al. (1997) Cost-effective
associations of in utero and early life near-roadway air monitoring strategies to estimate mean water table
pollution with trajectories of childhood body mass depth. Groundwater 35(6): 1089–1096.
index. Environmental Health 17(1): 64. Mickley JG, Moore TE, Schlichting CD, et al. (2019)
Kumar P, Morawska L, Martani C, et al. (2015) The rise of Measuring microenvironments for global change:
low-cost sensing for managing air pollution in cities. DIY environmental microcontroller units (EMUs).
Environment international 75: 199–205. Methods in Ecology and Evolution 10(4): 578–584.
Kusher D (2011) The making of Arduino. Available at: Mishra AK and Coulibaly P (2009) Developments in
https://spectrum.ieee.org/geek-life/hands-on/the-mak hydrometric network design: A review. Reviews of
ing-of-arduino (accessed 17 June 2019). Geophysics 47(2): RG2001.
Larson KM, Small EE, Gutmann ED, et al. (2008) Use Morawska L, Thai PK, Liu X, et al. (2018) Applications of
of GPS receivers as a soil moisture network for low-cost sensing technologies for air quality monitor-
water cycle studies. Geophysical Research Letters ing and exposure assessment: How far have they gone?
35: L24405. Environment International 116: 286–299.
Lavelle AM, Bury NR, O’Shea FT, et al. (2019) Morrison G, Fatoki OS, Persson L, et al. (2001) Assess-
Influence of urban river restoration on nitrogen ment of the impact of point source pollution from the
dynamics at the sediment-water interface. PLOS Keiskammahoek Sewage Treatment Plant on the
One 14(3): e0212690. Keiskamma River-pH, electrical conductivity, oxygen-
Lewis A and Edwards P (2016) Validate personal air- demanding substance (COD) and nutrients. Water SA
pollution sensors. Nature News 535(7610): 29. 27(4): 475–480.
Lim KY, Vu HN, Sim JY, et al. (2020) The use of Muller CL, Chapman L, Johnston S, et al. (2015) Crowd-
microclimatic data in authentic learning: A two-site sourcing for climate and atmospheric sciences: Current
case study between Hanoi and Singapore. In: Nedkov status and future potential. International Journal of
S, Zhelezov G, Ilieva N, Nikolova M and Koulov B Climatology 35(11): 3185–3203.
(eds) Smart Geography: 100 Years of the Bulgarian Munir S, Mayfield M, Coca D, et al. (2019) Analysing the
Geographical Society. Cham: Springer, 91–104. performance of low-cost air quality sensors, their
Lorenz C and Kunstmann H (2012) The hydrological cycle drivers, relative benefits and calibration in cities: A
in three state-of-the-art reanalyses: Intercomparison and case study in Sheffield. Environmental Monitoring and
performance analysis. Journal of Hydrometeorology Assessment 191(2): 94.
13(5): 1397–1420. Muzzi RW and Eadie BJ (2002) The design and perfor-
Lupascu M, Akhtar H, Smith TEL, et al. (2020) Post-fire mance of a sequencing sediment trap for lake research.
carbon dynamics in the tropical peat swamp forests of Marine Technology Society Journal 36(2): 23–28.
Brunei reveal long-term elevated CH4 flux. Global Overeem A, Leijnse H and Uijlenhoet R (2013) Country-
Change Biology 26(9): 5125–5145. wide rainfall maps from cellular communication net-
McCabe MF, Rodell M, Alsdorf DE, et al. (2017) The works. Proceedings of the National Academy of Sciences
future of Earth observation in hydrology. Hydrology 110(8): 2741–2745.
and Earth System Sciences 21: 3879–3914. Pearce JM (2015) Return on investment for open source
Mead MI, Popoola OAM, Stewart GB, et al. (2013) The scientific hardware development. Science and Public
use of electrochemical sensors for monitoring urban air Policy 43(2): 192–195.
quality in low-cost, high-density networks. Atmo- Piedrahita R, Xiang Y, Masson N, et al. (2014) The next
spheric Environment 70: 186–203. generation of low-cost personal air quality sensors for
Mendell MJ, Eliseeva EA, Davies MM, et al. (2013) quantitative exposure monitoring. Atmospheric Mea-
Association of classroom ventilation with reduced ill- surement Techniques 7(10): 3325–3336.
ness absence: A prospective study in California ele- Plantower (2016) PMS5003 series data manual. Available
mentary schools. Indoor Air 23(6): 515–528. at: https://www.aqmd.gov/docs/default-source/aq-spec/
Metzger M, Konrad A, Blendinger F, et al. (2018) Low- resources-page/plantower-pms5003-manual_v2-3.pdf
cost GRIN-lens-based Nephelometric turbidity sensing (accessed 4 June 2019).
in the range of 0.1–1000 NTU. Sensors 18(4): 1115.
Chan et al. 33
Plaza P, Sancristobal E, Carro G, et al. (2018) Arduino as Sundell J, Levin H, Nazaroff WW, et al. (2011) Ventilation
an educational tool to introduce robotics. In: 2018 IEEE rates and health: Multidisciplinary review of the sci-
international conference on teaching, assessment, and entific literature. Indoor Air 21(3): 191–204.
learning for engineering (TALE), Wollongong, Australia, Tauro F, Selker J, Van De Giesen N, et al. (2018) Mea-
4–7 December 2018, 1–8. New York: IEEE Publishing. surements and observations in the XXI century
Rai AC, Kumar P, Pilla F, et al. (2017) End-user per- (MOXXI): Innovation and multi-disciplinarity to sense
spective of low-cost sensors for outdoor air pollution the hydrological cycle. Hydrological Sciences Journal
monitoring. Science of The Total Environment 607: 63(2): 169–196.
691–705. Tebbs E, Wilson H, Mulligan M, et al. (2019) Satellite Soil
Rocket Scream (2020) Mini Ultra 8 MHz (Arduino Com- Moisture Observations: Applications in the UK and India.
patible). Available at: https://www.rocketscream.com/ Wallingford and Pune: The India-UK Water Centre.
blog/product/mini-ultra-8-mhz-arduino-compatible/ Teixeira D, Maron M and van Rensburg BJ (2019) Bioa-
(accessed 20 May 2020). coustic monitoring of animal vocal behavior for con-
Rockström J, Falkenmark M, Allan T, et al. (2014) The servation. Conservation Science and Practice 1(18):
unfolding water drama in the Anthropocene: Towards a e72.
resilience-based perspective on water for global sus- Terando AJ, Youngsteadt E, Meineke EK, et al. (2017) Ad
tainability. Ecohydrology 7(5): 1249–1261. hoc instrumentation methods in ecological studies
Sayahi T, Butterfield A and Kelly KE (2019) Long-term field produce highly biased temperature measurements.
evaluation of the Plantower PMS low-cost particulate Ecology and evolution 7(23):9890–9904.
matter sensors. Environmental Pollution 245: 932–940. Tetzlaff D, Carey SK, McNamara JP, et al. (2017) The
Schillereff DN, Chiverrell RC, Macdonald N, et al. (2016) essential value of long-term experimental data for
Hydrological thresholds and basin control over paleo- hydrology and water management. Water Resources
flood records in lakes. Geology 44(1): 43–46. Research 53(4): 2598–2604.
SeedStudio (2011) Open source project for radiation Thames21 (2019) Restoring Brent rivers and communities.
detection using Geiger Counter. Available at: https:// Available at: https://www.thames21.org.uk/brent-riv
github.com/Seeed-Studio/Geiger-Counter (accessed 1 ers-and-communities/ (accessed 6 June 2019).
February 2019). Thompson JE (2016) Crowd-sourced air quality studies: A
Sherman DJ, Houser C and Baas ACW (2013) Electronic review of the literature & portable sensors. Trends in
measurement techniques for field experiments in pro- Environmental Analytical Chemistry 11: 23–34.
cess geomorphology. In: Shroder JF (ed) Treatise on United Nations Environment Programme (2020) Urban
Geomorphology 14. San Diego: Academic Press, Air Action Platform. Available at: https://www.unenvir
195–221. onment.org/explore-topics/air/what-we-do/monitoring-
Siragusa EJ and Galton I (2000) Gain error correction air-quality/urban-air-action-platform (accessed 15 April
technique for pipelined analogue-to-digital converters. 2020).
Electronics Letters 36(7): 617–618. Urban MC, Bocedi G, Hendry AP, et al. (2016) Improving
Skarbøvik E, Stålnacke P, Bogen J, et al. (2012) Impact of the forecast for biodiversity under climate change.
sampling frequency on mean concentrations and esti- Science 353(6304): aad8466.
mated loads of suspended sediment in a Norwegian Van de Giesen N, Hut R and Selker J (2014) The Trans-
river: Implications for water management. Science of African Hydro-Meteorological Observatory (TAHMO).
the Total Environment 433: 462–471. Wiley Interdisciplinary Reviews: Water 1(4): 341–348.
Smith RB, Fecht D, Gulliver J, et al. (2017) Impact of Vicente ED, Duarte MA, Calvo AI, et al. (2015) Influence
London’s road traffic air and noise pollution on birth of operating conditions on chemical composition of
weight: Retrospective population based cohort study. particulate matter emissions from residential combus-
British Medical Journal 359: j5299. tion. Atmospheric Research 166: 92–100.
Stevens M, Vitos M, Altenbuchner J, et al. (2014) Taking Vorosmarty CJ, McIntyre PB, Gessner MO, et al. (2010)
participatory citizen science to extremes. IEEE Per- Global threats to human water security and river bio-
vasive Computing 13(2): 20–29. diversity. Nature 467(7315): 555–561.
Wagner RJ, Boulger RW Jr, Oblinger CJ, et al. (2006) Whittier JB, Westhoff JT, Paukert CP and Rotman RM
Guidelines and standard procedures for continuous (2020) Use of multiple temperature logger models can
water-quality monitors: Station operation, record alter conclusions. Water 12(3): 668.
computation, and data reporting (Report No. 1-D3), World Health Organization (WHO) (1996) Water Quality
Techniques and Methods. Reston: United States Geo- Assessments: A Guide to the Use of Biota, Sediments
logical Survey. and Water in Environmental Monitoring. Cambridge:
Ward RC, Loftis JC and McBride GB (1986) The ‘data-rich Cambridge University Press.
but information-poor’ syndrome in water quality moni- World Health Organization (WHO) (2004) Guidelines for
toring. Environmental Management 10(3): 291–297. Drinking-Water Quality: Recommendations (Vol. 1).
Wargocki P, Porras-Salazar JA, Contreras-Espinoza S, Geneva: World Health Organization.
et al. (2020) The relationships between classroom air World Meteorological Organization (WMO) (2019)
quality and children’s performance in school. Building The WMO Global Basic Observing Network – GBON
and Environment 173: 106749. (Executive Summary). Available at: https://www.wmo.int/
Wei Y, Jiao Y, An D, et al. (2019) Review of dissolved pages/prog/www/wigos/documents/GBON/GBON-exsum
oxygen detection technology: From laboratory anal- mary.pdf (accessed 1 May 2020).
ysis to online intelligent detection. Sensors 19(18): Wu MW and Lin YD (2001) Open source software
3995. development: An overview. Computer 34(6): 33–38.

Low-Cost Electronic Sensors For Environmental Research: Pitfalls and Opportunities

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

Low-Cost Electronic Sensors For Environmental Research: Pitfalls and Opportunities

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Low-Cost Electronic Sensors For Environmental Research: Pitfalls and Opportunities

Uploaded by

Copyright:

Available Formats

Main Article

Progress in Physical Geography

for environmental research: Article reuse guidelines:

Pitfalls and opportunities DOI: 10.1177/0309133320956567

Kristofer Chan Francis T O’Shea

I Introduction shown to hamper ground-truthing efforts (Lorenz

Component Description and function

Digital I/O Channels Pins for interfacing with peripherical devices

low-cost water table depth sensor (*£40) that

Plantower PMS3005 and Alphasense NO2-

effectively (r 2 ¼ 0.97, RSME ¼ 0.29 C;

board switches and lights) on reservoir volume

2.6 Time-sequencing lake sediment traps

Table 9. Number of pulse counts per second as

speed, belt-pulley driven, bench drill while

building and calibrating low-cost environmen- generally included in methodological sections

http://www.alphasense.com/index.php/products/ Bramer I, Anderson BJ, Bennie J, et al. (2018) Advances in

El-Jabi N and Caissie D (2019) Characterization of https://www.bafg.de/GRDC/EN/01_GRDC/13_dtbse/

You might also like