The ability to collect high quality environmental data inexpensively using open source equipment aids transparency, scientific reproducibility, and lowers cost barriers in science. Low-cost open source and open hardware environmental sensors are proliferating and becoming more reliable. Examples include a water quality sensor developed by Bas Wijnen et al (2014) that has been technically verified against professional equipment; and the Sensor.Community air quality project, currently producing air quality data globally from open source particulate matter equipment (https://sensor.community/en/).
Our development, the Sonic Kayak system, sets itself apart from these projects as the environmental data it collects is automatically geolocated and can be collected while moving, allowing very quick mapping without the need for multiple sets of equipment. A similar initiative, the Smartfin (https://smartfin.org/) is close in its aims, but is specific to surfboards and stand-up-paddleboards; its electronics are embedded within a fin and currently only measures temperature; it is not open source so it is not available for people to build or develop themselves (Brewin et al. 2020). The Sonic Kayak is an open source technology project for gathering environmental data from marine, estuarine, river or lake environments, which can then be mapped at a fine scale. The equipment is designed to fit onto any model of kayak.
The Sonic Kayak Version 1 (Griffiths et al. 2017) was equipped with temperature sensors and a hydrophone, recording the water temperature every second and underwater sound continually, together with the GPS coordinates, time and date. Sonification, the process of conveying information by using non-speech sounds (Dubus & Bresin 2013, Hermann et al. 2011), was an additional feature of the original version. Data is sonified in real time through an on-board speaker, allowing the paddler to hear the data as it comes in, and to identify or follow interesting occurrences such as temperature gradients.
The Sonic Kayak Version 1 sensors for water temperature and underwater sound offer new opportunities for data collection in climate change and anthropogenic noise pollution research. Anthrophony (human-made underwater noise) has been increasing since the industrial revolution and originates from a range of sources (Duarte et al., 2021), the most pervasive of which is from shipping. Vessels produce underwater noise at a range of frequencies depending on their size; for large ships this noise is predominantly below 100 Hz, while small boats usually produce underwater noise at higher frequencies, with concentrations around 1 kHz (Figure 2 in Duarte et al., 2021). Sea temperatures are typically measured using static sensors attached to buoys or other structures, or using low resolution satellite observations, while underwater sound is usually collected using sensors attached to the seabed, drifting devices, or towed sensors on motored research vessels (these sensors are discussed in more detail in Griffiths et al. 2017). The Sonic Kayak allows such data to be collected at fine scales that are difficult or impossible to obtain using existing off-the-shelf research equipment, and in areas that can be hard to reach including shallow waters or close to shores/cliffs.
Since publishing the original Sonic Kayak design, we have had requests to add a turbidity sensor to measure the cloudiness of water caused by particulates. These requests came from different sources: a harbour master wanting to monitor algal blooms in an EcoPort, a water company aiming to monitor cyanobacteria, and seaweed farmers looking for water quality readings. Similar applications could include mapping farm run-off or sewage outflows. We were also interested in incorporating additional features, including measuring particulate matter pollution in the air low over the water caused by boat engines and industry. This type of air pollution is not well studied, but has the potential to affect water users like swimmers or kayakers, as well as wildlife such as waterbirds or cetaceans (e.g. Rawson et al. 1995).
This article outlines the design of the new turbidity and air quality sensors and how to fit them into the Sonic Kayak, together with the new data sonification approach for these sensors. In addition we present proof-of principle data collected using Sonic Kayaks, including maps for water temperature, underwater sound, water turbidity and air particulate pollution. Fine-scale mapping of environmental data in difficult-to-reach areas like estuaries and close to coasts is an exciting step for professional researchers, however we believe the system also offers particularly interesting opportunities for citizen science and community-driven environmental activism. The Sonic Kayak system is highly flexible, and could equally be used on land, for example to collect particulate pollution data while walking or cycling around a city.
Hardware and Software
General design considerations
Version 1 of the Sonic Kayak (Griffiths et al. 2017) was designed to have the electronics, sensors, and two speakers all located at the front of the kayak. Since then, we worked with Access Lizard Adventure, an accessible kayak club, to redesign the Sonic Kayak system for people with visual impairments, to enable greater independence on the water. As part of that redesign we produced Version 2, moving the electronics and sensors to the back of the kayak and switching from cabled speakers to a single waterproof Bluetooth speaker; this speaker was moved to the middle of the kayak where there is usually a recess for holding a water bottle. These changes were largely introduced to reduce the risk of entanglement in the cabling, which is beneficial to every paddler. This new design also has the benefit of being quicker and easier to attach to the kayak, as the main electronics box now sits in the storage compartment at the back of the kayak and simply clips onto any available D-ring or strap that the kayak model has. During the Version 2 redesign process we also improved the waterproofing by using a watertight electronics box fitted with marine-grade cable glands; the box is IP66/68 rated, meaning it is resistant to full submersion under 1.5 m of water for up to 30 minutes.
The current design, Version 3, has brought its own challenges due to the addition of the new turbidity and air quality sensors. Two of the key design considerations for the new water turbidity sensor were (i) to keep it at a fixed depth to ensure consistent data, and (ii) to reduce the light from above the water that could affect the data collected. To achieve this, we built a structure that attaches to the side of the kayak, allowing the paddler to easily lift it out of the water when needed. With previous versions we were aware that the temperature sensor also suffered from not being at a fixed depth; in this version we addressed this issue by fixing it to the same rig as the turbidity sensor. The air quality sensor, by definition, needs to be open to the air, making full waterproofing impossible. If the kayak capsizes, this sensor would be destroyed without affecting the rest of the equipment; this is why the air quality sensor needed to be as low cost as possible and separate from the main electronics.
Together this means that the Sonic Kayak version 3 system is mainly housed in a waterproof box on the back of the kayak, with a small external box containing the air quality sensor on top, a hydrophone trailing from the back of the kayak, and a rig strapped using Velcro to one side handle of the kayak holding the temperature and turbidity sensors, together with a separate waterproof Bluetooth speaker that can be placed anywhere on the kayak (Figures 1, 2 and 3).
The sensors we chose needed to record and transmit data digitally and continually. Many commonly used research-grade sensors are designed to be deployed and later brought into a lab or office for the data to be downloaded, as such they store the data internally but do not output it in real time. These sensors are typically proprietary and cannot easily be modified, and are often remarkably expensive. For example, one unit of the very commonly used HOBO TidbiT water temperature logger costs $313.50, requires a $292.60 base station to download the data, and $234.30 software to view the data – a total of $840.40 (AUD; roughly £467 at the time of writing).
Detailed plans for building a complete Sonic Kayak (version 3) can be found in the project wiki https://github.com/fo-am/sonic-kayaks/wiki, and the new sensor design decisions are summarised in Table 1 and described in more detail below. The software is open source and available on GitHub https://github.com/fo-am/sonic-kayaks.
|ENVIRONMENTAL VARIABLE||SENSOR USED||REASONS FOR CHOICE||SPECIFICATIONS||COST|
|Temperature (under water)||DX Waterproof DS18B20 temperature sensor||This is a widely used, reliable, fast, accurate, and cheap waterproofed sensor.||Usable temperature range: –55 to 125°C (–67°F to +257°F), 9 to 12 bit selectable resolution, uses 1-Wire interface – requires only one digital pin for communication, ±0.5°C accuracy from –10°C to +85°C, query time is less than 750 ms.||Approx. £3|
|Sound (under water)||DolphinEar DE-PRO balanced hydrophone||Hydrophones are complex and vary enormously in quality and price; we are not aware of a way to build reliable DIY versions.
The DolphinEar is a mid-quality hydrophone with good balance between capability and cost.
It is straightforward to replace this sensor with a better or cheaper equivalent.
|Balanced Output: 600 ohms (approx.)
Frequency Range (Overall): 1 Hz – 24000 Hz
|Turbidity (under water)||Standard LED (White Clear Lens 3 mm LED 3000 mcd 20 mA) and LDR (8k–24 k 1M dark resistance. 5mm diameter) in custom-designed housing.||Easy to obtain, standard, and very cheap components. Custom housing is used to reduce external sources of light.
Pre-made sensors like the DFROBOT SEN0189 could be used instead, however full waterproofing and custom housing to exclude light and attach to the kayak would still be necessary.
|Measures voltage from the LDR. Higher voltages indicate clearer water with low turbidity, and lower voltages indicate higher turbidity.||Approx. 70p for the LED and LDR together.
Custom housing costs if home-printed are approximately £3.77 in printer filament (based on needing ~160 g of matte black recycled PLA from Filamentive).
Additional costs include resin and Sugru for sensor waterproofing.
|Air quality – particulate matter (above water)||PMS7003 (manufactured by Plantower)||Low cost, shown to be less affected by humidity than other similar models (Badura et al. 2018), and less affected by temporal drift (Bulot et al. 2019).||Measures PM1/2.5/10 concentrations (μg/m3) at ‘standard particle’ and ‘under atmospheric environment’, and the number of particles of each size in 0.1 litres of air. Measurements are once per second.||Approx. £24|
|Location (GPS/GNSS)||GlobalSat BU-353 S4||Water resistant, USB connection is simple, outputs data in NMEA (universal format – easy to extract location and position quality information).||48 channel receiver capable of producing NMEA data every second. Utilises the GPS satellite constellation. Horizontal positional accuracy of <2.5 m.||Approx. £45|
Particulate sensor design
Low cost sensors for measuring air quality (specifically particulate matter) have been implemented very successfully for citizen science in many settings, perhaps most comprehensively by Sensor.Community (https://sensor.community/en/). Reliable linear relationships have been demonstrated for various low-cost particulate matter sensors tested against professional research grade equipment (e.g., the TEOM™ Continuous Ambient Particulate Monitor from Thermo Scientific), however the low-cost sensors do appear to routinely over-estimate particulate levels (Badura et al. 2018). Low cost PM2.5 sensor measurements can be affected by humidity, mainly from fog as opposed to rain, as the water particles are similar in size to the pollution particles (Jayaratne et al. 2018) – which is of course a concern for use on water.
Of the four low-cost models of particulate sensors tested by Badura et al. (2018), the PMS7003 (manufactured by Plantower) and SDS011 (manufactured by Nova Fitness and used by the Sensor.Community project mentioned above) appear to perform best in terms of data replicability and comparisons with research-grade equipment. However humidity levels were shown to affect SDS011 such that the data was less comparable with the research-grade TEOM data the higher the humidity. This did not appear to be an issue with PMS7003. Temporal drift can also be a problem with low-cost PM sensors either from dust accumulation or the degradation of electrical components, but Bulot et al. (2019) saw no evidence for drift over time for the PMS7003 sensor. Because of these issues, we decided to work with the PMS7003 sensor instead of the more widely used SDS011 sensor.
We designed a custom 3D printable housing for the PMS7003 sensor to protect it from rain and splashes (Figure 4), including downward pointing tubes aligned with the inflow and outflow of the sensor which allow air in/out, but will stop ingress from water (unless submerged). Files for 3D printing are available via GitHub https://github.com/fo-am/sonic-kayaks/tree/master/hardware/3dp (pm-box is as pictured in Figure 4, pm-box-home-print is a version with wider air input/output nostrils that is also better suited to printing on a standard extrusion printer).
The PMS7003 sensor outputs particulate matter concentrations for PM1/2.5/10 at ‘standard particle’ (μg/m3; the concentration corrected to a standard atmospheric pressure) and ‘under atmospheric environment’ (μg/m3; the concentration in the air as it is when the sample was taken, which can be useful for example when taking measurements at high altitudes as air pressure changes the concentrations), as well as the number of particles of each size in 0.1 litres of air. For the purposes of this paper we have used the standard particle levels, having checked that they do not differ substantially from the atmospheric equivalent values at sea level.
We believe the best approach is to consider only relative values rather than to expect accurate absolute values, to primarily use the system for studying spatially localised particulate matter concentrations, and where possible to take multiple readings for the same area to confirm replicability. If absolute particulate matter concentrations are needed, we recommend calibrating the air quality sensor using professional-grade equipment that includes a drying mechanism for the air sample, or against an Automatic Urban and Rural Network (AURN) station (UK) or equivalent (there are currently no AURN stations in the over 100km of land west of Plymouth in the UK, which is the region where our trials took place).
Turbidity sensor design
Turbidity sensors give a measurement of the amount of suspended solids in water – the more suspended solids, the higher the turbidity (cloudiness) of the water. The most basic approach to measuring water turbidity is to use a Secchi disk. These are plain white or black and white circular disks that are lowered into the water, and the depth at which the disk is no longer visible is an approximate measure of the cloudiness of the water. This is a great low-key approach, but the result is affected by other factors such as the amount of daylight (Preisendorfer, 1986). More accurate equipment tends to use an enclosed light source and a light receptor, with the water placed in between, so that the amount of light that reaches the receptor from the light source gives a reading of how turbid the water is. There are several pre-existing publications on how to make open source turbidity sensors (Román-Herrera et al. 2016; Public Lab, 2017), and similar sensors to these have been shown to be able to detect different species of algal blooms (Parra et al. 2018).
For the Sonic Kayaks, a bespoke sensor assembly was needed due to the combined requirements of waterproofing, light proofing and robustly mounting the sensor under a kayak hull. Sonic Kayaks sonify sensor data in realtime, recording data continually. As such, we needed a turbidity sensor that was able to log real-time data and provide a digital output that could be integrated into the existing Sonic Kayak kit, as opposed to a conventional system where a one-off sample of water is taken and run through a separate piece of equipment in a laboratory. We based our design on Bachler’s (2019) prototype, which uses a Light Emitting Diode (LED) and Light Dependent Resistor (LDR) housed within a piece of hose pipe – the more cloudy the water is, the less light from the LED will reach the LDR.
We designed a custom 3D printable turbidity tube, suspended from a reinforced nylon webbing rig that attaches to the kayak handle using Velcro. This means that the sensor is kept at a reasonably fixed depth of approximately 60 cm below the waterline but can easily be lifted out of the water and attached/removed where the depth is too shallow. The rig also provided an opportunity to mount the temperature sensor at the same fixed depth. The turbidity tube itself was printed using selective laser sintering for strength, with a matte black nylon plastic to block light, and was designed with slatted end caps to allow water to flow freely through while keeping debris out of the tube, as well as to further minimise light entering the tube. The files for 3D printing are available via GitHub https://github.com/fo-am/sonic-kayaks/tree/master/hardware/3dp (turbid-tube and turbid-cap are suitable for laser sintering prints, or turbid-tube-home-print and turbid-cap-home-print are better suited for low cost extrusion printing). The end caps are held in place with silicone sealant to make them easy to remove for cleaning. The LED and LDR were set into small clear resin blocks for waterproofing (Figure 5a), orientated over component-sized holes drilled into the turbidity tube, and fixed in place with matte black Sugru to block out the light (Figure 5b). The cables run up the nylon webbing rig such that there is no pressure on any of the cable joints (Figure 5c).
The data output from the turbidity sensor is not completely straightforward. LDRs decrease resistance with light intensity so when more light hits the sensor, the voltage increases, leading to a higher numerical output. The numerical output is related to the voltage coming in, with an analogue to digital conversion (10 bit) applied such that 0 V = 0 and 3.3 V = 1023. If required, it is possible to do a lookup from the specific LDR sensor curve data to work out the voltage from the numerical output. As with the air quality sensor, the turbidity sensor can easily be calibrated using professional research equipment, and without calibration it is suitable for obtaining relative values and identifying spatial areas with high/low turbidity.
The light levels above water may influence the readings despite our efforts to block extraneous light, however tests where the kayak was paddled under the shade of low bridges indicated that this level of change in above water lighting conditions was not noticeable on the output, and similarly two different Sonic Kayak systems paddled around the same area on different days and at different times of day under different weather conditions produce remarkably similar results (Figure 6).
The particulate, turbidity and temperature sensors were connected to an Arduino Nano microcontroller, which was attached to the main Raspberry Pi system. The additional Arduino Nano provides analogue to digital conversion for the turbidity sensor, which the Pi lacks, and frees up a connection (UART) on the Pi (required for the PMS7003 and the GPS sensors). In this way, the Arduino connects to the various digital and analogue interfaces for each sensor, collects the data and provides it to the Raspberry Pi in a convenient form.
The 10 bit analogue to digital converter (ADC) on the Arduino uses the battery voltage as its reference, but is still susceptible to fluctuations on the power supply coming from other parts of the circuit. We found that the PMS7003 causes high frequency interference (either from the motor air pump or laser) which affected our turbidity measurements. We were unable to alleviate this interference with either the addition of extra bypass capacitors or via software filtering. In order to sidestep this issue, we found we could use the command interface on the PMS7003 to put the sensor to sleep for ten seconds while reading the turbidity sensor. A small delay is required for the PMS7003 to fully power down before reading a clean value from the ADC; ten seconds is an adequate duration that allows air to be drawn into the particulate sensor to get accurate periodic readings.
Once read by the Raspberry Pi over an i2c serial interface, the data is (i) appended to a log file stored on a USB stick in CSV format to be analysed later and (ii) sent to the sonification system, alongside error conditions so the paddler is aware of changes in the data or problems with any of the sensors.
Python scripts read and log sensor data and send it to Pure Data, an open source visual programming language for multimedia, for sonification. The GPS data is interpreted by an audio zone map to indicate areas for sampling (see below), playing indicator sounds when the user crosses into polygons pre-defined on the map. Due to the unpredictable nature of Bluetooth radio transmission, a watchdog process detects problems that can affect the hydrophone recording, such as the speaker being out of range, and can restart and reconnect to the speaker. Underwater recordings are split into 26 Mb (5 minute) mono WAV files and written to the USB stick, with sequence and trip IDs embedded in the filenames.
The Sonic Kayak version 1 sonification process consisted of a live-feed from the hydrophone so the paddler could hear the underwater sounds, together with a simple synthesised tone for the temperature sensor. The tone went up in pitch in higher temperature water, and down in pitch in lower temperature water, only making a sound when the temperature changed rather than continuously. For version 3, adding two new sensors might seem straightforward, but in terms of the sonification this required three separate sounds (representing temperature, turbidity, and air quality) that could easily be distinguished from each other, sounded good together and with the hydrophone feed, and didn’t overwhelm or annoy the listener/paddler. To handle the complex task of sonifying multiple data streams in different ways, we built a custom modular sonification system in Pure Data that allowed us to use different synthesis and sample manipulation techniques on separate sounds.
To guide our decision making, we made four separate types of sonification for the set of sensors, and produced an online survey to see which type people preferred. We also made mock environmental data sets where the data for each sensor followed either rising/falling/rising then falling/falling then rising patterns, and tested survey respondents to see whether they could tell what was happening with the environmental data just from the sonification. The questions and survey response data from 49 survey participants is available on Zenodo (https://zenodo.org/record/3923743#.X0j2OHWYVH4). We chose simple synthesised sounds which scored second for favourability (mean score = 3.41, on scale of 1–5), and scored first for interpretability (73% correct answers). A demo track with these sonifications is also available at: https://archive.org/details/skdemo_20200915.
Error detection is another important aspect of the audio feedback for citizen science use cases. As trips can last multiple hours, the paddler needs to be aware of problems as soon as they happen, rather than returning to base to discover the data is partially or completely unusable. To address this, if a sensor stops working the system uses synthesised speech to warn the paddler immediately that there is a problem. This approach makes it much easier for the user to keep track of potential problems, rather than using lights or a visual display while in the water. For some sensors, there are specific error states that need to be reported, such as GPS “no fix” error, where not enough satellites are acquired; in this case the system emits an audible “ping” sound, warning the user of the bad quality of positioning data.
To demonstrate the potential of the Sonic Kayaks, we performed tests in two locations in Falmouth Bay, Cornwall, UK (Figure 8), which is a Special Area of Conservation (SAC) and a Special Protection Area (SPA; for birds). The Bay is used by cargo shipping, cruise ships, fishing and recreational boating. The first location was Falmouth Harbour, where tests were performed on 29th July 2020. To build a robust case study we selected highly variable routes; the test area covered a range from a river estuary through to more open sea, including zones where people live on houseboats, areas adjacent to agricultural fields, an industrial dock, town, and marinas of various sizes. For this test we used three Sonic Kayak systems, each with turbidity, temperature, and air quality sensors. The second location was the Helford River, where tests were performed on 7th August 2020. The area is a flooded river valley categorized as a Site of Special Scientific Interest within the Cornwall Area of Outstanding Natural Beauty. It is typically considered a very quiet, clean and beautiful location, attracting holiday makers and leisure craft users (particularly yachts and kayaks). In this case we tested one Sonic Kayak system, with turbidity, temperature, air quality and hydrophone sensors. The locations included eelgrass and Maerl beds habitats, which are features protected within the SAC designation (JNCC 2015).
Data from the temperature, turbidity, and air quality sensors do not need any post-processing. Underwater sound data from the hydrophone is more complex; although it is collected continually, to obtain good quality recordings it is best to avoid any noise from the kayak itself, including noise from paddling, or drips coming from the paddles. This paddling noise is in a broad frequency range (1 – 6 kHz), with loudest sounds around 2 kHz. An example clip can be found at https://soundcloud.com/jokat/paddling-clip-sonic-kayak. To address this issue, during the tests we decided to stop kayaking at semi-regular intervals, roughly every 200 m, to gather a clean sound sample. To facilitate this, we made use of a feature built into the original Sonic Kayak system that allows users to define spatial zones in a map and assign them sounds. Triggered by GPS, these sounds were automatically played through the speakers when the kayak entered the zones. We placed zones approximately 200 m apart around the path that we were planning to paddle the kayaks, so that an alert ping sounded to prompt us to stop for a minute. Developed from the previous, more ad-hoc approach where samples were not taken periodically (Griffiths et al., 2017), this method enables a more systematic approach to sampling.
The .wav files were opened as spectrograms using the freely available open-source sound software Audacity (version 2.2.2) to aurally identify periods of time without paddling noise. The start and end time of these sampling periods were recorded with a 1 second temporal resolution. The samples were analysed in RStudio (version 1.2.5001; RStudio, Inc.) with R version 3.6.3 (R Development Core Team, 2020) using custom scripts adapted from the open source code PAMGuide in Merchant et al., (2015) and the package “tuneR” (version 1.3.3; Ligges 2018) to allow analysis only of the identified paddling-free clips. Merchant et al., (2015) includes a tutorial for the use of PAMGuide, which can be used to produce relative sound levels of an audio clip. All the code used for our sound analysis is available via GitHub (https://github.com/fo-am/sonic-kayaks/tree/master/analysis).
A correction factor was calculated by recording reference sine waves (10 Hz, 100 Hz, 1 kHz and 10 kHz) of known voltage measured using an Amprobe PM51A Pocket Digital Multimeter, through the entire system (Tascam iXZ pre-amp, TechRise USB sound card and Raspberry Pi) and comparing the output with the known input. The results were then interpolated every 1Hz to obtain the correction factor. The hydrophone sensitivity, provided by the manufacturer as a range (–180 to –183 dB; approximately flat 20 Hz – 20 kHz; although not calibrated), was specified in our processing as –181.5 dBV μPa–1. The correction factor and hydrophone sensitivity allow absolute sound values to be calculated; however, these should be interpreted with some caution as the hydrophone sensitivity is not calibrated.
To explore how the sound level varies according to changes in frequency, and to apply the frequency dependent corrections, fast Fourier transforms were applied to the sound samples in segments of 1 second duration (Hann window, 50% overlap). As a result, we obtained the sound power in 1 Hz bins. Several metrics were calculated by summing these values within various frequency ranges per second.
The broadband sound pressure level (SPLRMS; 20 Hz – 19 kHz) represents the overall sound level for the full frequency range effectively recorded with the hydrophone. Variations in sound levels according to changes in frequency (pitch) are represented by third octave levels (TOL). An octave represents a doubling in acoustic frequency from the lower limit to the upper limit of the band (frequency range) and third octave bands are 1/3 of an octave wide. The third octave bands are described by the centre frequency and the bandwidth increases with increasing frequency.
The sound power values were passed to the mapping script, averaged over space (time may vary) and then converted to decibels as the final step once all processing and averaging was completed. Measurements taken underwater are different to those taken in air; underwater monitoring uses a reference pressure of 1 µPa. The final units for our data are dB re 1 µPa.
Our figures present the broadband sound level and the third octave levels centred on 125 Hz, 1 kHz and 10 kHz, representing the underwater sound environment captured by the Sonic Kayak. The 125 Hz band is, along with the 63 Hz band, an indicator of shipping noise in the marine strategy framework directive (MSFD; Tasker et al., 2010). Previous research found the 125 Hz band to be louder in Falmouth Bay than the 63 Hz band (Garrett et al., 2016). We included the higher frequency bands (1 kHz and 10 kHz) to explore the contribution of sound in these frequency ranges. We were interested in these because the low frequencies of large commercial ships do not propagate in shallow water, and the small boats used in shallow coastal areas and estuaries produce sounds at these higher frequencies (Duarte et al., 2021).
The noise floor of the system was estimated by leaving the Sonic Kayak system to record for five minutes in a quiet space out of the water. The analysed recordings were above the noise floor for >99.9% of the time for all presented measures, with the exception of the 125 Hz third octave level (~99.7%).
Data collected with the different sensors on the Sonic Kayak are time-stamped and therefore matching them up is easily achievable. Each data point also has a position in space as provided by the on-board Global Navigation Satellite System (GNSS). The following processing and figure creation was conducted using the R software environment (R Development Core Team 2020), with the following packages: cowplot (Wilke 2019), dplyr (Wickham et al. 2020), ggplot2 (Wickham 2016), lubridate (Grolemund & Wickham 2011), purrr (Henry & Wickham 2019), raster (Hijmans 2020), readr (Wickham et al. 2018), RStoolbox (Leutner at al. 2019), sf (Pebesma 2018) & tidyr (Wickham & Henry 2019).
Regular 50 m wide grids were created spanning the extent of GNSS data for the Falmouth Harbour and Helford testing sites. These hexagon polygons were then clipped to a mean high water line freely available from Ordnance Survey (https://osdatahub.os.uk/downloads/open/BoundaryLine). Next, the sensor data were joined (using time as a common variable) to the GNSS data to create a spatial point dataset. These points were then intersected with the regular hexagon polygons, and the mean values of the sensor measurements were calculated and assigned to the hexagon. This process was repeated for each of the sensors. Not all of the sensors were simultaneously recording throughout the whole kayak trip (e.g., in the case of turbidity and air quality, the sensors alternated, as mentioned in the software section above). Absent data were treated as NA values and had no influence on the calculated means. If a hexagon intersected points only with NA values, then it was assigned an NA. These hexagons appear hollow in Figures 9, 10, 11, 12, indicating the path of the kayak, but the absence of data. Lastly, spatial data was projected to British National Grid (EPSG 27700) and overlaid on mosaicked aerial images (Sources: Esri, DigitalGlobe, GeoEye, i-cubed, USDA FSA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community). All data used in the visualisations are available via Zenodo (https://zenodo.org/record/4041588) and the code used is available via GitHub (https://github.com/fo-am/sonic-kayaks/tree/master/mapping).
The data visualisations from the test trips provide an indication of the potential of the Sonic Kayak system for fine-scale mapping, particularly close to the coast, but also in rivers and lakes.
Spatial and temporal variation are both evident. For example, at Falmouth Harbour, the particulate matter map shows a spike in particulate pollution in the air, North of where ‘The World’ cruise ship was moored and running its engines despite being stationary, with the wind coming from the South at the time of sampling (see Figure 9 for PM2.5, and Supplementary Material 1 for PM1 and PM10).
The turbidity sensor, which is perhaps the most likely to be problematic as it is not an off-the-shelf sensor, shows good replicability from multiple trips at Falmouth Harbour. Higher levels of water turbidity were detected in muddy tidal areas, and lower at a river outflow (Figure 6 for detailed version, Figure 10 for broader geographical area).
Our test data from Helford and Falmouth Harbour show water temperature variation at very fine scales, with warmer water in shallow and sheltered areas and colder water in deeper and more exposed areas (Figure 11 and Supplementary Material 1).
The seas around Cornwall have been modelled to have high levels of underwater noise (Farcas et al., 2020). However, the water depth in our Helford test area was no more than 10 m. Such shallow waters prevent the long-distance propagation of low frequency noise such as that generated by commercial shipping traffic, which is typically the focus of international concern and management (Tasker et al., 2010; Farcas et al., 2020; Merchant et al., 2016). Our results highlight the contribution of higher frequencies to the overall sound levels and the importance of small, recreational vessels in shallow coastal waters (Figure 12). The importance of these recreational vessels, which don’t carry tracking systems such as the Automatic Identification System (AIS) required on larger or passenger vessels, was also highlighted in shallow coastal waters in Denmark (Hermannsen et al., 2019). Our example data in the Helford test area indicates the precise geographical areas that are most affected by underwater noise, in particular towards the South where the majority of the boats travel (Figure 12 and Supplementary material 1).
We have demonstrated the capabilities of the Sonic Kayak system for gathering fine-scale data from a number of different environmental sensors. We believe the examples show potential both for professional researchers, enabling the collection of types of data that were otherwise difficult or impossible to obtain at a fraction of the cost of conventional research equipment, as well as for citizen scientists, activists, conservation groups or environmental managers to gather data for their own purposes. For example, by using a Sonic Kayak to gather underwater noise data, a conservation group or harbour manager could make better informed decisions on whether it might be necessary to limit the areas that boats can use, or the number of boats, type of boats or speed allowed in a specific area. We see particular opportunities for communities to use the system to highlight poor practice and lobby for environmental protection.
The sonifications allow paddlers to follow routes of interesting data. Users might hear sounds from the Sonic Kayak indicating a temperature gradient and choose to paddle around an area to gather more detailed information in that particular location, or hear underwater noise and decide to stop paddling and take a longer sample in a particular place. It would also be straightforward to modify the sonification approach in future versions, to notify the paddler when the data collected have crossed particular thresholds such as ‘safe’ particulate matter concentrations (10 μg/m3 annual mean, 25 μg/m3 24-hour mean are considered important thresholds for PM2.5; World Health Organisation 2005).
The kit is very quick and uncomplicated to deploy, and the system is flexible enough to adapt to different sampling designs. Spatial and temporal replication are straightforward to achieve, for example to see how water temperatures vary over the year, or even to detect longer term trends caused by climate change.
The cost of the whole system, including all sensors (temperature, turbidity, air quality and hydrophone) is approximately £1,100; this cost assumes access to a 3D printer, electronics tools, and other basic workshop tools that would typically be available in any community Makerspace. A significant part of this cost is taken up by the hydrophone, which can be obtained at approximately £300 (DolphinEar pro), and the waterproof bluetooth speaker, costing around £140 (Ultimate Ears Megaboom 3); these two components may be over or under the required specifications depending on the user’s aims. A full breakdown of indicative costs is indicated on the project wiki https://github.com/fo-am/sonic-kayaks/wiki.
We recommend that the sensor systems should be used for approximate, relative data mapping, rather than for gathering absolute values required for regulatory purposes or particular research applications. For professional researchers or those seeking data to guide policy, the system can be a useful tool for gathering data to inform the design of more precise secondary studies, or for obtaining more spatially or temporally detailed data following sensor calibration.
The data obtained using the system is very straightforward to handle, with the exception of the underwater sound data which requires more expertise and the manual selection of periods that are clear of paddling noise. In the future, it would be possible to modify the system to include on-board data processing for the sound data, to lower this barrier.
The frequency range of underwater sound recording is also a current limitation. Many sounds underwater extend into higher frequency ranges, including those from echosounders, vessels and marine fauna such as dolphins and porpoises (Duarte et al., 2021; Veirs et al., 2016). However, recording at a higher sample rate to capture the higher frequencies would also require a more expensive hydrophone as well as increasing the data storage and power consumption requirements of the Sonic Kayak. The frequency range currently covered by version 3 is considered to encompass most of the underwater sounds of interest in such shallow coastal environments, such as vessel noise and sounds from marine life including bivalves, crustaceans, fish, seals and many sounds from cetaceans (Duarte et al., 2021). Nevertheless, the frequency range could be further developed in the future, in particular where kayaking, dolphins and porpoises are likely to overlap.
In terms of the hardware, we are confident that the design is robust and highly waterproof, however entanglement and drag remain an issue with the turbidity/temperature sensor rig. While on the water, users can easily deal with any entanglement issues as the rig can be reached from the kayak seat; if retaining the underwater depth of these sensors is not an issue for the user then the sensors could be fitted close to the hull of the kayak using straps around the hull. An alternative prototype turbidity tube design allowing for this arrangement is available at https://github.com/fo-am/sonic-kayaks/blob/master/hardware/3dp/turbid-tube-home-print-under-hull.stl When using this design, we recommend placing a piece of 1cm thick neoprene between the tube and the hull to cushion the turbidity sensor components.
We made a video about the project (https://www.youtube.com/watch?v=puLXKj1AVAk) and launched a survey to get a broader range of feedback on the project (responses available in full at https://zenodo.org/record/4032599). Respondents highlighted the diversity of uses of the equipment, including its potential for informing decision making in conservation areas, for water companies, landowners or Government bodies to map pollution around the coastal sewer outlets and rivers, for connecting people to the quality of ‘their’ environment and promoting a feeling of engagement/ownership, for educational uses with children and others, and also for monitoring the use and noise distribution of Acoustic Deterrent Devices (a typical source of high frequencies in open cage salmon farms). One respondent stated that its main use was to allow ‘a more democratic and participative approach to looking after our natural environments’.
Respondents also suggested future developments could include additional sensors for pH, oxygen levels, and salinity. Others were interested in adding an underwater camera, for example to capture footage of pelagic marine species and pollution sources (discarded fish nets, waste, etc.) as qualitative data to complement quantitative data collection efforts.
The Sonic Kayak system has already been adapted for use on bicycles, to record and sonify air pollution levels in cities as part of a new sound art project led by Kaffe Matthews and the Bicrophonic Research Institute (https://sonicbikes.net/environmental-bike-2020/; https://github.com/bicrophonics).
From our perspective, we would like to further develop the Sonic Kayak system to make an off-the-shelf unit that could be purchased at low cost for those who are unable to make their own, and we would like to develop an online mapping portal for hosting, visualising, and sharing the data collected by users, potentially in real-time.
Data accessibility statement
Sonification survey data: https://zenodo.org/record/3923743#.X0j2OHWYVH4.
Sensor data from test trips: https://zenodo.org/record/4041588.
Feedback survey data: https://zenodo.org/record/4032599.
Hardware plans: https://github.com/fo-am/sonic-kayaks/wiki.
Data visualisation code: https://github.com/fo-am/sonic-kayaks/tree/master/mapping.