Given a good weather forecast for October in Scotland, sunny with a high of 13 °C, I wanted to test out the sensor in a fairly safe place. As I lived in St Andrews before, I knew that there was a manmade pool on Castle Sands that would be deep enough to do some initial tests with the sensor (< 2 m deep) and the water would be reasonable clear (Figure 3.6.1). There is a beach there and the kids were happy playing while I was doing some measuring. It is also a very pretty place and overlooked by the castle that was built around 1400 AD (Figure 3.6.2).
Figure 3.6.1 Castle Sand's pool, St Andrews, Fife, UK
Aim
The aim of this trial were to 1) pull the sensor around the pool to log some GPS data and 2) investigate the quality of the underwater photographs using the standard, lowest cost Raspberry Pi camera.
Method
The sensor was housed in a Draper 38080 small (18.8 x 11.5 x 4.7 cm) waterproof storage case, which was inserted into a foam kick board to give it buoyancy. It is important to note that the bottom of the case (i.e. the camera and depth sensor) must be below the surface of the water when the sensor is floating on the water. The sensor was then powered up and the waterproof case closed. I then towed the floating sensor around the pool for around 10 minutes while the kids were playing on the beach (Figure 3.6.2). The towing had to be slow enough to try and take non-blurry images, but fast enough that the photographs at 10 second intervals did not overlap. An ornamental conch was then placed below the camera to investigate how clear a photo of a conch would be in the deepest part of the pool.
Figure 3.6.2 testing the sensor at the Castle Sand's pool, St Andrews, Fife, UK
Saturday 12th October 2019
My first attempt was made on the Saturday. I towed the sensor around for 10 minutes in the very cold water. Unfortunately, something was amiss with the power bank and the Raspberry Pi did not record anything.
Monday 14th October 2019
Another lovely day in St Andrews was forecast for the following Monday. I took the kids to the beach again and towed the sensor around the pool for 20 minutes, then checked the sensor was collecting data. The photos were being logged, but with movement the leads to GPS had become loose and the locations had not been logged (Figure 3.6.5). The GPS was reconnected and the sensor was towed around the pool for another 20 minutes.
After drying off and warming up a bit, the photographs were downloaded. Individual filenames were copied and pasted into Microsoft Excel. The GPS coordinates were covered from the display format to degrees (i.e. DDMM.MMMM to DD.DDDD), saved as a .csv file and presented in QGIS (https://qgis.org/), a free GIS, where the depth of the pool could be plotted. Also, an example of estimating conch abundance using the ornamental conch is also presented below using the depth of the pool and horizontal/vertical angle of the lens, which were 53.5 ° and 41.4 °, respectively.
Results
GPS data
The GPS coordinates and depth of the photographs were imported into QGIS and presented on Google satellite images (Figure 3.6.3). For this test I walked at around 0.5 m s-1, i.e. 1 step every second. My route was initially up the middle (from the beach to the far end) and then in a rough zig-zag back down to the beach. I had to stop as my son's was getting bored and started to throw stones into the pool.
Figure 3.6.3 Top panels: Location of Castle Sands pool sampling site in St Andrews. Bottom panel: Individual measurement locations in the pool.
Photo quality
Sample photos of the bottom of the pool on Castle Sands show that individual stones and plants can be seen in each of the photos (Figure 3.6.4) in relatively turbid water (I should have brought a Secchi disk) that was 140 cm deep.
Figure 3.6.4 Sample photographs of the bottom of Castle Sands pool, St Andrews
Also the camera also passed the "ornamental conch" test, as the conch can be clearly seen at a depth of 140 cm (Figure 3.6.5). Unfortunately, the GPS sensor become detached for the Arduino, hence the odd message across the top of the image. Also, the length of the conch can be estimated using the ratio of length of the shell in the photograph to the length of the photograph multiplied by the actual width of the seafloor in the photograph (as shown below). In Figure 3.6.5 the conch is 2.4 cm, the photograph is 8.4 cm and the actual width of the seafloor is 100 cm, the calculated length of the conch is 28 cm while its actual length in 26 cm.
Figure 3.6.5 Example calculating conch density using an ornamental conch shell
Conclusions
The sensor in this configuration seemed to work well when plotting GPS data and taking photographs in water 140 cm deep. There was not much variability in the pool depth and it would be more interesting to see how well quickly the depth sensor changes. I would assume that the waters around Cayman are less turbid than the Castle Sands pool and it is clear then this needs to be tested in-situ. One area of concern was the quality of photographs as the sensor is towed. However, the photographs taken during this trial (Figure 3.6.4) suggest that at low tow speeds, approximately 0.5 m s-1, the images remain relatively sharp and air bubbles do not affect the photo.
One potential shortcoming of the current system is the size of the sample area. Currently, Cayman DoE use a quadrat of 24 square meters, this area is only visible by the current camera if the depth of the water is 5.9 m. Using the current camera, the quadrat area visible at 140 cm is only 0.77 square metres and a “fisheye” camera may be more useful. However, the high frequency of the photographs also mean that 360 quadrats can be measured in an hour compared to the 3 measured currently by the DoE and the total sample area will depend on the depth of the water. It should be reiterated that the current depth sensor will not work below depths of 50 cm.
Another weakness is this system as compared to the current method used by the DoE is the data on conch length and lip thickness are not measured. Here I must acknowledge only partial success, as an estimate on conch length can be made using the photographs but there is no way this method can be used to measure lip thickness of the shell. Despite some flaws, this sensor seems to go some way to providing a method to reduce the uncertainty in the abundance estimates in queen conch. As with all sensors, more test are required to make sure that they are fit for purpose.