2.1.2. Underwater Spectroscopic Techniques

Spectroscopic techniques can image in numerous, narrower wavelength bands across the whole visible light spectrum and mark an improvement over simple RGB imaging. The use of spectral data enables a more definitive discrimination between live coral, macroalgae, and other photoactive organisms by using the specific spectral "signature" or "fingerprints" associated with a certain organism or type of organism [60,61]. It also identifies whether corals are displaying a decline in 'normal health' by measuring the relative intensity of the spectral signatures arising from specific pigments associated with health such as chlorophyll. This can be achieved by using reference targets to correct for incident light variations, thereby normalising spectra so they can be compared between datasets to track changes in pigment intensity and thus bleaching.

Underwater spectrometry can be achieved by using laboratory spectrometers enclosed in waterproof housings with fibre optic probes to record radiance reflectance measurements. The fibre optic probes are held at an orthogonal angle to the solar incidence angle, approximately 0.5–1.0 cm from the target [36]. An accompanying reference measurement is required to normalise for variations in ambient illumination. This is achieved by taking a reflectance measurement from a well characterised, white Lambertian reflectance target such as polytetrafluorethylene (PTFE) or a Spectralon (Labsphere, USA). The reference spectrum enables a correction function to be applied to the data. Specialised spectrometers can also be employed for certain niche applications. For example, pulse amplitude modulation (PAM) fluorometers specifically look at fluorescence to determine the photosynthetic yield. Chlorophyll density can be used to determine relative electron transport rates of photosynthetic organisms to provide a measurement of photosynthetic efficiency [62]. This is a measure of how well chlorophyll converts light into energy and detects compromised tissues that are less efficient. The diving PAM I and II (Walz, Germany) are examples of underwater fluorometers and are the most commonly used devices in studies using this technique [3,63,64]. PAM devices do have limitations. Notably, a requirement for the sampling optical fibre probe to be held in near contact (<5 mm) with the sampled object for a long time (>30 s) to obtain accurate readings.

Spectrometers and fluorometers are able to generate more accurate spectral data but also suffer from many of the same pitfalls as RGB imaging. Data acquisition is typically slow when used to cover a whole reef system. This is mainly due to the small sampling area of the probes and the requirement to make point measurements. This limitation makes the technique particularly unsuitable for large-area surveys. Additionally, multiple points are often sampled on individual corals to obtain average spectra. However, the small number of measurements precludes confidence that these average spectra are truly representative of the whole organism or a whole reef system.

Conversely, spectroscopy techniques using imagers (multispectral and hyperspectral imaging) can generate spectra for every pixel in an image within one data acquisition. This makes the process of data collection quicker and more efficient, thereby facilitating the collection of datasets that are more comprehensive and representative. In turn, imagers can categorise and quantify colour. Spectral imagers generally comprise a dispersive element (either a prism or diffraction grating) or filter, which splits or filters incoming light into wavelengths, and an imaging detector such as a charged coupled device (CCD) or complementary metal–oxide–semiconductor device (CMOS).

Multispectral imagers record data across multiple spectral bands, typically between three and 15 bands [65]. Conversely, hyperspectral imaging records in hundreds of spectral bands, which means data may be collected and processed across the whole visible and/or near infrared spectrum with improved spectral resolution.

Previously, some multispectral systems have been deployed to assess specific marine monitoring cases. These included determining coral fluorescence using narrow bandpass filters [66] and filter wheel style imagers for classification via spectral discrimination [67]. Other imagers have been produced for applications such as the exploration of marine minerals and ores [68], but are not currently being used in coral monitoring surveys [69].

Underwater hyperspectral imaging (UHI) is a relatively new, emerging technology with limited published instances to date. Current diver operated hyperspectral systems such as the "HyperDiver" system [70] can generate hyperspectral and traditional RGB images simultaneously capturing synchronised high-resolution digital images, hyperspectral, and topographic data [70]. The system utilises a push-broom hyperspectral imager (Pika 2, Resonon Inc., Bozeman, MT, USA) with a spectral range of 400–900 nm sampled at ~1.5 nm resolution with 480 fixed bands and 640 spatial pixels [70].

Push broom or line scanning imaging methods acquire full spectral data one spatial line at a time. The line is imaged onto the entrance slit of a spectrometer, which disperses the light into its spectral components before reaching the sensor array. The composite image is constructed by either moving the slit across the image plane or by moving the entire system across the scene [71]. This is advantageous as spectral data can be gathered whilst the imager is moving, which provides both full spectral and spatial data. Other hyperspectral systems such as 'Full data cube snapshot' imagers work from fixed viewpoints, similar to traditional RGB imagers. In this case, a push broom effect is achieved by optically scanning a linear field of view across the hyperspectral detector within the device. The need for a stable platform and the delicate nature of the optics involved make them generally unsuitable for use in UHI.

UHI presents additional potential applications using an 'objects of interest' (OOI) identification technique, as described by Johnsen [72], which includes mapping and monitoring of seafloor habitats for minerals or soft versus hard bottom; seafloor pipeline inspections to determine type of material, cracks, rust, and leakage; shipwrecks (type and state of wood, nails, rust, and artefacts); deep-water coral reefs and sponge fields for species identification, area coverage and physiological state, and kelp forests (species identification, area coverage, physiological state, and growth rates of benthic organisms).

Current UHI technologies (outlined in Table 1) are generally bulky systems that are difficult to deploy and manoeuvre. For example, the "Hyperdiver" system [70] including all its additional sensors and payloads weighs 32 kg in air. Other sensors, specifically the tunable LED-based underwater multispectral imaging system (TuLUMIS) and ocean vision (UHI OV), are designed to be mounted on a UUV. The UUV provides the interface system to operate the camera as well as a translation platform. These are not easily deployed by a diver.

The use of UUVs does, however, eliminate the limitations imposed by diver reliance. For example, dive surveys require substantial amounts of time as there is a finite period a diver can spend underwater; on dives, this is usually dependent on air-tank capacity and depth. Subsequent dives can be achieved through the use of multiple air-tanks but ultimately, a diver will fatigue. The corresponding issue on UUV based surveys is battery life, although multiple batteries can be used to extend the survey time. Crucially, UUVs do not suffer fatigue and can be deployed longer than their human counterparts. A UUV can also cover a larger distance in a shorter time. For example, a 120 m squared area may take two scuba divers up to 2.5 h [4], equating to a surveying rate of 0.13 m2/s. Comparatively, a low-cost remotely operated vehicle (ROV) such as BlueROV2 can achieve survey rates of 1 m2/s. Key limitations to both divers and UUVs are repeatability and accuracy when surveying reefs because global positioning system(s) (GPS) do not work underwater. Acoustic transponder networks designed for UUVs create a way of translating GPS coordinates underwater and thus improve the repeatability and accuracy by recording accurate georeferenced data [57].

The use of UHI on UUVs is currently limited with only a few studies having been reported. One such study [73] used a prototype UHI system for mapping the seafloor for the automated identification of seabed, habitat, and OOI in coral reefs. Other studies [61], specifically using hyperspectral imaging with corals, have mainly focused on coverage and benthic discrimination with machine learning to classify corals and have not focused on assessing health or disease. The current generation of commercially available hyperspectral imagers are often cost prohibitive to both acquire and insure for marine studies. Consequently, there exists a need for technology development and application to study marine environments such as the surveyance of coral health.
