Ocean colour satellites can be used to measure light attenuation in the water column – a crucial factor when processing images.
By Derya Akkaynak and Tali Treibitz (Marine Imaging Lab, University of Haifa), Ben Loveday and Hayley Evers-King (PML)
Arguably one of the most challenging — and currently unsolved — problems in marine imaging is the robust, consistent, and accurate reconstruction of lost colours and contrast in underwater photographs.
While large image datasets taken on land can be analysed with a plethora of computer-vision and machine-learning algorithms that can segment, recognise, count, and classify objects, underwater datasets do not benefit from the full power of these methods because water degrades images too severely for automated analysis.
Existing methods that try to recover lost colours perform inconsistently. According to researchers at the University of Haifa this is because these methods rely on an equation that inaccurately represents how light travels in water. They have proposed a revised equation that corrects these inaccuracies, and they are now developing a method to efficiently remove the effects of water from underwater photographs.
They study coral reefs in Eilat, in the northernmost tip of the Red Sea. For every dataset they collect, they must also measure or estimate attenuation coefficients corresponding to water conditions on that day, and they use these coefficients to remove the degrading effects of water from images. These coefficients can be measured in the water using specialised instruments, but this is costly and impractical, so they have turned to freely-available satellite imagery for the estimation of the optical parameters they need (Figure 1, top right, click to expand).
They use Ocean and Land Colour Imager (OLCI) products from Sentinel-3, which are available on an almost daily basis, because the area around Eilat is a vast desert and free of clouds for 350 days of the year. The OLCI sensor offers a number of special benefits for this application — providing relatively high temporal (1–2 day revisit) and spatial (300 m) resolution, while also having suitable spectral characteristics for observing these types of ocean waters. Comparison of the products retrieved from OLCI with in situ measurements made from the team at the University of Haifa, confirm that the products are suitably accurate for their applications (Figure 2).
Until now, the analysis of large datasets from coral reefs, sea floor, and other oceanic habitats required a lot of manual effort. The robust method researchers at the University of Haifa are developing may open up troves of underwater imagery collected every day by automated analysis, and help scientists globally who are monitoring the threats faced by these important ocean ecosystems.