AI revolutionizes marine research in Andratx with the analysis of 80,000 underwater images

A study led by IMEDEA (CSIC-UIB) analyzes more than 80,000 underwater images and develops a way to estimate the relative abundance of fish by taking into account the volume of water observed by the camera.

A scientific team led by the Mediterranean Institute for Advanced Studies, IMEDEA (CSIC-UIB), has identified significant changes in fish communities along the Mallorcan coast thanks to a long-term underwater monitoring system installed in the port of Andratx. The analysis made it possible to reconstruct daily and seasonal patterns of numerous species commonly found in the western Mediterranean.

The study highlights that many of these species showed behaviors closely linked to sunlight, water temperature or marine productivity. The most striking result, however, was the appearance of an abrupt shift in the time series from 2022 onwards. Several species frequently found in the area, especially sparids of interest to artisanal and recreational fishing in the Balearic Islands, such as the white seabream or sard (Diplodus sargus), the common two-banded seabream or variada (Diplodus vulgaris), and the saddled seabream (Oblada melanura), began to appear more frequently and in higher concentrations within the monitored area.

The authors warn, however, that this phenomenon does not necessarily imply a change in fish populations. Although the Mediterranean experienced an intense marine heatwave in 2022, the ecological models used indicate that some of these species respond neutrally or even negatively to rising temperatures. According to the researchers, the observed increase could reflect local changes in the distribution or behavior of fish, without necessarily indicating an overall increase in populations.

Indeed, the study points to a combination of local and regional factors, such as possible physical changes in the wreck used as a reference point, the simultaneous impact of extreme climate events, or other environmental factors. The interaction among these factors may have altered the way different coastal species use the habitat.

The research was carried out between 2016 and 2024 using a fixed camera located at a depth of eight meters, next to a small wreck at the entrance to the natural harbor. The device automatically captured images every few minutes and sent them in real time to the research center for storage and analysis. Using this material, the researchers applied computer vision tools and deep learning systems, an artificial intelligence technique capable of automatically recognizing patterns in images, to detect and classify fish automatically. Over the nine years of the study, the system identified more than half a million individual fish belonging to 17 different taxonomic groups.

One of the study’s main advances was the development of a method to estimate local fish density more accurately. To do so, the researchers took into account the volume of water visible in each image, making it possible to correct for differences caused by changes in visibility, lighting, or camera position. Thanks to this system, the data can be compared more reliably across different times and environmental conditions.

The researchers stress that these types of tools open up new possibilities for studying how marine ecosystems change over time. The use of artificial intelligence makes it possible to automatically analyze large volumes of underwater images and detect ecological variations that would be very difficult to observe through occasional sampling. According to the authors, this methodology could be applied in future underwater observation networks to monitor the evolution of marine communities and detect rapid changes associated with environmental or climate pressures.

“The main contribution of the study is not only that we observed changes in a specific area, but that we demonstrated that artificial intelligence applied to underwater images can already be used to generate useful ecological indicators in long time series. We have moved from a more experimental phase to direct applications for monitoring fish communities and their relationship with the environment,” explains Ignasi A. Catalán, IMEDEA researcher and lead author of the study. “In addition, the study shows that simply counting fish in images is not enough. To compare long time series, years, or observation stations, we need more standardized metrics, such as how much volume of water the camera is actually observing,” Catalán adds.

This work is the result of several projects funded by the Union-Next Generation EU program and a collaboration between IMEDEA, UIB, LIMIA-IRFAP and SOCIB.