Routine biomonitoring is key to understanding how aquaculture affects the marine environment and the ecosystems around it. Accurate identification of marine species is essential for understanding biodiversity and ecosystem health, and benthic macroinvertebrates (macrobenthos) are widely used as indicators of environmental change in monitoring programs. However, their identification through morphology is labour-intensive and requires specialized taxonomic expertise. A promising alternative to this is environmental DNA (eDNA) sequencing, which detects traces of genetic material left behind by organisms in the environment, or, when sufficiently small, the organisms themselves. This approach – especially when focused on microorganisms – has already shown strong potential in aquaculture and other fields.
Some recent studies have moved beyond large bottom-dwelling animals (macrobenthos) and started looking at much smaller organisms, such as microbes and meiofauna. These groups are particularly attractive as bioindicators because they respond quickly to environmental changes, especially factors like redox potential and nutrient availability. Their small size and more even distribution in sediments also make them easier to study using eDNA based methods. However, this presents a challenge; unlike larger organisms, we still lack well-established biotic indices and indicator species for microbes and meiofauna (with some notable exceptions such as microgAMBI and nema-gAMBI). This is where data-driven approaches come in. Instead of relying on predefined indicator species, these methods use patterns in the data to build new tools for environmental assessment.
One such approach is Random Forest Regression (RFR), a machine learning method that has gained considerable attention recently for its ability to predict ecosystem quality. It has already been applied in salmon aquaculture, and we have recently tested it in a novel offshore tuna farming installation in the Bay of Biscay. Here, I summarize the results of these studies, with a particular focus on using physicochemical pressure indices as a less biased alternative to traditional morphology-based indices when defining "ground truth".