Machine-learning prediction of late-stage saprolegniosis risk in freshwater aquaculture from water-quality data

Abstract Saprolegniosis, also known as water mould disease, causes substantial mortality in freshwater animals, threatening endangered species and imposing major economic losses on aquaculture. However, evidence regarding its environmental drivers remains inconsistent. We used daily aggregated water-quality data and fish saprolegniosis mortality records from a landlocked salmon hatchery to identify environmental predictors of late-stage saprolegniosis and assess their potential for early warning. A weighted random forest predicted saprolegniosis risk from water-quality data for the preceding 28 days, achieving strong predictive performance (Brier score = 0.096; log loss = 0.341; ROC–AUC = 0.989). When applied to new monitoring data, the model indicated increasing disease risk 15 days before the first observed cases of the subsequent infection period. Model interpretation associated higher late-stage saprolegniosis risk with oxygen-rich water, moderate temperatures (12–17°C), and stable organic-matter conditions, whereas high turbidity and greater variability in UV-absorbing organic matter were associated with lower risk. This transparent, data-driven framework could support the development of early-warning approaches to tackle saprolegniosis in aquaculture.