Introduction
Saprolegniosis caused by Saprolegnia spp. oomycetes, also referred to as water mold, is a serious threat to fish farming and wild fish and other aquatic animals in fresh water environments. It is one of the major obstacles in fish production and even threatens some highly endangered species (Eronen et al. 2024). Currently, it is still inadequately understood which abiotic and biotic factors in water quality or other conditions in nature and at fish farms influence active saprolegniosis incidences (Pavi�� et al. 2022). Stressful conditions in aquaculture farms (temperature shock, additional infections, overcrowding, injuries) can weaken the fish��s immune system and increase its susceptibility to S. parasitica. However, it is still unclear how these interactions of different factors work (Lindholm-Lehto and Pylkk�� 2024).
Materials and Methods
Using daily water-quality and fish mortality data from Natural Resources Institute Finland (Luke)��s Enonkoski aquaculture infrastructure (2024-2025) rearing endangered landlocked salmon (Salmo salar m. sebago) for re-stocking, we developed an interpretable Random Forest model that predicts infection occurrence. Infection densities were estimated with a walk-forward two-part hurdle generalized additive model (GAM), and past-only lag and rolling features (1–28 days) were derived from water-quality variables.
Results
A weighted Random Forest using the 50 top-ranked features achieved strong probabilistic performance (Brier = 0.096 ± 0.037; LogLoss = 0.341 ± 0.194; ROC-AUC = 0.989). Calibrated forecasts correctly identified the start of a new infection period, with late-stage lethal infection probability rising several daysbefore the first detected cases. SHapley Additive exPlanations (SHAP) and permutation analyses revealed that infection risk was highest under oxygen-rich, moderately warm (12–17°C), and stable organic matter conditions, whereas high turbidity and strong UV254 (aromatic organic matter) variability reduced risk
Figure 1. Infection probability forecasts for an independent test period (September–October 2025). The blue line shows predicted infection probability, orange bars indicate observed infection density (normalised for visual comparison), and open red circles mark days with zero observed infection. The dashed horizontal line denotes the operational decision threshold (0.5). Infection probability begins to rise approximately 15 days before the first observed infection.
Discussion
An interpretable Random Forest model was able to quantify how water-quality dynamics over the preceding month predict late-stage lethal S. parasitica infections in an aquaculture system. By combining daily water-quality measurements with infection observations and model-based label estimation, we identified key environmental precursors and lag structures preceding infection events. Our goal was not only to achieve accurate prediction, but to provide mechanistic insight into the environmental context favouring S. parasitica growth and to demonstrate a transparent modelling framework for future early-warning applications.
Acknowledgments
The experiment was funded by the European Union through the European regional Fund (EAKR) The Finland 2021–2027 innovation and skills in Finland program, municipality of Enonkoski, and Hanka-Taimen Oy. We would like to thank the staff at the Natural Resources Institute Finland, Enonkoski.
References
Eronen, A., Janhunen, M., Hyv��rinen, P., Kortet, R., Karvonen, A., 2024. The Effects of hybridization and parasite infection on the survival and behaviour of endangered landlocked salmon subject to predation —Implications for genetic rescue. Evolut Appl 17, e70056. doi.org/10.1111/eva.70056
Lindholm-Lehto, P. C., Pylkk��, P., 2024. Saprolegniosis in aquaculture and how to control it? Aquacult Fish Fisheries 4, e2200. doi.org/10.1002/aff2.200
Pavi��, D., Grbin, D., Hudina, S., Zmrzljak, U.P., Miljanovi��, A., Ko��ir, R. et al. 2022. Tracing the oomycete pathogen Saprolegnia parasitica in aquaculture and the environment. Sci Rep 12, 16646. doi.org/10.1038/s41598-022-16553-0