Modelling biofouling growth on water quality sensors using survival analysis

Biofouling is the primary cause of maintenance and data gaps for online underwater water quality sensors, which are crucial for environmental compliance. This study proposes a novel method for predicting the likelihood and timing of biofouling events, moving beyond reactive sensor servicing and creating a proactive maintenance strategy. Utilising a comprehensive survival analysis dataset derived from time series data and service logs from 14 monitoring stations on the Fehmarnbelt Fixed Link project, we constructed a comprehensive survival dataset supported by Large Language Model (LLM) extraction of biofouling related events from unstructured service reports and quality-control notes. We evaluated multiple survival analysis models and found that a gradient boosted Cox proportional hazards model delivered the highest predictive performance (concordance index: 0.993). The model reliably identifies short-term fouling risk and provides decision support for optimising maintenance schedules. This model is highly promising as a decision support tool to optimise maintenance schedules, potentially reducing costs and environmental footprint while significantly extending sensor uptime.