Towards an operational phytoplankton bloom monitoring and forecasting system for the large deep Lake Geneva (France/Switzerland)

Phytoplankton blooms can affect lake ecosystem services, particularly in large lakes where their impacts are often localized in space and time. To anticipate areas at risk and support management responses, an operational framework for monitoring and forecasting phytoplankton blooms is proposed and applied to the large deep Lake Geneva (France/Switzerland). This alert system integrates in situ observations, satellite chlorophyll-a (Chl-a) and Secchi depth imagery, and three-dimensional hydrodynamic simulations within an operational workflow. The monitoring module analyzes near-real-time satellite Chl-a to identify areas exceeding the World Health Organization risk threshold of 10 µg Chl-a L - ¹, triggering local or global alerts indicative of surface bloom conditions potentially affecting water uses. The forecasting module predicts short-term surface bloom transport by advecting virtual particles within the hydrodynamic model. Model performance was evaluated using two historical surface bloom events, showing good agreement between observations and forecasts, with mean trajectory errors of 0.3-1.7 km and maximum errors below 6.1 km. Forecast skill decreased beyond three days and during bloom initiation or decay phases. To address the limitations of satellite observations, a third module detects deep Chl-a exceedances above 10 µg L - ¹ not visible by satellite. This module relies on a Random Forest machine learning model trained on a 10-year in situ dataset from the central lake station and successfully detects 78% of deep bloom events. By enabling daily monitoring and forecasting, the alert system complements conventional in situ programs and helps identify short-lived or spatially localized blooms. Its modular, open-access design facilitates transferability to other lakes.