Predicting sea surface salinity in a tidal estuary with machine learning

As an indicator of exchanges between watersheds, rivers and coastal seas, salinity may provide valuable information about the exposure, ecological health robustness marine ecosystems, including especially estuaries. The temporal variations are traditionally approached with numerical models based on a physical description hydrodynamic hydrological processes. However, as these require large computational resources, such approach is, in practice, rarely considered for rapid turnaround predictions requested by engineering operational applications dealing monitoring alternative efficient solution, we investigated here potential machine learning algorithms to mimic non-linear complex relationships series input parameters (such tide-induced free-surface elevation, river discharges wind velocity). Beyond regression methods, attention was dedicated popular approaches MultiLayer Perceptron, Support Vector Regression Random Forest. These were applied six-year observations sea surface at mouth Elorn estuary (bay Brest, western Brittany, France) compared from advanced model. In spite simple data, reproduced seasonal semi-diurnal characterised noticeable modulations low-salinity events during winter period. provided best estimations salinity, improving model events. This promotes exploitation complementary tool process-based models.