Seminar: Ivica Vilibić: Forecasting ocean surface currents using neural networks, high-frequency ocean radars and numerical weather prediction models

Monday, 21st September 2015, 13:15
Room F2, 1st floor, FMF, Jadranska 19, Ljubljana

Meteorology Group will host two seminars presented by our visitors from Institute of Oceanography and Fisheries, Split, Croatia. The seminars will be in English. Everyone is kindly invited to join.


The lecture will try to synthesize the major results obtained by the ongoing NEURAL project (, which has an aim to research and to build an efficient and reliable prototype of ocean surface current forecasting system, based on high-frequency (HF) radar measurements, numerical weather prediction (NWP) model outputs and neural network algorithms (Self-Organizing Maps). Such a possibility has been introduced when joint mapping of modelled mesoscale surface winds and surface currents measured by HF radars in a coastal area pointed to a high correlation between them. The technological component of the project includes installation of new HF radars in the coastal area of the middle Adriatic and development and implementation of data management procedures. The research component of the project includes an assessment of different combination of input variables (radial vs. Cartesian vectors, original vs. detided vs. filtered series, WRF-ARW vs. Aladin meteorological model, varying of the model domain) in order to get the best hindcasted surface currents. The operational component of the project is based on using the NWP operational products for short-term forecasting of surface currents. Both historical and newly observed HF radar current data and both reanalysis and operational NWP model runs are used. The hindcasted ocean currents are found to have lower rmse compared to the observations in respect to the operational ROMS model outputs. The disadvantage of suggested system is that it can be applied only in areas where long series of surface currents exist and where the recognized patterns can be properly ascribed to the forcing fields though a neural network approach.