Data-Driven Modeling Architecture of Wave Climate Along Compressional Margins

We present a data-driven modeling architecture of wave climate variability along continental shelves of compressional margins in Northern South America by defining a dynamical system from the variable wave energy flux, P in W/m, with {P R|P>=0}. We use wave energy flux data calculated from wave statistics, a reanalysis model of wave dynamics using discrete spatial locations and time realizations. We then determine the functional form of the system from sparse regression methods, numerical differentiation schemes, and model selection metrics. We further describe the latitudinal variability of P in a complex system framework to produce a reduced-order model (ROM) of the system of interest. We discover, reduce, and solve the symbolic expression that forecasts the coastal variability to obtain a parsimonious model (balancing accuracy and complexity) of nearshore wave energy flux. We train the discovered ROM using data from 1980 to 2000 to predict results from 2000 to 2010. We find significant agreement between extrapolated values and high-fidelity global simulations (NRMSE=15.14%), reducing the system significantly from a 13x87663 PDE to a 3x360 ODE.