A tree-based feature ranking approach to enhance emulation modelling of 3D hydrodynamic-ecological models

The paper presents an emulation modelling approach to reduce the computational burden associated with the use of process-based, distributed, dynamic models in planning and management of water resources. This approach, which relies on a procedure composed by few steps, leads to the identification of an emulation model, namely a simplified, computationally-efficient model built over a sample data-set produced via simulation of the original computationally expensive model. The core of the procedure is a feature ranking algorithm, based on Extremely Randomized Trees (Extra-Trees), through which the most relevant input variables to the emulation model are selected among the large set of candidate input variables associated with the original model. The calibration of the emulation model, in the form of an input-output relationship, is then performed using again Extra-Trees. The approach is demonstrated on a real-world case study (Googong Reservoir, AUS).