Research papersOn the use of surrogate-based modeling for the numerical analysis of Low Impact Development techniques

- The use of surrogate-based modeling for LIDs analysis was investigated.
- The kriging technique was used to approximate the response of a mechanistic model.
- Surrogate-based sensitivity analysis and global optimization were carried out.
- The model is validated on an independent set of data with good results.

Mechanistic models have proven to be accurate tools for the numerical analysis of the hydraulic behavior of Low Impact Development (LIDs) techniques. However, their widespread adoption has been limited by their computational cost. In this view, surrogate modeling is focused on developing and using a computationally inexpensive surrogate of the original model. While having been previously applied to various water-related and environmental modeling problems, no studies have used surrogate models for the analysis of LIDs. The aim of this research thus was to investigate the benefit of surrogate-based modeling in the numerical analysis of LIDs. The kriging technique was used to approximate the deterministic response of the widely used mechanistic model HYDRUS-2D, which was employed to simulate the variably-saturated hydraulic behavior of a contained stormwater filter. The Nash-Sutcliffe efficiency (NSE) index was used to compare the simulated and measured outflows and as the variable of interest for the construction of the response surface. The validated kriging model was first used to carry out a Global Sensitivity Analysis of the unknown soil hydraulic parameters of the filter layer, revealing that only the shape parameter α and the saturated hydraulic conductivity Ks significantly affected the model response. Next, the Particle Swarm Optimization algorithm was used to estimate their values. The NSE value of 0.85 indicated a good accuracy of estimated parameters. Finally, the calibrated model was validated against an independent set of measured outflows with a NSE value of 0.8, which again corroborated the reliability of the surrogate-based optimized parameters.