Particle Swarm Optimization for Inverse Modeling of Soils in Urban Green Stormwater Infrastructure Sites

Abstract

The measurement of soil parameters at green stormwater infrastructure (GSI) sites is a labor and time-intensive process. Use of machine learning and inverse modeling techniques to estimate soil parameters provides an answer to this issue. In this paper a particle swarm optimization (PSO) algorithm is used in conjunction with inverse modeling using Hydrus-1D to estimate soil parameters. The novelty of this work is the implementation of PSO to identify soil infiltration models in a functioning urban field site using data from deployed sensors. The linear bioinfiltration site, located in Philadelphia, Pennsylvania, has two layers of soil: a top layer designed for the site and a lower layer native to the site. The PSO was used to estimate parameters for each of these two soils, as well as the depth of the top engineered soil. The resulting simulation using the estimated parameters showed a promising fit to measured soil moisture data, an RMS error of 0.017 in validation testing, and the parameters themselves were estimated more accurately than assuming a standard soil type. This lays the groundwork for using PSO and inverse modeling in conjunction with continuous soil moisture monitoring to enable long-term continuous modeling of GSI sites to determine performance degradation and enable on-demand maintenance.