Development and uncertainty analysis of infiltration models using PSO and Monte Carlo method

Abstract

The present research aims to calibrate infiltration models including the Kostiakov (KM), modified Kostiakov (MKM), Novel (NM), Philip (PM), Horton (HM) and Soil Conservation Service (SCSM) models using the particle swarm optimization (PSO) algorithm. To satisfy this end, the published data related to the double-ring test in the Davood Rashid and Hunam regions located in Lorestan and Ilam provinces (western Iran) were applied. Then, Monte Carlo analysis (MCA) was used to model the uncertainty of the coefficients of the mentioned models. The lowest accuracy in the Hunam region is related to MKM with the statistical indices coefficient of determination (R 2 ) = 0.901 and root mean square error (RMSE) = 0.023, while in the same region, the highest accuracy is related to the NM with the statistical indices R 2 = 0.988 and RMSE = 0.008. For the Davood Rashid region, the lowest accuracy is related to the HM with the statistical indices R 2 = 0.890 and RMSE = 0.038. In this region, the highest accuracy is related to the NM and MKM with the statistical indices R 2 = 0.987 and RMSE = 0.012. To determine the uncertainty of the parameters of the infiltration models, 10,000 datasets were generated based on observed data and the MCA, and then their range at the 95% confidence level was calculated. According to the MCA, among the models used in the study areas, the NM has less uncertainty and the results of this model can be adopted with more reliability.

Impact of Storm Characteristics on Infiltration Dynamics in Sponge Cities Using SWMM

“…Consequently, numerous theoretical and empirical infiltration models have been developed for indirect estimation [5,6]. Infiltration models can be categorized into two types [7]: physically-based equations such as Horton [8][9][10], Green-Ampt [11], Soil Conservation Service [12], Swartzendruber [13], Kostiakov, Kostiakov-Lewis, and Philip; and empirical and data-driven methods including artificial neural networks [14], support vector machines [15], random-forest models [16], and Gene Expression Programming [17].…”