Estimating parameters of empirical infiltration models from the global dataset using machine learning

Infiltration is the key process of the hydrological cycle. Infiltration estimates are of paramount importance in flood and drought management, irrigation and drainage system design, groundwater recharge assessment, subsurface flow, and contaminant transport investigation and modelling. A large number of equations have been proposed to simulate and predict infiltration (Mishra et al., 2003). Both physics-based equations, e.g.: Brutsaert (1977), Green and Ampt (1911), Kutílek and Krejča (1987), Philip (1957), Swartzendruber (1987), and empirical equations, e.g. Kostiakov (1932), Horton (1940), Holtan (1961), Mezencev (1948) are in use. Infiltration measurements are both time consuming and labour-intensive and are therefore impractical for largescale projects. Such projects benefit from predictive models that relate the parameters of the infiltration equations to the readily available or more easily attainable site-specific data. Estimating the parameters of the infiltration equations from their soil and landscape properties has led to the development of special types of pedotransfer function (Pachepsky and Rawls, 2003). The parameters of various infiltration equations have been estimated using basic soil properties, © 2021 Institute of Agrophysics, Polish Academy of Sciences